entry_point
stringlengths 1
65
| original_triton_code
stringlengths 4.5k
619k
| python_code
stringlengths 208
60.9k
| triton_code
stringlengths 1.15k
275k
| repo_name
stringlengths 7
115
| module_name
stringlengths 1
65
| synthetic
bool 1
class | uuid
int64 0
18.5k
| licenses
sequencelengths 1
6
| stars
int64 0
19.8k
| sha
stringlengths 40
40
| repo_link
stringlengths 72
180
| pytorch_code
stringlengths 200
4.05k
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
RegL1 | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/5c/c5c76tomh2lks74yoej4jwcze4mhqjftq6po2to4epppt3h7t52a.py
# Topologically Sorted Source Nodes: [abs_1, z1], Original ATen: [aten.abs, aten.sum]
# Source node to ATen node mapping:
# abs_1 => abs_1
# z1 => sum_1
# Graph fragment:
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%primals_1,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%abs_1,), kwargs = {})
triton_per_fused_abs_sum_0 = async_compile.triton('triton_per_fused_abs_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {2: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_abs_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_abs_sum_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = tl_math.abs(tmp0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.sum(tmp2, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp4, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_2
buf1 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [abs_1, z1], Original ATen: [aten.abs, aten.sum]
stream0 = get_raw_stream(0)
triton_per_fused_abs_sum_0.run(primals_1, buf1, 1, 16, grid=grid(1), stream=stream0)
return (reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0), buf1, primals_1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class RegL1(nn.Module):
"""
Run Regression with L1
"""
def __init__(self, n_input, n_output):
super(RegL1, self).__init__()
self.linear = nn.Linear(n_input, n_output, bias=True)
def forward(self, x, training=True):
self.training = training
x = self.linear(x)
z1 = torch.sum(torch.abs(self.linear.weight))
self.training = True
return x, z1
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_input': 4, 'n_output': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_abs_sum_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = tl_math.abs(tmp0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.sum(tmp2, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp4, None)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_2
buf1 = empty_strided_cuda((), (), torch.float32)
get_raw_stream(0)
triton_per_fused_abs_sum_0[grid(1)](primals_1, buf1, 1, 16, XBLOCK=
1, num_warps=2, num_stages=1)
return reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0
), buf1, primals_1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0)
class RegL1New(nn.Module):
"""
Run Regression with L1
"""
def __init__(self, n_input, n_output):
super(RegL1New, self).__init__()
self.linear = nn.Linear(n_input, n_output, bias=True)
def forward(self, input_0):
primals_1 = self.linear.weight
primals_2 = self.linear.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0], output[1]
| rmporsch/ML_genetic_risk | RegL1 | false | 4,197 | [
"MIT"
] | 0 | 4e1a0510c94260e69f93639ff4104c5f85080d9f | https://github.com/rmporsch/ML_genetic_risk/tree/4e1a0510c94260e69f93639ff4104c5f85080d9f | import torch
import torch.nn as nn
class Model(nn.Module):
"""
Run Regression with L1
"""
def __init__(self, n_input, n_output):
super().__init__()
self.linear = nn.Linear(n_input, n_output, bias=True)
def forward(self, x, training=True):
self.training = training
x = self.linear(x)
z1 = torch.sum(torch.abs(self.linear.weight))
self.training = True
return x, z1
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
DecoderRNN | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ce/cce2l7xu6vkn6rmf2fwkbam27ms7thykgzluzgv2xydzvmji5gt4.py
# Topologically Sorted Source Nodes: [add, a_t, h_t], Original ATen: [aten.add, aten.tanh]
# Source node to ATen node mapping:
# a_t => add_1
# add => add
# h_t => tanh
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %view_3), kwargs = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_1,), kwargs = {})
triton_poi_fused_add_tanh_0 = async_compile.triton('triton_poi_fused_add_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_tanh_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_tanh_0(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x3 = xindex
x4 = xindex % 64
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (x3), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tmp7 = tmp5 + tmp6
tmp8 = tmp4 + tmp7
tmp9 = libdevice.tanh(tmp8)
tl.store(in_out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xk/cxkugsynlmnyrjhah42fewrhwovuvurnuv2qimo2qhxq27wjmq7q.py
# Topologically Sorted Source Nodes: [y_t_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# y_t_3 => amax_3, exp_3, sub_3
# Graph fragment:
# %amax_3 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_23, [1], True), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_23, %amax_3), kwargs = {})
# %exp_3 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_3,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/jf/cjfzp64ny4hf7wdw5wptah3hqv5fcsh5rrw4brz7uxcy6ad57n7h.py
# Topologically Sorted Source Nodes: [y_t_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# y_t_3 => div_3, sum_4
# Graph fragment:
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_3, [1], True), kwargs = {})
# %div_3 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_3, %sum_4), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (1, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf1)
buf2 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [add, a_t, h_t], Original ATen: [aten.add, aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_add_tanh_0.run(buf2, primals_1, primals_3, buf1, primals_7, 256, grid=grid(256), stream=stream0)
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf3)
buf4 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 64), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf4)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [add_2, a_t_1, h_t_1], Original ATen: [aten.add, aten.tanh]
triton_poi_fused_add_tanh_0.run(buf5, primals_1, primals_3, buf4, primals_7, 256, grid=grid(256), stream=stream0)
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf5, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf6)
buf7 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 128), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf7)
buf8 = reinterpret_tensor(buf6, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [add_4, a_t_2, h_t_2], Original ATen: [aten.add, aten.tanh]
triton_poi_fused_add_tanh_0.run(buf8, primals_1, primals_3, buf7, primals_7, 256, grid=grid(256), stream=stream0)
buf9 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf8, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf9)
buf10 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 192), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf10)
del primals_6
buf11 = reinterpret_tensor(buf9, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [add_6, a_t_3, h_t_3], Original ATen: [aten.add, aten.tanh]
triton_poi_fused_add_tanh_0.run(buf11, primals_1, primals_3, buf10, primals_7, 256, grid=grid(256), stream=stream0)
del buf10
del primals_1
del primals_3
del primals_7
buf12 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [o_t_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_9, reinterpret_tensor(buf11, (64, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf12)
del primals_9
buf13 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [y_t_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf12, buf13, 256, grid=grid(256), stream=stream0)
buf14 = reinterpret_tensor(buf12, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf12 # reuse
# Topologically Sorted Source Nodes: [y_t_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf13, buf14, 256, grid=grid(256), stream=stream0)
del buf13
return (buf14, buf11, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), reinterpret_tensor(primals_5, (16, 4), (4, 1), 0), buf2, reinterpret_tensor(primals_5, (16, 4), (4, 1), 64), buf5, reinterpret_tensor(primals_5, (16, 4), (4, 1), 128), buf8, reinterpret_tensor(primals_5, (16, 4), (4, 1), 192), buf11, buf14, primals_8, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.nn.functional as F
class DecoderRNN(nn.Module):
def __init__(self, T, d):
super().__init__()
self.T = T
self.d = d
self.W = nn.Linear(d, d)
self.U = nn.Linear(d, d)
self.V = nn.Linear(d, d)
self.b = nn.Parameter(torch.rand(1, d))
def forward(self, x, h_t):
y_t = None
for t in range(self.T):
a_t = self.b + self.W(h_t) + self.U(x[t])
h_t = torch.tanh(a_t)
o_t = self.V(h_t)
y_t = F.softmax(o_t, 1)
return y_t, h_t
def init_hidden(self):
return torch.zeros(1, self.d)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'T': 4, 'd': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_tanh_0(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2,
in_ptr3, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x3 = xindex
x4 = xindex % 64
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + x3, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tmp7 = tmp5 + tmp6
tmp8 = tmp4 + tmp7
tmp9 = libdevice.tanh(tmp8)
tl.store(in_out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (1, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_4, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf1)
buf2 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_add_tanh_0[grid(256)](buf2, primals_1, primals_3,
buf1, primals_7, 256, XBLOCK=128, num_warps=4, num_stages=1)
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf3)
buf4 = buf1
del buf1
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 64
), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf4)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused_add_tanh_0[grid(256)](buf5, primals_1, primals_3,
buf4, primals_7, 256, XBLOCK=128, num_warps=4, num_stages=1)
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf5, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf6)
buf7 = buf4
del buf4
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1),
128), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf7)
buf8 = reinterpret_tensor(buf6, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf6
triton_poi_fused_add_tanh_0[grid(256)](buf8, primals_1, primals_3,
buf7, primals_7, 256, XBLOCK=128, num_warps=4, num_stages=1)
buf9 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf8, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf9)
buf10 = buf7
del buf7
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1),
192), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf10)
del primals_6
buf11 = reinterpret_tensor(buf9, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf9
triton_poi_fused_add_tanh_0[grid(256)](buf11, primals_1, primals_3,
buf10, primals_7, 256, XBLOCK=128, num_warps=4, num_stages=1)
del buf10
del primals_1
del primals_3
del primals_7
buf12 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_9, reinterpret_tensor(buf11, (64, 4),
(4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf12)
del primals_9
buf13 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf12, buf13, 256, XBLOCK=
256, num_warps=4, num_stages=1)
buf14 = reinterpret_tensor(buf12, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf12
triton_poi_fused__softmax_2[grid(256)](buf13, buf14, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del buf13
return buf14, buf11, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0
), reinterpret_tensor(primals_5, (16, 4), (4, 1), 0
), buf2, reinterpret_tensor(primals_5, (16, 4), (4, 1), 64
), buf5, reinterpret_tensor(primals_5, (16, 4), (4, 1), 128
), buf8, reinterpret_tensor(primals_5, (16, 4), (4, 1), 192
), buf11, buf14, primals_8, primals_2
class DecoderRNNNew(nn.Module):
def __init__(self, T, d):
super().__init__()
self.T = T
self.d = d
self.W = nn.Linear(d, d)
self.U = nn.Linear(d, d)
self.V = nn.Linear(d, d)
self.b = nn.Parameter(torch.rand(1, d))
def init_hidden(self):
return torch.zeros(1, self.d)
def forward(self, input_0, input_1):
primals_1 = self.b
primals_2 = self.W.weight
primals_3 = self.W.bias
primals_6 = self.U.weight
primals_7 = self.U.bias
primals_8 = self.V.weight
primals_9 = self.V.bias
primals_4 = input_0
primals_5 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0], output[1]
| rish-16/SHA-RNN | DecoderRNN | false | 4,198 | [
"MIT"
] | 0 | 08c701396217f0b645de043963ff8ec4bf27e835 | https://github.com/rish-16/SHA-RNN/tree/08c701396217f0b645de043963ff8ec4bf27e835 | import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, T, d):
super().__init__()
self.T = T
self.d = d
self.W = nn.Linear(d, d)
self.U = nn.Linear(d, d)
self.V = nn.Linear(d, d)
self.b = nn.Parameter(torch.rand(1, d))
def forward(self, x, h_t):
y_t = None
for t in range(self.T):
a_t = self.b + self.W(h_t) + self.U(x[t])
h_t = torch.tanh(a_t)
o_t = self.V(h_t)
y_t = F.softmax(o_t, 1)
return y_t, h_t
def init_hidden(self):
return torch.zeros(1, self.d)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
SpatialAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/46/c46mg7rvdztu6n5oosf5c4if7ziag6obrxhwbn43lcdfibfuom7w.py
# Topologically Sorted Source Nodes: [temp_x], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# temp_x => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%mean, %getitem], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 16) % 2
x0 = xindex % 16
x2 = (xindex // 32)
x3 = xindex
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp9 = tmp7 + tmp8
tmp10 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp9 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp4, tmp13, tmp14)
tmp16 = tmp0 >= tmp3
tmp17 = tl.full([1], 2, tl.int64)
tmp18 = tmp0 < tmp17
tmp19 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp20 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp21 = triton_helpers.maximum(tmp19, tmp20)
tmp22 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp23 = triton_helpers.maximum(tmp21, tmp22)
tmp24 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp25 = triton_helpers.maximum(tmp23, tmp24)
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp16, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp15, tmp27)
tl.store(out_ptr0 + (x3), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/my/cmyzznlnv3aykf56hqxijqkl7hycovubzmkxdtihopwgojtdv2p3.py
# Topologically Sorted Source Nodes: [attention, x], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# attention => sigmoid
# x => mul
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %primals_1), kwargs = {})
triton_poi_fused_mul_sigmoid_1 = async_compile.triton('triton_poi_fused_mul_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_sigmoid_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_sigmoid_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x2 = (xindex // 64)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (x3), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x3), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 3, 3), (18, 9, 3, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [temp_x], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, buf0, 128, grid=grid(128), stream=stream0)
# Topologically Sorted Source Nodes: [temp_x_1], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention, x], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_1.run(buf1, primals_1, buf2, 256, grid=grid(256), stream=stream0)
return (buf2, primals_1, primals_2, buf0, buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 2, 3, 3), (18, 9, 3, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.utils.data
import torch
import torch.nn as nn
class SpatialAttention(nn.Module):
def __init__(self):
super(SpatialAttention, self).__init__()
self.conv1 = nn.Conv2d(in_channels=2, out_channels=1, kernel_size=3,
padding=1, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
avg_out = torch.mean(x, dim=1, keepdim=True)
max_out, _ = torch.max(x, dim=1, keepdim=True)
temp_x = torch.cat([avg_out, max_out], dim=1)
temp_x = self.conv1(temp_x)
attention = self.sigmoid(temp_x)
x = attention * x
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.utils.data
import torch
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 16 % 2
x0 = xindex % 16
x2 = xindex // 32
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp9 = tmp7 + tmp8
tmp10 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp11 = tmp9 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp4, tmp13, tmp14)
tmp16 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp19 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp20 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp21 = triton_helpers.maximum(tmp19, tmp20)
tmp22 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp23 = triton_helpers.maximum(tmp21, tmp22)
tmp24 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp25 = triton_helpers.maximum(tmp23, tmp24)
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp16, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp15, tmp27)
tl.store(out_ptr0 + x3, tmp28, xmask)
@triton.jit
def triton_poi_fused_mul_sigmoid_1(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x2 = xindex // 64
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr1 + x3, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + x3, tmp3, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 3, 3), (18, 9, 3, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(128)](primals_1, buf0, 128, XBLOCK=128,
num_warps=4, num_stages=1)
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_1[grid(256)](buf1, primals_1, buf2,
256, XBLOCK=256, num_warps=4, num_stages=1)
return buf2, primals_1, primals_2, buf0, buf1
class SpatialAttentionNew(nn.Module):
def __init__(self):
super(SpatialAttentionNew, self).__init__()
self.conv1 = nn.Conv2d(in_channels=2, out_channels=1, kernel_size=3,
padding=1, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.conv1.weight
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
| robvincen/robot_gradet | SpatialAttention | false | 4,199 | [
"BSD-3-Clause"
] | 0 | a39e3c772c72806dfc99e4d24d8787e0d1bdeef5 | https://github.com/robvincen/robot_gradet/tree/a39e3c772c72806dfc99e4d24d8787e0d1bdeef5 | import torch
import torch.utils.data
import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(in_channels=2, out_channels=1, kernel_size=3,
padding=1, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
avg_out = torch.mean(x, dim=1, keepdim=True)
max_out, _ = torch.max(x, dim=1, keepdim=True)
temp_x = torch.cat([avg_out, max_out], dim=1)
temp_x = self.conv1(temp_x)
attention = self.sigmoid(temp_x)
x = attention * x
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
QNet | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/cw/ccwssfx6qcvunhe3oavlbicw6nznahssfugkgjkuhgpjkulcytk6.py
# Topologically Sorted Source Nodes: [o1], Original ATen: [aten.leaky_relu]
# Source node to ATen node mapping:
# o1 => gt, mul, where
# Graph fragment:
# %gt : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%view_1, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.01), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %view_1, %mul), kwargs = {})
triton_poi_fused_leaky_relu_0 = async_compile.triton('triton_poi_fused_leaky_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_leaky_relu_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_leaky_relu_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.01
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5n/c5nnm5j7zky5bxmyowjq4lc7gdpqhzs7nzbrb3occhq5mr35r7m6.py
# Topologically Sorted Source Nodes: [o2], Original ATen: [aten.leaky_relu]
# Source node to ATen node mapping:
# o2 => gt_1, mul_1, where_1
# Graph fragment:
# %gt_1 : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%view_3, 0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_3, 0.01), kwargs = {})
# %where_1 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %view_3, %mul_1), kwargs = {})
triton_poi_fused_leaky_relu_1 = async_compile.triton('triton_poi_fused_leaky_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_leaky_relu_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_leaky_relu_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.01
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4), (4, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 16), (16, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 16), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 16), (256, 64, 16, 1), torch.bool)
buf2 = empty_strided_cuda((4, 4, 4, 16), (256, 64, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [o1], Original ATen: [aten.leaky_relu]
stream0 = get_raw_stream(0)
triton_poi_fused_leaky_relu_0.run(buf0, primals_2, buf1, buf2, 1024, grid=grid(1024), stream=stream0)
del buf0
del primals_2
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf2, (64, 16), (16, 1), 0), reinterpret_tensor(primals_4, (16, 4), (1, 16), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [o2], Original ATen: [aten.leaky_relu]
triton_poi_fused_leaky_relu_1.run(buf3, primals_5, buf4, buf5, 256, grid=grid(256), stream=stream0)
del buf3
del primals_5
return (buf5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, reinterpret_tensor(buf2, (64, 16), (16, 1), 0), buf4, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class QNet(nn.Module):
def __init__(self, in_size: 'int', out_size: 'int'):
super(QNet, self).__init__()
self.fc1 = nn.Linear(in_size, 16)
self.fc_out = nn.Linear(16, out_size)
self.act = nn.LeakyReLU()
def forward(self, x):
o1 = self.act(self.fc1(x))
o2 = self.act(self.fc_out(o1))
return o2
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_size': 4, 'out_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_leaky_relu_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.01
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp7, xmask)
@triton.jit
def triton_poi_fused_leaky_relu_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.01
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4), (4, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 16), (16, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 16), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 16), (256, 64, 16, 1), torch.bool)
buf2 = empty_strided_cuda((4, 4, 4, 16), (256, 64, 16, 1), torch.
float32)
get_raw_stream(0)
triton_poi_fused_leaky_relu_0[grid(1024)](buf0, primals_2, buf1,
buf2, 1024, XBLOCK=256, num_warps=4, num_stages=1)
del buf0
del primals_2
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (64, 16), (16, 1), 0),
reinterpret_tensor(primals_4, (16, 4), (1, 16), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_leaky_relu_1[grid(256)](buf3, primals_5, buf4,
buf5, 256, XBLOCK=128, num_warps=4, num_stages=1)
del buf3
del primals_5
return buf5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, reinterpret_tensor(buf2, (64, 16), (16, 1), 0
), buf4, primals_4
class QNetNew(nn.Module):
def __init__(self, in_size: 'int', out_size: 'int'):
super(QNetNew, self).__init__()
self.fc1 = nn.Linear(in_size, 16)
self.fc_out = nn.Linear(16, out_size)
self.act = nn.LeakyReLU()
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc_out.weight
primals_5 = self.fc_out.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| rosebin/gymlab | QNet | false | 4,200 | [
"BSD-3-Clause"
] | 0 | de97fc24e0ddf5e328a2aa732cc339b2371d92d1 | https://github.com/rosebin/gymlab/tree/de97fc24e0ddf5e328a2aa732cc339b2371d92d1 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, in_size: 'int', out_size: 'int'):
super().__init__()
self.fc1 = nn.Linear(in_size, 16)
self.fc_out = nn.Linear(16, out_size)
self.act = nn.LeakyReLU()
def forward(self, x):
o1 = self.act(self.fc1(x))
o2 = self.act(self.fc_out(o1))
return o2
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
L0Linear | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/7g/c7g7othpnsbmovs44m7nfet7tt65cm2vlyavdn5zikazjtnmjrqw.py
# Topologically Sorted Source Nodes: [sigmoid, mul, s, zeros_like, max_1, ones_like, mask, mul_1], Original ATen: [aten.sigmoid, aten.mul, aten.add, aten.zeros_like, aten.maximum, aten.ones_like, aten.minimum]
# Source node to ATen node mapping:
# mask => minimum
# max_1 => maximum
# mul => mul
# mul_1 => mul_1
# ones_like => full_default_1
# s => add
# sigmoid => sigmoid
# zeros_like => full_default
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%primals_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, 1.2000000000000002), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, -0.1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %maximum : [num_users=1] = call_function[target=torch.ops.aten.maximum.default](args = (%add, %full_default), kwargs = {})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %minimum : [num_users=1] = call_function[target=torch.ops.aten.minimum.default](args = (%maximum, %full_default_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %minimum), kwargs = {})
triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0 = async_compile.triton('triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tl.sigmoid(tmp1)
tmp3 = 1.2000000000000002
tmp4 = tmp2 * tmp3
tmp5 = -0.1
tmp6 = tmp4 + tmp5
tmp7 = 0.0
tmp8 = triton_helpers.maximum(tmp6, tmp7)
tmp9 = 1.0
tmp10 = triton_helpers.minimum(tmp8, tmp9)
tmp11 = tmp0 * tmp10
tl.store(out_ptr0 + (x0), tmp11, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sigmoid, mul, s, zeros_like, max_1, ones_like, mask, mul_1], Original ATen: [aten.sigmoid, aten.mul, aten.add, aten.zeros_like, aten.maximum, aten.ones_like, aten.minimum]
stream0 = get_raw_stream(0)
triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0.run(primals_2, primals_1, buf0, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del buf0
del primals_3
return (reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_1, primals_2, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
import torch.nn as nn
from torch.nn import functional as F
from torch.autograd import Variable
import logging as lg
def hard_sigmoid(x):
"""Hard Sigmoid function."""
return torch.min(torch.max(x, torch.zeros_like(x)), torch.ones_like(x))
class _L0Norm(nn.Module):
"""L0 norm."""
def __init__(self, origin, loc_mean: 'float'=0.0, loc_sdev: 'float'=
0.01, beta: 'float'=2 / 3, gamma: 'float'=-0.1, zeta: 'float'=1.1,
fix_temp: 'bool'=True, l02: 'bool'=False, l02_alpha=0.01):
"""Class of layers using L0 Norm.
:param origin: original layer such as nn.Linear(..), nn.Conv2d(..)
:param loc_mean: mean of the normal of initial location parameters
:param loc_sdev: standard deviation of initial location parameters
:param beta: initial temperature parameter
:param gamma: lower bound of "stretched" s
:param zeta: upper bound of "stretched" s
:param fix_temp: True if temperature is fixed
"""
super(_L0Norm, self).__init__()
self._origin = origin
self._size = self._origin.weight.size()
self.loc = nn.Parameter(torch.zeros(self._size).normal_(loc_mean,
loc_sdev))
self.temp = beta if fix_temp else nn.Parameter(torch.zeros(1).fill_
(beta))
self.register_buffer('uniform', torch.zeros(self._size))
self.gamma = gamma
self.zeta = zeta
self.gamma_zeta_ratio = np.log(-gamma / zeta)
self.l02_alpha = l02_alpha
self.l02 = l02
if self.l02:
assert self.l02_alpha > 0
lg.info('Using L_0,2 norm')
def _get_mask(self):
if self.training:
self.uniform.uniform_()
u = Variable(self.uniform)
s = F.sigmoid((torch.log(u) - torch.log(1 - u) + self.loc) /
self.temp)
s = s * (self.zeta - self.gamma) + self.gamma
penalty = F.sigmoid(self.loc - self.temp * self.gamma_zeta_ratio
).sum()
if self.l02:
l02Norm = (F.sigmoid(self.loc - self.temp * self.
gamma_zeta_ratio) * self._origin.weight ** 2).sum()
penalty = penalty + self.l02_alpha * l02Norm
else:
s = F.sigmoid(self.loc) * (self.zeta - self.gamma) + self.gamma
penalty = 0
return hard_sigmoid(s), penalty
class L0Linear(_L0Norm):
"""Linear model with L0 norm."""
def __init__(self, in_features: 'int', out_features: 'int', bias:
'bool'=True, **kwargs):
"""Linear model with L0 norm."""
super(L0Linear, self).__init__(nn.Linear(in_features, out_features,
bias=bias), **kwargs)
def forward(self, input):
"""Forward function with mask and penalty."""
mask, penalty = self._get_mask()
out = F.linear(input, self._origin.weight * mask, self._origin.bias)
return out, penalty
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_features': 4, 'out_features': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import numpy as np
import torch.nn as nn
from torch.nn import functional as F
from torch.autograd import Variable
import logging as lg
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0(
in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tl.sigmoid(tmp1)
tmp3 = 1.2000000000000002
tmp4 = tmp2 * tmp3
tmp5 = -0.1
tmp6 = tmp4 + tmp5
tmp7 = 0.0
tmp8 = triton_helpers.maximum(tmp6, tmp7)
tmp9 = 1.0
tmp10 = triton_helpers.minimum(tmp8, tmp9)
tmp11 = tmp0 * tmp10
tl.store(out_ptr0 + x0, tmp11, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_maximum_minimum_mul_ones_like_sigmoid_zeros_like_0[
grid(16)](primals_2, primals_1, buf0, 16, XBLOCK=16, num_warps=
1, num_stages=1)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_4, (64,
4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf1)
del buf0
del primals_3
return reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0
), primals_1, primals_2, reinterpret_tensor(primals_4, (64, 4), (4,
1), 0)
def hard_sigmoid(x):
"""Hard Sigmoid function."""
return torch.min(torch.max(x, torch.zeros_like(x)), torch.ones_like(x))
class _L0Norm(nn.Module):
"""L0 norm."""
def __init__(self, origin, loc_mean: 'float'=0.0, loc_sdev: 'float'=
0.01, beta: 'float'=2 / 3, gamma: 'float'=-0.1, zeta: 'float'=1.1,
fix_temp: 'bool'=True, l02: 'bool'=False, l02_alpha=0.01):
"""Class of layers using L0 Norm.
:param origin: original layer such as nn.Linear(..), nn.Conv2d(..)
:param loc_mean: mean of the normal of initial location parameters
:param loc_sdev: standard deviation of initial location parameters
:param beta: initial temperature parameter
:param gamma: lower bound of "stretched" s
:param zeta: upper bound of "stretched" s
:param fix_temp: True if temperature is fixed
"""
super(_L0Norm, self).__init__()
self._origin = origin
self._size = self._origin.weight.size()
self.loc = nn.Parameter(torch.zeros(self._size).normal_(loc_mean,
loc_sdev))
self.temp = beta if fix_temp else nn.Parameter(torch.zeros(1).fill_
(beta))
self.register_buffer('uniform', torch.zeros(self._size))
self.gamma = gamma
self.zeta = zeta
self.gamma_zeta_ratio = np.log(-gamma / zeta)
self.l02_alpha = l02_alpha
self.l02 = l02
if self.l02:
assert self.l02_alpha > 0
lg.info('Using L_0,2 norm')
def _get_mask(self):
if self.training:
self.uniform.uniform_()
u = Variable(self.uniform)
s = F.sigmoid((torch.log(u) - torch.log(1 - u) + self.loc) /
self.temp)
s = s * (self.zeta - self.gamma) + self.gamma
penalty = F.sigmoid(self.loc - self.temp * self.gamma_zeta_ratio
).sum()
if self.l02:
l02Norm = (F.sigmoid(self.loc - self.temp * self.
gamma_zeta_ratio) * self._origin.weight ** 2).sum()
penalty = penalty + self.l02_alpha * l02Norm
else:
s = F.sigmoid(self.loc) * (self.zeta - self.gamma) + self.gamma
penalty = 0
return hard_sigmoid(s), penalty
class L0LinearNew(_L0Norm):
"""Linear model with L0 norm."""
def __init__(self, in_features: 'int', out_features: 'int', bias:
'bool'=True, **kwargs):
"""Linear model with L0 norm."""
super(L0LinearNew, self).__init__(nn.Linear(in_features,
out_features, bias=bias), **kwargs)
def forward(self, input_0):
primals_1 = self.loc
primals_2 = self._origin.weight
primals_3 = self._origin.bias
primals_4 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0], output[1]
| rmporsch/ML_genetic_risk | L0Linear | false | 4,201 | [
"MIT"
] | 0 | 4e1a0510c94260e69f93639ff4104c5f85080d9f | https://github.com/rmporsch/ML_genetic_risk/tree/4e1a0510c94260e69f93639ff4104c5f85080d9f | import torch
import numpy as np
import torch.nn as nn
from torch.nn import functional as F
from torch.autograd import Variable
import logging as lg
def hard_sigmoid(x):
"""Hard Sigmoid function."""
return torch.min(torch.max(x, torch.zeros_like(x)), torch.ones_like(x))
class _L0Norm(nn.Module):
"""L0 norm."""
def __init__(self, origin, loc_mean: 'float'=0.0, loc_sdev: 'float'=
0.01, beta: 'float'=2 / 3, gamma: 'float'=-0.1, zeta: 'float'=1.1,
fix_temp: 'bool'=True, l02: 'bool'=False, l02_alpha=0.01):
"""Class of layers using L0 Norm.
:param origin: original layer such as nn.Linear(..), nn.Conv2d(..)
:param loc_mean: mean of the normal of initial location parameters
:param loc_sdev: standard deviation of initial location parameters
:param beta: initial temperature parameter
:param gamma: lower bound of "stretched" s
:param zeta: upper bound of "stretched" s
:param fix_temp: True if temperature is fixed
"""
super().__init__()
self._origin = origin
self._size = self._origin.weight.size()
self.loc = nn.Parameter(torch.zeros(self._size).normal_(loc_mean,
loc_sdev))
self.temp = beta if fix_temp else nn.Parameter(torch.zeros(1).fill_
(beta))
self.register_buffer('uniform', torch.zeros(self._size))
self.gamma = gamma
self.zeta = zeta
self.gamma_zeta_ratio = np.log(-gamma / zeta)
self.l02_alpha = l02_alpha
self.l02 = l02
if self.l02:
assert self.l02_alpha > 0
lg.info('Using L_0,2 norm')
def _get_mask(self):
if self.training:
self.uniform.uniform_()
u = Variable(self.uniform)
s = F.sigmoid((torch.log(u) - torch.log(1 - u) + self.loc) /
self.temp)
s = s * (self.zeta - self.gamma) + self.gamma
penalty = F.sigmoid(self.loc - self.temp * self.gamma_zeta_ratio
).sum()
if self.l02:
l02Norm = (F.sigmoid(self.loc - self.temp * self.
gamma_zeta_ratio) * self._origin.weight ** 2).sum()
penalty = penalty + self.l02_alpha * l02Norm
else:
s = F.sigmoid(self.loc) * (self.zeta - self.gamma) + self.gamma
penalty = 0
return hard_sigmoid(s), penalty
class Model(_L0Norm):
"""Linear model with L0 norm."""
def __init__(self, in_features: 'int', out_features: 'int', bias:
'bool'=True, **kwargs):
"""Linear model with L0 norm."""
super().__init__(nn.Linear(in_features, out_features,
bias=bias), **kwargs)
def forward(self, input):
"""Forward function with mask and penalty."""
mask, penalty = self._get_mask()
out = F.linear(input, self._origin.weight * mask, self._origin.bias)
return out, penalty
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
QNetwork | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/md/cmd3ewacyhu5w5hausgbjbmtnt5rr66cgczh4ibdypq7dz6p4v7g.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xc/cxcj5l7s6w5ttrq2fk2nirlbp44yesw6n2m2dnxtpcjjmym2njhr.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_1 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (128, 4), (4, 1))
assert_size_stride(primals_2, (128, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 128), (128, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (4, 64), (64, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 128), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 128), (2048, 512, 128, 1), 0); del buf0 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf6, 8192, grid=grid(8192), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 128), (128, 1), 0), reinterpret_tensor(primals_4, (128, 64), (1, 128), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf2 # reuse
buf5 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf5, 4096, grid=grid(4096), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(primals_6, (64, 4), (1, 64), 0), alpha=1, beta=1, out=buf4)
del primals_7
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 128), (128, 1), 0), reinterpret_tensor(buf3, (64, 64), (64, 1), 0), primals_6, buf5, primals_4, buf6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((128, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 128), (128, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn as nn
class QNetwork(nn.Module):
def __init__(self, state_size, action_size, seed):
super(QNetwork, self).__init__()
self.seed = torch.manual_seed(seed)
self.fc1 = nn.Linear(state_size, 128)
self.fc2 = nn.Linear(128, 64)
self.fc3 = nn.Linear(64, action_size)
def forward(self, state):
x = F.relu(self.fc1(state))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_size': 4, 'action_size': 4, 'seed': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (128, 4), (4, 1))
assert_size_stride(primals_2, (128,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 128), (128, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (4, 64), (64, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 128), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 128), (2048, 512, 128, 1), 0)
del buf0
buf6 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(8192)](buf1,
primals_2, buf6, 8192, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 128), (128, 1), 0),
reinterpret_tensor(primals_4, (128, 64), (1, 128), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf2
buf5 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.bool
)
triton_poi_fused_relu_threshold_backward_1[grid(4096)](buf3,
primals_5, buf5, 4096, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64),
(64, 1), 0), reinterpret_tensor(primals_6, (64, 4), (1, 64), 0),
alpha=1, beta=1, out=buf4)
del primals_7
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 128), (128, 1), 0
), reinterpret_tensor(buf3, (64, 64), (64, 1), 0
), primals_6, buf5, primals_4, buf6
class QNetworkNew(nn.Module):
def __init__(self, state_size, action_size, seed):
super(QNetworkNew, self).__init__()
self.seed = torch.manual_seed(seed)
self.fc1 = nn.Linear(state_size, 128)
self.fc2 = nn.Linear(128, 64)
self.fc3 = nn.Linear(64, action_size)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| royveshovda/deep-reinforcement-learning | QNetwork | false | 4,202 | [
"MIT"
] | 0 | 64ba7ef5ab44f095b7e8b29f6c4ff1585025981a | https://github.com/royveshovda/deep-reinforcement-learning/tree/64ba7ef5ab44f095b7e8b29f6c4ff1585025981a | import torch
import torch.nn.functional as F
import torch.nn as nn
class Model(nn.Module):
def __init__(self, state_size, action_size, seed):
super().__init__()
self.seed = torch.manual_seed(seed)
self.fc1 = nn.Linear(state_size, 128)
self.fc2 = nn.Linear(128, 64)
self.fc3 = nn.Linear(64, action_size)
def forward(self, state):
x = F.relu(self.fc1(state))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
Discriminator | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ms/cmsuzohbg5nq52jnvirovzkvykrzzko5xomu7zyu5e5u2lhegppw.py
# Topologically Sorted Source Nodes: [state_action], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# state_action => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_1, %primals_2], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + ((4*x1) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/dk/cdk3l2gdrurg6uj2ckltekowplr53hgrpb57mjds3uwpjtqhrw64.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x => tanh
# Graph fragment:
# %add_tensor_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_3, %primals_4), kwargs = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_tensor_3,), kwargs = {})
triton_poi_fused_tanh_1 = async_compile.triton('triton_poi_fused_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_tanh_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 2000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 500
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ls/clszevd4n3pthav5wqqc3tv5ntzabt6apuyy2ydb45lnc25ohanu.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x_1 => tanh_1
# Graph fragment:
# %add_tensor_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_2, %primals_6), kwargs = {})
# %tanh_1 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_tensor_2,), kwargs = {})
triton_poi_fused_tanh_2 = async_compile.triton('triton_poi_fused_tanh_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_tanh_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_tanh_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 300
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/rh/crhvyy3w3uejbzndu7qftnyc25sndrfzlmb3i2bzpyadobz7z7bm.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# x_3 => sigmoid
# Graph fragment:
# %add_tensor : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default, %primals_10), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_tensor,), kwargs = {})
triton_poi_fused_sigmoid_3 = async_compile.triton('triton_poi_fused_sigmoid_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_3', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_3(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (500, 8), (8, 1))
assert_size_stride(primals_4, (500, ), (1, ))
assert_size_stride(primals_5, (300, 500), (500, 1))
assert_size_stride(primals_6, (300, ), (1, ))
assert_size_stride(primals_7, (300, 300), (300, 1))
assert_size_stride(primals_8, (300, ), (1, ))
assert_size_stride(primals_9, (1, 300), (300, 1))
assert_size_stride(primals_10, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [state_action], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, primals_2, buf0, 32, grid=grid(32), stream=stream0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 500), (500, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf0, reinterpret_tensor(primals_3, (8, 500), (1, 8), 0), out=buf1)
del primals_3
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
triton_poi_fused_tanh_1.run(buf2, primals_4, 2000, grid=grid(2000), stream=stream0)
del primals_4
buf3 = empty_strided_cuda((4, 300), (300, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf2, reinterpret_tensor(primals_5, (500, 300), (1, 500), 0), out=buf3)
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
triton_poi_fused_tanh_2.run(buf4, primals_6, 1200, grid=grid(1200), stream=stream0)
del primals_6
buf5 = empty_strided_cuda((4, 300), (300, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf4, reinterpret_tensor(primals_7, (300, 300), (1, 300), 0), out=buf5)
buf6 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.tanh]
triton_poi_fused_tanh_2.run(buf6, primals_8, 1200, grid=grid(1200), stream=stream0)
del primals_8
buf7 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf6, reinterpret_tensor(primals_9, (300, 1), (1, 300), 0), out=buf7)
buf8 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_3.run(buf8, primals_10, 4, grid=grid(4), stream=stream0)
del primals_10
return (buf8, buf0, buf2, buf4, buf6, buf8, primals_9, primals_7, primals_5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((500, 8), (8, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((500, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((300, 500), (500, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((300, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((300, 300), (300, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((300, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((1, 300), (300, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class Discriminator(nn.Module):
def __init__(self, state_dim, action_dim):
super(Discriminator, self).__init__()
self.l1 = nn.Linear(state_dim + action_dim, 500)
self.l2 = nn.Linear(500, 300)
self.l3 = nn.Linear(300, 300)
self.l4 = nn.Linear(300, 1)
def forward(self, state, action):
state_action = torch.cat([state, action], 1)
x = torch.tanh(self.l1(state_action))
x = torch.tanh(self.l2(x))
x = torch.tanh(self.l3(x))
x = torch.sigmoid(self.l4(x))
return x
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4, 'action_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = xindex // 8
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (4 * x1 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 2000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 500
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_tanh_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 1200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 300
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_sigmoid_3(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (500, 8), (8, 1))
assert_size_stride(primals_4, (500,), (1,))
assert_size_stride(primals_5, (300, 500), (500, 1))
assert_size_stride(primals_6, (300,), (1,))
assert_size_stride(primals_7, (300, 300), (300, 1))
assert_size_stride(primals_8, (300,), (1,))
assert_size_stride(primals_9, (1, 300), (300, 1))
assert_size_stride(primals_10, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(32)](primals_1, primals_2, buf0, 32,
XBLOCK=32, num_warps=1, num_stages=1)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 500), (500, 1), torch.float32)
extern_kernels.mm(buf0, reinterpret_tensor(primals_3, (8, 500), (1,
8), 0), out=buf1)
del primals_3
buf2 = buf1
del buf1
triton_poi_fused_tanh_1[grid(2000)](buf2, primals_4, 2000, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((4, 300), (300, 1), torch.float32)
extern_kernels.mm(buf2, reinterpret_tensor(primals_5, (500, 300), (
1, 500), 0), out=buf3)
buf4 = buf3
del buf3
triton_poi_fused_tanh_2[grid(1200)](buf4, primals_6, 1200, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_6
buf5 = empty_strided_cuda((4, 300), (300, 1), torch.float32)
extern_kernels.mm(buf4, reinterpret_tensor(primals_7, (300, 300), (
1, 300), 0), out=buf5)
buf6 = buf5
del buf5
triton_poi_fused_tanh_2[grid(1200)](buf6, primals_8, 1200, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_8
buf7 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf6, reinterpret_tensor(primals_9, (300, 1), (1,
300), 0), out=buf7)
buf8 = buf7
del buf7
triton_poi_fused_sigmoid_3[grid(4)](buf8, primals_10, 4, XBLOCK=4,
num_warps=1, num_stages=1)
del primals_10
return buf8, buf0, buf2, buf4, buf6, buf8, primals_9, primals_7, primals_5
class DiscriminatorNew(nn.Module):
def __init__(self, state_dim, action_dim):
super(DiscriminatorNew, self).__init__()
self.l1 = nn.Linear(state_dim + action_dim, 500)
self.l2 = nn.Linear(500, 300)
self.l3 = nn.Linear(300, 300)
self.l4 = nn.Linear(300, 1)
def forward(self, input_0, input_1):
primals_3 = self.l1.weight
primals_4 = self.l1.bias
primals_5 = self.l2.weight
primals_6 = self.l2.bias
primals_7 = self.l3.weight
primals_8 = self.l3.bias
primals_9 = self.l4.weight
primals_10 = self.l4.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return output[0]
| rortiz9/meleeml | Discriminator | false | 4,203 | [
"MIT"
] | 0 | 9be4bf53a377dfb46dbb3b51f102f1bffc0124d2 | https://github.com/rortiz9/meleeml/tree/9be4bf53a377dfb46dbb3b51f102f1bffc0124d2 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, state_dim, action_dim):
super().__init__()
self.l1 = nn.Linear(state_dim + action_dim, 500)
self.l2 = nn.Linear(500, 300)
self.l3 = nn.Linear(300, 300)
self.l4 = nn.Linear(300, 1)
def forward(self, state, action):
state_action = torch.cat([state, action], 1)
x = torch.tanh(self.l1(state_action))
x = torch.tanh(self.l2(x))
x = torch.tanh(self.l3(x))
x = torch.sigmoid(self.l4(x))
return x
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [4, 4]
|
RelationalTransformerEncoderLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ih/cihzqunvhmsoa5e3orrwo2iez2setdyhxwrujxi7fwrn2aw7adgt.py
# Topologically Sorted Source Nodes: [q_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# q_1 => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%getitem, 1.0), kwargs = {})
triton_poi_fused_mul_0 = async_compile.triton('triton_poi_fused_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (12*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/iz/cizyqzghi7fnldgxtoz2clsnz4jvogk5fdvj7mi3tbe3tnb6g4kt.py
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_2 => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%getitem_2,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_1 = async_compile.triton('triton_poi_fused_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (8 + x0 + (12*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (8 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/54/c545jcidaxblcqouel6ztiwyytwzuaasohh4q2qelpa3u5sbntsf.py
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_1 => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%getitem_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_2 = async_compile.triton('triton_poi_fused_clone_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4 + x0 + (12*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (4 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/s4/cs472yivvl3yzse325afzknsz7ua5dqrqzmwls3lwujk3hte6xkl.py
# Topologically Sorted Source Nodes: [attn_output_weights_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_output_weights_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/zh/czh6tw7ngffcygnivwvcjex5edxy3ms4t27ymyn2hemxlpspxzq7.py
# Topologically Sorted Source Nodes: [attn_output_weights_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_output_weights_1 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_4 = async_compile.triton('triton_poi_fused__softmax_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/la/claxsme7yqsdzeonqq2iowvclxlentfbkqvzkxqtzqzk4g2snafx.py
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_3 => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_5,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_5 = async_compile.triton('triton_poi_fused_clone_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_5(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (16*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ns/cnspfsjjmvserkfymbru7x5vm2xumtyor5javdiv74jr3avx67rq.py
# Topologically Sorted Source Nodes: [src, src_1], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# src => add
# src_1 => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_5, %view_7), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/7u/c7uxwow3tztifyrr5oj6dotpbrh7qtup53xfydkt35y65ajtfwre.py
# Topologically Sorted Source Nodes: [src, src_1], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# src => add
# src_1 => add_1, add_2, mul_1, mul_2, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_5, %view_7), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_3, 1e-05), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_4), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, %primals_6), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_7), kwargs = {})
triton_poi_fused_add_native_layer_norm_7 = async_compile.triton('triton_poi_fused_add_native_layer_norm_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/rv/crvf2y3myjmfsbbg2ubhxgcshxxvsclo3cxqhn4x2kt2s5qfl37x.py
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# relu => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_10,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_8 = async_compile.triton('triton_poi_fused_relu_threshold_backward_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_8', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_8(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32768
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 2048
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/gi/cgiemrmhgokkex4ehtwipuydyw7qa56vlnfsfovgk46i4srp45yk.py
# Topologically Sorted Source Nodes: [src_2], Original ATen: [aten.add]
# Source node to ATen node mapping:
# src_2 => add_3
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %view_12), kwargs = {})
triton_poi_fused_add_9 = async_compile.triton('triton_poi_fused_add_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_9(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_out_ptr0 + (x2), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nn/cnnxhu2nhctulajy3rzg5wf2f3gd62kwlmtnwu5433qpihin4y26.py
# Topologically Sorted Source Nodes: [src_3], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# src_3 => add_4, rsqrt_1, var_mean_1
# Graph fragment:
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_3, [2]), kwargs = {correction: 0, keepdim: True})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_5, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_4,), kwargs = {})
triton_poi_fused_native_layer_norm_10 = async_compile.triton('triton_poi_fused_native_layer_norm_10', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_10', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_10(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-05
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + (x0), tmp8, xmask)
tl.store(out_ptr1 + (x0), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/73/c73z2hfes5izl473wn57vaku4rt2ae7swkdamlriywh5x5xt7g3z.py
# Topologically Sorted Source Nodes: [src_3], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# src_3 => add_4, add_5, mul_3, mul_4, rsqrt_1, sub_2, var_mean_1
# Graph fragment:
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_3, [2]), kwargs = {correction: 0, keepdim: True})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_5, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_4,), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_3, %getitem_6), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, %rsqrt_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %primals_12), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %primals_13), kwargs = {})
triton_poi_fused_native_layer_norm_11 = async_compile.triton('triton_poi_fused_native_layer_norm_11', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_11', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_11(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13 = args
args.clear()
assert_size_stride(primals_1, (12, 4), (4, 1))
assert_size_stride(primals_2, (12, ), (1, ))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (2048, 4), (4, 1))
assert_size_stride(primals_9, (2048, ), (1, ))
assert_size_stride(primals_10, (4, 2048), (2048, 1))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, ), (1, ))
assert_size_stride(primals_13, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 12), (12, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 12), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [q_1], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(buf0, primals_2, buf1, 64, grid=grid(64), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf0, primals_2, buf2, 64, grid=grid(64), stream=stream0)
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf0, primals_2, buf3, 64, grid=grid(64), stream=stream0)
del buf0
del primals_2
buf4 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_output_weights], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf1, (16, 4, 1), (1, 16, 0), 0), reinterpret_tensor(buf3, (16, 1, 4), (1, 0, 16), 0), out=buf4)
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_output_weights_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
buf6 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [attn_output_weights_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
buf7 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_output], Original ATen: [aten.bmm]
extern_kernels.bmm(buf6, reinterpret_tensor(buf2, (16, 4, 1), (1, 16, 0), 0), out=buf7)
buf8 = empty_strided_cuda((4, 16, 1), (16, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
triton_poi_fused_clone_5.run(buf7, buf8, 4, 16, grid=grid(4, 16), stream=stream0)
buf9 = reinterpret_tensor(buf7, (16, 4), (4, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [attn_output_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_4, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf9)
del primals_4
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [src, src_1], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(primals_5, buf9, buf10, buf11, 16, grid=grid(16), stream=stream0)
buf12 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [src, src_1], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_7.run(primals_5, buf9, buf10, buf11, primals_6, primals_7, buf12, 64, grid=grid(64), stream=stream0)
del primals_7
buf13 = empty_strided_cuda((16, 2048), (2048, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf12, (16, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 2048), (1, 4), 0), out=buf13)
buf14 = reinterpret_tensor(buf13, (4, 4, 2048), (8192, 2048, 1), 0); del buf13 # reuse
buf20 = empty_strided_cuda((4, 4, 2048), (8192, 2048, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_8.run(buf14, primals_9, buf20, 32768, grid=grid(32768), stream=stream0)
del primals_9
buf15 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf14, (16, 2048), (2048, 1), 0), reinterpret_tensor(primals_10, (2048, 4), (1, 2048), 0), out=buf15)
buf16 = reinterpret_tensor(buf15, (4, 4, 4), (16, 4, 1), 0); del buf15 # reuse
# Topologically Sorted Source Nodes: [src_2], Original ATen: [aten.add]
triton_poi_fused_add_9.run(buf16, buf12, primals_11, 64, grid=grid(64), stream=stream0)
del primals_11
buf17 = buf11; del buf11 # reuse
buf18 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [src_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_10.run(buf16, buf17, buf18, 16, grid=grid(16), stream=stream0)
buf19 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [src_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_11.run(buf16, buf17, buf18, primals_12, primals_13, buf19, 64, grid=grid(64), stream=stream0)
del buf17
del buf18
del primals_13
return (buf19, primals_5, primals_6, primals_12, buf6, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), buf9, reinterpret_tensor(buf12, (16, 4), (4, 1), 0), reinterpret_tensor(buf14, (16, 2048), (2048, 1), 0), buf16, primals_10, buf20, primals_8, primals_3, reinterpret_tensor(buf2, (16, 1, 4), (1, 1, 16), 0), reinterpret_tensor(buf1, (16, 1, 4), (1, 1, 16), 0), reinterpret_tensor(buf3, (16, 4, 1), (1, 16, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((12, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((2048, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((2048, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 2048), (2048, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import warnings
from torch import Tensor
from torch.nn import TransformerEncoderLayer
from torch.nn.functional import *
from torch.nn.modules.activation import MultiheadAttention
from torch.nn.modules.activation import xavier_uniform_
from torch.nn.modules.activation import xavier_normal_
from torch.nn.modules.activation import constant_
from torch.nn.modules.activation import Parameter
from typing import Optional
import torch.utils.data.dataset
from typing import Tuple
def relational_multi_head_attention_forward(query: 'Tensor', key: 'Tensor',
value: 'Tensor', relation: 'Tensor', embed_dim_to_check: 'int',
num_heads: 'int', in_proj_weight: 'Tensor', in_proj_bias: 'Tensor',
bias_k: 'Optional[Tensor]', bias_v: 'Optional[Tensor]', add_zero_attn:
'bool', dropout_p: 'float', out_proj_weight: 'Tensor', out_proj_bias:
'Tensor', training: 'bool'=True, key_padding_mask: 'Optional[Tensor]'=
None, need_weights: 'bool'=True, attn_mask: 'Optional[Tensor]'=None,
use_separate_proj_weight: 'bool'=False, q_proj_weight:
'Optional[Tensor]'=None, k_proj_weight: 'Optional[Tensor]'=None,
v_proj_weight: 'Optional[Tensor]'=None, static_k: 'Optional[Tensor]'=
None, static_v: 'Optional[Tensor]'=None, relation_type: 'str'=None
) ->Tuple[Tensor, Optional[Tensor]]:
"""
Args:
query, key, value: map a query and a set of key-value pairs to an output.
See "Attention Is All You Need" for more details.
relation: relation between queries and keys.
embed_dim_to_check: total dimension of the model.
num_heads: parallel attention heads.
in_proj_weight, in_proj_bias: input projection weight and bias.
bias_k, bias_v: bias of the key and value sequences to be added at dim=0.
add_zero_attn: add a new batch of zeros to the key and
value sequences at dim=1.
dropout_p: probability of an element to be zeroed.
out_proj_weight, out_proj_bias: the output projection weight and bias.
training: apply dropout if is ``True``.
key_padding_mask: if provided, specified padding elements in the key will
be ignored by the attention. This is an binary mask. When the value is True,
the corresponding value on the attention layer will be filled with -inf.
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
use_separate_proj_weight: the function accept the proj. weights for query, key,
and value in different forms. If false, in_proj_weight will be used, which is
a combination of q_proj_weight, k_proj_weight, v_proj_weight.
q_proj_weight, k_proj_weight, v_proj_weight, in_proj_bias: input projection weight and bias.
static_k, static_v: static key and value used for attention operators.
Shape:
Inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- relation: :math:`(L, S, N, E)` where L is the target sequence length, where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the zero positions
will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the value of ``False`` will be unchanged.
- attn_mask: 2D mask :math:`(L, S)` where L is the target sequence length, S is the source sequence length.
3D mask :math:`(N*num_heads, L, S)` where N is the batch size, L is the target sequence length,
S is the source sequence length. attn_mask ensures that position i is allowed to attend the unmasked
positions. If a ByteTensor is provided, the non-zero positions are not allowed to attend
while the zero positions will be unchanged. If a BoolTensor is provided, positions with ``True``
are not allowed to attend while ``False`` values will be unchanged. If a FloatTensor
is provided, it will be added to the attention weight.
- static_k: :math:`(N*num_heads, S, E/num_heads)`, where S is the source sequence length,
N is the batch size, E is the embedding dimension. E/num_heads is the head dimension.
- static_v: :math:`(N*num_heads, S, E/num_heads)`, where S is the source sequence length,
N is the batch size, E is the embedding dimension. E/num_heads is the head dimension.
Outputs:
- attn_output: :math:`(L, N, E)` where L is the target sequence length, N is the batch size,
E is the embedding dimension.
- attn_output_weights: :math:`(N, L, S)` where N is the batch size,
L is the target sequence length, S is the source sequence length.
"""
tens_ops = (query, key, value, in_proj_weight, in_proj_bias, bias_k,
bias_v, out_proj_weight, out_proj_bias)
if has_torch_function(tens_ops):
return handle_torch_function(multi_head_attention_forward, tens_ops,
query, key, value, embed_dim_to_check, num_heads,
in_proj_weight, in_proj_bias, bias_k, bias_v, add_zero_attn,
dropout_p, out_proj_weight, out_proj_bias, training=training,
key_padding_mask=key_padding_mask, need_weights=need_weights,
attn_mask=attn_mask, use_separate_proj_weight=
use_separate_proj_weight, q_proj_weight=q_proj_weight,
k_proj_weight=k_proj_weight, v_proj_weight=v_proj_weight,
static_k=static_k, static_v=static_v)
tgt_len, bsz, embed_dim = query.size()
assert embed_dim == embed_dim_to_check
assert key.size(0) == value.size(0) and key.size(1) == value.size(1)
head_dim = embed_dim // num_heads
assert head_dim * num_heads == embed_dim, 'embed_dim must be divisible by num_heads'
scaling = float(head_dim) ** -0.5
if not use_separate_proj_weight:
if (query is key or torch.equal(query, key)) and (key is value or
torch.equal(key, value)):
q, k, v = linear(query, in_proj_weight, in_proj_bias).chunk(3,
dim=-1)
elif key is value or torch.equal(key, value):
_b = in_proj_bias
_start = 0
_end = embed_dim
_w = in_proj_weight[_start:_end, :]
if _b is not None:
_b = _b[_start:_end]
q = linear(query, _w, _b)
if key is None:
assert value is None
k = None
v = None
else:
_b = in_proj_bias
_start = embed_dim
_end = None
_w = in_proj_weight[_start:, :]
if _b is not None:
_b = _b[_start:]
k, v = linear(key, _w, _b).chunk(2, dim=-1)
else:
_b = in_proj_bias
_start = 0
_end = embed_dim
_w = in_proj_weight[_start:_end, :]
if _b is not None:
_b = _b[_start:_end]
q = linear(query, _w, _b)
_b = in_proj_bias
_start = embed_dim
_end = embed_dim * 2
_w = in_proj_weight[_start:_end, :]
if _b is not None:
_b = _b[_start:_end]
k = linear(key, _w, _b)
_b = in_proj_bias
_start = embed_dim * 2
_end = None
_w = in_proj_weight[_start:, :]
if _b is not None:
_b = _b[_start:]
v = linear(value, _w, _b)
else:
q_proj_weight_non_opt = torch.jit._unwrap_optional(q_proj_weight)
len1, len2 = q_proj_weight_non_opt.size()
assert len1 == embed_dim and len2 == query.size(-1)
k_proj_weight_non_opt = torch.jit._unwrap_optional(k_proj_weight)
len1, len2 = k_proj_weight_non_opt.size()
assert len1 == embed_dim and len2 == key.size(-1)
v_proj_weight_non_opt = torch.jit._unwrap_optional(v_proj_weight)
len1, len2 = v_proj_weight_non_opt.size()
assert len1 == embed_dim and len2 == value.size(-1)
if in_proj_bias is not None:
q = linear(query, q_proj_weight_non_opt, in_proj_bias[0:embed_dim])
k = linear(key, k_proj_weight_non_opt, in_proj_bias[embed_dim:
embed_dim * 2])
v = linear(value, v_proj_weight_non_opt, in_proj_bias[embed_dim *
2:])
else:
q = linear(query, q_proj_weight_non_opt, in_proj_bias)
k = linear(key, k_proj_weight_non_opt, in_proj_bias)
v = linear(value, v_proj_weight_non_opt, in_proj_bias)
q = q * scaling
if attn_mask is not None:
assert attn_mask.dtype == torch.float32 or attn_mask.dtype == torch.float64 or attn_mask.dtype == torch.float16 or attn_mask.dtype == torch.uint8 or attn_mask.dtype == torch.bool, 'Only float, byte, and bool types are supported for attn_mask, not {}'.format(
attn_mask.dtype)
if attn_mask.dtype == torch.uint8:
warnings.warn(
'Byte tensor for attn_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead.'
)
attn_mask = attn_mask
if attn_mask.dim() == 2:
attn_mask = attn_mask.unsqueeze(0)
if list(attn_mask.size()) != [1, query.size(0), key.size(0)]:
raise RuntimeError(
'The size of the 2D attn_mask is not correct.')
elif attn_mask.dim() == 3:
if list(attn_mask.size()) != [bsz * num_heads, query.size(0),
key.size(0)]:
raise RuntimeError(
'The size of the 3D attn_mask is not correct.')
else:
raise RuntimeError("attn_mask's dimension {} is not supported".
format(attn_mask.dim()))
if key_padding_mask is not None and key_padding_mask.dtype == torch.uint8:
warnings.warn(
'Byte tensor for key_padding_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead.'
)
key_padding_mask = key_padding_mask
if bias_k is not None and bias_v is not None:
if static_k is None and static_v is None:
k = torch.cat([k, bias_k.repeat(1, bsz, 1)])
v = torch.cat([v, bias_v.repeat(1, bsz, 1)])
if attn_mask is not None:
attn_mask = pad(attn_mask, (0, 1))
if key_padding_mask is not None:
key_padding_mask = pad(key_padding_mask, (0, 1))
else:
assert static_k is None, 'bias cannot be added to static key.'
assert static_v is None, 'bias cannot be added to static value.'
else:
assert bias_k is None
assert bias_v is None
r = None
q = q.contiguous().view(tgt_len, bsz * num_heads, head_dim).transpose(0, 1)
if k is not None:
k = k.contiguous().view(-1, bsz * num_heads, head_dim).transpose(0, 1)
if v is not None:
v = v.contiguous().view(-1, bsz * num_heads, head_dim).transpose(0, 1)
if static_k is not None:
assert static_k.size(0) == bsz * num_heads
assert static_k.size(2) == head_dim
k = static_k
if static_v is not None:
assert static_v.size(0) == bsz * num_heads
assert static_v.size(2) == head_dim
v = static_v
src_len = k.size(1)
if relation is not None:
if relation_type == 'qk+r':
r = relation.contiguous().view(tgt_len, -1, bsz * num_heads, 1
).squeeze(3).permute(2, 0, 1)
elif relation_type == 'q(k+r)':
r = relation.contiguous().view(tgt_len, src_len, bsz *
num_heads, head_dim).permute(2, 0, 1)
r = r.view(bsz * num_heads, tgt_len * src_len, head_dim)
r = torch.bmm(r, q.unsqueeze(2).repeat(1, 1, src_len, 1).view(
bsz * num_heads, tgt_len * src_len, head_dim).transpose(1, 2)
).view(bsz * num_heads, tgt_len, src_len)
if key_padding_mask is not None:
assert key_padding_mask.size(0) == bsz
assert key_padding_mask.size(1) == src_len
if add_zero_attn:
src_len += 1
k = torch.cat([k, torch.zeros((k.size(0), 1) + k.size()[2:], dtype=
k.dtype, device=k.device)], dim=1)
v = torch.cat([v, torch.zeros((v.size(0), 1) + v.size()[2:], dtype=
v.dtype, device=v.device)], dim=1)
if attn_mask is not None:
attn_mask = pad(attn_mask, (0, 1))
if key_padding_mask is not None:
key_padding_mask = pad(key_padding_mask, (0, 1))
attn_output_weights = torch.bmm(q, k.transpose(1, 2))
if r is not None:
attn_output_weights = attn_output_weights + r
assert list(attn_output_weights.size()) == [bsz * num_heads, tgt_len,
src_len]
if attn_mask is not None:
if attn_mask.dtype == torch.bool:
attn_output_weights.masked_fill_(attn_mask, float('-inf'))
else:
attn_output_weights += attn_mask
if key_padding_mask is not None:
attn_output_weights = attn_output_weights.view(bsz, num_heads,
tgt_len, src_len)
attn_output_weights = attn_output_weights.masked_fill(key_padding_mask
.unsqueeze(1).unsqueeze(2), float('-inf'))
attn_output_weights = attn_output_weights.view(bsz * num_heads,
tgt_len, src_len)
attn_output_weights = softmax(attn_output_weights, dim=-1)
attn_output_weights = dropout(attn_output_weights, p=dropout_p,
training=training)
attn_output = torch.bmm(attn_output_weights, v)
assert list(attn_output.size()) == [bsz * num_heads, tgt_len, head_dim]
attn_output = attn_output.transpose(0, 1).contiguous().view(tgt_len,
bsz, embed_dim)
attn_output = linear(attn_output, out_proj_weight, out_proj_bias)
if need_weights:
attn_output_weights = attn_output_weights.view(bsz, num_heads,
tgt_len, src_len)
return attn_output, attn_output_weights.sum(dim=1) / num_heads
else:
return attn_output, None
class RelationalMultiheadAttention(MultiheadAttention):
"""Allows the model to jointly attend to information
from different representation subspaces.
See `Attention Is All You Need <https://arxiv.org/abs/1706.03762>`_
.. math::
\\text{MultiHead}(Q, K, V) = \\text{Concat}(head_1,\\dots,head_h)W^O
where :math:`head_i = \\text{Attention}(QW_i^Q, KW_i^K, VW_i^V)`.
Args:
embed_dim: total dimension of the model.
num_heads: parallel attention heads.
dropout: a Dropout layer on attn_output_weights. Default: 0.0.
bias: add bias as module parameter. Default: True.
add_bias_kv: add bias to the key and value sequences at dim=0.
add_zero_attn: add a new batch of zeros to the key and
value sequences at dim=1.
kdim: total number of features in key. Default: None.
vdim: total number of features in value. Default: None.
Note that if :attr:`kdim` and :attr:`vdim` are None, they will be set
to :attr:`embed_dim` such that query, key, and value have the same
number of features.
Examples::
>>> realational_multihead_attn = RelationalMultiheadAttention(embed_dim, num_heads)
>>> attn_output, attn_output_weights = realational_multihead_attn(query, key, value)
"""
bias_k: 'Optional[torch.Tensor]'
bias_v: 'Optional[torch.Tensor]'
def __init__(self, embed_dim, num_heads, dropout=0.0, bias=True,
add_bias_kv=False, add_zero_attn=False, kdim=None, vdim=None,
add_relation=False, rdim=None, relation_type=None):
super(RelationalMultiheadAttention, self).__init__(embed_dim=
embed_dim, num_heads=num_heads, dropout=dropout, bias=bias,
add_bias_kv=add_bias_kv, add_zero_attn=add_zero_attn, kdim=kdim,
vdim=vdim)
self.add_relation = add_relation
self.rdim = rdim if rdim is not None else embed_dim
self.relation_type = relation_type if relation_type else 'qk+r'
if self.add_relation:
if relation_type == 'qk+r':
self.r_proj_weight = Parameter(torch.Tensor(num_heads, self
.rdim))
self.r_proj_bias = Parameter(torch.empty(num_heads))
elif relation_type == 'q(k+r)':
self.r_proj_weight = Parameter(torch.Tensor(embed_dim, self
.rdim))
self.r_proj_bias = Parameter(torch.empty(embed_dim))
self._reset_parameters()
def _reset_parameters(self):
if self._qkv_same_embed_dim:
xavier_uniform_(self.in_proj_weight)
else:
xavier_uniform_(self.q_proj_weight)
xavier_uniform_(self.k_proj_weight)
xavier_uniform_(self.v_proj_weight)
if hasattr(self, 'add_relation') and self.add_relation:
xavier_uniform_(self.r_proj_weight)
constant_(self.r_proj_bias, 0.0)
if self.in_proj_bias is not None:
constant_(self.in_proj_bias, 0.0)
constant_(self.out_proj.bias, 0.0)
if self.bias_k is not None:
xavier_normal_(self.bias_k)
if self.bias_v is not None:
xavier_normal_(self.bias_v)
def __setstate__(self, state):
if '_qkv_same_embed_dim' not in state:
state['_qkv_same_embed_dim'] = True
super(MultiheadAttention, self).__setstate__(state)
def forward(self, query: 'Tensor', key: 'Tensor', value: 'Tensor',
relation_dict=None, key_padding_mask: 'Optional[Tensor]'=None,
need_weights: 'bool'=True, attn_mask: 'Optional[Tensor]'=None) ->Tuple[
Tensor, Optional[Tensor]]:
"""
Args:
query, key, value: map a query and a set of key-value pairs to an output.
See "Attention Is All You Need" for more details.
key_padding_mask: if provided, specified padding elements in the key will
be ignored by the attention. When given a binary mask and a value is True,
the corresponding value on the attention layer will be ignored. When given
a byte mask and a value is non-zero, the corresponding value on the attention
layer will be ignored
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
Shapes for inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the position
with the zero positions will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the value of ``False`` will be unchanged.
- attn_mask: if a 2D mask: :math:`(L, S)` where L is the target sequence length, S is the
source sequence length.
If a 3D mask: :math:`(N\\cdot\\text{num\\_heads}, L, S)` where N is the batch size, L is the target sequence
length, S is the source sequence length. ``attn_mask`` ensure that position i is allowed to attend
the unmasked positions. If a ByteTensor is provided, the non-zero positions are not allowed to attend
while the zero positions will be unchanged. If a BoolTensor is provided, positions with ``True``
is not allowed to attend while ``False`` values will be unchanged. If a FloatTensor
is provided, it will be added to the attention weight.
Shapes for outputs:
- attn_output: :math:`(L, N, E)` where L is the target sequence length, N is the batch size,
E is the embedding dimension.
- attn_output_weights: :math:`(N, L, S)` where N is the batch size,
L is the target sequence length, S is the source sequence length.
"""
if relation_dict is not None:
relation_labels = relation_dict['relation_labels']
relation_ids = relation_dict['relation_ids']
batch_index = relation_dict['batch_index']
pad_embedding = relation_dict['pad_embedding']
relation_labels = linear(relation_labels, self.r_proj_weight,
self.r_proj_bias)
pad_embedding = linear(pad_embedding.unsqueeze(0), self.
r_proj_weight, self.r_proj_bias).squeeze()
tgt_length, bsz, _ = query.size()
src_length, _, _ = key.size()
relation = pad_embedding.view(1, 1, 1, -1).repeat(bsz,
tgt_length, src_length, 1)
relation[batch_index, relation_ids[:, :, 0], relation_ids[:, :, 1]
] = relation_labels
relation = relation.permute(1, 2, 0, 3)
if not self._qkv_same_embed_dim:
return relational_multi_head_attention_forward(query, key,
value, relation, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.
bias_v, self.add_zero_attn, self.dropout, self.out_proj
.weight, self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask,
use_separate_proj_weight=True, q_proj_weight=self.
q_proj_weight, k_proj_weight=self.k_proj_weight,
v_proj_weight=self.v_proj_weight)
else:
return relational_multi_head_attention_forward(query, key,
value, relation, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.
bias_v, self.add_zero_attn, self.dropout, self.out_proj
.weight, self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask)
elif not self._qkv_same_embed_dim:
return relational_multi_head_attention_forward(query, key,
value, None, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.bias_v,
self.add_zero_attn, self.dropout, self.out_proj.weight,
self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask, use_separate_proj_weight
=True, q_proj_weight=self.q_proj_weight, k_proj_weight=self
.k_proj_weight, v_proj_weight=self.v_proj_weight)
else:
return relational_multi_head_attention_forward(query, key,
value, None, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.bias_v,
self.add_zero_attn, self.dropout, self.out_proj.weight,
self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask)
class RelationalTransformerEncoderLayer(TransformerEncoderLayer):
"""TransformerEncoderLayer is made up of self-attn and feedforward network.
This standard encoder layer is based on the paper "Attention Is All You Need".
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
in a different way during application.
Args:
d_model: the number of expected features in the input (required).
nhead: the number of heads in the multiheadattention models (required).
dim_feedforward: the dimension of the feedforward network model (default=2048).
dropout: the dropout value (default=0.1).
activation: the activation function of intermediate layer, relu or gelu (default=relu).
Examples::
>>> encoder_layer = RelationalTransformerEncoderLayer(d_model=512, nhead=8)
>>> src = torch.rand(10, 32, 512)
>>> rel = torch.rand(10, 10, 32, 512)
>>> out = encoder_layer(src, rel)
"""
def __init__(self, d_model, nhead, add_relation=False, dim_feedforward=
2048, dropout=0.1, activation='relu', relation_type=None):
super(RelationalTransformerEncoderLayer, self).__init__(d_model,
nhead, dim_feedforward=dim_feedforward, dropout=dropout,
activation=activation)
self.self_attn = RelationalMultiheadAttention(d_model, nhead,
add_relation=add_relation, dropout=dropout, relation_type=
relation_type)
def forward(self, src: 'Tensor', relation=None, src_mask:
'Optional[Tensor]'=None, src_key_padding_mask: 'Optional[Tensor]'=None
) ->Tensor:
"""Pass the input through the encoder layer.
Args:
src: the sequence to the encoder layer (required).
src_mask: the mask for the src sequence (optional).
src_key_padding_mask: the mask for the src keys per batch (optional).
Shape:
see the docs in Transformer class.
"""
src2 = self.self_attn(src, src, src, relation_dict=relation,
attn_mask=src_mask, key_padding_mask=src_key_padding_mask)[0]
src = src + self.dropout1(src2)
src = self.norm1(src)
src2 = self.linear2(self.dropout(self.activation(self.linear1(src))))
src = src + self.dropout2(src2)
src = self.norm2(src)
return src
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'d_model': 4, 'nhead': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import warnings
from torch import Tensor
from torch.nn import TransformerEncoderLayer
from torch.nn.functional import *
from torch.nn.modules.activation import MultiheadAttention
from torch.nn.modules.activation import xavier_uniform_
from torch.nn.modules.activation import xavier_normal_
from torch.nn.modules.activation import constant_
from torch.nn.modules.activation import Parameter
from typing import Optional
import torch.utils.data.dataset
from typing import Tuple
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 12 * x1), xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (8 + x0 + 12 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (8 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_poi_fused_clone_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4 + x0 + 12 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (4 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_5(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x1), xmask & ymask, eviction_policy=
'evict_last')
tl.store(out_ptr0 + (x1 + 16 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_8(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 2048
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused_add_9(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_out_ptr0 + x2, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_10(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-05
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp23, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_11(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13) = args
args.clear()
assert_size_stride(primals_1, (12, 4), (4, 1))
assert_size_stride(primals_2, (12,), (1,))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (2048, 4), (4, 1))
assert_size_stride(primals_9, (2048,), (1,))
assert_size_stride(primals_10, (4, 2048), (2048, 1))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4,), (1,))
assert_size_stride(primals_13, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 12), (12, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_5, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 12), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(64)](buf0, primals_2, buf1, 64, XBLOCK=
64, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clone_1[grid(64)](buf0, primals_2, buf2, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(64)](buf0, primals_2, buf3, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del buf0
del primals_2
buf4 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf1, (16, 4, 1), (1, 16, 0),
0), reinterpret_tensor(buf3, (16, 1, 4), (1, 0, 16), 0), out=buf4)
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_3[grid(256)](buf4, buf5, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf6 = buf4
del buf4
triton_poi_fused__softmax_4[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
buf7 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(buf6, reinterpret_tensor(buf2, (16, 4, 1), (1,
16, 0), 0), out=buf7)
buf8 = empty_strided_cuda((4, 16, 1), (16, 1, 1), torch.float32)
triton_poi_fused_clone_5[grid(4, 16)](buf7, buf8, 4, 16, XBLOCK=16,
YBLOCK=4, num_warps=1, num_stages=1)
buf9 = reinterpret_tensor(buf7, (16, 4), (4, 1), 0)
del buf7
extern_kernels.addmm(primals_4, reinterpret_tensor(buf8, (16, 4), (
4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf9)
del primals_4
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(16)](primals_5, buf9,
buf10, buf11, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf12 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_7[grid(64)](primals_5, buf9,
buf10, buf11, primals_6, primals_7, buf12, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_7
buf13 = empty_strided_cuda((16, 2048), (2048, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf12, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 2048), (1, 4), 0), out=buf13)
buf14 = reinterpret_tensor(buf13, (4, 4, 2048), (8192, 2048, 1), 0)
del buf13
buf20 = empty_strided_cuda((4, 4, 2048), (8192, 2048, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_8[grid(32768)](buf14,
primals_9, buf20, 32768, XBLOCK=128, num_warps=4, num_stages=1)
del primals_9
buf15 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf14, (16, 2048), (2048, 1),
0), reinterpret_tensor(primals_10, (2048, 4), (1, 2048), 0),
out=buf15)
buf16 = reinterpret_tensor(buf15, (4, 4, 4), (16, 4, 1), 0)
del buf15
triton_poi_fused_add_9[grid(64)](buf16, buf12, primals_11, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_11
buf17 = buf11
del buf11
buf18 = buf10
del buf10
triton_poi_fused_native_layer_norm_10[grid(16)](buf16, buf17, buf18,
16, XBLOCK=16, num_warps=1, num_stages=1)
buf19 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_11[grid(64)](buf16, buf17, buf18,
primals_12, primals_13, buf19, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf17
del buf18
del primals_13
return buf19, primals_5, primals_6, primals_12, buf6, reinterpret_tensor(
buf8, (16, 4), (4, 1), 0), buf9, reinterpret_tensor(buf12, (16, 4),
(4, 1), 0), reinterpret_tensor(buf14, (16, 2048), (2048, 1), 0
), buf16, primals_10, buf20, primals_8, primals_3, reinterpret_tensor(
buf2, (16, 1, 4), (1, 1, 16), 0), reinterpret_tensor(buf1, (16, 1,
4), (1, 1, 16), 0), reinterpret_tensor(buf3, (16, 4, 1), (1, 16, 1), 0)
def relational_multi_head_attention_forward(query: 'Tensor', key: 'Tensor',
value: 'Tensor', relation: 'Tensor', embed_dim_to_check: 'int',
num_heads: 'int', in_proj_weight: 'Tensor', in_proj_bias: 'Tensor',
bias_k: 'Optional[Tensor]', bias_v: 'Optional[Tensor]', add_zero_attn:
'bool', dropout_p: 'float', out_proj_weight: 'Tensor', out_proj_bias:
'Tensor', training: 'bool'=True, key_padding_mask: 'Optional[Tensor]'=
None, need_weights: 'bool'=True, attn_mask: 'Optional[Tensor]'=None,
use_separate_proj_weight: 'bool'=False, q_proj_weight:
'Optional[Tensor]'=None, k_proj_weight: 'Optional[Tensor]'=None,
v_proj_weight: 'Optional[Tensor]'=None, static_k: 'Optional[Tensor]'=
None, static_v: 'Optional[Tensor]'=None, relation_type: 'str'=None
) ->Tuple[Tensor, Optional[Tensor]]:
"""
Args:
query, key, value: map a query and a set of key-value pairs to an output.
See "Attention Is All You Need" for more details.
relation: relation between queries and keys.
embed_dim_to_check: total dimension of the model.
num_heads: parallel attention heads.
in_proj_weight, in_proj_bias: input projection weight and bias.
bias_k, bias_v: bias of the key and value sequences to be added at dim=0.
add_zero_attn: add a new batch of zeros to the key and
value sequences at dim=1.
dropout_p: probability of an element to be zeroed.
out_proj_weight, out_proj_bias: the output projection weight and bias.
training: apply dropout if is ``True``.
key_padding_mask: if provided, specified padding elements in the key will
be ignored by the attention. This is an binary mask. When the value is True,
the corresponding value on the attention layer will be filled with -inf.
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
use_separate_proj_weight: the function accept the proj. weights for query, key,
and value in different forms. If false, in_proj_weight will be used, which is
a combination of q_proj_weight, k_proj_weight, v_proj_weight.
q_proj_weight, k_proj_weight, v_proj_weight, in_proj_bias: input projection weight and bias.
static_k, static_v: static key and value used for attention operators.
Shape:
Inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- relation: :math:`(L, S, N, E)` where L is the target sequence length, where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the zero positions
will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the value of ``False`` will be unchanged.
- attn_mask: 2D mask :math:`(L, S)` where L is the target sequence length, S is the source sequence length.
3D mask :math:`(N*num_heads, L, S)` where N is the batch size, L is the target sequence length,
S is the source sequence length. attn_mask ensures that position i is allowed to attend the unmasked
positions. If a ByteTensor is provided, the non-zero positions are not allowed to attend
while the zero positions will be unchanged. If a BoolTensor is provided, positions with ``True``
are not allowed to attend while ``False`` values will be unchanged. If a FloatTensor
is provided, it will be added to the attention weight.
- static_k: :math:`(N*num_heads, S, E/num_heads)`, where S is the source sequence length,
N is the batch size, E is the embedding dimension. E/num_heads is the head dimension.
- static_v: :math:`(N*num_heads, S, E/num_heads)`, where S is the source sequence length,
N is the batch size, E is the embedding dimension. E/num_heads is the head dimension.
Outputs:
- attn_output: :math:`(L, N, E)` where L is the target sequence length, N is the batch size,
E is the embedding dimension.
- attn_output_weights: :math:`(N, L, S)` where N is the batch size,
L is the target sequence length, S is the source sequence length.
"""
tens_ops = (query, key, value, in_proj_weight, in_proj_bias, bias_k,
bias_v, out_proj_weight, out_proj_bias)
if has_torch_function(tens_ops):
return handle_torch_function(multi_head_attention_forward, tens_ops,
query, key, value, embed_dim_to_check, num_heads,
in_proj_weight, in_proj_bias, bias_k, bias_v, add_zero_attn,
dropout_p, out_proj_weight, out_proj_bias, training=training,
key_padding_mask=key_padding_mask, need_weights=need_weights,
attn_mask=attn_mask, use_separate_proj_weight=
use_separate_proj_weight, q_proj_weight=q_proj_weight,
k_proj_weight=k_proj_weight, v_proj_weight=v_proj_weight,
static_k=static_k, static_v=static_v)
tgt_len, bsz, embed_dim = query.size()
assert embed_dim == embed_dim_to_check
assert key.size(0) == value.size(0) and key.size(1) == value.size(1)
head_dim = embed_dim // num_heads
assert head_dim * num_heads == embed_dim, 'embed_dim must be divisible by num_heads'
scaling = float(head_dim) ** -0.5
if not use_separate_proj_weight:
if (query is key or torch.equal(query, key)) and (key is value or
torch.equal(key, value)):
q, k, v = linear(query, in_proj_weight, in_proj_bias).chunk(3,
dim=-1)
elif key is value or torch.equal(key, value):
_b = in_proj_bias
_start = 0
_end = embed_dim
_w = in_proj_weight[_start:_end, :]
if _b is not None:
_b = _b[_start:_end]
q = linear(query, _w, _b)
if key is None:
assert value is None
k = None
v = None
else:
_b = in_proj_bias
_start = embed_dim
_end = None
_w = in_proj_weight[_start:, :]
if _b is not None:
_b = _b[_start:]
k, v = linear(key, _w, _b).chunk(2, dim=-1)
else:
_b = in_proj_bias
_start = 0
_end = embed_dim
_w = in_proj_weight[_start:_end, :]
if _b is not None:
_b = _b[_start:_end]
q = linear(query, _w, _b)
_b = in_proj_bias
_start = embed_dim
_end = embed_dim * 2
_w = in_proj_weight[_start:_end, :]
if _b is not None:
_b = _b[_start:_end]
k = linear(key, _w, _b)
_b = in_proj_bias
_start = embed_dim * 2
_end = None
_w = in_proj_weight[_start:, :]
if _b is not None:
_b = _b[_start:]
v = linear(value, _w, _b)
else:
q_proj_weight_non_opt = torch.jit._unwrap_optional(q_proj_weight)
len1, len2 = q_proj_weight_non_opt.size()
assert len1 == embed_dim and len2 == query.size(-1)
k_proj_weight_non_opt = torch.jit._unwrap_optional(k_proj_weight)
len1, len2 = k_proj_weight_non_opt.size()
assert len1 == embed_dim and len2 == key.size(-1)
v_proj_weight_non_opt = torch.jit._unwrap_optional(v_proj_weight)
len1, len2 = v_proj_weight_non_opt.size()
assert len1 == embed_dim and len2 == value.size(-1)
if in_proj_bias is not None:
q = linear(query, q_proj_weight_non_opt, in_proj_bias[0:embed_dim])
k = linear(key, k_proj_weight_non_opt, in_proj_bias[embed_dim:
embed_dim * 2])
v = linear(value, v_proj_weight_non_opt, in_proj_bias[embed_dim *
2:])
else:
q = linear(query, q_proj_weight_non_opt, in_proj_bias)
k = linear(key, k_proj_weight_non_opt, in_proj_bias)
v = linear(value, v_proj_weight_non_opt, in_proj_bias)
q = q * scaling
if attn_mask is not None:
assert attn_mask.dtype == torch.float32 or attn_mask.dtype == torch.float64 or attn_mask.dtype == torch.float16 or attn_mask.dtype == torch.uint8 or attn_mask.dtype == torch.bool, 'Only float, byte, and bool types are supported for attn_mask, not {}'.format(
attn_mask.dtype)
if attn_mask.dtype == torch.uint8:
warnings.warn(
'Byte tensor for attn_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead.'
)
attn_mask = attn_mask
if attn_mask.dim() == 2:
attn_mask = attn_mask.unsqueeze(0)
if list(attn_mask.size()) != [1, query.size(0), key.size(0)]:
raise RuntimeError(
'The size of the 2D attn_mask is not correct.')
elif attn_mask.dim() == 3:
if list(attn_mask.size()) != [bsz * num_heads, query.size(0),
key.size(0)]:
raise RuntimeError(
'The size of the 3D attn_mask is not correct.')
else:
raise RuntimeError("attn_mask's dimension {} is not supported".
format(attn_mask.dim()))
if key_padding_mask is not None and key_padding_mask.dtype == torch.uint8:
warnings.warn(
'Byte tensor for key_padding_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead.'
)
key_padding_mask = key_padding_mask
if bias_k is not None and bias_v is not None:
if static_k is None and static_v is None:
k = torch.cat([k, bias_k.repeat(1, bsz, 1)])
v = torch.cat([v, bias_v.repeat(1, bsz, 1)])
if attn_mask is not None:
attn_mask = pad(attn_mask, (0, 1))
if key_padding_mask is not None:
key_padding_mask = pad(key_padding_mask, (0, 1))
else:
assert static_k is None, 'bias cannot be added to static key.'
assert static_v is None, 'bias cannot be added to static value.'
else:
assert bias_k is None
assert bias_v is None
r = None
q = q.contiguous().view(tgt_len, bsz * num_heads, head_dim).transpose(0, 1)
if k is not None:
k = k.contiguous().view(-1, bsz * num_heads, head_dim).transpose(0, 1)
if v is not None:
v = v.contiguous().view(-1, bsz * num_heads, head_dim).transpose(0, 1)
if static_k is not None:
assert static_k.size(0) == bsz * num_heads
assert static_k.size(2) == head_dim
k = static_k
if static_v is not None:
assert static_v.size(0) == bsz * num_heads
assert static_v.size(2) == head_dim
v = static_v
src_len = k.size(1)
if relation is not None:
if relation_type == 'qk+r':
r = relation.contiguous().view(tgt_len, -1, bsz * num_heads, 1
).squeeze(3).permute(2, 0, 1)
elif relation_type == 'q(k+r)':
r = relation.contiguous().view(tgt_len, src_len, bsz *
num_heads, head_dim).permute(2, 0, 1)
r = r.view(bsz * num_heads, tgt_len * src_len, head_dim)
r = torch.bmm(r, q.unsqueeze(2).repeat(1, 1, src_len, 1).view(
bsz * num_heads, tgt_len * src_len, head_dim).transpose(1, 2)
).view(bsz * num_heads, tgt_len, src_len)
if key_padding_mask is not None:
assert key_padding_mask.size(0) == bsz
assert key_padding_mask.size(1) == src_len
if add_zero_attn:
src_len += 1
k = torch.cat([k, torch.zeros((k.size(0), 1) + k.size()[2:], dtype=
k.dtype, device=k.device)], dim=1)
v = torch.cat([v, torch.zeros((v.size(0), 1) + v.size()[2:], dtype=
v.dtype, device=v.device)], dim=1)
if attn_mask is not None:
attn_mask = pad(attn_mask, (0, 1))
if key_padding_mask is not None:
key_padding_mask = pad(key_padding_mask, (0, 1))
attn_output_weights = torch.bmm(q, k.transpose(1, 2))
if r is not None:
attn_output_weights = attn_output_weights + r
assert list(attn_output_weights.size()) == [bsz * num_heads, tgt_len,
src_len]
if attn_mask is not None:
if attn_mask.dtype == torch.bool:
attn_output_weights.masked_fill_(attn_mask, float('-inf'))
else:
attn_output_weights += attn_mask
if key_padding_mask is not None:
attn_output_weights = attn_output_weights.view(bsz, num_heads,
tgt_len, src_len)
attn_output_weights = attn_output_weights.masked_fill(key_padding_mask
.unsqueeze(1).unsqueeze(2), float('-inf'))
attn_output_weights = attn_output_weights.view(bsz * num_heads,
tgt_len, src_len)
attn_output_weights = softmax(attn_output_weights, dim=-1)
attn_output_weights = dropout(attn_output_weights, p=dropout_p,
training=training)
attn_output = torch.bmm(attn_output_weights, v)
assert list(attn_output.size()) == [bsz * num_heads, tgt_len, head_dim]
attn_output = attn_output.transpose(0, 1).contiguous().view(tgt_len,
bsz, embed_dim)
attn_output = linear(attn_output, out_proj_weight, out_proj_bias)
if need_weights:
attn_output_weights = attn_output_weights.view(bsz, num_heads,
tgt_len, src_len)
return attn_output, attn_output_weights.sum(dim=1) / num_heads
else:
return attn_output, None
class RelationalMultiheadAttention(MultiheadAttention):
"""Allows the model to jointly attend to information
from different representation subspaces.
See `Attention Is All You Need <https://arxiv.org/abs/1706.03762>`_
.. math::
\\text{MultiHead}(Q, K, V) = \\text{Concat}(head_1,\\dots,head_h)W^O
where :math:`head_i = \\text{Attention}(QW_i^Q, KW_i^K, VW_i^V)`.
Args:
embed_dim: total dimension of the model.
num_heads: parallel attention heads.
dropout: a Dropout layer on attn_output_weights. Default: 0.0.
bias: add bias as module parameter. Default: True.
add_bias_kv: add bias to the key and value sequences at dim=0.
add_zero_attn: add a new batch of zeros to the key and
value sequences at dim=1.
kdim: total number of features in key. Default: None.
vdim: total number of features in value. Default: None.
Note that if :attr:`kdim` and :attr:`vdim` are None, they will be set
to :attr:`embed_dim` such that query, key, and value have the same
number of features.
Examples::
>>> realational_multihead_attn = RelationalMultiheadAttention(embed_dim, num_heads)
>>> attn_output, attn_output_weights = realational_multihead_attn(query, key, value)
"""
bias_k: 'Optional[torch.Tensor]'
bias_v: 'Optional[torch.Tensor]'
def __init__(self, embed_dim, num_heads, dropout=0.0, bias=True,
add_bias_kv=False, add_zero_attn=False, kdim=None, vdim=None,
add_relation=False, rdim=None, relation_type=None):
super(RelationalMultiheadAttention, self).__init__(embed_dim=
embed_dim, num_heads=num_heads, dropout=dropout, bias=bias,
add_bias_kv=add_bias_kv, add_zero_attn=add_zero_attn, kdim=kdim,
vdim=vdim)
self.add_relation = add_relation
self.rdim = rdim if rdim is not None else embed_dim
self.relation_type = relation_type if relation_type else 'qk+r'
if self.add_relation:
if relation_type == 'qk+r':
self.r_proj_weight = Parameter(torch.Tensor(num_heads, self
.rdim))
self.r_proj_bias = Parameter(torch.empty(num_heads))
elif relation_type == 'q(k+r)':
self.r_proj_weight = Parameter(torch.Tensor(embed_dim, self
.rdim))
self.r_proj_bias = Parameter(torch.empty(embed_dim))
self._reset_parameters()
def _reset_parameters(self):
if self._qkv_same_embed_dim:
xavier_uniform_(self.in_proj_weight)
else:
xavier_uniform_(self.q_proj_weight)
xavier_uniform_(self.k_proj_weight)
xavier_uniform_(self.v_proj_weight)
if hasattr(self, 'add_relation') and self.add_relation:
xavier_uniform_(self.r_proj_weight)
constant_(self.r_proj_bias, 0.0)
if self.in_proj_bias is not None:
constant_(self.in_proj_bias, 0.0)
constant_(self.out_proj.bias, 0.0)
if self.bias_k is not None:
xavier_normal_(self.bias_k)
if self.bias_v is not None:
xavier_normal_(self.bias_v)
def __setstate__(self, state):
if '_qkv_same_embed_dim' not in state:
state['_qkv_same_embed_dim'] = True
super(MultiheadAttention, self).__setstate__(state)
def forward(self, query: 'Tensor', key: 'Tensor', value: 'Tensor',
relation_dict=None, key_padding_mask: 'Optional[Tensor]'=None,
need_weights: 'bool'=True, attn_mask: 'Optional[Tensor]'=None) ->Tuple[
Tensor, Optional[Tensor]]:
"""
Args:
query, key, value: map a query and a set of key-value pairs to an output.
See "Attention Is All You Need" for more details.
key_padding_mask: if provided, specified padding elements in the key will
be ignored by the attention. When given a binary mask and a value is True,
the corresponding value on the attention layer will be ignored. When given
a byte mask and a value is non-zero, the corresponding value on the attention
layer will be ignored
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
Shapes for inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the position
with the zero positions will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while the position with the value of ``False`` will be unchanged.
- attn_mask: if a 2D mask: :math:`(L, S)` where L is the target sequence length, S is the
source sequence length.
If a 3D mask: :math:`(N\\cdot\\text{num\\_heads}, L, S)` where N is the batch size, L is the target sequence
length, S is the source sequence length. ``attn_mask`` ensure that position i is allowed to attend
the unmasked positions. If a ByteTensor is provided, the non-zero positions are not allowed to attend
while the zero positions will be unchanged. If a BoolTensor is provided, positions with ``True``
is not allowed to attend while ``False`` values will be unchanged. If a FloatTensor
is provided, it will be added to the attention weight.
Shapes for outputs:
- attn_output: :math:`(L, N, E)` where L is the target sequence length, N is the batch size,
E is the embedding dimension.
- attn_output_weights: :math:`(N, L, S)` where N is the batch size,
L is the target sequence length, S is the source sequence length.
"""
if relation_dict is not None:
relation_labels = relation_dict['relation_labels']
relation_ids = relation_dict['relation_ids']
batch_index = relation_dict['batch_index']
pad_embedding = relation_dict['pad_embedding']
relation_labels = linear(relation_labels, self.r_proj_weight,
self.r_proj_bias)
pad_embedding = linear(pad_embedding.unsqueeze(0), self.
r_proj_weight, self.r_proj_bias).squeeze()
tgt_length, bsz, _ = query.size()
src_length, _, _ = key.size()
relation = pad_embedding.view(1, 1, 1, -1).repeat(bsz,
tgt_length, src_length, 1)
relation[batch_index, relation_ids[:, :, 0], relation_ids[:, :, 1]
] = relation_labels
relation = relation.permute(1, 2, 0, 3)
if not self._qkv_same_embed_dim:
return relational_multi_head_attention_forward(query, key,
value, relation, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.
bias_v, self.add_zero_attn, self.dropout, self.out_proj
.weight, self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask,
use_separate_proj_weight=True, q_proj_weight=self.
q_proj_weight, k_proj_weight=self.k_proj_weight,
v_proj_weight=self.v_proj_weight)
else:
return relational_multi_head_attention_forward(query, key,
value, relation, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.
bias_v, self.add_zero_attn, self.dropout, self.out_proj
.weight, self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask)
elif not self._qkv_same_embed_dim:
return relational_multi_head_attention_forward(query, key,
value, None, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.bias_v,
self.add_zero_attn, self.dropout, self.out_proj.weight,
self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask, use_separate_proj_weight
=True, q_proj_weight=self.q_proj_weight, k_proj_weight=self
.k_proj_weight, v_proj_weight=self.v_proj_weight)
else:
return relational_multi_head_attention_forward(query, key,
value, None, self.embed_dim, self.num_heads, self.
in_proj_weight, self.in_proj_bias, self.bias_k, self.bias_v,
self.add_zero_attn, self.dropout, self.out_proj.weight,
self.out_proj.bias, training=self.training,
key_padding_mask=key_padding_mask, need_weights=
need_weights, attn_mask=attn_mask)
class RelationalTransformerEncoderLayerNew(TransformerEncoderLayer):
"""TransformerEncoderLayer is made up of self-attn and feedforward network.
This standard encoder layer is based on the paper "Attention Is All You Need".
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
in a different way during application.
Args:
d_model: the number of expected features in the input (required).
nhead: the number of heads in the multiheadattention models (required).
dim_feedforward: the dimension of the feedforward network model (default=2048).
dropout: the dropout value (default=0.1).
activation: the activation function of intermediate layer, relu or gelu (default=relu).
Examples::
>>> encoder_layer = RelationalTransformerEncoderLayer(d_model=512, nhead=8)
>>> src = torch.rand(10, 32, 512)
>>> rel = torch.rand(10, 10, 32, 512)
>>> out = encoder_layer(src, rel)
"""
def __init__(self, d_model, nhead, add_relation=False, dim_feedforward=
2048, dropout=0.1, activation='relu', relation_type=None):
super(RelationalTransformerEncoderLayerNew, self).__init__(d_model,
nhead, dim_feedforward=dim_feedforward, dropout=dropout,
activation=activation)
self.self_attn = RelationalMultiheadAttention(d_model, nhead,
add_relation=add_relation, dropout=dropout, relation_type=
relation_type)
def forward(self, input_0):
primals_1 = self.self_attn.in_proj_weight
primals_2 = self.self_attn.in_proj_bias
primals_3 = self.self_attn.out_proj.weight
primals_4 = self.self_attn.out_proj.bias
primals_8 = self.linear1.weight
primals_9 = self.linear1.bias
primals_10 = self.linear2.weight
primals_6 = self.linear2.bias
primals_7 = self.norm1.weight
primals_11 = self.norm1.bias
primals_12 = self.norm2.weight
primals_13 = self.norm2.bias
primals_5 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13])
return output[0]
| mfk3138/jiant | RelationalTransformerEncoderLayer | false | 4,204 | [
"MIT"
] | 0 | 6e67ff1ecb1bb98533c1019a86af4ad2c04c6a64 | https://github.com/mfk3138/jiant/tree/6e67ff1ecb1bb98533c1019a86af4ad2c04c6a64 | import torch
import warnings
from torch import Tensor
from torch.nn import TransformerEncoderLayer
from torch.nn.functional import *
from torch.nn.modules.activation import MultiheadAttention
from torch.nn.modules.activation import xavier_uniform_
from torch.nn.modules.activation import xavier_normal_
from torch.nn.modules.activation import constant_
from torch.nn.modules.activation import Parameter
from typing import Optional
import torch.utils.data.dataset
from typing import Tuple
def relational_multi_head_attention_forward(query: 'Tensor', key: 'Tensor',
value: 'Tensor', relation: 'Tensor', embed_dim_to_check: 'int',
num_heads: 'int', in_proj_weight: 'Tensor', in_proj_bias: 'Tensor',
bias_k: 'Optional[Tensor]', bias_v: 'Optional[Tensor]', add_zero_attn:
'bool', dropout_p: 'float', out_proj_weight: 'Tensor', out_proj_bias:
'Tensor', training: 'bool'=True, key_padding_mask: 'Optional[Tensor]'=
None, need_weights: 'bool'=True, attn_mask: 'Optional[Tensor]'=None,
use_separate_proj_weight: 'bool'=False, q_proj_weight:
'Optional[Tensor]'=None, k_proj_weight: 'Optional[Tensor]'=None,
v_proj_weight: 'Optional[Tensor]'=None, static_k: 'Optional[Tensor]'=
None, static_v: 'Optional[Tensor]'=None, relation_type: 'str'=None
) ->Tuple[Tensor, Optional[Tensor]]:
"""
Args:
query, key, value: map a query and a set of key-value pairs to an output.
See "Attention Is All You Need" for more details.
relation: relation between queries and keys.
embed_dim_to_check: total dimension of the model.
num_heads: parallel attention heads.
in_proj_weight, in_proj_bias: input projection weight and bias.
bias_k, bias_v: bias of the key and value sequences to be added at dim=0.
add_zero_attn: add a new batch of zeros to the key and
value sequences at dim=1.
dropout_p: probability of an element to be zeroed.
out_proj_weight, out_proj_bias: the output projection weight and bias.
training: apply dropout if is ``True``.
key_padding_mask: if provided, specified padding elements in the key will
be ignored by the attention. This is an binary mask. When the value is True,
the corresponding value on the attention layer will be filled with -inf.
need_weights: output attn_output_weights.
attn_mask: 2D or 3D mask that prevents attention to certain positions. A 2D mask will be broadcasted for all
the batches while a 3D mask allows to specify a different mask for the entries of each batch.
use_separate_proj_weight: the function accept the proj. weights for query, key,
and value in different forms. If false, in_proj_weight will be used, which is
a combination of q_proj_weight, k_proj_weight, v_proj_weight.
q_proj_weight, k_proj_weight, v_proj_weight, in_proj_bias: input projection weight and bias.
static_k, static_v: static key and value used for attention operators.
Shape:
Inputs:
- query: :math:`(L, N, E)` where L is the target sequence length, N is the batch size, E is
the embedding dimension.
- key: :math:`(S, N, E)`, where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- value: :math:`(S, N, E)` where S is the source sequence length, N is the batch size, E is
the embedding dimension.
- relation: :math:`(L, S, N, E)` where L is the target sequence length, where S is the source sequence length,
N is the batch size, E is the embedding dimension.
- key_padding_mask: :math:`(N, S)` where N is the batch size, S is the source sequence length.
If a ByteTensor is provided, the non-zero positions will be ignored while the zero positions
will be unchanged. If a BoolTensor is provided, the positions with the
value of ``True`` will be ignored while th
# ... truncated (>4000 chars) for memory efficiency |
LxmertAttentionOutput | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ai/cai32p2ssjvpyulvuzcicdszqe3thbavgxn4jeed6uatjnl7yq2s.py
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add => add
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_4), kwargs = {})
triton_poi_fused_add_0 = async_compile.triton('triton_poi_fused_add_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nk/cnkbkukjfarsysqlaadkg24xmqibk3adq5p7jyfnt6k6loydbn2r.py
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_2 => add_1, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [3]), kwargs = {correction: 0, keepdim: True})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
triton_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-12
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + (x0), tmp8, xmask)
tl.store(out_ptr1 + (x0), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/mn/cmntyljhuirhsdjg2yosgzllpkpxqedxgoyk6gunquq2rf3kl7u5.py
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_2 => add_1, add_2, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [3]), kwargs = {correction: 0, keepdim: True})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_5), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_6), kwargs = {})
triton_poi_fused_native_layer_norm_2 = async_compile.triton('triton_poi_fused_native_layer_norm_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(buf1, primals_2, primals_4, 256, grid=grid(256), stream=stream0)
del primals_2
del primals_4
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(buf1, buf2, buf3, 64, grid=grid(64), stream=stream0)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_2.run(buf1, buf2, buf3, primals_5, primals_6, buf4, 256, grid=grid(256), stream=stream0)
del buf2
del buf3
del primals_6
return (buf4, primals_5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import torch
from torch import nn
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, hidden_dropout_prob=
0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-12
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp23, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](buf1, primals_2, primals_4, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
del primals_4
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(64)](buf1, buf2, buf3, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_2[grid(256)](buf1, buf2, buf3,
primals_5, primals_6, buf4, 256, XBLOCK=256, num_warps=4,
num_stages=1)
del buf2
del buf3
del primals_6
return buf4, primals_5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1
class LxmertAttentionOutputNew(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, input_0, input_1):
primals_1 = self.dense.weight
primals_2 = self.dense.bias
primals_5 = self.LayerNorm.weight
primals_6 = self.LayerNorm.bias
primals_3 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| rsgit95/med_kg_txt_multimodal | LxmertAttentionOutput | false | 4,205 | [
"Apache-2.0"
] | 0 | 80355b0cf58e0571531ad6f9728c533110ca996d | https://github.com/rsgit95/med_kg_txt_multimodal/tree/80355b0cf58e0571531ad6f9728c533110ca996d | from _paritybench_helpers import _mock_config
import torch
from torch import nn
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, hidden_dropout_prob=
0.5)}]
|
Block | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/zw/czwemfbc42kjdc6k3rt2lguzq7gywvrwr3b4jc5tlwvxk2bj7thg.py
# Topologically Sorted Source Nodes: [layer_norm], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# layer_norm => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-05), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_3, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_2), kwargs = {})
triton_per_fused_native_layer_norm_0 = async_compile.triton('triton_per_fused_native_layer_norm_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[8192, 512],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_native_layer_norm_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 3, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_native_layer_norm_0(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, xnumel, rnumel):
xnumel = 8192
XBLOCK: tl.constexpr = 1
rnumel = 512
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (512*x0)), None)
tmp21 = tl.load(in_ptr1 + (r1), None, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr2 + (r1), None, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [RBLOCK])
tmp3 = tl.broadcast_to(tmp1, [RBLOCK])
tmp5 = triton_helpers.promote_to_tensor(tl.sum(tmp3, 0))
tmp6 = tl.full([1], 512, tl.int32)
tmp7 = tmp6.to(tl.float32)
tmp8 = tmp5 / tmp7
tmp9 = tmp1 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tl.broadcast_to(tmp10, [RBLOCK])
tmp13 = triton_helpers.promote_to_tensor(tl.sum(tmp11, 0))
tmp14 = 512.0
tmp15 = tmp13 / tmp14
tmp16 = 1e-05
tmp17 = tmp15 + tmp16
tmp18 = libdevice.rsqrt(tmp17)
tmp19 = tmp0 - tmp8
tmp20 = tmp19 * tmp18
tmp22 = tmp20 * tmp21
tmp24 = tmp22 + tmp23
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp18, None)
tl.store(out_ptr1 + (r1 + (512*x0)), tmp24, None)
tl.store(out_ptr0 + (x0), tmp8, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/65/c65pi2qtb6eyfexgoigw5cn2nkouol3ju46n5j4xo2rc4qkpo3ge.py
# Topologically Sorted Source Nodes: [x_1, layer_norm_1], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# layer_norm_1 => add_3, add_4, mul_2, mul_3, rsqrt_1, sub_1, var_mean_1
# x_1 => add_2
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %view_1), kwargs = {})
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_2, [3]), kwargs = {correction: 0, keepdim: True})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_2, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_3,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %getitem_3), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt_1), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %primals_6), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_3, %primals_7), kwargs = {})
triton_per_fused_add_native_layer_norm_1 = async_compile.triton('triton_per_fused_add_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[8192, 512],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32', 8: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_native_layer_norm_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 4, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_native_layer_norm_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, rnumel):
xnumel = 8192
XBLOCK: tl.constexpr = 1
rnumel = 512
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (512*x0)), None)
tmp1 = tl.load(in_ptr1 + (r1 + (512*x0)), None)
tmp23 = tl.load(in_ptr2 + (r1), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr3 + (r1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = tl.broadcast_to(tmp3, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tmp8 = tl.full([1], 512, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp3 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 512.0
tmp17 = tmp15 / tmp16
tmp18 = 1e-05
tmp19 = tmp17 + tmp18
tmp20 = libdevice.rsqrt(tmp19)
tmp21 = tmp2 - tmp10
tmp22 = tmp21 * tmp20
tmp24 = tmp22 * tmp23
tmp26 = tmp24 + tmp25
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp20, None)
tl.store(out_ptr1 + (r1 + (512*x0)), tmp26, None)
tl.store(out_ptr0 + (x0), tmp10, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/jg/cjghtxku4uxojc6qot4mjj5pxz2z5t6h4tdjsogj7sz5tuogsufs.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.gelu]
# Source node to ATen node mapping:
# x_3 => add_5, erf, mul_4, mul_5, mul_6
# Graph fragment:
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_3, 0.5), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_3, 0.7071067811865476), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%mul_5,), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1), kwargs = {})
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_4, %add_5), kwargs = {})
triton_poi_fused_gelu_2 = async_compile.triton('triton_poi_fused_gelu_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16777216],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_gelu_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_gelu_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16777216
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), None)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865476
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tl.store(out_ptr0 + (x0), tmp8, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/pa/cpavnfhh6w5k3ia3bkbpbo6pvpzyknjt46fji7qua27hjunmkmdc.py
# Topologically Sorted Source Nodes: [x_1, x_7], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x_1 => add_2
# x_7 => add_6
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %view_1), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %view_5), kwargs = {})
triton_poi_fused_add_3 = async_compile.triton('triton_poi_fused_add_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4194304],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_3', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_3(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 4194304
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr1 + (x2), None)
tmp3 = tl.load(in_out_ptr0 + (x2), None)
tmp4 = tl.load(in_ptr2 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tl.store(in_out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (512, ), (1, ))
assert_size_stride(primals_2, (512, ), (1, ))
assert_size_stride(primals_3, (4, 4, 512, 512), (1048576, 262144, 512, 1))
assert_size_stride(primals_4, (512, 512), (512, 1))
assert_size_stride(primals_5, (512, ), (1, ))
assert_size_stride(primals_6, (512, ), (1, ))
assert_size_stride(primals_7, (512, ), (1, ))
assert_size_stride(primals_8, (2048, 512), (512, 1))
assert_size_stride(primals_9, (2048, ), (1, ))
assert_size_stride(primals_10, (512, 2048), (2048, 1))
assert_size_stride(primals_11, (512, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 8192), torch.float32)
buf3 = reinterpret_tensor(buf1, (4, 4, 512, 1), (2048, 512, 1, 1), 0); del buf1 # reuse
buf4 = empty_strided_cuda((4, 4, 512, 512), (1048576, 262144, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [layer_norm], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_per_fused_native_layer_norm_0.run(buf3, primals_3, primals_1, primals_2, buf0, buf4, 8192, 512, grid=grid(8192), stream=stream0)
del primals_1
del primals_2
buf5 = empty_strided_cuda((8192, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf4, (8192, 512), (512, 1), 0), reinterpret_tensor(primals_4, (512, 512), (1, 512), 0), alpha=1, beta=1, out=buf5)
del primals_5
buf6 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 8192), torch.float32)
buf9 = reinterpret_tensor(buf7, (4, 4, 512, 1), (2048, 512, 1, 1), 0); del buf7 # reuse
buf10 = empty_strided_cuda((4, 4, 512, 512), (1048576, 262144, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1, layer_norm_1], Original ATen: [aten.add, aten.native_layer_norm]
triton_per_fused_add_native_layer_norm_1.run(buf9, primals_3, buf5, primals_6, primals_7, buf6, buf10, 8192, 512, grid=grid(8192), stream=stream0)
del primals_7
buf11 = empty_strided_cuda((8192, 2048), (2048, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_9, reinterpret_tensor(buf10, (8192, 512), (512, 1), 0), reinterpret_tensor(primals_8, (512, 2048), (1, 512), 0), alpha=1, beta=1, out=buf11)
del primals_9
buf12 = empty_strided_cuda((4, 4, 512, 2048), (4194304, 1048576, 2048, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.gelu]
triton_poi_fused_gelu_2.run(buf11, buf12, 16777216, grid=grid(16777216), stream=stream0)
buf13 = empty_strided_cuda((8192, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf12, (8192, 2048), (2048, 1), 0), reinterpret_tensor(primals_10, (2048, 512), (1, 2048), 0), out=buf13)
buf14 = reinterpret_tensor(buf13, (4, 4, 512, 512), (1048576, 262144, 512, 1), 0); del buf13 # reuse
# Topologically Sorted Source Nodes: [x_1, x_7], Original ATen: [aten.add]
triton_poi_fused_add_3.run(buf14, primals_3, buf5, primals_11, 4194304, grid=grid(4194304), stream=stream0)
del primals_11
return (buf14, primals_3, primals_6, buf0, buf3, reinterpret_tensor(buf4, (8192, 512), (512, 1), 0), buf5, buf6, buf9, reinterpret_tensor(buf10, (8192, 512), (512, 1), 0), buf11, reinterpret_tensor(buf12, (8192, 2048), (2048, 1), 0), primals_10, primals_8, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 512, 512), (1048576, 262144, 512, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((512, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((2048, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((2048, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((512, 2048), (2048, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch as th
from torch import nn
def drop_path(x, drop_prob: 'float'=0.0, training: 'bool'=False):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0.0 or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1)
random_tensor = keep_prob + th.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_()
output = x.div(keep_prob) * random_tensor
return output
class Attention(nn.Module):
def __init__(self, dim=2, num_heads=8, qkv_bias=False, qk_scale=None,
attn_drop=0.0, proj_drop=0.0):
super().__init__()
self.proj = nn.Linear(dim, dim)
def forward(self, x):
x = self.proj(x)
return x
class DropPath(th.nn.Module):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
def forward(self, x):
return drop_path(x, self.drop_prob, self.training)
class Mlp(th.nn.Module):
""" MLP as used in Vision Transformer, MLP-Mixer and related networks
"""
def __init__(self, in_features, hidden_features=None, out_features=None,
act_layer=th.nn.GELU, drop=0.0):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = th.nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = th.nn.Linear(hidden_features, out_features)
self.drop = th.nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
class Block(th.nn.Module):
def __init__(self, dim=512, num_heads=8, mlp_ratio=4.0, qkv_bias=False,
qk_scale=None, drop=0.0, attn_drop=0.0, drop_path=0.0, act_layer=th
.nn.GELU, norm_layer=th.nn.LayerNorm):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias,
qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
self.drop_path = DropPath(drop_path
) if drop_path > 0.0 else th.nn.Identity()
self.norm2 = norm_layer(dim)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim,
act_layer=act_layer, drop=drop)
def forward(self, x):
x = x + self.drop_path(self.attn(self.norm1(x)))
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
def get_inputs():
return [torch.rand([4, 4, 512, 512])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import torch as th
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_native_layer_norm_0(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, out_ptr0, out_ptr1, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 512 * x0), None)
tmp21 = tl.load(in_ptr1 + r1, None, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr2 + r1, None, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [RBLOCK])
tmp3 = tl.broadcast_to(tmp1, [RBLOCK])
tmp5 = triton_helpers.promote_to_tensor(tl.sum(tmp3, 0))
tmp6 = tl.full([1], 512, tl.int32)
tmp7 = tmp6.to(tl.float32)
tmp8 = tmp5 / tmp7
tmp9 = tmp1 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tl.broadcast_to(tmp10, [RBLOCK])
tmp13 = triton_helpers.promote_to_tensor(tl.sum(tmp11, 0))
tmp14 = 512.0
tmp15 = tmp13 / tmp14
tmp16 = 1e-05
tmp17 = tmp15 + tmp16
tmp18 = libdevice.rsqrt(tmp17)
tmp19 = tmp0 - tmp8
tmp20 = tmp19 * tmp18
tmp22 = tmp20 * tmp21
tmp24 = tmp22 + tmp23
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp18, None)
tl.store(out_ptr1 + (r1 + 512 * x0), tmp24, None)
tl.store(out_ptr0 + x0, tmp8, None)
@triton.jit
def triton_per_fused_add_native_layer_norm_1(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 512 * x0), None)
tmp1 = tl.load(in_ptr1 + (r1 + 512 * x0), None)
tmp23 = tl.load(in_ptr2 + r1, None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr3 + r1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = tl.broadcast_to(tmp3, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tmp8 = tl.full([1], 512, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp3 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 512.0
tmp17 = tmp15 / tmp16
tmp18 = 1e-05
tmp19 = tmp17 + tmp18
tmp20 = libdevice.rsqrt(tmp19)
tmp21 = tmp2 - tmp10
tmp22 = tmp21 * tmp20
tmp24 = tmp22 * tmp23
tmp26 = tmp24 + tmp25
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp20, None)
tl.store(out_ptr1 + (r1 + 512 * x0), tmp26, None)
tl.store(out_ptr0 + x0, tmp10, None)
@triton.jit
def triton_poi_fused_gelu_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, None)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865476
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tl.store(out_ptr0 + x0, tmp8, None)
@triton.jit
def triton_poi_fused_add_3(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_ptr0 + x2, None)
tmp1 = tl.load(in_ptr1 + x2, None)
tmp3 = tl.load(in_out_ptr0 + x2, None)
tmp4 = tl.load(in_ptr2 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tl.store(in_out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (512,), (1,))
assert_size_stride(primals_2, (512,), (1,))
assert_size_stride(primals_3, (4, 4, 512, 512), (1048576, 262144, 512, 1))
assert_size_stride(primals_4, (512, 512), (512, 1))
assert_size_stride(primals_5, (512,), (1,))
assert_size_stride(primals_6, (512,), (1,))
assert_size_stride(primals_7, (512,), (1,))
assert_size_stride(primals_8, (2048, 512), (512, 1))
assert_size_stride(primals_9, (2048,), (1,))
assert_size_stride(primals_10, (512, 2048), (2048, 1))
assert_size_stride(primals_11, (512,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 1), torch.
float32)
buf1 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 8192),
torch.float32)
buf3 = reinterpret_tensor(buf1, (4, 4, 512, 1), (2048, 512, 1, 1), 0)
del buf1
buf4 = empty_strided_cuda((4, 4, 512, 512), (1048576, 262144, 512,
1), torch.float32)
get_raw_stream(0)
triton_per_fused_native_layer_norm_0[grid(8192)](buf3, primals_3,
primals_1, primals_2, buf0, buf4, 8192, 512, num_warps=4,
num_stages=1)
del primals_1
del primals_2
buf5 = empty_strided_cuda((8192, 512), (512, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf4, (8192, 512
), (512, 1), 0), reinterpret_tensor(primals_4, (512, 512), (1,
512), 0), alpha=1, beta=1, out=buf5)
del primals_5
buf6 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 1), torch.
float32)
buf7 = empty_strided_cuda((4, 4, 512, 1), (2048, 512, 1, 8192),
torch.float32)
buf9 = reinterpret_tensor(buf7, (4, 4, 512, 1), (2048, 512, 1, 1), 0)
del buf7
buf10 = empty_strided_cuda((4, 4, 512, 512), (1048576, 262144, 512,
1), torch.float32)
triton_per_fused_add_native_layer_norm_1[grid(8192)](buf9,
primals_3, buf5, primals_6, primals_7, buf6, buf10, 8192, 512,
num_warps=4, num_stages=1)
del primals_7
buf11 = empty_strided_cuda((8192, 2048), (2048, 1), torch.float32)
extern_kernels.addmm(primals_9, reinterpret_tensor(buf10, (8192,
512), (512, 1), 0), reinterpret_tensor(primals_8, (512, 2048),
(1, 512), 0), alpha=1, beta=1, out=buf11)
del primals_9
buf12 = empty_strided_cuda((4, 4, 512, 2048), (4194304, 1048576,
2048, 1), torch.float32)
triton_poi_fused_gelu_2[grid(16777216)](buf11, buf12, 16777216,
XBLOCK=1024, num_warps=4, num_stages=1)
buf13 = empty_strided_cuda((8192, 512), (512, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf12, (8192, 2048), (2048, 1),
0), reinterpret_tensor(primals_10, (2048, 512), (1, 2048), 0),
out=buf13)
buf14 = reinterpret_tensor(buf13, (4, 4, 512, 512), (1048576,
262144, 512, 1), 0)
del buf13
triton_poi_fused_add_3[grid(4194304)](buf14, primals_3, buf5,
primals_11, 4194304, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_11
return buf14, primals_3, primals_6, buf0, buf3, reinterpret_tensor(buf4,
(8192, 512), (512, 1), 0), buf5, buf6, buf9, reinterpret_tensor(buf10,
(8192, 512), (512, 1), 0), buf11, reinterpret_tensor(buf12, (8192,
2048), (2048, 1), 0), primals_10, primals_8, primals_4
def drop_path(x, drop_prob: 'float'=0.0, training: 'bool'=False):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0.0 or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1)
random_tensor = keep_prob + th.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_()
output = x.div(keep_prob) * random_tensor
return output
class Attention(nn.Module):
def __init__(self, dim=2, num_heads=8, qkv_bias=False, qk_scale=None,
attn_drop=0.0, proj_drop=0.0):
super().__init__()
self.proj = nn.Linear(dim, dim)
def forward(self, x):
x = self.proj(x)
return x
class DropPath(th.nn.Module):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
def forward(self, x):
return drop_path(x, self.drop_prob, self.training)
class Mlp(th.nn.Module):
""" MLP as used in Vision Transformer, MLP-Mixer and related networks
"""
def __init__(self, in_features, hidden_features=None, out_features=None,
act_layer=th.nn.GELU, drop=0.0):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = th.nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = th.nn.Linear(hidden_features, out_features)
self.drop = th.nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
class BlockNew(th.nn.Module):
def __init__(self, dim=512, num_heads=8, mlp_ratio=4.0, qkv_bias=False,
qk_scale=None, drop=0.0, attn_drop=0.0, drop_path=0.0, act_layer=th
.nn.GELU, norm_layer=th.nn.LayerNorm):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias,
qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
self.drop_path = DropPath(drop_path
) if drop_path > 0.0 else th.nn.Identity()
self.norm2 = norm_layer(dim)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim,
act_layer=act_layer, drop=drop)
def forward(self, input_0):
primals_1 = self.norm1.weight
primals_2 = self.norm1.bias
primals_4 = self.attn.proj.weight
primals_5 = self.attn.proj.bias
primals_6 = self.norm2.weight
primals_7 = self.norm2.bias
primals_8 = self.mlp.fc1.weight
primals_9 = self.mlp.fc1.bias
primals_10 = self.mlp.fc2.weight
primals_11 = self.mlp.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
| q5628077/Transformer-in-RL | Block | false | 4,206 | [
"MIT"
] | 0 | 14679656779a372d91d9fbd89bd802b5ff34c200 | https://github.com/q5628077/Transformer-in-RL/tree/14679656779a372d91d9fbd89bd802b5ff34c200 | import torch
import torch as th
from torch import nn
def drop_path(x, drop_prob: 'float'=0.0, training: 'bool'=False):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0.0 or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1)
random_tensor = keep_prob + th.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_()
output = x.div(keep_prob) * random_tensor
return output
class Attention(nn.Module):
def __init__(self, dim=2, num_heads=8, qkv_bias=False, qk_scale=None,
attn_drop=0.0, proj_drop=0.0):
super().__init__()
self.proj = nn.Linear(dim, dim)
def forward(self, x):
x = self.proj(x)
return x
class DropPath(th.nn.Module):
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None):
super().__init__()
self.drop_prob = drop_prob
def forward(self, x):
return drop_path(x, self.drop_prob, self.training)
class Mlp(th.nn.Module):
""" MLP as used in Vision Transformer, MLP-Mixer and related networks
"""
def __init__(self, in_features, hidden_features=None, out_features=None,
act_layer=th.nn.GELU, drop=0.0):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = th.nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = th.nn.Linear(hidden_features, out_features)
self.drop = th.nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
class Model(th.nn.Module):
def __init__(self, dim=512, num_heads=8, mlp_ratio=4.0, qkv_bias=False,
qk_scale=None, drop=0.0, attn_drop=0.0, drop_path=0.0, act_layer=th
.nn.GELU, norm_layer=th.nn.LayerNorm):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias,
qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
self.drop_path = DropPath(drop_path
) if drop_path > 0.0 else th.nn.Identity()
self.norm2 = norm_layer(dim)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim,
act_layer=act_layer, drop=drop)
def forward(self, x):
x = x + self.drop_path(self.attn(self.norm1(x)))
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
def get_inputs():
return [torch.rand([4, 4, 512, 512])]
def get_init_inputs():
return []
|
Net | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ky/cky64l574tkwxzjewzevqyhty73x4t3q4p6d2tu2humfvstjwiaa.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_2 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 2048
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 32
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xc/cxcj5l7s6w5ttrq2fk2nirlbp44yesw6n2m2dnxtpcjjmym2njhr.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_3 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/h7/ch7av6xnulewt5b7odqowg5upc5aaxv4uylilvlgoap3w6rnompj.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_5 => relu_2
# Graph fragment:
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_5,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_2, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_2 = async_compile.triton('triton_poi_fused_relu_threshold_backward_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_2(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (32, 4), (4, 1))
assert_size_stride(primals_2, (32, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 32), (32, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (128, 64), (64, 1))
assert_size_stride(primals_7, (128, ), (1, ))
assert_size_stride(primals_8, (4, 128), (128, 1))
assert_size_stride(primals_9, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 32), (32, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 32), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 32), (512, 128, 32, 1), 0); del buf0 # reuse
buf9 = empty_strided_cuda((4, 4, 4, 32), (512, 128, 32, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf9, 2048, grid=grid(2048), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 32), (32, 1), 0), reinterpret_tensor(primals_4, (32, 64), (1, 32), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf2 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf8, 4096, grid=grid(4096), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(primals_6, (64, 128), (1, 64), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 128), (2048, 512, 128, 1), 0); del buf4 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_2.run(buf5, primals_7, buf7, 8192, grid=grid(8192), stream=stream0)
del primals_7
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_9, reinterpret_tensor(buf5, (64, 128), (128, 1), 0), reinterpret_tensor(primals_8, (128, 4), (1, 128), 0), alpha=1, beta=1, out=buf6)
del primals_9
return (reinterpret_tensor(buf6, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 32), (32, 1), 0), reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(buf5, (64, 128), (128, 1), 0), primals_8, buf7, primals_6, buf8, primals_4, buf9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((32, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 32), (32, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((128, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 128), (128, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class Net(nn.Module):
def __init__(self, state_dim, action_dim):
super(Net, self).__init__()
fc1_dim = 32
fc2_dim = 64
fc3_dim = 128
self.fc1 = nn.Linear(state_dim, fc1_dim)
self.fc2 = nn.Linear(fc1_dim, fc2_dim)
self.fc3 = nn.Linear(fc2_dim, fc3_dim)
self.fc_out = nn.Linear(fc3_dim, action_dim)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x)
x = self.fc2(x)
x = F.relu(x)
x = self.fc3(x)
x = F.relu(x)
return self.fc_out(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4, 'action_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 32
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused_relu_threshold_backward_2(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (32, 4), (4, 1))
assert_size_stride(primals_2, (32,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 32), (32, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (128, 64), (64, 1))
assert_size_stride(primals_7, (128,), (1,))
assert_size_stride(primals_8, (4, 128), (128, 1))
assert_size_stride(primals_9, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 32), (32, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 32), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 32), (512, 128, 32, 1), 0)
del buf0
buf9 = empty_strided_cuda((4, 4, 4, 32), (512, 128, 32, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(2048)](buf1,
primals_2, buf9, 2048, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 32), (32, 1), 0),
reinterpret_tensor(primals_4, (32, 64), (1, 32), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf2
buf8 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.bool
)
triton_poi_fused_relu_threshold_backward_1[grid(4096)](buf3,
primals_5, buf8, 4096, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 64), (64, 1), 0),
reinterpret_tensor(primals_6, (64, 128), (1, 64), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 128), (2048, 512, 128, 1), 0)
del buf4
buf7 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_2[grid(8192)](buf5,
primals_7, buf7, 8192, XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_9, reinterpret_tensor(buf5, (64, 128),
(128, 1), 0), reinterpret_tensor(primals_8, (128, 4), (1, 128),
0), alpha=1, beta=1, out=buf6)
del primals_9
return reinterpret_tensor(buf6, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 32), (32, 1), 0), reinterpret_tensor(
buf3, (64, 64), (64, 1), 0), reinterpret_tensor(buf5, (64, 128), (
128, 1), 0), primals_8, buf7, primals_6, buf8, primals_4, buf9
class NetNew(nn.Module):
def __init__(self, state_dim, action_dim):
super(NetNew, self).__init__()
fc1_dim = 32
fc2_dim = 64
fc3_dim = 128
self.fc1 = nn.Linear(state_dim, fc1_dim)
self.fc2 = nn.Linear(fc1_dim, fc2_dim)
self.fc3 = nn.Linear(fc2_dim, fc3_dim)
self.fc_out = nn.Linear(fc3_dim, action_dim)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_8 = self.fc_out.weight
primals_9 = self.fc_out.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
| ronekko/study_reinforcement_learning | Net | false | 4,207 | [
"MIT"
] | 0 | ef5201e3eae69c20f29b7f176b5a6de7ecdb856a | https://github.com/ronekko/study_reinforcement_learning/tree/ef5201e3eae69c20f29b7f176b5a6de7ecdb856a | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, state_dim, action_dim):
super().__init__()
fc1_dim = 32
fc2_dim = 64
fc3_dim = 128
self.fc1 = nn.Linear(state_dim, fc1_dim)
self.fc2 = nn.Linear(fc1_dim, fc2_dim)
self.fc3 = nn.Linear(fc2_dim, fc3_dim)
self.fc_out = nn.Linear(fc3_dim, action_dim)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x)
x = self.fc2(x)
x = F.relu(x)
x = self.fc3(x)
x = F.relu(x)
return self.fc_out(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
IReLU | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/t3/ct33p6ilro47ee445n2n3qpwlctivf63qsesypc37wtn4y6gjqwv.py
# Topologically Sorted Source Nodes: [clamp, mul, clamp_1, mul_1, add], Original ATen: [aten.clamp, aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# clamp => clamp_min
# clamp_1 => clamp_max
# mul => mul
# mul_1 => mul_1
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%arg0_1, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%clamp_min, 2.414213562373095), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%arg0_1, 0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%clamp_max, 0.41421356237309503), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
triton_poi_fused_add_clamp_mul_0 = async_compile.triton('triton_poi_fused_add_clamp_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_clamp_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_clamp_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.0
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp3 = 2.414213562373095
tmp4 = tmp2 * tmp3
tmp5 = triton_helpers.minimum(tmp0, tmp1)
tmp6 = 0.41421356237309503
tmp7 = tmp5 * tmp6
tmp8 = tmp4 + tmp7
tl.store(out_ptr0 + (x0), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [clamp, mul, clamp_1, mul_1, add], Original ATen: [aten.clamp, aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_clamp_mul_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
class IReLU(torch.nn.Module):
__constants__ = ['negative_slope', 'positive_slope']
negative_slope: 'float'
positive_slope: 'float'
def __init__(self, negative_slope=math.tan(math.pi / 8), positive_slope
=math.tan(3 * math.pi / 8)):
super(IReLU, self).__init__()
self.negative_slope = negative_slope
self.positive_slope = positive_slope
def forward(self, x):
return torch.clamp(x, min=0) * self.positive_slope + torch.clamp(x,
max=0) * self.negative_slope
def inv(self, y):
return torch.clamp(y, min=0) / self.positive_slope + torch.clamp(y,
max=0) / self.negative_slope
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import math
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_clamp_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.0
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp3 = 2.414213562373095
tmp4 = tmp2 * tmp3
tmp5 = triton_helpers.minimum(tmp0, tmp1)
tmp6 = 0.41421356237309503
tmp7 = tmp5 * tmp6
tmp8 = tmp4 + tmp7
tl.store(out_ptr0 + x0, tmp8, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_clamp_mul_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class IReLUNew(torch.nn.Module):
__constants__ = ['negative_slope', 'positive_slope']
negative_slope: 'float'
positive_slope: 'float'
def __init__(self, negative_slope=math.tan(math.pi / 8), positive_slope
=math.tan(3 * math.pi / 8)):
super(IReLUNew, self).__init__()
self.negative_slope = negative_slope
self.positive_slope = positive_slope
def inv(self, y):
return torch.clamp(y, min=0) / self.positive_slope + torch.clamp(y,
max=0) / self.negative_slope
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| rupumped/DFL | IReLU | false | 4,208 | [
"BSD-3-Clause"
] | 0 | a4e4d96b7ce7522cf7fee3c2cfdbb54eb7a473f2 | https://github.com/rupumped/DFL/tree/a4e4d96b7ce7522cf7fee3c2cfdbb54eb7a473f2 | import math
import torch
class Model(torch.nn.Module):
__constants__ = ['negative_slope', 'positive_slope']
negative_slope: 'float'
positive_slope: 'float'
def __init__(self, negative_slope=math.tan(math.pi / 8), positive_slope
=math.tan(3 * math.pi / 8)):
super().__init__()
self.negative_slope = negative_slope
self.positive_slope = positive_slope
def forward(self, x):
return torch.clamp(x, min=0) * self.positive_slope + torch.clamp(x,
max=0) * self.negative_slope
def inv(self, y):
return torch.clamp(y, min=0) / self.positive_slope + torch.clamp(y,
max=0) / self.negative_slope
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Affine | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/qr/cqr7rmxj75bvywume64cxzcwmqxcjc4bemo2iikheyw7jvomskyi.py
# Topologically Sorted Source Nodes: [mul, add], Original ATen: [aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %primals_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_3), kwargs = {})
triton_poi_fused_add_mul_0 = async_compile.triton('triton_poi_fused_add_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tl.store(out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 1, 4), (4, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (1, 1, 4), (4, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, add], Original ATen: [aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_0.run(primals_2, primals_1, primals_3, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_3
return (buf0, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((1, 1, 4), (4, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, 1, 4), (4, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class Affine(nn.Module):
def __init__(self, channel):
super().__init__()
self.g = nn.Parameter(torch.ones(1, 1, channel))
self.b = nn.Parameter(torch.zeros(1, 1, channel))
def forward(self, x):
return x * self.g + self.b
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'channel': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tl.store(out_ptr0 + x2, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 1, 4), (4, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (1, 1, 4), (4, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mul_0[grid(256)](primals_2, primals_1,
primals_3, buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
del primals_3
return buf0, primals_2
class AffineNew(nn.Module):
def __init__(self, channel):
super().__init__()
self.g = nn.Parameter(torch.ones(1, 1, channel))
self.b = nn.Parameter(torch.zeros(1, 1, channel))
def forward(self, input_0):
primals_1 = self.g
primals_3 = self.b
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | Affine | false | 4,209 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, channel):
super().__init__()
self.g = nn.Parameter(torch.ones(1, 1, channel))
self.b = nn.Parameter(torch.zeros(1, 1, channel))
def forward(self, x):
return x * self.g + self.b
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
ECAAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/l3/cl35tzbhrd24dhunkbb6gjs54aklpyr46oikqhoylcgmkcmhujil.py
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# y => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [-1, -2], True), kwargs = {})
triton_per_fused_mean_0 = async_compile.triton('triton_per_fused_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[16, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mean_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/it/citxitwi5k5revcoaspxziwrq6kifyromw64awswainul3shsed3.py
# Topologically Sorted Source Nodes: [y_2], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# y_2 => convolution
# Graph fragment:
# %convolution : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%permute, %primals_2, %primals_3, [1], [1], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tl.store(in_out_ptr0 + (x0), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xv/cxvgsfj3x2o5ls6evsy4rhywutbtjkwezlavric3plphgvn75mea.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %expand), kwargs = {})
triton_poi_fused_mul_2 = async_compile.triton('triton_poi_fused_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 1, 3), (3, 3, 1))
assert_size_stride(primals_3, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 16, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [y_2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (4, 1, 4), (4, 0, 1), 0), primals_2, stride=(1,), padding=(1,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 4), (4, 4, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [y_2], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf3, primals_3, 16, grid=grid(16), stream=stream0)
del primals_3
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(primals_1, buf3, buf4, 256, grid=grid(256), stream=stream0)
return (buf4, primals_1, primals_2, reinterpret_tensor(buf1, (4, 1, 4), (4, 1, 1), 0), buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 1, 3), (3, 3, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
from torch.nn import init
class ECAAttention(nn.Module):
def __init__(self, kernel_size=3):
super().__init__()
self.gap = nn.AdaptiveAvgPool2d(1)
self.conv = nn.Conv1d(1, 1, kernel_size=kernel_size, padding=(
kernel_size - 1) // 2)
self.sigmoid = nn.Sigmoid()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, x):
y = self.gap(x)
y = y.squeeze(-1).permute(0, 2, 1)
y = self.conv(y)
y = self.sigmoid(y)
y = y.permute(0, 2, 1).unsqueeze(-1)
return x * y.expand_as(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tl.store(in_out_ptr0 + x0, tmp3, xmask)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 16
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 1, 3), (3, 3, 1))
assert_size_stride(primals_3, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(16)](buf1, primals_1, 16, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (4, 1, 4
), (4, 0, 1), 0), primals_2, stride=(1,), padding=(1,),
dilation=(1,), transposed=False, output_padding=(0,), groups=1,
bias=None)
assert_size_stride(buf2, (4, 1, 4), (4, 4, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_1[grid(16)](buf3, primals_3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_2[grid(256)](primals_1, buf3, buf4, 256,
XBLOCK=128, num_warps=4, num_stages=1)
return buf4, primals_1, primals_2, reinterpret_tensor(buf1, (4, 1, 4),
(4, 1, 1), 0), buf3
class ECAAttentionNew(nn.Module):
def __init__(self, kernel_size=3):
super().__init__()
self.gap = nn.AdaptiveAvgPool2d(1)
self.conv = nn.Conv1d(1, 1, kernel_size=kernel_size, padding=(
kernel_size - 1) // 2)
self.sigmoid = nn.Sigmoid()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, input_0):
primals_2 = self.conv.weight
primals_3 = self.conv.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | ECAAttention | false | 4,210 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
from torch.nn import init
class Model(nn.Module):
def __init__(self, kernel_size=3):
super().__init__()
self.gap = nn.AdaptiveAvgPool2d(1)
self.conv = nn.Conv1d(1, 1, kernel_size=kernel_size, padding=(
kernel_size - 1) // 2)
self.sigmoid = nn.Sigmoid()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, x):
y = self.gap(x)
y = y.squeeze(-1).permute(0, 2, 1)
y = self.conv(y)
y = self.sigmoid(y)
y = y.permute(0, 2, 1).unsqueeze(-1)
return x * y.expand_as(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GTXAttentionOutput | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/cl/cclwzssgcha5xquex5qki6klim6hfpecfb3mn3wjv6nq7ppewan7.py
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.add]
# Source node to ATen node mapping:
# hidden_states_2 => add
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_4, %view_1), kwargs = {})
triton_poi_fused_add_0 = async_compile.triton('triton_poi_fused_add_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_out_ptr0 + (x2), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nk/cnkbkukjfarsysqlaadkg24xmqibk3adq5p7jyfnt6k6loydbn2r.py
# Topologically Sorted Source Nodes: [hidden_states_3], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_3 => add_1, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [3]), kwargs = {correction: 0, keepdim: True})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
triton_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-12
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + (x0), tmp8, xmask)
tl.store(out_ptr1 + (x0), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/mn/cmntyljhuirhsdjg2yosgzllpkpxqedxgoyk6gunquq2rf3kl7u5.py
# Topologically Sorted Source Nodes: [hidden_states_3], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_3 => add_1, add_2, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [3]), kwargs = {correction: 0, keepdim: True})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_5), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_6), kwargs = {})
triton_poi_fused_native_layer_norm_2 = async_compile.triton('triton_poi_fused_native_layer_norm_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(buf1, primals_4, primals_2, 256, grid=grid(256), stream=stream0)
del primals_2
del primals_4
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(buf1, buf2, buf3, 64, grid=grid(64), stream=stream0)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_2.run(buf1, buf2, buf3, primals_5, primals_6, buf4, 256, grid=grid(256), stream=stream0)
del buf2
del buf3
del primals_6
return (buf4, primals_5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import torch
from torch import nn
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, hidden_dropout_prob=
0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_out_ptr0 + x2, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-12
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp23, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](buf1, primals_4, primals_2, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
del primals_4
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(64)](buf1, buf2, buf3, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_2[grid(256)](buf1, buf2, buf3,
primals_5, primals_6, buf4, 256, XBLOCK=256, num_warps=4,
num_stages=1)
del buf2
del buf3
del primals_6
return buf4, primals_5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1
class GTXAttentionOutputNew(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, input_0, input_1):
primals_1 = self.dense.weight
primals_2 = self.dense.bias
primals_5 = self.LayerNorm.weight
primals_6 = self.LayerNorm.bias
primals_3 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| rsgit95/med_kg_txt_multimodal | GTXAttentionOutput | false | 4,211 | [
"Apache-2.0"
] | 0 | 80355b0cf58e0571531ad6f9728c533110ca996d | https://github.com/rsgit95/med_kg_txt_multimodal/tree/80355b0cf58e0571531ad6f9728c533110ca996d | from _paritybench_helpers import _mock_config
import torch
from torch import nn
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, hidden_dropout_prob=
0.5)}]
|
Actor | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ix/cixxyusyg44s2hkoufcgbrv3ix5ookwqjl4ia3xkv7bdqi4yrzus.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 25600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 400
x2 = xindex % 1600
x3 = (xindex // 1600)
tmp0 = tl.load(in_out_ptr0 + (x4), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x4), tmp4, xmask)
tl.store(out_ptr0 + (x2 + (1664*x3)), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ld/cld3palaxnfc7osbafgwkvp6zk52xgffwvw26mvcpxisjgbhf4qn.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_1 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 12800
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 200
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ld/cldtk5skh6gtzdd62vyilgjgd55ch7o62ebbhqgbpau5cmhd5sca.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_2 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/7v/c7vy54xspuettz5pgulxporznj2yqlyufnh2o2cvg7er4bnu4zox.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_2 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (400, 4), (4, 1))
assert_size_stride(primals_2, (400, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (200, 400), (400, 1))
assert_size_stride(primals_5, (200, ), (1, ))
assert_size_stride(primals_6, (4, 200), (200, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 400), (400, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 400), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 400), (6400, 1600, 400, 1), 0); del buf0 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 400), (6656, 1664, 400, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf8, 25600, grid=grid(25600), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 200), (200, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 400), (400, 1), 0), reinterpret_tensor(primals_4, (400, 200), (1, 400), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 200), (3200, 800, 200, 1), 0); del buf2 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 200), (3200, 800, 200, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf7, 12800, grid=grid(12800), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 200), (200, 1), 0), reinterpret_tensor(primals_6, (200, 4), (1, 200), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
return (buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 400), (400, 1), 0), reinterpret_tensor(buf3, (64, 200), (200, 1), 0), buf6, primals_6, buf7, primals_4, buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((400, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((400, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((200, 400), (400, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((200, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 200), (200, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class Actor(nn.Module):
def __init__(self, state_dim, action_dim):
super(Actor, self).__init__()
self.l1 = nn.Linear(state_dim, 400)
self.l2 = nn.Linear(400, 200)
self.l3 = nn.Linear(200, action_dim)
def forward(self, x):
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = nn.Softmax()(self.l3(x))
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4, 'action_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 25600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 400
x2 = xindex % 1600
x3 = xindex // 1600
tmp0 = tl.load(in_out_ptr0 + x4, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x4, tmp4, xmask)
tl.store(out_ptr0 + (x2 + 1664 * x3), tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 12800
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 200
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (400, 4), (4, 1))
assert_size_stride(primals_2, (400,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (200, 400), (400, 1))
assert_size_stride(primals_5, (200,), (1,))
assert_size_stride(primals_6, (4, 200), (200, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 400), (400, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 400), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 400), (6400, 1600, 400, 1), 0
)
del buf0
buf8 = empty_strided_cuda((4, 4, 4, 400), (6656, 1664, 400, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(25600)](buf1,
primals_2, buf8, 25600, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 200), (200, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 400), (400, 1), 0),
reinterpret_tensor(primals_4, (400, 200), (1, 400), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 200), (3200, 800, 200, 1), 0)
del buf2
buf7 = empty_strided_cuda((4, 4, 4, 200), (3200, 800, 200, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(12800)](buf3,
primals_5, buf7, 12800, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 200),
(200, 1), 0), reinterpret_tensor(primals_6, (200, 4), (1, 200),
0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(256)](buf4, buf5, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf4
triton_poi_fused__softmax_3[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
return buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 400), (400, 1), 0
), reinterpret_tensor(buf3, (64, 200), (200, 1), 0
), buf6, primals_6, buf7, primals_4, buf8
class ActorNew(nn.Module):
def __init__(self, state_dim, action_dim):
super(ActorNew, self).__init__()
self.l1 = nn.Linear(state_dim, 400)
self.l2 = nn.Linear(400, 200)
self.l3 = nn.Linear(200, action_dim)
def forward(self, input_0):
primals_1 = self.l1.weight
primals_2 = self.l1.bias
primals_4 = self.l2.weight
primals_5 = self.l2.bias
primals_6 = self.l3.weight
primals_7 = self.l3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| rortiz9/meleeml | Actor | false | 4,212 | [
"MIT"
] | 0 | 9be4bf53a377dfb46dbb3b51f102f1bffc0124d2 | https://github.com/rortiz9/meleeml/tree/9be4bf53a377dfb46dbb3b51f102f1bffc0124d2 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, state_dim, action_dim):
super().__init__()
self.l1 = nn.Linear(state_dim, 400)
self.l2 = nn.Linear(400, 200)
self.l3 = nn.Linear(200, action_dim)
def forward(self, x):
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = nn.Softmax()(self.l3(x))
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
PolicyNetwork | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/r3/cr3febcwm3t44fuoitsx3ou2p6xg4sk4f7unagmmrvffasxf47te.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_2 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/rm/crmfikkxblrhxfynyknfm2x3wwcwtibkjkkbyhzwmxqi4kmwkosl.py
# Topologically Sorted Source Nodes: [log_std_1], Original ATen: [aten.clamp, aten.ge, aten.le, aten.logical_and]
# Source node to ATen node mapping:
# log_std_1 => clamp_max, clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%view_7, -20), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 2), kwargs = {})
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%view_7, -20), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%view_7, 2), kwargs = {})
# %logical_and : [num_users=1] = call_function[target=torch.ops.aten.logical_and.default](args = (%ge, %le), kwargs = {})
triton_poi_fused_clamp_ge_le_logical_and_1 = async_compile.triton('triton_poi_fused_clamp_ge_le_logical_and_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_ge_le_logical_and_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clamp_ge_le_logical_and_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = -20.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = 2.0
tmp6 = triton_helpers.minimum(tmp4, tmp5)
tmp7 = tmp2 >= tmp3
tmp8 = tmp2 <= tmp5
tmp9 = tmp7 & tmp8
tl.store(out_ptr0 + (x2), tmp6, xmask)
tl.store(out_ptr1 + (x2), tmp9, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf9, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf8, 256, grid=grid(256), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [log_std_1], Original ATen: [aten.clamp, aten.ge, aten.le, aten.logical_and]
triton_poi_fused_clamp_ge_le_logical_and_1.run(buf5, primals_9, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf5
del primals_9
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(buf3, (64, 4), (4, 1), 0), buf7, primals_8, primals_6, buf8, primals_4, buf9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
from torch.nn import functional as F
from torch.distributions import Normal
class PolicyNetwork(nn.Module):
def __init__(self, state_dim, action_dim, hidden_dim, init_w=0.003,
log_std_min=-20, log_std_max=2):
super(PolicyNetwork, self).__init__()
self.log_std_min = log_std_min
self.log_std_max = log_std_max
self.linear1 = nn.Linear(state_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
self.mean_linear = nn.Linear(hidden_dim, action_dim)
self.mean_linear.weight.data.uniform_(-init_w, init_w)
self.mean_linear.bias.data.uniform_(-init_w, init_w)
self.log_std_linear = nn.Linear(hidden_dim, action_dim)
self.log_std_linear.weight.data.uniform_(-init_w, init_w)
self.log_std_linear.bias.data.uniform_(-init_w, init_w)
def forward(self, state):
x = F.relu(self.linear1(state))
x = F.relu(self.linear2(x))
mean = self.mean_linear(x)
log_std = self.log_std_linear(x)
log_std = torch.clamp(log_std, self.log_std_min, self.log_std_max)
return mean, log_std
def evaluate(self, state, epsilon=1e-06):
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(0, 1)
z = normal.sample()
action = torch.tanh(mean + std * z)
log_prob = Normal(mean, std).log_prob(mean + std * z) - torch.log(1 -
action.pow(2) + epsilon)
return action, log_prob, z, mean, log_std
def get_action(self, state):
state = torch.FloatTensor(state).unsqueeze(0)
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
return action[0]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4, 'action_dim': 4, 'hidden_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
from torch.distributions import Normal
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
@triton.jit
def triton_poi_fused_clamp_ge_le_logical_and_1(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = -20.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = 2.0
tmp6 = triton_helpers.minimum(tmp4, tmp5)
tmp7 = tmp2 >= tmp3
tmp8 = tmp2 <= tmp5
tmp9 = tmp7 & tmp8
tl.store(out_ptr0 + x2, tmp6, xmask)
tl.store(out_ptr1 + x2, tmp9, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf1,
primals_2, buf9, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf3,
primals_5, buf8, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_clamp_ge_le_logical_and_1[grid(256)](buf5,
primals_9, buf6, buf7, 256, XBLOCK=128, num_warps=4, num_stages=1)
del buf5
del primals_9
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(
buf3, (64, 4), (4, 1), 0
), buf7, primals_8, primals_6, buf8, primals_4, buf9
class PolicyNetworkNew(nn.Module):
def __init__(self, state_dim, action_dim, hidden_dim, init_w=0.003,
log_std_min=-20, log_std_max=2):
super(PolicyNetworkNew, self).__init__()
self.log_std_min = log_std_min
self.log_std_max = log_std_max
self.linear1 = nn.Linear(state_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
self.mean_linear = nn.Linear(hidden_dim, action_dim)
self.mean_linear.weight.data.uniform_(-init_w, init_w)
self.mean_linear.bias.data.uniform_(-init_w, init_w)
self.log_std_linear = nn.Linear(hidden_dim, action_dim)
self.log_std_linear.weight.data.uniform_(-init_w, init_w)
self.log_std_linear.bias.data.uniform_(-init_w, init_w)
def evaluate(self, state, epsilon=1e-06):
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(0, 1)
z = normal.sample()
action = torch.tanh(mean + std * z)
log_prob = Normal(mean, std).log_prob(mean + std * z) - torch.log(1 -
action.pow(2) + epsilon)
return action, log_prob, z, mean, log_std
def get_action(self, state):
state = torch.FloatTensor(state).unsqueeze(0)
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
return action[0]
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_6 = self.mean_linear.weight
primals_7 = self.mean_linear.bias
primals_8 = self.log_std_linear.weight
primals_9 = self.log_std_linear.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0], output[1]
| rtharungowda/Soft-Actor-Critic-Pytorch | PolicyNetwork | false | 4,213 | [
"MIT"
] | 0 | 0d2c20c6cfd4e578e0b7cff4525ddf0bc956812f | https://github.com/rtharungowda/Soft-Actor-Critic-Pytorch/tree/0d2c20c6cfd4e578e0b7cff4525ddf0bc956812f | import torch
import torch.nn as nn
from torch.nn import functional as F
from torch.distributions import Normal
class Model(nn.Module):
def __init__(self, state_dim, action_dim, hidden_dim, init_w=0.003,
log_std_min=-20, log_std_max=2):
super().__init__()
self.log_std_min = log_std_min
self.log_std_max = log_std_max
self.linear1 = nn.Linear(state_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
self.mean_linear = nn.Linear(hidden_dim, action_dim)
self.mean_linear.weight.data.uniform_(-init_w, init_w)
self.mean_linear.bias.data.uniform_(-init_w, init_w)
self.log_std_linear = nn.Linear(hidden_dim, action_dim)
self.log_std_linear.weight.data.uniform_(-init_w, init_w)
self.log_std_linear.bias.data.uniform_(-init_w, init_w)
def forward(self, state):
x = F.relu(self.linear1(state))
x = F.relu(self.linear2(x))
mean = self.mean_linear(x)
log_std = self.log_std_linear(x)
log_std = torch.clamp(log_std, self.log_std_min, self.log_std_max)
return mean, log_std
def evaluate(self, state, epsilon=1e-06):
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(0, 1)
z = normal.sample()
action = torch.tanh(mean + std * z)
log_prob = Normal(mean, std).log_prob(mean + std * z) - torch.log(1 -
action.pow(2) + epsilon)
return action, log_prob, z, mean, log_std
def get_action(self, state):
state = torch.FloatTensor(state).unsqueeze(0)
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
return action[0]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
Depth_Pointwise_Conv1d | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/o6/co6pflndmsdhmqwe2jfrf4itwvl27ku5p27kydz44oxklfdvmyvc.py
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv1d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%unsqueeze, %primals_1, %primals_2, [1], [2], [1], False, [0], 4), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 20
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 5)
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 1, 4), (4, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(reinterpret_tensor(primals_3, (1, 4, 4), (16, 4, 1), 0), primals_1, stride=(1,), padding=(2,), dilation=(1,), transposed=False, output_padding=(0,), groups=4, bias=None)
assert_size_stride(buf0, (1, 4, 5), (20, 5, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_2, 20, grid=grid(20), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 5), (0, 5, 1), 0), primals_4, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (1, 4, 5), (20, 5, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
triton_poi_fused_convolution_0.run(buf3, primals_5, 20, grid=grid(20), stream=stream0)
del primals_5
return (reinterpret_tensor(buf3, (4, 5), (5, 1), 0), primals_1, primals_4, reinterpret_tensor(primals_3, (1, 4, 4), (16, 4, 1), 0), buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 1, 4), (4, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class Depth_Pointwise_Conv1d(nn.Module):
def __init__(self, in_ch, out_ch, k):
super().__init__()
if k == 1:
self.depth_conv = nn.Identity()
else:
self.depth_conv = nn.Conv1d(in_channels=in_ch, out_channels=
in_ch, kernel_size=k, groups=in_ch, padding=k // 2)
self.pointwise_conv = nn.Conv1d(in_channels=in_ch, out_channels=
out_ch, kernel_size=1, groups=1)
def forward(self, x):
out = self.pointwise_conv(self.depth_conv(x))
return out
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'in_ch': 4, 'out_ch': 4, 'k': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 20
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 5
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 1, 4), (4, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(reinterpret_tensor(primals_3, (1,
4, 4), (16, 4, 1), 0), primals_1, stride=(1,), padding=(2,),
dilation=(1,), transposed=False, output_padding=(0,), groups=4,
bias=None)
assert_size_stride(buf0, (1, 4, 5), (20, 5, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(20)](buf1, primals_2, 20,
XBLOCK=32, num_warps=1, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 5
), (0, 5, 1), 0), primals_4, stride=(1,), padding=(0,),
dilation=(1,), transposed=False, output_padding=(0,), groups=1,
bias=None)
assert_size_stride(buf2, (1, 4, 5), (20, 5, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_0[grid(20)](buf3, primals_5, 20,
XBLOCK=32, num_warps=1, num_stages=1)
del primals_5
return reinterpret_tensor(buf3, (4, 5), (5, 1), 0
), primals_1, primals_4, reinterpret_tensor(primals_3, (1, 4, 4), (
16, 4, 1), 0), buf1
class Depth_Pointwise_Conv1dNew(nn.Module):
def __init__(self, in_ch, out_ch, k):
super().__init__()
if k == 1:
self.depth_conv = nn.Identity()
else:
self.depth_conv = nn.Conv1d(in_channels=in_ch, out_channels=
in_ch, kernel_size=k, groups=in_ch, padding=k // 2)
self.pointwise_conv = nn.Conv1d(in_channels=in_ch, out_channels=
out_ch, kernel_size=1, groups=1)
def forward(self, input_0):
primals_1 = self.depth_conv.weight
primals_2 = self.depth_conv.bias
primals_4 = self.pointwise_conv.weight
primals_5 = self.pointwise_conv.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | Depth_Pointwise_Conv1d | false | 4,214 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, in_ch, out_ch, k):
super().__init__()
if k == 1:
self.depth_conv = nn.Identity()
else:
self.depth_conv = nn.Conv1d(in_channels=in_ch, out_channels=
in_ch, kernel_size=k, groups=in_ch, padding=k // 2)
self.pointwise_conv = nn.Conv1d(in_channels=in_ch, out_channels=
out_ch, kernel_size=1, groups=1)
def forward(self, x):
out = self.pointwise_conv(self.depth_conv(x))
return out
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
ExternalAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/5q/c5qeh3pdyggwxnuvejg2thm5cuxzihqbmcufbsbx7b7rowvqqtu3.py
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x0 = xindex % 256
x2 = (xindex // 1024)
tmp0 = tl.load(in_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (256 + x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (512 + x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (768 + x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/g3/cg3hw563jssel4qcvpxgu54b5u5vs56ekmydhabf4bw4vpcp55xv.py
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_1 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x0 = xindex % 256
x2 = (xindex // 1024)
tmp0 = tl.load(in_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (256 + x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (512 + x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (768 + x0 + (1024*x2)), None, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6r/c6ref7by2nkxvxeoorqx3pmp6kdvasadwz73xoow3yx5up6n7fv4.py
# Topologically Sorted Source Nodes: [sum_1, attn_2], Original ATen: [aten.sum, aten.div]
# Source node to ATen node mapping:
# attn_2 => div_1
# sum_1 => sum_2
# Graph fragment:
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%div, [2], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%div, %sum_2), kwargs = {})
triton_poi_fused_div_sum_2 = async_compile.triton('triton_poi_fused_div_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_div_sum_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x0 = xindex % 64
x2 = (xindex // 256)
tmp0 = tl.load(in_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x0 + (256*x2)), None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (64 + x0 + (256*x2)), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (128 + x0 + (256*x2)), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (192 + x0 + (256*x2)), None, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 64), (64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf0, buf1, 4096, grid=grid(4096), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 4096, grid=grid(4096), stream=stream0)
buf3 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [sum_1, attn_2], Original ATen: [aten.sum, aten.div]
triton_poi_fused_div_sum_2.run(buf2, buf3, 4096, grid=grid(4096), stream=stream0)
del buf2
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(primals_3, (64, 4), (1, 64), 0), out=buf4)
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_2, (64, 4), (4, 1), 0), buf0, buf3, primals_3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 64), (64, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
from torch.nn import init
class ExternalAttention(nn.Module):
def __init__(self, d_model, S=64):
super().__init__()
self.mk = nn.Linear(d_model, S, bias=False)
self.mv = nn.Linear(S, d_model, bias=False)
self.softmax = nn.Softmax(dim=1)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries):
attn = self.mk(queries)
attn = self.softmax(attn)
attn = attn / torch.sum(attn, dim=2, keepdim=True)
out = self.mv(attn)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_model': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x0 = xindex % 256
x2 = xindex // 1024
tmp0 = tl.load(in_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + (x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (256 + x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (512 + x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (768 + x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, None)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x0 = xindex % 256
x2 = xindex // 1024
tmp0 = tl.load(in_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + (x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (256 + x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (512 + x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (768 + x0 + 1024 * x2), None, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, None)
@triton.jit
def triton_poi_fused_div_sum_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x0 = xindex % 64
x2 = xindex // 256
tmp0 = tl.load(in_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + (x0 + 256 * x2), None, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (64 + x0 + 256 * x2), None, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (128 + x0 + 256 * x2), None, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (192 + x0 + 256 * x2), None, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, None)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 64), (64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_2, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.
float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(4096)](buf0, buf1, 4096, XBLOCK=
256, num_warps=4, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.
float32)
triton_poi_fused__softmax_1[grid(4096)](buf1, buf2, 4096, XBLOCK=
128, num_warps=4, num_stages=1)
buf3 = buf1
del buf1
triton_poi_fused_div_sum_2[grid(4096)](buf2, buf3, 4096, XBLOCK=256,
num_warps=4, num_stages=1)
del buf2
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 64), (64, 1), 0),
reinterpret_tensor(primals_3, (64, 4), (1, 64), 0), out=buf4)
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_2, (64, 4), (4, 1), 0
), buf0, buf3, primals_3
class ExternalAttentionNew(nn.Module):
def __init__(self, d_model, S=64):
super().__init__()
self.mk = nn.Linear(d_model, S, bias=False)
self.mv = nn.Linear(S, d_model, bias=False)
self.softmax = nn.Softmax(dim=1)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, input_0):
primals_1 = self.mk.weight
primals_3 = self.mv.weight
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | ExternalAttention | false | 4,215 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
from torch.nn import init
class Model(nn.Module):
def __init__(self, d_model, S=64):
super().__init__()
self.mk = nn.Linear(d_model, S, bias=False)
self.mv = nn.Linear(S, d_model, bias=False)
self.softmax = nn.Softmax(dim=1)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries):
attn = self.mk(queries)
attn = self.softmax(attn)
attn = attn / torch.sum(attn, dim=2, keepdim=True)
out = self.mv(attn)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
SpatialAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/uc/cucdaa5tqnxykdmw5yqh7ir5ac35phopjcobljrg4rrtlnfjtuwd.py
# Topologically Sorted Source Nodes: [result], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# result => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%getitem, %mean], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 16) % 2
x0 = xindex % 16
x2 = (xindex // 32)
x3 = xindex
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp9 = triton_helpers.maximum(tmp7, tmp8)
tmp10 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = triton_helpers.maximum(tmp9, tmp10)
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tmp15 = tl.full([1], 2, tl.int64)
tmp16 = tmp0 < tmp15
tmp17 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp18 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp19 = tmp17 + tmp18
tmp20 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp21 = tmp19 + tmp20
tmp22 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp23 = tmp21 + tmp22
tmp24 = 4.0
tmp25 = tmp23 / tmp24
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp14, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp13, tmp27)
tl.store(out_ptr0 + (x3), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/go/cgofqcgduqrtcjakfd7uk3wkcrpwsqxispluihwsstry6ekodk2u.py
# Topologically Sorted Source Nodes: [output, output_1], Original ATen: [aten.convolution, aten.sigmoid]
# Source node to ATen node mapping:
# output => convolution
# output_1 => sigmoid
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%cat, %primals_2, %primals_3, [1, 1], [3, 3], [1, 1], False, [0, 0], 1), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_sigmoid_1 = async_compile.triton('triton_poi_fused_convolution_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_sigmoid_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 7, 7), (98, 49, 7, 1))
assert_size_stride(primals_3, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [result], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, buf0, 128, grid=grid(128), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1), padding=(3, 3), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [output, output_1], Original ATen: [aten.convolution, aten.sigmoid]
triton_poi_fused_convolution_sigmoid_1.run(buf2, primals_3, 64, grid=grid(64), stream=stream0)
del primals_3
return (buf2, primals_2, buf0, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 2, 7, 7), (98, 49, 7, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class SpatialAttention(nn.Module):
def __init__(self, kernel_size=7):
super().__init__()
self.conv = nn.Conv2d(2, 1, kernel_size=kernel_size, padding=
kernel_size // 2)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
max_result, _ = torch.max(x, dim=1, keepdim=True)
avg_result = torch.mean(x, dim=1, keepdim=True)
result = torch.cat([max_result, avg_result], 1)
output = self.conv(result)
output = self.sigmoid(output)
return output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 16 % 2
x0 = xindex % 16
x2 = xindex // 32
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp9 = triton_helpers.maximum(tmp7, tmp8)
tmp10 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp11 = triton_helpers.maximum(tmp9, tmp10)
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp17 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp18 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp19 = tmp17 + tmp18
tmp20 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp21 = tmp19 + tmp20
tmp22 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp23 = tmp21 + tmp22
tmp24 = 4.0
tmp25 = tmp23 / tmp24
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp14, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp13, tmp27)
tl.store(out_ptr0 + x3, tmp28, xmask)
@triton.jit
def triton_poi_fused_convolution_sigmoid_1(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 7, 7), (98, 49, 7, 1))
assert_size_stride(primals_3, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(128)](primals_1, buf0, 128, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1),
padding=(3, 3), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = buf1
del buf1
triton_poi_fused_convolution_sigmoid_1[grid(64)](buf2, primals_3,
64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_3
return buf2, primals_2, buf0, buf2
class SpatialAttentionNew(nn.Module):
def __init__(self, kernel_size=7):
super().__init__()
self.conv = nn.Conv2d(2, 1, kernel_size=kernel_size, padding=
kernel_size // 2)
self.sigmoid = nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.conv.weight
primals_3 = self.conv.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | SpatialAttention | false | 4,216 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, kernel_size=7):
super().__init__()
self.conv = nn.Conv2d(2, 1, kernel_size=kernel_size, padding=
kernel_size // 2)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
max_result, _ = torch.max(x, dim=1, keepdim=True)
avg_result = torch.mean(x, dim=1, keepdim=True)
result = torch.cat([max_result, avg_result], 1)
output = self.conv(result)
output = self.sigmoid(output)
return output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GTXSelfAttentionLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/x2/cx2hdvwyo7m5jvhhvtugzxqvmy6z4nsfhkkjhvgzbbm3cb6dsum2.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %mul_scalar : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%permute_default, 1.0), kwargs = {})
# %clone_default : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/iz/ciztqj6kop3hxov46yrmzprkzfir3eljcic4mkqznz2j5cfeaudr.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %add_tensor : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_default_2, %primals_8), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_tensor, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
# %sum_dim_int_list : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_default, [-1], True), kwargs = {})
# %eq_scalar : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%add_tensor, -inf), kwargs = {})
# %logical_not_default : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%eq_scalar,), kwargs = {})
# %any_dim : [num_users=1] = call_function[target=torch.ops.aten.any.dim](args = (%logical_not_default, -1, True), kwargs = {})
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = triton_helpers.maximum(tmp2, tmp5)
tmp9 = tmp7 + tmp8
tmp10 = triton_helpers.maximum(tmp6, tmp9)
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp10, tmp13)
tmp15 = tmp2 - tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tmp5 - tmp14
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp9 - tmp14
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp23 = tmp13 - tmp14
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = float("-inf")
tmp27 = tmp2 == tmp26
tmp28 = tmp27 == 0
tmp29 = tmp28.to(tl.int64)
tmp30 = (tmp29 != 0)
tmp31 = tmp5 == tmp26
tmp32 = tmp31 == 0
tmp33 = tmp32.to(tl.int64)
tmp34 = (tmp33 != 0)
tmp35 = tmp30 | tmp34
tmp36 = tmp9 == tmp26
tmp37 = tmp36 == 0
tmp38 = tmp37.to(tl.int64)
tmp39 = (tmp38 != 0)
tmp40 = tmp35 | tmp39
tmp41 = tmp13 == tmp26
tmp42 = tmp41 == 0
tmp43 = tmp42.to(tl.int64)
tmp44 = (tmp43 != 0)
tmp45 = tmp40 | tmp44
tl.store(out_ptr0 + (x2), tmp14, xmask)
tl.store(out_ptr1 + (x2), tmp25, xmask)
tl.store(out_ptr2 + (x2), tmp45, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/x5/cx5uvbfethxuwwkwxf3xaualzhlcwqsz4jxqpbhintggaypzjwqf.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %add_tensor : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_default_2, %primals_8), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_tensor, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_default, %sum_dim_int_list), kwargs = {})
# %logical_not_default_1 : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%any_dim,), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%logical_not_default_1, %full_default, %div_tensor), kwargs = {})
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i1', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = (xindex // 4)
x4 = xindex
x5 = xindex % 64
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last').to(tl.int1)
tmp2 = tl.load(in_out_ptr0 + (x4), xmask)
tmp3 = tl.load(in_ptr1 + (x5), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr3 + (x3), xmask, eviction_policy='evict_last')
tmp1 = tmp0 == 0
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tmp10 = 0.0
tmp11 = tl.where(tmp1, tmp10, tmp9)
tl.store(in_out_ptr0 + (x4), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vv/cvvnhithjvmvhfjufxwwzclfobkrgbyyteg66hp24r675f7elw4c.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %clone_default_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default_3,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6t/c6t5a5ere3lqjiu7zh3uu4oxmpdoujdaqqmeunxqapgzo4m74uav.py
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# context_layer_1 => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6m/c6mhj5zwirfhy5e4o45uaeov72uwfby4udubpm2fcz42iqvs2g57.py
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_2 => add_1
# hidden_states_3 => var_mean
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %view_17), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_1, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/i6/ci6ua4lfqzz3v6lbsh75noa7k5ird3udb6b5bjh7gxx4qxuz7gz3.py
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_2 => add_1
# hidden_states_3 => add_2, add_3, mul, mul_1, rsqrt, sub_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %view_17), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_2,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_1, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_11), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_12), kwargs = {})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(buf0, primals_2, buf3, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_0.run(buf1, primals_5, buf4, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 64), 0); del buf1 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf8 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.bool)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(buf5, primals_8, buf6, buf7, buf8, 64, grid=grid(64), stream=stream0)
buf9 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(buf9, buf8, primals_8, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf8
del primals_8
buf10 = reinterpret_tensor(buf7, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(buf2, primals_7, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_7
buf11 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf9, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf10, (16, 4, 1), (4, 1, 0), 0), out=buf11)
buf12 = reinterpret_tensor(buf6, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf11, buf12, 16, 4, grid=grid(16, 4), stream=stream0)
buf13 = reinterpret_tensor(buf11, (16, 4), (4, 1), 0); del buf11 # reuse
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf12, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf13)
del primals_10
buf14 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf15 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(primals_3, buf13, buf14, buf15, 16, grid=grid(16), stream=stream0)
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(primals_3, buf13, buf14, buf15, primals_11, primals_12, buf16, 64, grid=grid(64), stream=stream0)
del buf14
del buf15
del primals_12
return (buf16, primals_3, primals_11, buf9, reinterpret_tensor(buf10, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0), reinterpret_tensor(buf12, (16, 4), (4, 1), 0), buf13, primals_9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class GTXAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class GTXSelfAttentionLayer(nn.Module):
def __init__(self, config):
super().__init__()
self.self = GTXAttention(config)
self.output = GTXAttentionOutput(config)
def forward(self, input_tensor, attention_mask, output_attentions=False):
output = self.self(input_tensor, input_tensor, attention_mask,
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
attention_output = self.output(output[0], input_tensor)
outputs = (attention_output, attention_probs
) if output_attentions else (attention_output,)
return outputs
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, hidden_dropout_prob=0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = triton_helpers.maximum(tmp2, tmp5)
tmp9 = tmp7 + tmp8
tmp10 = triton_helpers.maximum(tmp6, tmp9)
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp10, tmp13)
tmp15 = tmp2 - tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tmp5 - tmp14
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp9 - tmp14
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp23 = tmp13 - tmp14
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = float('-inf')
tmp27 = tmp2 == tmp26
tmp28 = tmp27 == 0
tmp29 = tmp28.to(tl.int64)
tmp30 = tmp29 != 0
tmp31 = tmp5 == tmp26
tmp32 = tmp31 == 0
tmp33 = tmp32.to(tl.int64)
tmp34 = tmp33 != 0
tmp35 = tmp30 | tmp34
tmp36 = tmp9 == tmp26
tmp37 = tmp36 == 0
tmp38 = tmp37.to(tl.int64)
tmp39 = tmp38 != 0
tmp40 = tmp35 | tmp39
tmp41 = tmp13 == tmp26
tmp42 = tmp41 == 0
tmp43 = tmp42.to(tl.int64)
tmp44 = tmp43 != 0
tmp45 = tmp40 | tmp44
tl.store(out_ptr0 + x2, tmp14, xmask)
tl.store(out_ptr1 + x2, tmp25, xmask)
tl.store(out_ptr2 + x2, tmp45, xmask)
@triton.jit
def triton_poi_fused_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex // 4
x4 = xindex
x5 = xindex % 64
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp2 = tl.load(in_out_ptr0 + x4, xmask)
tmp3 = tl.load(in_ptr1 + x5, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr3 + x3, xmask, eviction_policy='evict_last')
tmp1 = tmp0 == 0
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tmp10 = 0.0
tmp11 = tl.where(tmp1, tmp10, tmp9)
tl.store(in_out_ptr0 + x4, tmp11, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(16, 4)](buf0, primals_2, buf3, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0)
del buf0
triton_poi_fused_0[grid(16, 4)](buf1, primals_5, buf4, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 64), 0)
del buf1
buf7 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf8 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.bool)
triton_poi_fused_1[grid(64)](buf5, primals_8, buf6, buf7, buf8, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf9 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused_2[grid(256)](buf9, buf8, primals_8, buf6, buf7,
256, XBLOCK=128, num_warps=4, num_stages=1)
del buf8
del primals_8
buf10 = reinterpret_tensor(buf7, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf7
triton_poi_fused_3[grid(16, 4)](buf2, primals_7, buf10, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_7
buf11 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf9, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf10, (16, 4, 1), (4, 1, 0), 0), out=buf11)
buf12 = reinterpret_tensor(buf6, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf6
triton_poi_fused_clone_4[grid(16, 4)](buf11, buf12, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf13 = reinterpret_tensor(buf11, (16, 4), (4, 1), 0)
del buf11
extern_kernels.addmm(primals_10, reinterpret_tensor(buf12, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf13)
del primals_10
buf14 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf15 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](primals_3, buf13,
buf14, buf15, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(64)](primals_3, buf13,
buf14, buf15, primals_11, primals_12, buf16, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf14
del buf15
del primals_12
return buf16, primals_3, primals_11, buf9, reinterpret_tensor(buf10, (
16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4,
1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0
), reinterpret_tensor(buf12, (16, 4), (4, 1), 0), buf13, primals_9
class GTXAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class GTXSelfAttentionLayerNew(nn.Module):
def __init__(self, config):
super().__init__()
self.self = GTXAttention(config)
self.output = GTXAttentionOutput(config)
def forward(self, input_0, input_1):
primals_1 = self.self.query.weight
primals_2 = self.self.query.bias
primals_4 = self.self.key.weight
primals_5 = self.self.key.bias
primals_6 = self.self.value.weight
primals_7 = self.self.value.bias
primals_9 = self.output.dense.weight
primals_10 = self.output.dense.bias
primals_11 = self.output.LayerNorm.weight
primals_12 = self.output.LayerNorm.bias
primals_3 = input_0
primals_8 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12])
return output[0]
| rsgit95/med_kg_txt_multimodal | GTXSelfAttentionLayer | false | 4,217 | [
"Apache-2.0"
] | 0 | 80355b0cf58e0571531ad6f9728c533110ca996d | https://github.com/rsgit95/med_kg_txt_multimodal/tree/80355b0cf58e0571531ad6f9728c533110ca996d | from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class GTXAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.self = GTXAttention(config)
self.output = GTXAttentionOutput(config)
def forward(self, input_tensor, attention_mask, output_attentions=False):
output = self.self(input_tensor, input_tensor, attention_mask,
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
atten
# ... truncated (>4000 chars) for memory efficiency |
MlpBlock | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/go/cgoenidctx3utbz75whfhze57bssfe7fzqjjztzbmsp3z7uj7v55.py
# Topologically Sorted Source Nodes: [gelu], Original ATen: [aten.gelu]
# Source node to ATen node mapping:
# gelu => add, erf, mul, mul_1, mul_2
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.5), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.7071067811865476), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%mul_1,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add), kwargs = {})
triton_poi_fused_gelu_0 = async_compile.triton('triton_poi_fused_gelu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_gelu_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_gelu_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32768
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), None)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865476
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tl.store(out_ptr0 + (x0), tmp8, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (512, 4), (4, 1))
assert_size_stride(primals_2, (512, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 512), (512, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 512), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 512), (8192, 2048, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [gelu], Original ATen: [aten.gelu]
stream0 = get_raw_stream(0)
triton_poi_fused_gelu_0.run(buf0, buf1, 32768, grid=grid(32768), stream=stream0)
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 512), (512, 1), 0), reinterpret_tensor(primals_4, (512, 4), (1, 512), 0), alpha=1, beta=1, out=buf2)
del primals_5
return (reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf0, reinterpret_tensor(buf1, (64, 512), (512, 1), 0), primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((512, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class MlpBlock(nn.Module):
def __init__(self, input_dim, mlp_dim=512):
super().__init__()
self.fc1 = nn.Linear(input_dim, mlp_dim)
self.gelu = nn.GELU()
self.fc2 = nn.Linear(mlp_dim, input_dim)
def forward(self, x):
return self.fc2(self.gelu(self.fc1(x)))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_gelu_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, None)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865476
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tl.store(out_ptr0 + x0, tmp8, None)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (512, 4), (4, 1))
assert_size_stride(primals_2, (512,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 512), (512, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 512), (512, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 512), (1, 4),
0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 512), (8192, 2048, 512, 1),
torch.float32)
get_raw_stream(0)
triton_poi_fused_gelu_0[grid(32768)](buf0, buf1, 32768, XBLOCK=256,
num_warps=4, num_stages=1)
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 512),
(512, 1), 0), reinterpret_tensor(primals_4, (512, 4), (1, 512),
0), alpha=1, beta=1, out=buf2)
del primals_5
return reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf0, reinterpret_tensor(buf1, (64, 512), (512, 1), 0), primals_4
class MlpBlockNew(nn.Module):
def __init__(self, input_dim, mlp_dim=512):
super().__init__()
self.fc1 = nn.Linear(input_dim, mlp_dim)
self.gelu = nn.GELU()
self.fc2 = nn.Linear(mlp_dim, input_dim)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | MlpBlock | false | 4,218 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, input_dim, mlp_dim=512):
super().__init__()
self.fc1 = nn.Linear(input_dim, mlp_dim)
self.gelu = nn.GELU()
self.fc2 = nn.Linear(mlp_dim, input_dim)
def forward(self, x):
return self.fc2(self.gelu(self.fc1(x)))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
LxmertCrossAttentionLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/x2/cx2hdvwyo7m5jvhhvtugzxqvmy6z4nsfhkkjhvgzbbm3cb6dsum2.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %mul_scalar : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%permute_default, 1.0), kwargs = {})
# %clone_default : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5j/c5jll3kxtd32cl7pwubrb5oky2mtzckfgip2xbwad7crvvp4zk4r.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_default_2, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_default_2, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/kt/cktnex5febczl2ac6zugjmcksgsd5kjdufazv65vtepuwob3cb7a.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %sum_dim_int_list : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_default, [-1], True), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_default, %sum_dim_int_list), kwargs = {})
# %eq_scalar : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%view_default_2, -inf), kwargs = {})
# %logical_not_default : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%eq_scalar,), kwargs = {})
# %any_dim : [num_users=1] = call_function[target=torch.ops.aten.any.dim](args = (%logical_not_default, -1, True), kwargs = {})
# %logical_not_default_1 : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%any_dim,), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%logical_not_default_1, %full_default, %div_tensor), kwargs = {})
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (x2), xmask)
tmp26 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp31 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = float("-inf")
tmp2 = tmp0 == tmp1
tmp3 = tmp2 == 0
tmp4 = tmp3.to(tl.int64)
tmp5 = (tmp4 != 0)
tmp7 = tmp6 == tmp1
tmp8 = tmp7 == 0
tmp9 = tmp8.to(tl.int64)
tmp10 = (tmp9 != 0)
tmp11 = tmp5 | tmp10
tmp13 = tmp12 == tmp1
tmp14 = tmp13 == 0
tmp15 = tmp14.to(tl.int64)
tmp16 = (tmp15 != 0)
tmp17 = tmp11 | tmp16
tmp19 = tmp18 == tmp1
tmp20 = tmp19 == 0
tmp21 = tmp20.to(tl.int64)
tmp22 = (tmp21 != 0)
tmp23 = tmp17 | tmp22
tmp24 = tmp23 == 0
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp32 = tmp30 + tmp31
tmp33 = tmp25 / tmp32
tmp34 = 0.0
tmp35 = tl.where(tmp24, tmp34, tmp33)
tl.store(out_ptr0 + (x2), tmp35, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vv/cvvnhithjvmvhfjufxwwzclfobkrgbyyteg66hp24r675f7elw4c.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %clone_default_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default_3,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6t/c6t5a5ere3lqjiu7zh3uu4oxmpdoujdaqqmeunxqapgzo4m74uav.py
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# context_layer_1 => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6m/c6mhj5zwirfhy5e4o45uaeov72uwfby4udubpm2fcz42iqvs2g57.py
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# hidden_states_2 => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/i6/ci6ua4lfqzz3v6lbsh75noa7k5ird3udb6b5bjh7gxx4qxuz7gz3.py
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# hidden_states_2 => add_1, add_2, mul, mul_1, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_11), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_12), kwargs = {})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(buf0, primals_2, buf3, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_0.run(buf1, primals_5, buf4, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(buf5, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf5
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(buf2, primals_8, buf8, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf9, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf11)
del primals_10
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(buf11, primals_3, buf12, buf13, 16, grid=grid(16), stream=stream0)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(buf11, primals_3, buf12, buf13, primals_11, primals_12, buf14, 64, grid=grid(64), stream=stream0)
del buf12
del buf13
del primals_12
return (buf14, primals_3, primals_11, reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), buf7, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0), reinterpret_tensor(buf10, (16, 4), (4, 1), 0), buf11, primals_9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class LxmertAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class LxmertCrossAttentionLayer(nn.Module):
def __init__(self, config):
super().__init__()
self.att = LxmertAttention(config)
self.output = LxmertAttentionOutput(config)
def forward(self, input_tensor, ctx_tensor, ctx_att_mask=None,
output_attentions=False):
output = self.att(input_tensor, ctx_tensor, ctx_att_mask,
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
attention_output = self.output(output[0], input_tensor)
outputs = (attention_output, attention_probs
) if output_attentions else (attention_output,)
return outputs
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, hidden_dropout_prob=0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr1 + x2, xmask)
tmp26 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp29 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp31 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = float('-inf')
tmp2 = tmp0 == tmp1
tmp3 = tmp2 == 0
tmp4 = tmp3.to(tl.int64)
tmp5 = tmp4 != 0
tmp7 = tmp6 == tmp1
tmp8 = tmp7 == 0
tmp9 = tmp8.to(tl.int64)
tmp10 = tmp9 != 0
tmp11 = tmp5 | tmp10
tmp13 = tmp12 == tmp1
tmp14 = tmp13 == 0
tmp15 = tmp14.to(tl.int64)
tmp16 = tmp15 != 0
tmp17 = tmp11 | tmp16
tmp19 = tmp18 == tmp1
tmp20 = tmp19 == 0
tmp21 = tmp20.to(tl.int64)
tmp22 = tmp21 != 0
tmp23 = tmp17 | tmp22
tmp24 = tmp23 == 0
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp32 = tmp30 + tmp31
tmp33 = tmp25 / tmp32
tmp34 = 0.0
tmp35 = tl.where(tmp24, tmp34, tmp33)
tl.store(out_ptr0 + x2, tmp35, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(16, 4)](buf0, primals_2, buf3, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0)
del buf0
triton_poi_fused_0[grid(16, 4)](buf1, primals_5, buf4, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_1[grid(256)](buf5, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_2[grid(256)](buf5, buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_3[grid(16, 4)](buf2, primals_8, buf8, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_4[grid(16, 4)](buf9, buf10, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.addmm(primals_10, reinterpret_tensor(buf10, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf11)
del primals_10
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](buf11, primals_3,
buf12, buf13, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(64)](buf11, primals_3,
buf12, buf13, primals_11, primals_12, buf14, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf12
del buf13
del primals_12
return buf14, primals_3, primals_11, reinterpret_tensor(primals_6, (16,
4), (4, 1), 0), buf7, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0
), reinterpret_tensor(buf10, (16, 4), (4, 1), 0), buf11, primals_9
class LxmertAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class LxmertCrossAttentionLayerNew(nn.Module):
def __init__(self, config):
super().__init__()
self.att = LxmertAttention(config)
self.output = LxmertAttentionOutput(config)
def forward(self, input_0, input_1):
primals_1 = self.att.query.weight
primals_2 = self.att.query.bias
primals_4 = self.att.key.weight
primals_5 = self.att.key.bias
primals_7 = self.att.value.weight
primals_8 = self.att.value.bias
primals_9 = self.output.dense.weight
primals_10 = self.output.dense.bias
primals_11 = self.output.LayerNorm.weight
primals_12 = self.output.LayerNorm.bias
primals_3 = input_0
primals_6 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12])
return output[0]
| rsgit95/med_kg_txt_multimodal | LxmertCrossAttentionLayer | false | 4,219 | [
"Apache-2.0"
] | 0 | 80355b0cf58e0571531ad6f9728c533110ca996d | https://github.com/rsgit95/med_kg_txt_multimodal/tree/80355b0cf58e0571531ad6f9728c533110ca996d | from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class LxmertAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.att = LxmertAttention(config)
self.output = LxmertAttentionOutput(config)
def forward(self, input_tensor, ctx_tensor, ctx_att_mask=None,
output_attentions=False):
output = self.att(input_tensor, ctx_tensor, ctx_att_mask,
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
attention_output = self.output(output[0], input_tensor)
outputs = (attention_output, attention_probs
) if output_attentions else (attention_output,)
return outputs
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, hidden_dropout_prob=0.5)}]
|
GTXCrossAttentionLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/x2/cx2hdvwyo7m5jvhhvtugzxqvmy6z4nsfhkkjhvgzbbm3cb6dsum2.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %mul_scalar : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%permute_default, 1.0), kwargs = {})
# %clone_default : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5j/c5jll3kxtd32cl7pwubrb5oky2mtzckfgip2xbwad7crvvp4zk4r.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_default_2, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_default_2, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/kt/cktnex5febczl2ac6zugjmcksgsd5kjdufazv65vtepuwob3cb7a.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %sum_dim_int_list : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_default, [-1], True), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_default, %sum_dim_int_list), kwargs = {})
# %eq_scalar : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%view_default_2, -inf), kwargs = {})
# %logical_not_default : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%eq_scalar,), kwargs = {})
# %any_dim : [num_users=1] = call_function[target=torch.ops.aten.any.dim](args = (%logical_not_default, -1, True), kwargs = {})
# %logical_not_default_1 : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%any_dim,), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%logical_not_default_1, %full_default, %div_tensor), kwargs = {})
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (x2), xmask)
tmp26 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp31 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = float("-inf")
tmp2 = tmp0 == tmp1
tmp3 = tmp2 == 0
tmp4 = tmp3.to(tl.int64)
tmp5 = (tmp4 != 0)
tmp7 = tmp6 == tmp1
tmp8 = tmp7 == 0
tmp9 = tmp8.to(tl.int64)
tmp10 = (tmp9 != 0)
tmp11 = tmp5 | tmp10
tmp13 = tmp12 == tmp1
tmp14 = tmp13 == 0
tmp15 = tmp14.to(tl.int64)
tmp16 = (tmp15 != 0)
tmp17 = tmp11 | tmp16
tmp19 = tmp18 == tmp1
tmp20 = tmp19 == 0
tmp21 = tmp20.to(tl.int64)
tmp22 = (tmp21 != 0)
tmp23 = tmp17 | tmp22
tmp24 = tmp23 == 0
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp32 = tmp30 + tmp31
tmp33 = tmp25 / tmp32
tmp34 = 0.0
tmp35 = tl.where(tmp24, tmp34, tmp33)
tl.store(out_ptr0 + (x2), tmp35, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vv/cvvnhithjvmvhfjufxwwzclfobkrgbyyteg66hp24r675f7elw4c.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %clone_default_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default_3,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6t/c6t5a5ere3lqjiu7zh3uu4oxmpdoujdaqqmeunxqapgzo4m74uav.py
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# context_layer_1 => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6m/c6mhj5zwirfhy5e4o45uaeov72uwfby4udubpm2fcz42iqvs2g57.py
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_2 => add
# hidden_states_3 => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %view_17), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/i6/ci6ua4lfqzz3v6lbsh75noa7k5ird3udb6b5bjh7gxx4qxuz7gz3.py
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states_2 => add
# hidden_states_3 => add_1, add_2, mul, mul_1, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %view_17), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_11), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_12), kwargs = {})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(buf0, primals_2, buf3, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_0.run(buf1, primals_5, buf4, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(buf5, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf5
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(buf2, primals_8, buf8, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf9, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf11)
del primals_10
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(primals_3, buf11, buf12, buf13, 16, grid=grid(16), stream=stream0)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_2, hidden_states_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(primals_3, buf11, buf12, buf13, primals_11, primals_12, buf14, 64, grid=grid(64), stream=stream0)
del buf12
del buf13
del primals_12
return (buf14, primals_3, primals_11, reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), buf7, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0), reinterpret_tensor(buf10, (16, 4), (4, 1), 0), buf11, primals_9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class GTXAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class GTXCrossAttentionLayer(nn.Module):
def __init__(self, config):
super().__init__()
self.att = GTXAttention(config)
self.output = GTXAttentionOutput(config)
def forward(self, input_tensor, ctx_tensor, ctx_att_mask=None,
KnowMix_indices=None, output_attentions=False):
if KnowMix_indices is None:
output = self.att(input_tensor, ctx_tensor, ctx_att_mask,
output_attentions=output_attentions)
elif isinstance(KnowMix_indices, int):
output = self.att(input_tensor[:, KnowMix_indices].unsqueeze(1),
ctx_tensor, ctx_att_mask, output_attentions=output_attentions)
else:
output = self.att(input_tensor[KnowMix_indices, :].unsqueeze(1),
ctx_tensor[KnowMix_indices, :], ctx_att_mask[
KnowMix_indices.unsqueeze(1), :].unsqueeze(1).unsqueeze(2),
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
attention_output = self.output(output[0], input_tensor, KnowMix_indices
)
outputs = (attention_output, attention_probs
) if output_attentions else (attention_output,)
return outputs
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, hidden_dropout_prob=0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr1 + x2, xmask)
tmp26 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp29 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp31 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = float('-inf')
tmp2 = tmp0 == tmp1
tmp3 = tmp2 == 0
tmp4 = tmp3.to(tl.int64)
tmp5 = tmp4 != 0
tmp7 = tmp6 == tmp1
tmp8 = tmp7 == 0
tmp9 = tmp8.to(tl.int64)
tmp10 = tmp9 != 0
tmp11 = tmp5 | tmp10
tmp13 = tmp12 == tmp1
tmp14 = tmp13 == 0
tmp15 = tmp14.to(tl.int64)
tmp16 = tmp15 != 0
tmp17 = tmp11 | tmp16
tmp19 = tmp18 == tmp1
tmp20 = tmp19 == 0
tmp21 = tmp20.to(tl.int64)
tmp22 = tmp21 != 0
tmp23 = tmp17 | tmp22
tmp24 = tmp23 == 0
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp32 = tmp30 + tmp31
tmp33 = tmp25 / tmp32
tmp34 = 0.0
tmp35 = tl.where(tmp24, tmp34, tmp33)
tl.store(out_ptr0 + x2, tmp35, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(16, 4)](buf0, primals_2, buf3, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0)
del buf0
triton_poi_fused_0[grid(16, 4)](buf1, primals_5, buf4, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_1[grid(256)](buf5, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_2[grid(256)](buf5, buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_3[grid(16, 4)](buf2, primals_8, buf8, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_4[grid(16, 4)](buf9, buf10, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.addmm(primals_10, reinterpret_tensor(buf10, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf11)
del primals_10
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](primals_3, buf11,
buf12, buf13, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(64)](primals_3, buf11,
buf12, buf13, primals_11, primals_12, buf14, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf12
del buf13
del primals_12
return buf14, primals_3, primals_11, reinterpret_tensor(primals_6, (16,
4), (4, 1), 0), buf7, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0
), reinterpret_tensor(buf10, (16, 4), (4, 1), 0), buf11, primals_9
class GTXAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class GTXCrossAttentionLayerNew(nn.Module):
def __init__(self, config):
super().__init__()
self.att = GTXAttention(config)
self.output = GTXAttentionOutput(config)
def forward(self, input_0, input_1):
primals_1 = self.att.query.weight
primals_2 = self.att.query.bias
primals_4 = self.att.key.weight
primals_5 = self.att.key.bias
primals_7 = self.att.value.weight
primals_8 = self.att.value.bias
primals_9 = self.output.dense.weight
primals_10 = self.output.dense.bias
primals_11 = self.output.LayerNorm.weight
primals_12 = self.output.LayerNorm.bias
primals_3 = input_0
primals_6 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12])
return output[0]
| rsgit95/med_kg_txt_multimodal | GTXCrossAttentionLayer | false | 4,220 | [
"Apache-2.0"
] | 0 | 80355b0cf58e0571531ad6f9728c533110ca996d | https://github.com/rsgit95/med_kg_txt_multimodal/tree/80355b0cf58e0571531ad6f9728c533110ca996d | from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class GTXAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class GTXAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor, KnowMix_indices=None):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
if KnowMix_indices is None:
hidden_states = input_tensor + hidden_states
else:
if isinstance(KnowMix_indices, int):
input_tensor[:, KnowMix_indices] = input_tensor[:,
KnowMix_indices] + hidden_states.squeeze(1)
else:
input_tensor[KnowMix_indices, :] = input_tensor[
KnowMix_indices, :] + hidden_states.squeeze(1)
hidden_states = input_tensor
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.att = GTXAttention(config)
self.output = GTXAttentionOutput(config)
def forward(self, input_tensor, ctx_tensor, ctx_att_mask=None,
KnowMix_indices=None, output_attentions=False):
if KnowMix_indices is None:
output = self.att(input_tensor, ctx_tensor, ctx_att_mask,
output_attentions=output_attentions)
# ... truncated (>4000 chars) for memory efficiency |
ConvEncoder | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/kn/cknyjwkwufnzzf4ya3scui55ownkmt5cdh3hggzwsfe3ch5fshzm.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4096], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4096
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y3 = yindex
y0 = yindex % 3
y1 = (yindex // 3)
tmp0 = tl.load(in_ptr0 + (x2 + (4096*y3)), ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (3*x2) + (12288*y1)), tmp0, ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/p6/cp6s4svoxgnzqeja6pzmabu3asmqyfoaympag6dtphmqncue7fik.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 96
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 3
y1 = (yindex // 3)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (3*x2) + (27*y1)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/wv/cwvtp6qflpb42kxrujmda5zselv7wvkz3fgp2tryo2ftsisaildr.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 2048
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 32
y1 = (yindex // 32)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (32*x2) + (288*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nw/cnwm6ljuusoqjcwr2jdx6p2ue7ldghxjdr3oe62stiuqhsboiczy.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 8192
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 64
y1 = (yindex // 64)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (64*x2) + (576*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ih/cihu7ohoiwwrblocurozhw6ihpzbq4oc43mseo4n6wd7ronp74tw.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_4 = async_compile.triton('triton_poi_fused_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 32768
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 128
y1 = (yindex // 128)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (128*x2) + (1152*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5c/c5cs3c3svcrznivu3zzny5tguj65spdtj2aitirh7fijbdkiv4cm.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_5 = async_compile.triton('triton_poi_fused_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[131072, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_5(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 131072
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = (yindex // 256)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (256*x2) + (2304*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/wm/cwmaacsffuosyjtapv3aqt4pkyqgob4jjpsl2qfwqddn7tfjjwjo.py
# Topologically Sorted Source Nodes: [net, relu], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# net => convolution
# relu => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_2, %primals_3, [2, 2], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_6 = async_compile.triton('triton_poi_fused_convolution_relu_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[131072],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_6', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_6(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 123008
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 32
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/kt/cktxa4ungjp46xeccvlgy4nvcb4x4hunx3vtik5eswj7ak2rlada.py
# Topologically Sorted Source Nodes: [net_1, relu_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# net_1 => convolution_1
# relu_1 => relu_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [2, 2], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_relu_7 = async_compile.triton('triton_poi_fused_convolution_relu_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[65536],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_7', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_7(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 57600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ky/ckylxdqvllfbkwuaq4pecelkx2s5gcvwuxkx6iawkomgwk6xhngy.py
# Topologically Sorted Source Nodes: [net_2, relu_2], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# net_2 => convolution_2
# relu_2 => relu_2
# Graph fragment:
# %convolution_2 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu_1, %primals_6, %primals_7, [2, 2], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_2,), kwargs = {})
triton_poi_fused_convolution_relu_8 = async_compile.triton('triton_poi_fused_convolution_relu_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_8', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_8(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 25088
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/r7/cr7uq5ur2j4jwzfgtl7exszppe4m5worsfpyxi3aykaqmannjwdx.py
# Topologically Sorted Source Nodes: [net_3, relu_3], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# net_3 => convolution_3
# relu_3 => relu_3
# Graph fragment:
# %convolution_3 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu_2, %primals_8, %primals_9, [2, 2], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_3 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_3,), kwargs = {})
triton_poi_fused_convolution_relu_9 = async_compile.triton('triton_poi_fused_convolution_relu_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_9(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 9216
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 256
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ow/cowazwlu4tabjis56afgwzpg3aatyheci4ezqs3pmtxm2jiuqqhs.py
# Topologically Sorted Source Nodes: [net_5, relu_4], Original ATen: [aten.mean, aten.relu]
# Source node to ATen node mapping:
# net_5 => mean
# relu_4 => relu_4
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view, [2]), kwargs = {})
# %relu_4 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%mean,), kwargs = {})
triton_poi_fused_mean_relu_10 = async_compile.triton('triton_poi_fused_mean_relu_10', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mean_relu_10', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mean_relu_10(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 2048
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 / tmp3
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tl.store(in_out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13 = args
args.clear()
assert_size_stride(primals_1, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_2, (32, 3, 3, 3), (27, 9, 3, 1))
assert_size_stride(primals_3, (32, ), (1, ))
assert_size_stride(primals_4, (64, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (128, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_7, (128, ), (1, ))
assert_size_stride(primals_8, (256, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_9, (256, ), (1, ))
assert_size_stride(primals_10, (512, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_11, (512, ), (1, ))
assert_size_stride(primals_12, (128, 512), (512, 1))
assert_size_stride(primals_13, (128, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 3, 64, 64), (12288, 1, 192, 3), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(primals_1, buf0, 12, 4096, grid=grid(12, 4096), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((32, 3, 3, 3), (27, 1, 9, 3), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(primals_2, buf1, 96, 9, grid=grid(96, 9), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 32, 3, 3), (288, 1, 96, 32), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(primals_4, buf2, 2048, 9, grid=grid(2048, 9), stream=stream0)
del primals_4
buf3 = empty_strided_cuda((128, 64, 3, 3), (576, 1, 192, 64), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(primals_6, buf3, 8192, 9, grid=grid(8192, 9), stream=stream0)
del primals_6
buf4 = empty_strided_cuda((256, 128, 3, 3), (1152, 1, 384, 128), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_4.run(primals_8, buf4, 32768, 9, grid=grid(32768, 9), stream=stream0)
del primals_8
buf5 = empty_strided_cuda((512, 256, 3, 3), (2304, 1, 768, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_5.run(primals_10, buf5, 131072, 9, grid=grid(131072, 9), stream=stream0)
del primals_10
# Topologically Sorted Source Nodes: [net], Original ATen: [aten.convolution]
buf6 = extern_kernels.convolution(buf0, buf1, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf6, (4, 32, 31, 31), (30752, 1, 992, 32))
buf7 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [net, relu], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_6.run(buf7, primals_3, 123008, grid=grid(123008), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [net_1], Original ATen: [aten.convolution]
buf8 = extern_kernels.convolution(buf7, buf2, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 64, 15, 15), (14400, 1, 960, 64))
buf9 = buf8; del buf8 # reuse
# Topologically Sorted Source Nodes: [net_1, relu_1], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_7.run(buf9, primals_5, 57600, grid=grid(57600), stream=stream0)
del primals_5
# Topologically Sorted Source Nodes: [net_2], Original ATen: [aten.convolution]
buf10 = extern_kernels.convolution(buf9, buf3, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 128, 7, 7), (6272, 1, 896, 128))
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [net_2, relu_2], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_8.run(buf11, primals_7, 25088, grid=grid(25088), stream=stream0)
del primals_7
# Topologically Sorted Source Nodes: [net_3], Original ATen: [aten.convolution]
buf12 = extern_kernels.convolution(buf11, buf4, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 256, 3, 3), (2304, 1, 768, 256))
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [net_3, relu_3], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_9.run(buf13, primals_9, 9216, grid=grid(9216), stream=stream0)
del primals_9
# Topologically Sorted Source Nodes: [net_4], Original ATen: [aten.convolution]
buf14 = extern_kernels.convolution(buf13, buf5, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf14, (4, 512, 1, 1), (512, 1, 512, 512))
buf15 = reinterpret_tensor(buf14, (4, 512), (512, 1), 0); del buf14 # reuse
# Topologically Sorted Source Nodes: [net_5, relu_4], Original ATen: [aten.mean, aten.relu]
triton_poi_fused_mean_relu_10.run(buf15, primals_11, 2048, grid=grid(2048), stream=stream0)
del primals_11
buf16 = empty_strided_cuda((4, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_13, buf15, reinterpret_tensor(primals_12, (512, 128), (1, 512), 0), alpha=1, beta=1, out=buf16)
del primals_13
return (buf16, buf0, buf1, buf2, buf3, buf4, buf5, buf7, buf9, buf11, buf13, buf15, primals_12, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 3, 64, 64), (12288, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((32, 3, 3, 3), (27, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 32, 3, 3), (288, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((128, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((256, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((512, 256, 3, 3), (2304, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((128, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class ConvEncoder(nn.Module):
""" Simple convolutional encoder network.
It consists of 5 convolutional layers, each downsampling the input by a
factor of 2, and a final fully-connected layer projecting the output to
c_dim dimensions.
Args:
c_dim (int): output dimension of latent embedding
"""
def __init__(self, c_dim=128):
super().__init__()
self.conv0 = nn.Conv2d(3, 32, 3, stride=2)
self.conv1 = nn.Conv2d(32, 64, 3, stride=2)
self.conv2 = nn.Conv2d(64, 128, 3, stride=2)
self.conv3 = nn.Conv2d(128, 256, 3, stride=2)
self.conv4 = nn.Conv2d(256, 512, 3, stride=2)
self.fc_out = nn.Linear(512, c_dim)
self.actvn = nn.ReLU()
def forward(self, x):
batch_size = x.size(0)
net = self.conv0(x)
net = self.conv1(self.actvn(net))
net = self.conv2(self.actvn(net))
net = self.conv3(self.actvn(net))
net = self.conv4(self.actvn(net))
net = net.view(batch_size, 512, -1).mean(2)
out = self.fc_out(self.actvn(net))
return out
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 12
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y3 = yindex
y0 = yindex % 3
y1 = yindex // 3
tmp0 = tl.load(in_ptr0 + (x2 + 4096 * y3), ymask, eviction_policy=
'evict_last')
tl.store(out_ptr0 + (y0 + 3 * x2 + 12288 * y1), tmp0, ymask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 96
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 3
y1 = yindex // 3
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask & ymask, eviction_policy=
'evict_last')
tl.store(out_ptr0 + (y0 + 3 * x2 + 27 * y1), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_2(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 32
y1 = yindex // 32
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 32 * x2 + 288 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 64
y1 = yindex // 64
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 64 * x2 + 576 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 128
y1 = yindex // 128
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 128 * x2 + 1152 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_5(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)
) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = yindex // 256
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 256 * x2 + 2304 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_6(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 123008
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 32
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_7(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 57600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_8(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 25088
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_9(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 9216
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 256
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_mean_relu_10(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 / tmp3
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tl.store(in_out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13) = args
args.clear()
assert_size_stride(primals_1, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_2, (32, 3, 3, 3), (27, 9, 3, 1))
assert_size_stride(primals_3, (32,), (1,))
assert_size_stride(primals_4, (64, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (128, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_7, (128,), (1,))
assert_size_stride(primals_8, (256, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_9, (256,), (1,))
assert_size_stride(primals_10, (512, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_11, (512,), (1,))
assert_size_stride(primals_12, (128, 512), (512, 1))
assert_size_stride(primals_13, (128,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 3, 64, 64), (12288, 1, 192, 3), torch
.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(12, 4096)](primals_1, buf0, 12, 4096,
XBLOCK=64, YBLOCK=16, num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((32, 3, 3, 3), (27, 1, 9, 3), torch.float32)
triton_poi_fused_1[grid(96, 9)](primals_2, buf1, 96, 9, XBLOCK=16,
YBLOCK=64, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 32, 3, 3), (288, 1, 96, 32), torch.
float32)
triton_poi_fused_2[grid(2048, 9)](primals_4, buf2, 2048, 9, XBLOCK=
16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((128, 64, 3, 3), (576, 1, 192, 64), torch
.float32)
triton_poi_fused_3[grid(8192, 9)](primals_6, buf3, 8192, 9, XBLOCK=
16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_6
buf4 = empty_strided_cuda((256, 128, 3, 3), (1152, 1, 384, 128),
torch.float32)
triton_poi_fused_4[grid(32768, 9)](primals_8, buf4, 32768, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_8
buf5 = empty_strided_cuda((512, 256, 3, 3), (2304, 1, 768, 256),
torch.float32)
triton_poi_fused_5[grid(131072, 9)](primals_10, buf5, 131072, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_10
buf6 = extern_kernels.convolution(buf0, buf1, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf6, (4, 32, 31, 31), (30752, 1, 992, 32))
buf7 = buf6
del buf6
triton_poi_fused_convolution_relu_6[grid(123008)](buf7, primals_3,
123008, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_3
buf8 = extern_kernels.convolution(buf7, buf2, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 64, 15, 15), (14400, 1, 960, 64))
buf9 = buf8
del buf8
triton_poi_fused_convolution_relu_7[grid(57600)](buf9, primals_5,
57600, XBLOCK=512, num_warps=4, num_stages=1)
del primals_5
buf10 = extern_kernels.convolution(buf9, buf3, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 128, 7, 7), (6272, 1, 896, 128))
buf11 = buf10
del buf10
triton_poi_fused_convolution_relu_8[grid(25088)](buf11, primals_7,
25088, XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
buf12 = extern_kernels.convolution(buf11, buf4, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 256, 3, 3), (2304, 1, 768, 256))
buf13 = buf12
del buf12
triton_poi_fused_convolution_relu_9[grid(9216)](buf13, primals_9,
9216, XBLOCK=256, num_warps=4, num_stages=1)
del primals_9
buf14 = extern_kernels.convolution(buf13, buf5, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf14, (4, 512, 1, 1), (512, 1, 512, 512))
buf15 = reinterpret_tensor(buf14, (4, 512), (512, 1), 0)
del buf14
triton_poi_fused_mean_relu_10[grid(2048)](buf15, primals_11, 2048,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_11
buf16 = empty_strided_cuda((4, 128), (128, 1), torch.float32)
extern_kernels.addmm(primals_13, buf15, reinterpret_tensor(
primals_12, (512, 128), (1, 512), 0), alpha=1, beta=1, out=buf16)
del primals_13
return (buf16, buf0, buf1, buf2, buf3, buf4, buf5, buf7, buf9, buf11,
buf13, buf15, primals_12)
class ConvEncoderNew(nn.Module):
""" Simple convolutional encoder network.
It consists of 5 convolutional layers, each downsampling the input by a
factor of 2, and a final fully-connected layer projecting the output to
c_dim dimensions.
Args:
c_dim (int): output dimension of latent embedding
"""
def __init__(self, c_dim=128):
super().__init__()
self.conv0 = nn.Conv2d(3, 32, 3, stride=2)
self.conv1 = nn.Conv2d(32, 64, 3, stride=2)
self.conv2 = nn.Conv2d(64, 128, 3, stride=2)
self.conv3 = nn.Conv2d(128, 256, 3, stride=2)
self.conv4 = nn.Conv2d(256, 512, 3, stride=2)
self.fc_out = nn.Linear(512, c_dim)
self.actvn = nn.ReLU()
def forward(self, input_0):
primals_2 = self.conv0.weight
primals_3 = self.conv0.bias
primals_4 = self.conv1.weight
primals_5 = self.conv1.bias
primals_6 = self.conv2.weight
primals_7 = self.conv2.bias
primals_8 = self.conv3.weight
primals_9 = self.conv3.bias
primals_10 = self.conv4.weight
primals_11 = self.conv4.bias
primals_12 = self.fc_out.weight
primals_13 = self.fc_out.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13])
return output[0]
| planetceres/differentiable_volumetric_rendering | ConvEncoder | false | 4,221 | [
"MIT"
] | 0 | f2fe46d139244c7642439ced23656db1e7f5c128 | https://github.com/planetceres/differentiable_volumetric_rendering/tree/f2fe46d139244c7642439ced23656db1e7f5c128 | import torch
from torch import nn
class Model(nn.Module):
""" Simple convolutional encoder network.
It consists of 5 convolutional layers, each downsampling the input by a
factor of 2, and a final fully-connected layer projecting the output to
c_dim dimensions.
Args:
c_dim (int): output dimension of latent embedding
"""
def __init__(self, c_dim=128):
super().__init__()
self.conv0 = nn.Conv2d(3, 32, 3, stride=2)
self.conv1 = nn.Conv2d(32, 64, 3, stride=2)
self.conv2 = nn.Conv2d(64, 128, 3, stride=2)
self.conv3 = nn.Conv2d(128, 256, 3, stride=2)
self.conv4 = nn.Conv2d(256, 512, 3, stride=2)
self.fc_out = nn.Linear(512, c_dim)
self.actvn = nn.ReLU()
def forward(self, x):
batch_size = x.size(0)
net = self.conv0(x)
net = self.conv1(self.actvn(net))
net = self.conv2(self.actvn(net))
net = self.conv3(self.actvn(net))
net = self.conv4(self.actvn(net))
net = net.view(batch_size, 512, -1).mean(2)
out = self.fc_out(self.actvn(net))
return out
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return []
|
DoubleAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/pw/cpw5jgywzg5ntkknxkt5orxsrrr5zq7a6eoteboi3ba7zrcxj2p7.py
# Topologically Sorted Source Nodes: [A], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# A => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_2, %primals_3, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5r/c5rhp4pv4cg6ddxhvgfx33emv6p7gwbi47umgsoy4tfzazsi4rak.py
# Topologically Sorted Source Nodes: [attention_maps], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attention_maps => amax, exp, sub, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [0], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [0], True), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + x2), xmask)
tmp6 = tl.load(in_ptr0 + (128 + x2), xmask)
tmp9 = tl.load(in_ptr0 + (192 + x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = triton_helpers.maximum(tmp2, tmp4)
tmp7 = tmp6 + tmp1
tmp8 = triton_helpers.maximum(tmp5, tmp7)
tmp10 = tmp9 + tmp1
tmp11 = triton_helpers.maximum(tmp8, tmp10)
tmp12 = tmp2 - tmp11
tmp13 = tl_math.exp(tmp12)
tmp14 = tmp4 - tmp11
tmp15 = tl_math.exp(tmp14)
tmp16 = tmp13 + tmp15
tmp17 = tmp7 - tmp11
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp10 - tmp11
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tl.store(out_ptr0 + (x2), tmp11, xmask)
tl.store(out_ptr1 + (x2), tmp22, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ps/cpsgkfkg5qfu5qkcgb4gvy54hsdn7yerusqeekt5z32d3ofd5f5x.py
# Topologically Sorted Source Nodes: [attention_maps], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attention_maps => amax, div, exp, sub, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [0], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [0], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 4
x4 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x4), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp5 = tl_math.exp(tmp4)
tmp7 = tmp5 / tmp6
tl.store(in_out_ptr0 + (x3), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_9, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [A], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4, 4), (64, 16, 4, 1))
# Topologically Sorted Source Nodes: [B], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(primals_1, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 4, 4), (64, 16, 4, 1))
# Topologically Sorted Source Nodes: [V], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(primals_1, primals_6, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 4, 4), (64, 16, 4, 1))
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [A], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf3, primals_3, 256, grid=grid(256), stream=stream0)
del primals_3
buf4 = empty_strided_cuda((1, 4, 16), (64, 16, 1), torch.float32)
buf5 = empty_strided_cuda((1, 4, 16), (64, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_maps], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, primals_5, buf4, buf5, 64, grid=grid(64), stream=stream0)
buf6 = reinterpret_tensor(buf1, (4, 4, 16), (64, 16, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [attention_maps], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf6, primals_5, buf4, buf5, 256, grid=grid(256), stream=stream0)
del primals_5
buf7 = buf5; del buf5 # reuse
buf8 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [attention_vectors], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf2, primals_7, buf7, buf8, 64, grid=grid(64), stream=stream0)
buf9 = reinterpret_tensor(buf2, (4, 4, 16), (64, 16, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [attention_vectors], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf9, primals_7, buf7, buf8, 256, grid=grid(256), stream=stream0)
del buf7
del primals_7
buf10 = reinterpret_tensor(buf8, (4, 4, 4), (16, 4, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [global_descriptors], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 4, 16), (64, 16, 1), 0), reinterpret_tensor(buf6, (4, 16, 4), (64, 1, 16), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 16), (64, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [tmpZ], Original ATen: [aten.bmm]
extern_kernels.bmm(buf10, buf9, out=buf11)
# Topologically Sorted Source Nodes: [tmpZ_2], Original ATen: [aten.convolution]
buf12 = extern_kernels.convolution(reinterpret_tensor(buf11, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_8, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 4, 4, 4), (64, 16, 4, 1))
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [tmpZ_2], Original ATen: [aten.convolution]
triton_poi_fused_convolution_0.run(buf13, primals_9, 256, grid=grid(256), stream=stream0)
del primals_9
return (buf13, primals_1, primals_2, primals_4, primals_6, primals_8, buf6, buf9, reinterpret_tensor(buf11, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf10, (4, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf3, (4, 16, 4), (64, 1, 16), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
from torch.nn import functional as F
from torch.nn import init
class DoubleAttention(nn.Module):
def __init__(self, in_channels, c_m, c_n, reconstruct=True):
super().__init__()
self.in_channels = in_channels
self.reconstruct = reconstruct
self.c_m = c_m
self.c_n = c_n
self.convA = nn.Conv2d(in_channels, c_m, 1)
self.convB = nn.Conv2d(in_channels, c_n, 1)
self.convV = nn.Conv2d(in_channels, c_n, 1)
if self.reconstruct:
self.conv_reconstruct = nn.Conv2d(c_m, in_channels, kernel_size=1)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, x):
b, c, h, w = x.shape
assert c == self.in_channels
A = self.convA(x)
B = self.convB(x)
V = self.convV(x)
tmpA = A.view(b, self.c_m, -1)
attention_maps = F.softmax(B.view(b, self.c_n, -1))
attention_vectors = F.softmax(V.view(b, self.c_n, -1))
global_descriptors = torch.bmm(tmpA, attention_maps.permute(0, 2, 1))
tmpZ = global_descriptors.matmul(attention_vectors)
tmpZ = tmpZ.view(b, self.c_m, h, w)
if self.reconstruct:
tmpZ = self.conv_reconstruct(tmpZ)
return tmpZ
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'c_m': 4, 'c_n': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 16
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + x2), xmask)
tmp6 = tl.load(in_ptr0 + (128 + x2), xmask)
tmp9 = tl.load(in_ptr0 + (192 + x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = triton_helpers.maximum(tmp2, tmp4)
tmp7 = tmp6 + tmp1
tmp8 = triton_helpers.maximum(tmp5, tmp7)
tmp10 = tmp9 + tmp1
tmp11 = triton_helpers.maximum(tmp8, tmp10)
tmp12 = tmp2 - tmp11
tmp13 = tl_math.exp(tmp12)
tmp14 = tmp4 - tmp11
tmp15 = tl_math.exp(tmp14)
tmp16 = tmp13 + tmp15
tmp17 = tmp7 - tmp11
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp10 - tmp11
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tl.store(out_ptr0 + x2, tmp11, xmask)
tl.store(out_ptr1 + x2, tmp22, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 4
x4 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x4, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp5 = tl_math.exp(tmp4)
tmp7 = tmp5 / tmp6
tl.store(in_out_ptr0 + x3, tmp7, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_9, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4, 4), (64, 16, 4, 1))
buf1 = extern_kernels.convolution(primals_1, primals_4, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 4, 4), (64, 16, 4, 1))
buf2 = extern_kernels.convolution(primals_1, primals_6, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 4, 4), (64, 16, 4, 1))
buf3 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(256)](buf3, primals_3, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_3
buf4 = empty_strided_cuda((1, 4, 16), (64, 16, 1), torch.float32)
buf5 = empty_strided_cuda((1, 4, 16), (64, 16, 1), torch.float32)
triton_poi_fused__softmax_1[grid(64)](buf1, primals_5, buf4, buf5,
64, XBLOCK=64, num_warps=1, num_stages=1)
buf6 = reinterpret_tensor(buf1, (4, 4, 16), (64, 16, 1), 0)
del buf1
triton_poi_fused__softmax_2[grid(256)](buf6, primals_5, buf4, buf5,
256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf7 = buf5
del buf5
buf8 = buf4
del buf4
triton_poi_fused__softmax_1[grid(64)](buf2, primals_7, buf7, buf8,
64, XBLOCK=64, num_warps=1, num_stages=1)
buf9 = reinterpret_tensor(buf2, (4, 4, 16), (64, 16, 1), 0)
del buf2
triton_poi_fused__softmax_2[grid(256)](buf9, primals_7, buf7, buf8,
256, XBLOCK=256, num_warps=4, num_stages=1)
del buf7
del primals_7
buf10 = reinterpret_tensor(buf8, (4, 4, 4), (16, 4, 1), 0)
del buf8
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 4, 16), (64, 16, 1),
0), reinterpret_tensor(buf6, (4, 16, 4), (64, 1, 16), 0), out=buf10
)
buf11 = empty_strided_cuda((4, 4, 16), (64, 16, 1), torch.float32)
extern_kernels.bmm(buf10, buf9, out=buf11)
buf12 = extern_kernels.convolution(reinterpret_tensor(buf11, (4, 4,
4, 4), (64, 16, 4, 1), 0), primals_8, stride=(1, 1), padding=(0,
0), dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=1, bias=None)
assert_size_stride(buf12, (4, 4, 4, 4), (64, 16, 4, 1))
buf13 = buf12
del buf12
triton_poi_fused_convolution_0[grid(256)](buf13, primals_9, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_9
return (buf13, primals_1, primals_2, primals_4, primals_6, primals_8,
buf6, buf9, reinterpret_tensor(buf11, (4, 4, 4, 4), (64, 16, 4, 1),
0), reinterpret_tensor(buf10, (4, 4, 4), (16, 1, 4), 0),
reinterpret_tensor(buf3, (4, 16, 4), (64, 1, 16), 0))
class DoubleAttentionNew(nn.Module):
def __init__(self, in_channels, c_m, c_n, reconstruct=True):
super().__init__()
self.in_channels = in_channels
self.reconstruct = reconstruct
self.c_m = c_m
self.c_n = c_n
self.convA = nn.Conv2d(in_channels, c_m, 1)
self.convB = nn.Conv2d(in_channels, c_n, 1)
self.convV = nn.Conv2d(in_channels, c_n, 1)
if self.reconstruct:
self.conv_reconstruct = nn.Conv2d(c_m, in_channels, kernel_size=1)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, input_0):
primals_2 = self.convA.weight
primals_3 = self.convA.bias
primals_4 = self.convB.weight
primals_5 = self.convB.bias
primals_6 = self.convV.weight
primals_7 = self.convV.bias
primals_8 = self.conv_reconstruct.weight
primals_9 = self.conv_reconstruct.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | DoubleAttention | false | 4,222 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
from torch.nn import functional as F
from torch.nn import init
class Model(nn.Module):
def __init__(self, in_channels, c_m, c_n, reconstruct=True):
super().__init__()
self.in_channels = in_channels
self.reconstruct = reconstruct
self.c_m = c_m
self.c_n = c_n
self.convA = nn.Conv2d(in_channels, c_m, 1)
self.convB = nn.Conv2d(in_channels, c_n, 1)
self.convV = nn.Conv2d(in_channels, c_n, 1)
if self.reconstruct:
self.conv_reconstruct = nn.Conv2d(c_m, in_channels, kernel_size=1)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, x):
b, c, h, w = x.shape
assert c == self.in_channels
A = self.convA(x)
B = self.convB(x)
V = self.convV(x)
tmpA = A.view(b, self.c_m, -1)
attention_maps = F.softmax(B.view(b, self.c_n, -1))
attention_vectors = F.softmax(V.view(b, self.c_n, -1))
global_descriptors = torch.bmm(tmpA, attention_maps.permute(0, 2, 1))
tmpZ = global_descriptors.matmul(attention_vectors)
tmpZ = tmpZ.view(b, self.c_m, h, w)
if self.reconstruct:
tmpZ = self.conv_reconstruct(tmpZ)
return tmpZ
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
LxmertSelfAttentionLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/x2/cx2hdvwyo7m5jvhhvtugzxqvmy6z4nsfhkkjhvgzbbm3cb6dsum2.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %mul_scalar : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%permute_default, 1.0), kwargs = {})
# %clone_default : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/iz/ciztqj6kop3hxov46yrmzprkzfir3eljcic4mkqznz2j5cfeaudr.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %add_tensor : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_default_2, %primals_8), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_tensor, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
# %sum_dim_int_list : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_default, [-1], True), kwargs = {})
# %eq_scalar : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%add_tensor, -inf), kwargs = {})
# %logical_not_default : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%eq_scalar,), kwargs = {})
# %any_dim : [num_users=1] = call_function[target=torch.ops.aten.any.dim](args = (%logical_not_default, -1, True), kwargs = {})
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = triton_helpers.maximum(tmp2, tmp5)
tmp9 = tmp7 + tmp8
tmp10 = triton_helpers.maximum(tmp6, tmp9)
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp10, tmp13)
tmp15 = tmp2 - tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tmp5 - tmp14
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp9 - tmp14
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp23 = tmp13 - tmp14
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = float("-inf")
tmp27 = tmp2 == tmp26
tmp28 = tmp27 == 0
tmp29 = tmp28.to(tl.int64)
tmp30 = (tmp29 != 0)
tmp31 = tmp5 == tmp26
tmp32 = tmp31 == 0
tmp33 = tmp32.to(tl.int64)
tmp34 = (tmp33 != 0)
tmp35 = tmp30 | tmp34
tmp36 = tmp9 == tmp26
tmp37 = tmp36 == 0
tmp38 = tmp37.to(tl.int64)
tmp39 = (tmp38 != 0)
tmp40 = tmp35 | tmp39
tmp41 = tmp13 == tmp26
tmp42 = tmp41 == 0
tmp43 = tmp42.to(tl.int64)
tmp44 = (tmp43 != 0)
tmp45 = tmp40 | tmp44
tl.store(out_ptr0 + (x2), tmp14, xmask)
tl.store(out_ptr1 + (x2), tmp25, xmask)
tl.store(out_ptr2 + (x2), tmp45, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/x5/cx5uvbfethxuwwkwxf3xaualzhlcwqsz4jxqpbhintggaypzjwqf.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %add_tensor : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_default_2, %primals_8), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_tensor, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_default, %sum_dim_int_list), kwargs = {})
# %logical_not_default_1 : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%any_dim,), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%logical_not_default_1, %full_default, %div_tensor), kwargs = {})
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i1', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = (xindex // 4)
x4 = xindex
x5 = xindex % 64
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last').to(tl.int1)
tmp2 = tl.load(in_out_ptr0 + (x4), xmask)
tmp3 = tl.load(in_ptr1 + (x5), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr3 + (x3), xmask, eviction_policy='evict_last')
tmp1 = tmp0 == 0
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tmp10 = 0.0
tmp11 = tl.where(tmp1, tmp10, tmp9)
tl.store(in_out_ptr0 + (x4), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vv/cvvnhithjvmvhfjufxwwzclfobkrgbyyteg66hp24r675f7elw4c.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %clone_default_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default_3,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6t/c6t5a5ere3lqjiu7zh3uu4oxmpdoujdaqqmeunxqapgzo4m74uav.py
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# context_layer_1 => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6m/c6mhj5zwirfhy5e4o45uaeov72uwfby4udubpm2fcz42iqvs2g57.py
# Topologically Sorted Source Nodes: [add_1, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add_1 => add_1
# hidden_states_2 => var_mean
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_1, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/i6/ci6ua4lfqzz3v6lbsh75noa7k5ird3udb6b5bjh7gxx4qxuz7gz3.py
# Topologically Sorted Source Nodes: [add_1, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add_1 => add_1
# hidden_states_2 => add_2, add_3, mul, mul_1, rsqrt, sub_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-12), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_2,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_1, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_11), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_12), kwargs = {})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(buf0, primals_2, buf3, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_0.run(buf1, primals_5, buf4, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 64), 0); del buf1 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf8 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.bool)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(buf5, primals_8, buf6, buf7, buf8, 64, grid=grid(64), stream=stream0)
buf9 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(buf9, buf8, primals_8, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf8
del primals_8
buf10 = reinterpret_tensor(buf7, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(buf2, primals_7, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_7
buf11 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf9, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf10, (16, 4, 1), (4, 1, 0), 0), out=buf11)
buf12 = reinterpret_tensor(buf6, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf11, buf12, 16, 4, grid=grid(16, 4), stream=stream0)
buf13 = reinterpret_tensor(buf11, (16, 4), (4, 1), 0); del buf11 # reuse
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf12, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf13)
del primals_10
buf14 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf15 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add_1, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(buf13, primals_3, buf14, buf15, 16, grid=grid(16), stream=stream0)
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add_1, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(buf13, primals_3, buf14, buf15, primals_11, primals_12, buf16, 64, grid=grid(64), stream=stream0)
del buf14
del buf15
del primals_12
return (buf16, primals_3, primals_11, buf9, reinterpret_tensor(buf10, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0), reinterpret_tensor(buf12, (16, 4), (4, 1), 0), buf13, primals_9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class LxmertAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class LxmertSelfAttentionLayer(nn.Module):
def __init__(self, config):
super().__init__()
self.self = LxmertAttention(config)
self.output = LxmertAttentionOutput(config)
def forward(self, input_tensor, attention_mask, output_attentions=False):
output = self.self(input_tensor, input_tensor, attention_mask,
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
attention_output = self.output(output[0], input_tensor)
outputs = (attention_output, attention_probs
) if output_attentions else (attention_output,)
return outputs
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, hidden_dropout_prob=0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = triton_helpers.maximum(tmp2, tmp5)
tmp9 = tmp7 + tmp8
tmp10 = triton_helpers.maximum(tmp6, tmp9)
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp10, tmp13)
tmp15 = tmp2 - tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tmp5 - tmp14
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp9 - tmp14
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp23 = tmp13 - tmp14
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = float('-inf')
tmp27 = tmp2 == tmp26
tmp28 = tmp27 == 0
tmp29 = tmp28.to(tl.int64)
tmp30 = tmp29 != 0
tmp31 = tmp5 == tmp26
tmp32 = tmp31 == 0
tmp33 = tmp32.to(tl.int64)
tmp34 = tmp33 != 0
tmp35 = tmp30 | tmp34
tmp36 = tmp9 == tmp26
tmp37 = tmp36 == 0
tmp38 = tmp37.to(tl.int64)
tmp39 = tmp38 != 0
tmp40 = tmp35 | tmp39
tmp41 = tmp13 == tmp26
tmp42 = tmp41 == 0
tmp43 = tmp42.to(tl.int64)
tmp44 = tmp43 != 0
tmp45 = tmp40 | tmp44
tl.store(out_ptr0 + x2, tmp14, xmask)
tl.store(out_ptr1 + x2, tmp25, xmask)
tl.store(out_ptr2 + x2, tmp45, xmask)
@triton.jit
def triton_poi_fused_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex // 4
x4 = xindex
x5 = xindex % 64
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp2 = tl.load(in_out_ptr0 + x4, xmask)
tmp3 = tl.load(in_ptr1 + x5, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr3 + x3, xmask, eviction_policy='evict_last')
tmp1 = tmp0 == 0
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tmp10 = 0.0
tmp11 = tl.where(tmp1, tmp10, tmp9)
tl.store(in_out_ptr0 + x4, tmp11, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-12
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(16, 4)](buf0, primals_2, buf3, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0)
del buf0
triton_poi_fused_0[grid(16, 4)](buf1, primals_5, buf4, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 64), 0)
del buf1
buf7 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf8 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.bool)
triton_poi_fused_1[grid(64)](buf5, primals_8, buf6, buf7, buf8, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf9 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused_2[grid(256)](buf9, buf8, primals_8, buf6, buf7,
256, XBLOCK=128, num_warps=4, num_stages=1)
del buf8
del primals_8
buf10 = reinterpret_tensor(buf7, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf7
triton_poi_fused_3[grid(16, 4)](buf2, primals_7, buf10, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_7
buf11 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf9, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf10, (16, 4, 1), (4, 1, 0), 0), out=buf11)
buf12 = reinterpret_tensor(buf6, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf6
triton_poi_fused_clone_4[grid(16, 4)](buf11, buf12, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf13 = reinterpret_tensor(buf11, (16, 4), (4, 1), 0)
del buf11
extern_kernels.addmm(primals_10, reinterpret_tensor(buf12, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf13)
del primals_10
buf14 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf15 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](buf13, primals_3,
buf14, buf15, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(64)](buf13, primals_3,
buf14, buf15, primals_11, primals_12, buf16, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf14
del buf15
del primals_12
return buf16, primals_3, primals_11, buf9, reinterpret_tensor(buf10, (
16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4,
1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0
), reinterpret_tensor(buf12, (16, 4), (4, 1), 0), buf13, primals_9
class LxmertAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class LxmertSelfAttentionLayerNew(nn.Module):
def __init__(self, config):
super().__init__()
self.self = LxmertAttention(config)
self.output = LxmertAttentionOutput(config)
def forward(self, input_0, input_1):
primals_1 = self.self.query.weight
primals_2 = self.self.query.bias
primals_4 = self.self.key.weight
primals_5 = self.self.key.bias
primals_6 = self.self.value.weight
primals_7 = self.self.value.bias
primals_9 = self.output.dense.weight
primals_10 = self.output.dense.bias
primals_11 = self.output.LayerNorm.weight
primals_12 = self.output.LayerNorm.bias
primals_3 = input_0
primals_8 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12])
return output[0]
| rsgit95/med_kg_txt_multimodal | LxmertSelfAttentionLayer | false | 4,223 | [
"Apache-2.0"
] | 0 | 80355b0cf58e0571531ad6f9728c533110ca996d | https://github.com/rsgit95/med_kg_txt_multimodal/tree/80355b0cf58e0571531ad6f9728c533110ca996d | from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
class LxmertAttention(nn.Module):
def __init__(self, config, ctx_dim=None):
super().__init__()
if config.hidden_size % config.num_attention_heads != 0:
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.head_size = self.num_attention_heads * self.attention_head_size
if ctx_dim is None:
ctx_dim = config.hidden_size
self.query = nn.Linear(config.hidden_size, self.head_size)
self.key = nn.Linear(ctx_dim, self.head_size)
self.value = nn.Linear(ctx_dim, self.head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, context, attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
mixed_key_layer = self.key(context)
mixed_value_layer = self.value(context)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class LxmertAttentionOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=1e-12)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.self = LxmertAttention(config)
self.output = LxmertAttentionOutput(config)
def forward(self, input_tensor, attention_mask, output_attentions=False):
output = self.self(input_tensor, input_tensor, attention_mask,
output_attentions=output_attentions)
if output_attentions:
attention_probs = output[1]
attention_output = self.output(output[0], input_tensor)
outputs = (attention_output, attention_probs
) if output_attentions else (attention_output,)
return outputs
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, hidden_dropout_prob=0.5)}]
|
SimplifiedScaledDotProductAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/yd/cydbtjoq352gcolmflbvu2nqkda7xg7q5hnvltb47jsg5dbmubym.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/hz/chz2sqsqk26mwhf2dxhgh44jfpu2er5yqjftwkzfav5ctqtx5e7f.py
# Topologically Sorted Source Nodes: [att_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# att_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/3f/c3fx6bzkalkw7u7askqdnz4rzlcoyqiec4r434sjc5x3axxgkrmr.py
# Topologically Sorted Source Nodes: [att_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# att_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 1), (16, 4, 1, 1))
assert_size_stride(primals_2, (4, 4, 4, 1), (16, 4, 1, 1))
assert_size_stride(primals_3, (4, 4, 4, 1), (16, 4, 1, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(primals_1, buf0, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 1, 4), (16, 4, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(primals_2, buf1, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf1, (16, 1, 4), (4, 0, 1), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [att_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf2, buf3, 256, grid=grid(256), stream=stream0)
buf4 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [att_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf3, buf4, 256, grid=grid(256), stream=stream0)
del buf3
buf5 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(primals_3, buf5, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_3
buf6 = reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 0), 0), out=buf6)
del buf4
buf7 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf6, buf7, 16, 4, grid=grid(16, 4), stream=stream0)
buf8 = reinterpret_tensor(buf6, (16, 4), (4, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf7, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf8)
del primals_4
del primals_5
return (reinterpret_tensor(buf8, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf7, (16, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 1), (16, 4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 1), (16, 4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 1), (16, 4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
from torch import nn
from torch.nn import init
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(SimplifiedScaledDotProductAttention, self).__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 1]), torch.rand([4, 4, 4, 1]), torch.rand(
[4, 4, 4, 1])]
def get_init_inputs():
return [[], {'d_model': 4, 'h': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 1), (16, 4, 1, 1))
assert_size_stride(primals_2, (4, 4, 4, 1), (16, 4, 1, 1))
assert_size_stride(primals_3, (4, 4, 4, 1), (16, 4, 1, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(16, 4)](primals_1, buf0, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 4, 1, 4), (16, 4, 4, 1), torch.float32)
triton_poi_fused_clone_0[grid(16, 4)](primals_2, buf1, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf1, (16, 1, 4), (4, 0, 1), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf2, buf3, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf4 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
triton_poi_fused__softmax_2[grid(256)](buf3, buf4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf3
buf5 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_clone_0[grid(16, 4)](primals_3, buf5, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf6 = reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 1), 0)
del buf0
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 0), 0), out=buf6)
del buf4
buf7 = buf5
del buf5
triton_poi_fused_clone_0[grid(16, 4)](buf6, buf7, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf8 = reinterpret_tensor(buf6, (16, 4), (4, 1), 0)
del buf6
extern_kernels.addmm(primals_5, reinterpret_tensor(buf7, (16, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf8)
del primals_4
del primals_5
return reinterpret_tensor(buf8, (4, 4, 4), (16, 4, 1), 0
), reinterpret_tensor(buf7, (16, 4), (4, 1), 0)
class SimplifiedScaledDotProductAttentionNew(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(SimplifiedScaledDotProductAttentionNew, self).__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, input_0, input_1, input_2):
primals_4 = self.fc_o.weight
primals_5 = self.fc_o.bias
primals_1 = input_0
primals_2 = input_1
primals_3 = input_2
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | SimplifiedScaledDotProductAttention | false | 4,224 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
import numpy as np
from torch import nn
from torch.nn import init
class Model(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super().__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 1]), torch.rand([4, 4, 4, 1]), torch.rand(
[4, 4, 4, 1])]
def get_init_inputs():
return [4, 4]
|
SpatialGroupEnhance | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/l3/cl35tzbhrd24dhunkbb6gjs54aklpyr46oikqhoylcgmkcmhujil.py
# Topologically Sorted Source Nodes: [adaptive_avg_pool2d], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# adaptive_avg_pool2d => mean
# Graph fragment:
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%view, [-1, -2], True), kwargs = {})
triton_per_fused_mean_0 = async_compile.triton('triton_per_fused_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[16, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mean_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/a5/ca56thseizupperaiqs5wiuvhcun4wc2jggffpd37a2xbjph27wh.py
# Topologically Sorted Source Nodes: [xn, xn_1, mean, t_1, std, std_1, mul_1, t_4, sigmoid], Original ATen: [aten.mul, aten.sum, aten.mean, aten.sub, aten.std, aten.add, aten.sigmoid]
# Source node to ATen node mapping:
# mean => mean_1
# mul_1 => mul_1
# sigmoid => sigmoid
# std => sqrt, var
# std_1 => add
# t_1 => sub
# t_4 => add_1
# xn => mul
# xn_1 => sum_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %mean), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1], True), kwargs = {})
# %mean_1 : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%view_1, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %mean_1), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%sub, [1]), kwargs = {correction: 1.0, keepdim: True})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt, 1e-05), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_2, %primals_2), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_3), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_1,), kwargs = {})
triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1 = async_compile.triton('triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 16],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: 'i32', 9: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': False, 'num_load': 10, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (64*x0)), xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + r1 + (64*x0)), xmask, other=0.0)
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (32 + r1 + (64*x0)), xmask, other=0.0)
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (48 + r1 + (64*x0)), xmask, other=0.0)
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp44 = tl.load(in_ptr2 + (0))
tmp45 = tl.broadcast_to(tmp44, [XBLOCK, RBLOCK])
tmp47 = tl.load(in_ptr3 + (0))
tmp48 = tl.broadcast_to(tmp47, [XBLOCK, RBLOCK])
tmp2 = tmp0 * tmp1
tmp5 = tmp3 * tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 * tmp12
tmp14 = tmp10 + tmp13
tmp15 = tl.broadcast_to(tmp14, [XBLOCK, RBLOCK])
tmp17 = tl.where(xmask, tmp15, 0)
tmp18 = tl.sum(tmp17, 1)[:, None]
tmp19 = 16.0
tmp20 = tmp18 / tmp19
tmp21 = tmp14 - tmp20
tmp22 = tl.broadcast_to(tmp21, [XBLOCK, RBLOCK])
tmp24 = tl.where(xmask, tmp22, 0)
tmp25 = tl.broadcast_to(tmp22, [XBLOCK, RBLOCK])
tmp27 = tl.where(xmask, tmp25, 0)
tmp28 = tl.sum(tmp27, 1)[:, None]
tmp29 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp30 = tmp29.to(tl.float32)
tmp31 = tmp28 / tmp30
tmp32 = tmp22 - tmp31
tmp33 = tmp32 * tmp32
tmp34 = tl.broadcast_to(tmp33, [XBLOCK, RBLOCK])
tmp36 = tl.where(xmask, tmp34, 0)
tmp37 = tl.sum(tmp36, 1)[:, None]
tmp38 = 15.0
tmp39 = tmp37 / tmp38
tmp40 = libdevice.sqrt(tmp39)
tmp41 = 1e-05
tmp42 = tmp40 + tmp41
tmp43 = tmp21 / tmp42
tmp46 = tmp43 * tmp45
tmp49 = tmp46 + tmp48
tmp50 = tl.sigmoid(tmp49)
tl.store(out_ptr0 + (r1 + (16*x0)), tmp14, xmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp20, xmask)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x0), tmp42, xmask)
tl.store(out_ptr1 + (r1 + (16*x0)), tmp50, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/oc/cocahikgrum5ddfascfd75hjn6cow7rd2ztrvh7tsckzl7n6vq6q.py
# Topologically Sorted Source Nodes: [mul_1, t_4, sigmoid, x_1], Original ATen: [aten.mul, aten.add, aten.sigmoid]
# Source node to ATen node mapping:
# mul_1 => mul_1
# sigmoid => sigmoid
# t_4 => add_1
# x_1 => mul_2
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_2, %primals_2), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_3), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_1,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %sigmoid), kwargs = {})
triton_poi_fused_add_mul_sigmoid_2 = async_compile.triton('triton_poi_fused_add_mul_sigmoid_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_sigmoid_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_sigmoid_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_3, (1, 1, 1, 1), (1, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
buf1 = reinterpret_tensor(buf0, (4, 4, 1, 1), (4, 1, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [adaptive_avg_pool2d], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 16, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf3 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf4 = reinterpret_tensor(buf3, (4, 1), (1, 1), 0); del buf3 # reuse
buf6 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf8 = reinterpret_tensor(buf6, (4, 1), (1, 1), 0); del buf6 # reuse
buf9 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [xn, xn_1, mean, t_1, std, std_1, mul_1, t_4, sigmoid], Original ATen: [aten.mul, aten.sum, aten.mean, aten.sub, aten.std, aten.add, aten.sigmoid]
triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1.run(buf4, buf8, primals_1, buf1, primals_2, primals_3, buf2, buf9, 4, 16, grid=grid(4), stream=stream0)
del buf2
buf10 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_1, t_4, sigmoid, x_1], Original ATen: [aten.mul, aten.add, aten.sigmoid]
triton_poi_fused_add_mul_sigmoid_2.run(primals_1, buf9, buf10, 256, grid=grid(256), stream=stream0)
del buf9
return (buf10, primals_1, primals_2, primals_3, buf1, buf4, buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
from torch.nn import init
class SpatialGroupEnhance(nn.Module):
def __init__(self, groups):
super().__init__()
self.groups = groups
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.weight = nn.Parameter(torch.zeros(1, groups, 1, 1))
self.bias = nn.Parameter(torch.zeros(1, groups, 1, 1))
self.sig = nn.Sigmoid()
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, x):
b, c, h, w = x.shape
x = x.view(b * self.groups, -1, h, w)
xn = x * self.avg_pool(x)
xn = xn.sum(dim=1, keepdim=True)
t = xn.view(b * self.groups, -1)
t = t - t.mean(dim=1, keepdim=True)
std = t.std(dim=1, keepdim=True) + 1e-05
t = t / std
t = t.view(b, self.groups, h, w)
t = t * self.weight + self.bias
t = t.view(b * self.groups, 1, h, w)
x = x * self.sig(t)
x = x.view(b, c, h, w)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'groups': 1}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1(in_out_ptr0,
in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1,
xnumel, rnumel, XBLOCK: tl.constexpr):
xnumel = 4
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 64 * x0), xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + r1 + 64 * x0), xmask, other=0.0)
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (32 + r1 + 64 * x0), xmask, other=0.0)
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (48 + r1 + 64 * x0), xmask, other=0.0)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp44 = tl.load(in_ptr2 + 0)
tmp45 = tl.broadcast_to(tmp44, [XBLOCK, RBLOCK])
tmp47 = tl.load(in_ptr3 + 0)
tmp48 = tl.broadcast_to(tmp47, [XBLOCK, RBLOCK])
tmp2 = tmp0 * tmp1
tmp5 = tmp3 * tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 * tmp12
tmp14 = tmp10 + tmp13
tmp15 = tl.broadcast_to(tmp14, [XBLOCK, RBLOCK])
tmp17 = tl.where(xmask, tmp15, 0)
tmp18 = tl.sum(tmp17, 1)[:, None]
tmp19 = 16.0
tmp20 = tmp18 / tmp19
tmp21 = tmp14 - tmp20
tmp22 = tl.broadcast_to(tmp21, [XBLOCK, RBLOCK])
tl.where(xmask, tmp22, 0)
tmp25 = tl.broadcast_to(tmp22, [XBLOCK, RBLOCK])
tmp27 = tl.where(xmask, tmp25, 0)
tmp28 = tl.sum(tmp27, 1)[:, None]
tmp29 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp30 = tmp29.to(tl.float32)
tmp31 = tmp28 / tmp30
tmp32 = tmp22 - tmp31
tmp33 = tmp32 * tmp32
tmp34 = tl.broadcast_to(tmp33, [XBLOCK, RBLOCK])
tmp36 = tl.where(xmask, tmp34, 0)
tmp37 = tl.sum(tmp36, 1)[:, None]
tmp38 = 15.0
tmp39 = tmp37 / tmp38
tmp40 = libdevice.sqrt(tmp39)
tmp41 = 1e-05
tmp42 = tmp40 + tmp41
tmp43 = tmp21 / tmp42
tmp46 = tmp43 * tmp45
tmp49 = tmp46 + tmp48
tmp50 = tl.sigmoid(tmp49)
tl.store(out_ptr0 + (r1 + 16 * x0), tmp14, xmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp20, xmask)
tl.debug_barrier()
tl.store(in_out_ptr1 + x0, tmp42, xmask)
tl.store(out_ptr1 + (r1 + 16 * x0), tmp50, xmask)
@triton.jit
def triton_poi_fused_add_mul_sigmoid_2(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_3, (1, 1, 1, 1), (1, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
buf1 = reinterpret_tensor(buf0, (4, 4, 1, 1), (4, 1, 1, 1), 0)
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(16)](buf1, primals_1, 16, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf2 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf3 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf4 = reinterpret_tensor(buf3, (4, 1), (1, 1), 0)
del buf3
buf6 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf8 = reinterpret_tensor(buf6, (4, 1), (1, 1), 0)
del buf6
buf9 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
triton_per_fused_add_mean_mul_sigmoid_std_sub_sum_1[grid(4)](buf4,
buf8, primals_1, buf1, primals_2, primals_3, buf2, buf9, 4, 16,
XBLOCK=1, num_warps=2, num_stages=1)
del buf2
buf10 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_mul_sigmoid_2[grid(256)](primals_1, buf9,
buf10, 256, XBLOCK=256, num_warps=4, num_stages=1)
del buf9
return buf10, primals_1, primals_2, primals_3, buf1, buf4, buf8
class SpatialGroupEnhanceNew(nn.Module):
def __init__(self, groups):
super().__init__()
self.groups = groups
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.weight = nn.Parameter(torch.zeros(1, groups, 1, 1))
self.bias = nn.Parameter(torch.zeros(1, groups, 1, 1))
self.sig = nn.Sigmoid()
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, input_0):
primals_2 = self.weight
primals_3 = self.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | SpatialGroupEnhance | false | 4,225 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
from torch import nn
from torch.nn import init
class Model(nn.Module):
def __init__(self, groups):
super().__init__()
self.groups = groups
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.weight = nn.Parameter(torch.zeros(1, groups, 1, 1))
self.bias = nn.Parameter(torch.zeros(1, groups, 1, 1))
self.sig = nn.Sigmoid()
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, x):
b, c, h, w = x.shape
x = x.view(b * self.groups, -1, h, w)
xn = x * self.avg_pool(x)
xn = xn.sum(dim=1, keepdim=True)
t = xn.view(b * self.groups, -1)
t = t - t.mean(dim=1, keepdim=True)
std = t.std(dim=1, keepdim=True) + 1e-05
t = t / std
t = t.view(b, self.groups, h, w)
t = t * self.weight + self.bias
t = t.view(b * self.groups, 1, h, w)
x = x * self.sig(t)
x = x.view(b, c, h, w)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [1]
|
ScaledDotProductAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/xe/cxeuttfzx4xq2jmzwzvkech4crjirky5wjckb34lnep5o6sog3uw.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1) + (64*x3)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x4), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/fn/cfnr6wn6wbusamhilcgctjberp7g5kksyakcze32k6ntswznc2de.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_1 = async_compile.triton('triton_poi_fused_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 16
y1 = (yindex // 16)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ka/ckaneo6wn23ipwgbubou64jdtwieswlrn7w7r7kqky4aagh3v6l3.py
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
# Source node to ATen node mapping:
# att_1 => exp
# wrapped_sqrt => full_default
# Graph fragment:
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([], 2.0), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %scalar_tensor_default : [num_users=2] = call_function[target=torch.ops.aten.scalar_tensor.default](args = (1,), kwargs = {dtype: torch.float32, device: cuda:0, pin_memory: False})
# %ge_scalar : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%full_default, 0), kwargs = {})
# %neg_default : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%scalar_tensor_default,), kwargs = {})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ge_scalar, %scalar_tensor_default, %neg_default), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_11, %where_self), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where_self, %full_default), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, %mul_tensor_1), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
triton_poi_fused__softmax_sqrt_2 = async_compile.triton('triton_poi_fused__softmax_sqrt_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_sqrt_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_sqrt_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp8 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = tl.full([1], 2.0, tl.float64)
tmp2 = tl.full([1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp9 = tmp8 * tmp6
tmp11 = tmp10 * tmp6
tmp12 = triton_helpers.maximum(tmp9, tmp11)
tmp14 = tmp13 * tmp6
tmp15 = triton_helpers.maximum(tmp12, tmp14)
tmp17 = tmp16 * tmp6
tmp18 = triton_helpers.maximum(tmp15, tmp17)
tmp19 = tmp7 - tmp18
tmp20 = tmp6.to(tl.float64)
tmp21 = tmp20 * tmp1
tmp22 = tmp21.to(tl.float32)
tmp23 = tmp19 / tmp22
tmp24 = tl_math.exp(tmp23)
tl.store(out_ptr0 + (x2), tmp24, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ry/cryn7ntc2gpkbfzbre3xh7lffx7zkbskw6oihbzsekkgajmdbki6.py
# Topologically Sorted Source Nodes: [att_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# att_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6b/c6busvilz5nn36jjet3bmw7cqddirh4sgalamjr3fsrp3sbsacfi.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_6,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1) + (64*x3)), xmask)
tl.store(out_ptr0 + (x4), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (16, 4), (4, 1))
assert_size_stride(primals_4, (16, ), (1, ))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16, ), (1, ))
assert_size_stride(primals_7, (16, 4), (4, 1))
assert_size_stride(primals_8, (16, ), (1, ))
assert_size_stride(primals_9, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_10, (4, 16), (16, 1))
assert_size_stride(primals_11, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 16), (1, 4), 0), out=buf0)
del primals_3
buf1 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf1)
del primals_5
buf2 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_9, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 16), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, primals_4, buf3, 256, grid=grid(256), stream=stream0)
del primals_4
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf1, primals_6, buf4, 64, 4, grid=grid(64, 4), stream=stream0)
del primals_6
buf5 = reinterpret_tensor(buf1, (16, 4, 4), (16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
triton_poi_fused__softmax_sqrt_2.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [att_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
buf8 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf2, primals_8, buf8, 256, grid=grid(256), stream=stream0)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf8, (16, 4, 4), (16, 4, 1), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf9, buf10, 256, grid=grid(256), stream=stream0)
del buf9
buf11 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, reinterpret_tensor(buf10, (16, 16), (16, 1), 0), reinterpret_tensor(primals_10, (16, 4), (1, 16), 0), alpha=1, beta=1, out=buf11)
del primals_11
return (reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (16, 4), (4, 1), 0), buf7, reinterpret_tensor(buf10, (16, 16), (16, 1), 0), primals_10, reinterpret_tensor(buf8, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf3, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf4, (16, 4, 4), (16, 1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
from torch import nn
from torch.nn import init
class ScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, d_k, d_v, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(ScaledDotProductAttention, self).__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = self.fc_k(keys).view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = self.fc_v(values).view(b_s, nk, self.h, self.d_v).permute(0, 2,
1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [[], {'d_model': 4, 'd_k': 4, 'd_v': 4, 'h': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1 + 64 * x3), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x4, tmp2, xmask)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 16
y1 = yindex // 16
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_sqrt_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp8 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp13 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp16 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = tl.full([1], 2.0, tl.float64)
tmp2 = tl.full([1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp9 = tmp8 * tmp6
tmp11 = tmp10 * tmp6
tmp12 = triton_helpers.maximum(tmp9, tmp11)
tmp14 = tmp13 * tmp6
tmp15 = triton_helpers.maximum(tmp12, tmp14)
tmp17 = tmp16 * tmp6
tmp18 = triton_helpers.maximum(tmp15, tmp17)
tmp19 = tmp7 - tmp18
tmp20 = tmp6.to(tl.float64)
tmp21 = tmp20 * tmp1
tmp22 = tmp21.to(tl.float32)
tmp23 = tmp19 / tmp22
tmp24 = tl_math.exp(tmp23)
tl.store(out_ptr0 + x2, tmp24, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1 + 64 * x3), xmask)
tl.store(out_ptr0 + x4, tmp0, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (16, 4), (4, 1))
assert_size_stride(primals_4, (16,), (1,))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16,), (1,))
assert_size_stride(primals_7, (16, 4), (4, 1))
assert_size_stride(primals_8, (16,), (1,))
assert_size_stride(primals_9, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_10, (4, 16), (16, 1))
assert_size_stride(primals_11, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 16), (1, 4), 0), out=buf0)
del primals_3
buf1 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf1)
del primals_5
buf2 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_9, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 16), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(256)](buf0, primals_4, buf3, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_4
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
triton_poi_fused_clone_1[grid(64, 4)](buf1, primals_6, buf4, 64, 4,
XBLOCK=4, YBLOCK=32, num_warps=4, num_stages=1)
del primals_6
buf5 = reinterpret_tensor(buf1, (16, 4, 4), (16, 4, 1), 0)
del buf1
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_sqrt_2[grid(256)](buf5, buf6, 256, XBLOCK
=128, num_warps=4, num_stages=1)
buf7 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused__softmax_3[grid(256)](buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf8 = buf6
del buf6
triton_poi_fused_clone_0[grid(256)](buf2, primals_8, buf8, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf8, (16, 4, 4), (16, 4, 1), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_4[grid(256)](buf9, buf10, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf9
buf11 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_11, reinterpret_tensor(buf10, (16, 16),
(16, 1), 0), reinterpret_tensor(primals_10, (16, 4), (1, 16), 0
), alpha=1, beta=1, out=buf11)
del primals_11
return reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0
), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_9, (16, 4), (4, 1), 0
), buf7, reinterpret_tensor(buf10, (16, 16), (16, 1), 0
), primals_10, reinterpret_tensor(buf8, (16, 4, 4), (16, 1, 4), 0
), reinterpret_tensor(buf3, (16, 4, 4), (16, 1, 4), 0
), reinterpret_tensor(buf4, (16, 4, 4), (16, 1, 4), 0)
class ScaledDotProductAttentionNew(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, d_k, d_v, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(ScaledDotProductAttentionNew, self).__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, input_0, input_1, input_2):
primals_3 = self.fc_q.weight
primals_4 = self.fc_q.bias
primals_5 = self.fc_k.weight
primals_6 = self.fc_k.bias
primals_7 = self.fc_v.weight
primals_8 = self.fc_v.bias
primals_10 = self.fc_o.weight
primals_11 = self.fc_o.bias
primals_1 = input_0
primals_2 = input_1
primals_9 = input_2
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | ScaledDotProductAttention | false | 4,226 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
import numpy as np
from torch import nn
from torch.nn import init
class Model(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, d_k, d_v, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super().__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = self.fc_k(keys).view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = self.fc_v(values).view(b_s, nk, self.h, self.d_v).permute(0, 2,
1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [4, 4, 4, 4]
|
AttentionHead | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/fz/cfzmg4qtw6jgry4nhlwopodzjz62ll3n3ykfox77hwd2crdnlh2w.py
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => exp
# Graph fragment:
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%bmm, 1), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, 2.0), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp3 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = 0.5
tmp16 = tmp14 * tmp15
tmp17 = tl_math.exp(tmp16)
tl.store(out_ptr0 + (x2), tmp17, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/kj/ckjtlefzavjukjsytvkak6ek26zmzexpcbnlwelx4k5kascjxlf3.py
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_4
del primals_5
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_9, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_7
del primals_8
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [temp], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf1, (4, 4, 4), (16, 1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf3, buf4, 64, grid=grid(64), stream=stream0)
buf5 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf4, buf5, 64, grid=grid(64), stream=stream0)
buf6 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [bmm_1], Original ATen: [aten.bmm]
extern_kernels.bmm(buf5, reinterpret_tensor(buf2, (4, 4, 4), (16, 4, 1), 0), out=buf6)
return (buf6, reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (16, 4), (4, 1), 0), buf5, reinterpret_tensor(buf2, (4, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf0, (4, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import Tensor
import torch.nn as nn
import torch.nn.functional as F
from torch.functional import Tensor
def scaled_dot_product_attention(query: 'torch.Tensor', key: 'torch.Tensor',
value: 'torch.Tensor') ->Tensor:
temp = query.bmm(key.transpose(1, 2))
scale = query.size(-1) ** 0.5
softmax = F.softmax(temp / scale, dim=-1)
return softmax.bmm(value)
class AttentionHead(nn.Module):
def __init__(self, dim_in: 'int', dim_k: 'int', dim_v: 'int'):
super().__init__()
self.q = nn.Linear(dim_in, dim_k)
self.k = nn.Linear(dim_in, dim_k)
self.v = nn.Linear(dim_in, dim_v)
def forward(self, query: 'torch.Tensor', key: 'torch.Tensor', value:
'Tensor') ->Tensor:
return scaled_dot_product_attention(self.q(query), self.k(key),
self.v(value))
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [[], {'dim_in': 4, 'dim_k': 4, 'dim_v': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import Tensor
import torch.nn as nn
import torch.nn.functional as F
from torch.functional import Tensor
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp3 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = 0.5
tmp16 = tmp14 * tmp15
tmp17 = tl_math.exp(tmp16)
tl.store(out_ptr0 + x2, tmp17, xmask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (16,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_6, (16,
4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf1)
del primals_4
del primals_5
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_9, (16,
4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf2)
del primals_7
del primals_8
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (4, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf1, (4, 4, 4), (16, 1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(64)](buf3, buf4, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf5 = buf3
del buf3
triton_poi_fused__softmax_1[grid(64)](buf4, buf5, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf6 = buf4
del buf4
extern_kernels.bmm(buf5, reinterpret_tensor(buf2, (4, 4, 4), (16, 4,
1), 0), out=buf6)
return buf6, reinterpret_tensor(primals_3, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_6, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_9, (16, 4), (4, 1), 0
), buf5, reinterpret_tensor(buf2, (4, 4, 4), (16, 1, 4), 0
), reinterpret_tensor(buf0, (4, 4, 4), (16, 1, 4), 0
), reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0)
def scaled_dot_product_attention(query: 'torch.Tensor', key: 'torch.Tensor',
value: 'torch.Tensor') ->Tensor:
temp = query.bmm(key.transpose(1, 2))
scale = query.size(-1) ** 0.5
softmax = F.softmax(temp / scale, dim=-1)
return softmax.bmm(value)
class AttentionHeadNew(nn.Module):
def __init__(self, dim_in: 'int', dim_k: 'int', dim_v: 'int'):
super().__init__()
self.q = nn.Linear(dim_in, dim_k)
self.k = nn.Linear(dim_in, dim_k)
self.v = nn.Linear(dim_in, dim_v)
def forward(self, input_0, input_1, input_2):
primals_1 = self.q.weight
primals_2 = self.q.bias
primals_4 = self.k.weight
primals_5 = self.k.bias
primals_7 = self.v.weight
primals_8 = self.v.bias
primals_3 = input_0
primals_6 = input_1
primals_9 = input_2
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
| sabernn/vit-pytorch | AttentionHead | false | 4,227 | [
"MIT"
] | 0 | 21a2671aa92adb941a56ae629f6089f550949fb2 | https://github.com/sabernn/vit-pytorch/tree/21a2671aa92adb941a56ae629f6089f550949fb2 | import torch
from torch import Tensor
import torch.nn as nn
import torch.nn.functional as F
from torch.functional import Tensor
def scaled_dot_product_attention(query: 'torch.Tensor', key: 'torch.Tensor',
value: 'torch.Tensor') ->Tensor:
temp = query.bmm(key.transpose(1, 2))
scale = query.size(-1) ** 0.5
softmax = F.softmax(temp / scale, dim=-1)
return softmax.bmm(value)
class Model(nn.Module):
def __init__(self, dim_in: 'int', dim_k: 'int', dim_v: 'int'):
super().__init__()
self.q = nn.Linear(dim_in, dim_k)
self.k = nn.Linear(dim_in, dim_k)
self.v = nn.Linear(dim_in, dim_v)
def forward(self, query: 'torch.Tensor', key: 'torch.Tensor', value:
'Tensor') ->Tensor:
return scaled_dot_product_attention(self.q(query), self.k(key),
self.v(value))
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [4, 4, 4]
|
SE_Connect | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/jn/cjnv5uptstyk4xaisuiw5kf5lbz3m33meejxhbfbsta5ozps7ijn.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# out => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [2]), kwargs = {})
triton_poi_fused_mean_0 = async_compile.triton('triton_poi_fused_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mean_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mean_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + (16*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + (16*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + (16*x1)), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ip/cip355a7nnydvbbk53yzzlgfxtclbx4sdaz6diiadsc7bk4g3ikp.py
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# out_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = 0.0
tmp7 = tmp5 <= tmp6
tl.store(in_out_ptr0 + (x0), tmp5, xmask)
tl.store(out_ptr0 + (x0), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/cm/ccmo3ssy4it32zgmnziqn6cih5z7bew4voyzb4nxpajh2zquk7fp.py
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# out_3 => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %unsqueeze), kwargs = {})
triton_poi_fused_mul_2 = async_compile.triton('triton_poi_fused_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + (x3), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
assert_size_stride(primals_3, (1, ), (1, ))
assert_size_stride(primals_4, (4, 1), (1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_poi_fused_mean_0.run(primals_1, buf0, 64, grid=grid(64), stream=stream0)
buf1 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 1), (1, 4), 0), out=buf1)
del primals_2
buf2 = reinterpret_tensor(buf1, (4, 4, 1), (4, 1, 1), 0); del buf1 # reuse
buf5 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf2, primals_3, buf5, 16, grid=grid(16), stream=stream0)
del primals_3
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf2, (16, 1), (1, 0), 0), reinterpret_tensor(primals_4, (1, 4), (1, 1), 0), alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(primals_1, buf3, buf4, 256, grid=grid(256), stream=stream0)
return (buf4, primals_1, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(buf2, (16, 1), (1, 1), 0), buf3, primals_4, buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn
import torch.nn as nn
class SE_Connect(nn.Module):
def __init__(self, channels, s=4):
super().__init__()
assert channels % s == 0, '{} % {} != 0'.format(channesl, s)
self.linear1 = nn.Linear(channels, channels // s)
self.linear2 = nn.Linear(channels // s, channels)
def forward(self, x):
out = x.mean(dim=2)
out = F.relu(self.linear1(out))
out = torch.sigmoid(self.linear2(out))
out = x * out.unsqueeze(2)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_mean_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + 16 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + 16 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + 16 * x1), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = 0.0
tmp7 = tmp5 <= tmp6
tl.store(in_out_ptr0 + x0, tmp5, xmask)
tl.store(out_ptr0 + x0, tmp7, xmask)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + x3, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
assert_size_stride(primals_3, (1,), (1,))
assert_size_stride(primals_4, (4, 1), (1, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mean_0[grid(64)](primals_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 1), (1, 4), 0), out=buf1)
del primals_2
buf2 = reinterpret_tensor(buf1, (4, 4, 1), (4, 1, 1), 0)
del buf1
buf5 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(16)](buf2,
primals_3, buf5, 16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf2, (16, 1), (
1, 0), 0), reinterpret_tensor(primals_4, (1, 4), (1, 1), 0),
alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_2[grid(256)](primals_1, buf3, buf4, 256,
XBLOCK=256, num_warps=4, num_stages=1)
return buf4, primals_1, reinterpret_tensor(buf0, (16, 4), (4, 1), 0
), reinterpret_tensor(buf2, (16, 1), (1, 1), 0), buf3, primals_4, buf5
class SE_ConnectNew(nn.Module):
def __init__(self, channels, s=4):
super().__init__()
assert channels % s == 0, '{} % {} != 0'.format(channesl, s)
self.linear1 = nn.Linear(channels, channels // s)
self.linear2 = nn.Linear(channels // s, channels)
def forward(self, input_0):
primals_2 = self.linear1.weight
primals_3 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| qlindazm/asv-subtools | SE_Connect | false | 4,228 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn.functional as F
import torch.nn
import torch.nn as nn
class Model(nn.Module):
def __init__(self, channels, s=4):
super().__init__()
assert channels % s == 0, '{} % {} != 0'.format(channesl, s)
self.linear1 = nn.Linear(channels, channels // s)
self.linear2 = nn.Linear(channels // s, channels)
def forward(self, x):
out = x.mean(dim=2)
out = F.relu(self.linear1(out))
out = torch.sigmoid(self.linear2(out))
out = x * out.unsqueeze(2)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
AttentiveStatsPool | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/lv/clvzonxnhy6fl5yrkxrr6ovbvqty6idyvwskbymkmtmh6lwq4ump.py
# Topologically Sorted Source Nodes: [conv1d, alpha], Original ATen: [aten.convolution, aten.tanh]
# Source node to ATen node mapping:
# alpha => tanh
# conv1d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1], [0], [1], False, [0], 1), kwargs = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_tanh_0 = async_compile.triton('triton_poi_fused_convolution_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_tanh_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 4) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x3), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/au/cau4pihcaptiev5y2ewn2o2nvrwhk7hogc72cofmmtbyv4rxc2oy.py
# Topologically Sorted Source Nodes: [conv1d_1], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv1d_1 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%tanh, %primals_4, %primals_5, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 4) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/hg/chg3iq6bscxmmxv5f7tuzgwycb4mgrimwfhv2nauw5rj4tt5cmv2.py
# Topologically Sorted Source Nodes: [alpha_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# alpha_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%convolution_1, [2], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%convolution_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/zu/czuvep3dmpmqmhiiliwubh4ghdt2qr27va67sszkua7trziinwov.py
# Topologically Sorted Source Nodes: [alpha_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# alpha_1 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [2], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/js/cjsld3wpkuqa2jsc3754couxrmwao7s4uswgaey3epfkafdnydkc.py
# Topologically Sorted Source Nodes: [mul, mean, pow_1, mul_1, sum_2, pow_2, residuals, clamp, std], Original ATen: [aten.mul, aten.sum, aten.pow, aten.sub, aten.clamp, aten.sqrt]
# Source node to ATen node mapping:
# clamp => clamp_min
# mean => sum_2
# mul => mul
# mul_1 => mul_1
# pow_1 => pow_1
# pow_2 => pow_2
# residuals => sub_1
# std => sqrt
# sum_2 => sum_3
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %primals_3), kwargs = {})
# %sum_2 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [2]), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%primals_3, 2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %pow_1), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [2]), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 2), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sum_3, %pow_2), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%sub_1, 1e-09), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%clamp_min,), kwargs = {})
triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4 = async_compile.triton('triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4(in_ptr0, in_ptr1, out_ptr0, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tmp5 = tmp3 * tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 * tmp12
tmp14 = tmp10 + tmp13
tmp15 = tmp1 * tmp1
tmp16 = tmp0 * tmp15
tmp17 = tmp4 * tmp4
tmp18 = tmp3 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp8 * tmp8
tmp21 = tmp7 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = tmp12 * tmp12
tmp24 = tmp11 * tmp23
tmp25 = tmp22 + tmp24
tmp26 = tmp14 * tmp14
tmp27 = tmp25 - tmp26
tmp28 = 1e-09
tmp29 = triton_helpers.maximum(tmp27, tmp28)
tmp30 = libdevice.sqrt(tmp29)
tl.store(out_ptr0 + (x0 + (8*x1)), tmp14, xmask)
tl.store(out_ptr2 + (x0 + (8*x1)), tmp30, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv1d, alpha], Original ATen: [aten.convolution, aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_tanh_0.run(buf1, primals_2, 64, grid=grid(64), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [conv1d_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 4), (16, 4, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [conv1d_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf3, primals_5, 64, grid=grid(64), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [alpha_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf3, buf4, 64, grid=grid(64), stream=stream0)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [alpha_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf4, buf5, 64, grid=grid(64), stream=stream0)
del buf4
buf9 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
buf6 = reinterpret_tensor(buf9, (4, 4), (8, 1), 0) # alias
buf8 = reinterpret_tensor(buf9, (4, 4), (8, 1), 4) # alias
# Topologically Sorted Source Nodes: [mul, mean, pow_1, mul_1, sum_2, pow_2, residuals, clamp, std], Original ATen: [aten.mul, aten.sum, aten.pow, aten.sub, aten.clamp, aten.sqrt]
triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4.run(buf5, primals_3, buf6, buf8, 16, grid=grid(16), stream=stream0)
del buf5
return (buf9, primals_1, primals_3, primals_4, buf1, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn
import torch.nn as nn
class AttentiveStatsPool(nn.Module):
def __init__(self, in_dim, bottleneck_dim):
super().__init__()
self.linear1 = nn.Conv1d(in_dim, bottleneck_dim, kernel_size=1)
self.linear2 = nn.Conv1d(bottleneck_dim, in_dim, kernel_size=1)
def forward(self, x):
alpha = torch.tanh(self.linear1(x))
alpha = torch.softmax(self.linear2(alpha), dim=2)
mean = torch.sum(alpha * x, dim=2)
residuals = torch.sum(alpha * x ** 2, dim=2) - mean ** 2
std = torch.sqrt(residuals.clamp(min=1e-09))
return torch.cat([mean, std], dim=1)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'in_dim': 4, 'bottleneck_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_tanh_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 4 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x3, tmp3, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 4 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4(in_ptr0, in_ptr1,
out_ptr0, out_ptr2, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x2, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x2), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 * tmp1
tmp5 = tmp3 * tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 * tmp12
tmp14 = tmp10 + tmp13
tmp15 = tmp1 * tmp1
tmp16 = tmp0 * tmp15
tmp17 = tmp4 * tmp4
tmp18 = tmp3 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp8 * tmp8
tmp21 = tmp7 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = tmp12 * tmp12
tmp24 = tmp11 * tmp23
tmp25 = tmp22 + tmp24
tmp26 = tmp14 * tmp14
tmp27 = tmp25 - tmp26
tmp28 = 1e-09
tmp29 = triton_helpers.maximum(tmp27, tmp28)
tmp30 = libdevice.sqrt(tmp29)
tl.store(out_ptr0 + (x0 + 8 * x1), tmp14, xmask)
tl.store(out_ptr2 + (x0 + 8 * x1), tmp30, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_tanh_0[grid(64)](buf1, primals_2, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 4), (16, 4, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_1[grid(64)](buf3, primals_5, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(64)](buf3, buf4, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_3[grid(64)](buf4, buf5, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf4
buf9 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
buf6 = reinterpret_tensor(buf9, (4, 4), (8, 1), 0)
buf8 = reinterpret_tensor(buf9, (4, 4), (8, 1), 4)
triton_poi_fused_clamp_mul_pow_sqrt_sub_sum_4[grid(16)](buf5,
primals_3, buf6, buf8, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf5
return buf9, primals_1, primals_3, primals_4, buf1, buf3
class AttentiveStatsPoolNew(nn.Module):
def __init__(self, in_dim, bottleneck_dim):
super().__init__()
self.linear1 = nn.Conv1d(in_dim, bottleneck_dim, kernel_size=1)
self.linear2 = nn.Conv1d(bottleneck_dim, in_dim, kernel_size=1)
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| qlindazm/asv-subtools | AttentiveStatsPool | false | 4,229 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn
import torch.nn as nn
class Model(nn.Module):
def __init__(self, in_dim, bottleneck_dim):
super().__init__()
self.linear1 = nn.Conv1d(in_dim, bottleneck_dim, kernel_size=1)
self.linear2 = nn.Conv1d(bottleneck_dim, in_dim, kernel_size=1)
def forward(self, x):
alpha = torch.tanh(self.linear1(x))
alpha = torch.softmax(self.linear2(alpha), dim=2)
mean = torch.sum(alpha * x, dim=2)
residuals = torch.sum(alpha * x ** 2, dim=2) - mean ** 2
std = torch.sqrt(residuals.clamp(min=1e-09))
return torch.cat([mean, std], dim=1)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
OutlookAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/6x/c6xdtz2w5en5fsdbgutlchlfsh4q7a2byarfiaglzh45nn222wce.py
# Topologically Sorted Source Nodes: [unfold], Original ATen: [aten.im2col]
# Source node to ATen node mapping:
# unfold => add
# Graph fragment:
# %add : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%unsqueeze, %unsqueeze_1), kwargs = {})
triton_poi_fused_im2col_0 = async_compile.triton('triton_poi_fused_im2col_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0,), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_im2col_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_im2col_0(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 12
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = x0 + x1
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/7h/c7howbh27c6wmleecf2uzap4cbx7ucljylpyhqy6ghbnqufvr5po.py
# Topologically Sorted Source Nodes: [avg_pool2d], Original ATen: [aten.avg_pool2d]
# Source node to ATen node mapping:
# avg_pool2d => avg_pool2d
# Graph fragment:
# %avg_pool2d : [num_users=1] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%permute_4, [1, 1], [1, 1], [0, 0], True), kwargs = {})
triton_poi_fused_avg_pool2d_1 = async_compile.triton('triton_poi_fused_avg_pool2d_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_avg_pool2d_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_avg_pool2d_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/qu/cquhnm52pqnjqmg225t6c6byordni5rqv6e4jn22ez5xcptmtplx.py
# Topologically Sorted Source Nodes: [attn_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_3 => div, exp, sum_1
# Graph fragment:
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_7, 1), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_tensor, 0.5), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%mul_tensor_1,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_per_fused__softmax_2 = async_compile.triton('triton_per_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1024, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_2(in_ptr0, in_ptr1, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 576
rnumel = 9
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = rindex < rnumel
r2 = rindex
x5 = xindex
x0 = xindex % 9
x4 = (xindex // 144)
x6 = xindex % 144
tmp0 = tl.load(in_ptr0 + (r2 + (9*x5)), rmask & xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + (r2 + (9*x0)), rmask & xmask, eviction_policy='evict_last', other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.where(rmask & xmask, tmp5, float("-inf"))
tmp8 = triton_helpers.max2(tmp7, 1)[:, None]
tmp9 = tmp4 - tmp8
tmp10 = 0.5
tmp11 = tmp9 * tmp10
tmp12 = tl_math.exp(tmp11)
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(rmask & xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = tmp12 / tmp16
tl.store(out_ptr2 + (r2 + (9*x6) + (1312*x4)), tmp17, rmask & xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ey/ceyb3am47ifxesuvzrr26eydaluvdgp3oxv24nkd3mmm7x7jwygu.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul => clone_2
# Graph fragment:
# %clone_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_3 = async_compile.triton('triton_poi_fused_clone_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 2304
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4) % 9
x2 = (xindex // 36) % 16
x0 = xindex % 4
x3 = (xindex // 576)
x4 = xindex
tmp0 = tl.load(in_ptr0 + ((4*(x1 // 3)) + (x2 // 4)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + ((4*(x1 % 3)) + (x2 % 4)), xmask, eviction_policy='evict_last')
tmp1 = tl.full([XBLOCK], 6, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert(((0 <= tmp4) & (tmp4 < 6)) | ~(xmask), "index out of bounds: 0 <= tmp4 < 6")
tmp7 = tmp6 + tmp1
tmp8 = tmp6 < 0
tmp9 = tl.where(tmp8, tmp7, tmp6)
tl.device_assert(((0 <= tmp9) & (tmp9 < 6)) | ~(xmask), "index out of bounds: 0 <= tmp9 < 6")
tmp11 = (-1) + tmp4
tmp12 = tl.full([1], 0, tl.int64)
tmp13 = tmp11 >= tmp12
tmp14 = tl.full([1], 4, tl.int64)
tmp15 = tmp11 < tmp14
tmp16 = (-1) + tmp9
tmp17 = tmp16 >= tmp12
tmp18 = tmp16 < tmp14
tmp19 = tmp13 & tmp15
tmp20 = tmp19 & tmp17
tmp21 = tmp20 & tmp18
tmp22 = tl.load(in_ptr1 + ((-20) + x0 + (4*tmp9) + (16*tmp4) + (64*x3)), tmp21 & xmask, other=0.0)
tl.store(out_ptr0 + (x4), tmp22, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/s2/cs2stujdhu7ikjsefsqrf3pzshjig4nfsxvbaum4e7txzyio474y.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
# Source node to ATen node mapping:
# matmul => bmm
# Graph fragment:
# %bmm : [num_users=1] = call_function[target=torch.ops.aten.bmm.default](args = (%view_7, %view_8), kwargs = {})
triton_poi_fused_bmm_4 = async_compile.triton('triton_poi_fused_bmm_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_bmm_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_bmm_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 5184
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 81
x1 = (xindex // 81)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (81*(x1 % 16)) + (1312*(x1 // 16))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/tv/ctvlqogy5r6ohjndwmx3qbdwvnnrjj2qm7iknoh6f6ckrtehwqir.py
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.col2im]
# Source node to ATen node mapping:
# out_1 => full_default
# Graph fragment:
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 6, 6], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
triton_poi_fused_col2im_5 = async_compile.triton('triton_poi_fused_col2im_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_col2im_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_col2im_5(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = 0.0
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/wn/cwny5emcwuqtopqfvgdcafgoxd3b3o6ewj73od3v5htm4uthw2ne.py
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.col2im]
# Source node to ATen node mapping:
# out_1 => index_put
# Graph fragment:
# %index_put : [num_users=1] = call_function[target=torch.ops.aten.index_put.default](args = (%full_default, [None, None, %unsqueeze_5, %add], %permute_9, True), kwargs = {})
triton_poi_fused_col2im_6 = async_compile.triton('triton_poi_fused_col2im_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_col2im_6', 'mutated_arg_names': ['out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_col2im_6(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 2304
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x7 = (xindex // 48) % 12
x9 = (xindex // 4) % 12
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 3
x3 = (xindex // 48) % 4
x4 = (xindex // 192) % 3
x5 = (xindex // 576)
tmp0 = tl.load(in_ptr0 + (x7), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (x9), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr1 + (x0 + (4*x2) + (12*x4) + (36*x1) + (144*x3) + (576*x5) + ((x2 + (3*x4)) // 9)), xmask)
tmp1 = tl.full([XBLOCK], 6, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert(((0 <= tmp4) & (tmp4 < 6)) | ~(xmask), "index out of bounds: 0 <= tmp4 < 6")
tmp7 = tmp6 + tmp1
tmp8 = tmp6 < 0
tmp9 = tl.where(tmp8, tmp7, tmp6)
tl.device_assert(((0 <= tmp9) & (tmp9 < 6)) | ~(xmask), "index out of bounds: 0 <= tmp9 < 6")
tl.atomic_add(out_ptr0 + (tmp9 + (6*tmp4) + (36*x0) + (144*x5)), tmp11, xmask, sem='relaxed')
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6j/c6j6zr7wslvz3frvfzwveytngq4nfxv75gmqi2vju57tya4iykk7.py
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# out_2 => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_10,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_7 = async_compile.triton('triton_poi_fused_clone_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_7(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
y1 = (yindex // 4) % 4
y0 = yindex % 4
x3 = xindex
y2 = (yindex // 16)
y5 = yindex
tmp0 = 1 + y1
tmp1 = tl.full([1, 1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1, 1], 6, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = 1 + y0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (7 + y0 + (6*y1) + (36*x3) + (144*y2)), tmp10 & xmask & ymask, eviction_policy='evict_last', other=0.0)
tl.store(out_ptr0 + (x3 + (4*y5)), tmp11, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/wv/cwvxucyxlsbx6r4eu4pwwxtgq2adykv2e5ulhy576dumppymjdrc.py
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.add]
# Source node to ATen node mapping:
# out_2 => add_4
# Graph fragment:
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_13, %primals_6), kwargs = {})
triton_poi_fused_add_8 = async_compile.triton('triton_poi_fused_add_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_8', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_8(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (81, 4), (4, 1))
assert_size_stride(primals_4, (81, ), (1, ))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((3, 4), (4, 1), torch.int64)
# Topologically Sorted Source Nodes: [unfold], Original ATen: [aten.im2col]
stream0 = get_raw_stream(0)
triton_poi_fused_im2col_0.run(buf1, 12, grid=grid(12), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 1, 16, 4), torch.float32)
# Topologically Sorted Source Nodes: [avg_pool2d], Original ATen: [aten.avg_pool2d]
triton_poi_fused_avg_pool2d_1.run(primals_1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = empty_strided_cuda((64, 81), (81, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 81), (1, 4), 0), out=buf3)
del primals_3
buf6 = empty_strided_cuda((4, 1, 16, 9, 9), (1312, 1312, 81, 9, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_3], Original ATen: [aten._softmax]
triton_per_fused__softmax_2.run(buf3, primals_4, buf6, 576, 9, grid=grid(576), stream=stream0)
del primals_4
buf7 = empty_strided_cuda((4, 1, 16, 9, 4), (576, 1, 36, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf1, buf0, buf7, 2304, grid=grid(2304), stream=stream0)
buf8 = reinterpret_tensor(buf3, (64, 9, 9), (81, 9, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
triton_poi_fused_bmm_4.run(buf6, buf8, 5184, grid=grid(5184), stream=stream0)
buf9 = empty_strided_cuda((64, 9, 4), (36, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(buf8, reinterpret_tensor(buf7, (64, 9, 4), (36, 4, 1), 0), out=buf9)
del buf8
buf10 = empty_strided_cuda((4, 4, 6, 6), (144, 36, 6, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.col2im]
triton_poi_fused_col2im_5.run(buf10, 576, grid=grid(576), stream=stream0)
buf11 = empty_strided_cuda((4, 4, 6, 6), (144, 36, 6, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.col2im]
triton_poi_fused_col2im_5.run(buf11, 576, grid=grid(576), stream=stream0)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.col2im]
triton_poi_fused_col2im_6.run(buf1, buf9, buf11, 2304, grid=grid(2304), stream=stream0)
del buf9
buf13 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.clone]
triton_poi_fused_clone_7.run(buf11, buf13, 64, 4, grid=grid(64, 4), stream=stream0)
del buf11
buf14 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf13, (64, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf14)
buf15 = reinterpret_tensor(buf14, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf14 # reuse
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.add]
triton_poi_fused_add_8.run(buf15, primals_6, 256, grid=grid(256), stream=stream0)
del primals_6
return (buf15, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), buf1, reinterpret_tensor(buf2, (64, 4), (4, 1), 0), buf6, buf10, reinterpret_tensor(buf13, (64, 4), (4, 1), 0), primals_5, reinterpret_tensor(buf7, (64, 4, 9), (36, 1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((81, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((81, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
from torch import nn
from torch.nn import functional as F
class OutlookAttention(nn.Module):
def __init__(self, dim, num_heads=1, kernel_size=3, padding=1, stride=1,
qkv_bias=False, attn_drop=0.1):
super().__init__()
self.dim = dim
self.num_heads = num_heads
self.head_dim = dim // num_heads
self.kernel_size = kernel_size
self.padding = padding
self.stride = stride
self.scale = self.head_dim ** -0.5
self.v_pj = nn.Linear(dim, dim, bias=qkv_bias)
self.attn = nn.Linear(dim, kernel_size ** 4 * num_heads)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = nn.Linear(dim, dim)
self.proj_drop = nn.Dropout(attn_drop)
self.unflod = nn.Unfold(kernel_size, padding, stride)
self.pool = nn.AvgPool2d(kernel_size=stride, stride=stride,
ceil_mode=True)
def forward(self, x):
B, H, W, C = x.shape
v = self.v_pj(x).permute(0, 3, 1, 2)
h, w = math.ceil(H / self.stride), math.ceil(W / self.stride)
v = self.unflod(v).reshape(B, self.num_heads, self.head_dim, self.
kernel_size * self.kernel_size, h * w).permute(0, 1, 4, 3, 2)
attn = self.pool(x.permute(0, 3, 1, 2)).permute(0, 2, 3, 1)
attn = self.attn(attn).reshape(B, h * w, self.num_heads, self.
kernel_size * self.kernel_size, self.kernel_size * self.kernel_size
).permute(0, 2, 1, 3, 4)
attn = self.scale * attn
attn = attn.softmax(-1)
attn = self.attn_drop(attn)
out = (attn @ v).permute(0, 1, 4, 3, 2).reshape(B, C * self.
kernel_size * self.kernel_size, h * w)
out = F.fold(out, output_size=(H, W), kernel_size=self.kernel_size,
padding=self.padding, stride=self.stride)
out = self.proj(out.permute(0, 2, 3, 1))
out = self.proj_drop(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_im2col_0(out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 12
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = x0 + x1
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_avg_pool2d_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_per_fused__softmax_2(in_ptr0, in_ptr1, out_ptr2, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 576
rnumel = 9
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
rmask = rindex < rnumel
r2 = rindex
x5 = xindex
x0 = xindex % 9
x4 = xindex // 144
x6 = xindex % 144
tmp0 = tl.load(in_ptr0 + (r2 + 9 * x5), rmask & xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + (r2 + 9 * x0), rmask & xmask, eviction_policy=
'evict_last', other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.where(rmask & xmask, tmp5, float('-inf'))
tmp8 = triton_helpers.max2(tmp7, 1)[:, None]
tmp9 = tmp4 - tmp8
tmp10 = 0.5
tmp11 = tmp9 * tmp10
tmp12 = tl_math.exp(tmp11)
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(rmask & xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = tmp12 / tmp16
tl.store(out_ptr2 + (r2 + 9 * x6 + 1312 * x4), tmp17, rmask & xmask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 2304
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4 % 9
x2 = xindex // 36 % 16
x0 = xindex % 4
x3 = xindex // 576
x4 = xindex
tmp0 = tl.load(in_ptr0 + (4 * (x1 // 3) + x2 // 4), xmask,
eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (4 * (x1 % 3) + x2 % 4), xmask,
eviction_policy='evict_last')
tmp1 = tl.full([XBLOCK], 6, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert((0 <= tmp4) & (tmp4 < 6) | ~xmask,
'index out of bounds: 0 <= tmp4 < 6')
tmp7 = tmp6 + tmp1
tmp8 = tmp6 < 0
tmp9 = tl.where(tmp8, tmp7, tmp6)
tl.device_assert((0 <= tmp9) & (tmp9 < 6) | ~xmask,
'index out of bounds: 0 <= tmp9 < 6')
tmp11 = -1 + tmp4
tmp12 = tl.full([1], 0, tl.int64)
tmp13 = tmp11 >= tmp12
tmp14 = tl.full([1], 4, tl.int64)
tmp15 = tmp11 < tmp14
tmp16 = -1 + tmp9
tmp17 = tmp16 >= tmp12
tmp18 = tmp16 < tmp14
tmp19 = tmp13 & tmp15
tmp20 = tmp19 & tmp17
tmp21 = tmp20 & tmp18
tmp22 = tl.load(in_ptr1 + (-20 + x0 + 4 * tmp9 + 16 * tmp4 + 64 * x3),
tmp21 & xmask, other=0.0)
tl.store(out_ptr0 + x4, tmp22, xmask)
@triton.jit
def triton_poi_fused_bmm_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 5184
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 81
x1 = xindex // 81
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 81 * (x1 % 16) + 1312 * (x1 // 16)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_col2im_5(out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = 0.0
tl.store(out_ptr0 + x0, tmp0, xmask)
@triton.jit
def triton_poi_fused_col2im_6(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 2304
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x7 = xindex // 48 % 12
x9 = xindex // 4 % 12
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 3
x3 = xindex // 48 % 4
x4 = xindex // 192 % 3
x5 = xindex // 576
tmp0 = tl.load(in_ptr0 + x7, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + x9, xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr1 + (x0 + 4 * x2 + 12 * x4 + 36 * x1 + 144 * x3 +
576 * x5 + (x2 + 3 * x4) // 9), xmask)
tmp1 = tl.full([XBLOCK], 6, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert((0 <= tmp4) & (tmp4 < 6) | ~xmask,
'index out of bounds: 0 <= tmp4 < 6')
tmp7 = tmp6 + tmp1
tmp8 = tmp6 < 0
tmp9 = tl.where(tmp8, tmp7, tmp6)
tl.device_assert((0 <= tmp9) & (tmp9 < 6) | ~xmask,
'index out of bounds: 0 <= tmp9 < 6')
tl.atomic_add(out_ptr0 + (tmp9 + 6 * tmp4 + 36 * x0 + 144 * x5), tmp11,
xmask, sem='relaxed')
@triton.jit
def triton_poi_fused_clone_7(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
y1 = yindex // 4 % 4
y0 = yindex % 4
x3 = xindex
y2 = yindex // 16
y5 = yindex
tmp0 = 1 + y1
tmp1 = tl.full([1, 1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1, 1], 6, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = 1 + y0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (7 + y0 + 6 * y1 + 36 * x3 + 144 * y2), tmp10 &
xmask & ymask, eviction_policy='evict_last', other=0.0)
tl.store(out_ptr0 + (x3 + 4 * y5), tmp11, xmask & ymask)
@triton.jit
def triton_poi_fused_add_8(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (81, 4), (4, 1))
assert_size_stride(primals_4, (81,), (1,))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((3, 4), (4, 1), torch.int64)
get_raw_stream(0)
triton_poi_fused_im2col_0[grid(12)](buf1, 12, XBLOCK=16, num_warps=
1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 1, 16, 4), torch.float32)
triton_poi_fused_avg_pool2d_1[grid(256)](primals_1, buf2, 256,
XBLOCK=256, num_warps=4, num_stages=1)
buf3 = empty_strided_cuda((64, 81), (81, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 81), (1, 4), 0), out=buf3)
del primals_3
buf6 = empty_strided_cuda((4, 1, 16, 9, 9), (1312, 1312, 81, 9, 1),
torch.float32)
triton_per_fused__softmax_2[grid(576)](buf3, primals_4, buf6, 576,
9, XBLOCK=8, num_warps=2, num_stages=1)
del primals_4
buf7 = empty_strided_cuda((4, 1, 16, 9, 4), (576, 1, 36, 4, 1),
torch.float32)
triton_poi_fused_clone_3[grid(2304)](buf1, buf0, buf7, 2304, XBLOCK
=256, num_warps=4, num_stages=1)
buf8 = reinterpret_tensor(buf3, (64, 9, 9), (81, 9, 1), 0)
del buf3
triton_poi_fused_bmm_4[grid(5184)](buf6, buf8, 5184, XBLOCK=128,
num_warps=4, num_stages=1)
buf9 = empty_strided_cuda((64, 9, 4), (36, 4, 1), torch.float32)
extern_kernels.bmm(buf8, reinterpret_tensor(buf7, (64, 9, 4), (36,
4, 1), 0), out=buf9)
del buf8
buf10 = empty_strided_cuda((4, 4, 6, 6), (144, 36, 6, 1), torch.float32
)
triton_poi_fused_col2im_5[grid(576)](buf10, 576, XBLOCK=256,
num_warps=4, num_stages=1)
buf11 = empty_strided_cuda((4, 4, 6, 6), (144, 36, 6, 1), torch.float32
)
triton_poi_fused_col2im_5[grid(576)](buf11, 576, XBLOCK=256,
num_warps=4, num_stages=1)
triton_poi_fused_col2im_6[grid(2304)](buf1, buf9, buf11, 2304,
XBLOCK=128, num_warps=4, num_stages=1)
del buf9
buf13 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
triton_poi_fused_clone_7[grid(64, 4)](buf11, buf13, 64, 4, XBLOCK=4,
YBLOCK=32, num_warps=4, num_stages=1)
del buf11
buf14 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf13, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf14)
buf15 = reinterpret_tensor(buf14, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf14
triton_poi_fused_add_8[grid(256)](buf15, primals_6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_6
return buf15, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0
), buf1, reinterpret_tensor(buf2, (64, 4), (4, 1), 0
), buf6, buf10, reinterpret_tensor(buf13, (64, 4), (4, 1), 0
), primals_5, reinterpret_tensor(buf7, (64, 4, 9), (36, 1, 4), 0)
class OutlookAttentionNew(nn.Module):
def __init__(self, dim, num_heads=1, kernel_size=3, padding=1, stride=1,
qkv_bias=False, attn_drop=0.1):
super().__init__()
self.dim = dim
self.num_heads = num_heads
self.head_dim = dim // num_heads
self.kernel_size = kernel_size
self.padding = padding
self.stride = stride
self.scale = self.head_dim ** -0.5
self.v_pj = nn.Linear(dim, dim, bias=qkv_bias)
self.attn = nn.Linear(dim, kernel_size ** 4 * num_heads)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = nn.Linear(dim, dim)
self.proj_drop = nn.Dropout(attn_drop)
self.unflod = nn.Unfold(kernel_size, padding, stride)
self.pool = nn.AvgPool2d(kernel_size=stride, stride=stride,
ceil_mode=True)
def forward(self, input_0):
primals_2 = self.v_pj.weight
primals_3 = self.attn.weight
primals_4 = self.attn.bias
primals_5 = self.proj.weight
primals_6 = self.proj.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | OutlookAttention | false | 4,230 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import math
import torch
from torch import nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self, dim, num_heads=1, kernel_size=3, padding=1, stride=1,
qkv_bias=False, attn_drop=0.1):
super().__init__()
self.dim = dim
self.num_heads = num_heads
self.head_dim = dim // num_heads
self.kernel_size = kernel_size
self.padding = padding
self.stride = stride
self.scale = self.head_dim ** -0.5
self.v_pj = nn.Linear(dim, dim, bias=qkv_bias)
self.attn = nn.Linear(dim, kernel_size ** 4 * num_heads)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = nn.Linear(dim, dim)
self.proj_drop = nn.Dropout(attn_drop)
self.unflod = nn.Unfold(kernel_size, padding, stride)
self.pool = nn.AvgPool2d(kernel_size=stride, stride=stride,
ceil_mode=True)
def forward(self, x):
B, H, W, C = x.shape
v = self.v_pj(x).permute(0, 3, 1, 2)
h, w = math.ceil(H / self.stride), math.ceil(W / self.stride)
v = self.unflod(v).reshape(B, self.num_heads, self.head_dim, self.
kernel_size * self.kernel_size, h * w).permute(0, 1, 4, 3, 2)
attn = self.pool(x.permute(0, 3, 1, 2)).permute(0, 2, 3, 1)
attn = self.attn(attn).reshape(B, h * w, self.num_heads, self.
kernel_size * self.kernel_size, self.kernel_size * self.kernel_size
).permute(0, 2, 1, 3, 4)
attn = self.scale * attn
attn = attn.softmax(-1)
attn = self.attn_drop(attn)
out = (attn @ v).permute(0, 1, 4, 3, 2).reshape(B, C * self.
kernel_size * self.kernel_size, h * w)
out = F.fold(out, output_size=(H, W), kernel_size=self.kernel_size,
padding=self.padding, stride=self.stride)
out = self.proj(out.permute(0, 2, 3, 1))
out = self.proj_drop(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
Critic | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/a2/ca2wr2cvkya5clovpxidv7ia56pdcyp7uq4omtpg5m2nr7ya3ryn.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_1,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_tanh_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 64), (64, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (1, 64), (64, 1))
assert_size_stride(primals_7, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_2, 4096, grid=grid(4096), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0), reinterpret_tensor(primals_4, (64, 64), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
triton_poi_fused_tanh_0.run(buf3, primals_5, 4096, grid=grid(4096), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(primals_6, (64, 1), (1, 64), 0), alpha=1, beta=1, out=buf5)
del primals_7
return (reinterpret_tensor(buf5, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, buf3, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((1, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class Critic(nn.Module):
def __init__(self, obs_dim: 'int'):
super().__init__()
self.fc1 = nn.Linear(obs_dim, 64)
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, 1)
def forward(self, x):
x = torch.tanh(self.fc1(x))
x = torch.tanh(self.fc2(x))
x = self.fc3(x)
return x.squeeze()
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'obs_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 64), (64, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (1, 64), (64, 1))
assert_size_stride(primals_7, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(4096)](buf1, primals_2, 4096, XBLOCK=
256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0),
reinterpret_tensor(primals_4, (64, 64), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf2
triton_poi_fused_tanh_0[grid(4096)](buf3, primals_5, 4096, XBLOCK=
256, num_warps=4, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64),
(64, 1), 0), reinterpret_tensor(primals_6, (64, 1), (1, 64), 0),
alpha=1, beta=1, out=buf5)
del primals_7
return reinterpret_tensor(buf5, (4, 4, 4), (16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, buf3, primals_6, primals_4
class CriticNew(nn.Module):
def __init__(self, obs_dim: 'int'):
super().__init__()
self.fc1 = nn.Linear(obs_dim, 64)
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, 1)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| raznem/rlex | Critic | false | 4,231 | [
"MIT"
] | 0 | d24b964d80067becc81d86f6ce87e5be413b7049 | https://github.com/raznem/rlex/tree/d24b964d80067becc81d86f6ce87e5be413b7049 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, obs_dim: 'int'):
super().__init__()
self.fc1 = nn.Linear(obs_dim, 64)
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, 1)
def forward(self, x):
x = torch.tanh(self.fc1(x))
x = torch.tanh(self.fc2(x))
x = self.fc3(x)
return x.squeeze()
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
TdnnAffine | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/cu/ccutvo2v4333pq6xhrg2zryqqwthm7dmmuqprvva2xdwiodpz5jn.py
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# outputs => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_2, %primals_3, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 4) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_3, 64, grid=grid(64), stream=stream0)
del primals_3
return (buf1, primals_1, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super(TdnnAffine, self).__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not None else 0
total_ops = y.nelement() * (m.input_dim * kernel_ops + bias_ops)
m.total_ops += torch.DoubleTensor([int(total_ops)])
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'output_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 4 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(64)](buf1, primals_3, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_3
return buf1, primals_1, primals_2
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffineNew(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super(TdnnAffineNew, self).__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not None else 0
total_ops = y.nelement() * (m.input_dim * kernel_ops + bias_ops)
m.total_ops += torch.DoubleTensor([int(total_ops)])
def forward(self, input_0):
primals_2 = self.weight
primals_3 = self.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| qlindazm/asv-subtools | TdnnAffine | false | 4,232 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn.functional as F
import torch.nn
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class Model(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super().__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not
# ... truncated (>4000 chars) for memory efficiency |
ChannelAttentionModule | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/p6/cp6vuooninjiuju55qtiu7u3yjx4izmj5jtvx2lkeddu4rheo45u.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048, 64], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 2048
xnumel = 49
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = (yindex // 512)
tmp0 = tl.load(in_ptr0 + (x2 + (49*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (512*x2) + (25088*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/oc/cochsno6wpkwamgsqz5legelnxxchuje5twfzhozvusus3e5bzmo.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 262144
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = (yindex // 512)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (512*x2) + (4608*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/qx/cqxjinzzdi527r7k6w42436njez2bx4cbmwgxdreh5ebe3jdfhoa.py
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# y => convolution
# Graph fragment:
# %convolution : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_2, %primals_3, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_2 = async_compile.triton('triton_poi_fused_convolution_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[131072],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 100352
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/d5/cd5fpdjs2anlog45byuw5kz6tjzeazlgaskaocoyyn6w5kpybr4e.py
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
# Source node to ATen node mapping:
# att_1 => div_1, exp, sum_1
# wrapped_sqrt => full_default
# Graph fragment:
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([], 7.0), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %scalar_tensor_default : [num_users=2] = call_function[target=torch.ops.aten.scalar_tensor.default](args = (1,), kwargs = {dtype: torch.float32, device: cuda:0, pin_memory: False})
# %ge_scalar : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%full_default, 0), kwargs = {})
# %neg_default : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%scalar_tensor_default,), kwargs = {})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ge_scalar, %scalar_tensor_default, %neg_default), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_6, %where_self), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where_self, %full_default), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, %mul_tensor_1), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_per_fused__softmax_sqrt_3 = async_compile.triton('triton_per_fused__softmax_sqrt_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[2048, 512],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_sqrt_3', 'mutated_arg_names': [], 'no_x_dim': True, 'num_load': 1, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_sqrt_3(in_ptr0, out_ptr2, xnumel, rnumel):
xnumel = 2048
XBLOCK: tl.constexpr = 1
rnumel = 512
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (512*x0)), None)
tmp1 = tl.full([1], 7.0, tl.float64)
tmp2 = tl.full([1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [RBLOCK])
tmp10 = triton_helpers.promote_to_tensor(triton_helpers.max2(tmp8, 0))
tmp11 = tmp7 - tmp10
tmp12 = tmp6.to(tl.float64)
tmp13 = tmp12 * tmp1
tmp14 = tmp13.to(tl.float32)
tmp15 = tmp11 / tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tl.broadcast_to(tmp16, [RBLOCK])
tmp19 = triton_helpers.promote_to_tensor(tl.sum(tmp17, 0))
tmp20 = tmp16 / tmp19
tl.store(out_ptr2 + (r1 + (512*x0)), tmp20, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 512, 1, 49), (25088, 49, 49, 1))
assert_size_stride(primals_2, (512, 512, 3, 3), (4608, 9, 3, 1))
assert_size_stride(primals_3, (512, ), (1, ))
assert_size_stride(primals_4, (49, 49), (49, 1))
assert_size_stride(primals_5, (49, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 512, 1, 49), (25088, 1, 25088, 512), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(primals_1, buf0, 2048, 49, grid=grid(2048, 49), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((512, 512, 3, 3), (4608, 1, 1536, 512), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(primals_2, buf1, 262144, 9, grid=grid(262144, 9), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf0, buf1, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 512, 1, 49), (25088, 1, 25088, 512))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf3, primals_3, 100352, grid=grid(100352), stream=stream0)
del primals_3
buf4 = empty_strided_cuda((4, 512, 512), (262144, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 512, 49), (25088, 1, 512), 0), reinterpret_tensor(buf3, (4, 49, 512), (25088, 512, 1), 0), out=buf4)
buf7 = empty_strided_cuda((4, 1, 512, 512), (262144, 1, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
triton_per_fused__softmax_sqrt_3.run(buf4, buf7, 2048, 512, grid=grid(2048), stream=stream0)
del buf4
buf8 = empty_strided_cuda((4, 512, 49), (25088, 49, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf7, (4, 512, 512), (262144, 512, 1), 0), reinterpret_tensor(buf3, (4, 512, 49), (25088, 1, 512), 0), out=buf8)
buf9 = empty_strided_cuda((2048, 49), (49, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf8, (2048, 49), (49, 1), 0), reinterpret_tensor(primals_4, (49, 49), (1, 49), 0), alpha=1, beta=1, out=buf9)
del primals_5
return (reinterpret_tensor(buf9, (4, 512, 49), (25088, 49, 1), 0), buf0, buf1, buf3, buf7, reinterpret_tensor(buf8, (2048, 49), (49, 1), 0), primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 512, 1, 49), (25088, 49, 49, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((512, 512, 3, 3), (4608, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((49, 49), (49, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((49, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
from torch import nn
from torch.nn import init
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(SimplifiedScaledDotProductAttention, self).__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class ChannelAttentionModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = SimplifiedScaledDotProductAttention(H * W, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1)
y = self.pa(y, y, y)
return y
def get_inputs():
return [torch.rand([4, 512, 1, 49])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import numpy as np
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 49
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = yindex // 512
tmp0 = tl.load(in_ptr0 + (x2 + 49 * y3), xmask, eviction_policy=
'evict_last')
tl.store(out_ptr0 + (y0 + 512 * x2 + 25088 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)
) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = yindex // 512
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 512 * x2 + 4608 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_convolution_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, None)
@triton.jit
def triton_per_fused__softmax_sqrt_3(in_ptr0, out_ptr2, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 512 * x0), None)
tmp1 = tl.full([1], 7.0, tl.float64)
tmp2 = tl.full([1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [RBLOCK])
tmp10 = triton_helpers.promote_to_tensor(triton_helpers.max2(tmp8, 0))
tmp11 = tmp7 - tmp10
tmp12 = tmp6.to(tl.float64)
tmp13 = tmp12 * tmp1
tmp14 = tmp13.to(tl.float32)
tmp15 = tmp11 / tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tl.broadcast_to(tmp16, [RBLOCK])
tmp19 = triton_helpers.promote_to_tensor(tl.sum(tmp17, 0))
tmp20 = tmp16 / tmp19
tl.store(out_ptr2 + (r1 + 512 * x0), tmp20, None)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 512, 1, 49), (25088, 49, 49, 1))
assert_size_stride(primals_2, (512, 512, 3, 3), (4608, 9, 3, 1))
assert_size_stride(primals_3, (512,), (1,))
assert_size_stride(primals_4, (49, 49), (49, 1))
assert_size_stride(primals_5, (49,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 512, 1, 49), (25088, 1, 25088, 512),
torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(2048, 49)](primals_1, buf0, 2048, 49,
XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((512, 512, 3, 3), (4608, 1, 1536, 512),
torch.float32)
triton_poi_fused_1[grid(262144, 9)](primals_2, buf1, 262144, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf0, buf1, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 512, 1, 49), (25088, 1, 25088, 512))
buf3 = buf2
del buf2
triton_poi_fused_convolution_2[grid(100352)](buf3, primals_3,
100352, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_3
buf4 = empty_strided_cuda((4, 512, 512), (262144, 512, 1), torch.
float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 512, 49), (25088, 1,
512), 0), reinterpret_tensor(buf3, (4, 49, 512), (25088, 512, 1
), 0), out=buf4)
buf7 = empty_strided_cuda((4, 1, 512, 512), (262144, 1, 512, 1),
torch.float32)
triton_per_fused__softmax_sqrt_3[grid(2048)](buf4, buf7, 2048, 512,
num_warps=4, num_stages=1)
del buf4
buf8 = empty_strided_cuda((4, 512, 49), (25088, 49, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf7, (4, 512, 512), (262144,
512, 1), 0), reinterpret_tensor(buf3, (4, 512, 49), (25088, 1,
512), 0), out=buf8)
buf9 = empty_strided_cuda((2048, 49), (49, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf8, (2048, 49),
(49, 1), 0), reinterpret_tensor(primals_4, (49, 49), (1, 49), 0
), alpha=1, beta=1, out=buf9)
del primals_5
return reinterpret_tensor(buf9, (4, 512, 49), (25088, 49, 1), 0
), buf0, buf1, buf3, buf7, reinterpret_tensor(buf8, (2048, 49), (49,
1), 0), primals_4
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(SimplifiedScaledDotProductAttention, self).__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class ChannelAttentionModuleNew(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = SimplifiedScaledDotProductAttention(H * W, h=1)
def forward(self, input_0):
primals_2 = self.cnn.weight
primals_3 = self.cnn.bias
primals_4 = self.pa.fc_o.weight
primals_5 = self.pa.fc_o.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | ChannelAttentionModule | false | 4,233 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
import numpy as np
from torch import nn
from torch.nn import init
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super().__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class Model(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = SimplifiedScaledDotProductAttention(H * W, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1)
y = self.pa(y, y, y)
return y
def get_inputs():
return [torch.rand([4, 512, 1, 49])]
def get_init_inputs():
return []
|
LDEPooling | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/px/cpxgh7rygczo7s5i6nrfdx75l2hi6cjdfexsy2ctyfflclxdb7ye.py
# Topologically Sorted Source Nodes: [r, pow_1, add, neg, pow_2, sum_1, mul, w], Original ATen: [aten.sub, aten.pow, aten.add, aten.neg, aten.sum, aten.mul, aten._softmax]
# Source node to ATen node mapping:
# add => add
# mul => mul
# neg => neg
# pow_1 => pow_1
# pow_2 => pow_2
# r => sub
# sum_1 => sum_1
# w => amax, exp, sub_1, sum_2
# Graph fragment:
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%unsqueeze, %primals_2), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%primals_3, 2), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_1, 1e-10), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%add,), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_2, [2], True), kwargs = {})
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%neg, %sum_1), kwargs = {})
# %amax : [num_users=2] = call_function[target=torch.ops.aten.amax.default](args = (%mul, [3], True), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %sum_2 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [3], True), kwargs = {})
triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0 = async_compile.triton('triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[16, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 64
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex
x0 = xindex % 4
x1 = (xindex // 4)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (r2), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (x0 + (16*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr2 + (r2), None, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (4 + x0 + (16*x1)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr2 + (64 + r2), None, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr1 + (8 + x0 + (16*x1)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + (128 + r2), None, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (12 + x0 + (16*x1)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr2 + (192 + r2), None, eviction_policy='evict_last')
tmp1 = tmp0 * tmp0
tmp2 = 1e-10
tmp3 = tmp1 + tmp2
tmp4 = -tmp3
tmp7 = tmp5 - tmp6
tmp8 = tmp7 * tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tmp21 = tmp19 - tmp20
tmp22 = tmp21 * tmp21
tmp23 = tmp18 + tmp22
tmp24 = tmp4 * tmp23
tmp25 = tl.broadcast_to(tmp24, [XBLOCK, RBLOCK])
tmp27 = tl.where(xmask, tmp25, float("-inf"))
tmp28 = triton_helpers.max2(tmp27, 1)[:, None]
tmp29 = tmp24 - tmp28
tmp30 = tl_math.exp(tmp29)
tmp31 = tl.broadcast_to(tmp30, [XBLOCK, RBLOCK])
tmp33 = tl.where(xmask, tmp31, 0)
tmp34 = tl.sum(tmp33, 1)[:, None]
tl.store(out_ptr0 + (r2 + (64*x3)), tmp24, xmask)
tl.store(out_ptr1 + (x3), tmp28, xmask)
tl.store(out_ptr2 + (x3), tmp34, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5c/c5cyg6tmgybx2znb56tli3mxu474a5mhwo2sfz4cyzeil7dyv77v.py
# Topologically Sorted Source Nodes: [r, w, mul_1, e], Original ATen: [aten.sub, aten._softmax, aten.mul, aten.mean]
# Source node to ATen node mapping:
# e => mean
# mul_1 => mul_1
# r => sub
# w => div, exp, sub_1
# Graph fragment:
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%unsqueeze, %primals_2), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %sub), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%mul_1, [1]), kwargs = {})
triton_poi_fused__softmax_mean_mul_sub_1 = async_compile.triton('triton_poi_fused__softmax_mean_mul_sub_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_mean_mul_sub_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 17, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mean_mul_sub_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x2 = (xindex // 256)
x4 = (xindex // 64)
x3 = xindex % 256
x5 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x2), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + (4*x2), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (4*x4), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x3), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (64 + x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr1 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr2 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr3 + (1 + (4*x4)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr0 + (128 + x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr1 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr2 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr3 + (2 + (4*x4)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr0 + (192 + x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp31 = tl.load(in_ptr1 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp34 = tl.load(in_ptr2 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp36 = tl.load(in_ptr3 + (3 + (4*x4)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tmp8 = tmp6 - tmp7
tmp9 = tmp5 * tmp8
tmp12 = tmp10 - tmp11
tmp13 = tl_math.exp(tmp12)
tmp15 = tmp13 / tmp14
tmp17 = tmp16 - tmp7
tmp18 = tmp15 * tmp17
tmp19 = tmp9 + tmp18
tmp22 = tmp20 - tmp21
tmp23 = tl_math.exp(tmp22)
tmp25 = tmp23 / tmp24
tmp27 = tmp26 - tmp7
tmp28 = tmp25 * tmp27
tmp29 = tmp19 + tmp28
tmp32 = tmp30 - tmp31
tmp33 = tl_math.exp(tmp32)
tmp35 = tmp33 / tmp34
tmp37 = tmp36 - tmp7
tmp38 = tmp35 * tmp37
tmp39 = tmp29 + tmp38
tmp40 = 4.0
tmp41 = tmp39 / tmp40
tl.store(out_ptr0 + (x5), tmp41, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 64), (64, 1))
assert_size_stride(primals_3, (64, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 64), (256, 64, 1024, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [r, pow_1, add, neg, pow_2, sum_1, mul, w], Original ATen: [aten.sub, aten.pow, aten.add, aten.neg, aten.sum, aten.mul, aten._softmax]
stream0 = get_raw_stream(0)
triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0.run(primals_3, primals_1, primals_2, buf0, buf1, buf2, 16, 64, grid=grid(16), stream=stream0)
buf3 = empty_strided_cuda((4, 4, 64), (256, 64, 1), torch.float32)
# Topologically Sorted Source Nodes: [r, w, mul_1, e], Original ATen: [aten.sub, aten._softmax, aten.mul, aten.mean]
triton_poi_fused__softmax_mean_mul_sub_1.run(buf0, buf1, buf2, primals_1, primals_2, buf3, 1024, grid=grid(1024), stream=stream0)
del buf0
return (reinterpret_tensor(buf3, (4, 256, 1), (256, 1, 1), 0), primals_1, primals_2, primals_3, buf1, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn
class LDEPooling(torch.nn.Module):
"""A novel learnable dictionary encoding layer.
Reference: Weicheng Cai, etc., "A NOVEL LEARNABLE DICTIONARY ENCODING LAYER FOR END-TO-END
LANGUAGE IDENTIFICATION", icassp, 2018
"""
def __init__(self, input_dim, c_num=64, eps=1e-10):
super(LDEPooling, self).__init__()
self.input_dim = input_dim
self.output_dim = input_dim * c_num
self.eps = eps
self.mu = torch.nn.Parameter(torch.randn(input_dim, c_num))
self.s = torch.nn.Parameter(torch.ones(c_num))
self.softmax_for_w = torch.nn.Softmax(dim=3)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
r = inputs.transpose(1, 2).unsqueeze(3) - self.mu
w = self.softmax_for_w(-(self.s ** 2 + self.eps) * torch.sum(r ** 2,
dim=2, keepdim=True))
e = torch.mean(w * r, dim=1)
return e.reshape(-1, self.output_dim, 1)
def get_output_dim(self):
return self.output_dim
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0(in_ptr0, in_ptr1,
in_ptr2, out_ptr0, out_ptr1, out_ptr2, xnumel, rnumel, XBLOCK: tl.constexpr
):
xnumel = 16
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex
x0 = xindex % 4
x1 = xindex // 4
x3 = xindex
tmp0 = tl.load(in_ptr0 + r2, None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (x0 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr2 + r2, None, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (4 + x0 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp10 = tl.load(in_ptr2 + (64 + r2), None, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr1 + (8 + x0 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp15 = tl.load(in_ptr2 + (128 + r2), None, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (12 + x0 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp20 = tl.load(in_ptr2 + (192 + r2), None, eviction_policy='evict_last')
tmp1 = tmp0 * tmp0
tmp2 = 1e-10
tmp3 = tmp1 + tmp2
tmp4 = -tmp3
tmp7 = tmp5 - tmp6
tmp8 = tmp7 * tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tmp21 = tmp19 - tmp20
tmp22 = tmp21 * tmp21
tmp23 = tmp18 + tmp22
tmp24 = tmp4 * tmp23
tmp25 = tl.broadcast_to(tmp24, [XBLOCK, RBLOCK])
tmp27 = tl.where(xmask, tmp25, float('-inf'))
tmp28 = triton_helpers.max2(tmp27, 1)[:, None]
tmp29 = tmp24 - tmp28
tmp30 = tl_math.exp(tmp29)
tmp31 = tl.broadcast_to(tmp30, [XBLOCK, RBLOCK])
tmp33 = tl.where(xmask, tmp31, 0)
tmp34 = tl.sum(tmp33, 1)[:, None]
tl.store(out_ptr0 + (r2 + 64 * x3), tmp24, xmask)
tl.store(out_ptr1 + x3, tmp28, xmask)
tl.store(out_ptr2 + x3, tmp34, xmask)
@triton.jit
def triton_poi_fused__softmax_mean_mul_sub_1(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x2 = xindex // 256
x4 = xindex // 64
x3 = xindex % 256
x5 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 256 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x2, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + 4 * x2, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + 4 * x4, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x3, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (64 + x0 + 256 * x2), xmask, eviction_policy=
'evict_last')
tmp11 = tl.load(in_ptr1 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp14 = tl.load(in_ptr2 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp16 = tl.load(in_ptr3 + (1 + 4 * x4), xmask, eviction_policy='evict_last'
)
tmp20 = tl.load(in_ptr0 + (128 + x0 + 256 * x2), xmask, eviction_policy
='evict_last')
tmp21 = tl.load(in_ptr1 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp24 = tl.load(in_ptr2 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp26 = tl.load(in_ptr3 + (2 + 4 * x4), xmask, eviction_policy='evict_last'
)
tmp30 = tl.load(in_ptr0 + (192 + x0 + 256 * x2), xmask, eviction_policy
='evict_last')
tmp31 = tl.load(in_ptr1 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp34 = tl.load(in_ptr2 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp36 = tl.load(in_ptr3 + (3 + 4 * x4), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tmp8 = tmp6 - tmp7
tmp9 = tmp5 * tmp8
tmp12 = tmp10 - tmp11
tmp13 = tl_math.exp(tmp12)
tmp15 = tmp13 / tmp14
tmp17 = tmp16 - tmp7
tmp18 = tmp15 * tmp17
tmp19 = tmp9 + tmp18
tmp22 = tmp20 - tmp21
tmp23 = tl_math.exp(tmp22)
tmp25 = tmp23 / tmp24
tmp27 = tmp26 - tmp7
tmp28 = tmp25 * tmp27
tmp29 = tmp19 + tmp28
tmp32 = tmp30 - tmp31
tmp33 = tl_math.exp(tmp32)
tmp35 = tmp33 / tmp34
tmp37 = tmp36 - tmp7
tmp38 = tmp35 * tmp37
tmp39 = tmp29 + tmp38
tmp40 = 4.0
tmp41 = tmp39 / tmp40
tl.store(out_ptr0 + x5, tmp41, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 64), (64, 1))
assert_size_stride(primals_3, (64,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 64), (256, 64, 1024, 1), torch.
float32)
buf1 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
get_raw_stream(0)
triton_per_fused__softmax_add_mul_neg_pow_sub_sum_0[grid(16)](primals_3
, primals_1, primals_2, buf0, buf1, buf2, 16, 64, XBLOCK=8,
num_warps=4, num_stages=1)
buf3 = empty_strided_cuda((4, 4, 64), (256, 64, 1), torch.float32)
triton_poi_fused__softmax_mean_mul_sub_1[grid(1024)](buf0, buf1,
buf2, primals_1, primals_2, buf3, 1024, XBLOCK=128, num_warps=4,
num_stages=1)
del buf0
return reinterpret_tensor(buf3, (4, 256, 1), (256, 1, 1), 0
), primals_1, primals_2, primals_3, buf1, buf2
class LDEPoolingNew(torch.nn.Module):
"""A novel learnable dictionary encoding layer.
Reference: Weicheng Cai, etc., "A NOVEL LEARNABLE DICTIONARY ENCODING LAYER FOR END-TO-END
LANGUAGE IDENTIFICATION", icassp, 2018
"""
def __init__(self, input_dim, c_num=64, eps=1e-10):
super(LDEPoolingNew, self).__init__()
self.input_dim = input_dim
self.output_dim = input_dim * c_num
self.eps = eps
self.mu = torch.nn.Parameter(torch.randn(input_dim, c_num))
self.s = torch.nn.Parameter(torch.ones(c_num))
self.softmax_for_w = torch.nn.Softmax(dim=3)
def get_output_dim(self):
return self.output_dim
def forward(self, input_0):
primals_2 = self.mu
primals_3 = self.s
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| qlindazm/asv-subtools | LDEPooling | false | 4,234 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn
class Model(torch.nn.Module):
"""A novel learnable dictionary encoding layer.
Reference: Weicheng Cai, etc., "A NOVEL LEARNABLE DICTIONARY ENCODING LAYER FOR END-TO-END
LANGUAGE IDENTIFICATION", icassp, 2018
"""
def __init__(self, input_dim, c_num=64, eps=1e-10):
super().__init__()
self.input_dim = input_dim
self.output_dim = input_dim * c_num
self.eps = eps
self.mu = torch.nn.Parameter(torch.randn(input_dim, c_num))
self.s = torch.nn.Parameter(torch.ones(c_num))
self.softmax_for_w = torch.nn.Softmax(dim=3)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
r = inputs.transpose(1, 2).unsqueeze(3) - self.mu
w = self.softmax_for_w(-(self.s ** 2 + self.eps) * torch.sum(r ** 2,
dim=2, keepdim=True))
e = torch.mean(w * r, dim=1)
return e.reshape(-1, self.output_dim, 1)
def get_output_dim(self):
return self.output_dim
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4]
|
SoftmaxAffineLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/2t/c2ttlvr2ngdhjnhfyvr646edvmboep3mriwuj3ygccebcw6w3uev.py
# Topologically Sorted Source Nodes: [outputs, log_softmax], Original ATen: [aten.convolution, aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => exp, sum_1
# outputs => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_2, %primals_3, [1], [0], [1], False, [0], 1), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, 1), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %div_tensor : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, 1.0), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
triton_poi_fused__log_softmax_convolution_0 = async_compile.triton('triton_poi_fused__log_softmax_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_convolution_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_convolution_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp6 = tl.load(in_ptr0 + (4 + x0 + (16*x1)), xmask)
tmp7 = tl.load(in_ptr1 + (1))
tmp8 = tl.broadcast_to(tmp7, [XBLOCK])
tmp12 = tl.load(in_ptr0 + (8 + x0 + (16*x1)), xmask)
tmp13 = tl.load(in_ptr1 + (2))
tmp14 = tl.broadcast_to(tmp13, [XBLOCK])
tmp18 = tl.load(in_ptr0 + (12 + x0 + (16*x1)), xmask)
tmp19 = tl.load(in_ptr1 + (3))
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = 1.0
tmp5 = tmp3 * tmp4
tmp9 = tmp6 + tmp8
tmp10 = tmp9 * tmp4
tmp11 = triton_helpers.maximum(tmp5, tmp10)
tmp15 = tmp12 + tmp14
tmp16 = tmp15 * tmp4
tmp17 = triton_helpers.maximum(tmp11, tmp16)
tmp21 = tmp18 + tmp20
tmp22 = tmp21 * tmp4
tmp23 = triton_helpers.maximum(tmp17, tmp22)
tmp24 = tmp5 - tmp23
tmp25 = tmp24 * tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp10 - tmp23
tmp28 = tmp27 * tmp4
tmp29 = tl_math.exp(tmp28)
tmp30 = tmp26 + tmp29
tmp31 = tmp16 - tmp23
tmp32 = tmp31 * tmp4
tmp33 = tl_math.exp(tmp32)
tmp34 = tmp30 + tmp33
tmp35 = tmp22 - tmp23
tmp36 = tmp35 * tmp4
tmp37 = tl_math.exp(tmp36)
tmp38 = tmp34 + tmp37
tl.store(out_ptr0 + (x2), tmp23, xmask)
tl.store(out_ptr1 + (x2), tmp38, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nk/cnkyhgk4ekrtmfxabvrxv7cvf54d7vrdimis4qzz5tw33rjuejwh.py
# Topologically Sorted Source Nodes: [outputs, log_softmax], Original ATen: [aten.convolution, aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => log, sub_1
# outputs => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_2, %primals_3, [1], [0], [1], False, [0], 1), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, 1), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %div_tensor : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, 1.0), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%div_tensor, %log), kwargs = {})
triton_poi_fused__log_softmax_convolution_1 = async_compile.triton('triton_poi_fused__log_softmax_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_convolution_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_convolution_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 4) % 4
x0 = xindex % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp3
tmp9 = tl_math.log(tmp8)
tmp10 = tmp7 - tmp9
tl.store(in_out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf2 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [outputs, log_softmax], Original ATen: [aten.convolution, aten._log_softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__log_softmax_convolution_0.run(buf0, primals_3, buf1, buf2, 16, grid=grid(16), stream=stream0)
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [outputs, log_softmax], Original ATen: [aten.convolution, aten._log_softmax]
triton_poi_fused__log_softmax_convolution_1.run(buf3, primals_3, buf1, buf2, 64, grid=grid(64), stream=stream0)
del buf1
del buf2
del primals_3
return (buf3, primals_1, primals_2, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super(TdnnAffine, self).__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not None else 0
total_ops = y.nelement() * (m.input_dim * kernel_ops + bias_ops)
m.total_ops += torch.DoubleTensor([int(total_ops)])
class SoftmaxAffineLayer(torch.nn.Module):
""" An usual 2-fold softmax layer with an affine transform.
@dim: which dim to apply softmax on
"""
def __init__(self, input_dim, output_dim, context=[0], dim=1, log=True,
bias=True, groups=1, t=1.0, special_init=False):
super(SoftmaxAffineLayer, self).__init__()
self.affine = TdnnAffine(input_dim, output_dim, context=context,
bias=bias, groups=groups)
self.t = t
if log:
self.softmax = torch.nn.LogSoftmax(dim=dim)
else:
self.softmax = torch.nn.Softmax(dim=dim)
if special_init:
torch.nn.init.xavier_uniform_(self.affine.weight, gain=torch.nn
.init.calculate_gain('sigmoid'))
def forward(self, inputs):
"""
@inputs: any, such as a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
return self.softmax(self.affine(inputs) / self.t)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'output_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn.functional as F
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused__log_softmax_convolution_0(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp6 = tl.load(in_ptr0 + (4 + x0 + 16 * x1), xmask)
tmp7 = tl.load(in_ptr1 + 1)
tmp8 = tl.broadcast_to(tmp7, [XBLOCK])
tmp12 = tl.load(in_ptr0 + (8 + x0 + 16 * x1), xmask)
tmp13 = tl.load(in_ptr1 + 2)
tmp14 = tl.broadcast_to(tmp13, [XBLOCK])
tmp18 = tl.load(in_ptr0 + (12 + x0 + 16 * x1), xmask)
tmp19 = tl.load(in_ptr1 + 3)
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = 1.0
tmp5 = tmp3 * tmp4
tmp9 = tmp6 + tmp8
tmp10 = tmp9 * tmp4
tmp11 = triton_helpers.maximum(tmp5, tmp10)
tmp15 = tmp12 + tmp14
tmp16 = tmp15 * tmp4
tmp17 = triton_helpers.maximum(tmp11, tmp16)
tmp21 = tmp18 + tmp20
tmp22 = tmp21 * tmp4
tmp23 = triton_helpers.maximum(tmp17, tmp22)
tmp24 = tmp5 - tmp23
tmp25 = tmp24 * tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp10 - tmp23
tmp28 = tmp27 * tmp4
tmp29 = tl_math.exp(tmp28)
tmp30 = tmp26 + tmp29
tmp31 = tmp16 - tmp23
tmp32 = tmp31 * tmp4
tmp33 = tl_math.exp(tmp32)
tmp34 = tmp30 + tmp33
tmp35 = tmp22 - tmp23
tmp36 = tmp35 * tmp4
tmp37 = tl_math.exp(tmp36)
tmp38 = tmp34 + tmp37
tl.store(out_ptr0 + x2, tmp23, xmask)
tl.store(out_ptr1 + x2, tmp38, xmask)
@triton.jit
def triton_poi_fused__log_softmax_convolution_1(in_out_ptr0, in_ptr0,
in_ptr1, in_ptr2, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 4 % 4
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp8 = tl.load(in_ptr2 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp3
tmp9 = tl_math.log(tmp8)
tmp10 = tmp7 - tmp9
tl.store(in_out_ptr0 + x3, tmp10, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf2 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__log_softmax_convolution_0[grid(16)](buf0,
primals_3, buf1, buf2, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf3 = buf0
del buf0
triton_poi_fused__log_softmax_convolution_1[grid(64)](buf3,
primals_3, buf1, buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
del buf1
del buf2
del primals_3
return buf3, primals_1, primals_2, buf3
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super(TdnnAffine, self).__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not None else 0
total_ops = y.nelement() * (m.input_dim * kernel_ops + bias_ops)
m.total_ops += torch.DoubleTensor([int(total_ops)])
class SoftmaxAffineLayerNew(torch.nn.Module):
""" An usual 2-fold softmax layer with an affine transform.
@dim: which dim to apply softmax on
"""
def __init__(self, input_dim, output_dim, context=[0], dim=1, log=True,
bias=True, groups=1, t=1.0, special_init=False):
super(SoftmaxAffineLayerNew, self).__init__()
self.affine = TdnnAffine(input_dim, output_dim, context=context,
bias=bias, groups=groups)
self.t = t
if log:
self.softmax = torch.nn.LogSoftmax(dim=dim)
else:
self.softmax = torch.nn.Softmax(dim=dim)
if special_init:
torch.nn.init.xavier_uniform_(self.affine.weight, gain=torch.nn
.init.calculate_gain('sigmoid'))
def forward(self, input_0):
primals_2 = self.affine.weight
primals_3 = self.affine.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| qlindazm/asv-subtools | SoftmaxAffineLayer | false | 4,235 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn.functional as F
import torch.nn
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super().__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is
# ... truncated (>4000 chars) for memory efficiency |
Net | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/dg/cdgsdqqvkke7xwl5f3y4wse55on55jonq2akapygqqfdobxlnfzx.py
# Topologically Sorted Source Nodes: [z, output], Original ATen: [aten.cat, aten._log_softmax]
# Source node to ATen node mapping:
# output => amax, exp, sub, sum_1
# z => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%convolution, %convolution],), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%cat, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%cat, %amax), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
triton_poi_fused__log_softmax_cat_0 = async_compile.triton('triton_poi_fused__log_softmax_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 12, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_cat_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp6 = tl.load(in_ptr1 + (0))
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp20 = tl.load(in_ptr1 + (1))
tmp21 = tl.broadcast_to(tmp20, [XBLOCK])
tmp32 = tl.load(in_ptr1 + (2))
tmp33 = tl.broadcast_to(tmp32, [XBLOCK])
tmp44 = tl.load(in_ptr1 + (3))
tmp45 = tl.broadcast_to(tmp44, [XBLOCK])
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (16*x1)), tmp4 & xmask, other=0.0)
tmp8 = tmp5 + tmp7
tmp9 = tl.full(tmp8.shape, 0.0, tmp8.dtype)
tmp10 = tl.where(tmp4, tmp8, tmp9)
tmp11 = tmp0 >= tmp3
tmp12 = tl.full([1], 8, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tl.load(in_ptr0 + (x0 + (16*((-4) + x1))), tmp11 & xmask, other=0.0)
tmp15 = tmp14 + tmp7
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp11, tmp15, tmp16)
tmp18 = tl.where(tmp4, tmp10, tmp17)
tmp19 = tl.load(in_ptr0 + (4 + x0 + (16*x1)), tmp4 & xmask, other=0.0)
tmp22 = tmp19 + tmp21
tmp23 = tl.full(tmp22.shape, 0.0, tmp22.dtype)
tmp24 = tl.where(tmp4, tmp22, tmp23)
tmp25 = tl.load(in_ptr0 + (4 + x0 + (16*((-4) + x1))), tmp11 & xmask, other=0.0)
tmp26 = tmp25 + tmp21
tmp27 = tl.full(tmp26.shape, 0.0, tmp26.dtype)
tmp28 = tl.where(tmp11, tmp26, tmp27)
tmp29 = tl.where(tmp4, tmp24, tmp28)
tmp30 = triton_helpers.maximum(tmp18, tmp29)
tmp31 = tl.load(in_ptr0 + (8 + x0 + (16*x1)), tmp4 & xmask, other=0.0)
tmp34 = tmp31 + tmp33
tmp35 = tl.full(tmp34.shape, 0.0, tmp34.dtype)
tmp36 = tl.where(tmp4, tmp34, tmp35)
tmp37 = tl.load(in_ptr0 + (8 + x0 + (16*((-4) + x1))), tmp11 & xmask, other=0.0)
tmp38 = tmp37 + tmp33
tmp39 = tl.full(tmp38.shape, 0.0, tmp38.dtype)
tmp40 = tl.where(tmp11, tmp38, tmp39)
tmp41 = tl.where(tmp4, tmp36, tmp40)
tmp42 = triton_helpers.maximum(tmp30, tmp41)
tmp43 = tl.load(in_ptr0 + (12 + x0 + (16*x1)), tmp4 & xmask, other=0.0)
tmp46 = tmp43 + tmp45
tmp47 = tl.full(tmp46.shape, 0.0, tmp46.dtype)
tmp48 = tl.where(tmp4, tmp46, tmp47)
tmp49 = tl.load(in_ptr0 + (12 + x0 + (16*((-4) + x1))), tmp11 & xmask, other=0.0)
tmp50 = tmp49 + tmp45
tmp51 = tl.full(tmp50.shape, 0.0, tmp50.dtype)
tmp52 = tl.where(tmp11, tmp50, tmp51)
tmp53 = tl.where(tmp4, tmp48, tmp52)
tmp54 = triton_helpers.maximum(tmp42, tmp53)
tmp55 = tmp18 - tmp54
tmp56 = tl_math.exp(tmp55)
tmp57 = tmp29 - tmp54
tmp58 = tl_math.exp(tmp57)
tmp59 = tmp56 + tmp58
tmp60 = tmp41 - tmp54
tmp61 = tl_math.exp(tmp60)
tmp62 = tmp59 + tmp61
tmp63 = tmp53 - tmp54
tmp64 = tl_math.exp(tmp63)
tmp65 = tmp62 + tmp64
tl.store(out_ptr0 + (x2), tmp54, xmask)
tl.store(out_ptr1 + (x2), tmp65, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/hu/chufshtfgijunlmszzic7i4cwvtzymt6wqmgz4g6trkzjukgnhg5.py
# Topologically Sorted Source Nodes: [z, output], Original ATen: [aten.cat, aten._log_softmax]
# Source node to ATen node mapping:
# output => log, sub, sub_1
# z => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%convolution, %convolution],), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%cat, %amax), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub, %log), kwargs = {})
triton_poi_fused__log_softmax_cat_1 = async_compile.triton('triton_poi_fused__log_softmax_cat_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_cat_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_cat_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 16)
x3 = xindex % 16
x1 = (xindex // 4) % 4
x0 = xindex % 4
x4 = xindex
tmp19 = tl.load(in_ptr2 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr3 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp0 = x2
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x3 + (16*x2)), tmp4 & xmask, other=0.0)
tmp6 = tl.load(in_ptr1 + (x1), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.full(tmp7.shape, 0.0, tmp7.dtype)
tmp9 = tl.where(tmp4, tmp7, tmp8)
tmp10 = tmp0 >= tmp3
tmp11 = tl.full([1], 8, tl.int64)
tmp12 = tmp0 < tmp11
tmp13 = tl.load(in_ptr0 + (x3 + (16*((-4) + x2))), tmp10 & xmask, other=0.0)
tmp14 = tl.load(in_ptr1 + (x1), tmp10 & xmask, eviction_policy='evict_last', other=0.0)
tmp15 = tmp13 + tmp14
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp10, tmp15, tmp16)
tmp18 = tl.where(tmp4, tmp9, tmp17)
tmp20 = tmp18 - tmp19
tmp22 = tl_math.log(tmp21)
tmp23 = tmp20 - tmp22
tl.store(out_ptr0 + (x4), tmp23, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [x0], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 2, 2), (16, 4, 2, 1))
buf1 = empty_strided_cuda((8, 1, 2, 2), (4, 32, 2, 1), torch.float32)
buf2 = empty_strided_cuda((8, 1, 2, 2), (4, 32, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [z, output], Original ATen: [aten.cat, aten._log_softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__log_softmax_cat_0.run(buf0, primals_2, buf1, buf2, 32, grid=grid(32), stream=stream0)
buf3 = empty_strided_cuda((8, 4, 2, 2), (16, 4, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [z, output], Original ATen: [aten.cat, aten._log_softmax]
triton_poi_fused__log_softmax_cat_1.run(buf0, primals_2, buf1, buf2, buf3, 128, grid=grid(128), stream=stream0)
del buf0
del buf1
del buf2
del primals_2
return (buf3, primals_1, primals_3, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class Net(nn.Module):
def __init__(self, Cin, Cout):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(Cin, Cout, (3, 3))
def forward(self, x):
x0 = self.conv1(x)
x1 = self.conv1(x)
z = torch.cat([x0, x1])
output = F.log_softmax(z, dim=1)
return output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'Cin': 4, 'Cout': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused__log_softmax_cat_0(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp6 = tl.load(in_ptr1 + 0)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp20 = tl.load(in_ptr1 + 1)
tmp21 = tl.broadcast_to(tmp20, [XBLOCK])
tmp32 = tl.load(in_ptr1 + 2)
tmp33 = tl.broadcast_to(tmp32, [XBLOCK])
tmp44 = tl.load(in_ptr1 + 3)
tmp45 = tl.broadcast_to(tmp44, [XBLOCK])
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 16 * x1), tmp4 & xmask, other=0.0)
tmp8 = tmp5 + tmp7
tmp9 = tl.full(tmp8.shape, 0.0, tmp8.dtype)
tmp10 = tl.where(tmp4, tmp8, tmp9)
tmp11 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp14 = tl.load(in_ptr0 + (x0 + 16 * (-4 + x1)), tmp11 & xmask, other=0.0)
tmp15 = tmp14 + tmp7
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp11, tmp15, tmp16)
tmp18 = tl.where(tmp4, tmp10, tmp17)
tmp19 = tl.load(in_ptr0 + (4 + x0 + 16 * x1), tmp4 & xmask, other=0.0)
tmp22 = tmp19 + tmp21
tmp23 = tl.full(tmp22.shape, 0.0, tmp22.dtype)
tmp24 = tl.where(tmp4, tmp22, tmp23)
tmp25 = tl.load(in_ptr0 + (4 + x0 + 16 * (-4 + x1)), tmp11 & xmask,
other=0.0)
tmp26 = tmp25 + tmp21
tmp27 = tl.full(tmp26.shape, 0.0, tmp26.dtype)
tmp28 = tl.where(tmp11, tmp26, tmp27)
tmp29 = tl.where(tmp4, tmp24, tmp28)
tmp30 = triton_helpers.maximum(tmp18, tmp29)
tmp31 = tl.load(in_ptr0 + (8 + x0 + 16 * x1), tmp4 & xmask, other=0.0)
tmp34 = tmp31 + tmp33
tmp35 = tl.full(tmp34.shape, 0.0, tmp34.dtype)
tmp36 = tl.where(tmp4, tmp34, tmp35)
tmp37 = tl.load(in_ptr0 + (8 + x0 + 16 * (-4 + x1)), tmp11 & xmask,
other=0.0)
tmp38 = tmp37 + tmp33
tmp39 = tl.full(tmp38.shape, 0.0, tmp38.dtype)
tmp40 = tl.where(tmp11, tmp38, tmp39)
tmp41 = tl.where(tmp4, tmp36, tmp40)
tmp42 = triton_helpers.maximum(tmp30, tmp41)
tmp43 = tl.load(in_ptr0 + (12 + x0 + 16 * x1), tmp4 & xmask, other=0.0)
tmp46 = tmp43 + tmp45
tmp47 = tl.full(tmp46.shape, 0.0, tmp46.dtype)
tmp48 = tl.where(tmp4, tmp46, tmp47)
tmp49 = tl.load(in_ptr0 + (12 + x0 + 16 * (-4 + x1)), tmp11 & xmask,
other=0.0)
tmp50 = tmp49 + tmp45
tmp51 = tl.full(tmp50.shape, 0.0, tmp50.dtype)
tmp52 = tl.where(tmp11, tmp50, tmp51)
tmp53 = tl.where(tmp4, tmp48, tmp52)
tmp54 = triton_helpers.maximum(tmp42, tmp53)
tmp55 = tmp18 - tmp54
tmp56 = tl_math.exp(tmp55)
tmp57 = tmp29 - tmp54
tmp58 = tl_math.exp(tmp57)
tmp59 = tmp56 + tmp58
tmp60 = tmp41 - tmp54
tmp61 = tl_math.exp(tmp60)
tmp62 = tmp59 + tmp61
tmp63 = tmp53 - tmp54
tmp64 = tl_math.exp(tmp63)
tmp65 = tmp62 + tmp64
tl.store(out_ptr0 + x2, tmp54, xmask)
tl.store(out_ptr1 + x2, tmp65, xmask)
@triton.jit
def triton_poi_fused__log_softmax_cat_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 16
x3 = xindex % 16
x1 = xindex // 4 % 4
x0 = xindex % 4
x4 = xindex
tmp19 = tl.load(in_ptr2 + (x0 + 4 * x2), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr3 + (x0 + 4 * x2), xmask, eviction_policy=
'evict_last')
tmp0 = x2
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x3 + 16 * x2), tmp4 & xmask, other=0.0)
tmp6 = tl.load(in_ptr1 + x1, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.full(tmp7.shape, 0.0, tmp7.dtype)
tmp9 = tl.where(tmp4, tmp7, tmp8)
tmp10 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp13 = tl.load(in_ptr0 + (x3 + 16 * (-4 + x2)), tmp10 & xmask, other=0.0)
tmp14 = tl.load(in_ptr1 + x1, tmp10 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp15 = tmp13 + tmp14
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp10, tmp15, tmp16)
tmp18 = tl.where(tmp4, tmp9, tmp17)
tmp20 = tmp18 - tmp19
tmp22 = tl_math.log(tmp21)
tmp23 = tmp20 - tmp22
tl.store(out_ptr0 + x4, tmp23, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 2, 2), (16, 4, 2, 1))
buf1 = empty_strided_cuda((8, 1, 2, 2), (4, 32, 2, 1), torch.float32)
buf2 = empty_strided_cuda((8, 1, 2, 2), (4, 32, 2, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__log_softmax_cat_0[grid(32)](buf0, primals_2, buf1,
buf2, 32, XBLOCK=32, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((8, 4, 2, 2), (16, 4, 2, 1), torch.float32)
triton_poi_fused__log_softmax_cat_1[grid(128)](buf0, primals_2,
buf1, buf2, buf3, 128, XBLOCK=128, num_warps=4, num_stages=1)
del buf0
del buf1
del buf2
del primals_2
return buf3, primals_1, primals_3, buf3
class NetNew(nn.Module):
def __init__(self, Cin, Cout):
super(NetNew, self).__init__()
self.conv1 = nn.Conv2d(Cin, Cout, (3, 3))
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| saeta/mlir-npcomp | Net | false | 4,236 | [
"Apache-2.0"
] | 0 | 85898aaf10ea30237ee1d66c977b966cf7fcf6d0 | https://github.com/saeta/mlir-npcomp/tree/85898aaf10ea30237ee1d66c977b966cf7fcf6d0 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, Cin, Cout):
super().__init__()
self.conv1 = nn.Conv2d(Cin, Cout, (3, 3))
def forward(self, x):
x0 = self.conv1(x)
x1 = self.conv1(x)
z = torch.cat([x0, x1])
output = F.log_softmax(z, dim=1)
return output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ChunkSeparationAffine | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/qz/cqzgictjpi53mvqpidmypr6gvl6skq6ldd43ow7utchkt5n2ang3.py
# Topologically Sorted Source Nodes: [inputs_1], Original ATen: [aten.constant_pad_nd]
# Source node to ATen node mapping:
# inputs_1 => constant_pad_nd_1
# Graph fragment:
# %constant_pad_nd_1 : [num_users=2] = call_function[target=torch.ops.aten.constant_pad_nd.default](args = (%slice_3, [0, 0], 0.0), kwargs = {})
triton_poi_fused_constant_pad_nd_0 = async_compile.triton('triton_poi_fused_constant_pad_nd_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_constant_pad_nd_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_constant_pad_nd_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 3
x1 = (xindex // 3)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (1 + x0 + (4*x1)), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nu/cnuyr7sdimes3uxey6dc2ueydyjlbpvwyujhycu27ud2jg4ujepp.py
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# cat => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%convolution, %convolution_1], 1), kwargs = {})
triton_poi_fused_cat_1 = async_compile.triton('triton_poi_fused_cat_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 2) % 4
x0 = xindex % 2
x2 = (xindex // 8)
x3 = xindex
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 2, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (2*x1) + (4*x2)), tmp4 & xmask, other=0.0)
tmp6 = tl.load(in_ptr1 + (x1), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.full(tmp7.shape, 0.0, tmp7.dtype)
tmp9 = tl.where(tmp4, tmp7, tmp8)
tmp10 = tmp0 >= tmp3
tmp11 = tl.full([1], 4, tl.int64)
tmp12 = tmp0 < tmp11
tmp13 = tl.load(in_ptr2 + (x0 + (2*((-2) + x1)) + (4*x2)), tmp10 & xmask, other=0.0)
tmp14 = tl.load(in_ptr3 + ((-2) + x1), tmp10 & xmask, eviction_policy='evict_last', other=0.0)
tmp15 = tmp13 + tmp14
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp10, tmp15, tmp16)
tmp18 = tl.where(tmp4, tmp9, tmp17)
tl.store(out_ptr0 + (x3), tmp18, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (2, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (2, ), (1, ))
assert_size_stride(primals_4, (2, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (2, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(2,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 2, 2), (4, 2, 1))
buf1 = empty_strided_cuda((4, 4, 3), (12, 3, 1), torch.float32)
# Topologically Sorted Source Nodes: [inputs_1], Original ATen: [aten.constant_pad_nd]
stream0 = get_raw_stream(0)
triton_poi_fused_constant_pad_nd_0.run(primals_1, buf1, 48, grid=grid(48), stream=stream0)
# Topologically Sorted Source Nodes: [outputs_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(2,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 2, 2), (4, 2, 1))
buf3 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
triton_poi_fused_cat_1.run(buf0, primals_3, buf2, primals_5, buf3, 32, grid=grid(32), stream=stream0)
del buf0
del buf2
del primals_3
del primals_5
return (buf3, primals_1, primals_2, primals_4, buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((2, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((2, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((2, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((2, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super(TdnnAffine, self).__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not None else 0
total_ops = y.nelement() * (m.input_dim * kernel_ops + bias_ops)
m.total_ops += torch.DoubleTensor([int(total_ops)])
class ChunkSeparationAffine(torch.nn.Module):
"""By this component, the chunk will be grouped to two parts, odd and even.
"""
def __init__(self, input_dim, output_dim, **options):
super(ChunkSeparationAffine, self).__init__()
self.input_dim = input_dim
self.output_dim = output_dim
self.odd = TdnnAffine(input_dim, output_dim // 2, stride=2, **options)
self.even = TdnnAffine(input_dim, output_dim // 2, stride=2, **options)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if inputs.shape[2] % 2 != 0:
inputs = F.pad(inputs, (0, 1), mode='constant', value=0)
return torch.cat((self.odd(inputs), self.even(inputs[:, :, 1:])), dim=1
)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'output_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn.functional as F
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_constant_pad_nd_0(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 3
x1 = xindex // 3
x2 = xindex
tmp0 = tl.load(in_ptr0 + (1 + x0 + 4 * x1), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_cat_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 2 % 4
x0 = xindex % 2
x2 = xindex // 8
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 2, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 2 * x1 + 4 * x2), tmp4 & xmask, other=0.0)
tmp6 = tl.load(in_ptr1 + x1, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.full(tmp7.shape, 0.0, tmp7.dtype)
tmp9 = tl.where(tmp4, tmp7, tmp8)
tmp10 = tmp0 >= tmp3
tl.full([1], 4, tl.int64)
tmp13 = tl.load(in_ptr2 + (x0 + 2 * (-2 + x1) + 4 * x2), tmp10 & xmask,
other=0.0)
tmp14 = tl.load(in_ptr3 + (-2 + x1), tmp10 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp15 = tmp13 + tmp14
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp10, tmp15, tmp16)
tmp18 = tl.where(tmp4, tmp9, tmp17)
tl.store(out_ptr0 + x3, tmp18, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (2, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (2,), (1,))
assert_size_stride(primals_4, (2, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (2,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(2,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 2, 2), (4, 2, 1))
buf1 = empty_strided_cuda((4, 4, 3), (12, 3, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_constant_pad_nd_0[grid(48)](primals_1, buf1, 48,
XBLOCK=64, num_warps=1, num_stages=1)
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(2,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 2, 2), (4, 2, 1))
buf3 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
triton_poi_fused_cat_1[grid(32)](buf0, primals_3, buf2, primals_5,
buf3, 32, XBLOCK=32, num_warps=1, num_stages=1)
del buf0
del buf2
del primals_3
del primals_5
return buf3, primals_1, primals_2, primals_4, buf1
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super(TdnnAffine, self).__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is not None else 0
total_ops = y.nelement() * (m.input_dim * kernel_ops + bias_ops)
m.total_ops += torch.DoubleTensor([int(total_ops)])
class ChunkSeparationAffineNew(torch.nn.Module):
"""By this component, the chunk will be grouped to two parts, odd and even.
"""
def __init__(self, input_dim, output_dim, **options):
super(ChunkSeparationAffineNew, self).__init__()
self.input_dim = input_dim
self.output_dim = output_dim
self.odd = TdnnAffine(input_dim, output_dim // 2, stride=2, **options)
self.even = TdnnAffine(input_dim, output_dim // 2, stride=2, **options)
def forward(self, input_0):
primals_2 = self.odd.weight
primals_3 = self.odd.bias
primals_4 = self.even.weight
primals_5 = self.even.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| qlindazm/asv-subtools | ChunkSeparationAffine | false | 4,237 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn.functional as F
import torch.nn
def to_device(device_object, tensor):
"""
Select device for non-parameters tensor w.r.t model or tensor which has been specified a device.
"""
if isinstance(device_object, torch.nn.Module):
next(device_object.parameters()).device
elif isinstance(device_object, torch.Tensor):
pass
return tensor
class TdnnAffine(torch.nn.Module):
""" An implemented tdnn affine component by conv1d
y = splice(w * x, context) + b
@input_dim: number of dims of frame <=> inputs channels of conv
@output_dim: number of layer nodes <=> outputs channels of conv
@context: a list of context
e.g. [-2,0,2]
If context is [0], then the TdnnAffine is equal to linear layer.
"""
def __init__(self, input_dim, output_dim, context=[0], bias=True, pad=
True, stride=1, groups=1, norm_w=False, norm_f=False):
super().__init__()
assert input_dim % groups == 0
for index in range(0, len(context) - 1):
if context[index] >= context[index + 1]:
raise ValueError(
'Context tuple {} is invalid, such as the order.'.
format(context))
self.input_dim = input_dim
self.output_dim = output_dim
self.context = context
self.bool_bias = bias
self.pad = pad
self.groups = groups
self.norm_w = norm_w
self.norm_f = norm_f
self.stride = stride
self.left_context = context[0] if context[0] < 0 else 0
self.right_context = context[-1] if context[-1] > 0 else 0
self.tot_context = self.right_context - self.left_context + 1
if self.tot_context > 1 and self.norm_f:
self.norm_f = False
None
kernel_size = self.tot_context,
self.weight = torch.nn.Parameter(torch.randn(output_dim, input_dim //
groups, *kernel_size))
if self.bool_bias:
self.bias = torch.nn.Parameter(torch.randn(output_dim))
else:
self.register_parameter('bias', None)
self.init_weight()
if len(context) != self.tot_context:
self.mask = torch.tensor([[[(1 if index in context else 0) for
index in range(self.left_context, self.right_context + 1)]]])
else:
self.mask = None
self.selected_device = False
def init_weight(self):
torch.nn.init.normal_(self.weight, 0.0, 0.01)
if self.bias is not None:
torch.nn.init.constant_(self.bias, 0.0)
def forward(self, inputs):
"""
@inputs: a 3-dimensional tensor (a batch), including [samples-index, frames-dim-index, frames-index]
"""
assert len(inputs.shape) == 3
assert inputs.shape[1] == self.input_dim
if self.pad:
inputs = F.pad(inputs, (-self.left_context, self.right_context),
mode='constant', value=0)
assert inputs.shape[2] >= self.tot_context
if not self.selected_device and self.mask is not None:
self.mask = to_device(self, self.mask)
self.selected_device = True
filters = (self.weight * self.mask if self.mask is not None else
self.weight)
if self.norm_w:
filters = F.normalize(filters, dim=1)
if self.norm_f:
inputs = F.normalize(inputs, dim=1)
outputs = F.conv1d(inputs, filters, self.bias, stride=self.stride,
padding=0, dilation=1, groups=self.groups)
return outputs
def extra_repr(self):
return (
'{input_dim}, {output_dim}, context={context}, bias={bool_bias}, stride={stride}, pad={pad}, groups={groups}, norm_w={norm_w}, norm_f={norm_f}'
.format(**self.__dict__))
@classmethod
def thop_count(self, m, x, y):
x = x[0]
kernel_ops = torch.zeros(m.weight.size()[2:]).numel()
bias_ops = 1 if m.bias is
# ... truncated (>4000 chars) for memory efficiency |
BartClassificationHead | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/nc/cncwsucylpsg2zmlivjfxu6vbd64ztxjndlsix2ysjtby3xohgk4.py
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# hidden_states_2 => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_1,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_tanh_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vz/cvzje67emefmtrrwfoiqmxlqwwubkdlwr3p3l5lnagwt3ifl22gu.py
# Topologically Sorted Source Nodes: [sent_scores_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# sent_scores_1 => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%squeeze, %primals_6), kwargs = {})
triton_poi_fused_mul_1 = async_compile.triton('triton_poi_fused_mul_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (x2), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [hidden_states_2], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_3, 256, grid=grid(256), stream=stream0)
del primals_3
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_4], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sent_scores_1], Original ATen: [aten.mul]
triton_poi_fused_mul_1.run(buf3, primals_6, buf4, 256, grid=grid(256), stream=stream0)
return (buf4, primals_6, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), buf1, buf3, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.utils.checkpoint
class BartClassificationHead(nn.Module):
"""Head for sentence-level classification tasks."""
def __init__(self, input_dim: 'int', inner_dim: 'int', pooler_dropout:
'float'):
super().__init__()
self.dense = nn.Linear(input_dim, inner_dim)
self.dropout = nn.Dropout(p=pooler_dropout)
self.out_proj = nn.Linear(inner_dim, 1, bias=True)
self.sigmoid = nn.Sigmoid()
def forward(self, hidden_states: 'torch.Tensor', mask: 'torch.Tensor'):
hidden_states = self.dropout(hidden_states)
hidden_states = self.dense(hidden_states)
hidden_states = torch.tanh(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.out_proj(hidden_states)
sent_scores = self.sigmoid(hidden_states)
sent_scores = sent_scores.squeeze(-1) * mask.float()
return sent_scores
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'inner_dim': 4, 'pooler_dropout': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_mul_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x2 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + x2, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1,), (1,))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(256)](buf1, primals_3, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_3
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0),
alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_1[grid(256)](buf3, primals_6, buf4, 256,
XBLOCK=128, num_warps=4, num_stages=1)
return buf4, primals_6, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0
), buf1, buf3, primals_4
class BartClassificationHeadNew(nn.Module):
"""Head for sentence-level classification tasks."""
def __init__(self, input_dim: 'int', inner_dim: 'int', pooler_dropout:
'float'):
super().__init__()
self.dense = nn.Linear(input_dim, inner_dim)
self.dropout = nn.Dropout(p=pooler_dropout)
self.out_proj = nn.Linear(inner_dim, 1, bias=True)
self.sigmoid = nn.Sigmoid()
def forward(self, input_0, input_1):
primals_2 = self.dense.weight
primals_3 = self.dense.bias
primals_4 = self.out_proj.weight
primals_5 = self.out_proj.bias
primals_1 = input_0
primals_6 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| sajastu/transformers-sent-curr | BartClassificationHead | false | 4,238 | [
"Apache-2.0"
] | 0 | 6dc41545c4ac298a010090fbca4b454c2eaf3dbb | https://github.com/sajastu/transformers-sent-curr/tree/6dc41545c4ac298a010090fbca4b454c2eaf3dbb | import torch
from torch import nn
import torch.utils.checkpoint
class Model(nn.Module):
"""Head for sentence-level classification tasks."""
def __init__(self, input_dim: 'int', inner_dim: 'int', pooler_dropout:
'float'):
super().__init__()
self.dense = nn.Linear(input_dim, inner_dim)
self.dropout = nn.Dropout(p=pooler_dropout)
self.out_proj = nn.Linear(inner_dim, 1, bias=True)
self.sigmoid = nn.Sigmoid()
def forward(self, hidden_states: 'torch.Tensor', mask: 'torch.Tensor'):
hidden_states = self.dropout(hidden_states)
hidden_states = self.dense(hidden_states)
hidden_states = torch.tanh(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.out_proj(hidden_states)
sent_scores = self.sigmoid(hidden_states)
sent_scores = sent_scores.squeeze(-1) * mask.float()
return sent_scores
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 0.5]
|
GroupedLinearLayer | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/v6/cv6odvhmmcyvquog4eo62pdliew53orxzwe2wfzampr64jy3ppa7.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x_5 => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_4, %primals_3), kwargs = {})
triton_poi_fused_add_0 = async_compile.triton('triton_poi_fused_add_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((1, 64, 4), (256, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(primals_1, (1, 64, 4), (4, 4, 1), 0), primals_2, out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 16, 4), (64, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(buf1, primals_3, 256, grid=grid(256), stream=stream0)
del primals_3
return (buf1, reinterpret_tensor(primals_1, (1, 4, 64), (4, 1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.utils.checkpoint
class GroupedLinearLayer(nn.Module):
def __init__(self, input_size, output_size, num_groups):
super().__init__()
self.input_size = input_size
self.output_size = output_size
self.num_groups = num_groups
self.group_in_dim = self.input_size // self.num_groups
self.group_out_dim = self.output_size // self.num_groups
self.weight = nn.Parameter(torch.empty(self.num_groups, self.
group_in_dim, self.group_out_dim))
self.bias = nn.Parameter(torch.empty(output_size))
def forward(self, hidden_states):
batch_size = list(hidden_states.size())[0]
x = torch.reshape(hidden_states, [-1, self.num_groups, self.
group_in_dim])
x = x.permute(1, 0, 2)
x = torch.matmul(x, self.weight)
x = x.permute(1, 0, 2)
x = torch.reshape(x, [batch_size, -1, self.output_size])
x = x + self.bias
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4, 'output_size': 4, 'num_groups': 1}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((1, 64, 4), (256, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(primals_1, (1, 64, 4), (4, 4,
1), 0), primals_2, out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 16, 4), (64, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](buf1, primals_3, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_3
return buf1, reinterpret_tensor(primals_1, (1, 4, 64), (4, 1, 4), 0)
class GroupedLinearLayerNew(nn.Module):
def __init__(self, input_size, output_size, num_groups):
super().__init__()
self.input_size = input_size
self.output_size = output_size
self.num_groups = num_groups
self.group_in_dim = self.input_size // self.num_groups
self.group_out_dim = self.output_size // self.num_groups
self.weight = nn.Parameter(torch.empty(self.num_groups, self.
group_in_dim, self.group_out_dim))
self.bias = nn.Parameter(torch.empty(output_size))
def forward(self, input_0):
primals_2 = self.weight
primals_3 = self.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| sajastu/transformers-sent-curr | GroupedLinearLayer | false | 4,239 | [
"Apache-2.0"
] | 0 | 6dc41545c4ac298a010090fbca4b454c2eaf3dbb | https://github.com/sajastu/transformers-sent-curr/tree/6dc41545c4ac298a010090fbca4b454c2eaf3dbb | import torch
from torch import nn
import torch.utils.checkpoint
class Model(nn.Module):
def __init__(self, input_size, output_size, num_groups):
super().__init__()
self.input_size = input_size
self.output_size = output_size
self.num_groups = num_groups
self.group_in_dim = self.input_size // self.num_groups
self.group_out_dim = self.output_size // self.num_groups
self.weight = nn.Parameter(torch.empty(self.num_groups, self.
group_in_dim, self.group_out_dim))
self.bias = nn.Parameter(torch.empty(output_size))
def forward(self, hidden_states):
batch_size = list(hidden_states.size())[0]
x = torch.reshape(hidden_states, [-1, self.num_groups, self.
group_in_dim])
x = x.permute(1, 0, 2)
x = torch.matmul(x, self.weight)
x = x.permute(1, 0, 2)
x = torch.reshape(x, [batch_size, -1, self.output_size])
x = x + self.bias
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 1]
|
HubertFeatureProjection | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/mz/cmzntnsms6lzyb35yqnfy7vd7osar32jl5popfgqekaoanmhac6c.py
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states => add, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1.0), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_native_layer_norm_0 = async_compile.triton('triton_poi_fused_native_layer_norm_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1.0
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + (x0), tmp8, xmask)
tl.store(out_ptr1 + (x0), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/lh/clhh73owbiuj4adasmetdqsot2nlmw2ljupnw2q4yt3du76mikww.py
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# hidden_states => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1.0), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_3, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_2), kwargs = {})
triton_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, ), (1, ))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_0.run(primals_3, buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(primals_3, buf0, buf1, primals_1, primals_2, buf2, 256, grid=grid(256), stream=stream0)
del buf0
del buf1
del primals_1
del primals_2
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
return (reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_3, reinterpret_tensor(buf2, (64, 4), (4, 1), 0), primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import torch
from torch import nn
import torch.utils.checkpoint
class HubertFeatureProjection(nn.Module):
def __init__(self, config):
super().__init__()
self.layer_norm = nn.LayerNorm(config.conv_dim[-1], eps=config.
layer_norm_eps)
self.projection = nn.Linear(config.conv_dim[-1], config.hidden_size)
self.dropout = nn.Dropout(config.feat_proj_dropout)
def forward(self, hidden_states):
hidden_states = self.layer_norm(hidden_states)
hidden_states = self.projection(hidden_states)
hidden_states = self.dropout(hidden_states)
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(conv_dim=[4, 4], layer_norm_eps=1,
hidden_size=4, feat_proj_dropout=0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1.0
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp23, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4,), (1,))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
get_raw_stream(0)
triton_poi_fused_native_layer_norm_0[grid(64)](primals_3, buf0,
buf1, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(256)](primals_3, buf0,
buf1, primals_1, primals_2, buf2, 256, XBLOCK=256, num_warps=4,
num_stages=1)
del buf0
del buf1
del primals_1
del primals_2
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf2, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf3)
del primals_5
return reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0
), primals_3, reinterpret_tensor(buf2, (64, 4), (4, 1), 0), primals_4
class HubertFeatureProjectionNew(nn.Module):
def __init__(self, config):
super().__init__()
self.layer_norm = nn.LayerNorm(config.conv_dim[-1], eps=config.
layer_norm_eps)
self.projection = nn.Linear(config.conv_dim[-1], config.hidden_size)
self.dropout = nn.Dropout(config.feat_proj_dropout)
def forward(self, input_0):
primals_1 = self.layer_norm.weight
primals_2 = self.layer_norm.bias
primals_4 = self.projection.weight
primals_5 = self.projection.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| sajastu/transformers-sent-curr | HubertFeatureProjection | false | 4,240 | [
"Apache-2.0"
] | 0 | 6dc41545c4ac298a010090fbca4b454c2eaf3dbb | https://github.com/sajastu/transformers-sent-curr/tree/6dc41545c4ac298a010090fbca4b454c2eaf3dbb | from _paritybench_helpers import _mock_config
import torch
from torch import nn
import torch.utils.checkpoint
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.layer_norm = nn.LayerNorm(config.conv_dim[-1], eps=config.
layer_norm_eps)
self.projection = nn.Linear(config.conv_dim[-1], config.hidden_size)
self.dropout = nn.Dropout(config.feat_proj_dropout)
def forward(self, hidden_states):
hidden_states = self.layer_norm(hidden_states)
hidden_states = self.projection(hidden_states)
hidden_states = self.dropout(hidden_states)
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(conv_dim=[4, 4], layer_norm_eps=1,
hidden_size=4, feat_proj_dropout=0.5)}]
|
Actor | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/a2/ca2wr2cvkya5clovpxidv7ia56pdcyp7uq4omtpg5m2nr7ya3ryn.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_1,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_tanh_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xk/cxkugsynlmnyrjhah42fewrhwovuvurnuv2qimo2qhxq27wjmq7q.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_2 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/jf/cjfzp64ny4hf7wdw5wptah3hqv5fcsh5rrw4brz7uxcy6ad57n7h.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_2 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 64), (64, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (4, 64), (64, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_2, 4096, grid=grid(4096), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0), reinterpret_tensor(primals_4, (64, 64), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
triton_poi_fused_tanh_0.run(buf3, primals_5, 4096, grid=grid(4096), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(primals_6, (64, 4), (1, 64), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
return (buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, buf3, buf6, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
from torch.distributions import Categorical
from torch.distributions import Normal
from torch.distributions import Independent
class Actor(nn.Module):
def __init__(self, obs_dim: 'int', ac_lim: 'float', ac_dim: 'int',
discrete: 'bool'=True):
super().__init__()
self.ac_dim = ac_dim
self.ac_lim = ac_lim
self.obs_dim = obs_dim
self.discrete = discrete
self.fc1 = nn.Linear(obs_dim, 64)
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, ac_dim)
if not self.discrete:
self.log_scale = nn.Parameter(0.3 * torch.ones(self.ac_dim),
requires_grad=True)
def forward(self, x):
x = torch.tanh(self.fc1(x))
x = torch.tanh(self.fc2(x))
if self.discrete:
x = torch.softmax(self.fc3(x), dim=1)
else:
x = torch.tanh(self.fc3(x))
return x
def act(self, obs, deterministic=False):
if self.discrete:
action_prob = self.forward(obs)
dist = Categorical(action_prob)
if deterministic:
action = torch.argmax(action_prob, dim=1)
else:
action = dist.sample()
else:
action_mean = self.forward(obs)
action_mean = action_mean * self.ac_lim
normal = Normal(action_mean, torch.exp(self.log_scale))
dist = Independent(normal, 1)
if deterministic:
action = action_mean.detach()
else:
action = dist.sample()
action_logprobs = dist.log_prob(torch.squeeze(action))
return action, action_logprobs
def get_actions_dist(self, obs):
if self.discrete:
action_prob = self.forward(obs)
dist = Categorical(action_prob)
else:
action_mean = self.forward(obs)
action_mean = action_mean * self.ac_lim
normal = Normal(action_mean, torch.exp(self.log_scale))
dist = Independent(normal, 1)
return dist
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'obs_dim': 4, 'ac_lim': 4, 'ac_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
from torch.distributions import Categorical
from torch.distributions import Normal
from torch.distributions import Independent
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, None)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 64), (64, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (4, 64), (64, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(4096)](buf1, primals_2, 4096, XBLOCK=
256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0),
reinterpret_tensor(primals_4, (64, 64), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf2
triton_poi_fused_tanh_0[grid(4096)](buf3, primals_5, 4096, XBLOCK=
256, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64),
(64, 1), 0), reinterpret_tensor(primals_6, (64, 4), (1, 64), 0),
alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf4, buf5, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf4
triton_poi_fused__softmax_2[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
return buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, buf3, buf6, primals_6, primals_4
class ActorNew(nn.Module):
def __init__(self, obs_dim: 'int', ac_lim: 'float', ac_dim: 'int',
discrete: 'bool'=True):
super().__init__()
self.ac_dim = ac_dim
self.ac_lim = ac_lim
self.obs_dim = obs_dim
self.discrete = discrete
self.fc1 = nn.Linear(obs_dim, 64)
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, ac_dim)
if not self.discrete:
self.log_scale = nn.Parameter(0.3 * torch.ones(self.ac_dim),
requires_grad=True)
def act(self, obs, deterministic=False):
if self.discrete:
action_prob = self.forward(obs)
dist = Categorical(action_prob)
if deterministic:
action = torch.argmax(action_prob, dim=1)
else:
action = dist.sample()
else:
action_mean = self.forward(obs)
action_mean = action_mean * self.ac_lim
normal = Normal(action_mean, torch.exp(self.log_scale))
dist = Independent(normal, 1)
if deterministic:
action = action_mean.detach()
else:
action = dist.sample()
action_logprobs = dist.log_prob(torch.squeeze(action))
return action, action_logprobs
def get_actions_dist(self, obs):
if self.discrete:
action_prob = self.forward(obs)
dist = Categorical(action_prob)
else:
action_mean = self.forward(obs)
action_mean = action_mean * self.ac_lim
normal = Normal(action_mean, torch.exp(self.log_scale))
dist = Independent(normal, 1)
return dist
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| raznem/rlex | Actor | false | 4,242 | [
"MIT"
] | 0 | d24b964d80067becc81d86f6ce87e5be413b7049 | https://github.com/raznem/rlex/tree/d24b964d80067becc81d86f6ce87e5be413b7049 | import torch
import torch.nn as nn
from torch.distributions import Categorical
from torch.distributions import Normal
from torch.distributions import Independent
class Model(nn.Module):
def __init__(self, obs_dim: 'int', ac_lim: 'float', ac_dim: 'int',
discrete: 'bool'=True):
super().__init__()
self.ac_dim = ac_dim
self.ac_lim = ac_lim
self.obs_dim = obs_dim
self.discrete = discrete
self.fc1 = nn.Linear(obs_dim, 64)
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, ac_dim)
if not self.discrete:
self.log_scale = nn.Parameter(0.3 * torch.ones(self.ac_dim),
requires_grad=True)
def forward(self, x):
x = torch.tanh(self.fc1(x))
x = torch.tanh(self.fc2(x))
if self.discrete:
x = torch.softmax(self.fc3(x), dim=1)
else:
x = torch.tanh(self.fc3(x))
return x
def act(self, obs, deterministic=False):
if self.discrete:
action_prob = self.forward(obs)
dist = Categorical(action_prob)
if deterministic:
action = torch.argmax(action_prob, dim=1)
else:
action = dist.sample()
else:
action_mean = self.forward(obs)
action_mean = action_mean * self.ac_lim
normal = Normal(action_mean, torch.exp(self.log_scale))
dist = Independent(normal, 1)
if deterministic:
action = action_mean.detach()
else:
action = dist.sample()
action_logprobs = dist.log_prob(torch.squeeze(action))
return action, action_logprobs
def get_actions_dist(self, obs):
if self.discrete:
action_prob = self.forward(obs)
dist = Categorical(action_prob)
else:
action_mean = self.forward(obs)
action_mean = action_mean * self.ac_lim
normal = Normal(action_mean, torch.exp(self.log_scale))
dist = Independent(normal, 1)
return dist
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
DiceLoss | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/il/cile5rubx7j2trpwwhxvyx2n7vmplsdhfcaobcwnijtfsgj3p43b.py
# Topologically Sorted Source Nodes: [mul, intersection, mul_1, add, sum_2, sum_3, add_1, add_2, dice, sub], Original ATen: [aten.mul, aten.sum, aten.add, aten.div, aten.rsub]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# dice => div
# intersection => sum_1
# mul => mul
# mul_1 => mul_1
# sub => sub
# sum_2 => sum_2
# sum_3 => sum_3
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %view_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%mul,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 2.0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, 1), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%view,), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%view_1,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, %sum_3), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, 1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add, %add_2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %div), kwargs = {})
triton_per_fused_add_div_mul_rsub_sum_0 = async_compile.triton('triton_per_fused_add_div_mul_rsub_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {3: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 4), equal_to_1=(3,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_mul_rsub_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 3, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_div_mul_rsub_sum_0(in_out_ptr0, in_ptr0, in_ptr1, xnumel, rnumel):
xnumel = 1
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp2 = tl.load(in_ptr1 + (r0), None)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = triton_helpers.promote_to_tensor(tl.sum(tmp4, 0))
tmp7 = tl.broadcast_to(tmp1, [RBLOCK])
tmp9 = triton_helpers.promote_to_tensor(tl.sum(tmp7, 0))
tmp10 = tl.broadcast_to(tmp2, [RBLOCK])
tmp12 = triton_helpers.promote_to_tensor(tl.sum(tmp10, 0))
tmp13 = 2.0
tmp14 = tmp6 * tmp13
tmp15 = 1.0
tmp16 = tmp14 + tmp15
tmp17 = tmp9 + tmp12
tmp18 = tmp17 + tmp15
tmp19 = tmp16 / tmp18
tmp20 = tmp15 - tmp19
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp20, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((), (), torch.float32)
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul, intersection, mul_1, add, sum_2, sum_3, add_1, add_2, dice, sub], Original ATen: [aten.mul, aten.sum, aten.add, aten.div, aten.rsub]
stream0 = get_raw_stream(0)
triton_per_fused_add_div_mul_rsub_sum_0.run(buf3, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class DiceLoss(nn.Module):
def __init__(self, weight=None, size_average=True):
super(DiceLoss, self).__init__()
def forward(self, inputs, targets, smooth=1):
inputs = torch.sigmoid(inputs)
inputs = inputs.view(-1)
targets = targets.view(-1)
intersection = (inputs * targets).sum()
dice = (2.0 * intersection + smooth) / (inputs.sum() + targets.sum(
) + smooth)
return 1 - dice
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_per_fused_add_div_mul_rsub_sum_0(in_out_ptr0, in_ptr0, in_ptr1,
xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp2 = tl.load(in_ptr1 + r0, None)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = triton_helpers.promote_to_tensor(tl.sum(tmp4, 0))
tmp7 = tl.broadcast_to(tmp1, [RBLOCK])
tmp9 = triton_helpers.promote_to_tensor(tl.sum(tmp7, 0))
tmp10 = tl.broadcast_to(tmp2, [RBLOCK])
tmp12 = triton_helpers.promote_to_tensor(tl.sum(tmp10, 0))
tmp13 = 2.0
tmp14 = tmp6 * tmp13
tmp15 = 1.0
tmp16 = tmp14 + tmp15
tmp17 = tmp9 + tmp12
tmp18 = tmp17 + tmp15
tmp19 = tmp16 / tmp18
tmp20 = tmp15 - tmp19
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp20, None)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((), (), torch.float32)
buf3 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_div_mul_rsub_sum_0[grid(1)](buf3, arg0_1,
arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf3,
class DiceLossNew(nn.Module):
def __init__(self, weight=None, size_average=True):
super(DiceLossNew, self).__init__()
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| salem-devloper/COVID-Lung-Segment | DiceLoss | false | 4,243 | [
"MIT"
] | 0 | 6896f6b0c56dac6d32e005afd4a94d59b1917b44 | https://github.com/salem-devloper/COVID-Lung-Segment/tree/6896f6b0c56dac6d32e005afd4a94d59b1917b44 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, weight=None, size_average=True):
super().__init__()
def forward(self, inputs, targets, smooth=1):
inputs = torch.sigmoid(inputs)
inputs = inputs.view(-1)
targets = targets.view(-1)
intersection = (inputs * targets).sum()
dice = (2.0 * intersection + smooth) / (inputs.sum() + targets.sum(
) + smooth)
return 1 - dice
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ImageTransformationNet | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/xi/cxi3ssslzv45liamqvbt6decmfms5gkzbjn7dtainfaa436qkyw3.py
# Topologically Sorted Source Nodes: [pad], Original ATen: [aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad => _unsafe_index, _unsafe_index_1
# Graph fragment:
# %_unsafe_index : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%primals_3, [None, None, %sub_1, None]), kwargs = {})
# %_unsafe_index_1 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index, [None, None, None, %sub_1]), kwargs = {})
triton_poi_fused_reflection_pad2d_0 = async_compile.triton('triton_poi_fused_reflection_pad2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[65536],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_reflection_pad2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_reflection_pad2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 62208
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 72
x1 = (xindex // 72) % 72
x2 = (xindex // 5184)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (4095 + ((-1)*(tl_math.abs((-63) + (tl_math.abs((-4) + x0))))) + ((-64)*(tl_math.abs((-63) + (tl_math.abs((-4) + x1))))) + (4096*x2)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/zp/czpuakvx3zciuzfmemejrltenkqbzqirfyy2fnfbmrorwkdndz6e.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
# Source node to ATen node mapping:
# x => convolution
# x_1 => add, rsqrt, var_mean
# Graph fragment:
# %convolution : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_1, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-05), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_red_fused__native_batch_norm_legit_convolution_1 = async_compile.triton('triton_red_fused__native_batch_norm_legit_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.reduction(
size_hints=[128, 4096],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused__native_batch_norm_legit_convolution_1', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_red_fused__native_batch_norm_legit_convolution_1(in_out_ptr0, in_out_ptr1, in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
xnumel = 128
rnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rbase = tl.arange(0, RBLOCK)[None, :]
x3 = xindex
x0 = xindex % 32
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp4_mean = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_m2 = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_weight = tl.zeros([XBLOCK, RBLOCK], tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r2 = rindex
tmp0 = tl.load(in_out_ptr0 + (r2 + (4096*x3)), rmask & xmask, eviction_policy='evict_first', other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp4_mean_next, tmp4_m2_next, tmp4_weight_next = triton_helpers.welford_reduce(
tmp3, tmp4_mean, tmp4_m2, tmp4_weight, roffset == 0
)
tmp4_mean = tl.where(rmask & xmask, tmp4_mean_next, tmp4_mean)
tmp4_m2 = tl.where(rmask & xmask, tmp4_m2_next, tmp4_m2)
tmp4_weight = tl.where(rmask & xmask, tmp4_weight_next, tmp4_weight)
tl.store(in_out_ptr0 + (r2 + (4096*x3)), tmp2, rmask & xmask)
tmp4_tmp, tmp5_tmp, tmp6_tmp = triton_helpers.welford(
tmp4_mean, tmp4_m2, tmp4_weight, 1
)
tmp4 = tmp4_tmp[:, None]
tmp5 = tmp5_tmp[:, None]
tmp6 = tmp6_tmp[:, None]
tl.store(out_ptr0 + (x3), tmp4, xmask)
tmp7 = 4096.0
tmp8 = tmp5 / tmp7
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = libdevice.rsqrt(tmp10)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x3), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/in/ciny2bql3sygecchlvr6rxw73jnhl7dgi3s5w2g2fefaoug53zzz.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.repeat]
# Source node to ATen node mapping:
# x_1 => repeat
# Graph fragment:
# %repeat : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_4, [4]), kwargs = {})
triton_poi_fused_repeat_2 = async_compile.triton('triton_poi_fused_repeat_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_repeat_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_repeat_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0 % 32), xmask)
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ii/ciidusl6utkne6h3zmwx3jccsnttcsdc42mtp3vanldcnxv4y7ov.py
# Topologically Sorted Source Nodes: [x_2, pad_1], Original ATen: [aten.relu, aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad_1 => _unsafe_index_2, _unsafe_index_3
# x_2 => relu
# Graph fragment:
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %_unsafe_index_2 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu, [None, None, %sub_6, None]), kwargs = {})
# %_unsafe_index_3 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_2, [None, None, None, %sub_6]), kwargs = {})
triton_poi_fused_reflection_pad2d_relu_3 = async_compile.triton('triton_poi_fused_reflection_pad2d_relu_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1048576],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_reflection_pad2d_relu_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 557568
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 66
x1 = (xindex // 66) % 66
x2 = (xindex // 4356)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (4095 + ((-1)*(tl_math.abs((-63) + (tl_math.abs((-1) + x0))))) + ((-64)*(tl_math.abs((-63) + (tl_math.abs((-1) + x1))))) + (4096*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x2), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/si/csiohvngy3nd4p3av6rdkonvlcuns665sjcyq5ggukrhfwpso4ay.py
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
# Source node to ATen node mapping:
# x_3 => convolution_1
# x_4 => add_2, rsqrt_1, var_mean_1
# Graph fragment:
# %convolution_1 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_3, %primals_6, %primals_7, [2, 2], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_2, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_2, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_2,), kwargs = {})
triton_per_fused__native_batch_norm_legit_convolution_4 = async_compile.triton('triton_per_fused__native_batch_norm_legit_convolution_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[256, 1024],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_convolution_4', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_convolution_4(in_out_ptr0, in_out_ptr1, in_ptr0, out_ptr0, xnumel, rnumel):
xnumel = 256
XBLOCK: tl.constexpr = 1
rnumel = 1024
RBLOCK: tl.constexpr = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r2 = rindex
x3 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (r2 + (1024*x3)), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = tl.broadcast_to(tmp3, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tmp8 = tl.full([1], 1024, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp3 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 1024.0
tmp17 = tmp15 / tmp16
tmp18 = 1e-05
tmp19 = tmp17 + tmp18
tmp20 = libdevice.rsqrt(tmp19)
tl.store(in_out_ptr0 + (r2 + (1024*x3)), tmp2, None)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x3), tmp20, None)
tl.store(out_ptr0 + (x3), tmp10, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/bo/cbop6byfkkzzjktajzua3ovnpvhy32nxb7dbv364jfeaxunlv7bo.py
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.repeat]
# Source node to ATen node mapping:
# x_4 => repeat_2
# Graph fragment:
# %repeat_2 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_8, [4]), kwargs = {})
triton_poi_fused_repeat_5 = async_compile.triton('triton_poi_fused_repeat_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_repeat_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_repeat_5(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0 % 64), xmask)
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/k6/ck6ljtglelyaqir7indwg3cp4wwudzqtlaof4xfdlyasdzhka7z5.py
# Topologically Sorted Source Nodes: [x_5, pad_2], Original ATen: [aten.relu, aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad_2 => _unsafe_index_4, _unsafe_index_5
# x_5 => relu_1
# Graph fragment:
# %relu_1 : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %_unsafe_index_4 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu_1, [None, None, %sub_11, None]), kwargs = {})
# %_unsafe_index_5 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_4, [None, None, None, %sub_11]), kwargs = {})
triton_poi_fused_reflection_pad2d_relu_6 = async_compile.triton('triton_poi_fused_reflection_pad2d_relu_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[524288],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_reflection_pad2d_relu_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 295936
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 34
x1 = (xindex // 34) % 34
x2 = (xindex // 1156)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (1023 + ((-1)*(tl_math.abs((-31) + (tl_math.abs((-1) + x0))))) + ((-32)*(tl_math.abs((-31) + (tl_math.abs((-1) + x1))))) + (1024*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x2), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/b3/cb3i36nfih3ah5aifo46hyitngbbqmrioka4h7sa3nz2vzd5toin.py
# Topologically Sorted Source Nodes: [x_6, x_7, x_8], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.relu]
# Source node to ATen node mapping:
# x_6 => convolution_2
# x_7 => add_4, repeat_4, repeat_5, rsqrt_2, var_mean_2
# x_8 => relu_2
# Graph fragment:
# %convolution_2 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_5, %primals_10, %primals_11, [2, 2], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %repeat_4 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_12, [4]), kwargs = {})
# %repeat_5 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_13, [4]), kwargs = {})
# %var_mean_2 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_4, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_4, 1e-05), kwargs = {})
# %rsqrt_2 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_4,), kwargs = {})
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_5,), kwargs = {})
triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7 = async_compile.triton('triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[512, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: 'i32', 10: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': True, 'num_load': 4, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, out_ptr3, xnumel, rnumel):
xnumel = 512
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + (x0 % 128), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 % 128), None, eviction_policy='evict_last')
tmp2 = tl.load(in_out_ptr0 + (r3 + (256*x0)), None)
tmp3 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp5 = tl.broadcast_to(tmp4, [RBLOCK])
tmp7 = tl.broadcast_to(tmp5, [RBLOCK])
tmp9 = triton_helpers.promote_to_tensor(tl.sum(tmp7, 0))
tmp10 = tl.full([1], 256, tl.int32)
tmp11 = tmp10.to(tl.float32)
tmp12 = tmp9 / tmp11
tmp13 = tmp5 - tmp12
tmp14 = tmp13 * tmp13
tmp15 = tl.broadcast_to(tmp14, [RBLOCK])
tmp17 = triton_helpers.promote_to_tensor(tl.sum(tmp15, 0))
tmp18 = 256.0
tmp19 = tmp17 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp4 - tmp12
tmp24 = tmp23 * tmp22
tmp25 = tmp24 * tmp0
tmp26 = tmp25 + tmp1
tmp27 = tl.full([1], 0, tl.int32)
tmp28 = triton_helpers.maximum(tmp27, tmp26)
tl.store(out_ptr0 + (x0), tmp0, None)
tl.store(out_ptr1 + (x0), tmp1, None)
tl.store(in_out_ptr0 + (r3 + (256*x0)), tmp4, None)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x0), tmp22, None)
tl.store(out_ptr3 + (r3 + (256*x0)), tmp28, None)
tl.store(out_ptr2 + (x0), tmp12, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/st/cstfzn4z33vdn3t4r76kkdoe3fox63ob7zbuq5lr4e2aj2wo3cfw.py
# Topologically Sorted Source Nodes: [pad_3], Original ATen: [aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad_3 => _unsafe_index_6, _unsafe_index_7
# Graph fragment:
# %_unsafe_index_6 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu_2, [None, None, %sub_16, None]), kwargs = {})
# %_unsafe_index_7 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_6, [None, None, None, %sub_16]), kwargs = {})
triton_poi_fused_reflection_pad2d_8 = async_compile.triton('triton_poi_fused_reflection_pad2d_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_reflection_pad2d_8', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_reflection_pad2d_8(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 165888
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 18
x1 = (xindex // 18) % 18
x2 = (xindex // 324)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (255 + ((-1)*(tl_math.abs((-15) + (tl_math.abs((-1) + x0))))) + ((-16)*(tl_math.abs((-15) + (tl_math.abs((-1) + x1))))) + (256*x2)), None, eviction_policy='evict_last')
tl.store(out_ptr0 + (x3), tmp0, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/e2/ce2xxpjelyctnuhefg5fuzcvwpa544akythto7ai5tgzpkjchqwu.py
# Topologically Sorted Source Nodes: [a, b], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
# Source node to ATen node mapping:
# a => convolution_3
# b => add_6, rsqrt_3, var_mean_3
# Graph fragment:
# %convolution_3 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_7, %primals_14, %primals_15, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %var_mean_3 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_6, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_6, 1e-05), kwargs = {})
# %rsqrt_3 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_6,), kwargs = {})
triton_per_fused__native_batch_norm_legit_convolution_9 = async_compile.triton('triton_per_fused__native_batch_norm_legit_convolution_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[512, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_convolution_9', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_convolution_9(in_out_ptr0, in_out_ptr1, in_ptr0, out_ptr0, xnumel, rnumel):
xnumel = 512
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r2 = rindex
x3 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (r2 + (256*x3)), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = tl.broadcast_to(tmp3, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tmp8 = tl.full([1], 256, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp3 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 256.0
tmp17 = tmp15 / tmp16
tmp18 = 1e-05
tmp19 = tmp17 + tmp18
tmp20 = libdevice.rsqrt(tmp19)
tl.store(in_out_ptr0 + (r2 + (256*x3)), tmp2, None)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x3), tmp20, None)
tl.store(out_ptr0 + (x3), tmp10, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/df/cdfz5yaux6hd3x6u7ywjjuon3rgwzpj6jchxqf6fmzsftmjj7luu.py
# Topologically Sorted Source Nodes: [b], Original ATen: [aten.repeat]
# Source node to ATen node mapping:
# b => repeat_6
# Graph fragment:
# %repeat_6 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_16, [4]), kwargs = {})
triton_poi_fused_repeat_10 = async_compile.triton('triton_poi_fused_repeat_10', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_repeat_10', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_repeat_10(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0 % 128), xmask)
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/72/c72anaicoavbg3ypt27amkloa7kkqjupcqqr7kifcj4pxrdujccb.py
# Topologically Sorted Source Nodes: [c, pad_4], Original ATen: [aten.relu, aten.reflection_pad2d]
# Source node to ATen node mapping:
# c => relu_3
# pad_4 => _unsafe_index_8, _unsafe_index_9
# Graph fragment:
# %relu_3 : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%view_7,), kwargs = {})
# %_unsafe_index_8 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu_3, [None, None, %sub_16, None]), kwargs = {})
# %_unsafe_index_9 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_8, [None, None, None, %sub_16]), kwargs = {})
triton_poi_fused_reflection_pad2d_relu_11 = async_compile.triton('triton_poi_fused_reflection_pad2d_relu_11', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_reflection_pad2d_relu_11', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_11(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 165888
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 18
x1 = (xindex // 18) % 18
x2 = (xindex // 324)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (255 + ((-1)*(tl_math.abs((-15) + (tl_math.abs((-1) + x0))))) + ((-16)*(tl_math.abs((-15) + (tl_math.abs((-1) + x1))))) + (256*x2)), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x2), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x2), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + (x3), tmp10, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/uw/cuwpi6qqweeebr62gd5qsxad6j5c7lwn7o4osxti7yhqwp7fq4gh.py
# Topologically Sorted Source Nodes: [d, e, add, x_9], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add => add_10
# d => convolution_4
# e => add_8, repeat_8, rsqrt_4, var_mean_4
# x_9 => relu_4
# Graph fragment:
# %convolution_4 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_9, %primals_18, %primals_19, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %repeat_8 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_20, [4]), kwargs = {})
# %var_mean_4 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_8, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_8 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_8, 1e-05), kwargs = {})
# %rsqrt_4 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_8,), kwargs = {})
# %add_10 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_9, %relu_2), kwargs = {})
# %relu_4 : [num_users=3] = call_function[target=torch.ops.aten.relu.default](args = (%add_10,), kwargs = {})
# %le_208 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_4, 0), kwargs = {})
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12 = async_compile.triton('triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[512, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*i1', 8: '*fp32', 9: 'i32', 10: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': True, 'num_load': 5, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr3, out_ptr4, xnumel, rnumel):
xnumel = 512
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + (x0 % 128), None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + (256*x0)), None)
tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
tmp27 = tl.load(in_out_ptr1 + (r3 + (256*x0)), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tmp31 = 0.0
tmp32 = tmp30 <= tmp31
tl.store(out_ptr0 + (x0), tmp0, None)
tl.store(in_out_ptr0 + (r3 + (256*x0)), tmp3, None)
tl.store(in_out_ptr1 + (r3 + (256*x0)), tmp30, None)
tl.store(out_ptr3 + (r3 + (256*x0)), tmp32, None)
tl.store(out_ptr4 + (x0), tmp22, None)
tl.store(out_ptr1 + (x0), tmp11, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/sf/csfx6lqqcqwu47chkxt2vd5vxfv4wdyrim5pel6dbqfe4g5taxyv.py
# Topologically Sorted Source Nodes: [d_1, e_1, add_1, x_10], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu]
# Source node to ATen node mapping:
# add_1 => add_15
# d_1 => convolution_6
# e_1 => add_13, repeat_12, rsqrt_6, var_mean_6
# x_10 => relu_6
# Graph fragment:
# %convolution_6 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_13, %primals_26, %primals_27, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %repeat_12 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_28, [4]), kwargs = {})
# %var_mean_6 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_12, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_13 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_12, 1e-05), kwargs = {})
# %rsqrt_6 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_13,), kwargs = {})
# %add_15 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_13, %relu_4), kwargs = {})
# %relu_6 : [num_users=3] = call_function[target=torch.ops.aten.relu.default](args = (%add_15,), kwargs = {})
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13 = async_compile.triton('triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[512, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: 'i32', 9: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'no_x_dim': True, 'num_load': 5, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr3, xnumel, rnumel):
xnumel = 512
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + (x0 % 128), None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + (256*x0)), None)
tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
tmp27 = tl.load(in_out_ptr1 + (r3 + (256*x0)), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tl.store(out_ptr0 + (x0), tmp0, None)
tl.store(in_out_ptr0 + (r3 + (256*x0)), tmp3, None)
tl.store(in_out_ptr1 + (r3 + (256*x0)), tmp30, None)
tl.store(out_ptr3 + (x0), tmp22, None)
tl.store(out_ptr1 + (x0), tmp11, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/4s/c4so2ujlsynrgjmhbb6bdgyvburbxi37bv7qd6z7gktzj7iyvro4.py
# Topologically Sorted Source Nodes: [d_2, e_2, add_2, x_11], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add_2 => add_20
# d_2 => convolution_8
# e_2 => add_18, repeat_16, rsqrt_8, var_mean_8
# x_11 => relu_8
# Graph fragment:
# %convolution_8 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_17, %primals_34, %primals_35, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %repeat_16 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_36, [4]), kwargs = {})
# %var_mean_8 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_16, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_18 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_16, 1e-05), kwargs = {})
# %rsqrt_8 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_18,), kwargs = {})
# %add_20 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %relu_6), kwargs = {})
# %relu_8 : [num_users=3] = call_function[target=torch.ops.aten.relu.default](args = (%add_20,), kwargs = {})
# %le_170 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_6, 0), kwargs = {})
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14 = async_compile.triton('triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[512, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*i1', 9: '*fp32', 10: 'i32', 11: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 5, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, out_ptr3, out_ptr4, out_ptr5, xnumel, rnumel):
xnumel = 512
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + (x0 % 128), None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + (256*x0)), None)
tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr3 + (r3 + (256*x0)), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tmp31 = 0.0
tmp32 = tmp27 <= tmp31
tl.store(out_ptr0 + (x0), tmp0, None)
tl.store(in_out_ptr0 + (r3 + (256*x0)), tmp3, None)
tl.store(out_ptr3 + (r3 + (256*x0)), tmp30, None)
tl.store(out_ptr4 + (r3 + (256*x0)), tmp32, None)
tl.store(out_ptr5 + (x0), tmp22, None)
tl.store(out_ptr1 + (x0), tmp11, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ly/clydupk2lm4bctgb264fhwzs5hdkqtahpqr54znsdhxwmqqhws72.py
# Topologically Sorted Source Nodes: [d_4, e_4, add_4, x_13], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add_4 => add_30
# d_4 => convolution_12
# e_4 => add_28, repeat_24, rsqrt_12, var_mean_12
# x_13 => relu_12
# Graph fragment:
# %convolution_12 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_25, %primals_50, %primals_51, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %repeat_24 : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_52, [4]), kwargs = {})
# %var_mean_12 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_24, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add_28 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_24, 1e-05), kwargs = {})
# %rsqrt_12 : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_28,), kwargs = {})
# %add_30 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_25, %relu_10), kwargs = {})
# %relu_12 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_30,), kwargs = {})
# %le_56 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_12, 0), kwargs = {})
# %le_94 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_10, 0), kwargs = {})
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15 = async_compile.triton('triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[512, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*i1', 9: '*i1', 10: '*fp32', 11: 'i32', 12: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 5, 'num_reduction': 4, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, out_ptr3, out_ptr4, out_ptr5, out_ptr6, xnumel, rnumel):
xnumel = 512
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + (x0 % 128), None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + (256*x0)), None)
tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr3 + (r3 + (256*x0)), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tmp31 = 0.0
tmp32 = tmp30 <= tmp31
tmp33 = tmp27 <= tmp31
tl.store(out_ptr0 + (x0), tmp0, None)
tl.store(in_out_ptr0 + (r3 + (256*x0)), tmp3, None)
tl.store(out_ptr3 + (r3 + (256*x0)), tmp30, None)
tl.store(out_ptr4 + (r3 + (256*x0)), tmp32, None)
tl.store(out_ptr5 + (r3 + (256*x0)), tmp33, None)
tl.store(out_ptr6 + (x0), tmp22, None)
tl.store(out_ptr1 + (x0), tmp11, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/no/cnojzm4gjo6slbfbwabtm5upo3o6uvnigj7no75s5aojpzw5wzr4.py
# Topologically Sorted Source Nodes: [x_14], Original ATen: [aten.arange]
# Source node to ATen node mapping:
# x_14 => iota_26
# Graph fragment:
# %iota_26 : [num_users=2] = call_function[target=torch.ops.prims.iota.default](args = (32,), kwargs = {start: 0, step: 1, dtype: torch.int64, device: cuda:0, requires_grad: False})
triton_poi_fused_arange_16 = async_compile.triton('triton_poi_fused_arange_16', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_arange_16', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_arange_16(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/2h/c2hheqavb5rllatxmrmvnofganisndoonyp3cu676lprf6s7ubpu.py
# Topologically Sorted Source Nodes: [x_14], Original ATen: [aten.arange, aten.add, aten.mul, aten._to_copy]
# Source node to ATen node mapping:
# x_14 => add_31, add_32, convert_element_type, convert_element_type_1, mul_26, mul_27
# Graph fragment:
# %mul_26 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%iota_26, 1), kwargs = {})
# %add_31 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_26, 0), kwargs = {})
# %convert_element_type : [num_users=1] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%add_31, torch.float32), kwargs = {})
# %add_32 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convert_element_type, 0.0), kwargs = {})
# %mul_27 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_32, 0.5), kwargs = {})
# %convert_element_type_1 : [num_users=3] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%mul_27, torch.int64), kwargs = {})
triton_poi_fused__to_copy_add_arange_mul_17 = async_compile.triton('triton_poi_fused__to_copy_add_arange_mul_17', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_add_arange_mul_17', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__to_copy_add_arange_mul_17(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tmp1 = tmp0.to(tl.float32)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tmp3.to(tl.int32)
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/mu/cmukjopi7zbd5nmm7ni2ywqm2h7wm5r7cfj622rlm7uituqip6vb.py
# Topologically Sorted Source Nodes: [x_14, pad_13], Original ATen: [aten._unsafe_index, aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad_13 => _unsafe_index_27, _unsafe_index_28
# x_14 => _unsafe_index_26
# Graph fragment:
# %_unsafe_index_26 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu_12, [None, None, %unsqueeze_52, %convert_element_type_1]), kwargs = {})
# %_unsafe_index_27 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_26, [None, None, %sub_11, None]), kwargs = {})
# %_unsafe_index_28 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_27, [None, None, None, %sub_11]), kwargs = {})
triton_poi_fused__unsafe_index_reflection_pad2d_18 = async_compile.triton('triton_poi_fused__unsafe_index_reflection_pad2d_18', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1048576],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_index_reflection_pad2d_18', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_index_reflection_pad2d_18(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 591872
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x1 = (xindex // 34) % 34
x0 = xindex % 34
x2 = (xindex // 1156)
x5 = xindex
tmp0 = tl.load(in_ptr0 + (31 + ((-1)*(tl_math.abs((-31) + (tl_math.abs((-1) + x1)))))), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (31 + ((-1)*(tl_math.abs((-31) + (tl_math.abs((-1) + x0)))))), None, eviction_policy='evict_last')
tmp1 = tl.full([XBLOCK], 16, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tmp6 = tmp5 + tmp1
tmp7 = tmp5 < 0
tmp8 = tl.where(tmp7, tmp6, tmp5)
tmp9 = tl.load(in_ptr1 + (tmp8 + (16*tmp4) + (256*x2)), None, eviction_policy='evict_last')
tl.store(out_ptr0 + (x5), tmp9, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/in/cinwskifi65bderdnldodgkrn3bylj6iisgi2lh5ipgy6c5ig2rj.py
# Topologically Sorted Source Nodes: [x_18], Original ATen: [aten.arange]
# Source node to ATen node mapping:
# x_18 => iota_30
# Graph fragment:
# %iota_30 : [num_users=2] = call_function[target=torch.ops.prims.iota.default](args = (64,), kwargs = {start: 0, step: 1, dtype: torch.int64, device: cuda:0, requires_grad: False})
triton_poi_fused_arange_19 = async_compile.triton('triton_poi_fused_arange_19', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_arange_19', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_arange_19(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/v4/cv42sdg4rewp3ehn56p4plhqpzwdog65e62nqvd6kjpilkf2iz72.py
# Topologically Sorted Source Nodes: [x_18], Original ATen: [aten.arange, aten.add, aten.mul, aten._to_copy]
# Source node to ATen node mapping:
# x_18 => add_37, add_38, convert_element_type_4, convert_element_type_5, mul_32, mul_33
# Graph fragment:
# %mul_32 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%iota_30, 1), kwargs = {})
# %add_37 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_32, 0), kwargs = {})
# %convert_element_type_4 : [num_users=1] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%add_37, torch.float32), kwargs = {})
# %add_38 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convert_element_type_4, 0.0), kwargs = {})
# %mul_33 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_38, 0.5), kwargs = {})
# %convert_element_type_5 : [num_users=3] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%mul_33, torch.int64), kwargs = {})
triton_poi_fused__to_copy_add_arange_mul_20 = async_compile.triton('triton_poi_fused__to_copy_add_arange_mul_20', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_add_arange_mul_20', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__to_copy_add_arange_mul_20(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tmp1 = tmp0.to(tl.float32)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tmp3.to(tl.int32)
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/3k/c3klftkyl244cvqegoif27ebt6qdixk6jchr3pvyvcb4kgjcn3nu.py
# Topologically Sorted Source Nodes: [x_17, x_18, pad_14], Original ATen: [aten.relu, aten._unsafe_index, aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad_14 => _unsafe_index_30, _unsafe_index_31
# x_17 => relu_13
# x_18 => _unsafe_index_29
# Graph fragment:
# %relu_13 : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%view_27,), kwargs = {})
# %_unsafe_index_29 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu_13, [None, None, %unsqueeze_57, %convert_element_type_5]), kwargs = {})
# %_unsafe_index_30 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_29, [None, None, %sub_6, None]), kwargs = {})
# %_unsafe_index_31 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_30, [None, None, None, %sub_6]), kwargs = {})
triton_poi_fused__unsafe_index_reflection_pad2d_relu_21 = async_compile.triton('triton_poi_fused__unsafe_index_reflection_pad2d_relu_21', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2097152],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_index_reflection_pad2d_relu_21', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_index_reflection_pad2d_relu_21(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1115136
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 66) % 66
x0 = xindex % 66
x2 = (xindex // 4356)
x5 = xindex
tmp0 = tl.load(in_ptr0 + (63 + ((-1)*(tl_math.abs((-63) + (tl_math.abs((-1) + x1)))))), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (63 + ((-1)*(tl_math.abs((-63) + (tl_math.abs((-1) + x0)))))), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr3 + (x2), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr4 + (x2), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr5 + (x2), xmask, eviction_policy='evict_last')
tmp1 = tl.full([XBLOCK], 32, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tmp6 = tmp5 + tmp1
tmp7 = tmp5 < 0
tmp8 = tl.where(tmp7, tmp6, tmp5)
tmp9 = tl.load(in_ptr1 + (tmp8 + (32*tmp4) + (1024*x2)), xmask, eviction_policy='evict_last')
tmp11 = tmp9 - tmp10
tmp13 = tmp11 * tmp12
tmp15 = tmp13 * tmp14
tmp17 = tmp15 + tmp16
tmp18 = tl.full([1], 0, tl.int32)
tmp19 = triton_helpers.maximum(tmp18, tmp17)
tl.store(out_ptr0 + (x5), tmp19, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/uu/cuuhbrvjuevz367ve2ajclcc6uy5k2vtz5hjdsrcf7qo736pfhc3.py
# Topologically Sorted Source Nodes: [x_21, pad_15], Original ATen: [aten.relu, aten.reflection_pad2d]
# Source node to ATen node mapping:
# pad_15 => _unsafe_index_32, _unsafe_index_33
# x_21 => relu_14
# Graph fragment:
# %relu_14 : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%view_29,), kwargs = {})
# %_unsafe_index_32 : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%relu_14, [None, None, %sub_1, None]), kwargs = {})
# %_unsafe_index_33 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index_32, [None, None, None, %sub_1]), kwargs = {})
triton_poi_fused_reflection_pad2d_relu_22 = async_compile.triton('triton_poi_fused_reflection_pad2d_relu_22', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1048576],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_reflection_pad2d_relu_22', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_22(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 663552
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 72
x1 = (xindex // 72) % 72
x2 = (xindex // 5184)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (4095 + ((-1)*(tl_math.abs((-63) + (tl_math.abs((-4) + x0))))) + ((-64)*(tl_math.abs((-63) + (tl_math.abs((-4) + x1))))) + (4096*x2)), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x2), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x2), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + (x3), tmp10, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/eu/ceuar2bmfx55pbipy63jgwlrppvhcxusapefl5k7l5zkznredoci.py
# Topologically Sorted Source Nodes: [x_22], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_22 => convolution_15
# Graph fragment:
# %convolution_15 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_33, %primals_62, %primals_63, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_23 = async_compile.triton('triton_poi_fused_convolution_23', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[65536],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_23', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_23(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 49152
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 4096) % 3
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19, primals_20, primals_21, primals_22, primals_23, primals_24, primals_25, primals_26, primals_27, primals_28, primals_29, primals_30, primals_31, primals_32, primals_33, primals_34, primals_35, primals_36, primals_37, primals_38, primals_39, primals_40, primals_41, primals_42, primals_43, primals_44, primals_45, primals_46, primals_47, primals_48, primals_49, primals_50, primals_51, primals_52, primals_53, primals_54, primals_55, primals_56, primals_57, primals_58, primals_59, primals_60, primals_61, primals_62, primals_63 = args
args.clear()
assert_size_stride(primals_1, (32, 3, 9, 9), (243, 81, 9, 1))
assert_size_stride(primals_2, (32, ), (1, ))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (32, ), (1, ))
assert_size_stride(primals_5, (32, ), (1, ))
assert_size_stride(primals_6, (64, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_7, (64, ), (1, ))
assert_size_stride(primals_8, (64, ), (1, ))
assert_size_stride(primals_9, (64, ), (1, ))
assert_size_stride(primals_10, (128, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_11, (128, ), (1, ))
assert_size_stride(primals_12, (128, ), (1, ))
assert_size_stride(primals_13, (128, ), (1, ))
assert_size_stride(primals_14, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_15, (128, ), (1, ))
assert_size_stride(primals_16, (128, ), (1, ))
assert_size_stride(primals_17, (128, ), (1, ))
assert_size_stride(primals_18, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_19, (128, ), (1, ))
assert_size_stride(primals_20, (128, ), (1, ))
assert_size_stride(primals_21, (128, ), (1, ))
assert_size_stride(primals_22, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_23, (128, ), (1, ))
assert_size_stride(primals_24, (128, ), (1, ))
assert_size_stride(primals_25, (128, ), (1, ))
assert_size_stride(primals_26, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_27, (128, ), (1, ))
assert_size_stride(primals_28, (128, ), (1, ))
assert_size_stride(primals_29, (128, ), (1, ))
assert_size_stride(primals_30, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_31, (128, ), (1, ))
assert_size_stride(primals_32, (128, ), (1, ))
assert_size_stride(primals_33, (128, ), (1, ))
assert_size_stride(primals_34, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_35, (128, ), (1, ))
assert_size_stride(primals_36, (128, ), (1, ))
assert_size_stride(primals_37, (128, ), (1, ))
assert_size_stride(primals_38, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_39, (128, ), (1, ))
assert_size_stride(primals_40, (128, ), (1, ))
assert_size_stride(primals_41, (128, ), (1, ))
assert_size_stride(primals_42, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_43, (128, ), (1, ))
assert_size_stride(primals_44, (128, ), (1, ))
assert_size_stride(primals_45, (128, ), (1, ))
assert_size_stride(primals_46, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_47, (128, ), (1, ))
assert_size_stride(primals_48, (128, ), (1, ))
assert_size_stride(primals_49, (128, ), (1, ))
assert_size_stride(primals_50, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_51, (128, ), (1, ))
assert_size_stride(primals_52, (128, ), (1, ))
assert_size_stride(primals_53, (128, ), (1, ))
assert_size_stride(primals_54, (64, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_55, (64, ), (1, ))
assert_size_stride(primals_56, (64, ), (1, ))
assert_size_stride(primals_57, (64, ), (1, ))
assert_size_stride(primals_58, (32, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_59, (32, ), (1, ))
assert_size_stride(primals_60, (32, ), (1, ))
assert_size_stride(primals_61, (32, ), (1, ))
assert_size_stride(primals_62, (3, 32, 9, 9), (2592, 81, 9, 1))
assert_size_stride(primals_63, (3, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 3, 72, 72), (15552, 5184, 72, 1), torch.float32)
# Topologically Sorted Source Nodes: [pad], Original ATen: [aten.reflection_pad2d]
stream0 = get_raw_stream(0)
triton_poi_fused_reflection_pad2d_0.run(primals_3, buf0, 62208, grid=grid(62208), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 32, 64, 64), (131072, 4096, 64, 1))
buf2 = buf1; del buf1 # reuse
buf5 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 1, 1), torch.float32)
buf6 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 128, 128), torch.float32)
buf8 = reinterpret_tensor(buf6, (1, 128, 1, 1), (128, 1, 1, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_red_fused__native_batch_norm_legit_convolution_1.run(buf2, buf8, primals_2, buf5, 128, 4096, grid=grid(128), stream=stream0)
del primals_2
buf3 = empty_strided_cuda((128, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.repeat]
triton_poi_fused_repeat_2.run(primals_4, buf3, 128, grid=grid(128), stream=stream0)
del primals_4
buf4 = empty_strided_cuda((128, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.repeat]
triton_poi_fused_repeat_2.run(primals_5, buf4, 128, grid=grid(128), stream=stream0)
del primals_5
buf9 = empty_strided_cuda((4, 32, 66, 66), (139392, 4356, 66, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2, pad_1], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_3.run(buf2, buf5, buf8, buf3, buf4, buf9, 557568, grid=grid(557568), stream=stream0)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf10 = extern_kernels.convolution(buf9, primals_6, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 64, 32, 32), (65536, 1024, 32, 1))
buf11 = buf10; del buf10 # reuse
buf14 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 1, 1), torch.float32)
buf15 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 256, 256), torch.float32)
buf17 = reinterpret_tensor(buf15, (1, 256, 1, 1), (256, 1, 1, 1), 0); del buf15 # reuse
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_4.run(buf11, buf17, primals_7, buf14, 256, 1024, grid=grid(256), stream=stream0)
del primals_7
buf12 = empty_strided_cuda((256, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.repeat]
triton_poi_fused_repeat_5.run(primals_8, buf12, 256, grid=grid(256), stream=stream0)
del primals_8
buf13 = empty_strided_cuda((256, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.repeat]
triton_poi_fused_repeat_5.run(primals_9, buf13, 256, grid=grid(256), stream=stream0)
del primals_9
buf18 = empty_strided_cuda((4, 64, 34, 34), (73984, 1156, 34, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_5, pad_2], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_6.run(buf11, buf14, buf17, buf12, buf13, buf18, 295936, grid=grid(295936), stream=stream0)
# Topologically Sorted Source Nodes: [x_6], Original ATen: [aten.convolution]
buf19 = extern_kernels.convolution(buf18, primals_10, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf19, (4, 128, 16, 16), (32768, 256, 16, 1))
buf21 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf22 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf20 = buf19; del buf19 # reuse
buf23 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.float32)
buf24 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf26 = reinterpret_tensor(buf24, (1, 512, 1, 1), (512, 1, 1, 1), 0); del buf24 # reuse
buf27 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_6, x_7, x_8], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.relu]
triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7.run(buf20, buf26, primals_12, primals_13, primals_11, buf21, buf22, buf23, buf27, 512, 256, grid=grid(512), stream=stream0)
del primals_11
del primals_12
del primals_13
buf28 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [pad_3], Original ATen: [aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_8.run(buf27, buf28, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [a], Original ATen: [aten.convolution]
buf29 = extern_kernels.convolution(buf28, primals_14, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf29, (4, 128, 16, 16), (32768, 256, 16, 1))
buf30 = buf29; del buf29 # reuse
buf33 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.float32)
buf34 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf36 = reinterpret_tensor(buf34, (1, 512, 1, 1), (512, 1, 1, 1), 0); del buf34 # reuse
# Topologically Sorted Source Nodes: [a, b], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_9.run(buf30, buf36, primals_15, buf33, 512, 256, grid=grid(512), stream=stream0)
del primals_15
buf31 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_16, buf31, 512, grid=grid(512), stream=stream0)
del primals_16
buf32 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_17, buf32, 512, grid=grid(512), stream=stream0)
del primals_17
buf37 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [c, pad_4], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_11.run(buf30, buf33, buf36, buf31, buf32, buf37, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [d], Original ATen: [aten.convolution]
buf38 = extern_kernels.convolution(buf37, primals_18, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf38, (4, 128, 16, 16), (32768, 256, 16, 1))
buf40 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf39 = buf38; del buf38 # reuse
buf41 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf45 = buf27; del buf27 # reuse
buf147 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.bool)
buf44 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
# Topologically Sorted Source Nodes: [d, e, add, x_9], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12.run(buf39, buf45, primals_20, primals_19, primals_21, buf40, buf41, buf147, buf44, 512, 256, grid=grid(512), stream=stream0)
del primals_19
del primals_20
del primals_21
buf46 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [pad_5], Original ATen: [aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_8.run(buf45, buf46, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [a_1], Original ATen: [aten.convolution]
buf47 = extern_kernels.convolution(buf46, primals_22, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf47, (4, 128, 16, 16), (32768, 256, 16, 1))
buf48 = buf47; del buf47 # reuse
buf51 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.float32)
buf52 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf54 = reinterpret_tensor(buf52, (1, 512, 1, 1), (512, 1, 1, 1), 0); del buf52 # reuse
# Topologically Sorted Source Nodes: [a_1, b_1], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_9.run(buf48, buf54, primals_23, buf51, 512, 256, grid=grid(512), stream=stream0)
del primals_23
buf49 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_1], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_24, buf49, 512, grid=grid(512), stream=stream0)
del primals_24
buf50 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_1], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_25, buf50, 512, grid=grid(512), stream=stream0)
del primals_25
buf55 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [c_1, pad_6], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_11.run(buf48, buf51, buf54, buf49, buf50, buf55, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [d_1], Original ATen: [aten.convolution]
buf56 = extern_kernels.convolution(buf55, primals_26, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf56, (4, 128, 16, 16), (32768, 256, 16, 1))
buf58 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf57 = buf56; del buf56 # reuse
buf59 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf63 = buf45; del buf45 # reuse
buf62 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
# Topologically Sorted Source Nodes: [d_1, e_1, add_1, x_10], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu]
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13.run(buf57, buf63, primals_28, primals_27, primals_29, buf58, buf59, buf62, 512, 256, grid=grid(512), stream=stream0)
del primals_27
del primals_28
del primals_29
buf64 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [pad_7], Original ATen: [aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_8.run(buf63, buf64, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [a_2], Original ATen: [aten.convolution]
buf65 = extern_kernels.convolution(buf64, primals_30, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf65, (4, 128, 16, 16), (32768, 256, 16, 1))
buf66 = buf65; del buf65 # reuse
buf69 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.float32)
buf70 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf72 = reinterpret_tensor(buf70, (1, 512, 1, 1), (512, 1, 1, 1), 0); del buf70 # reuse
# Topologically Sorted Source Nodes: [a_2, b_2], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_9.run(buf66, buf72, primals_31, buf69, 512, 256, grid=grid(512), stream=stream0)
del primals_31
buf67 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_2], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_32, buf67, 512, grid=grid(512), stream=stream0)
del primals_32
buf68 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_2], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_33, buf68, 512, grid=grid(512), stream=stream0)
del primals_33
buf73 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [c_2, pad_8], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_11.run(buf66, buf69, buf72, buf67, buf68, buf73, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [d_2], Original ATen: [aten.convolution]
buf74 = extern_kernels.convolution(buf73, primals_34, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf74, (4, 128, 16, 16), (32768, 256, 16, 1))
buf76 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf75 = buf74; del buf74 # reuse
buf77 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf81 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.float32)
buf146 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.bool)
buf80 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
# Topologically Sorted Source Nodes: [d_2, e_2, add_2, x_11], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14.run(buf75, primals_36, primals_35, primals_37, buf63, buf76, buf77, buf81, buf146, buf80, 512, 256, grid=grid(512), stream=stream0)
del primals_35
del primals_36
del primals_37
buf82 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [pad_9], Original ATen: [aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_8.run(buf81, buf82, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [a_3], Original ATen: [aten.convolution]
buf83 = extern_kernels.convolution(buf82, primals_38, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf83, (4, 128, 16, 16), (32768, 256, 16, 1))
buf84 = buf83; del buf83 # reuse
buf87 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.float32)
buf88 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf90 = reinterpret_tensor(buf88, (1, 512, 1, 1), (512, 1, 1, 1), 0); del buf88 # reuse
# Topologically Sorted Source Nodes: [a_3, b_3], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_9.run(buf84, buf90, primals_39, buf87, 512, 256, grid=grid(512), stream=stream0)
del primals_39
buf85 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_3], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_40, buf85, 512, grid=grid(512), stream=stream0)
del primals_40
buf86 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_3], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_41, buf86, 512, grid=grid(512), stream=stream0)
del primals_41
buf91 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [c_3, pad_10], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_11.run(buf84, buf87, buf90, buf85, buf86, buf91, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [d_3], Original ATen: [aten.convolution]
buf92 = extern_kernels.convolution(buf91, primals_42, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf92, (4, 128, 16, 16), (32768, 256, 16, 1))
buf94 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf93 = buf92; del buf92 # reuse
buf95 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf99 = buf63; del buf63 # reuse
buf145 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.bool)
buf98 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
# Topologically Sorted Source Nodes: [d_3, e_3, add_3, x_12], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14.run(buf93, primals_44, primals_43, primals_45, buf81, buf94, buf95, buf99, buf145, buf98, 512, 256, grid=grid(512), stream=stream0)
del primals_43
del primals_44
del primals_45
buf100 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [pad_11], Original ATen: [aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_8.run(buf99, buf100, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [a_4], Original ATen: [aten.convolution]
buf101 = extern_kernels.convolution(buf100, primals_46, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf101, (4, 128, 16, 16), (32768, 256, 16, 1))
buf102 = buf101; del buf101 # reuse
buf105 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.float32)
buf106 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf108 = reinterpret_tensor(buf106, (1, 512, 1, 1), (512, 1, 1, 1), 0); del buf106 # reuse
# Topologically Sorted Source Nodes: [a_4, b_4], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_9.run(buf102, buf108, primals_47, buf105, 512, 256, grid=grid(512), stream=stream0)
del primals_47
buf103 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_4], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_48, buf103, 512, grid=grid(512), stream=stream0)
del primals_48
buf104 = empty_strided_cuda((512, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [b_4], Original ATen: [aten.repeat]
triton_poi_fused_repeat_10.run(primals_49, buf104, 512, grid=grid(512), stream=stream0)
del primals_49
buf109 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1), torch.float32)
# Topologically Sorted Source Nodes: [c_4, pad_12], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_11.run(buf102, buf105, buf108, buf103, buf104, buf109, 165888, grid=grid(165888), stream=stream0)
# Topologically Sorted Source Nodes: [d_4], Original ATen: [aten.convolution]
buf110 = extern_kernels.convolution(buf109, primals_50, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf110, (4, 128, 16, 16), (32768, 256, 16, 1))
buf112 = empty_strided_cuda((512, ), (1, ), torch.float32)
buf111 = buf110; del buf110 # reuse
buf113 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
buf117 = buf81; del buf81 # reuse
buf143 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.bool)
buf144 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.bool)
buf116 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512), torch.float32)
# Topologically Sorted Source Nodes: [d_4, e_4, add_4, x_13], Original ATen: [aten.convolution, aten.repeat, aten._native_batch_norm_legit, aten.add, aten.relu, aten.threshold_backward]
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15.run(buf111, primals_52, primals_51, primals_53, buf99, buf112, buf113, buf117, buf143, buf144, buf116, 512, 256, grid=grid(512), stream=stream0)
del buf99
del primals_51
del primals_52
del primals_53
buf118 = empty_strided_cuda((32, ), (1, ), torch.int64)
# Topologically Sorted Source Nodes: [x_14], Original ATen: [aten.arange]
triton_poi_fused_arange_16.run(buf118, 32, grid=grid(32), stream=stream0)
buf119 = empty_strided_cuda((32, ), (1, ), torch.int64)
# Topologically Sorted Source Nodes: [x_14], Original ATen: [aten.arange, aten.add, aten.mul, aten._to_copy]
triton_poi_fused__to_copy_add_arange_mul_17.run(buf119, 32, grid=grid(32), stream=stream0)
buf120 = empty_strided_cuda((4, 128, 34, 34), (147968, 1156, 34, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_14, pad_13], Original ATen: [aten._unsafe_index, aten.reflection_pad2d]
triton_poi_fused__unsafe_index_reflection_pad2d_18.run(buf119, buf117, buf120, 591872, grid=grid(591872), stream=stream0)
del buf117
# Topologically Sorted Source Nodes: [x_15], Original ATen: [aten.convolution]
buf121 = extern_kernels.convolution(buf120, primals_54, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf121, (4, 64, 32, 32), (65536, 1024, 32, 1))
buf122 = buf121; del buf121 # reuse
buf125 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 1, 1), torch.float32)
buf126 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 256, 256), torch.float32)
buf128 = reinterpret_tensor(buf126, (1, 256, 1, 1), (256, 1, 1, 1), 0); del buf126 # reuse
# Topologically Sorted Source Nodes: [x_15, x_16], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_per_fused__native_batch_norm_legit_convolution_4.run(buf122, buf128, primals_55, buf125, 256, 1024, grid=grid(256), stream=stream0)
del primals_55
buf123 = empty_strided_cuda((256, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_16], Original ATen: [aten.repeat]
triton_poi_fused_repeat_5.run(primals_56, buf123, 256, grid=grid(256), stream=stream0)
del primals_56
buf124 = empty_strided_cuda((256, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_16], Original ATen: [aten.repeat]
triton_poi_fused_repeat_5.run(primals_57, buf124, 256, grid=grid(256), stream=stream0)
del primals_57
buf129 = empty_strided_cuda((64, ), (1, ), torch.int64)
# Topologically Sorted Source Nodes: [x_18], Original ATen: [aten.arange]
triton_poi_fused_arange_19.run(buf129, 64, grid=grid(64), stream=stream0)
buf130 = empty_strided_cuda((64, ), (1, ), torch.int64)
# Topologically Sorted Source Nodes: [x_18], Original ATen: [aten.arange, aten.add, aten.mul, aten._to_copy]
triton_poi_fused__to_copy_add_arange_mul_20.run(buf130, 64, grid=grid(64), stream=stream0)
buf131 = empty_strided_cuda((4, 64, 66, 66), (278784, 4356, 66, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_17, x_18, pad_14], Original ATen: [aten.relu, aten._unsafe_index, aten.reflection_pad2d]
triton_poi_fused__unsafe_index_reflection_pad2d_relu_21.run(buf130, buf122, buf125, buf128, buf123, buf124, buf131, 1115136, grid=grid(1115136), stream=stream0)
# Topologically Sorted Source Nodes: [x_19], Original ATen: [aten.convolution]
buf132 = extern_kernels.convolution(buf131, primals_58, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf132, (4, 32, 64, 64), (131072, 4096, 64, 1))
buf133 = buf132; del buf132 # reuse
buf136 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 1, 1), torch.float32)
buf137 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 128, 128), torch.float32)
buf139 = reinterpret_tensor(buf137, (1, 128, 1, 1), (128, 1, 1, 1), 0); del buf137 # reuse
# Topologically Sorted Source Nodes: [x_19, x_20], Original ATen: [aten.convolution, aten._native_batch_norm_legit]
triton_red_fused__native_batch_norm_legit_convolution_1.run(buf133, buf139, primals_59, buf136, 128, 4096, grid=grid(128), stream=stream0)
del primals_59
buf134 = empty_strided_cuda((128, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_20], Original ATen: [aten.repeat]
triton_poi_fused_repeat_2.run(primals_60, buf134, 128, grid=grid(128), stream=stream0)
del primals_60
buf135 = empty_strided_cuda((128, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [x_20], Original ATen: [aten.repeat]
triton_poi_fused_repeat_2.run(primals_61, buf135, 128, grid=grid(128), stream=stream0)
del primals_61
buf140 = empty_strided_cuda((4, 32, 72, 72), (165888, 5184, 72, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_21, pad_15], Original ATen: [aten.relu, aten.reflection_pad2d]
triton_poi_fused_reflection_pad2d_relu_22.run(buf133, buf136, buf139, buf134, buf135, buf140, 663552, grid=grid(663552), stream=stream0)
# Topologically Sorted Source Nodes: [x_22], Original ATen: [aten.convolution]
buf141 = extern_kernels.convolution(buf140, primals_62, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf141, (4, 3, 64, 64), (12288, 4096, 64, 1))
buf142 = buf141; del buf141 # reuse
# Topologically Sorted Source Nodes: [x_22], Original ATen: [aten.convolution]
triton_poi_fused_convolution_23.run(buf142, primals_63, 49152, grid=grid(49152), stream=stream0)
del primals_63
return (buf142, primals_1, primals_6, primals_10, primals_14, primals_18, primals_22, primals_26, primals_30, primals_34, primals_38, primals_42, primals_46, primals_50, primals_54, primals_58, primals_62, buf0, buf2, buf3, buf4, buf5, buf8, buf9, buf11, buf12, buf13, buf14, buf17, buf18, buf20, buf21, buf22, buf23, buf26, buf28, buf30, buf31, buf32, buf33, buf36, buf37, buf39, buf40, reinterpret_tensor(buf44, (512, ), (1, ), 0), buf46, buf48, buf49, buf50, buf51, buf54, buf55, buf57, buf58, reinterpret_tensor(buf62, (512, ), (1, ), 0), buf64, buf66, buf67, buf68, buf69, buf72, buf73, buf75, buf76, reinterpret_tensor(buf80, (512, ), (1, ), 0), buf82, buf84, buf85, buf86, buf87, buf90, buf91, buf93, buf94, reinterpret_tensor(buf98, (512, ), (1, ), 0), buf100, buf102, buf103, buf104, buf105, buf108, buf109, buf111, buf112, reinterpret_tensor(buf116, (512, ), (1, ), 0), buf118, buf119, buf120, buf122, buf123, buf124, buf125, buf128, buf129, buf130, buf131, buf133, buf134, buf135, buf136, buf139, buf140, buf143, reinterpret_tensor(buf113, (1, 512, 1, 1), (512, 1, 1, 1), 0), buf144, reinterpret_tensor(buf95, (1, 512, 1, 1), (512, 1, 1, 1), 0), buf145, reinterpret_tensor(buf77, (1, 512, 1, 1), (512, 1, 1, 1), 0), buf146, reinterpret_tensor(buf59, (1, 512, 1, 1), (512, 1, 1, 1), 0), buf147, reinterpret_tensor(buf41, (1, 512, 1, 1), (512, 1, 1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((32, 3, 9, 9), (243, 81, 9, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 64, 64), (12288, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((64, 32, 3, 3), (288, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((128, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_18 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_19 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_20 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_21 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_22 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_23 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_24 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_25 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_26 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_27 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_28 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_29 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_30 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_31 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_32 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_33 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_34 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_35 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_36 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_37 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_38 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_39 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_40 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_41 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_42 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_43 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_44 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_45 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_46 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_47 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_48 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_49 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_50 = rand_strided((128, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_51 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_52 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_53 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_54 = rand_strided((64, 128, 3, 3), (1152, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_55 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_56 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_57 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_58 = rand_strided((32, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_59 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_60 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_61 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_62 = rand_strided((3, 32, 9, 9), (2592, 81, 9, 1), device='cuda:0', dtype=torch.float32)
primals_63 = rand_strided((3, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19, primals_20, primals_21, primals_22, primals_23, primals_24, primals_25, primals_26, primals_27, primals_28, primals_29, primals_30, primals_31, primals_32, primals_33, primals_34, primals_35, primals_36, primals_37, primals_38, primals_39, primals_40, primals_41, primals_42, primals_43, primals_44, primals_45, primals_46, primals_47, primals_48, primals_49, primals_50, primals_51, primals_52, primals_53, primals_54, primals_55, primals_56, primals_57, primals_58, primals_59, primals_60, primals_61, primals_62, primals_63])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class ResidualBlock(nn.Module):
"""
Vanilla convolutional residual block from seminal paper by He et al.
Use of instance normalization suggested by Ulyanov et al. in
https://arxiv.org/pdf/1607.08022.pdf%C2%A0%C2%A0%C2%A0%C2%A0.
"""
def __init__(self, filters=128):
super(ResidualBlock, self).__init__()
self.conv1 = nn.Conv2d(filters, filters, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm1 = nn.InstanceNorm2d(filters, affine=True)
self.conv2 = nn.Conv2d(filters, filters, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm2 = nn.InstanceNorm2d(filters, affine=True)
def forward(self, x):
a = self.conv1(x)
b = self.in_norm1(a)
c = F.relu(b)
d = self.conv2(c)
e = self.in_norm2(d)
return F.relu(e + x)
class ImageTransformationNet(nn.Module):
"""
The image transformation network described in the paper by Johnson et al.,
with instance normalization as suggested by Ulyanov et al.
"""
def __init__(self, vangoh=False):
super(ImageTransformationNet, self).__init__()
self.conv1 = nn.Conv2d(3, 32, (9, 9), padding=(4, 4), padding_mode=
'reflect')
self.in_norm1 = nn.InstanceNorm2d(32, affine=True)
self.conv2 = nn.Conv2d(32, 64, (3, 3), padding=(1, 1), padding_mode
='reflect', stride=2)
self.in_norm2 = nn.InstanceNorm2d(64, affine=True)
self.conv3 = nn.Conv2d(64, 128, (3, 3), padding=(1, 1),
padding_mode='reflect', stride=2)
self.in_norm3 = nn.InstanceNorm2d(128, affine=True)
self.block1 = ResidualBlock()
self.block2 = ResidualBlock()
self.block3 = ResidualBlock()
self.block4 = ResidualBlock()
self.block5 = ResidualBlock()
self.conv4 = nn.Conv2d(128, 64, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm4 = nn.InstanceNorm2d(64, affine=True)
if vangoh:
self.conv5 = nn.ConvTranspose2d(64, 32, (3, 3), padding=(1, 1))
else:
self.conv5 = nn.Conv2d(64, 32, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm5 = nn.InstanceNorm2d(32, affine=True)
self.upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.conv6 = nn.Conv2d(32, 3, (9, 9), padding=(4, 4), padding_mode=
'reflect')
def forward(self, x):
x = self.conv1(x)
x = self.in_norm1(x)
x = F.relu(x)
x = self.conv2(x)
x = self.in_norm2(x)
x = F.relu(x)
x = self.conv3(x)
x = self.in_norm3(x)
x = F.relu(x)
x = self.block1(x)
x = self.block2(x)
x = self.block3(x)
x = self.block4(x)
x = self.block5(x)
x = self.upsample(x)
x = self.conv4(x)
x = self.in_norm4(x)
x = F.relu(x)
x = self.upsample(x)
x = self.conv5(x)
x = self.in_norm5(x)
x = F.relu(x)
x = self.conv6(x)
return x
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_reflection_pad2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 62208
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 72
x1 = xindex // 72 % 72
x2 = xindex // 5184
x3 = xindex
tmp0 = tl.load(in_ptr0 + (4095 + -1 * tl_math.abs(-63 + tl_math.abs(-4 +
x0)) + -64 * tl_math.abs(-63 + tl_math.abs(-4 + x1)) + 4096 * x2),
xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + x3, tmp0, xmask)
@triton.jit
def triton_red_fused__native_batch_norm_legit_convolution_1(in_out_ptr0,
in_out_ptr1, in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK: tl.constexpr,
RBLOCK: tl.constexpr):
xnumel = 128
rnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rbase = tl.arange(0, RBLOCK)[None, :]
x3 = xindex
x0 = xindex % 32
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp4_mean = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_m2 = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_weight = tl.zeros([XBLOCK, RBLOCK], tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r2 = rindex
tmp0 = tl.load(in_out_ptr0 + (r2 + 4096 * x3), rmask & xmask,
eviction_policy='evict_first', other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp4_mean_next, tmp4_m2_next, tmp4_weight_next = (triton_helpers.
welford_reduce(tmp3, tmp4_mean, tmp4_m2, tmp4_weight, roffset == 0)
)
tmp4_mean = tl.where(rmask & xmask, tmp4_mean_next, tmp4_mean)
tmp4_m2 = tl.where(rmask & xmask, tmp4_m2_next, tmp4_m2)
tmp4_weight = tl.where(rmask & xmask, tmp4_weight_next, tmp4_weight)
tl.store(in_out_ptr0 + (r2 + 4096 * x3), tmp2, rmask & xmask)
tmp4_tmp, tmp5_tmp, tmp6_tmp = triton_helpers.welford(tmp4_mean,
tmp4_m2, tmp4_weight, 1)
tmp4 = tmp4_tmp[:, None]
tmp5 = tmp5_tmp[:, None]
tmp6_tmp[:, None]
tl.store(out_ptr0 + x3, tmp4, xmask)
tmp7 = 4096.0
tmp8 = tmp5 / tmp7
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = libdevice.rsqrt(tmp10)
tl.debug_barrier()
tl.store(in_out_ptr1 + x3, tmp11, xmask)
@triton.jit
def triton_poi_fused_repeat_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0 % 32, xmask)
tl.store(out_ptr0 + x0, tmp0, xmask)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_3(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 557568
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 66
x1 = xindex // 66 % 66
x2 = xindex // 4356
x3 = xindex
tmp0 = tl.load(in_ptr0 + (4095 + -1 * tl_math.abs(-63 + tl_math.abs(-1 +
x0)) + -64 * tl_math.abs(-63 + tl_math.abs(-1 + x1)) + 4096 * x2),
xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x2, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_per_fused__native_batch_norm_legit_convolution_4(in_out_ptr0,
in_out_ptr1, in_ptr0, out_ptr0, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r2 = rindex
x3 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (r2 + 1024 * x3), None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = tl.broadcast_to(tmp3, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tmp8 = tl.full([1], 1024, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp3 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 1024.0
tmp17 = tmp15 / tmp16
tmp18 = 1e-05
tmp19 = tmp17 + tmp18
tmp20 = libdevice.rsqrt(tmp19)
tl.store(in_out_ptr0 + (r2 + 1024 * x3), tmp2, None)
tl.debug_barrier()
tl.store(in_out_ptr1 + x3, tmp20, None)
tl.store(out_ptr0 + x3, tmp10, None)
@triton.jit
def triton_poi_fused_repeat_5(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0 % 64, xmask)
tl.store(out_ptr0 + x0, tmp0, xmask)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 295936
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 34
x1 = xindex // 34 % 34
x2 = xindex // 1156
x3 = xindex
tmp0 = tl.load(in_ptr0 + (1023 + -1 * tl_math.abs(-31 + tl_math.abs(-1 +
x0)) + -32 * tl_math.abs(-31 + tl_math.abs(-1 + x1)) + 1024 * x2),
xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x2, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7(
in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1,
out_ptr2, out_ptr3, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + x0 % 128, None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x0 % 128, None, eviction_policy='evict_last')
tmp2 = tl.load(in_out_ptr0 + (r3 + 256 * x0), None)
tmp3 = tl.load(in_ptr2 + x1, None, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp5 = tl.broadcast_to(tmp4, [RBLOCK])
tmp7 = tl.broadcast_to(tmp5, [RBLOCK])
tmp9 = triton_helpers.promote_to_tensor(tl.sum(tmp7, 0))
tmp10 = tl.full([1], 256, tl.int32)
tmp11 = tmp10.to(tl.float32)
tmp12 = tmp9 / tmp11
tmp13 = tmp5 - tmp12
tmp14 = tmp13 * tmp13
tmp15 = tl.broadcast_to(tmp14, [RBLOCK])
tmp17 = triton_helpers.promote_to_tensor(tl.sum(tmp15, 0))
tmp18 = 256.0
tmp19 = tmp17 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp4 - tmp12
tmp24 = tmp23 * tmp22
tmp25 = tmp24 * tmp0
tmp26 = tmp25 + tmp1
tmp27 = tl.full([1], 0, tl.int32)
tmp28 = triton_helpers.maximum(tmp27, tmp26)
tl.store(out_ptr0 + x0, tmp0, None)
tl.store(out_ptr1 + x0, tmp1, None)
tl.store(in_out_ptr0 + (r3 + 256 * x0), tmp4, None)
tl.debug_barrier()
tl.store(in_out_ptr1 + x0, tmp22, None)
tl.store(out_ptr3 + (r3 + 256 * x0), tmp28, None)
tl.store(out_ptr2 + x0, tmp12, None)
@triton.jit
def triton_poi_fused_reflection_pad2d_8(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 18
x1 = xindex // 18 % 18
x2 = xindex // 324
x3 = xindex
tmp0 = tl.load(in_ptr0 + (255 + -1 * tl_math.abs(-15 + tl_math.abs(-1 +
x0)) + -16 * tl_math.abs(-15 + tl_math.abs(-1 + x1)) + 256 * x2),
None, eviction_policy='evict_last')
tl.store(out_ptr0 + x3, tmp0, None)
@triton.jit
def triton_per_fused__native_batch_norm_legit_convolution_9(in_out_ptr0,
in_out_ptr1, in_ptr0, out_ptr0, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r2 = rindex
x3 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (r2 + 256 * x3), None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = tl.broadcast_to(tmp3, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tmp8 = tl.full([1], 256, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp3 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 256.0
tmp17 = tmp15 / tmp16
tmp18 = 1e-05
tmp19 = tmp17 + tmp18
tmp20 = libdevice.rsqrt(tmp19)
tl.store(in_out_ptr0 + (r2 + 256 * x3), tmp2, None)
tl.debug_barrier()
tl.store(in_out_ptr1 + x3, tmp20, None)
tl.store(out_ptr0 + x3, tmp10, None)
@triton.jit
def triton_poi_fused_repeat_10(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0 % 128, xmask)
tl.store(out_ptr0 + x0, tmp0, xmask)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_11(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 18
x1 = xindex // 18 % 18
x2 = xindex // 324
x3 = xindex
tmp0 = tl.load(in_ptr0 + (255 + -1 * tl_math.abs(-15 + tl_math.abs(-1 +
x0)) + -16 * tl_math.abs(-15 + tl_math.abs(-1 + x1)) + 256 * x2),
None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x2, None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x2, None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + x3, tmp10, None)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12(
in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1,
out_ptr3, out_ptr4, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + x0 % 128, None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + 256 * x0), None)
tmp2 = tl.load(in_ptr1 + x1, None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + x1, None, eviction_policy='evict_last')
tmp27 = tl.load(in_out_ptr1 + (r3 + 256 * x0), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tmp31 = 0.0
tmp32 = tmp30 <= tmp31
tl.store(out_ptr0 + x0, tmp0, None)
tl.store(in_out_ptr0 + (r3 + 256 * x0), tmp3, None)
tl.store(in_out_ptr1 + (r3 + 256 * x0), tmp30, None)
tl.store(out_ptr3 + (r3 + 256 * x0), tmp32, None)
tl.store(out_ptr4 + x0, tmp22, None)
tl.store(out_ptr1 + x0, tmp11, None)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13(
in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1,
out_ptr3, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + x0 % 128, None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + 256 * x0), None)
tmp2 = tl.load(in_ptr1 + x1, None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + x1, None, eviction_policy='evict_last')
tmp27 = tl.load(in_out_ptr1 + (r3 + 256 * x0), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tl.store(out_ptr0 + x0, tmp0, None)
tl.store(in_out_ptr0 + (r3 + 256 * x0), tmp3, None)
tl.store(in_out_ptr1 + (r3 + 256 * x0), tmp30, None)
tl.store(out_ptr3 + x0, tmp22, None)
tl.store(out_ptr1 + x0, tmp11, None)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14(
in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1,
out_ptr3, out_ptr4, out_ptr5, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + x0 % 128, None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + 256 * x0), None)
tmp2 = tl.load(in_ptr1 + x1, None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + x1, None, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr3 + (r3 + 256 * x0), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tmp31 = 0.0
tmp32 = tmp27 <= tmp31
tl.store(out_ptr0 + x0, tmp0, None)
tl.store(in_out_ptr0 + (r3 + 256 * x0), tmp3, None)
tl.store(out_ptr3 + (r3 + 256 * x0), tmp30, None)
tl.store(out_ptr4 + (r3 + 256 * x0), tmp32, None)
tl.store(out_ptr5 + x0, tmp22, None)
tl.store(out_ptr1 + x0, tmp11, None)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15(
in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1,
out_ptr3, out_ptr4, out_ptr5, out_ptr6, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
x0 = xindex
r3 = rindex
x1 = xindex % 128
tmp0 = tl.load(in_ptr0 + x0 % 128, None, eviction_policy='evict_last')
tmp1 = tl.load(in_out_ptr0 + (r3 + 256 * x0), None)
tmp2 = tl.load(in_ptr1 + x1, None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + x1, None, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr3 + (r3 + 256 * x0), None)
tmp3 = tmp1 + tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = tl.broadcast_to(tmp4, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = tl.full([1], 256, tl.int32)
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 / tmp10
tmp12 = tmp4 - tmp11
tmp13 = tmp12 * tmp12
tmp14 = tl.broadcast_to(tmp13, [RBLOCK])
tmp16 = triton_helpers.promote_to_tensor(tl.sum(tmp14, 0))
tmp17 = tmp3 - tmp11
tmp18 = 256.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp24 = tmp23 * tmp0
tmp26 = tmp24 + tmp25
tmp28 = tmp26 + tmp27
tmp29 = tl.full([1], 0, tl.int32)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tmp31 = 0.0
tmp32 = tmp30 <= tmp31
tmp33 = tmp27 <= tmp31
tl.store(out_ptr0 + x0, tmp0, None)
tl.store(in_out_ptr0 + (r3 + 256 * x0), tmp3, None)
tl.store(out_ptr3 + (r3 + 256 * x0), tmp30, None)
tl.store(out_ptr4 + (r3 + 256 * x0), tmp32, None)
tl.store(out_ptr5 + (r3 + 256 * x0), tmp33, None)
tl.store(out_ptr6 + x0, tmp22, None)
tl.store(out_ptr1 + x0, tmp11, None)
@triton.jit
def triton_poi_fused_arange_16(out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tl.store(out_ptr0 + x0, tmp0, xmask)
@triton.jit
def triton_poi_fused__to_copy_add_arange_mul_17(out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tmp1 = tmp0.to(tl.float32)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tmp3.to(tl.int32)
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_poi_fused__unsafe_index_reflection_pad2d_18(in_ptr0, in_ptr1,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x1 = xindex // 34 % 34
x0 = xindex % 34
x2 = xindex // 1156
x5 = xindex
tmp0 = tl.load(in_ptr0 + (31 + -1 * tl_math.abs(-31 + tl_math.abs(-1 +
x1))), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (31 + -1 * tl_math.abs(-31 + tl_math.abs(-1 +
x0))), None, eviction_policy='evict_last')
tmp1 = tl.full([XBLOCK], 16, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tmp6 = tmp5 + tmp1
tmp7 = tmp5 < 0
tmp8 = tl.where(tmp7, tmp6, tmp5)
tmp9 = tl.load(in_ptr1 + (tmp8 + 16 * tmp4 + 256 * x2), None,
eviction_policy='evict_last')
tl.store(out_ptr0 + x5, tmp9, None)
@triton.jit
def triton_poi_fused_arange_19(out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tl.store(out_ptr0 + x0, tmp0, xmask)
@triton.jit
def triton_poi_fused__to_copy_add_arange_mul_20(out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tmp1 = tmp0.to(tl.float32)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tmp3.to(tl.int32)
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_poi_fused__unsafe_index_reflection_pad2d_relu_21(in_ptr0,
in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 1115136
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 66 % 66
x0 = xindex % 66
x2 = xindex // 4356
x5 = xindex
tmp0 = tl.load(in_ptr0 + (63 + -1 * tl_math.abs(-63 + tl_math.abs(-1 +
x1))), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (63 + -1 * tl_math.abs(-63 + tl_math.abs(-1 +
x0))), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr3 + x2, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr4 + x2, xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr5 + x2, xmask, eviction_policy='evict_last')
tmp1 = tl.full([XBLOCK], 32, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tmp6 = tmp5 + tmp1
tmp7 = tmp5 < 0
tmp8 = tl.where(tmp7, tmp6, tmp5)
tmp9 = tl.load(in_ptr1 + (tmp8 + 32 * tmp4 + 1024 * x2), xmask,
eviction_policy='evict_last')
tmp11 = tmp9 - tmp10
tmp13 = tmp11 * tmp12
tmp15 = tmp13 * tmp14
tmp17 = tmp15 + tmp16
tmp18 = tl.full([1], 0, tl.int32)
tmp19 = triton_helpers.maximum(tmp18, tmp17)
tl.store(out_ptr0 + x5, tmp19, xmask)
@triton.jit
def triton_poi_fused_reflection_pad2d_relu_22(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 72
x1 = xindex // 72 % 72
x2 = xindex // 5184
x3 = xindex
tmp0 = tl.load(in_ptr0 + (4095 + -1 * tl_math.abs(-63 + tl_math.abs(-4 +
x0)) + -64 * tl_math.abs(-63 + tl_math.abs(-4 + x1)) + 4096 * x2),
None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x2, None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x2, None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tl.store(out_ptr0 + x3, tmp10, None)
@triton.jit
def triton_poi_fused_convolution_23(in_out_ptr0, in_ptr0, xnumel, XBLOCK:
tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 4096 % 3
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13, primals_14, primals_15, primals_16, primals_17,
primals_18, primals_19, primals_20, primals_21, primals_22,
primals_23, primals_24, primals_25, primals_26, primals_27,
primals_28, primals_29, primals_30, primals_31, primals_32,
primals_33, primals_34, primals_35, primals_36, primals_37,
primals_38, primals_39, primals_40, primals_41, primals_42,
primals_43, primals_44, primals_45, primals_46, primals_47,
primals_48, primals_49, primals_50, primals_51, primals_52,
primals_53, primals_54, primals_55, primals_56, primals_57,
primals_58, primals_59, primals_60, primals_61, primals_62, primals_63
) = args
args.clear()
assert_size_stride(primals_1, (32, 3, 9, 9), (243, 81, 9, 1))
assert_size_stride(primals_2, (32,), (1,))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (32,), (1,))
assert_size_stride(primals_5, (32,), (1,))
assert_size_stride(primals_6, (64, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_7, (64,), (1,))
assert_size_stride(primals_8, (64,), (1,))
assert_size_stride(primals_9, (64,), (1,))
assert_size_stride(primals_10, (128, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_11, (128,), (1,))
assert_size_stride(primals_12, (128,), (1,))
assert_size_stride(primals_13, (128,), (1,))
assert_size_stride(primals_14, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_15, (128,), (1,))
assert_size_stride(primals_16, (128,), (1,))
assert_size_stride(primals_17, (128,), (1,))
assert_size_stride(primals_18, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_19, (128,), (1,))
assert_size_stride(primals_20, (128,), (1,))
assert_size_stride(primals_21, (128,), (1,))
assert_size_stride(primals_22, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_23, (128,), (1,))
assert_size_stride(primals_24, (128,), (1,))
assert_size_stride(primals_25, (128,), (1,))
assert_size_stride(primals_26, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_27, (128,), (1,))
assert_size_stride(primals_28, (128,), (1,))
assert_size_stride(primals_29, (128,), (1,))
assert_size_stride(primals_30, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_31, (128,), (1,))
assert_size_stride(primals_32, (128,), (1,))
assert_size_stride(primals_33, (128,), (1,))
assert_size_stride(primals_34, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_35, (128,), (1,))
assert_size_stride(primals_36, (128,), (1,))
assert_size_stride(primals_37, (128,), (1,))
assert_size_stride(primals_38, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_39, (128,), (1,))
assert_size_stride(primals_40, (128,), (1,))
assert_size_stride(primals_41, (128,), (1,))
assert_size_stride(primals_42, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_43, (128,), (1,))
assert_size_stride(primals_44, (128,), (1,))
assert_size_stride(primals_45, (128,), (1,))
assert_size_stride(primals_46, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_47, (128,), (1,))
assert_size_stride(primals_48, (128,), (1,))
assert_size_stride(primals_49, (128,), (1,))
assert_size_stride(primals_50, (128, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_51, (128,), (1,))
assert_size_stride(primals_52, (128,), (1,))
assert_size_stride(primals_53, (128,), (1,))
assert_size_stride(primals_54, (64, 128, 3, 3), (1152, 9, 3, 1))
assert_size_stride(primals_55, (64,), (1,))
assert_size_stride(primals_56, (64,), (1,))
assert_size_stride(primals_57, (64,), (1,))
assert_size_stride(primals_58, (32, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_59, (32,), (1,))
assert_size_stride(primals_60, (32,), (1,))
assert_size_stride(primals_61, (32,), (1,))
assert_size_stride(primals_62, (3, 32, 9, 9), (2592, 81, 9, 1))
assert_size_stride(primals_63, (3,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 3, 72, 72), (15552, 5184, 72, 1),
torch.float32)
get_raw_stream(0)
triton_poi_fused_reflection_pad2d_0[grid(62208)](primals_3, buf0,
62208, XBLOCK=256, num_warps=4, num_stages=1)
del primals_3
buf1 = extern_kernels.convolution(buf0, primals_1, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 32, 64, 64), (131072, 4096, 64, 1))
buf2 = buf1
del buf1
buf5 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 1, 1), torch.float32
)
buf6 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 128, 128), torch
.float32)
buf8 = reinterpret_tensor(buf6, (1, 128, 1, 1), (128, 1, 1, 1), 0)
del buf6
triton_red_fused__native_batch_norm_legit_convolution_1[grid(128)](buf2
, buf8, primals_2, buf5, 128, 4096, XBLOCK=1, RBLOCK=2048,
num_warps=16, num_stages=1)
del primals_2
buf3 = empty_strided_cuda((128,), (1,), torch.float32)
triton_poi_fused_repeat_2[grid(128)](primals_4, buf3, 128, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_4
buf4 = empty_strided_cuda((128,), (1,), torch.float32)
triton_poi_fused_repeat_2[grid(128)](primals_5, buf4, 128, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_5
buf9 = empty_strided_cuda((4, 32, 66, 66), (139392, 4356, 66, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_3[grid(557568)](buf2, buf5,
buf8, buf3, buf4, buf9, 557568, XBLOCK=512, num_warps=8,
num_stages=1)
buf10 = extern_kernels.convolution(buf9, primals_6, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 64, 32, 32), (65536, 1024, 32, 1))
buf11 = buf10
del buf10
buf14 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 1, 1), torch.
float32)
buf15 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 256, 256),
torch.float32)
buf17 = reinterpret_tensor(buf15, (1, 256, 1, 1), (256, 1, 1, 1), 0)
del buf15
triton_per_fused__native_batch_norm_legit_convolution_4[grid(256)](
buf11, buf17, primals_7, buf14, 256, 1024, num_warps=8,
num_stages=1)
del primals_7
buf12 = empty_strided_cuda((256,), (1,), torch.float32)
triton_poi_fused_repeat_5[grid(256)](primals_8, buf12, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_8
buf13 = empty_strided_cuda((256,), (1,), torch.float32)
triton_poi_fused_repeat_5[grid(256)](primals_9, buf13, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_9
buf18 = empty_strided_cuda((4, 64, 34, 34), (73984, 1156, 34, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_6[grid(295936)](buf11, buf14,
buf17, buf12, buf13, buf18, 295936, XBLOCK=1024, num_warps=4,
num_stages=1)
buf19 = extern_kernels.convolution(buf18, primals_10, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf19, (4, 128, 16, 16), (32768, 256, 16, 1))
buf21 = empty_strided_cuda((512,), (1,), torch.float32)
buf22 = empty_strided_cuda((512,), (1,), torch.float32)
buf20 = buf19
del buf19
buf23 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.
float32)
buf24 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf26 = reinterpret_tensor(buf24, (1, 512, 1, 1), (512, 1, 1, 1), 0)
del buf24
buf27 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.float32)
triton_per_fused__native_batch_norm_legit_convolution_relu_repeat_7[
grid(512)](buf20, buf26, primals_12, primals_13, primals_11,
buf21, buf22, buf23, buf27, 512, 256, num_warps=2, num_stages=1)
del primals_11
del primals_12
del primals_13
buf28 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_8[grid(165888)](buf27, buf28,
165888, XBLOCK=512, num_warps=8, num_stages=1)
buf29 = extern_kernels.convolution(buf28, primals_14, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf29, (4, 128, 16, 16), (32768, 256, 16, 1))
buf30 = buf29
del buf29
buf33 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.
float32)
buf34 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf36 = reinterpret_tensor(buf34, (1, 512, 1, 1), (512, 1, 1, 1), 0)
del buf34
triton_per_fused__native_batch_norm_legit_convolution_9[grid(512)](
buf30, buf36, primals_15, buf33, 512, 256, num_warps=2,
num_stages=1)
del primals_15
buf31 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_16, buf31, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_16
buf32 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_17, buf32, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_17
buf37 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_11[grid(165888)](buf30,
buf33, buf36, buf31, buf32, buf37, 165888, XBLOCK=1024,
num_warps=4, num_stages=1)
buf38 = extern_kernels.convolution(buf37, primals_18, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf38, (4, 128, 16, 16), (32768, 256, 16, 1))
buf40 = empty_strided_cuda((512,), (1,), torch.float32)
buf39 = buf38
del buf38
buf41 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf45 = buf27
del buf27
buf147 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.bool)
buf44 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_12[
grid(512)](buf39, buf45, primals_20, primals_19, primals_21,
buf40, buf41, buf147, buf44, 512, 256, num_warps=2, num_stages=1)
del primals_19
del primals_20
del primals_21
buf46 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_8[grid(165888)](buf45, buf46,
165888, XBLOCK=512, num_warps=8, num_stages=1)
buf47 = extern_kernels.convolution(buf46, primals_22, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf47, (4, 128, 16, 16), (32768, 256, 16, 1))
buf48 = buf47
del buf47
buf51 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.
float32)
buf52 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf54 = reinterpret_tensor(buf52, (1, 512, 1, 1), (512, 1, 1, 1), 0)
del buf52
triton_per_fused__native_batch_norm_legit_convolution_9[grid(512)](
buf48, buf54, primals_23, buf51, 512, 256, num_warps=2,
num_stages=1)
del primals_23
buf49 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_24, buf49, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_24
buf50 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_25, buf50, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_25
buf55 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_11[grid(165888)](buf48,
buf51, buf54, buf49, buf50, buf55, 165888, XBLOCK=1024,
num_warps=4, num_stages=1)
buf56 = extern_kernels.convolution(buf55, primals_26, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf56, (4, 128, 16, 16), (32768, 256, 16, 1))
buf58 = empty_strided_cuda((512,), (1,), torch.float32)
buf57 = buf56
del buf56
buf59 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf63 = buf45
del buf45
buf62 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_13[
grid(512)](buf57, buf63, primals_28, primals_27, primals_29,
buf58, buf59, buf62, 512, 256, num_warps=2, num_stages=1)
del primals_27
del primals_28
del primals_29
buf64 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_8[grid(165888)](buf63, buf64,
165888, XBLOCK=512, num_warps=8, num_stages=1)
buf65 = extern_kernels.convolution(buf64, primals_30, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf65, (4, 128, 16, 16), (32768, 256, 16, 1))
buf66 = buf65
del buf65
buf69 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.
float32)
buf70 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf72 = reinterpret_tensor(buf70, (1, 512, 1, 1), (512, 1, 1, 1), 0)
del buf70
triton_per_fused__native_batch_norm_legit_convolution_9[grid(512)](
buf66, buf72, primals_31, buf69, 512, 256, num_warps=2,
num_stages=1)
del primals_31
buf67 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_32, buf67, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_32
buf68 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_33, buf68, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_33
buf73 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_11[grid(165888)](buf66,
buf69, buf72, buf67, buf68, buf73, 165888, XBLOCK=1024,
num_warps=4, num_stages=1)
buf74 = extern_kernels.convolution(buf73, primals_34, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf74, (4, 128, 16, 16), (32768, 256, 16, 1))
buf76 = empty_strided_cuda((512,), (1,), torch.float32)
buf75 = buf74
del buf74
buf77 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf81 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.float32)
buf146 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.bool)
buf80 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14[
grid(512)](buf75, primals_36, primals_35, primals_37, buf63,
buf76, buf77, buf81, buf146, buf80, 512, 256, num_warps=2,
num_stages=1)
del primals_35
del primals_36
del primals_37
buf82 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_8[grid(165888)](buf81, buf82,
165888, XBLOCK=512, num_warps=8, num_stages=1)
buf83 = extern_kernels.convolution(buf82, primals_38, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf83, (4, 128, 16, 16), (32768, 256, 16, 1))
buf84 = buf83
del buf83
buf87 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.
float32)
buf88 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf90 = reinterpret_tensor(buf88, (1, 512, 1, 1), (512, 1, 1, 1), 0)
del buf88
triton_per_fused__native_batch_norm_legit_convolution_9[grid(512)](
buf84, buf90, primals_39, buf87, 512, 256, num_warps=2,
num_stages=1)
del primals_39
buf85 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_40, buf85, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_40
buf86 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_41, buf86, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_41
buf91 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_11[grid(165888)](buf84,
buf87, buf90, buf85, buf86, buf91, 165888, XBLOCK=1024,
num_warps=4, num_stages=1)
buf92 = extern_kernels.convolution(buf91, primals_42, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf92, (4, 128, 16, 16), (32768, 256, 16, 1))
buf94 = empty_strided_cuda((512,), (1,), torch.float32)
buf93 = buf92
del buf92
buf95 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf99 = buf63
del buf63
buf145 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.bool)
buf98 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_14[
grid(512)](buf93, primals_44, primals_43, primals_45, buf81,
buf94, buf95, buf99, buf145, buf98, 512, 256, num_warps=2,
num_stages=1)
del primals_43
del primals_44
del primals_45
buf100 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_8[grid(165888)](buf99, buf100,
165888, XBLOCK=512, num_warps=8, num_stages=1)
buf101 = extern_kernels.convolution(buf100, primals_46, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf101, (4, 128, 16, 16), (32768, 256, 16, 1))
buf102 = buf101
del buf101
buf105 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 1, 1), torch.
float32)
buf106 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf108 = reinterpret_tensor(buf106, (1, 512, 1, 1), (512, 1, 1, 1), 0)
del buf106
triton_per_fused__native_batch_norm_legit_convolution_9[grid(512)](
buf102, buf108, primals_47, buf105, 512, 256, num_warps=2,
num_stages=1)
del primals_47
buf103 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_48, buf103, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_48
buf104 = empty_strided_cuda((512,), (1,), torch.float32)
triton_poi_fused_repeat_10[grid(512)](primals_49, buf104, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_49
buf109 = empty_strided_cuda((4, 128, 18, 18), (41472, 324, 18, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_11[grid(165888)](buf102,
buf105, buf108, buf103, buf104, buf109, 165888, XBLOCK=1024,
num_warps=4, num_stages=1)
buf110 = extern_kernels.convolution(buf109, primals_50, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf110, (4, 128, 16, 16), (32768, 256, 16, 1))
buf112 = empty_strided_cuda((512,), (1,), torch.float32)
buf111 = buf110
del buf110
buf113 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
buf117 = buf81
del buf81
buf143 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.bool)
buf144 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.bool)
buf116 = empty_strided_cuda((1, 512, 1, 1), (512, 1, 512, 512),
torch.float32)
triton_per_fused__native_batch_norm_legit_add_convolution_relu_repeat_threshold_backward_15[
grid(512)](buf111, primals_52, primals_51, primals_53, buf99,
buf112, buf113, buf117, buf143, buf144, buf116, 512, 256,
num_warps=2, num_stages=1)
del buf99
del primals_51
del primals_52
del primals_53
buf118 = empty_strided_cuda((32,), (1,), torch.int64)
triton_poi_fused_arange_16[grid(32)](buf118, 32, XBLOCK=32,
num_warps=1, num_stages=1)
buf119 = empty_strided_cuda((32,), (1,), torch.int64)
triton_poi_fused__to_copy_add_arange_mul_17[grid(32)](buf119, 32,
XBLOCK=32, num_warps=1, num_stages=1)
buf120 = empty_strided_cuda((4, 128, 34, 34), (147968, 1156, 34, 1),
torch.float32)
triton_poi_fused__unsafe_index_reflection_pad2d_18[grid(591872)](buf119
, buf117, buf120, 591872, XBLOCK=1024, num_warps=4, num_stages=1)
del buf117
buf121 = extern_kernels.convolution(buf120, primals_54, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf121, (4, 64, 32, 32), (65536, 1024, 32, 1))
buf122 = buf121
del buf121
buf125 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 1, 1), torch.
float32)
buf126 = empty_strided_cuda((1, 256, 1, 1), (256, 1, 256, 256),
torch.float32)
buf128 = reinterpret_tensor(buf126, (1, 256, 1, 1), (256, 1, 1, 1), 0)
del buf126
triton_per_fused__native_batch_norm_legit_convolution_4[grid(256)](
buf122, buf128, primals_55, buf125, 256, 1024, num_warps=8,
num_stages=1)
del primals_55
buf123 = empty_strided_cuda((256,), (1,), torch.float32)
triton_poi_fused_repeat_5[grid(256)](primals_56, buf123, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_56
buf124 = empty_strided_cuda((256,), (1,), torch.float32)
triton_poi_fused_repeat_5[grid(256)](primals_57, buf124, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_57
buf129 = empty_strided_cuda((64,), (1,), torch.int64)
triton_poi_fused_arange_19[grid(64)](buf129, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf130 = empty_strided_cuda((64,), (1,), torch.int64)
triton_poi_fused__to_copy_add_arange_mul_20[grid(64)](buf130, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf131 = empty_strided_cuda((4, 64, 66, 66), (278784, 4356, 66, 1),
torch.float32)
triton_poi_fused__unsafe_index_reflection_pad2d_relu_21[grid(1115136)](
buf130, buf122, buf125, buf128, buf123, buf124, buf131, 1115136,
XBLOCK=512, num_warps=8, num_stages=1)
buf132 = extern_kernels.convolution(buf131, primals_58, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf132, (4, 32, 64, 64), (131072, 4096, 64, 1))
buf133 = buf132
del buf132
buf136 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 1, 1), torch.
float32)
buf137 = empty_strided_cuda((1, 128, 1, 1), (128, 1, 128, 128),
torch.float32)
buf139 = reinterpret_tensor(buf137, (1, 128, 1, 1), (128, 1, 1, 1), 0)
del buf137
triton_red_fused__native_batch_norm_legit_convolution_1[grid(128)](
buf133, buf139, primals_59, buf136, 128, 4096, XBLOCK=1, RBLOCK
=2048, num_warps=16, num_stages=1)
del primals_59
buf134 = empty_strided_cuda((128,), (1,), torch.float32)
triton_poi_fused_repeat_2[grid(128)](primals_60, buf134, 128,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_60
buf135 = empty_strided_cuda((128,), (1,), torch.float32)
triton_poi_fused_repeat_2[grid(128)](primals_61, buf135, 128,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_61
buf140 = empty_strided_cuda((4, 32, 72, 72), (165888, 5184, 72, 1),
torch.float32)
triton_poi_fused_reflection_pad2d_relu_22[grid(663552)](buf133,
buf136, buf139, buf134, buf135, buf140, 663552, XBLOCK=512,
num_warps=8, num_stages=1)
buf141 = extern_kernels.convolution(buf140, primals_62, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf141, (4, 3, 64, 64), (12288, 4096, 64, 1))
buf142 = buf141
del buf141
triton_poi_fused_convolution_23[grid(49152)](buf142, primals_63,
49152, XBLOCK=256, num_warps=4, num_stages=1)
del primals_63
return (buf142, primals_1, primals_6, primals_10, primals_14,
primals_18, primals_22, primals_26, primals_30, primals_34,
primals_38, primals_42, primals_46, primals_50, primals_54,
primals_58, primals_62, buf0, buf2, buf3, buf4, buf5, buf8, buf9,
buf11, buf12, buf13, buf14, buf17, buf18, buf20, buf21, buf22,
buf23, buf26, buf28, buf30, buf31, buf32, buf33, buf36, buf37,
buf39, buf40, reinterpret_tensor(buf44, (512,), (1,), 0), buf46,
buf48, buf49, buf50, buf51, buf54, buf55, buf57, buf58,
reinterpret_tensor(buf62, (512,), (1,), 0), buf64, buf66, buf67,
buf68, buf69, buf72, buf73, buf75, buf76, reinterpret_tensor(buf80,
(512,), (1,), 0), buf82, buf84, buf85, buf86, buf87, buf90, buf91,
buf93, buf94, reinterpret_tensor(buf98, (512,), (1,), 0), buf100,
buf102, buf103, buf104, buf105, buf108, buf109, buf111, buf112,
reinterpret_tensor(buf116, (512,), (1,), 0), buf118, buf119, buf120,
buf122, buf123, buf124, buf125, buf128, buf129, buf130, buf131,
buf133, buf134, buf135, buf136, buf139, buf140, buf143,
reinterpret_tensor(buf113, (1, 512, 1, 1), (512, 1, 1, 1), 0),
buf144, reinterpret_tensor(buf95, (1, 512, 1, 1), (512, 1, 1, 1), 0
), buf145, reinterpret_tensor(buf77, (1, 512, 1, 1), (512, 1, 1, 1),
0), buf146, reinterpret_tensor(buf59, (1, 512, 1, 1), (512, 1, 1, 1
), 0), buf147, reinterpret_tensor(buf41, (1, 512, 1, 1), (512, 1, 1,
1), 0))
class ResidualBlock(nn.Module):
"""
Vanilla convolutional residual block from seminal paper by He et al.
Use of instance normalization suggested by Ulyanov et al. in
https://arxiv.org/pdf/1607.08022.pdf%C2%A0%C2%A0%C2%A0%C2%A0.
"""
def __init__(self, filters=128):
super(ResidualBlock, self).__init__()
self.conv1 = nn.Conv2d(filters, filters, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm1 = nn.InstanceNorm2d(filters, affine=True)
self.conv2 = nn.Conv2d(filters, filters, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm2 = nn.InstanceNorm2d(filters, affine=True)
def forward(self, x):
a = self.conv1(x)
b = self.in_norm1(a)
c = F.relu(b)
d = self.conv2(c)
e = self.in_norm2(d)
return F.relu(e + x)
class ImageTransformationNetNew(nn.Module):
"""
The image transformation network described in the paper by Johnson et al.,
with instance normalization as suggested by Ulyanov et al.
"""
def __init__(self, vangoh=False):
super(ImageTransformationNetNew, self).__init__()
self.conv1 = nn.Conv2d(3, 32, (9, 9), padding=(4, 4), padding_mode=
'reflect')
self.in_norm1 = nn.InstanceNorm2d(32, affine=True)
self.conv2 = nn.Conv2d(32, 64, (3, 3), padding=(1, 1), padding_mode
='reflect', stride=2)
self.in_norm2 = nn.InstanceNorm2d(64, affine=True)
self.conv3 = nn.Conv2d(64, 128, (3, 3), padding=(1, 1),
padding_mode='reflect', stride=2)
self.in_norm3 = nn.InstanceNorm2d(128, affine=True)
self.block1 = ResidualBlock()
self.block2 = ResidualBlock()
self.block3 = ResidualBlock()
self.block4 = ResidualBlock()
self.block5 = ResidualBlock()
self.conv4 = nn.Conv2d(128, 64, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm4 = nn.InstanceNorm2d(64, affine=True)
if vangoh:
self.conv5 = nn.ConvTranspose2d(64, 32, (3, 3), padding=(1, 1))
else:
self.conv5 = nn.Conv2d(64, 32, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm5 = nn.InstanceNorm2d(32, affine=True)
self.upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.conv6 = nn.Conv2d(32, 3, (9, 9), padding=(4, 4), padding_mode=
'reflect')
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.in_norm1.weight
primals_5 = self.in_norm1.bias
primals_6 = self.conv2.weight
primals_7 = self.conv2.bias
primals_8 = self.in_norm2.weight
primals_9 = self.in_norm2.bias
primals_10 = self.conv3.weight
primals_11 = self.conv3.bias
primals_12 = self.in_norm3.weight
primals_13 = self.in_norm3.bias
primals_14 = self.block1.conv1.weight
primals_15 = self.block1.conv1.bias
primals_16 = self.block1.in_norm1.weight
primals_17 = self.block1.in_norm1.bias
primals_18 = self.block1.conv2.weight
primals_19 = self.block1.conv2.bias
primals_20 = self.block1.in_norm2.weight
primals_21 = self.block1.in_norm2.bias
primals_22 = self.block2.conv1.weight
primals_23 = self.block2.conv1.bias
primals_24 = self.block2.in_norm1.weight
primals_25 = self.block2.in_norm1.bias
primals_26 = self.block2.conv2.weight
primals_27 = self.block2.conv2.bias
primals_28 = self.block2.in_norm2.weight
primals_29 = self.block2.in_norm2.bias
primals_30 = self.block3.conv1.weight
primals_31 = self.block3.conv1.bias
primals_32 = self.block3.in_norm1.weight
primals_33 = self.block3.in_norm1.bias
primals_34 = self.block3.conv2.weight
primals_35 = self.block3.conv2.bias
primals_36 = self.block3.in_norm2.weight
primals_37 = self.block3.in_norm2.bias
primals_38 = self.block4.conv1.weight
primals_39 = self.block4.conv1.bias
primals_40 = self.block4.in_norm1.weight
primals_41 = self.block4.in_norm1.bias
primals_42 = self.block4.conv2.weight
primals_43 = self.block4.conv2.bias
primals_44 = self.block4.in_norm2.weight
primals_45 = self.block4.in_norm2.bias
primals_46 = self.block5.conv1.weight
primals_47 = self.block5.conv1.bias
primals_48 = self.block5.in_norm1.weight
primals_49 = self.block5.in_norm1.bias
primals_50 = self.block5.conv2.weight
primals_51 = self.block5.conv2.bias
primals_52 = self.block5.in_norm2.weight
primals_53 = self.block5.in_norm2.bias
primals_54 = self.conv4.weight
primals_55 = self.conv4.bias
primals_56 = self.in_norm4.weight
primals_57 = self.in_norm4.bias
primals_58 = self.conv5.weight
primals_59 = self.conv5.bias
primals_60 = self.in_norm5.weight
primals_61 = self.in_norm5.bias
primals_62 = self.conv6.weight
primals_63 = self.conv6.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13, primals_14,
primals_15, primals_16, primals_17, primals_18, primals_19,
primals_20, primals_21, primals_22, primals_23, primals_24,
primals_25, primals_26, primals_27, primals_28, primals_29,
primals_30, primals_31, primals_32, primals_33, primals_34,
primals_35, primals_36, primals_37, primals_38, primals_39,
primals_40, primals_41, primals_42, primals_43, primals_44,
primals_45, primals_46, primals_47, primals_48, primals_49,
primals_50, primals_51, primals_52, primals_53, primals_54,
primals_55, primals_56, primals_57, primals_58, primals_59,
primals_60, primals_61, primals_62, primals_63])
return output[0]
| rileypsmith/Fast-Style-Transfer | ImageTransformationNet | false | 4,244 | [
"MIT"
] | 0 | 8b2164f8bc6d63530f914610b6c5c5c1b0f4ffd5 | https://github.com/rileypsmith/Fast-Style-Transfer/tree/8b2164f8bc6d63530f914610b6c5c5c1b0f4ffd5 | import torch
import torch.nn as nn
import torch.nn.functional as F
class ResidualBlock(nn.Module):
"""
Vanilla convolutional residual block from seminal paper by He et al.
Use of instance normalization suggested by Ulyanov et al. in
https://arxiv.org/pdf/1607.08022.pdf%C2%A0%C2%A0%C2%A0%C2%A0.
"""
def __init__(self, filters=128):
super().__init__()
self.conv1 = nn.Conv2d(filters, filters, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm1 = nn.InstanceNorm2d(filters, affine=True)
self.conv2 = nn.Conv2d(filters, filters, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm2 = nn.InstanceNorm2d(filters, affine=True)
def forward(self, x):
a = self.conv1(x)
b = self.in_norm1(a)
c = F.relu(b)
d = self.conv2(c)
e = self.in_norm2(d)
return F.relu(e + x)
class Model(nn.Module):
"""
The image transformation network described in the paper by Johnson et al.,
with instance normalization as suggested by Ulyanov et al.
"""
def __init__(self, vangoh=False):
super().__init__()
self.conv1 = nn.Conv2d(3, 32, (9, 9), padding=(4, 4), padding_mode=
'reflect')
self.in_norm1 = nn.InstanceNorm2d(32, affine=True)
self.conv2 = nn.Conv2d(32, 64, (3, 3), padding=(1, 1), padding_mode
='reflect', stride=2)
self.in_norm2 = nn.InstanceNorm2d(64, affine=True)
self.conv3 = nn.Conv2d(64, 128, (3, 3), padding=(1, 1),
padding_mode='reflect', stride=2)
self.in_norm3 = nn.InstanceNorm2d(128, affine=True)
self.block1 = ResidualBlock()
self.block2 = ResidualBlock()
self.block3 = ResidualBlock()
self.block4 = ResidualBlock()
self.block5 = ResidualBlock()
self.conv4 = nn.Conv2d(128, 64, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm4 = nn.InstanceNorm2d(64, affine=True)
if vangoh:
self.conv5 = nn.ConvTranspose2d(64, 32, (3, 3), padding=(1, 1))
else:
self.conv5 = nn.Conv2d(64, 32, (3, 3), padding=(1, 1),
padding_mode='reflect')
self.in_norm5 = nn.InstanceNorm2d(32, affine=True)
self.upsample = nn.Upsample(scale_factor=2, mode='nearest')
self.conv6 = nn.Conv2d(32, 3, (9, 9), padding=(4, 4), padding_mode=
'reflect')
def forward(self, x):
x = self.conv1(x)
x = self.in_norm1(x)
x = F.relu(x)
x = self.conv2(x)
x = self.in_norm2(x)
x = F.relu(x)
x = self.conv3(x)
x = self.in_norm3(x)
x = F.relu(x)
x = self.block1(x)
x = self.block2(x)
x = self.block3(x)
x = self.block4(x)
x = self.block5(x)
x = self.upsample(x)
x = self.conv4(x)
x = self.in_norm4(x)
x = F.relu(x)
x = self.upsample(x)
x = self.conv5(x)
x = self.in_norm5(x)
x = F.relu(x)
x = self.conv6(x)
return x
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return []
|
LayerNormCustom | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/zf/czfnaeipqg4a3qzttb2l6zy5ng44vshk3lfmp25jc2er665hxsmw.py
# Topologically Sorted Source Nodes: [u, sub], Original ATen: [aten.mean, aten.sub]
# Source node to ATen node mapping:
# sub => sub
# u => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_1, %mean), kwargs = {})
triton_poi_fused_mean_sub_0 = async_compile.triton('triton_poi_fused_mean_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mean_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mean_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = tmp0 - tmp9
tl.store(out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/k3/ck3awyjmlyoxvkizg2opx6vtglv26uioox7nr33aabc2cmbcxgpr.py
# Topologically Sorted Source Nodes: [pow_1, s, add, sqrt, x, mul, add_1], Original ATen: [aten.pow, aten.mean, aten.add, aten.sqrt, aten.div, aten.mul]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# mul => mul
# pow_1 => pow_1
# s => mean_1
# sqrt => sqrt
# x => div
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%pow_1, [-1], True), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean_1, 1e-12), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add,), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %sqrt), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %div), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_3), kwargs = {})
triton_poi_fused_add_div_mean_mul_pow_sqrt_1 = async_compile.triton('triton_poi_fused_add_div_mean_mul_pow_sqrt_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_mean_mul_pow_sqrt_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 7, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_mean_mul_pow_sqrt_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp2 * tmp2
tmp5 = tmp4 * tmp4
tmp6 = tmp3 + tmp5
tmp8 = tmp7 * tmp7
tmp9 = tmp6 + tmp8
tmp11 = tmp10 * tmp10
tmp12 = tmp9 + tmp11
tmp13 = 4.0
tmp14 = tmp12 / tmp13
tmp15 = 1e-12
tmp16 = tmp14 + tmp15
tmp17 = libdevice.sqrt(tmp16)
tmp18 = tmp1 / tmp17
tmp19 = tmp0 * tmp18
tmp21 = tmp19 + tmp20
tl.store(out_ptr0 + (x2), tmp21, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [u, sub], Original ATen: [aten.mean, aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_mean_sub_0.run(primals_1, buf0, 256, grid=grid(256), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pow_1, s, add, sqrt, x, mul, add_1], Original ATen: [aten.pow, aten.mean, aten.add, aten.sqrt, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_pow_sqrt_1.run(primals_2, buf0, primals_3, buf1, 256, grid=grid(256), stream=stream0)
del buf0
del primals_2
del primals_3
return (buf1, primals_1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class LayerNormCustom(nn.Module):
"""A layernorm module in the TF style (epsilon inside the square root)."""
def __init__(self, n_hidden, variance_epsilon=1e-12):
super().__init__()
self.gamma = nn.Parameter(torch.ones(n_hidden))
self.beta = nn.Parameter(torch.zeros(n_hidden))
self.variance_epsilon = variance_epsilon
def forward(self, x):
u = x.mean(1, keepdim=True)
s = (x - u).pow(2).mean(-1, keepdim=True)
x = (x - u) / torch.sqrt(s + self.variance_epsilon)
return self.gamma * x + self.beta
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_hidden': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_mean_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = tmp0 - tmp9
tl.store(out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_div_mean_mul_pow_sqrt_1(in_ptr0, in_ptr1, in_ptr2,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp20 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp2 * tmp2
tmp5 = tmp4 * tmp4
tmp6 = tmp3 + tmp5
tmp8 = tmp7 * tmp7
tmp9 = tmp6 + tmp8
tmp11 = tmp10 * tmp10
tmp12 = tmp9 + tmp11
tmp13 = 4.0
tmp14 = tmp12 / tmp13
tmp15 = 1e-12
tmp16 = tmp14 + tmp15
tmp17 = libdevice.sqrt(tmp16)
tmp18 = tmp1 / tmp17
tmp19 = tmp0 * tmp18
tmp21 = tmp19 + tmp20
tl.store(out_ptr0 + x2, tmp21, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mean_sub_0[grid(256)](primals_1, buf0, 256, XBLOCK
=128, num_warps=4, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_pow_sqrt_1[grid(256)](primals_2,
buf0, primals_3, buf1, 256, XBLOCK=128, num_warps=4, num_stages=1)
del buf0
del primals_2
del primals_3
return buf1, primals_1
class LayerNormCustomNew(nn.Module):
"""A layernorm module in the TF style (epsilon inside the square root)."""
def __init__(self, n_hidden, variance_epsilon=1e-12):
super().__init__()
self.gamma = nn.Parameter(torch.ones(n_hidden))
self.beta = nn.Parameter(torch.zeros(n_hidden))
self.variance_epsilon = variance_epsilon
def forward(self, input_0):
primals_2 = self.gamma
primals_3 = self.beta
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| renebidart/pytorch-cifar | LayerNormCustom | false | 4,245 | [
"MIT"
] | 0 | 8f623299c25f7f219bab34bc7df41fe24232b1af | https://github.com/renebidart/pytorch-cifar/tree/8f623299c25f7f219bab34bc7df41fe24232b1af | import torch
import torch.nn as nn
class Model(nn.Module):
"""A layernorm module in the TF style (epsilon inside the square root)."""
def __init__(self, n_hidden, variance_epsilon=1e-12):
super().__init__()
self.gamma = nn.Parameter(torch.ones(n_hidden))
self.beta = nn.Parameter(torch.zeros(n_hidden))
self.variance_epsilon = variance_epsilon
def forward(self, x):
u = x.mean(1, keepdim=True)
s = (x - u).pow(2).mean(-1, keepdim=True)
x = (x - u) / torch.sqrt(s + self.variance_epsilon)
return self.gamma * x + self.beta
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
IBertLMHead | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/k2/ck2ensfw3jymkm3sdnn2b3sukex4hedkmwsqjeuwykarc22y3nse.py
# Topologically Sorted Source Nodes: [mul, pow_1, mul_1, add, mul_2, tanh, add_1, x_1, x_2], Original ATen: [aten.mul, aten.pow, aten.add, aten.tanh, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# pow_1 => pow_1
# tanh => tanh
# x_1 => mul_3
# x_2 => var_mean
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.5), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%view_1, 3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_1, 0.044715), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %mul_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, 0.7978845608028654), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%mul_2,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%tanh, 1), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add_1), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%mul_3, [3]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0 = async_compile.triton('triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp36 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = tmp0 * tmp0
tmp4 = tmp3 * tmp0
tmp5 = 0.044715
tmp6 = tmp4 * tmp5
tmp7 = tmp0 + tmp6
tmp8 = 0.7978845608028654
tmp9 = tmp7 * tmp8
tmp10 = libdevice.tanh(tmp9)
tmp11 = 1.0
tmp12 = tmp10 + tmp11
tmp13 = tmp2 * tmp12
tmp15 = tmp14 * tmp1
tmp16 = tmp14 * tmp14
tmp17 = tmp16 * tmp14
tmp18 = tmp17 * tmp5
tmp19 = tmp14 + tmp18
tmp20 = tmp19 * tmp8
tmp21 = libdevice.tanh(tmp20)
tmp22 = tmp21 + tmp11
tmp23 = tmp15 * tmp22
tmp24 = tmp13 + tmp23
tmp26 = tmp25 * tmp1
tmp27 = tmp25 * tmp25
tmp28 = tmp27 * tmp25
tmp29 = tmp28 * tmp5
tmp30 = tmp25 + tmp29
tmp31 = tmp30 * tmp8
tmp32 = libdevice.tanh(tmp31)
tmp33 = tmp32 + tmp11
tmp34 = tmp26 * tmp33
tmp35 = tmp24 + tmp34
tmp37 = tmp36 * tmp1
tmp38 = tmp36 * tmp36
tmp39 = tmp38 * tmp36
tmp40 = tmp39 * tmp5
tmp41 = tmp36 + tmp40
tmp42 = tmp41 * tmp8
tmp43 = libdevice.tanh(tmp42)
tmp44 = tmp43 + tmp11
tmp45 = tmp37 * tmp44
tmp46 = tmp35 + tmp45
tmp47 = 4.0
tmp48 = tmp46 / tmp47
tmp49 = tmp13 - tmp48
tmp50 = tmp49 * tmp49
tmp51 = tmp23 - tmp48
tmp52 = tmp51 * tmp51
tmp53 = tmp50 + tmp52
tmp54 = tmp34 - tmp48
tmp55 = tmp54 * tmp54
tmp56 = tmp53 + tmp55
tmp57 = tmp45 - tmp48
tmp58 = tmp57 * tmp57
tmp59 = tmp56 + tmp58
tmp60 = tmp59 / tmp47
tl.store(out_ptr0 + (x0), tmp48, xmask)
tl.store(out_ptr1 + (x0), tmp60, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/4h/c4hcwvzif5h6zqdmh45kywbrslvqedlkzfpdkfds26puiwek6kyk.py
# Topologically Sorted Source Nodes: [mul, pow_1, mul_1, add, mul_2, tanh, add_1, x_1, x_2], Original ATen: [aten.mul, aten.pow, aten.add, aten.tanh, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# pow_1 => pow_1
# tanh => tanh
# x_1 => mul_3
# x_2 => add_2, add_3, mul_4, mul_5, rsqrt, sub
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.5), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%view_1, 3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_1, 0.044715), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %mul_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, 0.7978845608028654), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%mul_2,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%tanh, 1), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add_1), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1.0), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_2,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_3, %getitem_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_4, %primals_4), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_5, %primals_5), kwargs = {})
triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1 = async_compile.triton('triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp14 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = tmp0 * tmp0
tmp4 = tmp3 * tmp0
tmp5 = 0.044715
tmp6 = tmp4 * tmp5
tmp7 = tmp0 + tmp6
tmp8 = 0.7978845608028654
tmp9 = tmp7 * tmp8
tmp10 = libdevice.tanh(tmp9)
tmp11 = 1.0
tmp12 = tmp10 + tmp11
tmp13 = tmp2 * tmp12
tmp15 = tmp13 - tmp14
tmp17 = tmp16 + tmp11
tmp18 = libdevice.rsqrt(tmp17)
tmp19 = tmp15 * tmp18
tmp21 = tmp19 * tmp20
tmp23 = tmp21 + tmp22
tl.store(out_ptr0 + (x2), tmp23, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [mul, pow_1, mul_1, add, mul_2, tanh, add_1, x_1, x_2], Original ATen: [aten.mul, aten.pow, aten.add, aten.tanh, aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0.run(buf0, buf1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, pow_1, mul_1, add, mul_2, tanh, add_1, x_1, x_2], Original ATen: [aten.mul, aten.pow, aten.add, aten.tanh, aten.native_layer_norm]
triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1.run(buf0, buf1, buf2, primals_4, primals_5, buf3, 256, grid=grid(256), stream=stream0)
del buf1
del buf2
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_4, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf0, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), primals_6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
import torch.utils.checkpoint
def gelu(x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 *
torch.pow(x, 3))))
class IBertLMHead(nn.Module):
"""I-BERT Head for masked language modeling."""
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.layer_norm = nn.LayerNorm(config.hidden_size, eps=config.
layer_norm_eps)
self.decoder = nn.Linear(config.hidden_size, config.vocab_size,
bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
self.decoder.bias = self.bias
def forward(self, features, **kwargs):
x = self.dense(features)
x = gelu(x)
x = self.layer_norm(x)
x = self.decoder(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, layer_norm_eps=1,
vocab_size=4)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import math
from torch import nn
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0(in_ptr0, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp36 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = tmp0 * tmp0
tmp4 = tmp3 * tmp0
tmp5 = 0.044715
tmp6 = tmp4 * tmp5
tmp7 = tmp0 + tmp6
tmp8 = 0.7978845608028654
tmp9 = tmp7 * tmp8
tmp10 = libdevice.tanh(tmp9)
tmp11 = 1.0
tmp12 = tmp10 + tmp11
tmp13 = tmp2 * tmp12
tmp15 = tmp14 * tmp1
tmp16 = tmp14 * tmp14
tmp17 = tmp16 * tmp14
tmp18 = tmp17 * tmp5
tmp19 = tmp14 + tmp18
tmp20 = tmp19 * tmp8
tmp21 = libdevice.tanh(tmp20)
tmp22 = tmp21 + tmp11
tmp23 = tmp15 * tmp22
tmp24 = tmp13 + tmp23
tmp26 = tmp25 * tmp1
tmp27 = tmp25 * tmp25
tmp28 = tmp27 * tmp25
tmp29 = tmp28 * tmp5
tmp30 = tmp25 + tmp29
tmp31 = tmp30 * tmp8
tmp32 = libdevice.tanh(tmp31)
tmp33 = tmp32 + tmp11
tmp34 = tmp26 * tmp33
tmp35 = tmp24 + tmp34
tmp37 = tmp36 * tmp1
tmp38 = tmp36 * tmp36
tmp39 = tmp38 * tmp36
tmp40 = tmp39 * tmp5
tmp41 = tmp36 + tmp40
tmp42 = tmp41 * tmp8
tmp43 = libdevice.tanh(tmp42)
tmp44 = tmp43 + tmp11
tmp45 = tmp37 * tmp44
tmp46 = tmp35 + tmp45
tmp47 = 4.0
tmp48 = tmp46 / tmp47
tmp49 = tmp13 - tmp48
tmp50 = tmp49 * tmp49
tmp51 = tmp23 - tmp48
tmp52 = tmp51 * tmp51
tmp53 = tmp50 + tmp52
tmp54 = tmp34 - tmp48
tmp55 = tmp54 * tmp54
tmp56 = tmp53 + tmp55
tmp57 = tmp45 - tmp48
tmp58 = tmp57 * tmp57
tmp59 = tmp56 + tmp58
tmp60 = tmp59 / tmp47
tl.store(out_ptr0 + x0, tmp48, xmask)
tl.store(out_ptr1 + x0, tmp60, xmask)
@triton.jit
def triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp14 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = tmp0 * tmp0
tmp4 = tmp3 * tmp0
tmp5 = 0.044715
tmp6 = tmp4 * tmp5
tmp7 = tmp0 + tmp6
tmp8 = 0.7978845608028654
tmp9 = tmp7 * tmp8
tmp10 = libdevice.tanh(tmp9)
tmp11 = 1.0
tmp12 = tmp10 + tmp11
tmp13 = tmp2 * tmp12
tmp15 = tmp13 - tmp14
tmp17 = tmp16 + tmp11
tmp18 = libdevice.rsqrt(tmp17)
tmp19 = tmp15 * tmp18
tmp21 = tmp19 * tmp20
tmp23 = tmp21 + tmp22
tl.store(out_ptr0 + x2, tmp23, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mul_native_layer_norm_pow_tanh_0[grid(64)](buf0,
buf1, buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_mul_native_layer_norm_pow_tanh_1[grid(256)](buf0,
buf1, buf2, primals_4, primals_5, buf3, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf1
del buf2
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf4)
del primals_7
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), primals_4, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf0, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), primals_6
def gelu(x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 *
torch.pow(x, 3))))
class IBertLMHeadNew(nn.Module):
"""I-BERT Head for masked language modeling."""
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.layer_norm = nn.LayerNorm(config.hidden_size, eps=config.
layer_norm_eps)
self.decoder = nn.Linear(config.hidden_size, config.vocab_size,
bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
self.decoder.bias = self.bias
def forward(self, input_0):
primals_2 = self.bias
primals_1 = self.dense.weight
primals_4 = self.dense.bias
primals_5 = self.layer_norm.weight
primals_7 = self.layer_norm.bias
primals_6 = self.decoder.weight
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| sajastu/transformers-sent-curr | IBertLMHead | false | 4,246 | [
"Apache-2.0"
] | 0 | 6dc41545c4ac298a010090fbca4b454c2eaf3dbb | https://github.com/sajastu/transformers-sent-curr/tree/6dc41545c4ac298a010090fbca4b454c2eaf3dbb | from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
import torch.utils.checkpoint
def gelu(x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 *
torch.pow(x, 3))))
class Model(nn.Module):
"""I-BERT Head for masked language modeling."""
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.layer_norm = nn.LayerNorm(config.hidden_size, eps=config.
layer_norm_eps)
self.decoder = nn.Linear(config.hidden_size, config.vocab_size,
bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
self.decoder.bias = self.bias
def forward(self, features, **kwargs):
x = self.dense(features)
x = gelu(x)
x = self.layer_norm(x)
x = self.decoder(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, layer_norm_eps=1,
vocab_size=4)}]
|
PatchSequential | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ts/ctsserxsekrwyqxewuhis2fi56hdzb4aa4jktxyotcweoqftjdww.py
# Topologically Sorted Source Nodes: [cat_1], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# cat_1 => cat_1
# Graph fragment:
# %cat_1 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%getitem_4, %getitem_5, %getitem_6, %getitem_7], -1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 16, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x5 = (xindex // 16)
x6 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = x1
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp7 & tmp4
tmp9 = tl.load(in_ptr0 + ((16*x2) + (64*x3) + ((16*x3) % 16)), tmp8 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tmp5 >= tmp3
tmp11 = tl.full([1], 2, tl.int64)
tmp12 = tmp5 < tmp11
tmp13 = tmp10 & tmp12
tmp14 = tmp13 & tmp4
tmp15 = tl.load(in_ptr0 + (4 + (16*x5)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp16 = tmp5 >= tmp11
tmp17 = tl.full([1], 3, tl.int64)
tmp18 = tmp5 < tmp17
tmp19 = tmp16 & tmp18
tmp20 = tmp19 & tmp4
tmp21 = tl.load(in_ptr0 + (8 + (16*x5)), tmp20 & xmask, eviction_policy='evict_last', other=0.0)
tmp22 = tmp5 >= tmp17
tmp23 = tl.full([1], 4, tl.int64)
tmp24 = tmp5 < tmp23
tmp25 = tmp22 & tmp4
tmp26 = tl.load(in_ptr0 + (12 + (16*x5)), tmp25 & xmask, eviction_policy='evict_last', other=0.0)
tmp27 = tl.where(tmp19, tmp21, tmp26)
tmp28 = tl.where(tmp13, tmp15, tmp27)
tmp29 = tl.where(tmp7, tmp9, tmp28)
tmp30 = tl.full(tmp29.shape, 0.0, tmp29.dtype)
tmp31 = tl.where(tmp4, tmp29, tmp30)
tmp32 = tmp0 >= tmp3
tmp33 = tmp0 < tmp11
tmp34 = tmp32 & tmp33
tmp35 = tmp7 & tmp34
tmp36 = tl.load(in_ptr0 + (1 + (16*x5)), tmp35 & xmask, eviction_policy='evict_last', other=0.0)
tmp37 = tmp13 & tmp34
tmp38 = tl.load(in_ptr0 + (5 + (16*x5)), tmp37 & xmask, eviction_policy='evict_last', other=0.0)
tmp39 = tmp19 & tmp34
tmp40 = tl.load(in_ptr0 + (9 + (16*x5)), tmp39 & xmask, eviction_policy='evict_last', other=0.0)
tmp41 = tmp22 & tmp34
tmp42 = tl.load(in_ptr0 + (13 + (16*x5)), tmp41 & xmask, eviction_policy='evict_last', other=0.0)
tmp43 = tl.where(tmp19, tmp40, tmp42)
tmp44 = tl.where(tmp13, tmp38, tmp43)
tmp45 = tl.where(tmp7, tmp36, tmp44)
tmp46 = tl.full(tmp45.shape, 0.0, tmp45.dtype)
tmp47 = tl.where(tmp34, tmp45, tmp46)
tmp48 = tmp0 >= tmp11
tmp49 = tmp0 < tmp17
tmp50 = tmp48 & tmp49
tmp51 = tmp7 & tmp50
tmp52 = tl.load(in_ptr0 + (2 + (16*x5)), tmp51 & xmask, eviction_policy='evict_last', other=0.0)
tmp53 = tmp13 & tmp50
tmp54 = tl.load(in_ptr0 + (6 + (16*x5)), tmp53 & xmask, eviction_policy='evict_last', other=0.0)
tmp55 = tmp19 & tmp50
tmp56 = tl.load(in_ptr0 + (10 + (16*x5)), tmp55 & xmask, eviction_policy='evict_last', other=0.0)
tmp57 = tmp22 & tmp50
tmp58 = tl.load(in_ptr0 + (14 + (16*x5)), tmp57 & xmask, eviction_policy='evict_last', other=0.0)
tmp59 = tl.where(tmp19, tmp56, tmp58)
tmp60 = tl.where(tmp13, tmp54, tmp59)
tmp61 = tl.where(tmp7, tmp52, tmp60)
tmp62 = tl.full(tmp61.shape, 0.0, tmp61.dtype)
tmp63 = tl.where(tmp50, tmp61, tmp62)
tmp64 = tmp0 >= tmp17
tmp65 = tmp0 < tmp23
tmp66 = tmp7 & tmp64
tmp67 = tl.load(in_ptr0 + (3 + (16*x5)), tmp66 & xmask, eviction_policy='evict_last', other=0.0)
tmp68 = tmp13 & tmp64
tmp69 = tl.load(in_ptr0 + (7 + (16*x5)), tmp68 & xmask, eviction_policy='evict_last', other=0.0)
tmp70 = tmp19 & tmp64
tmp71 = tl.load(in_ptr0 + (11 + (16*x5)), tmp70 & xmask, eviction_policy='evict_last', other=0.0)
tmp72 = tmp22 & tmp64
tmp73 = tl.load(in_ptr0 + (15 + (16*x5)), tmp72 & xmask, eviction_policy='evict_last', other=0.0)
tmp74 = tl.where(tmp19, tmp71, tmp73)
tmp75 = tl.where(tmp13, tmp69, tmp74)
tmp76 = tl.where(tmp7, tmp67, tmp75)
tmp77 = tl.full(tmp76.shape, 0.0, tmp76.dtype)
tmp78 = tl.where(tmp64, tmp76, tmp77)
tmp79 = tl.where(tmp50, tmp63, tmp78)
tmp80 = tl.where(tmp34, tmp47, tmp79)
tmp81 = tl.where(tmp4, tmp31, tmp80)
tl.store(out_ptr0 + (x6), tmp81, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 4, 4, 4), (64, 64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat_1], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import warnings
from typing import Dict
from typing import Optional
from typing import Tuple
import torch.nn as nn
import torch.nn.functional as F
from typing import cast
from typing import List
from typing import Union
from torch.distributions import Bernoulli
from itertools import zip_longest
from collections import OrderedDict
from typing import Any
from typing import Iterator
from typing import NamedTuple
from torch.nn.modules.utils import _pair
from math import pi
def _adapted_sampling(shape: 'Union[Tuple, torch.Size]', dist:
'torch.distributions.Distribution', same_on_batch=False) ->torch.Tensor:
"""The uniform sampling function that accepts 'same_on_batch'.
If same_on_batch is True, all values generated will be exactly same given a batch_size (shape[0]).
By default, same_on_batch is set to False.
"""
if same_on_batch:
return dist.sample((1, *shape[1:])).repeat(shape[0], *([1] * (len(
shape) - 1)))
return dist.sample(shape)
def _transform_output_shape(output:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', shape: 'Tuple'
) ->Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
"""Collapse the broadcasted batch dimensions an input tensor to be the specified shape.
Args:
input: torch.Tensor
shape: List/tuple of int
Returns:
torch.Tensor
"""
is_tuple = isinstance(output, tuple)
out_tensor: 'torch.Tensor'
trans_matrix: 'Optional[torch.Tensor]'
if is_tuple:
out_tensor, trans_matrix = cast(Tuple[torch.Tensor, torch.Tensor],
output)
else:
out_tensor = cast(torch.Tensor, output)
trans_matrix = None
if trans_matrix is not None:
if len(out_tensor.shape) > len(shape):
assert trans_matrix.shape[0
] == 1, f'Dimension 0 of transformation matrix is expected to be 1, got {trans_matrix.shape[0]}'
trans_matrix = trans_matrix.squeeze(0)
for dim in range(len(out_tensor.shape) - len(shape)):
assert out_tensor.shape[0
] == 1, f'Dimension {dim} of input is expected to be 1, got {out_tensor.shape[0]}'
out_tensor = out_tensor.squeeze(0)
return (out_tensor, trans_matrix) if is_tuple else out_tensor
def _transform_input(input: 'torch.Tensor') ->torch.Tensor:
"""Reshape an input tensor to be (*, C, H, W). Accept either (H, W), (C, H, W) or (*, C, H, W).
Args:
input: torch.Tensor
Returns:
torch.Tensor
"""
if not torch.is_tensor(input):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if len(input.shape) not in [2, 3, 4]:
raise ValueError(
f'Input size must have a shape of either (H, W), (C, H, W) or (*, C, H, W). Got {input.shape}'
)
if len(input.shape) == 2:
input = input.unsqueeze(0)
if len(input.shape) == 3:
input = input.unsqueeze(0)
return input
def _validate_input_dtype(input: 'torch.Tensor', accepted_dtypes: 'List'
) ->None:
"""Check if the dtype of the input tensor is in the range of accepted_dtypes
Args:
input: torch.Tensor
accepted_dtypes: List. e.g. [torch.float32, torch.float64]
"""
if input.dtype not in accepted_dtypes:
raise TypeError(
f'Expected input of {accepted_dtypes}. Got {input.dtype}')
def _extract_device_dtype(tensor_list: 'List[Optional[Any]]') ->Tuple[torch
.device, torch.dtype]:
"""Check if all the input are in the same device (only if when they are torch.Tensor).
If so, it would return a tuple of (device, dtype). Default: (cpu, ``get_default_dtype()``).
Returns:
[torch.device, torch.dtype]
"""
device, dtype = None, None
for tensor in tensor_list:
if tensor is not None:
if not isinstance(tensor, (torch.Tensor,)):
continue
_device = tensor.device
_dtype = tensor.dtype
if device is None and dtype is None:
device = _device
dtype = _dtype
elif device != _device or dtype != _dtype:
raise ValueError(
f'Passed values are not in the same device and dtype.Got ({device}, {dtype}) and ({_device}, {_dtype}).'
)
if device is None:
device = torch.device('cpu')
if dtype is None:
dtype = torch.get_default_dtype()
return device, dtype
def _joint_range_check(ranged_factor: 'torch.Tensor', name: 'str', bounds:
'Optional[Tuple[float, float]]'=None) ->None:
"""check if bounds[0] <= ranged_factor[0] <= ranged_factor[1] <= bounds[1]"""
if bounds is None:
bounds = float('-inf'), float('inf')
if ranged_factor.dim() == 1 and len(ranged_factor) == 2:
if not bounds[0] <= ranged_factor[0] or not bounds[1] >= ranged_factor[
1]:
raise ValueError(
f'{name} out of bounds. Expected inside {bounds}, got {ranged_factor}.'
)
if not bounds[0] <= ranged_factor[0] <= ranged_factor[1] <= bounds[1]:
raise ValueError(
f'{name}[0] should be smaller than {name}[1] got {ranged_factor}'
)
else:
raise TypeError(
f'{name} should be a tensor with length 2 whose values between {bounds}. Got {ranged_factor}.'
)
def _singular_range_check(ranged_factor: 'torch.Tensor', name: 'str',
bounds: 'Optional[Tuple[float, float]]'=None, skip_none: 'bool'=False,
mode: 'str'='2d') ->None:
"""check if bounds[0] <= ranged_factor[0] <= bounds[1] and bounds[0] <= ranged_factor[1] <= bounds[1]"""
if mode == '2d':
dim_size = 2
elif mode == '3d':
dim_size = 3
else:
raise ValueError(f"'mode' shall be either 2d or 3d. Got {mode}")
if skip_none and ranged_factor is None:
return
if bounds is None:
bounds = float('-inf'), float('inf')
if ranged_factor.dim() == 1 and len(ranged_factor) == dim_size:
for f in ranged_factor:
if not bounds[0] <= f <= bounds[1]:
raise ValueError(
f'{name} out of bounds. Expected inside {bounds}, got {ranged_factor}.'
)
else:
raise TypeError(
f'{name} should be a float number or a tuple with length {dim_size} whose values between {bounds}.Got {ranged_factor}'
)
def _range_bound(factor:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]', name:
'str', center: 'float'=0.0, bounds: 'Tuple[float, float]'=(0, float(
'inf')), check: 'Optional[str]'='joint', device: 'torch.device'=torch.
device('cpu'), dtype: 'torch.dtype'=torch.get_default_dtype()
) ->torch.Tensor:
"""Check inputs and compute the corresponding factor bounds"""
if not isinstance(factor, torch.Tensor):
factor = torch.tensor(factor, device=device, dtype=dtype)
factor_bound: 'torch.Tensor'
if factor.dim() == 0:
if factor < 0:
raise ValueError(
f'If {name} is a single number number, it must be non negative. Got {factor}'
)
factor_bound = factor.repeat(2) * torch.tensor([-1.0, 1.0], device=
factor.device, dtype=factor.dtype) + center
factor_bound = factor_bound.clamp(bounds[0], bounds[1])
else:
factor_bound = torch.as_tensor(factor, device=device, dtype=dtype)
if check is not None:
if check == 'joint':
_joint_range_check(factor_bound, name, bounds)
elif check == 'singular':
_singular_range_check(factor_bound, name, bounds)
else:
raise NotImplementedError(f"methods '{check}' not implemented.")
return factor_bound
def adjust_brightness(input: 'torch.Tensor', brightness_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust Brightness of an image.
.. image:: _static/img/adjust_brightness.png
This implementation aligns OpenCV, not PIL. Hence, the output differs from TorchVision.
The input image is expected to be in the range of [0, 1].
Args:
input: image to be adjusted in the shape of :math:`(*, N)`.
brightness_factor: Brightness adjust factor per element
in the batch. 0 does not modify the input image while any other number modify the
brightness.
Return:
Adjusted image in the shape of :math:`(*, N)`.
Example:
>>> x = torch.ones(1, 1, 2, 2)
>>> adjust_brightness(x, 1.)
tensor([[[[1., 1.],
[1., 1.]]]])
>>> x = torch.ones(2, 5, 3, 3)
>>> y = torch.tensor([0.25, 0.50])
>>> adjust_brightness(x, y).shape
torch.Size([2, 5, 3, 3])
"""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(brightness_factor, (float, torch.Tensor)):
raise TypeError(
f'The factor should be either a float or torch.Tensor. Got {type(brightness_factor)}'
)
if isinstance(brightness_factor, float):
brightness_factor = torch.tensor([brightness_factor])
brightness_factor = brightness_factor.to(input.device)
for _ in input.shape[1:]:
brightness_factor = torch.unsqueeze(brightness_factor, dim=-1)
x_adjust: 'torch.Tensor' = input + brightness_factor
out: 'torch.Tensor' = torch.clamp(x_adjust, 0.0, 1.0)
return out
def adjust_contrast(input: 'torch.Tensor', contrast_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust Contrast of an image.
.. image:: _static/img/adjust_contrast.png
This implementation aligns OpenCV, not PIL. Hence, the output differs from TorchVision.
The input image is expected to be in the range of [0, 1].
Args:
input: Image to be adjusted in the shape of :math:`(*, N)`.
contrast_factor: Contrast adjust factor per element
in the batch. 0 generates a completely black image, 1 does not modify
the input image while any other non-negative number modify the
brightness by this factor.
Return:
Adjusted image in the shape of :math:`(*, N)`.
Example:
>>> x = torch.ones(1, 1, 2, 2)
>>> adjust_contrast(x, 0.5)
tensor([[[[0.5000, 0.5000],
[0.5000, 0.5000]]]])
>>> x = torch.ones(2, 5, 3, 3)
>>> y = torch.tensor([0.65, 0.50])
>>> adjust_contrast(x, y).shape
torch.Size([2, 5, 3, 3])
"""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(contrast_factor, (float, torch.Tensor)):
raise TypeError(
f'The factor should be either a float or torch.Tensor. Got {type(contrast_factor)}'
)
if isinstance(contrast_factor, float):
contrast_factor = torch.tensor([contrast_factor])
contrast_factor = contrast_factor.to(input.device)
if (contrast_factor < 0).any():
raise ValueError(
f'Contrast factor must be non-negative. Got {contrast_factor}')
for _ in input.shape[1:]:
contrast_factor = torch.unsqueeze(contrast_factor, dim=-1)
x_adjust: 'torch.Tensor' = input * contrast_factor
out: 'torch.Tensor' = torch.clamp(x_adjust, 0.0, 1.0)
return out
def adjust_hue_raw(input: 'torch.Tensor', hue_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust hue of an image. Expecting input to be in hsv format already."""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(hue_factor, (float, torch.Tensor)):
raise TypeError(
f'The hue_factor should be a float number or torch.Tensor in the range between [-PI, PI]. Got {type(hue_factor)}'
)
if isinstance(hue_factor, float):
hue_factor = torch.as_tensor(hue_factor)
hue_factor = hue_factor
for _ in input.shape[1:]:
hue_factor = torch.unsqueeze(hue_factor, dim=-1)
h, s, v = torch.chunk(input, chunks=3, dim=-3)
divisor: 'float' = 2 * pi
h_out: 'torch.Tensor' = torch.fmod(h + hue_factor, divisor)
out: 'torch.Tensor' = torch.cat([h_out, s, v], dim=-3)
return out
def hsv_to_rgb(image: 'torch.Tensor') ->torch.Tensor:
"""Convert an image from HSV to RGB.
The H channel values are assumed to be in the range 0..2pi. S and V are in the range 0..1.
Args:
image: HSV Image to be converted to HSV with shape of :math:`(*, 3, H, W)`.
Returns:
RGB version of the image with shape of :math:`(*, 3, H, W)`.
Example:
>>> input = torch.rand(2, 3, 4, 5)
>>> output = hsv_to_rgb(input) # 2x3x4x5
"""
if not isinstance(image, torch.Tensor):
raise TypeError('Input type is not a torch.Tensor. Got {}'.format(
type(image)))
if len(image.shape) < 3 or image.shape[-3] != 3:
raise ValueError('Input size must have a shape of (*, 3, H, W). Got {}'
.format(image.shape))
h: 'torch.Tensor' = image[..., 0, :, :] / (2 * math.pi)
s: 'torch.Tensor' = image[..., 1, :, :]
v: 'torch.Tensor' = image[..., 2, :, :]
hi: 'torch.Tensor' = torch.floor(h * 6) % 6
f: 'torch.Tensor' = h * 6 % 6 - hi
one: 'torch.Tensor' = torch.tensor(1.0).to(image.device)
p: 'torch.Tensor' = v * (one - s)
q: 'torch.Tensor' = v * (one - f * s)
t: 'torch.Tensor' = v * (one - (one - f) * s)
hi = hi.long()
indices: 'torch.Tensor' = torch.stack([hi, hi + 6, hi + 12], dim=-3)
out = torch.stack((v, q, p, p, t, v, t, v, v, q, p, p, p, p, t, v, v, q
), dim=-3)
out = torch.gather(out, -3, indices)
return out
def rgb_to_hsv(image: 'torch.Tensor', eps: 'float'=1e-06) ->torch.Tensor:
"""Convert an image from RGB to HSV.
.. image:: _static/img/rgb_to_hsv.png
The image data is assumed to be in the range of (0, 1).
Args:
image: RGB Image to be converted to HSV with shape of :math:`(*, 3, H, W)`.
eps: scalar to enforce numarical stability.
Returns:
HSV version of the image with shape of :math:`(*, 3, H, W)`.
The H channel values are in the range 0..2pi. S and V are in the range 0..1.
Example:
>>> input = torch.rand(2, 3, 4, 5)
>>> output = rgb_to_hsv(input) # 2x3x4x5
"""
if not isinstance(image, torch.Tensor):
raise TypeError('Input type is not a torch.Tensor. Got {}'.format(
type(image)))
if len(image.shape) < 3 or image.shape[-3] != 3:
raise ValueError('Input size must have a shape of (*, 3, H, W). Got {}'
.format(image.shape))
maxc, _ = image.max(-3)
maxc_mask = image == maxc.unsqueeze(-3)
_, max_indices = ((maxc_mask.cumsum(-3) == 1) & maxc_mask).max(-3)
minc: 'torch.Tensor' = image.min(-3)[0]
v: 'torch.Tensor' = maxc
deltac: 'torch.Tensor' = maxc - minc
s: 'torch.Tensor' = deltac / (v + eps)
deltac = torch.where(deltac == 0, torch.ones_like(deltac, device=deltac
.device, dtype=deltac.dtype), deltac)
maxc_tmp = maxc.unsqueeze(-3) - image
rc: 'torch.Tensor' = maxc_tmp[..., 0, :, :]
gc: 'torch.Tensor' = maxc_tmp[..., 1, :, :]
bc: 'torch.Tensor' = maxc_tmp[..., 2, :, :]
h = torch.stack([bc - gc, 2.0 * deltac + rc - bc, 4.0 * deltac + gc -
rc], dim=-3)
h = torch.gather(h, dim=-3, index=max_indices[..., None, :, :])
h = h.squeeze(-3)
h = h / deltac
h = h / 6.0 % 1.0
h = 2 * math.pi * h
return torch.stack([h, s, v], dim=-3)
def adjust_hue(input: 'torch.Tensor', hue_factor: 'Union[float, torch.Tensor]'
) ->torch.Tensor:
"""Adjust hue of an image.
.. image:: _static/img/adjust_hue.png
The input image is expected to be an RGB image in the range of [0, 1].
Args:
input: Image to be adjusted in the shape of :math:`(*, 3, H, W)`.
hue_factor: How much to shift the hue channel. Should be in [-PI, PI]. PI
and -PI give complete reversal of hue channel in HSV space in positive and negative
direction respectively. 0 means no shift. Therefore, both -PI and PI will give an
image with complementary colors while 0 gives the original image.
Return:
Adjusted image in the shape of :math:`(*, 3, H, W)`.
Example:
>>> x = torch.ones(1, 3, 2, 2)
>>> adjust_hue(x, 3.141516).shape
torch.Size([1, 3, 2, 2])
>>> x = torch.ones(2, 3, 3, 3)
>>> y = torch.ones(2) * 3.141516
>>> adjust_hue(x, y).shape
torch.Size([2, 3, 3, 3])
"""
x_hsv: 'torch.Tensor' = rgb_to_hsv(input)
x_adjusted: 'torch.Tensor' = adjust_hue_raw(x_hsv, hue_factor)
out: 'torch.Tensor' = hsv_to_rgb(x_adjusted)
return out
def adjust_saturation_raw(input: 'torch.Tensor', saturation_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust color saturation of an image. Expecting input to be in hsv format already."""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(saturation_factor, (float, torch.Tensor)):
raise TypeError(
f'The saturation_factor should be a float number or torch.Tensor.Got {type(saturation_factor)}'
)
if isinstance(saturation_factor, float):
saturation_factor = torch.as_tensor(saturation_factor)
saturation_factor = saturation_factor.to(input.device)
for _ in input.shape[1:]:
saturation_factor = torch.unsqueeze(saturation_factor, dim=-1)
h, s, v = torch.chunk(input, chunks=3, dim=-3)
s_out: 'torch.Tensor' = torch.clamp(s * saturation_factor, min=0, max=1)
out: 'torch.Tensor' = torch.cat([h, s_out, v], dim=-3)
return out
def adjust_saturation(input: 'torch.Tensor', saturation_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust color saturation of an image.
.. image:: _static/img/adjust_saturation.png
The input image is expected to be an RGB image in the range of [0, 1].
Args:
input: Image/Tensor to be adjusted in the shape of :math:`(*, 3, H, W)`.
saturation_factor: How much to adjust the saturation. 0 will give a black
and white image, 1 will give the original image while 2 will enhance the saturation by a factor of 2.
Return:
Adjusted image in the shape of :math:`(*, 3, H, W)`.
Example:
>>> x = torch.ones(1, 3, 3, 3)
>>> adjust_saturation(x, 2.).shape
torch.Size([1, 3, 3, 3])
>>> x = torch.ones(2, 3, 3, 3)
>>> y = torch.tensor([1., 2.])
>>> adjust_saturation(x, y).shape
torch.Size([2, 3, 3, 3])
"""
x_hsv: 'torch.Tensor' = rgb_to_hsv(input)
x_adjusted: 'torch.Tensor' = adjust_saturation_raw(x_hsv, saturation_factor
)
out: 'torch.Tensor' = hsv_to_rgb(x_adjusted)
return out
def _extract_tensor_patchesnd(input: 'torch.Tensor', window_sizes:
'Tuple[int, ...]', strides: 'Tuple[int, ...]') ->torch.Tensor:
batch_size, num_channels = input.size()[:2]
dims = range(2, input.dim())
for dim, patch_size, stride in zip(dims, window_sizes, strides):
input = input.unfold(dim, patch_size, stride)
input = input.permute(0, *dims, 1, *[(dim + len(dims)) for dim in dims]
).contiguous()
return input.view(batch_size, -1, num_channels, *window_sizes)
def extract_tensor_patches(input: 'torch.Tensor', window_size:
'Union[int, Tuple[int, int]]', stride: 'Union[int, Tuple[int, int]]'=1,
padding: 'Union[int, Tuple[int, int]]'=0) ->torch.Tensor:
"""Function that extract patches from tensors and stack them.
See :class:`~kornia.contrib.ExtractTensorPatches` for details.
"""
if not torch.is_tensor(input):
raise TypeError('Input input type is not a torch.Tensor. Got {}'.
format(type(input)))
if not len(input.shape) == 4:
raise ValueError('Invalid input shape, we expect BxCxHxW. Got: {}'.
format(input.shape))
if padding:
pad_vert, pad_horz = _pair(padding)
input = F.pad(input, [pad_horz, pad_horz, pad_vert, pad_vert])
return _extract_tensor_patchesnd(input, _pair(window_size), _pair(stride))
class _BasicAugmentationBase(nn.Module):
"""_BasicAugmentationBase base class for customized augmentation implementations.
Plain augmentation base class without the functionality of transformation matrix calculations.
By default, the random computations will be happened on CPU with ``torch.get_default_dtype()``.
To change this behaviour, please use ``set_rng_device_and_dtype``.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation
probabilities element-wisely.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def __init__(self, p: 'float'=0.5, p_batch: 'float'=1.0, same_on_batch:
'bool'=False, keepdim: 'bool'=False) ->None:
super(_BasicAugmentationBase, self).__init__()
self.p = p
self.p_batch = p_batch
self.same_on_batch = same_on_batch
self.keepdim = keepdim
self._params: 'Dict[str, torch.Tensor]' = {}
if p != 0.0 or p != 1.0:
self._p_gen = Bernoulli(self.p)
if p_batch != 0.0 or p_batch != 1.0:
self._p_batch_gen = Bernoulli(self.p_batch)
self.set_rng_device_and_dtype(torch.device('cpu'), torch.
get_default_dtype())
def __repr__(self) ->str:
return (
f'p={self.p}, p_batch={self.p_batch}, same_on_batch={self.same_on_batch}'
)
def __unpack_input__(self, input: 'torch.Tensor') ->torch.Tensor:
return input
def __check_batching__(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]'):
"""Check if a transformation matrix is returned,
it has to be in the same batching mode as output."""
raise NotImplementedError
def transform_tensor(self, input: 'torch.Tensor') ->torch.Tensor:
"""Standardize input tensors."""
raise NotImplementedError
def generate_parameters(self, batch_shape: 'torch.Size') ->Dict[str,
torch.Tensor]:
return {}
def apply_transform(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->torch.Tensor:
raise NotImplementedError
def set_rng_device_and_dtype(self, device: 'torch.device', dtype:
'torch.dtype') ->None:
"""Change the random generation device and dtype.
Note:
The generated random numbers are not reproducible across different devices and dtypes.
"""
self.device = device
self.dtype = dtype
def __batch_prob_generator__(self, batch_shape: 'torch.Size', p:
'float', p_batch: 'float', same_on_batch: 'bool') ->torch.Tensor:
batch_prob: 'torch.Tensor'
if p_batch == 1:
batch_prob = torch.tensor([True])
elif p_batch == 0:
batch_prob = torch.tensor([False])
else:
batch_prob = _adapted_sampling((1,), self._p_batch_gen,
same_on_batch).bool()
if batch_prob.sum().item() == 1:
elem_prob: 'torch.Tensor'
if p == 1:
elem_prob = torch.tensor([True] * batch_shape[0])
elif p == 0:
elem_prob = torch.tensor([False] * batch_shape[0])
else:
elem_prob = _adapted_sampling((batch_shape[0],), self.
_p_gen, same_on_batch).bool()
batch_prob = batch_prob * elem_prob
else:
batch_prob = batch_prob.repeat(batch_shape[0])
return batch_prob
def forward_parameters(self, batch_shape):
to_apply = self.__batch_prob_generator__(batch_shape, self.p, self.
p_batch, self.same_on_batch)
_params = self.generate_parameters(torch.Size((int(to_apply.sum().
item()), *batch_shape[1:])))
if _params is None:
_params = {}
_params['batch_prob'] = to_apply
return _params
def apply_func(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->Union[torch.Tensor, Tuple[torch.Tensor,
torch.Tensor]]:
input = self.transform_tensor(input)
return self.apply_transform(input, params)
def forward(self, input: 'torch.Tensor', params:
'Optional[Dict[str, torch.Tensor]]'=None) ->Union[torch.Tensor,
Tuple[torch.Tensor, torch.Tensor]]:
in_tensor = self.__unpack_input__(input)
self.__check_batching__(input)
ori_shape = in_tensor.shape
in_tensor = self.transform_tensor(in_tensor)
batch_shape = in_tensor.shape
if params is None:
params = self.forward_parameters(batch_shape)
self._params = params
output = self.apply_func(input, self._params)
return _transform_output_shape(output, ori_shape
) if self.keepdim else output
class _AugmentationBase(_BasicAugmentationBase):
"""_AugmentationBase base class for customized augmentation implementations.
Advanced augmentation base class with the functionality of transformation matrix calculations.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation probabilities
element-wisely for a batch.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
return_transform (bool): if ``True`` return the matrix describing the geometric transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def __init__(self, return_transform: 'bool'=None, same_on_batch: 'bool'
=False, p: 'float'=0.5, p_batch: 'float'=1.0, keepdim: 'bool'=False
) ->None:
super(_AugmentationBase, self).__init__(p, p_batch=p_batch,
same_on_batch=same_on_batch, keepdim=keepdim)
self.p = p
self.p_batch = p_batch
self.return_transform = return_transform
def __repr__(self) ->str:
return super().__repr__(
) + f', return_transform={self.return_transform}'
def identity_matrix(self, input: 'torch.Tensor') ->torch.Tensor:
raise NotImplementedError
def compute_transformation(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->torch.Tensor:
raise NotImplementedError
def apply_transform(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]', transform: 'Optional[torch.Tensor]'=None
) ->torch.Tensor:
raise NotImplementedError
def __unpack_input__(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]') ->Tuple[
torch.Tensor, Optional[torch.Tensor]]:
if isinstance(input, tuple):
in_tensor = input[0]
in_transformation = input[1]
return in_tensor, in_transformation
in_tensor = input
return in_tensor, None
def apply_func(self, in_tensor: 'torch.Tensor', in_transform:
'Optional[torch.Tensor]', params: 'Dict[str, torch.Tensor]',
return_transform: 'bool'=False) ->Union[torch.Tensor, Tuple[torch.
Tensor, torch.Tensor]]:
to_apply = params['batch_prob']
if torch.sum(to_apply) == 0:
output = in_tensor
trans_matrix = self.identity_matrix(in_tensor)
elif torch.sum(to_apply) == len(to_apply):
trans_matrix = self.compute_transformation(in_tensor, params)
output = self.apply_transform(in_tensor, params, trans_matrix)
else:
output = in_tensor.clone()
trans_matrix = self.identity_matrix(in_tensor)
trans_matrix[to_apply] = self.compute_transformation(in_tensor[
to_apply], params)
output[to_apply] = self.apply_transform(in_tensor[to_apply],
params, trans_matrix[to_apply])
self._transform_matrix = trans_matrix
if return_transform:
out_transformation = (trans_matrix if in_transform is None else
trans_matrix @ in_transform)
return output, out_transformation
if in_transform is not None:
return output, in_transform
return output
def forward(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', params:
'Optional[Dict[str, torch.Tensor]]'=None, return_transform:
'Optional[bool]'=None) ->Union[torch.Tensor, Tuple[torch.Tensor,
torch.Tensor]]:
in_tensor, in_transform = self.__unpack_input__(input)
self.__check_batching__(input)
ori_shape = in_tensor.shape
in_tensor = self.transform_tensor(in_tensor)
batch_shape = in_tensor.shape
if return_transform is None:
return_transform = self.return_transform
return_transform = cast(bool, return_transform)
if params is None:
params = self.forward_parameters(batch_shape)
if 'batch_prob' not in params:
params['batch_prob'] = torch.tensor([True] * batch_shape[0])
warnings.warn(
'`batch_prob` is not found in params. Will assume applying on all data.'
)
self._params = params
output = self.apply_func(in_tensor, in_transform, self._params,
return_transform)
return _transform_output_shape(output, ori_shape
) if self.keepdim else output
class AugmentationBase2D(_AugmentationBase):
"""AugmentationBase2D base class for customized augmentation implementations.
For any augmentation, the implementation of "generate_parameters" and "apply_transform" are required while the
"compute_transformation" is only required when passing "return_transform" as True.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation probabilities
element-wisely for a batch.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
return_transform (bool): if ``True`` return the matrix describing the geometric transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def __check_batching__(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]'):
if isinstance(input, tuple):
inp, mat = input
if len(inp.shape) == 4:
assert len(mat.shape
) == 3, 'Input tensor is in batch mode but transformation matrix is not'
assert mat.shape[0] == inp.shape[0
], f'In batch dimension, input has {inp.shape[0]}but transformation matrix has {mat.shape[0]}'
elif len(inp.shape) == 3 or len(inp.shape) == 2:
assert len(mat.shape
) == 2, 'Input tensor is in non-batch mode but transformation matrix is not'
else:
raise ValueError(
f'Unrecognized output shape. Expected 2, 3, or 4, got {len(inp.shape)}'
)
def transform_tensor(self, input: 'torch.Tensor') ->torch.Tensor:
"""Convert any incoming (H, W), (C, H, W) and (B, C, H, W) into (B, C, H, W)."""
_validate_input_dtype(input, accepted_dtypes=[torch.float16, torch.
float32, torch.float64])
return _transform_input(input)
def identity_matrix(self, input) ->torch.Tensor:
"""Return 3x3 identity matrix."""
return kornia.eye_like(3, input)
class IntensityAugmentationBase2D(AugmentationBase2D):
"""IntensityAugmentationBase2D base class for customized intensity augmentation implementations.
For any augmentation, the implementation of "generate_parameters" and "apply_transform" are required while the
"compute_transformation" is only required when passing "return_transform" as True.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation probabilities
element-wisely for a batch.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
return_transform (bool): if ``True`` return the matrix describing the geometric transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def compute_transformation(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->torch.Tensor:
return self.identity_matrix(input)
class ParamItem(NamedTuple):
name: 'str'
data: 'Union[dict, list]'
class ImageSequential(nn.Sequential):
"""Sequential for creating kornia image processing pipeline.
Args:
*args : a list of kornia augmentation and image operation modules.
same_on_batch: apply the same transformation across the batch.
If None, it will not overwrite the function-wise settings.
return_transform: if ``True`` return the matrix describing the transformation
applied to each. If None, it will not overwrite the function-wise settings.
keepdim: whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). If None, it will not overwrite the function-wise settings.
random_apply: randomly select a sublist (order agnostic) of args to
apply transformation.
If int, a fixed number of transformations will be selected.
If (a,), x number of transformations (a <= x <= len(args)) will be selected.
If (a, b), x number of transformations (a <= x <= b) will be selected.
If True, the whole list of args will be processed as a sequence in a random order.
If False, the whole list of args will be processed as a sequence in original order.
Returns:
Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]: the tensor (, and the transformation matrix)
has been sequentially modified by the args.
Examples:
>>> import kornia
>>> input = torch.randn(2, 3, 5, 6)
>>> aug_list = ImageSequential(
... kornia.color.BgrToRgb(),
... kornia.augmentation.ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.0),
... kornia.filters.MedianBlur((3, 3)),
... kornia.augmentation.RandomAffine(360, p=1.0),
... kornia.enhance.Invert(),
... return_transform=True,
... same_on_batch=True,
... random_apply=10,
... )
>>> out = aug_list(input)
>>> out[0].shape, out[1].shape
(torch.Size([2, 3, 5, 6]), torch.Size([2, 3, 3]))
Reproduce with provided params.
>>> out2 = aug_list(input, params=aug_list._params)
>>> torch.equal(out[0], out2[0]), torch.equal(out[1], out2[1])
(True, True)
Note:
Transformation matrix returned only considers the transformation applied in ``kornia.augmentation`` module.
Those transformations in ``kornia.geometry`` will not be taken into account.
"""
def __init__(self, *args: nn.Module, same_on_batch: Optional[bool]=None,
return_transform: Optional[bool]=None, keepdim: Optional[bool]=None,
random_apply: Union[int, bool, Tuple[int, int]]=False) ->None:
self.same_on_batch = same_on_batch
self.return_transform = return_transform
self.keepdim = keepdim
_args = OrderedDict()
for idx, arg in enumerate(args):
if not isinstance(arg, nn.Module):
raise NotImplementedError(
f'Only nn.Module are supported at this moment. Got {arg}.')
if isinstance(arg, _AugmentationBase):
if same_on_batch is not None:
arg.same_on_batch = same_on_batch
if return_transform is not None:
arg.return_transform = return_transform
if keepdim is not None:
arg.keepdim = keepdim
_args.update({f'{arg.__class__.__name__}_{idx}': arg})
super(ImageSequential, self).__init__(_args)
self._params: 'List[Any]' = []
self.random_apply: 'Union[Tuple[int, int], bool]'
if random_apply:
if isinstance(random_apply, (bool,)) and random_apply is True:
self.random_apply = len(args), len(args) + 1
elif isinstance(random_apply, (int,)):
self.random_apply = random_apply, random_apply + 1
elif isinstance(random_apply, (tuple,)) and len(random_apply
) == 2 and isinstance(random_apply[0], (int,)) and isinstance(
random_apply[1], (int,)):
self.random_apply = random_apply[0], random_apply[1] + 1
elif isinstance(random_apply, (tuple,)) and len(random_apply
) == 1 and isinstance(random_apply[0], (int,)):
self.random_apply = random_apply[0], len(args) + 1
else:
raise ValueError(
f'Non-readable random_apply. Got {random_apply}.')
assert isinstance(self.random_apply, (tuple,)) and len(self.
random_apply) == 2 and isinstance(self.random_apply[0], (int,)
) and isinstance(self.random_apply[0], (int,)
), f'Expect a tuple of (int, int). Got {self.random_apply}.'
else:
self.random_apply = False
def _get_child_sequence(self) ->Iterator[Tuple[str, nn.Module]]:
if self.random_apply:
num_samples = int(torch.randint(*self.random_apply, (1,)).item())
indices = torch.multinomial(torch.ones((len(self),)),
num_samples, replacement=True if num_samples > len(self) else
False)
return self._get_children_by_indices(indices)
return self.named_children()
def _get_children_by_indices(self, indices: 'torch.Tensor') ->Iterator[
Tuple[str, nn.Module]]:
modules = list(self.named_children())
for idx in indices:
yield modules[idx]
def _get_children_by_module_names(self, names: 'List[str]') ->Iterator[
Tuple[str, nn.Module]]:
modules = list(self.named_children())
for name in names:
yield modules[list(dict(self.named_children()).keys()).index(name)]
def get_forward_sequence(self, params: 'Optional[List[ParamItem]]'=None
) ->Iterator[Tuple[str, nn.Module]]:
if params is None:
named_modules = self._get_child_sequence()
else:
named_modules = self._get_children_by_module_names([p.name for
p in params])
return named_modules
def apply_to_input(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]',
module_name: 'str', module: 'Optional[nn.Module]'=None, param:
'Optional[ParamItem]'=None) ->Union[torch.Tensor, Tuple[torch.
Tensor, torch.Tensor]]:
if module is None:
module = self.get_submodule(module_name)
if param is not None:
assert module_name == param.name
_param = param.data
else:
_param = None
if isinstance(module, (_AugmentationBase, ImageSequential)
) and _param is None:
input = module(input)
self._params.append(ParamItem(module_name, module._params))
elif isinstance(module, (_AugmentationBase, ImageSequential)
) and _param is not None:
input = module(input, params=_param)
self._params.append(ParamItem(module_name, _param))
else:
assert _param == {
} or _param is None, f'Non-augmentaion operation {module_name} require empty parameters. Got {module}.'
if isinstance(input, (tuple, list)):
input = module(input[0]), input[1]
else:
input = module(input)
self._params.append(ParamItem(module_name, {}))
return input
def forward(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', params:
'Optional[List[ParamItem]]'=None) ->Union[torch.Tensor, Tuple[torch
.Tensor, torch.Tensor]]:
self._params = []
named_modules = self.get_forward_sequence(params)
params = [] if params is None else params
for (name, module), param in zip_longest(named_modules, params):
input = self.apply_to_input(input, name, module, param=param)
return input
class ColorJitter(IntensityAugmentationBase2D):
"""Applies a random transformation to the brightness, contrast, saturation and hue of a tensor image.
.. image:: _static/img/ColorJitter.png
Args:
p: probability of applying the transformation.
brightness: The brightness factor to apply.
contrast: The contrast factor to apply.
saturation: The saturation factor to apply.
hue: The hue factor to apply.
return_transform: if ``True`` return the matrix describing the transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch: apply the same transformation across the batch.
keepdim: whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False).
Shape:
- Input: :math:`(C, H, W)` or :math:`(B, C, H, W)`, Optional: :math:`(B, 3, 3)`
- Output: :math:`(B, C, H, W)`
Note:
Input tensor must be float and normalized into [0, 1] for the best differentiability support.
Additionally, this function accepts another transformation tensor (:math:`(B, 3, 3)`), then the
applied transformation will be merged int to the input transformation tensor and returned.
Examples:
>>> rng = torch.manual_seed(0)
>>> inputs = torch.ones(1, 3, 3, 3)
>>> aug = ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.)
>>> aug(inputs)
tensor([[[[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993]],
<BLANKLINE>
[[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993]],
<BLANKLINE>
[[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993]]]])
"""
def __init__(self, brightness:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]'=0.0,
contrast:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]'=0.0,
saturation:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]'=0.0,
hue: 'Union[torch.Tensor, float, Tuple[float, float], List[float]]'
=0.0, return_transform: 'bool'=False, same_on_batch: 'bool'=False,
p: 'float'=1.0, keepdim: 'bool'=False) ->None:
super(ColorJitter, self).__init__(p=p, return_transform=
return_transform, same_on_batch=same_on_batch, keepdim=keepdim)
self._device, self._dtype = _extract_device_dtype([brightness,
contrast, hue, saturation])
self.brightness = brightness
self.contrast = contrast
self.saturation = saturation
self.hue = hue
def __repr__(self) ->str:
repr = (
f'brightness={self.brightness}, contrast={self.contrast}, saturation={self.saturation}, hue={self.hue}'
)
return self.__class__.__name__ + f'({repr}, {super().__repr__()})'
def generate_parameters(self, batch_shape: 'torch.Size') ->Dict[str,
torch.Tensor]:
brightness: 'torch.Tensor' = _range_bound(self.brightness,
'brightness', center=1.0, bounds=(0, 2), device=self._device,
dtype=self._dtype)
contrast: 'torch.Tensor' = _range_bound(self.contrast, 'contrast',
center=1.0, device=self._device, dtype=self._dtype)
saturation: 'torch.Tensor' = _range_bound(self.saturation,
'saturation', center=1.0, device=self._device, dtype=self._dtype)
hue: 'torch.Tensor' = _range_bound(self.hue, 'hue', bounds=(-0.5,
0.5), device=self._device, dtype=self._dtype)
return rg.random_color_jitter_generator(batch_shape[0], brightness,
contrast, saturation, hue, self.same_on_batch, self.device,
self.dtype)
def apply_transform(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]', transform: 'Optional[torch.Tensor]'=None
) ->torch.Tensor:
transforms = [lambda img: adjust_brightness(img, params[
'brightness_factor'] - 1), lambda img: adjust_contrast(img,
params['contrast_factor']), lambda img: adjust_saturation(img,
params['saturation_factor']), lambda img: adjust_hue(img,
params['hue_factor'] * 2 * pi)]
jittered = input
for idx in params['order'].tolist():
t = transforms[idx]
jittered = t(jittered)
return jittered
class PatchSequential(ImageSequential):
"""Container for performing patch-level image processing.
.. image:: https://kornia-tutorials.readthedocs.io/en/latest/_images/data_patch_sequential_5_1.png
PatchSequential breaks input images into patches by a given grid size, which will be resembled back
afterwards. Different image processing and augmentation methods will be performed on each patch region.
Args:
*args: a list of processing modules.
grid_size: controls the grid board seperation.
padding: same or valid padding. If same padding, it will pad to include all pixels if the input
tensor cannot be divisible by grid_size. If valid padding, the redundent border will be removed.
same_on_batch: apply the same transformation across the batch.
If None, it will not overwrite the function-wise settings.
keepdim: whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). If None, it will not overwrite the function-wise settings.
patchwise_apply: apply image processing args will be applied patch-wisely.
if ``True``, the number of args must be equal to grid number.
if ``False``, the image processing args will be applied as a sequence to all patches. Default: False.
random_apply: randomly select a sublist (order agnostic) of args to
apply transformation.
If ``int`` (batchwise mode only), a fixed number of transformations will be selected.
If ``(a,)`` (batchwise mode only), x number of transformations (a <= x <= len(args)) will be selected.
If ``(a, b)`` (batchwise mode only), x number of transformations (a <= x <= b) will be selected.
If ``True``, the whole list of args will be processed in a random order.
If ``False``, the whole list of args will be processed in original order.
Return:
List[Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]]: the tensor (, and the transformation matrix)
has been sequentially modified by the args.
Examples:
>>> import kornia.augmentation as K
>>> input = torch.randn(2, 3, 224, 224)
>>> seq = PatchSequential(
... ImageSequential(
... K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=0.5),
... K.RandomPerspective(0.2, p=0.5),
... K.RandomSolarize(0.1, 0.1, p=0.5),
... ),
... K.RandomAffine(360, p=1.0),
... ImageSequential(
... K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=0.5),
... K.RandomPerspective(0.2, p=0.5),
... K.RandomSolarize(0.1, 0.1, p=0.5),
... ),
... K.RandomSolarize(0.1, 0.1, p=0.1),
... grid_size=(2,2),
... patchwise_apply=False,
... same_on_batch=True,
... random_apply=True,
... )
>>> out = seq(input)
>>> out.shape
torch.Size([2, 3, 224, 224])
>>> out1 = seq(input, seq._params)
>>> torch.equal(out, out1)
True
"""
def __init__(self, *args: nn.Module, grid_size: Tuple[int, int]=(4, 4),
padding: str='same', same_on_batch: Optional[bool]=None, keepdim:
Optional[bool]=None, patchwise_apply: bool=False, random_apply:
Union[int, bool, Tuple[int, int]]=False) ->None:
_random_apply: 'Optional[Union[int, Tuple[int, int]]]'
if patchwise_apply and random_apply is True:
_random_apply = grid_size[0] * grid_size[1], grid_size[0
] * grid_size[1]
elif patchwise_apply and random_apply is False:
assert len(args) == grid_size[0] * grid_size[1
], f'The number of processing modules must be equal with grid size.Got {len(args)} and {grid_size[0] * grid_size[1]}.'
_random_apply = random_apply
elif patchwise_apply and isinstance(random_apply, (int, tuple)):
raise ValueError(
f'Only boolean value allowed when `patchwise_apply` is set to True. Got {random_apply}.'
)
else:
_random_apply = random_apply
super(PatchSequential, self).__init__(*args, same_on_batch=
same_on_batch, return_transform=False, keepdim=keepdim,
random_apply=_random_apply)
assert padding in ['same', 'valid'
], f'`padding` must be either `same` or `valid`. Got {padding}.'
self.grid_size = grid_size
self.padding = padding
self.patchwise_apply = patchwise_apply
def is_intensity_only(self) ->bool:
"""Check if all transformations are intensity-based.
Note: patch processing would break the continuity of labels (e.g. bbounding boxes, masks).
"""
for arg in self.children():
if isinstance(arg, (ImageSequential,)):
for _arg in arg.children():
if not isinstance(_arg, IntensityAugmentationBase2D):
return False
elif not isinstance(_arg, IntensityAugmentationBase2D):
return False
return True
def __repeat_param_across_patches__(self, param: 'torch.Tensor',
patch_num: 'int') ->torch.Tensor:
"""Repeat parameters across patches.
The input is shaped as (B, ...), while to output (B * patch_num, ...), which
to guarentee that the same transformation would happen for each patch index.
(B1, B2, ..., Bn) => (B1, ... Bn, B1, ..., Bn, ..., B1, ..., Bn)
| pt_size | | pt_size | ..., | pt_size |
"""
repeated = torch.cat([param] * patch_num, dim=0)
return repeated
def compute_padding(self, input: 'torch.Tensor', padding: 'str',
grid_size: 'Optional[Tuple[int, int]]'=None) ->Tuple[int, int, int, int
]:
if grid_size is None:
grid_size = self.grid_size
if padding == 'valid':
ph, pw = input.size(-2) // grid_size[0], input.size(-1
) // grid_size[1]
return -pw // 2, pw // 2 - pw, -ph // 2, ph // 2 - ph
elif padding == 'same':
ph = input.size(-2) - input.size(-2) // grid_size[0] * grid_size[0]
pw = input.size(-1) - input.size(-1) // grid_size[1] * grid_size[1]
return pw // 2, pw - pw // 2, ph // 2, ph - ph // 2
else:
raise NotImplementedError(
f"Expect `padding` as either 'valid' or 'same'. Got {padding}."
)
def extract_patches(self, input: 'torch.Tensor', grid_size:
'Optional[Tuple[int, int]]'=None, pad:
'Optional[Tuple[int, int, int, int]]'=None) ->torch.Tensor:
"""Extract patches from tensor.
Example:
>>> import kornia.augmentation as K
>>> pas = PatchSequential(K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.0))
>>> pas.extract_patches(torch.arange(16).view(1, 1, 4, 4), grid_size=(2, 2))
tensor([[[[[ 0, 1],
[ 4, 5]]],
<BLANKLINE>
<BLANKLINE>
[[[ 2, 3],
[ 6, 7]]],
<BLANKLINE>
<BLANKLINE>
[[[ 8, 9],
[12, 13]]],
<BLANKLINE>
<BLANKLINE>
[[[10, 11],
[14, 15]]]]])
>>> pas.extract_patches(torch.arange(54).view(1, 1, 6, 9), grid_size=(2, 2), pad=(-1, -1, -2, -2))
tensor([[[[[19, 20, 21]]],
<BLANKLINE>
<BLANKLINE>
[[[22, 23, 24]]],
<BLANKLINE>
<BLANKLINE>
[[[28, 29, 30]]],
<BLANKLINE>
<BLANKLINE>
[[[31, 32, 33]]]]])
"""
if pad is not None:
input = torch.nn.functional.pad(input, list(pad))
if grid_size is None:
grid_size = self.grid_size
window_size = input.size(-2) // grid_size[-2], input.size(-1
) // grid_size[-1]
stride = window_size
return extract_tensor_patches(input, window_size, stride)
def restore_from_patches(self, patches: 'torch.Tensor', grid_size:
'Tuple[int, int]'=(4, 4), pad:
'Optional[Tuple[int, int, int, int]]'=None) ->torch.Tensor:
"""Restore input from patches.
Example:
>>> import kornia.augmentation as K
>>> pas = PatchSequential(K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.0))
>>> out = pas.extract_patches(torch.arange(16).view(1, 1, 4, 4), grid_size=(2, 2))
>>> pas.restore_from_patches(out, grid_size=(2, 2))
tensor([[[[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]]]])
"""
if grid_size is None:
grid_size = self.grid_size
patches_tensor = patches.view(-1, grid_size[0], grid_size[1], *
patches.shape[-3:])
restored_tensor = torch.cat(torch.chunk(patches_tensor, grid_size[0
], dim=1), -2).squeeze(1)
restored_tensor = torch.cat(torch.chunk(restored_tensor, grid_size[
1], dim=1), -1).squeeze(1)
if pad is not None:
restored_tensor = torch.nn.functional.pad(restored_tensor, [(-i
) for i in pad])
return restored_tensor
def forward_patchwise(self, input: 'torch.Tensor', params:
'Optional[List[List[ParamItem]]]'=None) ->torch.Tensor:
if params is None:
params = [[]] * input.size(1)
auglist = [self.get_forward_sequence() for _ in range(input.
size(1))]
else:
auglist = [self.get_forward_sequence(p) for p in params]
assert input.size(0) == len(auglist) == len(params)
out = []
self._params = []
for inp, proc, param in zip(input, auglist, params):
o = []
p = []
for inp_pat, (proc_name, proc_pat), _param in zip_longest(inp,
proc, param):
if isinstance(proc_pat, (_AugmentationBase, ImageSequential)):
o.append(proc_pat(inp_pat[None], _param.data if _param
is not None else None))
p.append(ParamItem(proc_name, proc_pat._params))
else:
o.append(proc_pat(inp_pat[None]))
p.append(ParamItem(proc_name, {}))
out.append(torch.cat(o, dim=0))
self._params.append(p)
input = torch.stack(out, dim=0)
return input
def forward_batchwise(self, input: 'torch.Tensor', params:
'Optional[List[ParamItem]]'=None) ->torch.Tensor:
if self.same_on_batch:
batch_shape = input.size(1), *input.shape[-3:]
patch_num = input.size(0)
else:
batch_shape = input.size(0) * input.size(1), *input.shape[-3:]
if params is None:
params = []
for name, aug in self.get_forward_sequence():
if isinstance(aug, _AugmentationBase):
aug.same_on_batch = False
param = aug.forward_parameters(batch_shape)
if self.same_on_batch:
for k, v in param.items():
if not (k == 'order' and isinstance(aug,
ColorJitter)):
param.update({k: self.
__repeat_param_across_patches__(v,
patch_num)})
aug.same_on_batch = True
else:
param = None
params.append(ParamItem(name, param))
input = super().forward(input.view(-1, *input.shape[-3:]), params)
return input
def forward(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', params:
'Optional[Union[List[ParamItem], List[List[ParamItem]]]]'=None
) ->Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
"""Input transformation will be returned if input is a tuple."""
if isinstance(input, (tuple,)):
pad = self.compute_padding(input[0], self.padding)
input = self.extract_patches(input[0], self.grid_size, pad), input[
1]
else:
pad = self.compute_padding(input, self.padding)
input = self.extract_patches(input, self.grid_size, pad)
if not self.patchwise_apply:
params = cast(List[ParamItem], params)
if isinstance(input, (tuple,)):
input = self.forward_batchwise(input[0], params), input[1]
else:
input = self.forward_batchwise(input, params)
else:
params = cast(List[List[ParamItem]], params)
if isinstance(input, (tuple,)):
input = self.forward_patchwise(input[0], params), input[1]
else:
input = self.forward_patchwise(input, params)
if isinstance(input, (tuple,)):
input = self.restore_from_patches(input[0], self.grid_size, pad=pad
), input[1]
else:
input = self.restore_from_patches(input, self.grid_size, pad=pad)
return input
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import math
import warnings
from typing import Dict
from typing import Optional
from typing import Tuple
import torch.nn as nn
import torch.nn.functional as F
from typing import cast
from typing import List
from typing import Union
from torch.distributions import Bernoulli
from itertools import zip_longest
from collections import OrderedDict
from typing import Any
from typing import Iterator
from typing import NamedTuple
from torch.nn.modules.utils import _pair
from math import pi
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x5 = xindex // 16
x6 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = x1
tmp7 = tmp5 < tmp3
tmp8 = tmp7 & tmp4
tmp9 = tl.load(in_ptr0 + (16 * x2 + 64 * x3 + 16 * x3 % 16), tmp8 &
xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tmp5 >= tmp3
tmp11 = tl.full([1], 2, tl.int64)
tmp12 = tmp5 < tmp11
tmp13 = tmp10 & tmp12
tmp14 = tmp13 & tmp4
tmp15 = tl.load(in_ptr0 + (4 + 16 * x5), tmp14 & xmask, eviction_policy
='evict_last', other=0.0)
tmp16 = tmp5 >= tmp11
tmp17 = tl.full([1], 3, tl.int64)
tmp18 = tmp5 < tmp17
tmp19 = tmp16 & tmp18
tmp20 = tmp19 & tmp4
tmp21 = tl.load(in_ptr0 + (8 + 16 * x5), tmp20 & xmask, eviction_policy
='evict_last', other=0.0)
tmp22 = tmp5 >= tmp17
tl.full([1], 4, tl.int64)
tmp25 = tmp22 & tmp4
tmp26 = tl.load(in_ptr0 + (12 + 16 * x5), tmp25 & xmask,
eviction_policy='evict_last', other=0.0)
tmp27 = tl.where(tmp19, tmp21, tmp26)
tmp28 = tl.where(tmp13, tmp15, tmp27)
tmp29 = tl.where(tmp7, tmp9, tmp28)
tmp30 = tl.full(tmp29.shape, 0.0, tmp29.dtype)
tmp31 = tl.where(tmp4, tmp29, tmp30)
tmp32 = tmp0 >= tmp3
tmp33 = tmp0 < tmp11
tmp34 = tmp32 & tmp33
tmp35 = tmp7 & tmp34
tmp36 = tl.load(in_ptr0 + (1 + 16 * x5), tmp35 & xmask, eviction_policy
='evict_last', other=0.0)
tmp37 = tmp13 & tmp34
tmp38 = tl.load(in_ptr0 + (5 + 16 * x5), tmp37 & xmask, eviction_policy
='evict_last', other=0.0)
tmp39 = tmp19 & tmp34
tmp40 = tl.load(in_ptr0 + (9 + 16 * x5), tmp39 & xmask, eviction_policy
='evict_last', other=0.0)
tmp41 = tmp22 & tmp34
tmp42 = tl.load(in_ptr0 + (13 + 16 * x5), tmp41 & xmask,
eviction_policy='evict_last', other=0.0)
tmp43 = tl.where(tmp19, tmp40, tmp42)
tmp44 = tl.where(tmp13, tmp38, tmp43)
tmp45 = tl.where(tmp7, tmp36, tmp44)
tmp46 = tl.full(tmp45.shape, 0.0, tmp45.dtype)
tmp47 = tl.where(tmp34, tmp45, tmp46)
tmp48 = tmp0 >= tmp11
tmp49 = tmp0 < tmp17
tmp50 = tmp48 & tmp49
tmp51 = tmp7 & tmp50
tmp52 = tl.load(in_ptr0 + (2 + 16 * x5), tmp51 & xmask, eviction_policy
='evict_last', other=0.0)
tmp53 = tmp13 & tmp50
tmp54 = tl.load(in_ptr0 + (6 + 16 * x5), tmp53 & xmask, eviction_policy
='evict_last', other=0.0)
tmp55 = tmp19 & tmp50
tmp56 = tl.load(in_ptr0 + (10 + 16 * x5), tmp55 & xmask,
eviction_policy='evict_last', other=0.0)
tmp57 = tmp22 & tmp50
tmp58 = tl.load(in_ptr0 + (14 + 16 * x5), tmp57 & xmask,
eviction_policy='evict_last', other=0.0)
tmp59 = tl.where(tmp19, tmp56, tmp58)
tmp60 = tl.where(tmp13, tmp54, tmp59)
tmp61 = tl.where(tmp7, tmp52, tmp60)
tmp62 = tl.full(tmp61.shape, 0.0, tmp61.dtype)
tmp63 = tl.where(tmp50, tmp61, tmp62)
tmp64 = tmp0 >= tmp17
tmp66 = tmp7 & tmp64
tmp67 = tl.load(in_ptr0 + (3 + 16 * x5), tmp66 & xmask, eviction_policy
='evict_last', other=0.0)
tmp68 = tmp13 & tmp64
tmp69 = tl.load(in_ptr0 + (7 + 16 * x5), tmp68 & xmask, eviction_policy
='evict_last', other=0.0)
tmp70 = tmp19 & tmp64
tmp71 = tl.load(in_ptr0 + (11 + 16 * x5), tmp70 & xmask,
eviction_policy='evict_last', other=0.0)
tmp72 = tmp22 & tmp64
tmp73 = tl.load(in_ptr0 + (15 + 16 * x5), tmp72 & xmask,
eviction_policy='evict_last', other=0.0)
tmp74 = tl.where(tmp19, tmp71, tmp73)
tmp75 = tl.where(tmp13, tmp69, tmp74)
tmp76 = tl.where(tmp7, tmp67, tmp75)
tmp77 = tl.full(tmp76.shape, 0.0, tmp76.dtype)
tmp78 = tl.where(tmp64, tmp76, tmp77)
tmp79 = tl.where(tmp50, tmp63, tmp78)
tmp80 = tl.where(tmp34, tmp47, tmp79)
tmp81 = tl.where(tmp4, tmp31, tmp80)
tl.store(out_ptr0 + x6, tmp81, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 4, 4, 4), (64, 64, 16, 4, 1),
torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(256)](arg0_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0),
def _adapted_sampling(shape: 'Union[Tuple, torch.Size]', dist:
'torch.distributions.Distribution', same_on_batch=False) ->torch.Tensor:
"""The uniform sampling function that accepts 'same_on_batch'.
If same_on_batch is True, all values generated will be exactly same given a batch_size (shape[0]).
By default, same_on_batch is set to False.
"""
if same_on_batch:
return dist.sample((1, *shape[1:])).repeat(shape[0], *([1] * (len(
shape) - 1)))
return dist.sample(shape)
def _transform_output_shape(output:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', shape: 'Tuple'
) ->Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
"""Collapse the broadcasted batch dimensions an input tensor to be the specified shape.
Args:
input: torch.Tensor
shape: List/tuple of int
Returns:
torch.Tensor
"""
is_tuple = isinstance(output, tuple)
out_tensor: 'torch.Tensor'
trans_matrix: 'Optional[torch.Tensor]'
if is_tuple:
out_tensor, trans_matrix = cast(Tuple[torch.Tensor, torch.Tensor],
output)
else:
out_tensor = cast(torch.Tensor, output)
trans_matrix = None
if trans_matrix is not None:
if len(out_tensor.shape) > len(shape):
assert trans_matrix.shape[0
] == 1, f'Dimension 0 of transformation matrix is expected to be 1, got {trans_matrix.shape[0]}'
trans_matrix = trans_matrix.squeeze(0)
for dim in range(len(out_tensor.shape) - len(shape)):
assert out_tensor.shape[0
] == 1, f'Dimension {dim} of input is expected to be 1, got {out_tensor.shape[0]}'
out_tensor = out_tensor.squeeze(0)
return (out_tensor, trans_matrix) if is_tuple else out_tensor
def _transform_input(input: 'torch.Tensor') ->torch.Tensor:
"""Reshape an input tensor to be (*, C, H, W). Accept either (H, W), (C, H, W) or (*, C, H, W).
Args:
input: torch.Tensor
Returns:
torch.Tensor
"""
if not torch.is_tensor(input):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if len(input.shape) not in [2, 3, 4]:
raise ValueError(
f'Input size must have a shape of either (H, W), (C, H, W) or (*, C, H, W). Got {input.shape}'
)
if len(input.shape) == 2:
input = input.unsqueeze(0)
if len(input.shape) == 3:
input = input.unsqueeze(0)
return input
def _validate_input_dtype(input: 'torch.Tensor', accepted_dtypes: 'List'
) ->None:
"""Check if the dtype of the input tensor is in the range of accepted_dtypes
Args:
input: torch.Tensor
accepted_dtypes: List. e.g. [torch.float32, torch.float64]
"""
if input.dtype not in accepted_dtypes:
raise TypeError(
f'Expected input of {accepted_dtypes}. Got {input.dtype}')
def _extract_device_dtype(tensor_list: 'List[Optional[Any]]') ->Tuple[torch
.device, torch.dtype]:
"""Check if all the input are in the same device (only if when they are torch.Tensor).
If so, it would return a tuple of (device, dtype). Default: (cpu, ``get_default_dtype()``).
Returns:
[torch.device, torch.dtype]
"""
device, dtype = None, None
for tensor in tensor_list:
if tensor is not None:
if not isinstance(tensor, (torch.Tensor,)):
continue
_device = tensor.device
_dtype = tensor.dtype
if device is None and dtype is None:
device = _device
dtype = _dtype
elif device != _device or dtype != _dtype:
raise ValueError(
f'Passed values are not in the same device and dtype.Got ({device}, {dtype}) and ({_device}, {_dtype}).'
)
if device is None:
device = torch.device('cpu')
if dtype is None:
dtype = torch.get_default_dtype()
return device, dtype
def _joint_range_check(ranged_factor: 'torch.Tensor', name: 'str', bounds:
'Optional[Tuple[float, float]]'=None) ->None:
"""check if bounds[0] <= ranged_factor[0] <= ranged_factor[1] <= bounds[1]"""
if bounds is None:
bounds = float('-inf'), float('inf')
if ranged_factor.dim() == 1 and len(ranged_factor) == 2:
if not bounds[0] <= ranged_factor[0] or not bounds[1] >= ranged_factor[
1]:
raise ValueError(
f'{name} out of bounds. Expected inside {bounds}, got {ranged_factor}.'
)
if not bounds[0] <= ranged_factor[0] <= ranged_factor[1] <= bounds[1]:
raise ValueError(
f'{name}[0] should be smaller than {name}[1] got {ranged_factor}'
)
else:
raise TypeError(
f'{name} should be a tensor with length 2 whose values between {bounds}. Got {ranged_factor}.'
)
def _singular_range_check(ranged_factor: 'torch.Tensor', name: 'str',
bounds: 'Optional[Tuple[float, float]]'=None, skip_none: 'bool'=False,
mode: 'str'='2d') ->None:
"""check if bounds[0] <= ranged_factor[0] <= bounds[1] and bounds[0] <= ranged_factor[1] <= bounds[1]"""
if mode == '2d':
dim_size = 2
elif mode == '3d':
dim_size = 3
else:
raise ValueError(f"'mode' shall be either 2d or 3d. Got {mode}")
if skip_none and ranged_factor is None:
return
if bounds is None:
bounds = float('-inf'), float('inf')
if ranged_factor.dim() == 1 and len(ranged_factor) == dim_size:
for f in ranged_factor:
if not bounds[0] <= f <= bounds[1]:
raise ValueError(
f'{name} out of bounds. Expected inside {bounds}, got {ranged_factor}.'
)
else:
raise TypeError(
f'{name} should be a float number or a tuple with length {dim_size} whose values between {bounds}.Got {ranged_factor}'
)
def _range_bound(factor:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]', name:
'str', center: 'float'=0.0, bounds: 'Tuple[float, float]'=(0, float(
'inf')), check: 'Optional[str]'='joint', device: 'torch.device'=torch.
device('cpu'), dtype: 'torch.dtype'=torch.get_default_dtype()
) ->torch.Tensor:
"""Check inputs and compute the corresponding factor bounds"""
if not isinstance(factor, torch.Tensor):
factor = torch.tensor(factor, device=device, dtype=dtype)
factor_bound: 'torch.Tensor'
if factor.dim() == 0:
if factor < 0:
raise ValueError(
f'If {name} is a single number number, it must be non negative. Got {factor}'
)
factor_bound = factor.repeat(2) * torch.tensor([-1.0, 1.0], device=
factor.device, dtype=factor.dtype) + center
factor_bound = factor_bound.clamp(bounds[0], bounds[1])
else:
factor_bound = torch.as_tensor(factor, device=device, dtype=dtype)
if check is not None:
if check == 'joint':
_joint_range_check(factor_bound, name, bounds)
elif check == 'singular':
_singular_range_check(factor_bound, name, bounds)
else:
raise NotImplementedError(f"methods '{check}' not implemented.")
return factor_bound
def adjust_brightness(input: 'torch.Tensor', brightness_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust Brightness of an image.
.. image:: _static/img/adjust_brightness.png
This implementation aligns OpenCV, not PIL. Hence, the output differs from TorchVision.
The input image is expected to be in the range of [0, 1].
Args:
input: image to be adjusted in the shape of :math:`(*, N)`.
brightness_factor: Brightness adjust factor per element
in the batch. 0 does not modify the input image while any other number modify the
brightness.
Return:
Adjusted image in the shape of :math:`(*, N)`.
Example:
>>> x = torch.ones(1, 1, 2, 2)
>>> adjust_brightness(x, 1.)
tensor([[[[1., 1.],
[1., 1.]]]])
>>> x = torch.ones(2, 5, 3, 3)
>>> y = torch.tensor([0.25, 0.50])
>>> adjust_brightness(x, y).shape
torch.Size([2, 5, 3, 3])
"""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(brightness_factor, (float, torch.Tensor)):
raise TypeError(
f'The factor should be either a float or torch.Tensor. Got {type(brightness_factor)}'
)
if isinstance(brightness_factor, float):
brightness_factor = torch.tensor([brightness_factor])
brightness_factor = brightness_factor.to(input.device)
for _ in input.shape[1:]:
brightness_factor = torch.unsqueeze(brightness_factor, dim=-1)
x_adjust: 'torch.Tensor' = input + brightness_factor
out: 'torch.Tensor' = torch.clamp(x_adjust, 0.0, 1.0)
return out
def adjust_contrast(input: 'torch.Tensor', contrast_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust Contrast of an image.
.. image:: _static/img/adjust_contrast.png
This implementation aligns OpenCV, not PIL. Hence, the output differs from TorchVision.
The input image is expected to be in the range of [0, 1].
Args:
input: Image to be adjusted in the shape of :math:`(*, N)`.
contrast_factor: Contrast adjust factor per element
in the batch. 0 generates a completely black image, 1 does not modify
the input image while any other non-negative number modify the
brightness by this factor.
Return:
Adjusted image in the shape of :math:`(*, N)`.
Example:
>>> x = torch.ones(1, 1, 2, 2)
>>> adjust_contrast(x, 0.5)
tensor([[[[0.5000, 0.5000],
[0.5000, 0.5000]]]])
>>> x = torch.ones(2, 5, 3, 3)
>>> y = torch.tensor([0.65, 0.50])
>>> adjust_contrast(x, y).shape
torch.Size([2, 5, 3, 3])
"""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(contrast_factor, (float, torch.Tensor)):
raise TypeError(
f'The factor should be either a float or torch.Tensor. Got {type(contrast_factor)}'
)
if isinstance(contrast_factor, float):
contrast_factor = torch.tensor([contrast_factor])
contrast_factor = contrast_factor.to(input.device)
if (contrast_factor < 0).any():
raise ValueError(
f'Contrast factor must be non-negative. Got {contrast_factor}')
for _ in input.shape[1:]:
contrast_factor = torch.unsqueeze(contrast_factor, dim=-1)
x_adjust: 'torch.Tensor' = input * contrast_factor
out: 'torch.Tensor' = torch.clamp(x_adjust, 0.0, 1.0)
return out
def adjust_hue_raw(input: 'torch.Tensor', hue_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust hue of an image. Expecting input to be in hsv format already."""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(hue_factor, (float, torch.Tensor)):
raise TypeError(
f'The hue_factor should be a float number or torch.Tensor in the range between [-PI, PI]. Got {type(hue_factor)}'
)
if isinstance(hue_factor, float):
hue_factor = torch.as_tensor(hue_factor)
hue_factor = hue_factor
for _ in input.shape[1:]:
hue_factor = torch.unsqueeze(hue_factor, dim=-1)
h, s, v = torch.chunk(input, chunks=3, dim=-3)
divisor: 'float' = 2 * pi
h_out: 'torch.Tensor' = torch.fmod(h + hue_factor, divisor)
out: 'torch.Tensor' = torch.cat([h_out, s, v], dim=-3)
return out
def hsv_to_rgb(image: 'torch.Tensor') ->torch.Tensor:
"""Convert an image from HSV to RGB.
The H channel values are assumed to be in the range 0..2pi. S and V are in the range 0..1.
Args:
image: HSV Image to be converted to HSV with shape of :math:`(*, 3, H, W)`.
Returns:
RGB version of the image with shape of :math:`(*, 3, H, W)`.
Example:
>>> input = torch.rand(2, 3, 4, 5)
>>> output = hsv_to_rgb(input) # 2x3x4x5
"""
if not isinstance(image, torch.Tensor):
raise TypeError('Input type is not a torch.Tensor. Got {}'.format(
type(image)))
if len(image.shape) < 3 or image.shape[-3] != 3:
raise ValueError('Input size must have a shape of (*, 3, H, W). Got {}'
.format(image.shape))
h: 'torch.Tensor' = image[..., 0, :, :] / (2 * math.pi)
s: 'torch.Tensor' = image[..., 1, :, :]
v: 'torch.Tensor' = image[..., 2, :, :]
hi: 'torch.Tensor' = torch.floor(h * 6) % 6
f: 'torch.Tensor' = h * 6 % 6 - hi
one: 'torch.Tensor' = torch.tensor(1.0).to(image.device)
p: 'torch.Tensor' = v * (one - s)
q: 'torch.Tensor' = v * (one - f * s)
t: 'torch.Tensor' = v * (one - (one - f) * s)
hi = hi.long()
indices: 'torch.Tensor' = torch.stack([hi, hi + 6, hi + 12], dim=-3)
out = torch.stack((v, q, p, p, t, v, t, v, v, q, p, p, p, p, t, v, v, q
), dim=-3)
out = torch.gather(out, -3, indices)
return out
def rgb_to_hsv(image: 'torch.Tensor', eps: 'float'=1e-06) ->torch.Tensor:
"""Convert an image from RGB to HSV.
.. image:: _static/img/rgb_to_hsv.png
The image data is assumed to be in the range of (0, 1).
Args:
image: RGB Image to be converted to HSV with shape of :math:`(*, 3, H, W)`.
eps: scalar to enforce numarical stability.
Returns:
HSV version of the image with shape of :math:`(*, 3, H, W)`.
The H channel values are in the range 0..2pi. S and V are in the range 0..1.
Example:
>>> input = torch.rand(2, 3, 4, 5)
>>> output = rgb_to_hsv(input) # 2x3x4x5
"""
if not isinstance(image, torch.Tensor):
raise TypeError('Input type is not a torch.Tensor. Got {}'.format(
type(image)))
if len(image.shape) < 3 or image.shape[-3] != 3:
raise ValueError('Input size must have a shape of (*, 3, H, W). Got {}'
.format(image.shape))
maxc, _ = image.max(-3)
maxc_mask = image == maxc.unsqueeze(-3)
_, max_indices = ((maxc_mask.cumsum(-3) == 1) & maxc_mask).max(-3)
minc: 'torch.Tensor' = image.min(-3)[0]
v: 'torch.Tensor' = maxc
deltac: 'torch.Tensor' = maxc - minc
s: 'torch.Tensor' = deltac / (v + eps)
deltac = torch.where(deltac == 0, torch.ones_like(deltac, device=deltac
.device, dtype=deltac.dtype), deltac)
maxc_tmp = maxc.unsqueeze(-3) - image
rc: 'torch.Tensor' = maxc_tmp[..., 0, :, :]
gc: 'torch.Tensor' = maxc_tmp[..., 1, :, :]
bc: 'torch.Tensor' = maxc_tmp[..., 2, :, :]
h = torch.stack([bc - gc, 2.0 * deltac + rc - bc, 4.0 * deltac + gc -
rc], dim=-3)
h = torch.gather(h, dim=-3, index=max_indices[..., None, :, :])
h = h.squeeze(-3)
h = h / deltac
h = h / 6.0 % 1.0
h = 2 * math.pi * h
return torch.stack([h, s, v], dim=-3)
def adjust_hue(input: 'torch.Tensor', hue_factor: 'Union[float, torch.Tensor]'
) ->torch.Tensor:
"""Adjust hue of an image.
.. image:: _static/img/adjust_hue.png
The input image is expected to be an RGB image in the range of [0, 1].
Args:
input: Image to be adjusted in the shape of :math:`(*, 3, H, W)`.
hue_factor: How much to shift the hue channel. Should be in [-PI, PI]. PI
and -PI give complete reversal of hue channel in HSV space in positive and negative
direction respectively. 0 means no shift. Therefore, both -PI and PI will give an
image with complementary colors while 0 gives the original image.
Return:
Adjusted image in the shape of :math:`(*, 3, H, W)`.
Example:
>>> x = torch.ones(1, 3, 2, 2)
>>> adjust_hue(x, 3.141516).shape
torch.Size([1, 3, 2, 2])
>>> x = torch.ones(2, 3, 3, 3)
>>> y = torch.ones(2) * 3.141516
>>> adjust_hue(x, y).shape
torch.Size([2, 3, 3, 3])
"""
x_hsv: 'torch.Tensor' = rgb_to_hsv(input)
x_adjusted: 'torch.Tensor' = adjust_hue_raw(x_hsv, hue_factor)
out: 'torch.Tensor' = hsv_to_rgb(x_adjusted)
return out
def adjust_saturation_raw(input: 'torch.Tensor', saturation_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust color saturation of an image. Expecting input to be in hsv format already."""
if not isinstance(input, torch.Tensor):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if not isinstance(saturation_factor, (float, torch.Tensor)):
raise TypeError(
f'The saturation_factor should be a float number or torch.Tensor.Got {type(saturation_factor)}'
)
if isinstance(saturation_factor, float):
saturation_factor = torch.as_tensor(saturation_factor)
saturation_factor = saturation_factor.to(input.device)
for _ in input.shape[1:]:
saturation_factor = torch.unsqueeze(saturation_factor, dim=-1)
h, s, v = torch.chunk(input, chunks=3, dim=-3)
s_out: 'torch.Tensor' = torch.clamp(s * saturation_factor, min=0, max=1)
out: 'torch.Tensor' = torch.cat([h, s_out, v], dim=-3)
return out
def adjust_saturation(input: 'torch.Tensor', saturation_factor:
'Union[float, torch.Tensor]') ->torch.Tensor:
"""Adjust color saturation of an image.
.. image:: _static/img/adjust_saturation.png
The input image is expected to be an RGB image in the range of [0, 1].
Args:
input: Image/Tensor to be adjusted in the shape of :math:`(*, 3, H, W)`.
saturation_factor: How much to adjust the saturation. 0 will give a black
and white image, 1 will give the original image while 2 will enhance the saturation by a factor of 2.
Return:
Adjusted image in the shape of :math:`(*, 3, H, W)`.
Example:
>>> x = torch.ones(1, 3, 3, 3)
>>> adjust_saturation(x, 2.).shape
torch.Size([1, 3, 3, 3])
>>> x = torch.ones(2, 3, 3, 3)
>>> y = torch.tensor([1., 2.])
>>> adjust_saturation(x, y).shape
torch.Size([2, 3, 3, 3])
"""
x_hsv: 'torch.Tensor' = rgb_to_hsv(input)
x_adjusted: 'torch.Tensor' = adjust_saturation_raw(x_hsv, saturation_factor
)
out: 'torch.Tensor' = hsv_to_rgb(x_adjusted)
return out
def _extract_tensor_patchesnd(input: 'torch.Tensor', window_sizes:
'Tuple[int, ...]', strides: 'Tuple[int, ...]') ->torch.Tensor:
batch_size, num_channels = input.size()[:2]
dims = range(2, input.dim())
for dim, patch_size, stride in zip(dims, window_sizes, strides):
input = input.unfold(dim, patch_size, stride)
input = input.permute(0, *dims, 1, *[(dim + len(dims)) for dim in dims]
).contiguous()
return input.view(batch_size, -1, num_channels, *window_sizes)
def extract_tensor_patches(input: 'torch.Tensor', window_size:
'Union[int, Tuple[int, int]]', stride: 'Union[int, Tuple[int, int]]'=1,
padding: 'Union[int, Tuple[int, int]]'=0) ->torch.Tensor:
"""Function that extract patches from tensors and stack them.
See :class:`~kornia.contrib.ExtractTensorPatches` for details.
"""
if not torch.is_tensor(input):
raise TypeError('Input input type is not a torch.Tensor. Got {}'.
format(type(input)))
if not len(input.shape) == 4:
raise ValueError('Invalid input shape, we expect BxCxHxW. Got: {}'.
format(input.shape))
if padding:
pad_vert, pad_horz = _pair(padding)
input = F.pad(input, [pad_horz, pad_horz, pad_vert, pad_vert])
return _extract_tensor_patchesnd(input, _pair(window_size), _pair(stride))
class _BasicAugmentationBase(nn.Module):
"""_BasicAugmentationBase base class for customized augmentation implementations.
Plain augmentation base class without the functionality of transformation matrix calculations.
By default, the random computations will be happened on CPU with ``torch.get_default_dtype()``.
To change this behaviour, please use ``set_rng_device_and_dtype``.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation
probabilities element-wisely.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def __init__(self, p: 'float'=0.5, p_batch: 'float'=1.0, same_on_batch:
'bool'=False, keepdim: 'bool'=False) ->None:
super(_BasicAugmentationBase, self).__init__()
self.p = p
self.p_batch = p_batch
self.same_on_batch = same_on_batch
self.keepdim = keepdim
self._params: 'Dict[str, torch.Tensor]' = {}
if p != 0.0 or p != 1.0:
self._p_gen = Bernoulli(self.p)
if p_batch != 0.0 or p_batch != 1.0:
self._p_batch_gen = Bernoulli(self.p_batch)
self.set_rng_device_and_dtype(torch.device('cpu'), torch.
get_default_dtype())
def __repr__(self) ->str:
return (
f'p={self.p}, p_batch={self.p_batch}, same_on_batch={self.same_on_batch}'
)
def __unpack_input__(self, input: 'torch.Tensor') ->torch.Tensor:
return input
def __check_batching__(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]'):
"""Check if a transformation matrix is returned,
it has to be in the same batching mode as output."""
raise NotImplementedError
def transform_tensor(self, input: 'torch.Tensor') ->torch.Tensor:
"""Standardize input tensors."""
raise NotImplementedError
def generate_parameters(self, batch_shape: 'torch.Size') ->Dict[str,
torch.Tensor]:
return {}
def apply_transform(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->torch.Tensor:
raise NotImplementedError
def set_rng_device_and_dtype(self, device: 'torch.device', dtype:
'torch.dtype') ->None:
"""Change the random generation device and dtype.
Note:
The generated random numbers are not reproducible across different devices and dtypes.
"""
self.device = device
self.dtype = dtype
def __batch_prob_generator__(self, batch_shape: 'torch.Size', p:
'float', p_batch: 'float', same_on_batch: 'bool') ->torch.Tensor:
batch_prob: 'torch.Tensor'
if p_batch == 1:
batch_prob = torch.tensor([True])
elif p_batch == 0:
batch_prob = torch.tensor([False])
else:
batch_prob = _adapted_sampling((1,), self._p_batch_gen,
same_on_batch).bool()
if batch_prob.sum().item() == 1:
elem_prob: 'torch.Tensor'
if p == 1:
elem_prob = torch.tensor([True] * batch_shape[0])
elif p == 0:
elem_prob = torch.tensor([False] * batch_shape[0])
else:
elem_prob = _adapted_sampling((batch_shape[0],), self.
_p_gen, same_on_batch).bool()
batch_prob = batch_prob * elem_prob
else:
batch_prob = batch_prob.repeat(batch_shape[0])
return batch_prob
def forward_parameters(self, batch_shape):
to_apply = self.__batch_prob_generator__(batch_shape, self.p, self.
p_batch, self.same_on_batch)
_params = self.generate_parameters(torch.Size((int(to_apply.sum().
item()), *batch_shape[1:])))
if _params is None:
_params = {}
_params['batch_prob'] = to_apply
return _params
def apply_func(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->Union[torch.Tensor, Tuple[torch.Tensor,
torch.Tensor]]:
input = self.transform_tensor(input)
return self.apply_transform(input, params)
def forward(self, input: 'torch.Tensor', params:
'Optional[Dict[str, torch.Tensor]]'=None) ->Union[torch.Tensor,
Tuple[torch.Tensor, torch.Tensor]]:
in_tensor = self.__unpack_input__(input)
self.__check_batching__(input)
ori_shape = in_tensor.shape
in_tensor = self.transform_tensor(in_tensor)
batch_shape = in_tensor.shape
if params is None:
params = self.forward_parameters(batch_shape)
self._params = params
output = self.apply_func(input, self._params)
return _transform_output_shape(output, ori_shape
) if self.keepdim else output
class _AugmentationBase(_BasicAugmentationBase):
"""_AugmentationBase base class for customized augmentation implementations.
Advanced augmentation base class with the functionality of transformation matrix calculations.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation probabilities
element-wisely for a batch.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
return_transform (bool): if ``True`` return the matrix describing the geometric transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def __init__(self, return_transform: 'bool'=None, same_on_batch: 'bool'
=False, p: 'float'=0.5, p_batch: 'float'=1.0, keepdim: 'bool'=False
) ->None:
super(_AugmentationBase, self).__init__(p, p_batch=p_batch,
same_on_batch=same_on_batch, keepdim=keepdim)
self.p = p
self.p_batch = p_batch
self.return_transform = return_transform
def __repr__(self) ->str:
return super().__repr__(
) + f', return_transform={self.return_transform}'
def identity_matrix(self, input: 'torch.Tensor') ->torch.Tensor:
raise NotImplementedError
def compute_transformation(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->torch.Tensor:
raise NotImplementedError
def apply_transform(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]', transform: 'Optional[torch.Tensor]'=None
) ->torch.Tensor:
raise NotImplementedError
def __unpack_input__(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]') ->Tuple[
torch.Tensor, Optional[torch.Tensor]]:
if isinstance(input, tuple):
in_tensor = input[0]
in_transformation = input[1]
return in_tensor, in_transformation
in_tensor = input
return in_tensor, None
def apply_func(self, in_tensor: 'torch.Tensor', in_transform:
'Optional[torch.Tensor]', params: 'Dict[str, torch.Tensor]',
return_transform: 'bool'=False) ->Union[torch.Tensor, Tuple[torch.
Tensor, torch.Tensor]]:
to_apply = params['batch_prob']
if torch.sum(to_apply) == 0:
output = in_tensor
trans_matrix = self.identity_matrix(in_tensor)
elif torch.sum(to_apply) == len(to_apply):
trans_matrix = self.compute_transformation(in_tensor, params)
output = self.apply_transform(in_tensor, params, trans_matrix)
else:
output = in_tensor.clone()
trans_matrix = self.identity_matrix(in_tensor)
trans_matrix[to_apply] = self.compute_transformation(in_tensor[
to_apply], params)
output[to_apply] = self.apply_transform(in_tensor[to_apply],
params, trans_matrix[to_apply])
self._transform_matrix = trans_matrix
if return_transform:
out_transformation = (trans_matrix if in_transform is None else
trans_matrix @ in_transform)
return output, out_transformation
if in_transform is not None:
return output, in_transform
return output
def forward(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', params:
'Optional[Dict[str, torch.Tensor]]'=None, return_transform:
'Optional[bool]'=None) ->Union[torch.Tensor, Tuple[torch.Tensor,
torch.Tensor]]:
in_tensor, in_transform = self.__unpack_input__(input)
self.__check_batching__(input)
ori_shape = in_tensor.shape
in_tensor = self.transform_tensor(in_tensor)
batch_shape = in_tensor.shape
if return_transform is None:
return_transform = self.return_transform
return_transform = cast(bool, return_transform)
if params is None:
params = self.forward_parameters(batch_shape)
if 'batch_prob' not in params:
params['batch_prob'] = torch.tensor([True] * batch_shape[0])
warnings.warn(
'`batch_prob` is not found in params. Will assume applying on all data.'
)
self._params = params
output = self.apply_func(in_tensor, in_transform, self._params,
return_transform)
return _transform_output_shape(output, ori_shape
) if self.keepdim else output
class AugmentationBase2D(_AugmentationBase):
"""AugmentationBase2D base class for customized augmentation implementations.
For any augmentation, the implementation of "generate_parameters" and "apply_transform" are required while the
"compute_transformation" is only required when passing "return_transform" as True.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation probabilities
element-wisely for a batch.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
return_transform (bool): if ``True`` return the matrix describing the geometric transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def __check_batching__(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]'):
if isinstance(input, tuple):
inp, mat = input
if len(inp.shape) == 4:
assert len(mat.shape
) == 3, 'Input tensor is in batch mode but transformation matrix is not'
assert mat.shape[0] == inp.shape[0
], f'In batch dimension, input has {inp.shape[0]}but transformation matrix has {mat.shape[0]}'
elif len(inp.shape) == 3 or len(inp.shape) == 2:
assert len(mat.shape
) == 2, 'Input tensor is in non-batch mode but transformation matrix is not'
else:
raise ValueError(
f'Unrecognized output shape. Expected 2, 3, or 4, got {len(inp.shape)}'
)
def transform_tensor(self, input: 'torch.Tensor') ->torch.Tensor:
"""Convert any incoming (H, W), (C, H, W) and (B, C, H, W) into (B, C, H, W)."""
_validate_input_dtype(input, accepted_dtypes=[torch.float16, torch.
float32, torch.float64])
return _transform_input(input)
def identity_matrix(self, input) ->torch.Tensor:
"""Return 3x3 identity matrix."""
return kornia.eye_like(3, input)
class IntensityAugmentationBase2D(AugmentationBase2D):
"""IntensityAugmentationBase2D base class for customized intensity augmentation implementations.
For any augmentation, the implementation of "generate_parameters" and "apply_transform" are required while the
"compute_transformation" is only required when passing "return_transform" as True.
Args:
p (float): probability for applying an augmentation. This param controls the augmentation probabilities
element-wisely for a batch.
p_batch (float): probability for applying an augmentation to a batch. This param controls the augmentation
probabilities batch-wisely.
return_transform (bool): if ``True`` return the matrix describing the geometric transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch (bool): apply the same transformation across the batch. Default: False.
keepdim (bool): whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). Default: False.
"""
def compute_transformation(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]') ->torch.Tensor:
return self.identity_matrix(input)
class ParamItem(NamedTuple):
name: 'str'
data: 'Union[dict, list]'
class ImageSequential(nn.Sequential):
"""Sequential for creating kornia image processing pipeline.
Args:
*args : a list of kornia augmentation and image operation modules.
same_on_batch: apply the same transformation across the batch.
If None, it will not overwrite the function-wise settings.
return_transform: if ``True`` return the matrix describing the transformation
applied to each. If None, it will not overwrite the function-wise settings.
keepdim: whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). If None, it will not overwrite the function-wise settings.
random_apply: randomly select a sublist (order agnostic) of args to
apply transformation.
If int, a fixed number of transformations will be selected.
If (a,), x number of transformations (a <= x <= len(args)) will be selected.
If (a, b), x number of transformations (a <= x <= b) will be selected.
If True, the whole list of args will be processed as a sequence in a random order.
If False, the whole list of args will be processed as a sequence in original order.
Returns:
Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]: the tensor (, and the transformation matrix)
has been sequentially modified by the args.
Examples:
>>> import kornia
>>> input = torch.randn(2, 3, 5, 6)
>>> aug_list = ImageSequential(
... kornia.color.BgrToRgb(),
... kornia.augmentation.ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.0),
... kornia.filters.MedianBlur((3, 3)),
... kornia.augmentation.RandomAffine(360, p=1.0),
... kornia.enhance.Invert(),
... return_transform=True,
... same_on_batch=True,
... random_apply=10,
... )
>>> out = aug_list(input)
>>> out[0].shape, out[1].shape
(torch.Size([2, 3, 5, 6]), torch.Size([2, 3, 3]))
Reproduce with provided params.
>>> out2 = aug_list(input, params=aug_list._params)
>>> torch.equal(out[0], out2[0]), torch.equal(out[1], out2[1])
(True, True)
Note:
Transformation matrix returned only considers the transformation applied in ``kornia.augmentation`` module.
Those transformations in ``kornia.geometry`` will not be taken into account.
"""
def __init__(self, *args: nn.Module, same_on_batch: Optional[bool]=None,
return_transform: Optional[bool]=None, keepdim: Optional[bool]=None,
random_apply: Union[int, bool, Tuple[int, int]]=False) ->None:
self.same_on_batch = same_on_batch
self.return_transform = return_transform
self.keepdim = keepdim
_args = OrderedDict()
for idx, arg in enumerate(args):
if not isinstance(arg, nn.Module):
raise NotImplementedError(
f'Only nn.Module are supported at this moment. Got {arg}.')
if isinstance(arg, _AugmentationBase):
if same_on_batch is not None:
arg.same_on_batch = same_on_batch
if return_transform is not None:
arg.return_transform = return_transform
if keepdim is not None:
arg.keepdim = keepdim
_args.update({f'{arg.__class__.__name__}_{idx}': arg})
super(ImageSequential, self).__init__(_args)
self._params: 'List[Any]' = []
self.random_apply: 'Union[Tuple[int, int], bool]'
if random_apply:
if isinstance(random_apply, (bool,)) and random_apply is True:
self.random_apply = len(args), len(args) + 1
elif isinstance(random_apply, (int,)):
self.random_apply = random_apply, random_apply + 1
elif isinstance(random_apply, (tuple,)) and len(random_apply
) == 2 and isinstance(random_apply[0], (int,)) and isinstance(
random_apply[1], (int,)):
self.random_apply = random_apply[0], random_apply[1] + 1
elif isinstance(random_apply, (tuple,)) and len(random_apply
) == 1 and isinstance(random_apply[0], (int,)):
self.random_apply = random_apply[0], len(args) + 1
else:
raise ValueError(
f'Non-readable random_apply. Got {random_apply}.')
assert isinstance(self.random_apply, (tuple,)) and len(self.
random_apply) == 2 and isinstance(self.random_apply[0], (int,)
) and isinstance(self.random_apply[0], (int,)
), f'Expect a tuple of (int, int). Got {self.random_apply}.'
else:
self.random_apply = False
def _get_child_sequence(self) ->Iterator[Tuple[str, nn.Module]]:
if self.random_apply:
num_samples = int(torch.randint(*self.random_apply, (1,)).item())
indices = torch.multinomial(torch.ones((len(self),)),
num_samples, replacement=True if num_samples > len(self) else
False)
return self._get_children_by_indices(indices)
return self.named_children()
def _get_children_by_indices(self, indices: 'torch.Tensor') ->Iterator[
Tuple[str, nn.Module]]:
modules = list(self.named_children())
for idx in indices:
yield modules[idx]
def _get_children_by_module_names(self, names: 'List[str]') ->Iterator[
Tuple[str, nn.Module]]:
modules = list(self.named_children())
for name in names:
yield modules[list(dict(self.named_children()).keys()).index(name)]
def get_forward_sequence(self, params: 'Optional[List[ParamItem]]'=None
) ->Iterator[Tuple[str, nn.Module]]:
if params is None:
named_modules = self._get_child_sequence()
else:
named_modules = self._get_children_by_module_names([p.name for
p in params])
return named_modules
def apply_to_input(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]',
module_name: 'str', module: 'Optional[nn.Module]'=None, param:
'Optional[ParamItem]'=None) ->Union[torch.Tensor, Tuple[torch.
Tensor, torch.Tensor]]:
if module is None:
module = self.get_submodule(module_name)
if param is not None:
assert module_name == param.name
_param = param.data
else:
_param = None
if isinstance(module, (_AugmentationBase, ImageSequential)
) and _param is None:
input = module(input)
self._params.append(ParamItem(module_name, module._params))
elif isinstance(module, (_AugmentationBase, ImageSequential)
) and _param is not None:
input = module(input, params=_param)
self._params.append(ParamItem(module_name, _param))
else:
assert _param == {
} or _param is None, f'Non-augmentaion operation {module_name} require empty parameters. Got {module}.'
if isinstance(input, (tuple, list)):
input = module(input[0]), input[1]
else:
input = module(input)
self._params.append(ParamItem(module_name, {}))
return input
def forward(self, input:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', params:
'Optional[List[ParamItem]]'=None) ->Union[torch.Tensor, Tuple[torch
.Tensor, torch.Tensor]]:
self._params = []
named_modules = self.get_forward_sequence(params)
params = [] if params is None else params
for (name, module), param in zip_longest(named_modules, params):
input = self.apply_to_input(input, name, module, param=param)
return input
class ColorJitter(IntensityAugmentationBase2D):
"""Applies a random transformation to the brightness, contrast, saturation and hue of a tensor image.
.. image:: _static/img/ColorJitter.png
Args:
p: probability of applying the transformation.
brightness: The brightness factor to apply.
contrast: The contrast factor to apply.
saturation: The saturation factor to apply.
hue: The hue factor to apply.
return_transform: if ``True`` return the matrix describing the transformation applied to each
input tensor. If ``False`` and the input is a tuple the applied transformation
wont be concatenated.
same_on_batch: apply the same transformation across the batch.
keepdim: whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False).
Shape:
- Input: :math:`(C, H, W)` or :math:`(B, C, H, W)`, Optional: :math:`(B, 3, 3)`
- Output: :math:`(B, C, H, W)`
Note:
Input tensor must be float and normalized into [0, 1] for the best differentiability support.
Additionally, this function accepts another transformation tensor (:math:`(B, 3, 3)`), then the
applied transformation will be merged int to the input transformation tensor and returned.
Examples:
>>> rng = torch.manual_seed(0)
>>> inputs = torch.ones(1, 3, 3, 3)
>>> aug = ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.)
>>> aug(inputs)
tensor([[[[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993]],
<BLANKLINE>
[[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993]],
<BLANKLINE>
[[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993],
[0.9993, 0.9993, 0.9993]]]])
"""
def __init__(self, brightness:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]'=0.0,
contrast:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]'=0.0,
saturation:
'Union[torch.Tensor, float, Tuple[float, float], List[float]]'=0.0,
hue: 'Union[torch.Tensor, float, Tuple[float, float], List[float]]'
=0.0, return_transform: 'bool'=False, same_on_batch: 'bool'=False,
p: 'float'=1.0, keepdim: 'bool'=False) ->None:
super(ColorJitter, self).__init__(p=p, return_transform=
return_transform, same_on_batch=same_on_batch, keepdim=keepdim)
self._device, self._dtype = _extract_device_dtype([brightness,
contrast, hue, saturation])
self.brightness = brightness
self.contrast = contrast
self.saturation = saturation
self.hue = hue
def __repr__(self) ->str:
repr = (
f'brightness={self.brightness}, contrast={self.contrast}, saturation={self.saturation}, hue={self.hue}'
)
return self.__class__.__name__ + f'({repr}, {super().__repr__()})'
def generate_parameters(self, batch_shape: 'torch.Size') ->Dict[str,
torch.Tensor]:
brightness: 'torch.Tensor' = _range_bound(self.brightness,
'brightness', center=1.0, bounds=(0, 2), device=self._device,
dtype=self._dtype)
contrast: 'torch.Tensor' = _range_bound(self.contrast, 'contrast',
center=1.0, device=self._device, dtype=self._dtype)
saturation: 'torch.Tensor' = _range_bound(self.saturation,
'saturation', center=1.0, device=self._device, dtype=self._dtype)
hue: 'torch.Tensor' = _range_bound(self.hue, 'hue', bounds=(-0.5,
0.5), device=self._device, dtype=self._dtype)
return rg.random_color_jitter_generator(batch_shape[0], brightness,
contrast, saturation, hue, self.same_on_batch, self.device,
self.dtype)
def apply_transform(self, input: 'torch.Tensor', params:
'Dict[str, torch.Tensor]', transform: 'Optional[torch.Tensor]'=None
) ->torch.Tensor:
transforms = [lambda img: adjust_brightness(img, params[
'brightness_factor'] - 1), lambda img: adjust_contrast(img,
params['contrast_factor']), lambda img: adjust_saturation(img,
params['saturation_factor']), lambda img: adjust_hue(img,
params['hue_factor'] * 2 * pi)]
jittered = input
for idx in params['order'].tolist():
t = transforms[idx]
jittered = t(jittered)
return jittered
class PatchSequentialNew(ImageSequential):
"""Container for performing patch-level image processing.
.. image:: https://kornia-tutorials.readthedocs.io/en/latest/_images/data_patch_sequential_5_1.png
PatchSequential breaks input images into patches by a given grid size, which will be resembled back
afterwards. Different image processing and augmentation methods will be performed on each patch region.
Args:
*args: a list of processing modules.
grid_size: controls the grid board seperation.
padding: same or valid padding. If same padding, it will pad to include all pixels if the input
tensor cannot be divisible by grid_size. If valid padding, the redundent border will be removed.
same_on_batch: apply the same transformation across the batch.
If None, it will not overwrite the function-wise settings.
keepdim: whether to keep the output shape the same as input (True) or broadcast it
to the batch form (False). If None, it will not overwrite the function-wise settings.
patchwise_apply: apply image processing args will be applied patch-wisely.
if ``True``, the number of args must be equal to grid number.
if ``False``, the image processing args will be applied as a sequence to all patches. Default: False.
random_apply: randomly select a sublist (order agnostic) of args to
apply transformation.
If ``int`` (batchwise mode only), a fixed number of transformations will be selected.
If ``(a,)`` (batchwise mode only), x number of transformations (a <= x <= len(args)) will be selected.
If ``(a, b)`` (batchwise mode only), x number of transformations (a <= x <= b) will be selected.
If ``True``, the whole list of args will be processed in a random order.
If ``False``, the whole list of args will be processed in original order.
Return:
List[Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]]: the tensor (, and the transformation matrix)
has been sequentially modified by the args.
Examples:
>>> import kornia.augmentation as K
>>> input = torch.randn(2, 3, 224, 224)
>>> seq = PatchSequential(
... ImageSequential(
... K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=0.5),
... K.RandomPerspective(0.2, p=0.5),
... K.RandomSolarize(0.1, 0.1, p=0.5),
... ),
... K.RandomAffine(360, p=1.0),
... ImageSequential(
... K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=0.5),
... K.RandomPerspective(0.2, p=0.5),
... K.RandomSolarize(0.1, 0.1, p=0.5),
... ),
... K.RandomSolarize(0.1, 0.1, p=0.1),
... grid_size=(2,2),
... patchwise_apply=False,
... same_on_batch=True,
... random_apply=True,
... )
>>> out = seq(input)
>>> out.shape
torch.Size([2, 3, 224, 224])
>>> out1 = seq(input, seq._params)
>>> torch.equal(out, out1)
True
"""
def __init__(self, *args: nn.Module, grid_size: Tuple[int, int]=(4, 4),
padding: str='same', same_on_batch: Optional[bool]=None, keepdim:
Optional[bool]=None, patchwise_apply: bool=False, random_apply:
Union[int, bool, Tuple[int, int]]=False) ->None:
_random_apply: 'Optional[Union[int, Tuple[int, int]]]'
if patchwise_apply and random_apply is True:
_random_apply = grid_size[0] * grid_size[1], grid_size[0
] * grid_size[1]
elif patchwise_apply and random_apply is False:
assert len(args) == grid_size[0] * grid_size[1
], f'The number of processing modules must be equal with grid size.Got {len(args)} and {grid_size[0] * grid_size[1]}.'
_random_apply = random_apply
elif patchwise_apply and isinstance(random_apply, (int, tuple)):
raise ValueError(
f'Only boolean value allowed when `patchwise_apply` is set to True. Got {random_apply}.'
)
else:
_random_apply = random_apply
super(PatchSequentialNew, self).__init__(*args, same_on_batch=
same_on_batch, return_transform=False, keepdim=keepdim,
random_apply=_random_apply)
assert padding in ['same', 'valid'
], f'`padding` must be either `same` or `valid`. Got {padding}.'
self.grid_size = grid_size
self.padding = padding
self.patchwise_apply = patchwise_apply
def is_intensity_only(self) ->bool:
"""Check if all transformations are intensity-based.
Note: patch processing would break the continuity of labels (e.g. bbounding boxes, masks).
"""
for arg in self.children():
if isinstance(arg, (ImageSequential,)):
for _arg in arg.children():
if not isinstance(_arg, IntensityAugmentationBase2D):
return False
elif not isinstance(_arg, IntensityAugmentationBase2D):
return False
return True
def __repeat_param_across_patches__(self, param: 'torch.Tensor',
patch_num: 'int') ->torch.Tensor:
"""Repeat parameters across patches.
The input is shaped as (B, ...), while to output (B * patch_num, ...), which
to guarentee that the same transformation would happen for each patch index.
(B1, B2, ..., Bn) => (B1, ... Bn, B1, ..., Bn, ..., B1, ..., Bn)
| pt_size | | pt_size | ..., | pt_size |
"""
repeated = torch.cat([param] * patch_num, dim=0)
return repeated
def compute_padding(self, input: 'torch.Tensor', padding: 'str',
grid_size: 'Optional[Tuple[int, int]]'=None) ->Tuple[int, int, int, int
]:
if grid_size is None:
grid_size = self.grid_size
if padding == 'valid':
ph, pw = input.size(-2) // grid_size[0], input.size(-1
) // grid_size[1]
return -pw // 2, pw // 2 - pw, -ph // 2, ph // 2 - ph
elif padding == 'same':
ph = input.size(-2) - input.size(-2) // grid_size[0] * grid_size[0]
pw = input.size(-1) - input.size(-1) // grid_size[1] * grid_size[1]
return pw // 2, pw - pw // 2, ph // 2, ph - ph // 2
else:
raise NotImplementedError(
f"Expect `padding` as either 'valid' or 'same'. Got {padding}."
)
def extract_patches(self, input: 'torch.Tensor', grid_size:
'Optional[Tuple[int, int]]'=None, pad:
'Optional[Tuple[int, int, int, int]]'=None) ->torch.Tensor:
"""Extract patches from tensor.
Example:
>>> import kornia.augmentation as K
>>> pas = PatchSequential(K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.0))
>>> pas.extract_patches(torch.arange(16).view(1, 1, 4, 4), grid_size=(2, 2))
tensor([[[[[ 0, 1],
[ 4, 5]]],
<BLANKLINE>
<BLANKLINE>
[[[ 2, 3],
[ 6, 7]]],
<BLANKLINE>
<BLANKLINE>
[[[ 8, 9],
[12, 13]]],
<BLANKLINE>
<BLANKLINE>
[[[10, 11],
[14, 15]]]]])
>>> pas.extract_patches(torch.arange(54).view(1, 1, 6, 9), grid_size=(2, 2), pad=(-1, -1, -2, -2))
tensor([[[[[19, 20, 21]]],
<BLANKLINE>
<BLANKLINE>
[[[22, 23, 24]]],
<BLANKLINE>
<BLANKLINE>
[[[28, 29, 30]]],
<BLANKLINE>
<BLANKLINE>
[[[31, 32, 33]]]]])
"""
if pad is not None:
input = torch.nn.functional.pad(input, list(pad))
if grid_size is None:
grid_size = self.grid_size
window_size = input.size(-2) // grid_size[-2], input.size(-1
) // grid_size[-1]
stride = window_size
return extract_tensor_patches(input, window_size, stride)
def restore_from_patches(self, patches: 'torch.Tensor', grid_size:
'Tuple[int, int]'=(4, 4), pad:
'Optional[Tuple[int, int, int, int]]'=None) ->torch.Tensor:
"""Restore input from patches.
Example:
>>> import kornia.augmentation as K
>>> pas = PatchSequential(K.ColorJitter(0.1, 0.1, 0.1, 0.1, p=1.0))
>>> out = pas.extract_patches(torch.arange(16).view(1, 1, 4, 4), grid_size=(2, 2))
>>> pas.restore_from_patches(out, grid_size=(2, 2))
tensor([[[[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]]]])
"""
if grid_size is None:
grid_size = self.grid_size
patches_tensor = patches.view(-1, grid_size[0], grid_size[1], *
patches.shape[-3:])
restored_tensor = torch.cat(torch.chunk(patches_tensor, grid_size[0
], dim=1), -2).squeeze(1)
restored_tensor = torch.cat(torch.chunk(restored_tensor, grid_size[
1], dim=1), -1).squeeze(1)
if pad is not None:
restored_tensor = torch.nn.functional.pad(restored_tensor, [(-i
) for i in pad])
return restored_tensor
def forward_patchwise(self, input: 'torch.Tensor', params:
'Optional[List[List[ParamItem]]]'=None) ->torch.Tensor:
if params is None:
params = [[]] * input.size(1)
auglist = [self.get_forward_sequence() for _ in range(input.
size(1))]
else:
auglist = [self.get_forward_sequence(p) for p in params]
assert input.size(0) == len(auglist) == len(params)
out = []
self._params = []
for inp, proc, param in zip(input, auglist, params):
o = []
p = []
for inp_pat, (proc_name, proc_pat), _param in zip_longest(inp,
proc, param):
if isinstance(proc_pat, (_AugmentationBase, ImageSequential)):
o.append(proc_pat(inp_pat[None], _param.data if _param
is not None else None))
p.append(ParamItem(proc_name, proc_pat._params))
else:
o.append(proc_pat(inp_pat[None]))
p.append(ParamItem(proc_name, {}))
out.append(torch.cat(o, dim=0))
self._params.append(p)
input = torch.stack(out, dim=0)
return input
def forward_batchwise(self, input: 'torch.Tensor', params:
'Optional[List[ParamItem]]'=None) ->torch.Tensor:
if self.same_on_batch:
batch_shape = input.size(1), *input.shape[-3:]
patch_num = input.size(0)
else:
batch_shape = input.size(0) * input.size(1), *input.shape[-3:]
if params is None:
params = []
for name, aug in self.get_forward_sequence():
if isinstance(aug, _AugmentationBase):
aug.same_on_batch = False
param = aug.forward_parameters(batch_shape)
if self.same_on_batch:
for k, v in param.items():
if not (k == 'order' and isinstance(aug,
ColorJitter)):
param.update({k: self.
__repeat_param_across_patches__(v,
patch_num)})
aug.same_on_batch = True
else:
param = None
params.append(ParamItem(name, param))
input = super().forward(input.view(-1, *input.shape[-3:]), params)
return input
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| rozumden/kornia | PatchSequential | false | 4,247 | [
"ECL-2.0",
"Apache-2.0"
] | 0 | f62f324b201eea50e1e50db3fbf3e968e0a337c5 | https://github.com/rozumden/kornia/tree/f62f324b201eea50e1e50db3fbf3e968e0a337c5 | import math
import torch
import warnings
from typing import Dict
from typing import Optional
from typing import Tuple
import torch.nn as nn
import torch.nn.functional as F
from typing import cast
from typing import List
from typing import Union
from torch.distributions import Bernoulli
from itertools import zip_longest
from collections import OrderedDict
from typing import Any
from typing import Iterator
from typing import NamedTuple
from torch.nn.modules.utils import _pair
from math import pi
def _adapted_sampling(shape: 'Union[Tuple, torch.Size]', dist:
'torch.distributions.Distribution', same_on_batch=False) ->torch.Tensor:
"""The uniform sampling function that accepts 'same_on_batch'.
If same_on_batch is True, all values generated will be exactly same given a batch_size (shape[0]).
By default, same_on_batch is set to False.
"""
if same_on_batch:
return dist.sample((1, *shape[1:])).repeat(shape[0], *([1] * (len(
shape) - 1)))
return dist.sample(shape)
def _transform_output_shape(output:
'Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]', shape: 'Tuple'
) ->Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
"""Collapse the broadcasted batch dimensions an input tensor to be the specified shape.
Args:
input: torch.Tensor
shape: List/tuple of int
Returns:
torch.Tensor
"""
is_tuple = isinstance(output, tuple)
out_tensor: 'torch.Tensor'
trans_matrix: 'Optional[torch.Tensor]'
if is_tuple:
out_tensor, trans_matrix = cast(Tuple[torch.Tensor, torch.Tensor],
output)
else:
out_tensor = cast(torch.Tensor, output)
trans_matrix = None
if trans_matrix is not None:
if len(out_tensor.shape) > len(shape):
assert trans_matrix.shape[0
] == 1, f'Dimension 0 of transformation matrix is expected to be 1, got {trans_matrix.shape[0]}'
trans_matrix = trans_matrix.squeeze(0)
for dim in range(len(out_tensor.shape) - len(shape)):
assert out_tensor.shape[0
] == 1, f'Dimension {dim} of input is expected to be 1, got {out_tensor.shape[0]}'
out_tensor = out_tensor.squeeze(0)
return (out_tensor, trans_matrix) if is_tuple else out_tensor
def _transform_input(input: 'torch.Tensor') ->torch.Tensor:
"""Reshape an input tensor to be (*, C, H, W). Accept either (H, W), (C, H, W) or (*, C, H, W).
Args:
input: torch.Tensor
Returns:
torch.Tensor
"""
if not torch.is_tensor(input):
raise TypeError(f'Input type is not a torch.Tensor. Got {type(input)}')
if len(input.shape) not in [2, 3, 4]:
raise ValueError(
f'Input size must have a shape of either (H, W), (C, H, W) or (*, C, H, W). Got {input.shape}'
)
if len(input.shape) == 2:
input = input.unsqueeze(0)
if len(input.shape) == 3:
input = input.unsqueeze(0)
return input
def _validate_input_dtype(input: 'torch.Tensor', accepted_dtypes: 'List'
) ->None:
"""Check if the dtype of the input tensor is in the range of accepted_dtypes
Args:
input: torch.Tensor
accepted_dtypes: List. e.g. [torch.float32, torch.float64]
"""
if input.dtype not in accepted_dtypes:
raise TypeError(
f'Expected input of {accepted_dtypes}. Got {input.dtype}')
def _extract_device_dtype(tensor_list: 'List[Optional[Any]]') ->Tuple[torch
.device, torch.dtype]:
"""Check if all the input are in the same device (only if when they are torch.Tensor).
If so, it would return a tuple of (device, dtype). Default: (cpu, ``get_default_dtype()``).
Returns:
[torch.device, torch.dtype]
"""
device, dtype = None, None
for tensor in tensor_list:
if tensor is not None:
if not isinstance(tensor, (torch.Tensor,)):
continue
_device = tensor.device
_dtype = tensor.d
# ... truncated (>4000 chars) for memory efficiency |
DAModule | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/p6/cp6vuooninjiuju55qtiu7u3yjx4izmj5jtvx2lkeddu4rheo45u.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048, 64], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 2048
xnumel = 49
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = (yindex // 512)
tmp0 = tl.load(in_ptr0 + (x2 + (49*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (512*x2) + (25088*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/oc/cochsno6wpkwamgsqz5legelnxxchuje5twfzhozvusus3e5bzmo.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 262144
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = (yindex // 512)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (512*x2) + (4608*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/kn/cknbstongwqs3wrbq3bfnzsirzut4ooksppthlbpnxztongpfx6s.py
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# linear => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_2 = async_compile.triton('triton_poi_fused_clone_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[131072],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 100352
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5z/c5zbvh6jpdpmmwxvjaaeyoc5gh4nmeke3mou5c7akohm5ejrn7eh.py
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
# Source node to ATen node mapping:
# att_1 => div_1, exp, sum_1
# wrapped_sqrt => full_default
# Graph fragment:
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([], 22.62741699796952), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %scalar_tensor_default_1 : [num_users=3] = call_function[target=torch.ops.aten.scalar_tensor.default](args = (1,), kwargs = {dtype: torch.float32, device: cuda:0, pin_memory: False})
# %ge_scalar_1 : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%full_default, 0), kwargs = {})
# %neg_default_1 : [num_users=2] = call_function[target=torch.ops.aten.neg.default](args = (%scalar_tensor_default_1,), kwargs = {})
# %where_self_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ge_scalar_1, %scalar_tensor_default_1, %neg_default_1), kwargs = {})
# %mul_tensor_2 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_12, %where_self_1), kwargs = {})
# %amax_default_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor_2, [-1], True), kwargs = {})
# %sub_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor_2, %amax_default_1), kwargs = {})
# %mul_tensor_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where_self_1, %full_default), kwargs = {})
# %div_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor_1, %mul_tensor_3), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor_1,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_per_fused__softmax_sqrt_3 = async_compile.triton('triton_per_fused__softmax_sqrt_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[256, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_sqrt_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_sqrt_3(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 196
rnumel = 49
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
x2 = xindex % 49
x3 = (xindex // 49)
tmp0 = tl.load(in_ptr0 + (r1 + (49*x0)), rmask & xmask, other=0.0)
tmp1 = tl.full([1, 1], 22.62741699796952, tl.float64)
tmp2 = tl.full([1, 1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [XBLOCK, RBLOCK])
tmp10 = tl.where(rmask & xmask, tmp8, float("-inf"))
tmp11 = triton_helpers.max2(tmp10, 1)[:, None]
tmp12 = tmp7 - tmp11
tmp13 = tmp6.to(tl.float64)
tmp14 = tmp13 * tmp1
tmp15 = tmp14.to(tl.float32)
tmp16 = tmp12 / tmp15
tmp17 = tl_math.exp(tmp16)
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp20 = tl.where(rmask & xmask, tmp18, 0)
tmp21 = tl.sum(tmp20, 1)[:, None]
tmp22 = tmp17 / tmp21
tl.store(out_ptr2 + (r1 + (49*x2) + (2432*x3)), tmp22, rmask & xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/rm/crmnosmbl45e7ixhpxrfpqt6wmsmhumqck6slidwiut6es7773qu.py
# Topologically Sorted Source Nodes: [wrapped_sqrt_1, att_4], Original ATen: [aten.sqrt, aten._softmax]
# Source node to ATen node mapping:
# att_4 => div_3, exp_1, sum_2
# wrapped_sqrt_1 => full_default_1
# Graph fragment:
# %scalar_tensor_default_1 : [num_users=3] = call_function[target=torch.ops.aten.scalar_tensor.default](args = (1,), kwargs = {dtype: torch.float32, device: cuda:0, pin_memory: False})
# %neg_default_1 : [num_users=2] = call_function[target=torch.ops.aten.neg.default](args = (%scalar_tensor_default_1,), kwargs = {})
# %full_default_1 : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([], 7.0), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %ge_scalar : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%full_default_1, 0), kwargs = {})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ge_scalar, %scalar_tensor_default_1, %neg_default_1), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_25, %where_self), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where_self, %full_default_1), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, %mul_tensor_1), kwargs = {})
# %exp_1 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_1, [-1], True), kwargs = {})
# %div_3 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_1, %sum_2), kwargs = {})
triton_per_fused__softmax_sqrt_4 = async_compile.triton('triton_per_fused__softmax_sqrt_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[2048, 512],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_sqrt_4', 'mutated_arg_names': [], 'no_x_dim': True, 'num_load': 1, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_sqrt_4(in_ptr0, out_ptr2, xnumel, rnumel):
xnumel = 2048
XBLOCK: tl.constexpr = 1
rnumel = 512
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (512*x0)), None)
tmp1 = tl.full([1], 7.0, tl.float64)
tmp2 = tl.full([1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [RBLOCK])
tmp10 = triton_helpers.promote_to_tensor(triton_helpers.max2(tmp8, 0))
tmp11 = tmp7 - tmp10
tmp12 = tmp6.to(tl.float64)
tmp13 = tmp12 * tmp1
tmp14 = tmp13.to(tl.float32)
tmp15 = tmp11 / tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tl.broadcast_to(tmp16, [RBLOCK])
tmp19 = triton_helpers.promote_to_tensor(tl.sum(tmp17, 0))
tmp20 = tmp16 / tmp19
tl.store(out_ptr2 + (r1 + (512*x0)), tmp20, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ls/clskvd4qdnewc7cysnrd6bzcenviumrko6rb6ldndg35xjjrfbzr.py
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add => add_3
# Graph fragment:
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_32, %view_33), kwargs = {})
triton_poi_fused_add_5 = async_compile.triton('triton_poi_fused_add_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256, 512], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_5', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_5(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 196
xnumel = 512
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 49
y1 = (yindex // 49)
tmp0 = tl.load(in_out_ptr0 + (x2 + (512*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (y0 + (49*x2) + (25088*y1)), xmask & ymask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x2 + (512*y3)), tmp6, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15 = args
args.clear()
assert_size_stride(primals_1, (4, 512, 1, 49), (25088, 49, 49, 1))
assert_size_stride(primals_2, (512, 512, 3, 3), (4608, 9, 3, 1))
assert_size_stride(primals_3, (512, ), (1, ))
assert_size_stride(primals_4, (512, 512), (512, 1))
assert_size_stride(primals_5, (512, ), (1, ))
assert_size_stride(primals_6, (512, 512), (512, 1))
assert_size_stride(primals_7, (512, ), (1, ))
assert_size_stride(primals_8, (512, 512), (512, 1))
assert_size_stride(primals_9, (512, ), (1, ))
assert_size_stride(primals_10, (512, 512), (512, 1))
assert_size_stride(primals_11, (512, ), (1, ))
assert_size_stride(primals_12, (512, 512, 3, 3), (4608, 9, 3, 1))
assert_size_stride(primals_13, (512, ), (1, ))
assert_size_stride(primals_14, (49, 49), (49, 1))
assert_size_stride(primals_15, (49, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 512, 1, 49), (25088, 1, 25088, 512), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(primals_1, buf0, 2048, 49, grid=grid(2048, 49), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((512, 512, 3, 3), (4608, 1, 1536, 512), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(primals_2, buf1, 262144, 9, grid=grid(262144, 9), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((512, 512, 3, 3), (4608, 1, 1536, 512), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(primals_12, buf2, 262144, 9, grid=grid(262144, 9), stream=stream0)
del primals_12
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.convolution]
buf3 = extern_kernels.convolution(buf0, buf1, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 512, 1, 49), (25088, 1, 25088, 512))
buf4 = reinterpret_tensor(buf3, (4, 49, 512), (25088, 512, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf4, primals_3, 100352, grid=grid(100352), stream=stream0)
del primals_3
buf5 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf4, (196, 512), (512, 1), 0), reinterpret_tensor(primals_4, (512, 512), (1, 512), 0), out=buf5)
buf6 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf4, (196, 512), (512, 1), 0), reinterpret_tensor(primals_6, (512, 512), (1, 512), 0), out=buf6)
buf7 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf4, (196, 512), (512, 1), 0), reinterpret_tensor(primals_8, (512, 512), (1, 512), 0), out=buf7)
buf8 = reinterpret_tensor(buf5, (4, 49, 512), (25088, 512, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.add]
triton_poi_fused_clone_2.run(buf8, primals_5, 100352, grid=grid(100352), stream=stream0)
del primals_5
buf9 = reinterpret_tensor(buf6, (4, 49, 512), (25088, 512, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.add]
triton_poi_fused_clone_2.run(buf9, primals_7, 100352, grid=grid(100352), stream=stream0)
del primals_7
buf10 = empty_strided_cuda((4, 49, 49), (2401, 49, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (4, 512, 49), (25088, 1, 512), 0), out=buf10)
buf13 = empty_strided_cuda((4, 1, 49, 49), (2432, 49, 49, 1), torch.float32)
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
triton_per_fused__softmax_sqrt_3.run(buf10, buf13, 196, 49, grid=grid(196), stream=stream0)
del buf10
buf14 = reinterpret_tensor(buf7, (4, 49, 512), (25088, 512, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.add]
triton_poi_fused_clone_2.run(buf14, primals_9, 100352, grid=grid(100352), stream=stream0)
del primals_9
buf15 = empty_strided_cuda((4, 49, 512), (25088, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf13, (4, 49, 49), (2432, 49, 1), 0), buf14, out=buf15)
buf16 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf15, (196, 512), (512, 1), 0), reinterpret_tensor(primals_10, (512, 512), (1, 512), 0), out=buf16)
# Topologically Sorted Source Nodes: [y_2], Original ATen: [aten.convolution]
buf17 = extern_kernels.convolution(buf0, buf2, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf17, (4, 512, 1, 49), (25088, 1, 25088, 512))
buf18 = buf17; del buf17 # reuse
# Topologically Sorted Source Nodes: [y_2], Original ATen: [aten.convolution]
triton_poi_fused_clone_2.run(buf18, primals_13, 100352, grid=grid(100352), stream=stream0)
del primals_13
buf19 = empty_strided_cuda((4, 512, 512), (262144, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_2], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf18, (4, 512, 49), (25088, 1, 512), 0), reinterpret_tensor(buf18, (4, 49, 512), (25088, 512, 1), 0), out=buf19)
buf22 = empty_strided_cuda((4, 1, 512, 512), (262144, 1, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [wrapped_sqrt_1, att_4], Original ATen: [aten.sqrt, aten._softmax]
triton_per_fused__softmax_sqrt_4.run(buf19, buf22, 2048, 512, grid=grid(2048), stream=stream0)
del buf19
buf23 = empty_strided_cuda((4, 512, 49), (25088, 49, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_3], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf22, (4, 512, 512), (262144, 512, 1), 0), reinterpret_tensor(buf18, (4, 512, 49), (25088, 1, 512), 0), out=buf23)
buf24 = empty_strided_cuda((2048, 49), (49, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf23, (2048, 49), (49, 1), 0), reinterpret_tensor(primals_14, (49, 49), (1, 49), 0), out=buf24)
buf25 = reinterpret_tensor(buf16, (4, 512, 1, 49), (25088, 1, 25088, 512), 0); del buf16 # reuse
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
triton_poi_fused_add_5.run(buf25, primals_11, buf24, primals_15, 196, 512, grid=grid(196, 512), stream=stream0)
del buf24
del primals_11
del primals_15
return (buf25, buf0, buf1, buf2, reinterpret_tensor(buf4, (196, 512), (512, 1), 0), buf13, reinterpret_tensor(buf15, (196, 512), (512, 1), 0), buf18, buf22, reinterpret_tensor(buf23, (2048, 49), (49, 1), 0), primals_14, primals_10, reinterpret_tensor(buf14, (4, 512, 49), (25088, 1, 512), 0), reinterpret_tensor(buf8, (4, 512, 49), (25088, 1, 512), 0), buf9, primals_8, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 512, 1, 49), (25088, 49, 49, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((512, 512, 3, 3), (4608, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((512, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((512, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((512, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((512, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((512, 512, 3, 3), (4608, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((49, 49), (49, 1), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((49, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
from torch import nn
from torch.nn import init
class ScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, d_k, d_v, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(ScaledDotProductAttention, self).__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = self.fc_k(keys).view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = self.fc_v(values).view(b_s, nk, self.h, self.d_v).permute(0, 2,
1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class PositionAttentionModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = ScaledDotProductAttention(d_model, d_k=d_model, d_v=
d_model, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1).permute(0, 2, 1)
y = self.pa(y, y, y)
return y
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(SimplifiedScaledDotProductAttention, self).__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class ChannelAttentionModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = SimplifiedScaledDotProductAttention(H * W, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1)
y = self.pa(y, y, y)
return y
class DAModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.position_attention_module = PositionAttentionModule(d_model=
512, kernel_size=3, H=7, W=7)
self.channel_attention_module = ChannelAttentionModule(d_model=512,
kernel_size=3, H=7, W=7)
def forward(self, input):
bs, c, h, w = input.shape
p_out = self.position_attention_module(input)
c_out = self.channel_attention_module(input)
p_out = p_out.permute(0, 2, 1).view(bs, c, h, w)
c_out = c_out.view(bs, c, h, w)
return p_out + c_out
def get_inputs():
return [torch.rand([4, 512, 1, 49])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import numpy as np
from torch import nn
from torch.nn import init
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 49
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = yindex // 512
tmp0 = tl.load(in_ptr0 + (x2 + 49 * y3), xmask, eviction_policy=
'evict_last')
tl.store(out_ptr0 + (y0 + 512 * x2 + 25088 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)
) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = yindex // 512
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 512 * x2 + 4608 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_clone_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, None)
@triton.jit
def triton_per_fused__softmax_sqrt_3(in_ptr0, out_ptr2, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 196
rnumel = 49
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
x2 = xindex % 49
x3 = xindex // 49
tmp0 = tl.load(in_ptr0 + (r1 + 49 * x0), rmask & xmask, other=0.0)
tmp1 = tl.full([1, 1], 22.62741699796952, tl.float64)
tmp2 = tl.full([1, 1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [XBLOCK, RBLOCK])
tmp10 = tl.where(rmask & xmask, tmp8, float('-inf'))
tmp11 = triton_helpers.max2(tmp10, 1)[:, None]
tmp12 = tmp7 - tmp11
tmp13 = tmp6.to(tl.float64)
tmp14 = tmp13 * tmp1
tmp15 = tmp14.to(tl.float32)
tmp16 = tmp12 / tmp15
tmp17 = tl_math.exp(tmp16)
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp20 = tl.where(rmask & xmask, tmp18, 0)
tmp21 = tl.sum(tmp20, 1)[:, None]
tmp22 = tmp17 / tmp21
tl.store(out_ptr2 + (r1 + 49 * x2 + 2432 * x3), tmp22, rmask & xmask)
@triton.jit
def triton_per_fused__softmax_sqrt_4(in_ptr0, out_ptr2, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 512 * x0), None)
tmp1 = tl.full([1], 7.0, tl.float64)
tmp2 = tl.full([1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [RBLOCK])
tmp10 = triton_helpers.promote_to_tensor(triton_helpers.max2(tmp8, 0))
tmp11 = tmp7 - tmp10
tmp12 = tmp6.to(tl.float64)
tmp13 = tmp12 * tmp1
tmp14 = tmp13.to(tl.float32)
tmp15 = tmp11 / tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = tl.broadcast_to(tmp16, [RBLOCK])
tmp19 = triton_helpers.promote_to_tensor(tl.sum(tmp17, 0))
tmp20 = tmp16 / tmp19
tl.store(out_ptr2 + (r1 + 512 * x0), tmp20, None)
@triton.jit
def triton_poi_fused_add_5(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 196
xnumel = 512
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 49
y1 = yindex // 49
tmp0 = tl.load(in_out_ptr0 + (x2 + 512 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + x2, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (y0 + 49 * x2 + 25088 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x2 + 512 * y3), tmp6, xmask & ymask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13, primals_14, primals_15) = args
args.clear()
assert_size_stride(primals_1, (4, 512, 1, 49), (25088, 49, 49, 1))
assert_size_stride(primals_2, (512, 512, 3, 3), (4608, 9, 3, 1))
assert_size_stride(primals_3, (512,), (1,))
assert_size_stride(primals_4, (512, 512), (512, 1))
assert_size_stride(primals_5, (512,), (1,))
assert_size_stride(primals_6, (512, 512), (512, 1))
assert_size_stride(primals_7, (512,), (1,))
assert_size_stride(primals_8, (512, 512), (512, 1))
assert_size_stride(primals_9, (512,), (1,))
assert_size_stride(primals_10, (512, 512), (512, 1))
assert_size_stride(primals_11, (512,), (1,))
assert_size_stride(primals_12, (512, 512, 3, 3), (4608, 9, 3, 1))
assert_size_stride(primals_13, (512,), (1,))
assert_size_stride(primals_14, (49, 49), (49, 1))
assert_size_stride(primals_15, (49,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 512, 1, 49), (25088, 1, 25088, 512),
torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(2048, 49)](primals_1, buf0, 2048, 49,
XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((512, 512, 3, 3), (4608, 1, 1536, 512),
torch.float32)
triton_poi_fused_1[grid(262144, 9)](primals_2, buf1, 262144, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((512, 512, 3, 3), (4608, 1, 1536, 512),
torch.float32)
triton_poi_fused_1[grid(262144, 9)](primals_12, buf2, 262144, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_12
buf3 = extern_kernels.convolution(buf0, buf1, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 512, 1, 49), (25088, 1, 25088, 512))
buf4 = reinterpret_tensor(buf3, (4, 49, 512), (25088, 512, 1), 0)
del buf3
triton_poi_fused_clone_2[grid(100352)](buf4, primals_3, 100352,
XBLOCK=1024, num_warps=4, num_stages=1)
del primals_3
buf5 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf4, (196, 512), (512, 1), 0),
reinterpret_tensor(primals_4, (512, 512), (1, 512), 0), out=buf5)
buf6 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf4, (196, 512), (512, 1), 0),
reinterpret_tensor(primals_6, (512, 512), (1, 512), 0), out=buf6)
buf7 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf4, (196, 512), (512, 1), 0),
reinterpret_tensor(primals_8, (512, 512), (1, 512), 0), out=buf7)
buf8 = reinterpret_tensor(buf5, (4, 49, 512), (25088, 512, 1), 0)
del buf5
triton_poi_fused_clone_2[grid(100352)](buf8, primals_5, 100352,
XBLOCK=1024, num_warps=4, num_stages=1)
del primals_5
buf9 = reinterpret_tensor(buf6, (4, 49, 512), (25088, 512, 1), 0)
del buf6
triton_poi_fused_clone_2[grid(100352)](buf9, primals_7, 100352,
XBLOCK=1024, num_warps=4, num_stages=1)
del primals_7
buf10 = empty_strided_cuda((4, 49, 49), (2401, 49, 1), torch.float32)
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (4, 512, 49), (
25088, 1, 512), 0), out=buf10)
buf13 = empty_strided_cuda((4, 1, 49, 49), (2432, 49, 49, 1), torch
.float32)
triton_per_fused__softmax_sqrt_3[grid(196)](buf10, buf13, 196, 49,
XBLOCK=1, num_warps=2, num_stages=1)
del buf10
buf14 = reinterpret_tensor(buf7, (4, 49, 512), (25088, 512, 1), 0)
del buf7
triton_poi_fused_clone_2[grid(100352)](buf14, primals_9, 100352,
XBLOCK=1024, num_warps=4, num_stages=1)
del primals_9
buf15 = empty_strided_cuda((4, 49, 512), (25088, 512, 1), torch.float32
)
extern_kernels.bmm(reinterpret_tensor(buf13, (4, 49, 49), (2432, 49,
1), 0), buf14, out=buf15)
buf16 = empty_strided_cuda((196, 512), (512, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf15, (196, 512), (512, 1), 0
), reinterpret_tensor(primals_10, (512, 512), (1, 512), 0), out
=buf16)
buf17 = extern_kernels.convolution(buf0, buf2, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf17, (4, 512, 1, 49), (25088, 1, 25088, 512))
buf18 = buf17
del buf17
triton_poi_fused_clone_2[grid(100352)](buf18, primals_13, 100352,
XBLOCK=1024, num_warps=4, num_stages=1)
del primals_13
buf19 = empty_strided_cuda((4, 512, 512), (262144, 512, 1), torch.
float32)
extern_kernels.bmm(reinterpret_tensor(buf18, (4, 512, 49), (25088,
1, 512), 0), reinterpret_tensor(buf18, (4, 49, 512), (25088,
512, 1), 0), out=buf19)
buf22 = empty_strided_cuda((4, 1, 512, 512), (262144, 1, 512, 1),
torch.float32)
triton_per_fused__softmax_sqrt_4[grid(2048)](buf19, buf22, 2048,
512, num_warps=4, num_stages=1)
del buf19
buf23 = empty_strided_cuda((4, 512, 49), (25088, 49, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf22, (4, 512, 512), (262144,
512, 1), 0), reinterpret_tensor(buf18, (4, 512, 49), (25088, 1,
512), 0), out=buf23)
buf24 = empty_strided_cuda((2048, 49), (49, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf23, (2048, 49), (49, 1), 0),
reinterpret_tensor(primals_14, (49, 49), (1, 49), 0), out=buf24)
buf25 = reinterpret_tensor(buf16, (4, 512, 1, 49), (25088, 1, 25088,
512), 0)
del buf16
triton_poi_fused_add_5[grid(196, 512)](buf25, primals_11, buf24,
primals_15, 196, 512, XBLOCK=16, YBLOCK=256, num_warps=8,
num_stages=1)
del buf24
del primals_11
del primals_15
return buf25, buf0, buf1, buf2, reinterpret_tensor(buf4, (196, 512), (
512, 1), 0), buf13, reinterpret_tensor(buf15, (196, 512), (512, 1), 0
), buf18, buf22, reinterpret_tensor(buf23, (2048, 49), (49, 1), 0
), primals_14, primals_10, reinterpret_tensor(buf14, (4, 512, 49),
(25088, 1, 512), 0), reinterpret_tensor(buf8, (4, 512, 49), (25088,
1, 512), 0), buf9, primals_8, primals_6, primals_4
class ScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, d_k, d_v, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(ScaledDotProductAttention, self).__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = self.fc_k(keys).view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = self.fc_v(values).view(b_s, nk, self.h, self.d_v).permute(0, 2,
1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class PositionAttentionModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = ScaledDotProductAttention(d_model, d_k=d_model, d_v=
d_model, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1).permute(0, 2, 1)
y = self.pa(y, y, y)
return y
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super(SimplifiedScaledDotProductAttention, self).__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = queries.view(b_s, nq, self.h, self.d_k).permute(0, 2, 1, 3)
k = keys.view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = values.view(b_s, nk, self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class ChannelAttentionModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = SimplifiedScaledDotProductAttention(H * W, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1)
y = self.pa(y, y, y)
return y
class DAModuleNew(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.position_attention_module = PositionAttentionModule(d_model=
512, kernel_size=3, H=7, W=7)
self.channel_attention_module = ChannelAttentionModule(d_model=512,
kernel_size=3, H=7, W=7)
def forward(self, input_0):
primals_2 = self.position_attention_module.cnn.weight
primals_3 = self.position_attention_module.cnn.bias
primals_4 = self.position_attention_module.pa.fc_q.weight
primals_5 = self.position_attention_module.pa.fc_q.bias
primals_6 = self.position_attention_module.pa.fc_k.weight
primals_7 = self.position_attention_module.pa.fc_k.bias
primals_8 = self.position_attention_module.pa.fc_v.weight
primals_9 = self.position_attention_module.pa.fc_v.bias
primals_10 = self.position_attention_module.pa.fc_o.weight
primals_11 = self.position_attention_module.pa.fc_o.bias
primals_12 = self.channel_attention_module.cnn.weight
primals_13 = self.channel_attention_module.cnn.bias
primals_14 = self.channel_attention_module.pa.fc_o.weight
primals_15 = self.channel_attention_module.pa.fc_o.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13, primals_14,
primals_15])
return output[0]
| rushirajsherlocked/External-Attention-pytorch | DAModule | false | 4,248 | [
"MIT"
] | 0 | 7d6814b2d90909adf81c62f3f8a89e30a59d6481 | https://github.com/rushirajsherlocked/External-Attention-pytorch/tree/7d6814b2d90909adf81c62f3f8a89e30a59d6481 | import torch
import numpy as np
from torch import nn
from torch.nn import init
class ScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, d_k, d_v, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super().__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.weight, mode='fan_out')
if m.bias is not None:
init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
init.normal_(m.weight, std=0.001)
if m.bias is not None:
init.constant_(m.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = self.fc_k(keys).view(b_s, nk, self.h, self.d_k).permute(0, 2, 3, 1)
v = self.fc_v(values).view(b_s, nk, self.h, self.d_v).permute(0, 2,
1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = att * attention_weights
if attention_mask is not None:
att = att.masked_fill(attention_mask, -np.inf)
att = torch.softmax(att, -1)
att = self.dropout(att)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
class PositionAttentionModule(nn.Module):
def __init__(self, d_model=512, kernel_size=3, H=7, W=7):
super().__init__()
self.cnn = nn.Conv2d(d_model, d_model, kernel_size=kernel_size,
padding=(kernel_size - 1) // 2)
self.pa = ScaledDotProductAttention(d_model, d_k=d_model, d_v=
d_model, h=1)
def forward(self, x):
bs, c, _h, _w = x.shape
y = self.cnn(x)
y = y.view(bs, c, -1).permute(0, 2, 1)
y = self.pa(y, y, y)
return y
class SimplifiedScaledDotProductAttention(nn.Module):
"""
Scaled dot-product attention
"""
def __init__(self, d_model, h, dropout=0.1):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
"""
super().__init__()
self.d_model = d_model
self.d_k = d_model // h
self.d_v = d_model // h
self.h = h
self.fc_o = nn.Linear(h * self.d_v, d_model)
self.dropout = nn.Dropout(dropout)
self.init_weights()
def init_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal_(m.we
# ... truncated (>4000 chars) for memory efficiency |
MaskedWordPredictions | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/k6/ck6o2ucwdqtvjyw7bruyzgade2k6iruvl53t2wmqy2xkgypurpgf.py
# Topologically Sorted Source Nodes: [mul, truediv, erf, add, hidden_states_1, u, sub, pow_1, s], Original ATen: [aten.mul, aten.div, aten.erf, aten.add, aten.mean, aten.sub, aten.pow]
# Source node to ATen node mapping:
# add => add
# erf => erf
# hidden_states_1 => mul_1
# mul => mul
# pow_1 => pow_1
# s => mean_1
# sub => sub
# truediv => div
# u => mean
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.5), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_1, 1.4142135623730951), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%div,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1.0), kwargs = {})
# %mul_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%mul_1, [-1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_1, %mean), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%pow_1, [-1], True), kwargs = {})
triton_poi_fused_add_div_erf_mean_mul_pow_sub_0 = async_compile.triton('triton_poi_fused_add_div_erf_mean_mul_pow_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_erf_mean_mul_pow_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_erf_mean_mul_pow_sub_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tmp10 = tmp9 * tmp1
tmp11 = tmp9 * tmp3
tmp12 = libdevice.erf(tmp11)
tmp13 = tmp12 + tmp6
tmp14 = tmp10 * tmp13
tmp15 = tmp8 + tmp14
tmp17 = tmp16 * tmp1
tmp18 = tmp16 * tmp3
tmp19 = libdevice.erf(tmp18)
tmp20 = tmp19 + tmp6
tmp21 = tmp17 * tmp20
tmp22 = tmp15 + tmp21
tmp24 = tmp23 * tmp1
tmp25 = tmp23 * tmp3
tmp26 = libdevice.erf(tmp25)
tmp27 = tmp26 + tmp6
tmp28 = tmp24 * tmp27
tmp29 = tmp22 + tmp28
tmp30 = 4.0
tmp31 = tmp29 / tmp30
tmp32 = tmp8 - tmp31
tmp33 = tmp32 * tmp32
tmp34 = tmp14 - tmp31
tmp35 = tmp34 * tmp34
tmp36 = tmp33 + tmp35
tmp37 = tmp21 - tmp31
tmp38 = tmp37 * tmp37
tmp39 = tmp36 + tmp38
tmp40 = tmp28 - tmp31
tmp41 = tmp40 * tmp40
tmp42 = tmp39 + tmp41
tmp43 = tmp42 / tmp30
tl.store(out_ptr0 + (x0), tmp31, xmask)
tl.store(out_ptr1 + (x0), tmp43, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/pg/cpgs2sqnuixouquussvupjwsl3m3pfglz6posqku5lqt2uwjaend.py
# Topologically Sorted Source Nodes: [mul, truediv, erf, add, hidden_states_1, sub, add_1, sqrt, x, mul_2, hidden_states_2], Original ATen: [aten.mul, aten.div, aten.erf, aten.add, aten.sub, aten.sqrt]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# erf => erf
# hidden_states_1 => mul_1
# hidden_states_2 => add_2
# mul => mul
# mul_2 => mul_2
# sqrt => sqrt
# sub => sub
# truediv => div
# x => div_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.5), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_1, 1.4142135623730951), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%div,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1.0), kwargs = {})
# %mul_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_1, %mean), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean_1, 1e-12), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add_1,), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %sqrt), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_4, %div_1), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_5), kwargs = {})
triton_poi_fused_add_div_erf_mul_sqrt_sub_1 = async_compile.triton('triton_poi_fused_add_div_erf_mul_sqrt_sub_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_erf_mul_sqrt_sub_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_erf_mul_sqrt_sub_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp10 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = 0.7071067811865475
tmp5 = tmp1 * tmp4
tmp6 = libdevice.erf(tmp5)
tmp7 = 1.0
tmp8 = tmp6 + tmp7
tmp9 = tmp3 * tmp8
tmp11 = tmp9 - tmp10
tmp13 = 1e-12
tmp14 = tmp12 + tmp13
tmp15 = libdevice.sqrt(tmp14)
tmp16 = tmp11 / tmp15
tmp17 = tmp0 * tmp16
tmp19 = tmp17 + tmp18
tl.store(out_ptr0 + (x2), tmp19, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/54/c54scjtvk2o35gotra5iaxkdvh5gkumub7zcwgocxnyspbrgsbol.py
# Topologically Sorted Source Nodes: [hidden_states_3], Original ATen: [aten.add]
# Source node to ATen node mapping:
# hidden_states_3 => add_3
# Graph fragment:
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_7), kwargs = {})
triton_poi_fused_add_2 = async_compile.triton('triton_poi_fused_add_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [mul, truediv, erf, add, hidden_states_1, u, sub, pow_1, s], Original ATen: [aten.mul, aten.div, aten.erf, aten.add, aten.mean, aten.sub, aten.pow]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_erf_mean_mul_pow_sub_0.run(buf0, buf1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, truediv, erf, add, hidden_states_1, sub, add_1, sqrt, x, mul_2, hidden_states_2], Original ATen: [aten.mul, aten.div, aten.erf, aten.add, aten.sub, aten.sqrt]
triton_poi_fused_add_div_erf_mul_sqrt_sub_1.run(primals_4, buf0, buf1, buf2, primals_5, buf3, 256, grid=grid(256), stream=stream0)
del buf1
del buf2
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [hidden_states_3], Original ATen: [aten.add]
triton_poi_fused_add_2.run(buf5, primals_7, 256, grid=grid(256), stream=stream0)
del primals_7
return (buf5, primals_4, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf0, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), primals_6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
def gelu(x):
"""Gaussian Error Linear Unitという活性化関数です。
LeLUが0でカクっと不連続なので、そこを連続になるように滑らかにした形のLeLUです。
"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
class BertLayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12):
"""LayerNormalization層です。
学習済みモデルをそのままロードするため、学習済みモデルの変数名に変えています。
オリジナルのGitHubの実装から変数名を変えています。
weight→gamma、bias→beta
"""
super(BertLayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(hidden_size))
self.beta = nn.Parameter(torch.zeros(hidden_size))
self.variance_epsilon = eps
def forward(self, x):
u = x.mean(-1, keepdim=True)
s = (x - u).pow(2).mean(-1, keepdim=True)
x = (x - u) / torch.sqrt(s + self.variance_epsilon)
return self.gamma * x + self.beta
class BertPredictionHeadTransform(nn.Module):
"""MaskedWordPredictionsにて、BERTからの特徴量を変換するモジュール（入出力のサイズは同じ）"""
def __init__(self, config):
super(BertPredictionHeadTransform, self).__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.transform_act_fn = gelu
self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
def forward(self, hidden_states):
"""hidden_statesはsequence_output:[minibatch, seq_len, hidden_size]"""
hidden_states = self.dense(hidden_states)
hidden_states = self.transform_act_fn(hidden_states)
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class MaskedWordPredictions(nn.Module):
def __init__(self, config):
"""事前学習課題：Masked Language Model用のモジュール
元の[2]の実装では、BertLMPredictionHeadという名前です。
"""
super(MaskedWordPredictions, self).__init__()
self.transform = BertPredictionHeadTransform(config)
self.decoder = nn.Linear(in_features=config.hidden_size,
out_features=config.vocab_size, bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
def forward(self, hidden_states):
"""
hidden_states：BERTからの出力[batch_size, seq_len, hidden_size]
"""
hidden_states = self.transform(hidden_states)
hidden_states = self.decoder(hidden_states) + self.bias
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, vocab_size=4)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_div_erf_mean_mul_pow_sub_0(in_ptr0, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp23 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tmp10 = tmp9 * tmp1
tmp11 = tmp9 * tmp3
tmp12 = libdevice.erf(tmp11)
tmp13 = tmp12 + tmp6
tmp14 = tmp10 * tmp13
tmp15 = tmp8 + tmp14
tmp17 = tmp16 * tmp1
tmp18 = tmp16 * tmp3
tmp19 = libdevice.erf(tmp18)
tmp20 = tmp19 + tmp6
tmp21 = tmp17 * tmp20
tmp22 = tmp15 + tmp21
tmp24 = tmp23 * tmp1
tmp25 = tmp23 * tmp3
tmp26 = libdevice.erf(tmp25)
tmp27 = tmp26 + tmp6
tmp28 = tmp24 * tmp27
tmp29 = tmp22 + tmp28
tmp30 = 4.0
tmp31 = tmp29 / tmp30
tmp32 = tmp8 - tmp31
tmp33 = tmp32 * tmp32
tmp34 = tmp14 - tmp31
tmp35 = tmp34 * tmp34
tmp36 = tmp33 + tmp35
tmp37 = tmp21 - tmp31
tmp38 = tmp37 * tmp37
tmp39 = tmp36 + tmp38
tmp40 = tmp28 - tmp31
tmp41 = tmp40 * tmp40
tmp42 = tmp39 + tmp41
tmp43 = tmp42 / tmp30
tl.store(out_ptr0 + x0, tmp31, xmask)
tl.store(out_ptr1 + x0, tmp43, xmask)
@triton.jit
def triton_poi_fused_add_div_erf_mul_sqrt_sub_1(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp10 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = 0.7071067811865475
tmp5 = tmp1 * tmp4
tmp6 = libdevice.erf(tmp5)
tmp7 = 1.0
tmp8 = tmp6 + tmp7
tmp9 = tmp3 * tmp8
tmp11 = tmp9 - tmp10
tmp13 = 1e-12
tmp14 = tmp12 + tmp13
tmp15 = libdevice.sqrt(tmp14)
tmp16 = tmp11 / tmp15
tmp17 = tmp0 * tmp16
tmp19 = tmp17 + tmp18
tl.store(out_ptr0 + x2, tmp19, xmask)
@triton.jit
def triton_poi_fused_add_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_div_erf_mean_mul_pow_sub_0[grid(64)](buf0,
buf1, buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_div_erf_mul_sqrt_sub_1[grid(256)](primals_4,
buf0, buf1, buf2, primals_5, buf3, 256, XBLOCK=256, num_warps=4,
num_stages=1)
del buf1
del buf2
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf4
triton_poi_fused_add_2[grid(256)](buf5, primals_7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_7
return buf5, primals_4, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf0, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), primals_6
def gelu(x):
"""Gaussian Error Linear Unitという活性化関数です。
LeLUが0でカクっと不連続なので、そこを連続になるように滑らかにした形のLeLUです。
"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
class BertLayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12):
"""LayerNormalization層です。
学習済みモデルをそのままロードするため、学習済みモデルの変数名に変えています。
オリジナルのGitHubの実装から変数名を変えています。
weight→gamma、bias→beta
"""
super(BertLayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(hidden_size))
self.beta = nn.Parameter(torch.zeros(hidden_size))
self.variance_epsilon = eps
def forward(self, x):
u = x.mean(-1, keepdim=True)
s = (x - u).pow(2).mean(-1, keepdim=True)
x = (x - u) / torch.sqrt(s + self.variance_epsilon)
return self.gamma * x + self.beta
class BertPredictionHeadTransform(nn.Module):
"""MaskedWordPredictionsにて、BERTからの特徴量を変換するモジュール（入出力のサイズは同じ）"""
def __init__(self, config):
super(BertPredictionHeadTransform, self).__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.transform_act_fn = gelu
self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
def forward(self, hidden_states):
"""hidden_statesはsequence_output:[minibatch, seq_len, hidden_size]"""
hidden_states = self.dense(hidden_states)
hidden_states = self.transform_act_fn(hidden_states)
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class MaskedWordPredictionsNew(nn.Module):
def __init__(self, config):
"""事前学習課題：Masked Language Model用のモジュール
元の[2]の実装では、BertLMPredictionHeadという名前です。
"""
super(MaskedWordPredictionsNew, self).__init__()
self.transform = BertPredictionHeadTransform(config)
self.decoder = nn.Linear(in_features=config.hidden_size,
out_features=config.vocab_size, bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
def forward(self, input_0):
primals_2 = self.bias
primals_1 = self.transform.dense.weight
primals_4 = self.transform.dense.bias
primals_5 = self.transform.LayerNorm.gamma
primals_7 = self.transform.LayerNorm.beta
primals_6 = self.decoder.weight
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| kimihitosugiyama/text_analysis | MaskedWordPredictions | false | 4,249 | [
"Apache-2.0"
] | 0 | 8f51022957928c31e52af1e0fd407daca3addb40 | https://github.com/kimihitosugiyama/text_analysis/tree/8f51022957928c31e52af1e0fd407daca3addb40 | from _paritybench_helpers import _mock_config
import math
import torch
from torch import nn
def gelu(x):
"""Gaussian Error Linear Unitという活性化関数です。
LeLUが0でカクっと不連続なので、そこを連続になるように滑らかにした形のLeLUです。
"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
class BertLayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12):
"""LayerNormalization層です。
学習済みモデルをそのままロードするため、学習済みモデルの変数名に変えています。
オリジナルのGitHubの実装から変数名を変えています。
weight→gamma、bias→beta
"""
super().__init__()
self.gamma = nn.Parameter(torch.ones(hidden_size))
self.beta = nn.Parameter(torch.zeros(hidden_size))
self.variance_epsilon = eps
def forward(self, x):
u = x.mean(-1, keepdim=True)
s = (x - u).pow(2).mean(-1, keepdim=True)
x = (x - u) / torch.sqrt(s + self.variance_epsilon)
return self.gamma * x + self.beta
class BertPredictionHeadTransform(nn.Module):
"""MaskedWordPredictionsにて、BERTからの特徴量を変換するモジュール（入出力のサイズは同じ）"""
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.transform_act_fn = gelu
self.LayerNorm = BertLayerNorm(config.hidden_size, eps=1e-12)
def forward(self, hidden_states):
"""hidden_statesはsequence_output:[minibatch, seq_len, hidden_size]"""
hidden_states = self.dense(hidden_states)
hidden_states = self.transform_act_fn(hidden_states)
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class Model(nn.Module):
def __init__(self, config):
"""事前学習課題：Masked Language Model用のモジュール
元の[2]の実装では、BertLMPredictionHeadという名前です。
"""
super().__init__()
self.transform = BertPredictionHeadTransform(config)
self.decoder = nn.Linear(in_features=config.hidden_size,
out_features=config.vocab_size, bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
def forward(self, hidden_states):
"""
hidden_states：BERTからの出力[batch_size, seq_len, hidden_size]
"""
hidden_states = self.transform(hidden_states)
hidden_states = self.decoder(hidden_states) + self.bias
return hidden_states
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Conv1dLinear | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/iu/ciuxern2omgit5ovksuiwlddxkww6e3pkid4q2h3sauzn5rbd35z.py
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv1d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%permute, %primals_2, %primals_3, [1], [1], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/24/c247lstrpia4xuwjphutxw2zn36dsajpgqm4zjinubb3jymwwkd3.py
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# linear => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_1 = async_compile.triton('triton_poi_fused_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 3
y1 = (yindex // 3)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (3*x2) + (12*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1, 1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/ht/chtsnjuwkomolt5fx5l7wyg4kjhr7idbfplladpp2ezlbh6j4v67.py
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.add]
# Source node to ATen node mapping:
# linear => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_5), kwargs = {})
triton_poi_fused_add_2 = async_compile.triton('triton_poi_fused_add_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/74/c74kvo3dqgusjb7cihtkodmg7zdl3agdwtohp7aozh6d6fuoyxrj.py
# Topologically Sorted Source Nodes: [conv1d, relu], Original ATen: [aten.convolution, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# conv1d => convolution
# relu => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%permute, %primals_2, %primals_3, [1], [1], [1], False, [0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_convolution_relu_threshold_backward_3 = async_compile.triton('triton_poi_fused_convolution_relu_threshold_backward_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_threshold_backward_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_threshold_backward_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 3) % 4
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(out_ptr0 + (x3), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(primals_1, buf0, 16, 4, grid=grid(16, 4), stream=stream0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1,), padding=(1,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 3), (12, 3, 1))
del buf0
buf2 = empty_strided_cuda((4, 3, 4), (12, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf1, primals_3, buf2, 12, 4, grid=grid(12, 4), stream=stream0)
buf3 = empty_strided_cuda((12, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (12, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
buf4 = reinterpret_tensor(buf3, (4, 3, 4), (12, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.add]
triton_poi_fused_add_2.run(buf4, primals_5, 48, grid=grid(48), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((4, 4, 3), (12, 3, 1), torch.bool)
# Topologically Sorted Source Nodes: [conv1d, relu], Original ATen: [aten.convolution, aten.relu, aten.threshold_backward]
triton_poi_fused_convolution_relu_threshold_backward_3.run(buf1, primals_3, buf5, 48, grid=grid(48), stream=stream0)
del buf1
del primals_3
return (buf4, primals_2, reinterpret_tensor(primals_1, (4, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf2, (12, 4), (4, 1), 0), primals_4, buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn
class Conv1dLinear(torch.nn.Module):
"""Conv1D + Linear for Transformer block.
A variant of MultiLayeredConv1d, which replaces second conv-layer to linear.
"""
def __init__(self, in_chans, hidden_chans, kernel_size, dropout_rate):
"""Initialize Conv1dLinear module.
Args:
in_chans (int): Number of input channels.
hidden_chans (int): Number of hidden channels.
kernel_size (int): Kernel size of conv1d.
dropout_rate (float): Dropout rate.
"""
super(Conv1dLinear, self).__init__()
self.w_1 = torch.nn.Conv1d(in_chans, hidden_chans, kernel_size,
stride=1, padding=(kernel_size - 1) // 2)
self.w_2 = torch.nn.Linear(hidden_chans, in_chans)
self.dropout = torch.nn.Dropout(dropout_rate)
def forward(self, x):
"""Calculate forward propagation.
Args:
x (Tensor): Batch of input tensors (B, *, in_chans).
Returns:
Tensor: Batch of output tensors (B, *, hidden_chans)
"""
x = torch.relu(self.w_1(x.transpose(-1, 1))).transpose(-1, 1)
return self.w_2(self.dropout(x))
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'in_chans': 4, 'hidden_chans': 4, 'kernel_size': 4,
'dropout_rate': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_0(in_ptr0, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 3
y1 = yindex // 3
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 3 * x2 + 12 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1, 1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_add_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_threshold_backward_3(in_ptr0, in_ptr1,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 3 % 4
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(out_ptr0 + x3, tmp6, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(16, 4)](primals_1, buf0, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1,),
padding=(1,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 3), (12, 3, 1))
del buf0
buf2 = empty_strided_cuda((4, 3, 4), (12, 4, 1), torch.float32)
triton_poi_fused_clone_1[grid(12, 4)](buf1, primals_3, buf2, 12, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((12, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (12, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
buf4 = reinterpret_tensor(buf3, (4, 3, 4), (12, 4, 1), 0)
del buf3
triton_poi_fused_add_2[grid(48)](buf4, primals_5, 48, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((4, 4, 3), (12, 3, 1), torch.bool)
triton_poi_fused_convolution_relu_threshold_backward_3[grid(48)](buf1,
primals_3, buf5, 48, XBLOCK=64, num_warps=1, num_stages=1)
del buf1
del primals_3
return buf4, primals_2, reinterpret_tensor(primals_1, (4, 4, 4), (16, 1,
4), 0), reinterpret_tensor(buf2, (12, 4), (4, 1), 0), primals_4, buf5
class Conv1dLinearNew(torch.nn.Module):
"""Conv1D + Linear for Transformer block.
A variant of MultiLayeredConv1d, which replaces second conv-layer to linear.
"""
def __init__(self, in_chans, hidden_chans, kernel_size, dropout_rate):
"""Initialize Conv1dLinear module.
Args:
in_chans (int): Number of input channels.
hidden_chans (int): Number of hidden channels.
kernel_size (int): Kernel size of conv1d.
dropout_rate (float): Dropout rate.
"""
super(Conv1dLinearNew, self).__init__()
self.w_1 = torch.nn.Conv1d(in_chans, hidden_chans, kernel_size,
stride=1, padding=(kernel_size - 1) // 2)
self.w_2 = torch.nn.Linear(hidden_chans, in_chans)
self.dropout = torch.nn.Dropout(dropout_rate)
def forward(self, input_0):
primals_1 = self.w_1.weight
primals_3 = self.w_1.bias
primals_4 = self.w_2.weight
primals_5 = self.w_2.bias
primals_2 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| qlindazm/asv-subtools | Conv1dLinear | false | 4,250 | [
"Apache-2.0"
] | 0 | fe1d31db9f3268622016babe944201f6ff81ed56 | https://github.com/qlindazm/asv-subtools/tree/fe1d31db9f3268622016babe944201f6ff81ed56 | import torch
import torch.nn
class Model(torch.nn.Module):
"""Conv1D + Linear for Transformer block.
A variant of MultiLayeredConv1d, which replaces second conv-layer to linear.
"""
def __init__(self, in_chans, hidden_chans, kernel_size, dropout_rate):
"""Initialize Conv1dLinear module.
Args:
in_chans (int): Number of input channels.
hidden_chans (int): Number of hidden channels.
kernel_size (int): Kernel size of conv1d.
dropout_rate (float): Dropout rate.
"""
super().__init__()
self.w_1 = torch.nn.Conv1d(in_chans, hidden_chans, kernel_size,
stride=1, padding=(kernel_size - 1) // 2)
self.w_2 = torch.nn.Linear(hidden_chans, in_chans)
self.dropout = torch.nn.Dropout(dropout_rate)
def forward(self, x):
"""Calculate forward propagation.
Args:
x (Tensor): Batch of input tensors (B, *, in_chans).
Returns:
Tensor: Batch of output tensors (B, *, hidden_chans)
"""
x = torch.relu(self.w_1(x.transpose(-1, 1))).transpose(-1, 1)
return self.w_2(self.dropout(x))
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'in_chans': 4, 'hidden_chans': 4, 'kernel_size': 4,
'dropout_rate': 0.5}]
|
PositionWiseFeedForward | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/za/czajqsynheujzbcanx5rs2m5vkipgvedkdayw54ylduw3bknhzng.py
# Topologically Sorted Source Nodes: [conv2d, mul, truediv, erf, add, mul_1], Original ATen: [aten.convolution, aten.mul, aten.div, aten.erf, aten.add]
# Source node to ATen node mapping:
# add => add
# conv2d => convolution
# erf => erf
# mul => mul
# mul_1 => mul_1
# truediv => div
# Graph fragment:
# %convolution : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, 0.5), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%convolution, 1.4142135623730951), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%div,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1.0), kwargs = {})
# %mul_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add), kwargs = {})
triton_poi_fused_add_convolution_div_erf_mul_0 = async_compile.triton('triton_poi_fused_add_convolution_div_erf_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_convolution_div_erf_mul_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_convolution_div_erf_mul_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 16
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.5
tmp4 = tmp2 * tmp3
tmp5 = 0.7071067811865475
tmp6 = tmp2 * tmp5
tmp7 = libdevice.erf(tmp6)
tmp8 = 1.0
tmp9 = tmp7 + tmp8
tmp10 = tmp4 * tmp9
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
tl.store(out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/32/c32v7egt4mupqssam3gmac2qgv3ujprjybthsgweflmot256qqw7.py
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d_1 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%mul_1, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 16, 1, 1), (16, 1, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = buf0; del buf0 # reuse
buf2 = empty_strided_cuda((4, 16, 4, 4), (256, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv2d, mul, truediv, erf, add, mul_1], Original ATen: [aten.convolution, aten.mul, aten.div, aten.erf, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_convolution_div_erf_mul_0.run(buf1, primals_2, buf2, 1024, grid=grid(1024), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf3 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 4, 4), (64, 16, 4, 1))
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf4, primals_5, 256, grid=grid(256), stream=stream0)
del primals_5
return (buf4, primals_1, primals_3, primals_4, buf1, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((16, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 16, 1, 1), (16, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
def gelu(x):
"""Implementation of the gelu activation function by Hugging Face"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
class PositionWiseFeedForward(nn.Module):
""" FeedForward Neural Networks for each position """
def __init__(self, n_hidden):
super().__init__()
self.fc1 = nn.Conv2d(n_hidden, n_hidden * 4, kernel_size=1, bias=True)
self.fc2 = nn.Conv2d(n_hidden * 4, n_hidden, kernel_size=1, bias=True)
def forward(self, x):
return self.fc2(gelu(self.fc1(x)))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_hidden': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_convolution_div_erf_mul_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 16
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.5
tmp4 = tmp2 * tmp3
tmp5 = 0.7071067811865475
tmp6 = tmp2 * tmp5
tmp7 = libdevice.erf(tmp6)
tmp8 = 1.0
tmp9 = tmp7 + tmp8
tmp10 = tmp4 * tmp9
tl.store(in_out_ptr0 + x3, tmp2, xmask)
tl.store(out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 16, 1, 1), (16, 1, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = buf0
del buf0
buf2 = empty_strided_cuda((4, 16, 4, 4), (256, 16, 4, 1), torch.float32
)
get_raw_stream(0)
triton_poi_fused_add_convolution_div_erf_mul_0[grid(1024)](buf1,
primals_2, buf2, 1024, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf3 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 4, 4), (64, 16, 4, 1))
buf4 = buf3
del buf3
triton_poi_fused_convolution_1[grid(256)](buf4, primals_5, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
return buf4, primals_1, primals_3, primals_4, buf1, buf2
def gelu(x):
"""Implementation of the gelu activation function by Hugging Face"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
class PositionWiseFeedForwardNew(nn.Module):
""" FeedForward Neural Networks for each position """
def __init__(self, n_hidden):
super().__init__()
self.fc1 = nn.Conv2d(n_hidden, n_hidden * 4, kernel_size=1, bias=True)
self.fc2 = nn.Conv2d(n_hidden * 4, n_hidden, kernel_size=1, bias=True)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| renebidart/pytorch-cifar | PositionWiseFeedForward | false | 4,251 | [
"MIT"
] | 0 | 8f623299c25f7f219bab34bc7df41fe24232b1af | https://github.com/renebidart/pytorch-cifar/tree/8f623299c25f7f219bab34bc7df41fe24232b1af | import math
import torch
import torch.nn as nn
def gelu(x):
"""Implementation of the gelu activation function by Hugging Face"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
class Model(nn.Module):
""" FeedForward Neural Networks for each position """
def __init__(self, n_hidden):
super().__init__()
self.fc1 = nn.Conv2d(n_hidden, n_hidden * 4, kernel_size=1, bias=True)
self.fc2 = nn.Conv2d(n_hidden * 4, n_hidden, kernel_size=1, bias=True)
def forward(self, x):
return self.fc2(gelu(self.fc1(x)))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
FC_Decoder | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/az/cazao7d5hdb3kcfc76acvd3yerra6cq3h4spci3xujm27v6xwinj.py
# Topologically Sorted Source Nodes: [h3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# h3 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[65536],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 65536
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 1024
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/b3/cb3q7iauu5ues625wabsv7x5gf5ycfqrby4odqljsmjvbvhzfz5h.py
# Topologically Sorted Source Nodes: [sigmoid], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# sigmoid => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_3,), kwargs = {})
triton_poi_fused_sigmoid_1 = async_compile.triton('triton_poi_fused_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[65536],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 50176
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 784
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (1024, 4), (4, 1))
assert_size_stride(primals_2, (1024, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (784, 1024), (1024, 1))
assert_size_stride(primals_5, (784, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 1024), (1024, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 1024), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 1024), (16384, 4096, 1024, 1), 0); del buf0 # reuse
buf4 = empty_strided_cuda((4, 4, 4, 1024), (16384, 4096, 1024, 1), torch.bool)
# Topologically Sorted Source Nodes: [h3], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf4, 65536, grid=grid(65536), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 784), (784, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 1024), (1024, 1), 0), reinterpret_tensor(primals_4, (1024, 784), (1, 1024), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 784), (12544, 3136, 784, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [sigmoid], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_1.run(buf3, primals_5, 50176, grid=grid(50176), stream=stream0)
del primals_5
return (buf3, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 1024), (1024, 1), 0), buf3, primals_4, buf4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((1024, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1024, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((784, 1024), (1024, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((784, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class FC_Decoder(nn.Module):
def __init__(self, embedding_size):
super(FC_Decoder, self).__init__()
self.fc3 = nn.Linear(embedding_size, 1024)
self.fc4 = nn.Linear(1024, 784)
def forward(self, z):
h3 = F.relu(self.fc3(z))
return torch.sigmoid(self.fc4(h3))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'embedding_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 1024
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 50176
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 784
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (1024, 4), (4, 1))
assert_size_stride(primals_2, (1024,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (784, 1024), (1024, 1))
assert_size_stride(primals_5, (784,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 1024), (1024, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 1024), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 1024), (16384, 4096, 1024,
1), 0)
del buf0
buf4 = empty_strided_cuda((4, 4, 4, 1024), (16384, 4096, 1024, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(65536)](buf1,
primals_2, buf4, 65536, XBLOCK=512, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 784), (784, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 1024), (1024, 1), 0
), reinterpret_tensor(primals_4, (1024, 784), (1, 1024), 0),
out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 784), (12544, 3136, 784,
1), 0)
del buf2
triton_poi_fused_sigmoid_1[grid(50176)](buf3, primals_5, 50176,
XBLOCK=512, num_warps=4, num_stages=1)
del primals_5
return buf3, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 1024), (1024, 1), 0
), buf3, primals_4, buf4
class FC_DecoderNew(nn.Module):
def __init__(self, embedding_size):
super(FC_DecoderNew, self).__init__()
self.fc3 = nn.Linear(embedding_size, 1024)
self.fc4 = nn.Linear(1024, 784)
def forward(self, input_0):
primals_1 = self.fc3.weight
primals_2 = self.fc3.bias
primals_4 = self.fc4.weight
primals_5 = self.fc4.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| saksham36/LangGrounding | FC_Decoder | false | 4,252 | [
"MIT"
] | 0 | 89ee9e5b8090e61e6bf7bf2b3e1dd45edf9664b7 | https://github.com/saksham36/LangGrounding/tree/89ee9e5b8090e61e6bf7bf2b3e1dd45edf9664b7 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, embedding_size):
super().__init__()
self.fc3 = nn.Linear(embedding_size, 1024)
self.fc4 = nn.Linear(1024, 784)
def forward(self, z):
h3 = F.relu(self.fc3(z))
return torch.sigmoid(self.fc4(h3))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
Word2Vec | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/td/ctdj5kazgiki6gdaadhqtp2x7tq2ee5ey5hqqdcoqmp54jyhf74f.py
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => amax, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_3, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_3, %amax), kwargs = {})
triton_poi_fused__log_softmax_0 = async_compile.triton('triton_poi_fused__log_softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/u5/cu5ua3ykbptipkew3i3zng4a7a4hy4f6xs547ovdooepce7uyfwz.py
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => exp, log, sub_1, sum_1
# Graph fragment:
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub, %log), kwargs = {})
triton_poi_fused__log_softmax_1 = async_compile.triton('triton_poi_fused__log_softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl_math.exp(tmp1)
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp2 + tmp4
tmp7 = tl_math.exp(tmp6)
tmp8 = tmp5 + tmp7
tmp10 = tl_math.exp(tmp9)
tmp11 = tmp8 + tmp10
tmp12 = tl_math.log(tmp11)
tmp13 = tmp0 - tmp12
tl.store(out_ptr0 + (x3), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_2
del primals_3
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, buf0, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_5
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__log_softmax_0.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_1.run(buf2, buf3, 256, grid=grid(256), stream=stream0)
del buf2
return (buf3, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), buf0, buf3, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class Word2Vec(nn.Module):
def __init__(self, features, embedding_size):
super().__init__()
0.5 / embedding_size
self.fc1 = nn.Linear(features, embedding_size)
self.fc2 = nn.Linear(embedding_size, features)
def forward(self, one_hot):
x = self.fc1(one_hot.float())
x = self.fc2(x)
log_softmax = torch.nn.functional.log_softmax(x, dim=1)
return log_softmax
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'features': 4, 'embedding_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__log_softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
@triton.jit
def triton_poi_fused__log_softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl_math.exp(tmp1)
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp2 + tmp4
tmp7 = tl_math.exp(tmp6)
tmp8 = tmp5 + tmp7
tmp10 = tl_math.exp(tmp9)
tmp11 = tmp8 + tmp10
tmp12 = tl_math.log(tmp11)
tmp13 = tmp0 - tmp12
tl.store(out_ptr0 + x3, tmp13, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_1, (64,
4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_2
del primals_3
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, buf0, reinterpret_tensor(primals_4,
(4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_5
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__log_softmax_0[grid(256)](buf1, buf2, 256, XBLOCK=
128, num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf1
triton_poi_fused__log_softmax_1[grid(256)](buf2, buf3, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del buf2
return buf3, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0
), buf0, buf3, primals_4
class Word2VecNew(nn.Module):
def __init__(self, features, embedding_size):
super().__init__()
0.5 / embedding_size
self.fc1 = nn.Linear(features, embedding_size)
self.fc2 = nn.Linear(embedding_size, features)
def forward(self, input_0):
primals_2 = self.fc1.weight
primals_3 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| salmanedhi/NNTI-WS2021-NLP-Project | Word2Vec | false | 4,253 | [
"MIT"
] | 0 | 5b0a8f1258ef4e835a6e647082a8286078a0bdd6 | https://github.com/salmanedhi/NNTI-WS2021-NLP-Project/tree/5b0a8f1258ef4e835a6e647082a8286078a0bdd6 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, features, embedding_size):
super().__init__()
0.5 / embedding_size
self.fc1 = nn.Linear(features, embedding_size)
self.fc2 = nn.Linear(embedding_size, features)
def forward(self, one_hot):
x = self.fc1(one_hot.float())
x = self.fc2(x)
log_softmax = torch.nn.functional.log_softmax(x, dim=1)
return log_softmax
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Beta | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/vl/cvlekk7mg6q4mdm2ff42mapcgym5nn43grvi5dksbd4j5rg2sv57.py
# Topologically Sorted Source Nodes: [softplus, alpha], Original ATen: [aten.softplus, aten.add]
# Source node to ATen node mapping:
# alpha => add
# softplus => exp, gt, log1p, where
# Graph fragment:
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%arg0_1, 20), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%arg0_1,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %arg0_1, %log1p), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%where, 1), kwargs = {})
triton_poi_fused_add_softplus_0 = async_compile.triton('triton_poi_fused_add_softplus_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_softplus_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_softplus_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 20.0
tmp2 = tmp0 > tmp1
tmp3 = tl_math.exp(tmp0)
tmp4 = libdevice.log1p(tmp3)
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tl.store(out_ptr0 + (x0), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [softplus, alpha], Original ATen: [aten.softplus, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_softplus_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
buf1 = empty_strided_cuda((4, 0, 4, 4), (0, 16, 4, 1), torch.float32)
return (buf0, buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.functional as F
import torch.nn.functional as F
class BoundedBeta(torch.distributions.Beta):
def log_prob(self, x):
return super().log_prob((x + 1) / 2)
class Beta(nn.Module):
def __init__(self, action_dim):
super(Beta, self).__init__()
self.action_dim = action_dim
def forward(self, alpha_beta):
alpha = 1 + F.softplus(alpha_beta[:, :self.action_dim])
beta = 1 + F.softplus(alpha_beta[:, self.action_dim:])
return alpha, beta
def sample(self, x, deterministic):
if deterministic is False:
action = self.evaluate(x).sample()
else:
return self.evaluate(x).mean
return 2 * action - 1
def evaluate(self, x):
alpha, beta = self(x)
return BoundedBeta(alpha, beta)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'action_dim': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_softplus_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 20.0
tmp2 = tmp0 > tmp1
tmp3 = tl_math.exp(tmp0)
tmp4 = libdevice.log1p(tmp3)
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tl.store(out_ptr0 + x0, tmp7, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_softplus_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((4, 0, 4, 4), (0, 16, 4, 1), torch.float32)
return buf0, buf1
class BoundedBeta(torch.distributions.Beta):
def log_prob(self, x):
return super().log_prob((x + 1) / 2)
class BetaNew(nn.Module):
def __init__(self, action_dim):
super(BetaNew, self).__init__()
self.action_dim = action_dim
def sample(self, x, deterministic):
if deterministic is False:
action = self.evaluate(x).sample()
else:
return self.evaluate(x).mean
return 2 * action - 1
def evaluate(self, x):
alpha, beta = self(x)
return BoundedBeta(alpha, beta)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0], output[1]
| samarth-robo/apex | Beta | false | 4,254 | [
"MIT"
] | 0 | db24044acacd0fcd006886eb1677eaa2f2beedad | https://github.com/samarth-robo/apex/tree/db24044acacd0fcd006886eb1677eaa2f2beedad | import torch
import torch.nn as nn
import torch.functional as F
import torch.nn.functional as F
class BoundedBeta(torch.distributions.Beta):
def log_prob(self, x):
return super().log_prob((x + 1) / 2)
class Model(nn.Module):
def __init__(self, action_dim):
super().__init__()
self.action_dim = action_dim
def forward(self, alpha_beta):
alpha = 1 + F.softplus(alpha_beta[:, :self.action_dim])
beta = 1 + F.softplus(alpha_beta[:, self.action_dim:])
return alpha, beta
def sample(self, x, deterministic):
if deterministic is False:
action = self.evaluate(x).sample()
else:
return self.evaluate(x).mean
return 2 * action - 1
def evaluate(self, x):
alpha, beta = self(x)
return BoundedBeta(alpha, beta)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
Actor | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ix/cixxyusyg44s2hkoufcgbrv3ix5ookwqjl4ia3xkv7bdqi4yrzus.py
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# out_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 25600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 400
x2 = xindex % 1600
x3 = (xindex // 1600)
tmp0 = tl.load(in_out_ptr0 + (x4), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x4), tmp4, xmask)
tl.store(out_ptr0 + (x2 + (1664*x3)), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/op/coptu6xep3awc4lajb4xivopppqmjtx3zy7ebtazm45rqvyeknds.py
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# out_3 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 19200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 300
x2 = (xindex // 1200)
x3 = xindex % 1200
tmp0 = tl.load(in_ptr0 + (x4), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(out_ptr0 + (x3 + (1216*x2)), tmp4, xmask)
tl.store(out_ptr1 + (x3 + (1280*x2)), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/as/casrc7bf7ghsendgi7tkqxk3hj4ic6aqb4rmkxzuk5dhbidznia7.py
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.relu, aten.view]
# Source node to ATen node mapping:
# out_3 => relu_1
# out_4 => view_4
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %view_4 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%relu_1, [64, 300]), kwargs = {})
triton_poi_fused_relu_view_2 = async_compile.triton('triton_poi_fused_relu_view_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_view_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_view_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 19200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 300
x1 = (xindex // 300)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (300*(x1 % 4)) + (1216*(x1 // 4))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xp/cxpqywcqam7evubfwwa5zmt733w2zov6otomgqgpramgjdsnjg5k.py
# Topologically Sorted Source Nodes: [out_5], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# out_5 => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_5,), kwargs = {})
triton_poi_fused_sigmoid_3 = async_compile.triton('triton_poi_fused_sigmoid_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_3', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_3(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (400, 4), (4, 1))
assert_size_stride(primals_2, (400, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (300, 400), (400, 1))
assert_size_stride(primals_5, (300, ), (1, ))
assert_size_stride(primals_6, (4, 300), (300, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 400), (400, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 400), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 400), (6400, 1600, 400, 1), 0); del buf0 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 400), (6656, 1664, 400, 1), torch.bool)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf8, 25600, grid=grid(25600), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 300), (300, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 400), (400, 1), 0), reinterpret_tensor(primals_4, (400, 300), (1, 400), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4, 300), (4864, 1216, 300, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 300), (5120, 1280, 300, 1), torch.bool)
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf2, primals_5, buf3, buf7, 19200, grid=grid(19200), stream=stream0)
del primals_5
buf4 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.relu, aten.view]
triton_poi_fused_relu_view_2.run(buf3, buf4, 19200, grid=grid(19200), stream=stream0)
del buf3
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf4, reinterpret_tensor(primals_6, (300, 4), (1, 300), 0), out=buf5)
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [out_5], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_3.run(buf6, primals_7, 256, grid=grid(256), stream=stream0)
del primals_7
return (buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 400), (400, 1), 0), buf4, buf6, primals_6, buf7, primals_4, buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((400, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((400, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((300, 400), (400, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((300, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 300), (300, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
import torch.onnx
import torch.testing
class Actor(nn.Module):
def __init__(self, nb_states, nb_actions, hidden1=400, hidden2=300):
super(Actor, self).__init__()
self.fc1 = nn.Linear(nb_states, hidden1)
self.fc2 = nn.Linear(hidden1, hidden2)
self.fc3 = nn.Linear(hidden2, nb_actions)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, x):
out = self.fc1(x)
out = self.relu(out)
out = self.fc2(out)
out = self.relu(out)
out = self.fc3(out)
out = self.sigmoid(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nb_states': 4, 'nb_actions': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
import torch.onnx
import torch.testing
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 25600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 400
x2 = xindex % 1600
x3 = xindex // 1600
tmp0 = tl.load(in_out_ptr0 + x4, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x4, tmp4, xmask)
tl.store(out_ptr0 + (x2 + 1664 * x3), tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 19200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 300
x2 = xindex // 1200
x3 = xindex % 1200
tmp0 = tl.load(in_ptr0 + x4, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(out_ptr0 + (x3 + 1216 * x2), tmp4, xmask)
tl.store(out_ptr1 + (x3 + 1280 * x2), tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_view_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 19200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 300
x1 = xindex // 300
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 300 * (x1 % 4) + 1216 * (x1 // 4)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_sigmoid_3(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (400, 4), (4, 1))
assert_size_stride(primals_2, (400,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (300, 400), (400, 1))
assert_size_stride(primals_5, (300,), (1,))
assert_size_stride(primals_6, (4, 300), (300, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 400), (400, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 400), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 400), (6400, 1600, 400, 1), 0
)
del buf0
buf8 = empty_strided_cuda((4, 4, 4, 400), (6656, 1664, 400, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(25600)](buf1,
primals_2, buf8, 25600, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 300), (300, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 400), (400, 1), 0),
reinterpret_tensor(primals_4, (400, 300), (1, 400), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4, 300), (4864, 1216, 300, 1),
torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 300), (5120, 1280, 300, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(19200)](buf2,
primals_5, buf3, buf7, 19200, XBLOCK=128, num_warps=4, num_stages=1
)
del primals_5
buf4 = buf2
del buf2
triton_poi_fused_relu_view_2[grid(19200)](buf3, buf4, 19200, XBLOCK
=256, num_warps=4, num_stages=1)
del buf3
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(buf4, reinterpret_tensor(primals_6, (300, 4), (1,
300), 0), out=buf5)
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused_sigmoid_3[grid(256)](buf6, primals_7, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_7
return buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 400), (400, 1), 0
), buf4, buf6, primals_6, buf7, primals_4, buf8
class ActorNew(nn.Module):
def __init__(self, nb_states, nb_actions, hidden1=400, hidden2=300):
super(ActorNew, self).__init__()
self.fc1 = nn.Linear(nb_states, hidden1)
self.fc2 = nn.Linear(hidden1, hidden2)
self.fc3 = nn.Linear(hidden2, nb_actions)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| saman-aghazadeh/distiller | Actor | false | 4,255 | [
"Apache-2.0"
] | 0 | 7e8d3e6193c807f7c55d8453f64e1bc3c02eee30 | https://github.com/saman-aghazadeh/distiller/tree/7e8d3e6193c807f7c55d8453f64e1bc3c02eee30 | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
import torch.onnx
import torch.testing
class Model(nn.Module):
def __init__(self, nb_states, nb_actions, hidden1=400, hidden2=300):
super().__init__()
self.fc1 = nn.Linear(nb_states, hidden1)
self.fc2 = nn.Linear(hidden1, hidden2)
self.fc3 = nn.Linear(hidden2, nb_actions)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, x):
out = self.fc1(x)
out = self.relu(out)
out = self.fc2(out)
out = self.relu(out)
out = self.fc3(out)
out = self.sigmoid(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Beta2 | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/up/cup6o5cbzbunxrgsjove7ghgc4dkxlfpf5zncfpxjlrgo5zjrjpe.py
# Topologically Sorted Source Nodes: [mean, sub_2, exp, var, truediv_1, mul_1, sub_3, sub, truediv, pow_2, mul, alpha, beta], Original ATen: [aten.sigmoid, aten.rsub, aten.exp, aten.pow, aten.div, aten.mul, aten.sub]
# Source node to ATen node mapping:
# alpha => sub_1
# beta => sub_4
# exp => exp
# mean => sigmoid
# mul => mul
# mul_1 => mul_1
# pow_2 => pow_2
# sub => sub
# sub_2 => sub_2
# sub_3 => sub_3
# truediv => div
# truediv_1 => div_1
# var => pow_1
# Graph fragment:
# %sigmoid : [num_users=5] = call_function[target=torch.ops.aten.sigmoid.default](args = (%arg0_1,), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%arg1_1,), kwargs = {})
# %pow_1 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%exp, 2), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_2, %pow_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_1, %sigmoid), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_1, 1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %pow_1), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sigmoid, 2), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %pow_2), kwargs = {})
# %sub_1 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %sigmoid), kwargs = {})
# %sub_4 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_3, %sub_1), kwargs = {})
triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0 = async_compile.triton('triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp4 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.sigmoid(tmp0)
tmp2 = 1.0
tmp3 = tmp2 - tmp1
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp5 * tmp5
tmp7 = tmp3 / tmp6
tmp8 = tmp1 * tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp9 - tmp1
tmp11 = tmp7 * tmp1
tmp12 = tmp11 - tmp2
tmp13 = tmp12 - tmp10
tl.store(out_ptr0 + (x2), tmp10, xmask)
tl.store(out_ptr1 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (1, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean, sub_2, exp, var, truediv_1, mul_1, sub_3, sub, truediv, pow_2, mul, alpha, beta], Original ATen: [aten.sigmoid, aten.rsub, aten.exp, aten.pow, aten.div, aten.mul, aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0.run(arg0_1, arg1_1, buf0, buf1, 256, grid=grid(256), stream=stream0)
del arg0_1
del arg1_1
return (buf0, buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
import torch.nn as nn
class BoundedBeta(torch.distributions.Beta):
def log_prob(self, x):
return super().log_prob((x + 1) / 2)
class Beta2(nn.Module):
def __init__(self, action_dim, init_std=0.25, learn_std=False):
super(Beta2, self).__init__()
assert init_std < 0.5, 'Beta distribution has a max std dev of 0.5'
self.action_dim = action_dim
self.logstd = nn.Parameter(torch.ones(1, action_dim) * np.log(
init_std), requires_grad=learn_std)
self.learn_std = learn_std
def forward(self, x):
mean = torch.sigmoid(x)
var = self.logstd.exp().pow(2)
"""
alpha = ((1 - mu) / sigma^2 - 1 / mu) * mu^2
beta = alpha * (1 / mu - 1)
Implemented slightly differently for numerical stability.
"""
alpha = (1 - mean) / var * mean.pow(2) - mean
beta = (1 - mean) / var * mean - 1 - alpha
return alpha, beta
def sample(self, x, deterministic):
if deterministic is False:
action = self.evaluate(x).sample()
else:
return self.evaluate(x).mean
return 2 * action - 1
def evaluate(self, x):
alpha, beta = self(x)
return BoundedBeta(alpha, beta)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'action_dim': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
import numpy as np
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0(in_ptr0, in_ptr1,
out_ptr0, out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp4 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.sigmoid(tmp0)
tmp2 = 1.0
tmp3 = tmp2 - tmp1
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp5 * tmp5
tmp7 = tmp3 / tmp6
tmp8 = tmp1 * tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp9 - tmp1
tmp11 = tmp7 * tmp1
tmp12 = tmp11 - tmp2
tmp13 = tmp12 - tmp10
tl.store(out_ptr0 + x2, tmp10, xmask)
tl.store(out_ptr1 + x2, tmp13, xmask)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (1, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_div_exp_mul_pow_rsub_sigmoid_sub_0[grid(256)](arg0_1,
arg1_1, buf0, buf1, 256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0, buf1
class BoundedBeta(torch.distributions.Beta):
def log_prob(self, x):
return super().log_prob((x + 1) / 2)
class Beta2New(nn.Module):
def __init__(self, action_dim, init_std=0.25, learn_std=False):
super(Beta2New, self).__init__()
assert init_std < 0.5, 'Beta distribution has a max std dev of 0.5'
self.action_dim = action_dim
self.logstd = nn.Parameter(torch.ones(1, action_dim) * np.log(
init_std), requires_grad=learn_std)
self.learn_std = learn_std
def sample(self, x, deterministic):
if deterministic is False:
action = self.evaluate(x).sample()
else:
return self.evaluate(x).mean
return 2 * action - 1
def evaluate(self, x):
alpha, beta = self(x)
return BoundedBeta(alpha, beta)
def forward(self, input_0):
arg1_1 = self.logstd
arg0_1 = input_0
output = call([arg0_1, arg1_1])
return output[0], output[1]
| samarth-robo/apex | Beta2 | false | 4,256 | [
"MIT"
] | 0 | db24044acacd0fcd006886eb1677eaa2f2beedad | https://github.com/samarth-robo/apex/tree/db24044acacd0fcd006886eb1677eaa2f2beedad | import torch
import numpy as np
import torch.nn as nn
class BoundedBeta(torch.distributions.Beta):
def log_prob(self, x):
return super().log_prob((x + 1) / 2)
class Model(nn.Module):
def __init__(self, action_dim, init_std=0.25, learn_std=False):
super().__init__()
assert init_std < 0.5, 'Beta distribution has a max std dev of 0.5'
self.action_dim = action_dim
self.logstd = nn.Parameter(torch.ones(1, action_dim) * np.log(
init_std), requires_grad=learn_std)
self.learn_std = learn_std
def forward(self, x):
mean = torch.sigmoid(x)
var = self.logstd.exp().pow(2)
"""
alpha = ((1 - mu) / sigma^2 - 1 / mu) * mu^2
beta = alpha * (1 / mu - 1)
Implemented slightly differently for numerical stability.
"""
alpha = (1 - mean) / var * mean.pow(2) - mean
beta = (1 - mean) / var * mean - 1 - alpha
return alpha, beta
def sample(self, x, deterministic):
if deterministic is False:
action = self.evaluate(x).sample()
else:
return self.evaluate(x).mean
return 2 * action - 1
def evaluate(self, x):
alpha, beta = self(x)
return BoundedBeta(alpha, beta)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
BertPooler2 | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/yj/cyjqjllbc3bikte2womjnjqgo2d7wrcb5dt3tpl2sgyogireta4f.py
# Topologically Sorted Source Nodes: [pooled_output], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# pooled_output => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%select,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/2g/c2gw7362i2a6wsfdx2sxyywx4o6ronjg6goebvdn44w6gpjsxpbc.py
# Topologically Sorted Source Nodes: [pooled_output, pooled_output_1], Original ATen: [aten.add, aten.tanh]
# Source node to ATen node mapping:
# pooled_output => add
# pooled_output_1 => tanh
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_3), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add,), kwargs = {})
triton_poi_fused_add_tanh_1 = async_compile.triton('triton_poi_fused_add_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_tanh_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pooled_output], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(primals_1, buf0, 64, grid=grid(64), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pooled_output], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf1)
del primals_2
buf2 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [pooled_output, pooled_output_1], Original ATen: [aten.add, aten.tanh]
triton_poi_fused_add_tanh_1.run(buf2, primals_3, 64, grid=grid(64), stream=stream0)
del primals_3
return (buf2, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import torch
from torch import nn
import torch.nn.parallel
import torch.optim
from torch.utils.data import *
import torch.nn.functional
class BertPooler2(nn.Module):
def __init__(self, config):
super(BertPooler2, self).__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.activation = nn.Tanh()
def forward(self, hidden_states):
first_token_tensor = hidden_states[:, 1]
pooled_output = self.dense(first_token_tensor)
pooled_output = self.activation(pooled_output)
return pooled_output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
import torch.nn.parallel
import torch.optim
from torch.utils.data import *
import torch.nn.functional
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_add_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64)](primals_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf1)
del primals_2
buf2 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0)
del buf1
triton_poi_fused_add_tanh_1[grid(64)](buf2, primals_3, 64, XBLOCK=
64, num_warps=1, num_stages=1)
del primals_3
return buf2, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), buf2
class BertPooler2New(nn.Module):
def __init__(self, config):
super(BertPooler2New, self).__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.activation = nn.Tanh()
def forward(self, input_0):
primals_2 = self.dense.weight
primals_3 = self.dense.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| samuelyu2002/PACS | BertPooler2 | false | 4,257 | [
"MIT"
] | 0 | 5010b2f0d20933b0647e3d6230d673e1830249ec | https://github.com/samuelyu2002/PACS/tree/5010b2f0d20933b0647e3d6230d673e1830249ec | from _paritybench_helpers import _mock_config
import torch
from torch import nn
import torch.nn.parallel
import torch.optim
from torch.utils.data import *
import torch.nn.functional
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.activation = nn.Tanh()
def forward(self, hidden_states):
first_token_tensor = hidden_states[:, 1]
pooled_output = self.dense(first_token_tensor)
pooled_output = self.activation(pooled_output)
return pooled_output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ModelWithDuplicates | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/x7/cx7zib5vfcs4tjugjecjyojpxio3h7wkcy5bqp7pc5phvne4zdgj.py
# Topologically Sorted Source Nodes: [x, x_1, x_2], Original ATen: [aten.convolution, aten.relu, aten.tanh, aten.threshold_backward]
# Source node to ATen node mapping:
# x => convolution
# x_1 => relu
# x_2 => tanh
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%relu,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_convolution_relu_tanh_threshold_backward_0 = async_compile.triton('triton_poi_fused_convolution_relu_tanh_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_tanh_threshold_backward_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_tanh_threshold_backward_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 144000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 3600) % 10
x0 = xindex % 3600
x4 = (xindex // 3600)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = libdevice.tanh(tmp4)
tmp6 = 0.0
tmp7 = tmp4 <= tmp6
tl.store(out_ptr0 + (x3), tmp5, xmask)
tl.store(out_ptr1 + (x0 + (3712*x4)), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/sl/cslyor46ejkl5lvclqvfd2qnnvpo2y3hutdhtpmver5xbwv2l3ek.py
# Topologically Sorted Source Nodes: [x_3, x_4, x_5], Original ATen: [aten.convolution, aten.relu, aten.tanh, aten.threshold_backward]
# Source node to ATen node mapping:
# x_3 => convolution_1
# x_4 => relu_1
# x_5 => tanh_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%tanh, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_1,), kwargs = {})
# %tanh_1 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%relu_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_convolution_relu_tanh_threshold_backward_1 = async_compile.triton('triton_poi_fused_convolution_relu_tanh_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[524288],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_tanh_threshold_backward_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_tanh_threshold_backward_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 269120
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 3364) % 20
x0 = xindex % 3364
x4 = (xindex // 3364)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = libdevice.tanh(tmp4)
tmp6 = 0.0
tmp7 = tmp4 <= tmp6
tl.store(out_ptr0 + (x3), tmp5, xmask)
tl.store(out_ptr1 + (x0 + (3456*x4)), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (10, 3, 5, 5), (75, 25, 5, 1))
assert_size_stride(primals_2, (10, ), (1, ))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (20, 10, 3, 3), (90, 9, 3, 1))
assert_size_stride(primals_5, (20, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 10, 60, 60), (36000, 3600, 60, 1))
buf1 = empty_strided_cuda((4, 10, 60, 60), (36000, 3600, 60, 1), torch.float32)
buf5 = empty_strided_cuda((4, 10, 60, 60), (37120, 3712, 60, 1), torch.bool)
# Topologically Sorted Source Nodes: [x, x_1, x_2], Original ATen: [aten.convolution, aten.relu, aten.tanh, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_tanh_threshold_backward_0.run(buf0, primals_2, buf1, buf5, 144000, grid=grid(144000), stream=stream0)
del buf0
del primals_2
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 20, 58, 58), (67280, 3364, 58, 1))
buf3 = empty_strided_cuda((4, 20, 58, 58), (67280, 3364, 58, 1), torch.float32)
buf4 = empty_strided_cuda((4, 20, 58, 58), (69120, 3456, 58, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_3, x_4, x_5], Original ATen: [aten.convolution, aten.relu, aten.tanh, aten.threshold_backward]
triton_poi_fused_convolution_relu_tanh_threshold_backward_1.run(buf2, primals_5, buf3, buf4, 269120, grid=grid(269120), stream=stream0)
del buf2
del primals_5
return (buf3, primals_1, primals_3, primals_4, buf1, buf3, buf4, buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((10, 3, 5, 5), (75, 25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((10, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 64, 64), (12288, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((20, 10, 3, 3), (90, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((20, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from collections import OrderedDict
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
import torch.onnx
import torch.testing
class ModelWithDuplicates(nn.Module):
def __init__(self):
super(ModelWithDuplicates, self).__init__()
self.conv1 = nn.Conv2d(3, 10, 5)
self.post_conv1 = nn.ModuleList([nn.ReLU(), nn.Tanh()])
self.conv2 = nn.Conv2d(10, 20, 3)
self.post_conv2 = self.post_conv1
self.expected_mlist_to_dmlist = OrderedDict([('post_conv1', [
'post_conv1']), ('post_conv2', ['post_conv2'])])
self.expected_list_contents_name_changes = OrderedDict([(
'post_conv1.0', 'post_conv1_0'), ('post_conv1.1',
'post_conv1_1'), ('post_conv2.0', 'post_conv2_0'), (
'post_conv2.1', 'post_conv2_1')])
def forward(self, x):
x = self.conv1(x)
for m in self.post_conv1:
x = m(x)
x = self.conv2(x)
for m in self.post_conv2:
x = m(x)
return x
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
from collections import OrderedDict
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
import torch.onnx
import torch.testing
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_convolution_relu_tanh_threshold_backward_0(in_ptr0,
in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 144000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 3600 % 10
x0 = xindex % 3600
x4 = xindex // 3600
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = libdevice.tanh(tmp4)
tmp6 = 0.0
tmp7 = tmp4 <= tmp6
tl.store(out_ptr0 + x3, tmp5, xmask)
tl.store(out_ptr1 + (x0 + 3712 * x4), tmp7, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_tanh_threshold_backward_1(in_ptr0,
in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 269120
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 3364 % 20
x0 = xindex % 3364
x4 = xindex // 3364
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = libdevice.tanh(tmp4)
tmp6 = 0.0
tmp7 = tmp4 <= tmp6
tl.store(out_ptr0 + x3, tmp5, xmask)
tl.store(out_ptr1 + (x0 + 3456 * x4), tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (10, 3, 5, 5), (75, 25, 5, 1))
assert_size_stride(primals_2, (10,), (1,))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (20, 10, 3, 3), (90, 9, 3, 1))
assert_size_stride(primals_5, (20,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 10, 60, 60), (36000, 3600, 60, 1))
buf1 = empty_strided_cuda((4, 10, 60, 60), (36000, 3600, 60, 1),
torch.float32)
buf5 = empty_strided_cuda((4, 10, 60, 60), (37120, 3712, 60, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_convolution_relu_tanh_threshold_backward_0[grid(
144000)](buf0, primals_2, buf1, buf5, 144000, XBLOCK=1024,
num_warps=4, num_stages=1)
del buf0
del primals_2
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 20, 58, 58), (67280, 3364, 58, 1))
buf3 = empty_strided_cuda((4, 20, 58, 58), (67280, 3364, 58, 1),
torch.float32)
buf4 = empty_strided_cuda((4, 20, 58, 58), (69120, 3456, 58, 1),
torch.bool)
triton_poi_fused_convolution_relu_tanh_threshold_backward_1[grid(
269120)](buf2, primals_5, buf3, buf4, 269120, XBLOCK=512,
num_warps=8, num_stages=1)
del buf2
del primals_5
return buf3, primals_1, primals_3, primals_4, buf1, buf3, buf4, buf5
class ModelWithDuplicatesNew(nn.Module):
def __init__(self):
super(ModelWithDuplicatesNew, self).__init__()
self.conv1 = nn.Conv2d(3, 10, 5)
self.post_conv1 = nn.ModuleList([nn.ReLU(), nn.Tanh()])
self.conv2 = nn.Conv2d(10, 20, 3)
self.post_conv2 = self.post_conv1
self.expected_mlist_to_dmlist = OrderedDict([('post_conv1', [
'post_conv1']), ('post_conv2', ['post_conv2'])])
self.expected_list_contents_name_changes = OrderedDict([(
'post_conv1.0', 'post_conv1_0'), ('post_conv1.1',
'post_conv1_1'), ('post_conv2.0', 'post_conv2_0'), (
'post_conv2.1', 'post_conv2_1')])
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| saman-aghazadeh/distiller | ModelWithDuplicates | false | 4,258 | [
"Apache-2.0"
] | 0 | 7e8d3e6193c807f7c55d8453f64e1bc3c02eee30 | https://github.com/saman-aghazadeh/distiller/tree/7e8d3e6193c807f7c55d8453f64e1bc3c02eee30 | import torch
from collections import OrderedDict
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
import torch.onnx
import torch.testing
class Model(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(3, 10, 5)
self.post_conv1 = nn.ModuleList([nn.ReLU(), nn.Tanh()])
self.conv2 = nn.Conv2d(10, 20, 3)
self.post_conv2 = self.post_conv1
self.expected_mlist_to_dmlist = OrderedDict([('post_conv1', [
'post_conv1']), ('post_conv2', ['post_conv2'])])
self.expected_list_contents_name_changes = OrderedDict([(
'post_conv1.0', 'post_conv1_0'), ('post_conv1.1',
'post_conv1_1'), ('post_conv2.0', 'post_conv2_0'), (
'post_conv2.1', 'post_conv2_1')])
def forward(self, x):
x = self.conv1(x)
for m in self.post_conv1:
x = m(x)
x = self.conv2(x)
for m in self.post_conv2:
x = m(x)
return x
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return []
|
MySimpleNet | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/nu/cnuuaznpt4szfn74bn46qfjkdypvlkfa5x44ywjpperdjt2a66rj.py
# Topologically Sorted Source Nodes: [X], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# X => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 640
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 10
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/mp/cmpdsbnpgfsr7uwb7env74mojrq3nlzieqot6rnnkfpbzkkensbi.py
# Topologically Sorted Source Nodes: [X_2], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# X_2 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/z5/cz5xs7y3thsep5yn6qoths757rduuevog6mtea3nqr4nwnh2olnx.py
# Topologically Sorted Source Nodes: [X_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# X_3 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 32)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/nv/cnvo7i3x3dm4mdtrcmoddo2p4odl6hgahimnieftjxkqwe7ehw54.py
# Topologically Sorted Source Nodes: [X_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# X_3 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 32)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (10, 4), (4, 1))
assert_size_stride(primals_2, (10, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 10), (10, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (2, 4), (4, 1))
assert_size_stride(primals_7, (2, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 10), (10, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 10), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 10), (160, 40, 10, 1), 0); del buf0 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 10), (160, 40, 10, 1), torch.bool)
# Topologically Sorted Source Nodes: [X], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf8, 640, grid=grid(640), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 10), (10, 1), 0), reinterpret_tensor(primals_4, (10, 4), (1, 10), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [X_2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf7, 256, grid=grid(256), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 2), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 2), (32, 8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [X_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf4, buf5, 128, grid=grid(128), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 2), (32, 8, 2, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [X_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf5, buf6, 128, grid=grid(128), stream=stream0)
del buf5
return (buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 10), (10, 1), 0), reinterpret_tensor(buf3, (64, 4), (4, 1), 0), buf6, primals_6, buf7, primals_4, buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((10, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((10, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 10), (10, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((2, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((2, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
from torch import nn
class MySimpleNet(nn.Module):
"""
Very simple 2-layer net, slightly adapted from the docs:
https://skorch.readthedocs.io/en/stable/user/quickstart.html
"""
def __init__(self, num_in, num_feat, num_hidden=10, nonlin=F.relu):
super(MySimpleNet, self).__init__()
self.dense0 = nn.Linear(num_in, num_hidden)
self.nonlin = nonlin
self.dropout = nn.Dropout(0.5)
self.dense1 = nn.Linear(num_hidden, num_feat)
self.output = nn.Linear(num_feat, 2)
def forward(self, X, **kwargs):
X = self.nonlin(self.dense0(X))
X = self.dropout(X)
X = F.relu(self.dense1(X))
X = F.softmax(self.output(X))
return X
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_in': 4, 'num_feat': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn.functional as F
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 640
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 10
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = xindex // 32
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = xindex // 32
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (10, 4), (4, 1))
assert_size_stride(primals_2, (10,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 10), (10, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (2, 4), (4, 1))
assert_size_stride(primals_7, (2,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 10), (10, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 10), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 10), (160, 40, 10, 1), 0)
del buf0
buf8 = empty_strided_cuda((4, 4, 4, 10), (160, 40, 10, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(640)](buf1,
primals_2, buf8, 640, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 10), (10, 1), 0),
reinterpret_tensor(primals_4, (10, 4), (1, 10), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(256)](buf3,
primals_5, buf7, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 2), (1, 4), 0),
alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 2), (32, 8, 2, 1), torch.float32)
triton_poi_fused__softmax_2[grid(128)](buf4, buf5, 128, XBLOCK=128,
num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 2), (32, 8, 2, 1), 0)
del buf4
triton_poi_fused__softmax_3[grid(128)](buf5, buf6, 128, XBLOCK=128,
num_warps=4, num_stages=1)
del buf5
return buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 10), (10, 1), 0), reinterpret_tensor(
buf3, (64, 4), (4, 1), 0), buf6, primals_6, buf7, primals_4, buf8
class MySimpleNetNew(nn.Module):
"""
Very simple 2-layer net, slightly adapted from the docs:
https://skorch.readthedocs.io/en/stable/user/quickstart.html
"""
def __init__(self, num_in, num_feat, num_hidden=10, nonlin=F.relu):
super(MySimpleNetNew, self).__init__()
self.dense0 = nn.Linear(num_in, num_hidden)
self.nonlin = nonlin
self.dropout = nn.Dropout(0.5)
self.dense1 = nn.Linear(num_hidden, num_feat)
self.output = nn.Linear(num_feat, 2)
def forward(self, input_0):
primals_1 = self.dense0.weight
primals_2 = self.dense0.bias
primals_4 = self.dense1.weight
primals_5 = self.dense1.bias
primals_6 = self.output.weight
primals_7 = self.output.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| samxu0823/anfis-pytorch | MySimpleNet | false | 4,259 | [
"MIT"
] | 0 | b4ec3f0e8259963800e9e0a2904a580d1e56cc1c | https://github.com/samxu0823/anfis-pytorch/tree/b4ec3f0e8259963800e9e0a2904a580d1e56cc1c | import torch
import torch.nn.functional as F
from torch import nn
class Model(nn.Module):
"""
Very simple 2-layer net, slightly adapted from the docs:
https://skorch.readthedocs.io/en/stable/user/quickstart.html
"""
def __init__(self, num_in, num_feat, num_hidden=10, nonlin=F.relu):
super().__init__()
self.dense0 = nn.Linear(num_in, num_hidden)
self.nonlin = nonlin
self.dropout = nn.Dropout(0.5)
self.dense1 = nn.Linear(num_hidden, num_feat)
self.output = nn.Linear(num_feat, 2)
def forward(self, X, **kwargs):
X = self.nonlin(self.dense0(X))
X = self.dropout(X)
X = F.relu(self.dense1(X))
X = F.softmax(self.output(X))
return X
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
BahdanauAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ao/caoovxtqrx42gvkmjirowqmmbh6kppvfh5ebrzzv4kzkgwm2umii.py
# Topologically Sorted Source Nodes: [processed_query], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# processed_query => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1)), xmask)
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/bg/cbgmsaps4ljzc6rkbd4imsj3jo73tgvkd46dy7obklnnvintmaea.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.mv]
# Source node to ATen node mapping:
# out => mul, sum_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_4, %primals_5), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1]), kwargs = {})
triton_poi_fused_mv_1 = async_compile.triton('triton_poi_fused_mv_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mv_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 12, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mv_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*(x0 // 4)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + ((4*(x0 % 4)) + (16*(x0 // 16))), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (1 + (4*(x0 // 4))), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (1 + (4*(x0 % 4)) + (16*(x0 // 16))), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr2 + (1))
tmp12 = tl.broadcast_to(tmp11, [XBLOCK])
tmp15 = tl.load(in_ptr0 + (2 + (4*(x0 // 4))), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (2 + (4*(x0 % 4)) + (16*(x0 // 16))), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr2 + (2))
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp23 = tl.load(in_ptr0 + (3 + (4*(x0 // 4))), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr1 + (3 + (4*(x0 % 4)) + (16*(x0 // 16))), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr2 + (3))
tmp28 = tl.broadcast_to(tmp27, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tmp6 = tmp3 * tmp5
tmp9 = tmp7 + tmp8
tmp10 = libdevice.tanh(tmp9)
tmp13 = tmp10 * tmp12
tmp14 = tmp6 + tmp13
tmp17 = tmp15 + tmp16
tmp18 = libdevice.tanh(tmp17)
tmp21 = tmp18 * tmp20
tmp22 = tmp14 + tmp21
tmp25 = tmp23 + tmp24
tmp26 = libdevice.tanh(tmp25)
tmp29 = tmp26 * tmp28
tmp30 = tmp22 + tmp29
tl.store(out_ptr0 + (x0), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/hg/chg3iq6bscxmmxv5f7tuzgwycb4mgrimwfhv2nauw5rj4tt5cmv2.py
# Topologically Sorted Source Nodes: [scores_normalized], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# scores_normalized => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/zu/czuvep3dmpmqmhiiliwubh4ghdt2qr27va67sszkua7trziinwov.py
# Topologically Sorted Source Nodes: [scores_normalized], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# scores_normalized => div, sum_2
# Graph fragment:
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div : [num_users=3] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_2), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [processed_query], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(primals_2, buf0, 64, grid=grid(64), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [processed_query], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [processed_key], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(primals_1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [processed_key], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
del primals_4
buf4 = empty_strided_cuda((64, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.mv]
triton_poi_fused_mv_1.run(buf1, buf3, primals_5, buf4, 64, grid=grid(64), stream=stream0)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [scores_normalized], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf4, buf5, 64, grid=grid(64), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4), (16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [scores_normalized], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf5, buf6, 64, grid=grid(64), stream=stream0)
buf7 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [context], Original ATen: [aten.bmm]
extern_kernels.bmm(buf6, reinterpret_tensor(primals_1, (4, 4, 4), (4, 16, 1), 0), out=buf7)
return (reinterpret_tensor(buf7, (4, 4, 4), (4, 16, 1), 0), reinterpret_tensor(buf6, (4, 4, 4), (4, 16, 1), 0), primals_5, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), buf1, reinterpret_tensor(buf2, (16, 4), (4, 1), 0), buf3, buf6, reinterpret_tensor(primals_1, (4, 4, 4), (4, 1, 16), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
from torch.nn.parameter import Parameter
import torch.onnx
import torch.testing
class EltwiseAdd(nn.Module):
def __init__(self, inplace=False):
"""Element-wise addition"""
super().__init__()
self.inplace = inplace
def forward(self, *input):
res = input[0]
if self.inplace:
for t in input[1:]:
res += t
else:
for t in input[1:]:
res = res + t
return res
class EltwiseMult(nn.Module):
def __init__(self, inplace=False):
"""Element-wise multiplication"""
super().__init__()
self.inplace = inplace
def forward(self, *input):
res = input[0]
if self.inplace:
for t in input[1:]:
res *= t
else:
for t in input[1:]:
res = res * t
return res
class Matmul(nn.Module):
"""
A wrapper module for matmul operation between 2 tensors.
"""
def __init__(self):
super(Matmul, self).__init__()
def forward(self, a: 'torch.Tensor', b: 'torch.Tensor'):
return a.matmul(b)
class BatchMatmul(nn.Module):
"""
A wrapper module for torch.bmm operation between 2 tensors.
"""
def __init__(self):
super(BatchMatmul, self).__init__()
def forward(self, a: 'torch.Tensor', b: 'torch.Tensor'):
return torch.bmm(a, b)
class BahdanauAttention(nn.Module):
"""
It should be very similar to tf.contrib.seq2seq.BahdanauAttention
"""
def __init__(self, query_size, key_size, num_units, normalize=False,
dropout=0, batch_first=False):
super(BahdanauAttention, self).__init__()
self.normalize = normalize
self.batch_first = batch_first
self.num_units = num_units
self.linear_q = nn.Linear(query_size, num_units, bias=False)
self.linear_k = nn.Linear(key_size, num_units, bias=False)
self.linear_att = Parameter(torch.Tensor(num_units))
self.dropout = nn.Dropout(dropout)
self.mask = None
self.eltwiseadd_qk = EltwiseAdd()
self.eltwiseadd_norm_bias = EltwiseAdd()
self.eltwisemul_norm_scaler = EltwiseMult()
self.tanh = nn.Tanh()
self.matmul_score = Matmul()
self.softmax_att = nn.Softmax(dim=-1)
self.context_matmul = BatchMatmul()
if self.normalize:
self.normalize_scalar = Parameter(torch.Tensor(1))
self.normalize_bias = Parameter(torch.Tensor(num_units))
else:
self.register_parameter('normalize_scalar', None)
self.register_parameter('normalize_bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1.0 / math.sqrt(self.num_units)
self.linear_att.data.uniform_(-stdv, stdv)
if self.normalize:
self.normalize_scalar.data.fill_(stdv)
self.normalize_bias.data.zero_()
def set_mask(self, context_len, context):
"""
sets self.mask which is applied before softmax
ones for inactive context fields, zeros for active context fields
:param context_len: b
:param context: if batch_first: (b x t_k x n) else: (t_k x b x n)
self.mask: (b x t_k)
"""
if self.batch_first:
max_len = context.size(1)
else:
max_len = context.size(0)
indices = torch.arange(0, max_len, dtype=torch.int64, device=
context.device)
self.mask = indices >= context_len.unsqueeze(1)
def calc_score(self, att_query, att_keys):
"""
Calculate Bahdanau score
:param att_query: b x t_q x n
:param att_keys: b x t_k x n
return b x t_q x t_k scores
"""
b, t_k, n = att_keys.size()
t_q = att_query.size(1)
att_query = att_query.unsqueeze(2).expand(b, t_q, t_k, n)
att_keys = att_keys.unsqueeze(1).expand(b, t_q, t_k, n)
sum_qk = self.eltwiseadd_qk(att_query, att_keys)
if self.normalize:
sum_qk = self.eltwiseadd_norm_bias(sum_qk, self.normalize_bias)
tmp = self.linear_att
linear_att = tmp / tmp.norm()
linear_att = linear_att
linear_att = self.eltwisemul_norm_scaler(linear_att, self.
normalize_scalar)
else:
linear_att = self.linear_att
out = self.matmul_score(self.tanh(sum_qk), linear_att)
return out
def forward(self, query, keys):
"""
:param query: if batch_first: (b x t_q x n) else: (t_q x b x n)
:param keys: if batch_first: (b x t_k x n) else (t_k x b x n)
:returns: (context, scores_normalized)
context: if batch_first: (b x t_q x n) else (t_q x b x n)
scores_normalized: if batch_first (b x t_q x t_k) else (t_q x b x t_k)
"""
if not self.batch_first:
keys = keys.transpose(0, 1)
if query.dim() == 3:
query = query.transpose(0, 1)
if query.dim() == 2:
single_query = True
query = query.unsqueeze(1)
else:
single_query = False
b = query.size(0)
t_k = keys.size(1)
t_q = query.size(1)
processed_query = self.linear_q(query)
processed_key = self.linear_k(keys)
scores = self.calc_score(processed_query, processed_key)
if self.mask is not None:
mask = self.mask.unsqueeze(1).expand(b, t_q, t_k)
scores.data.masked_fill_(mask, -65504.0)
scores_normalized = self.softmax_att(scores)
scores_normalized = self.dropout(scores_normalized)
context = self.context_matmul(scores_normalized, keys)
if single_query:
context = context.squeeze(1)
scores_normalized = scores_normalized.squeeze(1)
elif not self.batch_first:
context = context.transpose(0, 1)
scores_normalized = scores_normalized.transpose(0, 1)
return context, scores_normalized
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'query_size': 4, 'key_size': 4, 'num_units': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import math
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
from torch.nn.parameter import Parameter
import torch.onnx
import torch.testing
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1), xmask)
tl.store(out_ptr0 + x3, tmp0, xmask)
@triton.jit
def triton_poi_fused_mv_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * (x0 // 4), xmask, eviction_policy='evict_last'
)
tmp1 = tl.load(in_ptr1 + (4 * (x0 % 4) + 16 * (x0 // 16)), xmask,
eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (1 + 4 * (x0 // 4)), xmask, eviction_policy=
'evict_last')
tmp8 = tl.load(in_ptr1 + (1 + 4 * (x0 % 4) + 16 * (x0 // 16)), xmask,
eviction_policy='evict_last')
tmp11 = tl.load(in_ptr2 + 1)
tmp12 = tl.broadcast_to(tmp11, [XBLOCK])
tmp15 = tl.load(in_ptr0 + (2 + 4 * (x0 // 4)), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + (2 + 4 * (x0 % 4) + 16 * (x0 // 16)), xmask,
eviction_policy='evict_last')
tmp19 = tl.load(in_ptr2 + 2)
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp23 = tl.load(in_ptr0 + (3 + 4 * (x0 // 4)), xmask, eviction_policy=
'evict_last')
tmp24 = tl.load(in_ptr1 + (3 + 4 * (x0 % 4) + 16 * (x0 // 16)), xmask,
eviction_policy='evict_last')
tmp27 = tl.load(in_ptr2 + 3)
tmp28 = tl.broadcast_to(tmp27, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tmp6 = tmp3 * tmp5
tmp9 = tmp7 + tmp8
tmp10 = libdevice.tanh(tmp9)
tmp13 = tmp10 * tmp12
tmp14 = tmp6 + tmp13
tmp17 = tmp15 + tmp16
tmp18 = libdevice.tanh(tmp17)
tmp21 = tmp18 * tmp20
tmp22 = tmp14 + tmp21
tmp25 = tmp23 + tmp24
tmp26 = libdevice.tanh(tmp25)
tmp29 = tmp26 * tmp28
tmp30 = tmp22 + tmp29
tl.store(out_ptr0 + x0, tmp30, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64)](primals_2, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clone_0[grid(64)](primals_1, buf2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
del primals_4
buf4 = empty_strided_cuda((64,), (1,), torch.float32)
triton_poi_fused_mv_1[grid(64)](buf1, buf3, primals_5, buf4, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(64)](buf4, buf5, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4), (16, 4, 1), 0)
del buf4
triton_poi_fused__softmax_3[grid(64)](buf5, buf6, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf7 = buf5
del buf5
extern_kernels.bmm(buf6, reinterpret_tensor(primals_1, (4, 4, 4), (
4, 16, 1), 0), out=buf7)
return reinterpret_tensor(buf7, (4, 4, 4), (4, 16, 1), 0
), reinterpret_tensor(buf6, (4, 4, 4), (4, 16, 1), 0
), primals_5, reinterpret_tensor(buf0, (16, 4), (4, 1), 0
), buf1, reinterpret_tensor(buf2, (16, 4), (4, 1), 0
), buf3, buf6, reinterpret_tensor(primals_1, (4, 4, 4), (4, 1, 16), 0)
class EltwiseAdd(nn.Module):
def __init__(self, inplace=False):
"""Element-wise addition"""
super().__init__()
self.inplace = inplace
def forward(self, *input):
res = input[0]
if self.inplace:
for t in input[1:]:
res += t
else:
for t in input[1:]:
res = res + t
return res
class EltwiseMult(nn.Module):
def __init__(self, inplace=False):
"""Element-wise multiplication"""
super().__init__()
self.inplace = inplace
def forward(self, *input):
res = input[0]
if self.inplace:
for t in input[1:]:
res *= t
else:
for t in input[1:]:
res = res * t
return res
class Matmul(nn.Module):
"""
A wrapper module for matmul operation between 2 tensors.
"""
def __init__(self):
super(Matmul, self).__init__()
def forward(self, a: 'torch.Tensor', b: 'torch.Tensor'):
return a.matmul(b)
class BatchMatmul(nn.Module):
"""
A wrapper module for torch.bmm operation between 2 tensors.
"""
def __init__(self):
super(BatchMatmul, self).__init__()
def forward(self, a: 'torch.Tensor', b: 'torch.Tensor'):
return torch.bmm(a, b)
class BahdanauAttentionNew(nn.Module):
"""
It should be very similar to tf.contrib.seq2seq.BahdanauAttention
"""
def __init__(self, query_size, key_size, num_units, normalize=False,
dropout=0, batch_first=False):
super(BahdanauAttentionNew, self).__init__()
self.normalize = normalize
self.batch_first = batch_first
self.num_units = num_units
self.linear_q = nn.Linear(query_size, num_units, bias=False)
self.linear_k = nn.Linear(key_size, num_units, bias=False)
self.linear_att = Parameter(torch.Tensor(num_units))
self.dropout = nn.Dropout(dropout)
self.mask = None
self.eltwiseadd_qk = EltwiseAdd()
self.eltwiseadd_norm_bias = EltwiseAdd()
self.eltwisemul_norm_scaler = EltwiseMult()
self.tanh = nn.Tanh()
self.matmul_score = Matmul()
self.softmax_att = nn.Softmax(dim=-1)
self.context_matmul = BatchMatmul()
if self.normalize:
self.normalize_scalar = Parameter(torch.Tensor(1))
self.normalize_bias = Parameter(torch.Tensor(num_units))
else:
self.register_parameter('normalize_scalar', None)
self.register_parameter('normalize_bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1.0 / math.sqrt(self.num_units)
self.linear_att.data.uniform_(-stdv, stdv)
if self.normalize:
self.normalize_scalar.data.fill_(stdv)
self.normalize_bias.data.zero_()
def set_mask(self, context_len, context):
"""
sets self.mask which is applied before softmax
ones for inactive context fields, zeros for active context fields
:param context_len: b
:param context: if batch_first: (b x t_k x n) else: (t_k x b x n)
self.mask: (b x t_k)
"""
if self.batch_first:
max_len = context.size(1)
else:
max_len = context.size(0)
indices = torch.arange(0, max_len, dtype=torch.int64, device=
context.device)
self.mask = indices >= context_len.unsqueeze(1)
def calc_score(self, att_query, att_keys):
"""
Calculate Bahdanau score
:param att_query: b x t_q x n
:param att_keys: b x t_k x n
return b x t_q x t_k scores
"""
b, t_k, n = att_keys.size()
t_q = att_query.size(1)
att_query = att_query.unsqueeze(2).expand(b, t_q, t_k, n)
att_keys = att_keys.unsqueeze(1).expand(b, t_q, t_k, n)
sum_qk = self.eltwiseadd_qk(att_query, att_keys)
if self.normalize:
sum_qk = self.eltwiseadd_norm_bias(sum_qk, self.normalize_bias)
tmp = self.linear_att
linear_att = tmp / tmp.norm()
linear_att = linear_att
linear_att = self.eltwisemul_norm_scaler(linear_att, self.
normalize_scalar)
else:
linear_att = self.linear_att
out = self.matmul_score(self.tanh(sum_qk), linear_att)
return out
def forward(self, input_0, input_1):
primals_5 = self.linear_att
primals_3 = self.linear_q.weight
primals_4 = self.linear_k.weight
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0], output[1]
| saman-aghazadeh/distiller | BahdanauAttention | false | 4,260 | [
"Apache-2.0"
] | 0 | 7e8d3e6193c807f7c55d8453f64e1bc3c02eee30 | https://github.com/saman-aghazadeh/distiller/tree/7e8d3e6193c807f7c55d8453f64e1bc3c02eee30 | import math
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from torch.optim.lr_scheduler import *
import torch.optim.lr_scheduler
import torch.quantization
from torch.nn.parameter import Parameter
import torch.onnx
import torch.testing
class EltwiseAdd(nn.Module):
def __init__(self, inplace=False):
"""Element-wise addition"""
super().__init__()
self.inplace = inplace
def forward(self, *input):
res = input[0]
if self.inplace:
for t in input[1:]:
res += t
else:
for t in input[1:]:
res = res + t
return res
class EltwiseMult(nn.Module):
def __init__(self, inplace=False):
"""Element-wise multiplication"""
super().__init__()
self.inplace = inplace
def forward(self, *input):
res = input[0]
if self.inplace:
for t in input[1:]:
res *= t
else:
for t in input[1:]:
res = res * t
return res
class Matmul(nn.Module):
"""
A wrapper module for matmul operation between 2 tensors.
"""
def __init__(self):
super().__init__()
def forward(self, a: 'torch.Tensor', b: 'torch.Tensor'):
return a.matmul(b)
class BatchMatmul(nn.Module):
"""
A wrapper module for torch.bmm operation between 2 tensors.
"""
def __init__(self):
super().__init__()
def forward(self, a: 'torch.Tensor', b: 'torch.Tensor'):
return torch.bmm(a, b)
class Model(nn.Module):
"""
It should be very similar to tf.contrib.seq2seq.BahdanauAttention
"""
def __init__(self, query_size, key_size, num_units, normalize=False,
dropout=0, batch_first=False):
super().__init__()
self.normalize = normalize
self.batch_first = batch_first
self.num_units = num_units
self.linear_q = nn.Linear(query_size, num_units, bias=False)
self.linear_k = nn.Linear(key_size, num_units, bias=False)
self.linear_att = Parameter(torch.Tensor(num_units))
self.dropout = nn.Dropout(dropout)
self.mask = None
self.eltwiseadd_qk = EltwiseAdd()
self.eltwiseadd_norm_bias = EltwiseAdd()
self.eltwisemul_norm_scaler = EltwiseMult()
self.tanh = nn.Tanh()
self.matmul_score = Matmul()
self.softmax_att = nn.Softmax(dim=-1)
self.context_matmul = BatchMatmul()
if self.normalize:
self.normalize_scalar = Parameter(torch.Tensor(1))
self.normalize_bias = Parameter(torch.Tensor(num_units))
else:
self.register_parameter('normalize_scalar', None)
self.register_parameter('normalize_bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1.0 / math.sqrt(self.num_units)
self.linear_att.data.uniform_(-stdv, stdv)
if self.normalize:
self.normalize_scalar.data.fill_(stdv)
self.normalize_bias.data.zero_()
def set_mask(self, context_len, context):
"""
sets self.mask which is applied before softmax
ones for inactive context fields, zeros for active context fields
:param context_len: b
:param context: if batch_first: (b x t_k x n) else: (t_k x b x n)
self.mask: (b x t_k)
"""
if self.batch_first:
max_len = context.size(1)
else:
max_len = context.size(0)
indices = torch.arange(0, max_len, dtype=torch.int64, device=
context.device)
self.mask = indices >= context_len.unsqueeze(1)
def calc_score(self, att_query, att_keys):
"""
Calculate Bahdanau score
:param att_query: b x t_q x n
:param att_keys: b x t_k x n
return b x t_q x t_k scores
"""
b, t_k, n = att_keys.size()
t_q = att_query.size(1)
# ... truncated (>4000 chars) for memory efficiency |
GaussMembFunc | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/4z/c4z5f5sypbvdzckwli2vgwoo37wqsejweryvcktm6nccmzns6jtu.py
# Topologically Sorted Source Nodes: [sub, pow_1, neg, pow_2, mul, truediv, val], Original ATen: [aten.sub, aten.pow, aten.neg, aten.mul, aten.div, aten.exp]
# Source node to ATen node mapping:
# mul => mul
# neg => neg
# pow_1 => pow_1
# pow_2 => pow_2
# sub => sub
# truediv => div
# val => exp
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_2, %primals_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%pow_1,), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%primals_3, 2), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_2, 2), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%neg, %mul), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%div,), kwargs = {})
triton_poi_fused_div_exp_mul_neg_pow_sub_0 = async_compile.triton('triton_poi_fused_div_exp_mul_neg_pow_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_exp_mul_neg_pow_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_div_exp_mul_neg_pow_sub_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp6 = tl.load(in_ptr2 + (0))
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp3 = tmp0 - tmp2
tmp4 = tmp3 * tmp3
tmp5 = -tmp4
tmp8 = tmp7 * tmp7
tmp9 = 2.0
tmp10 = tmp8 * tmp9
tmp11 = tmp5 / tmp10
tmp12 = tl_math.exp(tmp11)
tl.store(out_ptr0 + (x0), tmp12, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (), ())
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (), ())
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sub, pow_1, neg, pow_2, mul, truediv, val], Original ATen: [aten.sub, aten.pow, aten.neg, aten.mul, aten.div, aten.exp]
stream0 = get_raw_stream(0)
triton_poi_fused_div_exp_mul_neg_pow_sub_0.run(primals_2, primals_1, primals_3, buf0, 256, grid=grid(256), stream=stream0)
return (buf0, primals_1, primals_2, primals_3, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((), (), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((), (), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
def _mk_param(val):
"""Make a torch parameter from a scalar value"""
if isinstance(val, torch.Tensor):
val = val.item()
return torch.nn.Parameter(torch.tensor(val, dtype=torch.float))
class GaussMembFunc(torch.nn.Module):
"""
Gaussian membership functions, defined by two parameters:
mu, the mean (center)
sigma, the standard deviation.
"""
def __init__(self, mu, sigma):
super(GaussMembFunc, self).__init__()
self.register_parameter('mu', _mk_param(mu))
self.register_parameter('sigma', _mk_param(sigma))
def forward(self, x):
val = torch.exp(-torch.pow(x - self.mu, 2) / (2 * self.sigma ** 2))
return val
def pretty(self):
return 'GaussMembFunc {} {}'.format(self.mu, self.sigma)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'mu': 4, 'sigma': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_div_exp_mul_neg_pow_sub_0(in_ptr0, in_ptr1, in_ptr2,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp6 = tl.load(in_ptr2 + 0)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp3 = tmp0 - tmp2
tmp4 = tmp3 * tmp3
tmp5 = -tmp4
tmp8 = tmp7 * tmp7
tmp9 = 2.0
tmp10 = tmp8 * tmp9
tmp11 = tmp5 / tmp10
tmp12 = tl_math.exp(tmp11)
tl.store(out_ptr0 + x0, tmp12, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (), ())
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (), ())
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_div_exp_mul_neg_pow_sub_0[grid(256)](primals_2,
primals_1, primals_3, buf0, 256, XBLOCK=128, num_warps=4,
num_stages=1)
return buf0, primals_1, primals_2, primals_3, buf0
def _mk_param(val):
"""Make a torch parameter from a scalar value"""
if isinstance(val, torch.Tensor):
val = val.item()
return torch.nn.Parameter(torch.tensor(val, dtype=torch.float))
class GaussMembFuncNew(torch.nn.Module):
"""
Gaussian membership functions, defined by two parameters:
mu, the mean (center)
sigma, the standard deviation.
"""
def __init__(self, mu, sigma):
super(GaussMembFuncNew, self).__init__()
self.register_parameter('mu', _mk_param(mu))
self.register_parameter('sigma', _mk_param(sigma))
def pretty(self):
return 'GaussMembFunc {} {}'.format(self.mu, self.sigma)
def forward(self, input_0):
primals_1 = self.mu
primals_3 = self.sigma
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| samxu0823/anfis-pytorch | GaussMembFunc | false | 4,261 | [
"MIT"
] | 0 | b4ec3f0e8259963800e9e0a2904a580d1e56cc1c | https://github.com/samxu0823/anfis-pytorch/tree/b4ec3f0e8259963800e9e0a2904a580d1e56cc1c | import torch
def _mk_param(val):
"""Make a torch parameter from a scalar value"""
if isinstance(val, torch.Tensor):
val = val.item()
return torch.nn.Parameter(torch.tensor(val, dtype=torch.float))
class Model(torch.nn.Module):
"""
Gaussian membership functions, defined by two parameters:
mu, the mean (center)
sigma, the standard deviation.
"""
def __init__(self, mu, sigma):
super().__init__()
self.register_parameter('mu', _mk_param(mu))
self.register_parameter('sigma', _mk_param(sigma))
def forward(self, x):
val = torch.exp(-torch.pow(x - self.mu, 2) / (2 * self.sigma ** 2))
return val
def pretty(self):
return 'GaussMembFunc {} {}'.format(self.mu, self.sigma)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
qy | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/6q/c6q46q7lsepa4jw5qgcgbc5kiud5wm57hubk6vfo4gk47vl2tprk.py
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# h => relu
# Graph fragment:
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%primals_1,), kwargs = {})
triton_poi_fused_relu_0 = async_compile.triton('triton_poi_fused_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_0.run(primals_1, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [loc_y], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(buf0, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_2
del primals_3
return (reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf0, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn as nn
class qy(nn.Module):
def __init__(self, d_dim, x_dim, y_dim, z_dim):
super(qy, self).__init__()
self.fc1 = nn.Linear(z_dim, y_dim)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.fc1.bias.data.zero_()
def forward(self, zy):
h = F.relu(zy)
loc_y = self.fc1(h)
return loc_y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_dim': 4, 'x_dim': 4, 'y_dim': 4, 'z_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_relu_0[grid(256)](primals_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(buf0, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf1)
del primals_2
del primals_3
return reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(buf0, (64, 4), (4, 1), 0)
class qyNew(nn.Module):
def __init__(self, d_dim, x_dim, y_dim, z_dim):
super(qyNew, self).__init__()
self.fc1 = nn.Linear(z_dim, y_dim)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.fc1.bias.data.zero_()
def forward(self, input_0):
primals_2 = self.fc1.weight
primals_3 = self.fc1.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| sautami26/DIVA | qy | false | 4,262 | [
"MIT"
] | 0 | 52af683db216cb6e2ac777597fd9ec744ce7c8f2 | https://github.com/sautami26/DIVA/tree/52af683db216cb6e2ac777597fd9ec744ce7c8f2 | import torch
import torch.nn.functional as F
import torch.nn as nn
class Model(nn.Module):
def __init__(self, d_dim, x_dim, y_dim, z_dim):
super().__init__()
self.fc1 = nn.Linear(z_dim, y_dim)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.fc1.bias.data.zero_()
def forward(self, zy):
h = F.relu(zy)
loc_y = self.fc1(h)
return loc_y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4, 4]
|
BertAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/x2/cx2hdvwyo7m5jvhhvtugzxqvmy6z4nsfhkkjhvgzbbm3cb6dsum2.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %mul_scalar : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%permute_default, 1.0), kwargs = {})
# %clone_default : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5j/c5jll3kxtd32cl7pwubrb5oky2mtzckfgip2xbwad7crvvp4zk4r.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_default_2, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_default_2, %amax_default), kwargs = {})
# %exp_default : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_tensor,), kwargs = {})
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/kt/cktnex5febczl2ac6zugjmcksgsd5kjdufazv65vtepuwob3cb7a.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %sum_dim_int_list : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_default, [-1], True), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_default, %sum_dim_int_list), kwargs = {})
# %eq_scalar : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%view_default_2, -inf), kwargs = {})
# %logical_not_default : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%eq_scalar,), kwargs = {})
# %any_dim : [num_users=1] = call_function[target=torch.ops.aten.any.dim](args = (%logical_not_default, -1, True), kwargs = {})
# %logical_not_default_1 : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%any_dim,), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%logical_not_default_1, %full_default, %div_tensor), kwargs = {})
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (x2), xmask)
tmp26 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp31 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = float("-inf")
tmp2 = tmp0 == tmp1
tmp3 = tmp2 == 0
tmp4 = tmp3.to(tl.int64)
tmp5 = (tmp4 != 0)
tmp7 = tmp6 == tmp1
tmp8 = tmp7 == 0
tmp9 = tmp8.to(tl.int64)
tmp10 = (tmp9 != 0)
tmp11 = tmp5 | tmp10
tmp13 = tmp12 == tmp1
tmp14 = tmp13 == 0
tmp15 = tmp14.to(tl.int64)
tmp16 = (tmp15 != 0)
tmp17 = tmp11 | tmp16
tmp19 = tmp18 == tmp1
tmp20 = tmp19 == 0
tmp21 = tmp20.to(tl.int64)
tmp22 = (tmp21 != 0)
tmp23 = tmp17 | tmp22
tmp24 = tmp23 == 0
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp32 = tmp30 + tmp31
tmp33 = tmp25 / tmp32
tmp34 = 0.0
tmp35 = tl.where(tmp24, tmp34, tmp33)
tl.store(out_ptr0 + (x2), tmp35, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vv/cvvnhithjvmvhfjufxwwzclfobkrgbyyteg66hp24r675f7elw4c.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %clone_default_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_default_3,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6t/c6t5a5ere3lqjiu7zh3uu4oxmpdoujdaqqmeunxqapgzo4m74uav.py
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# context_layer_1 => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6m/c6mhj5zwirfhy5e4o45uaeov72uwfby4udubpm2fcz42iqvs2g57.py
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# hidden_states_2 => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/l3/cl3bnd5pv2p4ydfmlj74bv4mbiwr2ntrdvbubnjubetyhosmxag6.py
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# hidden_states_2 => add_1, add_2, mul, mul_1, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1.0), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_10), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_11), kwargs = {})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(buf0, primals_2, buf3, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_0.run(buf1, primals_5, buf4, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(buf5, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf5
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(buf2, primals_7, buf8, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_7
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [context_layer_1], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf9, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [hidden_states], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_9, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf11)
del primals_9
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(buf11, primals_3, buf12, buf13, 16, grid=grid(16), stream=stream0)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, hidden_states_2], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(buf11, primals_3, buf12, buf13, primals_10, primals_11, buf14, 64, grid=grid(64), stream=stream0)
del buf12
del buf13
del primals_11
return (buf14, primals_3, primals_10, buf7, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0), reinterpret_tensor(buf10, (16, 4), (4, 1), 0), buf11, primals_8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import math
import torch
import torch.nn as nn
import torch.utils.checkpoint
class BertSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
if (config.hidden_size % config.num_attention_heads != 0 and not
hasattr(config, 'embedding_size')):
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.all_head_size = (self.num_attention_heads * self.
attention_head_size)
self.query = nn.Linear(config.hidden_size, self.all_head_size)
self.key = nn.Linear(config.hidden_size, self.all_head_size)
self.value = nn.Linear(config.hidden_size, self.all_head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
self.position_embedding_type = getattr(config,
'position_embedding_type', 'absolute')
if (self.position_embedding_type == 'relative_key' or self.
position_embedding_type == 'relative_key_query'):
self.max_position_embeddings = config.max_position_embeddings
self.distance_embedding = nn.Embedding(2 * config.
max_position_embeddings - 1, self.attention_head_size)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, attention_mask=None, head_mask=None,
encoder_hidden_states=None, encoder_attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
if encoder_hidden_states is not None:
mixed_key_layer = self.key(encoder_hidden_states)
mixed_value_layer = self.value(encoder_hidden_states)
attention_mask = encoder_attention_mask
else:
mixed_key_layer = self.key(hidden_states)
mixed_value_layer = self.value(hidden_states)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
if (self.position_embedding_type == 'relative_key' or self.
position_embedding_type == 'relative_key_query'):
seq_length = hidden_states.size()[1]
position_ids_l = torch.arange(seq_length, dtype=torch.long,
device=hidden_states.device).view(-1, 1)
position_ids_r = torch.arange(seq_length, dtype=torch.long,
device=hidden_states.device).view(1, -1)
distance = position_ids_l - position_ids_r
positional_embedding = self.distance_embedding(distance + self.
max_position_embeddings - 1)
positional_embedding = positional_embedding
if self.position_embedding_type == 'relative_key':
relative_position_scores = torch.einsum('bhld,lrd->bhlr',
query_layer, positional_embedding)
attention_scores = attention_scores + relative_position_scores
elif self.position_embedding_type == 'relative_key_query':
relative_position_scores_query = torch.einsum('bhld,lrd->bhlr',
query_layer, positional_embedding)
relative_position_scores_key = torch.einsum('bhrd,lrd->bhlr',
key_layer, positional_embedding)
attention_scores = (attention_scores +
relative_position_scores_query +
relative_position_scores_key)
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
if head_mask is not None:
attention_probs = attention_probs * head_mask
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.
all_head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class BertSelfOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.
layer_norm_eps)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class BertAttention(nn.Module):
def __init__(self, config):
super().__init__()
self.self = BertSelfAttention(config)
self.output = BertSelfOutput(config)
self.pruned_heads = set()
def prune_heads(self, heads):
if len(heads) == 0:
return
heads, index = find_pruneable_heads_and_indices(heads, self.self.
num_attention_heads, self.self.attention_head_size, self.
pruned_heads)
self.self.query = prune_linear_layer(self.self.query, index)
self.self.key = prune_linear_layer(self.self.key, index)
self.self.value = prune_linear_layer(self.self.value, index)
self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
self.self.num_attention_heads = self.self.num_attention_heads - len(
heads)
self.self.all_head_size = (self.self.attention_head_size * self.
self.num_attention_heads)
self.pruned_heads = self.pruned_heads.union(heads)
def forward(self, hidden_states, attention_mask=None, head_mask=None,
encoder_hidden_states=None, encoder_attention_mask=None,
output_attentions=False):
self_outputs = self.self(hidden_states, attention_mask, head_mask,
encoder_hidden_states, encoder_attention_mask, output_attentions)
attention_output = self.output(self_outputs[0], hidden_states)
outputs = (attention_output,) + self_outputs[1:]
return outputs
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(hidden_size=4, num_attention_heads=
4, attention_probs_dropout_prob=0.5, position_embedding_type=4,
layer_norm_eps=1, hidden_dropout_prob=0.5)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import math
import torch.nn as nn
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr1 + x2, xmask)
tmp26 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp29 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp31 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = float('-inf')
tmp2 = tmp0 == tmp1
tmp3 = tmp2 == 0
tmp4 = tmp3.to(tl.int64)
tmp5 = tmp4 != 0
tmp7 = tmp6 == tmp1
tmp8 = tmp7 == 0
tmp9 = tmp8.to(tl.int64)
tmp10 = tmp9 != 0
tmp11 = tmp5 | tmp10
tmp13 = tmp12 == tmp1
tmp14 = tmp13 == 0
tmp15 = tmp14.to(tl.int64)
tmp16 = tmp15 != 0
tmp17 = tmp11 | tmp16
tmp19 = tmp18 == tmp1
tmp20 = tmp19 == 0
tmp21 = tmp20.to(tl.int64)
tmp22 = tmp21 != 0
tmp23 = tmp17 | tmp22
tmp24 = tmp23 == 0
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp32 = tmp30 + tmp31
tmp33 = tmp25 / tmp32
tmp34 = 0.0
tmp35 = tl.where(tmp24, tmp34, tmp33)
tl.store(out_ptr0 + x2, tmp35, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(16, 4)](buf0, primals_2, buf3, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_2
buf4 = reinterpret_tensor(buf0, (4, 4, 1, 4), (16, 4, 4, 1), 0)
del buf0
triton_poi_fused_0[grid(16, 4)](buf1, primals_5, buf4, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_1[grid(256)](buf5, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_2[grid(256)](buf5, buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_3[grid(16, 4)](buf2, primals_7, buf8, 16, 4,
XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_7
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_4[grid(16, 4)](buf9, buf10, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.addmm(primals_9, reinterpret_tensor(buf10, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf11)
del primals_9
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](buf11, primals_3,
buf12, buf13, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(64)](buf11, primals_3,
buf12, buf13, primals_10, primals_11, buf14, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf12
del buf13
del primals_11
return buf14, primals_3, primals_10, buf7, reinterpret_tensor(buf8, (16,
1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 4), 0
), reinterpret_tensor(buf10, (16, 4), (4, 1), 0), buf11, primals_8
class BertSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
if (config.hidden_size % config.num_attention_heads != 0 and not
hasattr(config, 'embedding_size')):
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.all_head_size = (self.num_attention_heads * self.
attention_head_size)
self.query = nn.Linear(config.hidden_size, self.all_head_size)
self.key = nn.Linear(config.hidden_size, self.all_head_size)
self.value = nn.Linear(config.hidden_size, self.all_head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
self.position_embedding_type = getattr(config,
'position_embedding_type', 'absolute')
if (self.position_embedding_type == 'relative_key' or self.
position_embedding_type == 'relative_key_query'):
self.max_position_embeddings = config.max_position_embeddings
self.distance_embedding = nn.Embedding(2 * config.
max_position_embeddings - 1, self.attention_head_size)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, attention_mask=None, head_mask=None,
encoder_hidden_states=None, encoder_attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
if encoder_hidden_states is not None:
mixed_key_layer = self.key(encoder_hidden_states)
mixed_value_layer = self.value(encoder_hidden_states)
attention_mask = encoder_attention_mask
else:
mixed_key_layer = self.key(hidden_states)
mixed_value_layer = self.value(hidden_states)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
if (self.position_embedding_type == 'relative_key' or self.
position_embedding_type == 'relative_key_query'):
seq_length = hidden_states.size()[1]
position_ids_l = torch.arange(seq_length, dtype=torch.long,
device=hidden_states.device).view(-1, 1)
position_ids_r = torch.arange(seq_length, dtype=torch.long,
device=hidden_states.device).view(1, -1)
distance = position_ids_l - position_ids_r
positional_embedding = self.distance_embedding(distance + self.
max_position_embeddings - 1)
positional_embedding = positional_embedding
if self.position_embedding_type == 'relative_key':
relative_position_scores = torch.einsum('bhld,lrd->bhlr',
query_layer, positional_embedding)
attention_scores = attention_scores + relative_position_scores
elif self.position_embedding_type == 'relative_key_query':
relative_position_scores_query = torch.einsum('bhld,lrd->bhlr',
query_layer, positional_embedding)
relative_position_scores_key = torch.einsum('bhrd,lrd->bhlr',
key_layer, positional_embedding)
attention_scores = (attention_scores +
relative_position_scores_query +
relative_position_scores_key)
attention_scores = attention_scores / math.sqrt(self.
attention_head_size)
if attention_mask is not None:
attention_scores = attention_scores + attention_mask
attention_probs = nn.Softmax(dim=-1)(attention_scores)
attention_probs = self.dropout(attention_probs)
if head_mask is not None:
attention_probs = attention_probs * head_mask
context_layer = torch.matmul(attention_probs, value_layer)
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
new_context_layer_shape = context_layer.size()[:-2] + (self.
all_head_size,)
context_layer = context_layer.view(*new_context_layer_shape)
outputs = (context_layer, attention_probs) if output_attentions else (
context_layer,)
return outputs
class BertSelfOutput(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.
layer_norm_eps)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(self, hidden_states, input_tensor):
hidden_states = self.dense(hidden_states)
hidden_states = self.dropout(hidden_states)
hidden_states = self.LayerNorm(hidden_states + input_tensor)
return hidden_states
class BertAttentionNew(nn.Module):
def __init__(self, config):
super().__init__()
self.self = BertSelfAttention(config)
self.output = BertSelfOutput(config)
self.pruned_heads = set()
def prune_heads(self, heads):
if len(heads) == 0:
return
heads, index = find_pruneable_heads_and_indices(heads, self.self.
num_attention_heads, self.self.attention_head_size, self.
pruned_heads)
self.self.query = prune_linear_layer(self.self.query, index)
self.self.key = prune_linear_layer(self.self.key, index)
self.self.value = prune_linear_layer(self.self.value, index)
self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
self.self.num_attention_heads = self.self.num_attention_heads - len(
heads)
self.self.all_head_size = (self.self.attention_head_size * self.
self.num_attention_heads)
self.pruned_heads = self.pruned_heads.union(heads)
def forward(self, input_0):
primals_1 = self.self.query.weight
primals_2 = self.self.query.bias
primals_4 = self.self.key.weight
primals_5 = self.self.key.bias
primals_6 = self.self.value.weight
primals_7 = self.self.value.bias
primals_8 = self.output.dense.weight
primals_9 = self.output.dense.bias
primals_10 = self.output.LayerNorm.weight
primals_11 = self.output.LayerNorm.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
| Hzfinfdu/Black-Box-Tuning | BertAttention | false | 4,263 | [
"MIT"
] | 0 | 64eb5505875dc1b242c6f0a2a2f07e4000c24cb4 | https://github.com/Hzfinfdu/Black-Box-Tuning/tree/64eb5505875dc1b242c6f0a2a2f07e4000c24cb4 | from _paritybench_helpers import _mock_config
import math
import torch
import torch.nn as nn
import torch.utils.checkpoint
class BertSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
if (config.hidden_size % config.num_attention_heads != 0 and not
hasattr(config, 'embedding_size')):
raise ValueError(
'The hidden size (%d) is not a multiple of the number of attention heads (%d)'
% (config.hidden_size, config.num_attention_heads))
self.num_attention_heads = config.num_attention_heads
self.attention_head_size = int(config.hidden_size / config.
num_attention_heads)
self.all_head_size = (self.num_attention_heads * self.
attention_head_size)
self.query = nn.Linear(config.hidden_size, self.all_head_size)
self.key = nn.Linear(config.hidden_size, self.all_head_size)
self.value = nn.Linear(config.hidden_size, self.all_head_size)
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
self.position_embedding_type = getattr(config,
'position_embedding_type', 'absolute')
if (self.position_embedding_type == 'relative_key' or self.
position_embedding_type == 'relative_key_query'):
self.max_position_embeddings = config.max_position_embeddings
self.distance_embedding = nn.Embedding(2 * config.
max_position_embeddings - 1, self.attention_head_size)
def transpose_for_scores(self, x):
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.
attention_head_size)
x = x.view(*new_x_shape)
return x.permute(0, 2, 1, 3)
def forward(self, hidden_states, attention_mask=None, head_mask=None,
encoder_hidden_states=None, encoder_attention_mask=None,
output_attentions=False):
mixed_query_layer = self.query(hidden_states)
if encoder_hidden_states is not None:
mixed_key_layer = self.key(encoder_hidden_states)
mixed_value_layer = self.value(encoder_hidden_states)
attention_mask = encoder_attention_mask
else:
mixed_key_layer = self.key(hidden_states)
mixed_value_layer = self.value(hidden_states)
query_layer = self.transpose_for_scores(mixed_query_layer)
key_layer = self.transpose_for_scores(mixed_key_layer)
value_layer = self.transpose_for_scores(mixed_value_layer)
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1,
-2))
if (self.position_embedding_type == 'relative_key' or self.
position_embedding_type == 'relative_key_query'):
seq_length = hidden_states.size()[1]
position_ids_l = torch.arange(seq_length, dtype=torch.long,
device=hidden_states.device).view(-1, 1)
position_ids_r = torch.arange(seq_length, dtype=torch.long,
device=hidden_states.device).view(1, -1)
distance = position_ids_l - position_ids_r
positional_embedding = self.distance_embedding(distance + self.
max_position_embeddings - 1)
positional_embedding = positional_embedding
if self.position_embedding_type == 'relative_key':
relative_position_scores = torch.einsum('bhld,lrd->bhlr',
query_layer, positional_embedding)
attention_scores = attention_scores + relative_position_scores
elif self.position_embedding_type == 'relative_key_query':
relative_position_scores_query = torch.einsum('bhld,lrd->bhlr',
query_layer, positional_embedding)
relative_position_scores_key = torch.einsum('bhrd,lrd->bhlr',
key_layer, positional_embedding)
attention_scores = (attention_scores +
relative_position_scores_query +
# ... truncated (>4000 chars) for memory efficiency |
down | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/e6/ce6fwb7cuhy3qppzvzwzq3dqytlyhklktwnjhzdza6cxmtqodq25.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.avg_pool2d]
# Source node to ATen node mapping:
# x => avg_pool2d
# Graph fragment:
# %avg_pool2d : [num_users=2] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%primals_1, [2, 2]), kwargs = {})
triton_poi_fused_avg_pool2d_0 = async_compile.triton('triton_poi_fused_avg_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_avg_pool2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 32
x1 = (xindex // 32)
x2 = xindex
tmp0 = tl.load(in_ptr0 + ((2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + (2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (65 + (2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tl.store(out_ptr0 + (x2), tmp8, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/j6/cj6n5qdz5f6f2g4oatwbm2xfskl6mdyix2skekye6ilanaqhphqv.py
# Topologically Sorted Source Nodes: [conv2d, x_1], Original ATen: [aten.convolution, aten.leaky_relu]
# Source node to ATen node mapping:
# conv2d => convolution
# x_1 => gt, mul, where
# Graph fragment:
# %convolution : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%avg_pool2d, %primals_2, %primals_3, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %gt : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%convolution, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, 0.1), kwargs = {})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt, %convolution, %mul), kwargs = {})
triton_poi_fused_convolution_leaky_relu_1 = async_compile.triton('triton_poi_fused_convolution_leaky_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_leaky_relu_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_leaky_relu_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 15376
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 961) % 4
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.1
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x3), tmp4, xmask)
tl.store(out_ptr1 + (x3), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/zg/czgcv4szvzqiitflipuf2423tmuqp5ogktsqdb2cvy5thsi6rpqj.py
# Topologically Sorted Source Nodes: [conv2d_1, x_2], Original ATen: [aten.convolution, aten.leaky_relu]
# Source node to ATen node mapping:
# conv2d_1 => convolution_1
# x_2 => gt_1, mul_1, where_1
# Graph fragment:
# %convolution_1 : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%where, %primals_4, %primals_5, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %gt_1 : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%convolution_1, 0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution_1, 0.1), kwargs = {})
# %where_1 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %convolution_1, %mul_1), kwargs = {})
triton_poi_fused_convolution_leaky_relu_2 = async_compile.triton('triton_poi_fused_convolution_leaky_relu_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_leaky_relu_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_leaky_relu_2(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 14400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 900) % 4
x2 = (xindex // 3600)
x4 = xindex % 3600
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.1
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x4 + (3712*x2)), tmp4, xmask)
tl.store(out_ptr1 + (x3), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 64, 64), (16384, 4096, 64, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 32, 32), (4096, 1024, 32, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.avg_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_avg_pool2d_0.run(primals_1, buf0, 16384, grid=grid(16384), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 31, 31), (3844, 961, 31, 1))
buf2 = empty_strided_cuda((4, 4, 31, 31), (3844, 961, 31, 1), torch.bool)
buf3 = empty_strided_cuda((4, 4, 31, 31), (3844, 961, 31, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv2d, x_1], Original ATen: [aten.convolution, aten.leaky_relu]
triton_poi_fused_convolution_leaky_relu_1.run(buf1, primals_3, buf2, buf3, 15376, grid=grid(15376), stream=stream0)
del buf1
del primals_3
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf3, primals_4, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 4, 30, 30), (3600, 900, 30, 1))
buf5 = empty_strided_cuda((4, 4, 30, 30), (3712, 900, 30, 1), torch.bool)
buf6 = empty_strided_cuda((4, 4, 30, 30), (3600, 900, 30, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv2d_1, x_2], Original ATen: [aten.convolution, aten.leaky_relu]
triton_poi_fused_convolution_leaky_relu_2.run(buf4, primals_5, buf5, buf6, 14400, grid=grid(14400), stream=stream0)
del buf4
del primals_5
return (buf6, primals_2, primals_4, buf0, buf2, buf3, buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 64, 64), (16384, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch.functional import F
import torch.nn as nn
import torch.nn.functional as F
class down(nn.Module):
"""
A class for creating neural network blocks containing layers:
Average Pooling --> Convlution + Leaky ReLU --> Convolution + Leaky ReLU
This is used in the UNet Class to create a UNet like NN architecture.
...
Methods
-------
forward(x)
Returns output tensor after passing input `x` to the neural network
block.
"""
def __init__(self, inChannels, outChannels, filterSize):
"""
Parameters
----------
inChannels : int
number of input channels for the first convolutional layer.
outChannels : int
number of output channels for the first convolutional layer.
This is also used as input and output channels for the
second convolutional layer.
filterSize : int
filter size for the convolution filter. input N would create
a N x N filter.
"""
super(down, self).__init__()
self.conv1 = nn.Conv2d(inChannels, outChannels, filterSize, stride=
1, padding=int((filterSize - 1) / 2))
self.conv2 = nn.Conv2d(outChannels, outChannels, filterSize, stride
=1, padding=int((filterSize - 1) / 2))
def forward(self, x):
"""
Returns output tensor after passing input `x` to the neural network
block.
Parameters
----------
x : tensor
input to the NN block.
Returns
-------
tensor
output of the NN block.
"""
x = F.avg_pool2d(x, 2)
x = F.leaky_relu(self.conv1(x), negative_slope=0.1)
x = F.leaky_relu(self.conv2(x), negative_slope=0.1)
return x
def get_inputs():
return [torch.rand([4, 4, 64, 64])]
def get_init_inputs():
return [[], {'inChannels': 4, 'outChannels': 4, 'filterSize': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 32
x1 = xindex // 32
x2 = xindex
tmp0 = tl.load(in_ptr0 + (2 * x0 + 128 * x1), None, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 128 * x1), None, eviction_policy
='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + 2 * x0 + 128 * x1), None,
eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (65 + 2 * x0 + 128 * x1), None,
eviction_policy='evict_last')
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tl.store(out_ptr0 + x2, tmp8, None)
@triton.jit
def triton_poi_fused_convolution_leaky_relu_1(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 15376
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 961 % 4
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.1
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + x3, tmp4, xmask)
tl.store(out_ptr1 + x3, tmp7, xmask)
@triton.jit
def triton_poi_fused_convolution_leaky_relu_2(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 14400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 900 % 4
x2 = xindex // 3600
x4 = xindex % 3600
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.1
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x4 + 3712 * x2), tmp4, xmask)
tl.store(out_ptr1 + x3, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 64, 64), (16384, 4096, 64, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 32, 32), (4096, 1024, 32, 1),
torch.float32)
get_raw_stream(0)
triton_poi_fused_avg_pool2d_0[grid(16384)](primals_1, buf0, 16384,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_1
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 31, 31), (3844, 961, 31, 1))
buf2 = empty_strided_cuda((4, 4, 31, 31), (3844, 961, 31, 1), torch
.bool)
buf3 = empty_strided_cuda((4, 4, 31, 31), (3844, 961, 31, 1), torch
.float32)
triton_poi_fused_convolution_leaky_relu_1[grid(15376)](buf1,
primals_3, buf2, buf3, 15376, XBLOCK=256, num_warps=4, num_stages=1
)
del buf1
del primals_3
buf4 = extern_kernels.convolution(buf3, primals_4, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 4, 30, 30), (3600, 900, 30, 1))
buf5 = empty_strided_cuda((4, 4, 30, 30), (3712, 900, 30, 1), torch
.bool)
buf6 = empty_strided_cuda((4, 4, 30, 30), (3600, 900, 30, 1), torch
.float32)
triton_poi_fused_convolution_leaky_relu_2[grid(14400)](buf4,
primals_5, buf5, buf6, 14400, XBLOCK=256, num_warps=4, num_stages=1
)
del buf4
del primals_5
return buf6, primals_2, primals_4, buf0, buf2, buf3, buf5
class downNew(nn.Module):
"""
A class for creating neural network blocks containing layers:
Average Pooling --> Convlution + Leaky ReLU --> Convolution + Leaky ReLU
This is used in the UNet Class to create a UNet like NN architecture.
...
Methods
-------
forward(x)
Returns output tensor after passing input `x` to the neural network
block.
"""
def __init__(self, inChannels, outChannels, filterSize):
"""
Parameters
----------
inChannels : int
number of input channels for the first convolutional layer.
outChannels : int
number of output channels for the first convolutional layer.
This is also used as input and output channels for the
second convolutional layer.
filterSize : int
filter size for the convolution filter. input N would create
a N x N filter.
"""
super(downNew, self).__init__()
self.conv1 = nn.Conv2d(inChannels, outChannels, filterSize, stride=
1, padding=int((filterSize - 1) / 2))
self.conv2 = nn.Conv2d(outChannels, outChannels, filterSize, stride
=1, padding=int((filterSize - 1) / 2))
def forward(self, input_0):
primals_2 = self.conv1.weight
primals_3 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| samuelpietri/Super-SloMo | down | false | 4,264 | [
"MIT"
] | 0 | e20eaa5550c30737be42b61f8e82e731cfd17457 | https://github.com/samuelpietri/Super-SloMo/tree/e20eaa5550c30737be42b61f8e82e731cfd17457 | import torch
from torch.functional import F
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
"""
A class for creating neural network blocks containing layers:
Average Pooling --> Convlution + Leaky ReLU --> Convolution + Leaky ReLU
This is used in the UNet Class to create a UNet like NN architecture.
...
Methods
-------
forward(x)
Returns output tensor after passing input `x` to the neural network
block.
"""
def __init__(self, inChannels, outChannels, filterSize):
"""
Parameters
----------
inChannels : int
number of input channels for the first convolutional layer.
outChannels : int
number of output channels for the first convolutional layer.
This is also used as input and output channels for the
second convolutional layer.
filterSize : int
filter size for the convolution filter. input N would create
a N x N filter.
"""
super().__init__()
self.conv1 = nn.Conv2d(inChannels, outChannels, filterSize, stride=
1, padding=int((filterSize - 1) / 2))
self.conv2 = nn.Conv2d(outChannels, outChannels, filterSize, stride
=1, padding=int((filterSize - 1) / 2))
def forward(self, x):
"""
Returns output tensor after passing input `x` to the neural network
block.
Parameters
----------
x : tensor
input to the NN block.
Returns
-------
tensor
output of the NN block.
"""
x = F.avg_pool2d(x, 2)
x = F.leaky_relu(self.conv1(x), negative_slope=0.1)
x = F.leaky_relu(self.conv2(x), negative_slope=0.1)
return x
def get_inputs():
return [torch.rand([4, 4, 64, 64])]
def get_init_inputs():
return [4, 4, 4]
|
BellMembFunc | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/eb/cebvdz62qer4oz5rwrmgolin6f7xce7tjvzcl2p4jvitglhropzk.py
# Topologically Sorted Source Nodes: [sub, truediv, dist, pow_2, add, reciprocal], Original ATen: [aten.sub, aten.div, aten.pow, aten.add, aten.reciprocal]
# Source node to ATen node mapping:
# add => add
# dist => pow_1
# pow_2 => pow_2
# reciprocal => reciprocal
# sub => sub
# truediv => div
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_2, %primals_1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %primals_3), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%div, 2), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Tensor](args = (%pow_1, %primals_4), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_2, 1), kwargs = {})
# %reciprocal : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%add,), kwargs = {})
triton_poi_fused_add_div_pow_reciprocal_sub_0 = async_compile.triton('triton_poi_fused_add_div_pow_reciprocal_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_pow_reciprocal_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_pow_reciprocal_sub_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr2 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp8 = tl.load(in_ptr3 + (0))
tmp9 = tl.broadcast_to(tmp8, [XBLOCK])
tmp3 = tmp0 - tmp2
tmp6 = tmp3 / tmp5
tmp7 = tmp6 * tmp6
tmp10 = libdevice.pow(tmp7, tmp9)
tmp11 = 1.0
tmp12 = tmp10 + tmp11
tmp13 = tl.full([1], 1, tl.int32)
tmp14 = tmp13 / tmp12
tl.store(out_ptr0 + (x0), tmp14, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (), ())
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (), ())
assert_size_stride(primals_4, (), ())
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sub, truediv, dist, pow_2, add, reciprocal], Original ATen: [aten.sub, aten.div, aten.pow, aten.add, aten.reciprocal]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_pow_reciprocal_sub_0.run(primals_2, primals_1, primals_3, primals_4, buf0, 256, grid=grid(256), stream=stream0)
return (buf0, primals_1, primals_2, primals_3, primals_4, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((), (), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((), (), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((), (), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
def _mk_param(val):
"""Make a torch parameter from a scalar value"""
if isinstance(val, torch.Tensor):
val = val.item()
return torch.nn.Parameter(torch.tensor(val, dtype=torch.float))
class BellMembFunc(torch.nn.Module):
"""
Generalised Bell membership function; defined by three parameters:
a, the half-width (at the crossover point)
b, controls the slope at the crossover point (which is -b/2a)
c, the center point
"""
def __init__(self, a, b, c):
super(BellMembFunc, self).__init__()
self.register_parameter('a', _mk_param(a))
self.register_parameter('b', _mk_param(b))
self.register_parameter('c', _mk_param(c))
self.b.register_hook(BellMembFunc.b_log_hook)
@staticmethod
def b_log_hook(grad):
"""
Possibility of a log(0) in the grad for b, giving a nan.
Fix this by replacing any nan in the grad with ~0.
"""
grad[torch.isnan(grad)] = 1e-09
return grad
def forward(self, x):
dist = torch.pow((x - self.c) / self.a, 2)
return torch.reciprocal(1 + torch.pow(dist, self.b))
def pretty(self):
return 'BellMembFunc {} {} {}'.format(self.a, self.b, self.c)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'a': 4, 'b': 4, 'c': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_div_pow_reciprocal_sub_0(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr2 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp8 = tl.load(in_ptr3 + 0)
tmp9 = tl.broadcast_to(tmp8, [XBLOCK])
tmp3 = tmp0 - tmp2
tmp6 = tmp3 / tmp5
tmp7 = tmp6 * tmp6
tmp10 = libdevice.pow(tmp7, tmp9)
tmp11 = 1.0
tmp12 = tmp10 + tmp11
tmp13 = tl.full([1], 1, tl.int32)
tmp14 = tmp13 / tmp12
tl.store(out_ptr0 + x0, tmp14, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (), ())
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (), ())
assert_size_stride(primals_4, (), ())
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_div_pow_reciprocal_sub_0[grid(256)](primals_2,
primals_1, primals_3, primals_4, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
return buf0, primals_1, primals_2, primals_3, primals_4, buf0
def _mk_param(val):
"""Make a torch parameter from a scalar value"""
if isinstance(val, torch.Tensor):
val = val.item()
return torch.nn.Parameter(torch.tensor(val, dtype=torch.float))
class BellMembFuncNew(torch.nn.Module):
"""
Generalised Bell membership function; defined by three parameters:
a, the half-width (at the crossover point)
b, controls the slope at the crossover point (which is -b/2a)
c, the center point
"""
def __init__(self, a, b, c):
super(BellMembFuncNew, self).__init__()
self.register_parameter('a', _mk_param(a))
self.register_parameter('b', _mk_param(b))
self.register_parameter('c', _mk_param(c))
self.b.register_hook(BellMembFuncNew.b_log_hook)
@staticmethod
def b_log_hook(grad):
"""
Possibility of a log(0) in the grad for b, giving a nan.
Fix this by replacing any nan in the grad with ~0.
"""
grad[torch.isnan(grad)] = 1e-09
return grad
def pretty(self):
return 'BellMembFunc {} {} {}'.format(self.a, self.b, self.c)
def forward(self, input_0):
primals_1 = self.a
primals_3 = self.b
primals_4 = self.c
primals_2 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| samxu0823/anfis-pytorch | BellMembFunc | false | 4,265 | [
"MIT"
] | 0 | b4ec3f0e8259963800e9e0a2904a580d1e56cc1c | https://github.com/samxu0823/anfis-pytorch/tree/b4ec3f0e8259963800e9e0a2904a580d1e56cc1c | import torch
def _mk_param(val):
"""Make a torch parameter from a scalar value"""
if isinstance(val, torch.Tensor):
val = val.item()
return torch.nn.Parameter(torch.tensor(val, dtype=torch.float))
class Model(torch.nn.Module):
"""
Generalised Bell membership function; defined by three parameters:
a, the half-width (at the crossover point)
b, controls the slope at the crossover point (which is -b/2a)
c, the center point
"""
def __init__(self, a, b, c):
super().__init__()
self.register_parameter('a', _mk_param(a))
self.register_parameter('b', _mk_param(b))
self.register_parameter('c', _mk_param(c))
self.b.register_hook(BellMembFunc.b_log_hook)
@staticmethod
def b_log_hook(grad):
"""
Possibility of a log(0) in the grad for b, giving a nan.
Fix this by replacing any nan in the grad with ~0.
"""
grad[torch.isnan(grad)] = 1e-09
return grad
def forward(self, x):
dist = torch.pow((x - self.c) / self.a, 2)
return torch.reciprocal(1 + torch.pow(dist, self.b))
def pretty(self):
return 'BellMembFunc {} {} {}'.format(self.a, self.b, self.c)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
qd | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/6q/c6q46q7lsepa4jw5qgcgbc5kiud5wm57hubk6vfo4gk47vl2tprk.py
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# h => relu
# Graph fragment:
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%primals_1,), kwargs = {})
triton_poi_fused_relu_0 = async_compile.triton('triton_poi_fused_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_0.run(primals_1, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [loc_d], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(buf0, (64, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_2
del primals_3
return (reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf0, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn as nn
class qd(nn.Module):
def __init__(self, d_dim, x_dim, y_dim, z_dim):
super(qd, self).__init__()
self.fc1 = nn.Linear(z_dim, d_dim)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.fc1.bias.data.zero_()
def forward(self, zd):
h = F.relu(zd)
loc_d = self.fc1(h)
return loc_d
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_dim': 4, 'x_dim': 4, 'y_dim': 4, 'z_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_relu_0[grid(256)](primals_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(buf0, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf1)
del primals_2
del primals_3
return reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(buf0, (64, 4), (4, 1), 0)
class qdNew(nn.Module):
def __init__(self, d_dim, x_dim, y_dim, z_dim):
super(qdNew, self).__init__()
self.fc1 = nn.Linear(z_dim, d_dim)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.fc1.bias.data.zero_()
def forward(self, input_0):
primals_2 = self.fc1.weight
primals_3 = self.fc1.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| sautami26/DIVA | qd | false | 4,266 | [
"MIT"
] | 0 | 52af683db216cb6e2ac777597fd9ec744ce7c8f2 | https://github.com/sautami26/DIVA/tree/52af683db216cb6e2ac777597fd9ec744ce7c8f2 | import torch
import torch.nn.functional as F
import torch.nn as nn
class Model(nn.Module):
def __init__(self, d_dim, x_dim, y_dim, z_dim):
super().__init__()
self.fc1 = nn.Linear(z_dim, d_dim)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.fc1.bias.data.zero_()
def forward(self, zd):
h = F.relu(zd)
loc_d = self.fc1(h)
return loc_d
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4, 4]
|
ConditionalBatchNorm2d | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/7h/c7h472oam6vlzx7yb5mtc3kdy3p5vvd7jgap45mgevlf5geawq4f.py
# Topologically Sorted Source Nodes: [matmul, norm], Original ATen: [aten.mv, aten.linalg_vector_norm]
# Source node to ATen node mapping:
# matmul => mul_2, sum_1
# norm => pow_1, sum_2
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute, %primals_4), kwargs = {})
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_2, [1]), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_1, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, None), kwargs = {})
triton_per_fused_linalg_vector_norm_mv_0 = async_compile.triton('triton_per_fused_linalg_vector_norm_mv_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 4],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {4: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=(4,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_linalg_vector_norm_mv_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_linalg_vector_norm_mv_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (4 + r0), None)
tmp5 = tl.load(in_ptr1 + (1))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (8 + r0), None)
tmp10 = tl.load(in_ptr1 + (2))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (12 + r0), None)
tmp15 = tl.load(in_ptr1 + (3))
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.sum(tmp20, 1)[:, None]
tl.store(out_ptr0 + (tl.broadcast_to(r0, [XBLOCK, RBLOCK])), tmp18, None)
tl.store(out_ptr1 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp22, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vn/cvni2qgjd2juowltkhd53nvyjor5s5rs2db3273tfh2p6in366cc.py
# Topologically Sorted Source Nodes: [norm, add, v, matmul_1, norm_1, add_1, u, sigma], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div, aten.mv, aten.dot]
# Source node to ATen node mapping:
# add => add_1
# add_1 => add_2
# matmul_1 => mul_3, sum_3
# norm => pow_2
# norm_1 => pow_3, pow_4, sum_4
# sigma => mul_5, sum_6
# u => div_1
# v => div
# Graph fragment:
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 0.5), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_2, 0.0001), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, %add_1), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %div), kwargs = {})
# %sum_3 : [num_users=3] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_3, [1]), kwargs = {})
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_3, 2), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_3, None), kwargs = {})
# %pow_4 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_4, 0.5), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_4, 0.0001), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_3, %add_2), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_1, %sum_3), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%mul_5,), kwargs = {})
triton_per_fused_add_div_dot_linalg_vector_norm_mv_1 = async_compile.triton('triton_per_fused_add_div_dot_linalg_vector_norm_mv_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 4],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {4: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=(4,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_dot_linalg_vector_norm_mv_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 2, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_div_dot_linalg_vector_norm_mv_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (4*r0), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp3 = tl.load(in_ptr2 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp10 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr1 + (1))
tmp12 = tl.broadcast_to(tmp11, [XBLOCK, RBLOCK])
tmp16 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr1 + (2))
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp22 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr1 + (3))
tmp24 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp5 = libdevice.sqrt(tmp4)
tmp6 = 0.0001
tmp7 = tmp5 + tmp6
tmp8 = tmp2 / tmp7
tmp9 = tmp0 * tmp8
tmp13 = tmp12 / tmp7
tmp14 = tmp10 * tmp13
tmp15 = tmp9 + tmp14
tmp19 = tmp18 / tmp7
tmp20 = tmp16 * tmp19
tmp21 = tmp15 + tmp20
tmp25 = tmp24 / tmp7
tmp26 = tmp22 * tmp25
tmp27 = tmp21 + tmp26
tmp28 = tmp27 * tmp27
tmp29 = tl.broadcast_to(tmp28, [XBLOCK, RBLOCK])
tmp31 = tl.sum(tmp29, 1)[:, None]
tmp32 = libdevice.sqrt(tmp31)
tmp33 = tmp32 + tmp6
tmp34 = tmp27 / tmp33
tmp35 = tmp34 * tmp27
tmp36 = tl.broadcast_to(tmp35, [XBLOCK, RBLOCK])
tmp38 = tl.sum(tmp36, 1)[:, None]
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp38, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/qa/cqacjmg5thruiiympgmeg563tgu547tvricd27codfxv36x3vdcv.py
# Topologically Sorted Source Nodes: [truediv_2], Original ATen: [aten.div]
# Source node to ATen node mapping:
# truediv_2 => div_2
# Graph fragment:
# %div_2 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_5, %expand), kwargs = {})
triton_poi_fused_div_2 = async_compile.triton('triton_poi_fused_div_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_div_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 / tmp2
tl.store(out_ptr0 + (x0), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/7p/c7pbdjyinlr4jzoslxhdn74su3e6kah3nd373g2lidkiorfsskik.py
# Topologically Sorted Source Nodes: [out, mul, out_1], Original ATen: [aten._native_batch_norm_legit_no_training, aten.mul, aten.add]
# Source node to ATen node mapping:
# mul => mul_10
# out => mul_1, sub
# out_1 => add_6
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_1, %unsqueeze_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %unsqueeze_3), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_6, %mul_1), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_10, %view_7), kwargs = {})
triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3 = async_compile.triton('triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = (xindex // 16)
x4 = xindex
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_ptr0 + (x3), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x4), None)
tmp4 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (x1), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr4 + (x3), None, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 + tmp1
tmp5 = tmp3 - tmp4
tmp7 = 0.0001
tmp8 = tmp6 + tmp7
tmp9 = libdevice.sqrt(tmp8)
tmp10 = tl.full([1], 1, tl.int32)
tmp11 = tmp10 / tmp9
tmp12 = tmp11 * tmp1
tmp13 = tmp5 * tmp12
tmp14 = tmp2 * tmp13
tmp16 = tmp14 + tmp15
tl.store(out_ptr0 + (x4), tmp16, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8 = args
args.clear()
assert_size_stride(primals_1, (64, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf1 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [matmul, norm], Original ATen: [aten.mv, aten.linalg_vector_norm]
stream0 = get_raw_stream(0)
triton_per_fused_linalg_vector_norm_mv_0.run(primals_5, primals_4, buf0, buf1, 1, 4, grid=grid(1), stream=stream0)
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [norm, add, v, matmul_1, norm_1, add_1, u, sigma], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div, aten.mv, aten.dot]
triton_per_fused_add_div_dot_linalg_vector_norm_mv_1.run(buf4, primals_5, buf0, buf1, 1, 4, grid=grid(1), stream=stream0)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv_2], Original ATen: [aten.div]
triton_poi_fused_div_2.run(primals_5, buf4, buf5, 16, grid=grid(16), stream=stream0)
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), reinterpret_tensor(buf5, (4, 4), (1, 4), 0), out=buf6)
buf7 = buf0; del buf0 # reuse
buf8 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [matmul_2, norm_2], Original ATen: [aten.mv, aten.linalg_vector_norm]
triton_per_fused_linalg_vector_norm_mv_0.run(primals_8, primals_7, buf7, buf8, 1, 4, grid=grid(1), stream=stream0)
buf10 = buf1; del buf1 # reuse
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [norm_2, add_3, v_1, matmul_3, norm_3, add_4, u_1, sigma_1], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div, aten.mv, aten.dot]
triton_per_fused_add_div_dot_linalg_vector_norm_mv_1.run(buf11, primals_8, buf7, buf8, 1, 4, grid=grid(1), stream=stream0)
del buf7
del buf8
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv_5], Original ATen: [aten.div]
triton_poi_fused_div_2.run(primals_8, buf11, buf12, 16, grid=grid(16), stream=stream0)
del buf11
buf13 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [beta], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), reinterpret_tensor(buf12, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((64, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out, mul, out_1], Original ATen: [aten._native_batch_norm_legit_no_training, aten.mul, aten.add]
triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3.run(buf6, primals_1, primals_2, primals_3, buf13, buf14, 4096, grid=grid(4096), stream=stream0)
del buf13
del buf6
return (buf14, buf5, buf12, primals_1, primals_2, primals_3, primals_4, primals_5, primals_7, primals_8, reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
from torch.nn import Parameter
def l2normalize(v, eps=0.0001):
return v / (v.norm() + eps)
class SpectralNorm(nn.Module):
def __init__(self, module, name='weight', power_iterations=1):
super(SpectralNorm, self).__init__()
self.module = module
self.name = name
self.power_iterations = power_iterations
if not self._made_params():
self._make_params()
def _update_u_v(self):
u = getattr(self.module, self.name + '_u')
v = getattr(self.module, self.name + '_v')
w = getattr(self.module, self.name + '_bar')
height = w.data.shape[0]
_w = w.view(height, -1)
for _ in range(self.power_iterations):
v = l2normalize(torch.matmul(_w.t(), u))
u = l2normalize(torch.matmul(_w, v))
sigma = u.dot(_w.mv(v))
setattr(self.module, self.name, w / sigma.expand_as(w))
def _made_params(self):
try:
getattr(self.module, self.name + '_u')
getattr(self.module, self.name + '_v')
getattr(self.module, self.name + '_bar')
return True
except AttributeError:
return False
def _make_params(self):
w = getattr(self.module, self.name)
height = w.data.shape[0]
w.view(height, -1).data.shape[1]
u = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
v = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
u.data = l2normalize(u.data)
v.data = l2normalize(v.data)
w_bar = Parameter(w.data)
del self.module._parameters[self.name]
self.module.register_parameter(self.name + '_u', u)
self.module.register_parameter(self.name + '_v', v)
self.module.register_parameter(self.name + '_bar', w_bar)
def forward(self, *args):
self._update_u_v()
return self.module.forward(*args)
class ConditionalBatchNorm2d(nn.Module):
def __init__(self, num_features, num_classes, eps=0.0001, momentum=0.1):
super().__init__()
self.num_features = num_features
self.bn = nn.BatchNorm2d(num_features, affine=False, eps=eps,
momentum=momentum)
self.gamma_embed = SpectralNorm(nn.Linear(num_classes, num_features,
bias=False))
self.beta_embed = SpectralNorm(nn.Linear(num_classes, num_features,
bias=False))
def forward(self, x, y):
out = self.bn(x)
gamma = self.gamma_embed(y) + 1
beta = self.beta_embed(y)
out = gamma.view(-1, self.num_features, 1, 1) * out + beta.view(-1,
self.num_features, 1, 1)
return out
def get_inputs():
return [torch.rand([64, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_features': 4, 'num_classes': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
from torch.nn import Parameter
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_linalg_vector_norm_mv_0(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, rnumel, XBLOCK: tl.constexpr):
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (4 + r0), None)
tmp5 = tl.load(in_ptr1 + 1)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (8 + r0), None)
tmp10 = tl.load(in_ptr1 + 2)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (12 + r0), None)
tmp15 = tl.load(in_ptr1 + 3)
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.sum(tmp20, 1)[:, None]
tl.store(out_ptr0 + tl.broadcast_to(r0, [XBLOCK, RBLOCK]), tmp18, None)
tl.store(out_ptr1 + tl.full([XBLOCK, 1], 0, tl.int32), tmp22, None)
@triton.jit
def triton_per_fused_add_div_dot_linalg_vector_norm_mv_1(in_out_ptr0,
in_ptr0, in_ptr1, in_ptr2, xnumel, rnumel, XBLOCK: tl.constexpr):
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + 4 * r0, None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp3 = tl.load(in_ptr2 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp10 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr1 + 1)
tmp12 = tl.broadcast_to(tmp11, [XBLOCK, RBLOCK])
tmp16 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr1 + 2)
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp22 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr1 + 3)
tmp24 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp5 = libdevice.sqrt(tmp4)
tmp6 = 0.0001
tmp7 = tmp5 + tmp6
tmp8 = tmp2 / tmp7
tmp9 = tmp0 * tmp8
tmp13 = tmp12 / tmp7
tmp14 = tmp10 * tmp13
tmp15 = tmp9 + tmp14
tmp19 = tmp18 / tmp7
tmp20 = tmp16 * tmp19
tmp21 = tmp15 + tmp20
tmp25 = tmp24 / tmp7
tmp26 = tmp22 * tmp25
tmp27 = tmp21 + tmp26
tmp28 = tmp27 * tmp27
tmp29 = tl.broadcast_to(tmp28, [XBLOCK, RBLOCK])
tmp31 = tl.sum(tmp29, 1)[:, None]
tmp32 = libdevice.sqrt(tmp31)
tmp33 = tmp32 + tmp6
tmp34 = tmp27 / tmp33
tmp35 = tmp34 * tmp27
tmp36 = tl.broadcast_to(tmp35, [XBLOCK, RBLOCK])
tmp38 = tl.sum(tmp36, 1)[:, None]
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp38, None)
@triton.jit
def triton_poi_fused_div_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 / tmp2
tl.store(out_ptr0 + x0, tmp3, xmask)
@triton.jit
def triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3(in_ptr0,
in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex // 16
x4 = xindex
x1 = xindex // 16 % 4
tmp0 = tl.load(in_ptr0 + x3, None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x4, None)
tmp4 = tl.load(in_ptr2 + x1, None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + x1, None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr4 + x3, None, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 + tmp1
tmp5 = tmp3 - tmp4
tmp7 = 0.0001
tmp8 = tmp6 + tmp7
tmp9 = libdevice.sqrt(tmp8)
tmp10 = tl.full([1], 1, tl.int32)
tmp11 = tmp10 / tmp9
tmp12 = tmp11 * tmp1
tmp13 = tmp5 * tmp12
tmp14 = tmp2 * tmp13
tmp16 = tmp14 + tmp15
tl.store(out_ptr0 + x4, tmp16, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8) = args
args.clear()
assert_size_stride(primals_1, (64, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4,), (1,), torch.float32)
buf1 = empty_strided_cuda((), (), torch.float32)
get_raw_stream(0)
triton_per_fused_linalg_vector_norm_mv_0[grid(1)](primals_5,
primals_4, buf0, buf1, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3
del buf3
triton_per_fused_add_div_dot_linalg_vector_norm_mv_1[grid(1)](buf4,
primals_5, buf0, buf1, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_div_2[grid(16)](primals_5, buf4, buf5, 16, XBLOCK=
16, num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (64, 4), (4, 1), 0),
reinterpret_tensor(buf5, (4, 4), (1, 4), 0), out=buf6)
buf7 = buf0
del buf0
buf8 = buf4
del buf4
triton_per_fused_linalg_vector_norm_mv_0[grid(1)](primals_8,
primals_7, buf7, buf8, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
buf10 = buf1
del buf1
buf11 = buf10
del buf10
triton_per_fused_add_div_dot_linalg_vector_norm_mv_1[grid(1)](buf11,
primals_8, buf7, buf8, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del buf7
del buf8
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_div_2[grid(16)](primals_8, buf11, buf12, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del buf11
buf13 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (64, 4), (4, 1), 0),
reinterpret_tensor(buf12, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((64, 4, 4, 4), (64, 16, 4, 1), torch.float32
)
triton_poi_fused__native_batch_norm_legit_no_training_add_mul_3[grid
(4096)](buf6, primals_1, primals_2, primals_3, buf13, buf14,
4096, XBLOCK=256, num_warps=4, num_stages=1)
del buf13
del buf6
return (buf14, buf5, buf12, primals_1, primals_2, primals_3, primals_4,
primals_5, primals_7, primals_8, reinterpret_tensor(primals_6, (64,
4), (4, 1), 0))
def l2normalize(v, eps=0.0001):
return v / (v.norm() + eps)
class SpectralNorm(nn.Module):
def __init__(self, module, name='weight', power_iterations=1):
super(SpectralNorm, self).__init__()
self.module = module
self.name = name
self.power_iterations = power_iterations
if not self._made_params():
self._make_params()
def _update_u_v(self):
u = getattr(self.module, self.name + '_u')
v = getattr(self.module, self.name + '_v')
w = getattr(self.module, self.name + '_bar')
height = w.data.shape[0]
_w = w.view(height, -1)
for _ in range(self.power_iterations):
v = l2normalize(torch.matmul(_w.t(), u))
u = l2normalize(torch.matmul(_w, v))
sigma = u.dot(_w.mv(v))
setattr(self.module, self.name, w / sigma.expand_as(w))
def _made_params(self):
try:
getattr(self.module, self.name + '_u')
getattr(self.module, self.name + '_v')
getattr(self.module, self.name + '_bar')
return True
except AttributeError:
return False
def _make_params(self):
w = getattr(self.module, self.name)
height = w.data.shape[0]
w.view(height, -1).data.shape[1]
u = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
v = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
u.data = l2normalize(u.data)
v.data = l2normalize(v.data)
w_bar = Parameter(w.data)
del self.module._parameters[self.name]
self.module.register_parameter(self.name + '_u', u)
self.module.register_parameter(self.name + '_v', v)
self.module.register_parameter(self.name + '_bar', w_bar)
def forward(self, *args):
self._update_u_v()
return self.module.forward(*args)
class ConditionalBatchNorm2dNew(nn.Module):
def __init__(self, num_features, num_classes, eps=0.0001, momentum=0.1):
super().__init__()
self.num_features = num_features
self.bn = nn.BatchNorm2d(num_features, affine=False, eps=eps,
momentum=momentum)
self.gamma_embed = SpectralNorm(nn.Linear(num_classes, num_features,
bias=False))
self.beta_embed = SpectralNorm(nn.Linear(num_classes, num_features,
bias=False))
def forward(self, input_0, input_1):
primals_2 = self.gamma_embed.module.weight_u
primals_3 = self.gamma_embed.module.weight_v
primals_5 = self.gamma_embed.module.weight_bar
primals_4 = self.beta_embed.module.weight_u
primals_7 = self.beta_embed.module.weight_v
primals_8 = self.beta_embed.module.weight_bar
primals_1 = input_0
primals_6 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0]
| samuelemarro/anne | ConditionalBatchNorm2d | false | 4,267 | [
"MIT"
] | 0 | 918022eb029a46fbfd1589369e9817f570d5651c | https://github.com/samuelemarro/anne/tree/918022eb029a46fbfd1589369e9817f570d5651c | import torch
import torch.nn as nn
from torch.nn import Parameter
def l2normalize(v, eps=0.0001):
return v / (v.norm() + eps)
class SpectralNorm(nn.Module):
def __init__(self, module, name='weight', power_iterations=1):
super().__init__()
self.module = module
self.name = name
self.power_iterations = power_iterations
if not self._made_params():
self._make_params()
def _update_u_v(self):
u = getattr(self.module, self.name + '_u')
v = getattr(self.module, self.name + '_v')
w = getattr(self.module, self.name + '_bar')
height = w.data.shape[0]
_w = w.view(height, -1)
for _ in range(self.power_iterations):
v = l2normalize(torch.matmul(_w.t(), u))
u = l2normalize(torch.matmul(_w, v))
sigma = u.dot(_w.mv(v))
setattr(self.module, self.name, w / sigma.expand_as(w))
def _made_params(self):
try:
getattr(self.module, self.name + '_u')
getattr(self.module, self.name + '_v')
getattr(self.module, self.name + '_bar')
return True
except AttributeError:
return False
def _make_params(self):
w = getattr(self.module, self.name)
height = w.data.shape[0]
w.view(height, -1).data.shape[1]
u = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
v = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
u.data = l2normalize(u.data)
v.data = l2normalize(v.data)
w_bar = Parameter(w.data)
del self.module._parameters[self.name]
self.module.register_parameter(self.name + '_u', u)
self.module.register_parameter(self.name + '_v', v)
self.module.register_parameter(self.name + '_bar', w_bar)
def forward(self, *args):
self._update_u_v()
return self.module.forward(*args)
class Model(nn.Module):
def __init__(self, num_features, num_classes, eps=0.0001, momentum=0.1):
super().__init__()
self.num_features = num_features
self.bn = nn.BatchNorm2d(num_features, affine=False, eps=eps,
momentum=momentum)
self.gamma_embed = SpectralNorm(nn.Linear(num_classes, num_features,
bias=False))
self.beta_embed = SpectralNorm(nn.Linear(num_classes, num_features,
bias=False))
def forward(self, x, y):
out = self.bn(x)
gamma = self.gamma_embed(y) + 1
beta = self.beta_embed(y)
out = gamma.view(-1, self.num_features, 1, 1) * out + beta.view(-1,
self.num_features, 1, 1)
return out
def get_inputs():
return [torch.rand([64, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
GlobalAvgPool1d | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/an/canusqrnw4njor7uwvf6vo7b6joi5xh6pmd66qfugkeeoq5wo34u.py
# Topologically Sorted Source Nodes: [avg_pool1d], Original ATen: [aten.avg_pool2d]
# Source node to ATen node mapping:
# avg_pool1d => avg_pool2d
# Graph fragment:
# %avg_pool2d : [num_users=1] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%unsqueeze, [1, 4], [1, 4]), kwargs = {})
triton_poi_fused_avg_pool2d_0 = async_compile.triton('triton_poi_fused_avg_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_avg_pool2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tl.store(out_ptr0 + (x0), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
# Topologically Sorted Source Nodes: [avg_pool1d], Original ATen: [aten.avg_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_avg_pool2d_0.run(arg0_1, buf0, 16, grid=grid(16), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf0, (4, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from abc import abstractmethod
from torch.nn import functional
class AvgPool(nn.Module):
"""
AvgPool Module.
"""
def __init__(self):
super().__init__()
@abstractmethod
def forward(self, input_tensor):
pass
class GlobalAvgPool1d(AvgPool):
"""
GlobalAvgPool1d Module.
"""
def forward(self, input_tensor):
return functional.avg_pool1d(input_tensor, input_tensor.size()[2:]
).view(input_tensor.size()[:2])
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from abc import abstractmethod
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tl.store(out_ptr0 + x0, tmp8, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_avg_pool2d_0[grid(16)](arg0_1, buf0, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del arg0_1
return reinterpret_tensor(buf0, (4, 4), (4, 1), 0),
class AvgPool(nn.Module):
"""
AvgPool Module.
"""
def __init__(self):
super().__init__()
@abstractmethod
def forward(self, input_tensor):
pass
class GlobalAvgPool1dNew(AvgPool):
"""
GlobalAvgPool1d Module.
"""
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| savan77/nni | GlobalAvgPool1d | false | 4,268 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
from abc import abstractmethod
from torch.nn import functional
class AvgPool(nn.Module):
"""
AvgPool Module.
"""
def __init__(self):
super().__init__()
@abstractmethod
def forward(self, input_tensor):
pass
class Model(AvgPool):
"""
GlobalAvgPool1d Module.
"""
def forward(self, input_tensor):
return functional.avg_pool1d(input_tensor, input_tensor.size()[2:]
).view(input_tensor.size()[:2])
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return []
|
DeepQNetwork | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/dw/cdw2zafobuhspvpds5vqtjyxlmdcpljwkwnnptboqhmdn4cucexl.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%squeeze,), kwargs = {})
# %le_3 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%squeeze_3, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
tl.store(out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/d6/cd6bapa2aseihiysk43m4ehqqsq6x6isa2ykn4ebi4nl4fn2uk5x.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_1 => relu_1
# Graph fragment:
# %relu_1 : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%squeeze_4,), kwargs = {})
# %le_2 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%squeeze_7, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
tl.store(out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/65/c65tc5xm2a3caq7ssexgjgocmcydeqgh76csqhvewry2b7btv4fj.py
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_4 => relu_3
# Graph fragment:
# %add_tensor : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default, %primals_9), kwargs = {})
# %relu_3 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor,), kwargs = {})
triton_poi_fused_relu_2 = async_compile.triton('triton_poi_fused_relu_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 2048
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (32, 4, 9, 9), (324, 81, 9, 1))
assert_size_stride(primals_3, (32, ), (1, ))
assert_size_stride(primals_4, (64, 32, 5, 5), (800, 25, 5, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (64, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_7, (64, ), (1, ))
assert_size_stride(primals_8, (512, 16), (16, 1))
assert_size_stride(primals_9, (512, ), (1, ))
assert_size_stride(primals_10, (4, 512), (512, 1))
assert_size_stride(primals_11, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1, 4, 4, 4), (64, 16, 4, 1), 0), primals_2, stride=(4, 4), padding=(4, 4), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (1, 32, 1, 1), (32, 1, 1, 1))
buf1 = reinterpret_tensor(buf0, (32, 1, 1), (1, 1, 1), 0); del buf0 # reuse
buf11 = empty_strided_cuda((32, 1, 1), (1, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_3, buf11, 32, grid=grid(32), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 32, 1, 1), (0, 1, 0, 0), 0), primals_4, stride=(2, 2), padding=(2, 2), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (1, 64, 1, 1), (64, 1, 1, 1))
buf3 = reinterpret_tensor(buf2, (64, 1, 1), (1, 1, 1), 0); del buf2 # reuse
buf10 = empty_strided_cuda((64, 1, 1), (1, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf10, 64, grid=grid(64), stream=stream0)
del primals_5
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(reinterpret_tensor(buf3, (1, 64, 1, 1), (0, 1, 0, 0), 0), primals_6, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (1, 64, 1, 1), (64, 1, 1, 1))
buf5 = reinterpret_tensor(buf4, (64, 1, 1), (1, 1, 1), 0); del buf4 # reuse
buf9 = empty_strided_cuda((64, 1, 1), (1, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf5, primals_7, buf9, 64, grid=grid(64), stream=stream0)
del primals_7
buf6 = empty_strided_cuda((4, 512), (512, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf5, (4, 16), (16, 1), 0), reinterpret_tensor(primals_8, (16, 512), (1, 16), 0), out=buf6)
buf7 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
triton_poi_fused_relu_2.run(buf7, primals_9, 2048, grid=grid(2048), stream=stream0)
del primals_9
buf8 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, buf7, reinterpret_tensor(primals_10, (512, 4), (1, 512), 0), alpha=1, beta=1, out=buf8)
del primals_11
return (buf8, primals_2, primals_4, primals_6, reinterpret_tensor(primals_1, (1, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf1, (1, 32, 1, 1), (32, 1, 1, 1), 0), reinterpret_tensor(buf3, (1, 64, 1, 1), (64, 1, 1, 1), 0), reinterpret_tensor(buf5, (4, 16), (16, 1), 0), buf7, primals_10, primals_8, buf9, buf10, buf11, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((32, 4, 9, 9), (324, 81, 9, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 32, 5, 5), (800, 25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((64, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((512, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 512), (512, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class DeepQNetwork(nn.Module):
def __init__(self, imagesize, num_input_frames, num_actions, **kwargs):
super(DeepQNetwork, self).__init__()
self.conv1 = nn.Conv2d(in_channels=num_input_frames, out_channels=
32, kernel_size=9, stride=4, padding=4)
self.conv2 = nn.Conv2d(in_channels=32, out_channels=64, kernel_size
=5, stride=2, padding=2)
self.conv3 = nn.Conv2d(in_channels=64, out_channels=64, kernel_size
=3, stride=1, padding=1)
in_features = 64 * imagesize[0] * imagesize[1] // 64
self.fc1 = nn.Linear(in_features=in_features, out_features=512)
self.fc2 = nn.Linear(in_features=512, out_features=num_actions)
def forward(self, x):
batch_size = x.size(0)
x = F.relu(self.conv1(x), inplace=True)
x = F.relu(self.conv2(x), inplace=True)
x = F.relu(self.conv3(x), inplace=True)
x = x.view(batch_size, -1)
x = F.relu(self.fc1(x), inplace=True)
x = self.fc2(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'imagesize': [4, 4], 'num_input_frames': 4, 'num_actions': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x0, tmp4, xmask)
tl.store(out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x0, tmp4, xmask)
tl.store(out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 512
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (32, 4, 9, 9), (324, 81, 9, 1))
assert_size_stride(primals_3, (32,), (1,))
assert_size_stride(primals_4, (64, 32, 5, 5), (800, 25, 5, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (64, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_7, (64,), (1,))
assert_size_stride(primals_8, (512, 16), (16, 1))
assert_size_stride(primals_9, (512,), (1,))
assert_size_stride(primals_10, (4, 512), (512, 1))
assert_size_stride(primals_11, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1,
4, 4, 4), (64, 16, 4, 1), 0), primals_2, stride=(4, 4), padding
=(4, 4), dilation=(1, 1), transposed=False, output_padding=(0,
0), groups=1, bias=None)
assert_size_stride(buf0, (1, 32, 1, 1), (32, 1, 1, 1))
buf1 = reinterpret_tensor(buf0, (32, 1, 1), (1, 1, 1), 0)
del buf0
buf11 = empty_strided_cuda((32, 1, 1), (1, 1, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(32)](buf1,
primals_3, buf11, 32, XBLOCK=32, num_warps=1, num_stages=1)
del primals_3
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 32,
1, 1), (0, 1, 0, 0), 0), primals_4, stride=(2, 2), padding=(2,
2), dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=1, bias=None)
assert_size_stride(buf2, (1, 64, 1, 1), (64, 1, 1, 1))
buf3 = reinterpret_tensor(buf2, (64, 1, 1), (1, 1, 1), 0)
del buf2
buf10 = empty_strided_cuda((64, 1, 1), (1, 1, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(64)](buf3,
primals_5, buf10, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_5
buf4 = extern_kernels.convolution(reinterpret_tensor(buf3, (1, 64,
1, 1), (0, 1, 0, 0), 0), primals_6, stride=(1, 1), padding=(1,
1), dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=1, bias=None)
assert_size_stride(buf4, (1, 64, 1, 1), (64, 1, 1, 1))
buf5 = reinterpret_tensor(buf4, (64, 1, 1), (1, 1, 1), 0)
del buf4
buf9 = empty_strided_cuda((64, 1, 1), (1, 1, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(64)](buf5,
primals_7, buf9, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_7
buf6 = empty_strided_cuda((4, 512), (512, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf5, (4, 16), (16, 1), 0),
reinterpret_tensor(primals_8, (16, 512), (1, 16), 0), out=buf6)
buf7 = buf6
del buf6
triton_poi_fused_relu_2[grid(2048)](buf7, primals_9, 2048, XBLOCK=
256, num_warps=4, num_stages=1)
del primals_9
buf8 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_11, buf7, reinterpret_tensor(
primals_10, (512, 4), (1, 512), 0), alpha=1, beta=1, out=buf8)
del primals_11
return buf8, primals_2, primals_4, primals_6, reinterpret_tensor(primals_1,
(1, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf1, (1, 32,
1, 1), (32, 1, 1, 1), 0), reinterpret_tensor(buf3, (1, 64, 1, 1), (
64, 1, 1, 1), 0), reinterpret_tensor(buf5, (4, 16), (16, 1), 0
), buf7, primals_10, primals_8, buf9, buf10, buf11
class DeepQNetworkNew(nn.Module):
def __init__(self, imagesize, num_input_frames, num_actions, **kwargs):
super(DeepQNetworkNew, self).__init__()
self.conv1 = nn.Conv2d(in_channels=num_input_frames, out_channels=
32, kernel_size=9, stride=4, padding=4)
self.conv2 = nn.Conv2d(in_channels=32, out_channels=64, kernel_size
=5, stride=2, padding=2)
self.conv3 = nn.Conv2d(in_channels=64, out_channels=64, kernel_size
=3, stride=1, padding=1)
in_features = 64 * imagesize[0] * imagesize[1] // 64
self.fc1 = nn.Linear(in_features=in_features, out_features=512)
self.fc2 = nn.Linear(in_features=512, out_features=num_actions)
def forward(self, input_0):
primals_2 = self.conv1.weight
primals_3 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_6 = self.conv3.weight
primals_7 = self.conv3.bias
primals_8 = self.fc1.weight
primals_9 = self.fc1.bias
primals_10 = self.fc2.weight
primals_11 = self.fc2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
| sanmusane/AIGames | DeepQNetwork | false | 4,269 | [
"MIT"
] | 0 | 3f4eecdd02089911d1989e40e2b336e13b800e55 | https://github.com/sanmusane/AIGames/tree/3f4eecdd02089911d1989e40e2b336e13b800e55 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, imagesize, num_input_frames, num_actions, **kwargs):
super().__init__()
self.conv1 = nn.Conv2d(in_channels=num_input_frames, out_channels=
32, kernel_size=9, stride=4, padding=4)
self.conv2 = nn.Conv2d(in_channels=32, out_channels=64, kernel_size
=5, stride=2, padding=2)
self.conv3 = nn.Conv2d(in_channels=64, out_channels=64, kernel_size
=3, stride=1, padding=1)
in_features = 64 * imagesize[0] * imagesize[1] // 64
self.fc1 = nn.Linear(in_features=in_features, out_features=512)
self.fc2 = nn.Linear(in_features=512, out_features=num_actions)
def forward(self, x):
batch_size = x.size(0)
x = F.relu(self.conv1(x), inplace=True)
x = F.relu(self.conv2(x), inplace=True)
x = F.relu(self.conv3(x), inplace=True)
x = x.view(batch_size, -1)
x = F.relu(self.fc1(x), inplace=True)
x = self.fc2(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Mask | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/pl/cpls7julgyzyzgsc5ycrh5sravin2piuyc3s5guflad7adet6qmj.py
# Topologically Sorted Source Nodes: [eq, zeros_like, where], Original ATen: [aten.eq, aten.zeros_like, aten.where]
# Source node to ATen node mapping:
# eq => eq
# where => where
# zeros_like => full_default
# Graph fragment:
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%permute, 1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq, %arg1_1, %full_default), kwargs = {})
triton_poi_fused_eq_where_zeros_like_0 = async_compile.triton('triton_poi_fused_eq_where_zeros_like_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_eq_where_zeros_like_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_eq_where_zeros_like_0(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2 + (4*y0)), xmask & ymask, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 == tmp1
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tl.store(out_ptr0 + (y0 + (4*x2) + (16*y1)), tmp5, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4), (4, 1))
assert_size_stride(arg1_1, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 1, 4), torch.float32)
# Topologically Sorted Source Nodes: [eq, zeros_like, where], Original ATen: [aten.eq, aten.zeros_like, aten.where]
stream0 = get_raw_stream(0)
triton_poi_fused_eq_where_zeros_like_0.run(arg0_1, arg1_1, buf0, 16, 4, grid=grid(16, 4), stream=stream0)
del arg0_1
del arg1_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class Mask(nn.Module):
def forward(self, seq, mask):
seq_mask = torch.unsqueeze(mask, 2)
seq_mask = torch.transpose(seq_mask.repeat(1, 1, seq.size()[1]), 1, 2)
return seq.where(torch.eq(seq_mask, 1), torch.zeros_like(seq))
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_eq_where_zeros_like_0(in_ptr0, in_ptr1, out_ptr0,
ynumel, xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (x2 + 4 * y1), xmask & ymask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr1 + (x2 + 4 * y0), xmask & ymask, eviction_policy=
'evict_last')
tmp1 = 1.0
tmp2 = tmp0 == tmp1
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tl.store(out_ptr0 + (y0 + 4 * x2 + 16 * y1), tmp5, xmask & ymask)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4), (4, 1))
assert_size_stride(arg1_1, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused_eq_where_zeros_like_0[grid(16, 4)](arg0_1, arg1_1,
buf0, 16, 4, XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class MaskNew(nn.Module):
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| savan77/nni | Mask | false | 4,270 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class Model(nn.Module):
def forward(self, seq, mask):
seq_mask = torch.unsqueeze(mask, 2)
seq_mask = torch.transpose(seq_mask.repeat(1, 1, seq.size()[1]), 1, 2)
return seq.where(torch.eq(seq_mask, 1), torch.zeros_like(seq))
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return []
|
Pooling | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/dz/cdz3nkgyrhben4dg5ahsmw55wko3y32durc6eb6vfqmjdr6gb3ir.py
# Topologically Sorted Source Nodes: [avg_pool2d], Original ATen: [aten.avg_pool2d]
# Source node to ATen node mapping:
# avg_pool2d => avg_pool2d
# Graph fragment:
# %avg_pool2d : [num_users=1] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%arg0_1, [3, 3], [1, 1], [1, 1], False, False), kwargs = {})
triton_poi_fused_avg_pool2d_0 = async_compile.triton('triton_poi_fused_avg_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_avg_pool2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4) % 4
x0 = xindex % 4
x4 = xindex
tmp0 = (-1) + x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tmp2 & tmp4
tmp6 = (-1) + x0
tmp7 = tmp6 >= tmp1
tmp8 = tmp6 < tmp3
tmp9 = tmp7 & tmp8
tmp10 = tmp5 & tmp9
tmp11 = tl.load(in_ptr0 + ((-5) + x4), tmp10 & xmask, other=0.0)
tmp12 = x0
tmp13 = tmp12 >= tmp1
tmp14 = tmp12 < tmp3
tmp15 = tmp13 & tmp14
tmp16 = tmp5 & tmp15
tmp17 = tl.load(in_ptr0 + ((-4) + x4), tmp16 & xmask, other=0.0)
tmp18 = tmp17 + tmp11
tmp19 = 1 + x0
tmp20 = tmp19 >= tmp1
tmp21 = tmp19 < tmp3
tmp22 = tmp20 & tmp21
tmp23 = tmp5 & tmp22
tmp24 = tl.load(in_ptr0 + ((-3) + x4), tmp23 & xmask, other=0.0)
tmp25 = tmp24 + tmp18
tmp26 = x1
tmp27 = tmp26 >= tmp1
tmp28 = tmp26 < tmp3
tmp29 = tmp27 & tmp28
tmp30 = tmp29 & tmp9
tmp31 = tl.load(in_ptr0 + ((-1) + x4), tmp30 & xmask, other=0.0)
tmp32 = tmp31 + tmp25
tmp33 = tmp29 & tmp15
tmp34 = tl.load(in_ptr0 + (x4), tmp33 & xmask, other=0.0)
tmp35 = tmp34 + tmp32
tmp36 = tmp29 & tmp22
tmp37 = tl.load(in_ptr0 + (1 + x4), tmp36 & xmask, other=0.0)
tmp38 = tmp37 + tmp35
tmp39 = 1 + x1
tmp40 = tmp39 >= tmp1
tmp41 = tmp39 < tmp3
tmp42 = tmp40 & tmp41
tmp43 = tmp42 & tmp9
tmp44 = tl.load(in_ptr0 + (3 + x4), tmp43 & xmask, other=0.0)
tmp45 = tmp44 + tmp38
tmp46 = tmp42 & tmp15
tmp47 = tl.load(in_ptr0 + (4 + x4), tmp46 & xmask, other=0.0)
tmp48 = tmp47 + tmp45
tmp49 = tmp42 & tmp22
tmp50 = tl.load(in_ptr0 + (5 + x4), tmp49 & xmask, other=0.0)
tmp51 = tmp50 + tmp48
tmp52 = (((0) * ((0) >= ((-1) + x0)) + ((-1) + x0) * (((-1) + x0) > (0)))*((0) * ((0) >= ((-1) + x1)) + ((-1) + x1) * (((-1) + x1) > (0)))) + (((4) * ((4) <= (2 + x0)) + (2 + x0) * ((2 + x0) < (4)))*((4) * ((4) <= (2 + x1)) + (2 + x1) * ((2 + x1) < (4)))) + ((-1)*((0) * ((0) >= ((-1) + x0)) + ((-1) + x0) * (((-1) + x0) > (0)))*((4) * ((4) <= (2 + x1)) + (2 + x1) * ((2 + x1) < (4)))) + ((-1)*((0) * ((0) >= ((-1) + x1)) + ((-1) + x1) * (((-1) + x1) > (0)))*((4) * ((4) <= (2 + x0)) + (2 + x0) * ((2 + x0) < (4))))
tmp53 = tmp51 / tmp52
tl.store(out_ptr0 + (x4), tmp53, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [avg_pool2d], Original ATen: [aten.avg_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_avg_pool2d_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class ReLUConvBN(nn.Module):
"""
Parameters
---
C_in: int
the number of input channels
C_out: int
the number of output channels
stride: int
stride of the convolution
padding: int
zero-padding added to both sides of the input
dilation: int
spacing between kernel elements
bn_affine: bool
If set to ``True``, ``torch.nn.BatchNorm2d`` will have learnable affine parameters. Default: True
bn_momentun: float
the value used for the running_mean and running_var computation. Default: 0.1
bn_track_running_stats: bool
When set to ``True``, ``torch.nn.BatchNorm2d`` tracks the running mean and variance. Default: True
"""
def __init__(self, C_in, C_out, kernel_size, stride, padding, dilation,
bn_affine=True, bn_momentum=0.1, bn_track_running_stats=True):
super(ReLUConvBN, self).__init__()
self.op = nn.Sequential(nn.ReLU(inplace=False), nn.Conv2d(C_in,
C_out, kernel_size, stride=stride, padding=padding, dilation=
dilation, bias=False), nn.BatchNorm2d(C_out, affine=bn_affine,
momentum=bn_momentum, track_running_stats=bn_track_running_stats))
def forward(self, x):
"""
Parameters
---
x: torch.Tensor
input tensor
"""
return self.op(x)
class Pooling(nn.Module):
"""
Parameters
---
C_in: int
the number of input channels
C_out: int
the number of output channels
stride: int
stride of the convolution
bn_affine: bool
If set to ``True``, ``torch.nn.BatchNorm2d`` will have learnable affine parameters. Default: True
bn_momentun: float
the value used for the running_mean and running_var computation. Default: 0.1
bn_track_running_stats: bool
When set to ``True``, ``torch.nn.BatchNorm2d`` tracks the running mean and variance. Default: True
"""
def __init__(self, C_in, C_out, stride, bn_affine=True, bn_momentum=0.1,
bn_track_running_stats=True):
super(Pooling, self).__init__()
if C_in == C_out:
self.preprocess = None
else:
self.preprocess = ReLUConvBN(C_in, C_out, 1, 1, 0, 0, bn_affine,
bn_momentum, bn_track_running_stats)
self.op = nn.AvgPool2d(3, stride=stride, padding=1,
count_include_pad=False)
def forward(self, x):
"""
Parameters
---
x: torch.Tensor
input tensor
"""
if self.preprocess:
x = self.preprocess(x)
return self.op(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'C_in': 4, 'C_out': 4, 'stride': 1}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4 % 4
x0 = xindex % 4
x4 = xindex
tmp0 = -1 + x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tmp2 & tmp4
tmp6 = -1 + x0
tmp7 = tmp6 >= tmp1
tmp8 = tmp6 < tmp3
tmp9 = tmp7 & tmp8
tmp10 = tmp5 & tmp9
tmp11 = tl.load(in_ptr0 + (-5 + x4), tmp10 & xmask, other=0.0)
tmp12 = x0
tmp13 = tmp12 >= tmp1
tmp14 = tmp12 < tmp3
tmp15 = tmp13 & tmp14
tmp16 = tmp5 & tmp15
tmp17 = tl.load(in_ptr0 + (-4 + x4), tmp16 & xmask, other=0.0)
tmp18 = tmp17 + tmp11
tmp19 = 1 + x0
tmp20 = tmp19 >= tmp1
tmp21 = tmp19 < tmp3
tmp22 = tmp20 & tmp21
tmp23 = tmp5 & tmp22
tmp24 = tl.load(in_ptr0 + (-3 + x4), tmp23 & xmask, other=0.0)
tmp25 = tmp24 + tmp18
tmp26 = x1
tmp27 = tmp26 >= tmp1
tmp28 = tmp26 < tmp3
tmp29 = tmp27 & tmp28
tmp30 = tmp29 & tmp9
tmp31 = tl.load(in_ptr0 + (-1 + x4), tmp30 & xmask, other=0.0)
tmp32 = tmp31 + tmp25
tmp33 = tmp29 & tmp15
tmp34 = tl.load(in_ptr0 + x4, tmp33 & xmask, other=0.0)
tmp35 = tmp34 + tmp32
tmp36 = tmp29 & tmp22
tmp37 = tl.load(in_ptr0 + (1 + x4), tmp36 & xmask, other=0.0)
tmp38 = tmp37 + tmp35
tmp39 = 1 + x1
tmp40 = tmp39 >= tmp1
tmp41 = tmp39 < tmp3
tmp42 = tmp40 & tmp41
tmp43 = tmp42 & tmp9
tmp44 = tl.load(in_ptr0 + (3 + x4), tmp43 & xmask, other=0.0)
tmp45 = tmp44 + tmp38
tmp46 = tmp42 & tmp15
tmp47 = tl.load(in_ptr0 + (4 + x4), tmp46 & xmask, other=0.0)
tmp48 = tmp47 + tmp45
tmp49 = tmp42 & tmp22
tmp50 = tl.load(in_ptr0 + (5 + x4), tmp49 & xmask, other=0.0)
tmp51 = tmp50 + tmp48
tmp52 = (0 * (0 >= -1 + x0) + (-1 + x0) * (-1 + x0 > 0)) * (0 * (0 >= -
1 + x1) + (-1 + x1) * (-1 + x1 > 0)) + (4 * (4 <= 2 + x0) + (2 + x0
) * (2 + x0 < 4)) * (4 * (4 <= 2 + x1) + (2 + x1) * (2 + x1 < 4)
) + -1 * (0 * (0 >= -1 + x0) + (-1 + x0) * (-1 + x0 > 0)) * (4 * (4 <=
2 + x1) + (2 + x1) * (2 + x1 < 4)) + -1 * (0 * (0 >= -1 + x1) + (-1 +
x1) * (-1 + x1 > 0)) * (4 * (4 <= 2 + x0) + (2 + x0) * (2 + x0 < 4))
tmp53 = tmp51 / tmp52
tl.store(out_ptr0 + x4, tmp53, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_avg_pool2d_0[grid(256)](arg0_1, buf0, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class ReLUConvBN(nn.Module):
"""
Parameters
---
C_in: int
the number of input channels
C_out: int
the number of output channels
stride: int
stride of the convolution
padding: int
zero-padding added to both sides of the input
dilation: int
spacing between kernel elements
bn_affine: bool
If set to ``True``, ``torch.nn.BatchNorm2d`` will have learnable affine parameters. Default: True
bn_momentun: float
the value used for the running_mean and running_var computation. Default: 0.1
bn_track_running_stats: bool
When set to ``True``, ``torch.nn.BatchNorm2d`` tracks the running mean and variance. Default: True
"""
def __init__(self, C_in, C_out, kernel_size, stride, padding, dilation,
bn_affine=True, bn_momentum=0.1, bn_track_running_stats=True):
super(ReLUConvBN, self).__init__()
self.op = nn.Sequential(nn.ReLU(inplace=False), nn.Conv2d(C_in,
C_out, kernel_size, stride=stride, padding=padding, dilation=
dilation, bias=False), nn.BatchNorm2d(C_out, affine=bn_affine,
momentum=bn_momentum, track_running_stats=bn_track_running_stats))
def forward(self, x):
"""
Parameters
---
x: torch.Tensor
input tensor
"""
return self.op(x)
class PoolingNew(nn.Module):
"""
Parameters
---
C_in: int
the number of input channels
C_out: int
the number of output channels
stride: int
stride of the convolution
bn_affine: bool
If set to ``True``, ``torch.nn.BatchNorm2d`` will have learnable affine parameters. Default: True
bn_momentun: float
the value used for the running_mean and running_var computation. Default: 0.1
bn_track_running_stats: bool
When set to ``True``, ``torch.nn.BatchNorm2d`` tracks the running mean and variance. Default: True
"""
def __init__(self, C_in, C_out, stride, bn_affine=True, bn_momentum=0.1,
bn_track_running_stats=True):
super(PoolingNew, self).__init__()
if C_in == C_out:
self.preprocess = None
else:
self.preprocess = ReLUConvBN(C_in, C_out, 1, 1, 0, 0, bn_affine,
bn_momentum, bn_track_running_stats)
self.op = nn.AvgPool2d(3, stride=stride, padding=1,
count_include_pad=False)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| savan77/nni | Pooling | false | 4,271 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class ReLUConvBN(nn.Module):
"""
Parameters
---
C_in: int
the number of input channels
C_out: int
the number of output channels
stride: int
stride of the convolution
padding: int
zero-padding added to both sides of the input
dilation: int
spacing between kernel elements
bn_affine: bool
If set to ``True``, ``torch.nn.BatchNorm2d`` will have learnable affine parameters. Default: True
bn_momentun: float
the value used for the running_mean and running_var computation. Default: 0.1
bn_track_running_stats: bool
When set to ``True``, ``torch.nn.BatchNorm2d`` tracks the running mean and variance. Default: True
"""
def __init__(self, C_in, C_out, kernel_size, stride, padding, dilation,
bn_affine=True, bn_momentum=0.1, bn_track_running_stats=True):
super().__init__()
self.op = nn.Sequential(nn.ReLU(inplace=False), nn.Conv2d(C_in,
C_out, kernel_size, stride=stride, padding=padding, dilation=
dilation, bias=False), nn.BatchNorm2d(C_out, affine=bn_affine,
momentum=bn_momentum, track_running_stats=bn_track_running_stats))
def forward(self, x):
"""
Parameters
---
x: torch.Tensor
input tensor
"""
return self.op(x)
class Model(nn.Module):
"""
Parameters
---
C_in: int
the number of input channels
C_out: int
the number of output channels
stride: int
stride of the convolution
bn_affine: bool
If set to ``True``, ``torch.nn.BatchNorm2d`` will have learnable affine parameters. Default: True
bn_momentun: float
the value used for the running_mean and running_var computation. Default: 0.1
bn_track_running_stats: bool
When set to ``True``, ``torch.nn.BatchNorm2d`` tracks the running mean and variance. Default: True
"""
def __init__(self, C_in, C_out, stride, bn_affine=True, bn_momentum=0.1,
bn_track_running_stats=True):
super().__init__()
if C_in == C_out:
self.preprocess = None
else:
self.preprocess = ReLUConvBN(C_in, C_out, 1, 1, 0, 0, bn_affine,
bn_momentum, bn_track_running_stats)
self.op = nn.AvgPool2d(3, stride=stride, padding=1,
count_include_pad=False)
def forward(self, x):
"""
Parameters
---
x: torch.Tensor
input tensor
"""
if self.preprocess:
x = self.preprocess(x)
return self.op(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 1]
|
BackboneModel1 | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/vg/cvgzll7advxze7fwtfxuvvxp6awpd565f4oliajayj6ukdru5c2v.py
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), None)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tl.store(in_out_ptr0 + (x0), tmp3, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_2, (1, ), (1, ))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 1, 64, 64), (4096, 4096, 64, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_2, 16384, grid=grid(16384), stream=stream0)
del primals_2
return (buf1, primals_1, primals_3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((1, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 1, 64, 64), (4096, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class BackboneModel1(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 1, 1, 1)
def forward(self, x):
return self.conv1(x)
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, None)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tl.store(in_out_ptr0 + x0, tmp3, None)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_2, (1,), (1,))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 1, 64, 64), (4096, 4096, 64, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(16384)](buf1, primals_2, 16384,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
return buf1, primals_1, primals_3
class BackboneModel1New(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 1, 1, 1)
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| savan77/nni | BackboneModel1 | false | 4,272 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class Model(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 1, 1, 1)
def forward(self, x):
return self.conv1(x)
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return []
|
InteractiveKLLoss | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/mc/cmc44gqwlbgitm3uqkuiwz6fe3jirwculg7zmyndeuqzyyqzyok7.py
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => exp_1
# Graph fragment:
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, 1), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, 4), kwargs = {})
# %exp_1 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp3 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = 0.25
tmp16 = tmp14 * tmp15
tmp17 = tl_math.exp(tmp16)
tl.store(out_ptr0 + (x3), tmp17, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xg/cxg6geasclvgycjnyaybokxud5rdp2fe6eropfaplher4ysvlw4g.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %mul_tensor_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 1), kwargs = {})
# %amax_default_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor_1, [1], True), kwargs = {})
# %sub_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor_1, %amax_default_1), kwargs = {})
# %div_tensor_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor_1, 4), kwargs = {})
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp3 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = 0.25
tmp16 = tmp14 * tmp15
tl.store(out_ptr0 + (x3), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/25/c252yxkeoy2jzoudseyd3vkmxj6p5ehtiqnttglx2n27knsfiyad.py
# Topologically Sorted Source Nodes: [softmax, kl_div, log_softmax], Original ATen: [aten._softmax, aten.xlogy, aten._log_softmax, aten.mul, aten.sub, aten.mean]
# Source node to ATen node mapping:
# kl_div => eq, full_default, full_default_1, isnan, log_1, mean, mul, mul_1, sub_3, where, where_1
# log_softmax => exp, log, sub_1, sum_1
# softmax => div_2, sum_2
# Graph fragment:
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_1, [1], True), kwargs = {})
# %div_2 : [num_users=5] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp_1, %sum_2), kwargs = {})
# %isnan : [num_users=1] = call_function[target=torch.ops.aten.isnan.default](args = (%div_2,), kwargs = {})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], nan), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%div_2, 0), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%div_2,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_2, %log_1), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %mul_1), kwargs = {})
# %where_1 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%isnan, %full_default_1, %where), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor_1,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%div_tensor_1, %log), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_2, %sub_1), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_1, %mul), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sub_3,), kwargs = {})
triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2 = async_compile.triton('triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {3: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 4), equal_to_1=(3,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 10, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2(in_out_ptr0, in_ptr0, in_ptr1, xnumel, rnumel):
xnumel = 1
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r3 = rindex
r0 = rindex % 16
r2 = (rindex // 64)
tmp0 = tl.load(in_ptr0 + (r3), None)
tmp1 = tl.load(in_ptr0 + (r0 + (64*r2)), None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr1 + (r3), None)
tmp18 = tl.load(in_ptr1 + (r0 + (64*r2)), None, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + (16 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr1 + (32 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr1 + (48 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tmp9 = libdevice.isnan(tmp8).to(tl.int1)
tmp10 = 0.0
tmp11 = tmp8 == tmp10
tmp12 = tl_math.log(tmp8)
tmp13 = tmp8 * tmp12
tmp14 = tl.where(tmp11, tmp10, tmp13)
tmp15 = float("nan")
tmp16 = tl.where(tmp9, tmp15, tmp14)
tmp19 = tl_math.exp(tmp18)
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tl_math.log(tmp28)
tmp30 = tmp17 - tmp29
tmp31 = tmp8 * tmp30
tmp32 = tmp16 - tmp31
tmp33 = tl.broadcast_to(tmp32, [RBLOCK])
tmp35 = triton_helpers.promote_to_tensor(tl.sum(tmp33, 0))
tmp36 = 256.0
tmp37 = tmp35 / tmp36
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp37, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(arg1_1, buf0, 256, grid=grid(256), stream=stream0)
del arg1_1
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(arg0_1, buf2, 256, grid=grid(256), stream=stream0)
del arg0_1
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [softmax, kl_div, log_softmax], Original ATen: [aten._softmax, aten.xlogy, aten._log_softmax, aten.mul, aten.sub, aten.mean]
triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2.run(buf4, buf0, buf2, 1, 256, grid=grid(1), stream=stream0)
del buf0
del buf2
return (buf4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class InteractiveKLLoss(nn.Module):
def __init__(self, temperature):
super().__init__()
self.temperature = temperature
self.kl_loss = nn.KLDivLoss()
def forward(self, student, teacher):
return self.kl_loss(F.log_softmax(student / self.temperature, dim=1
), F.softmax(teacher / self.temperature, dim=1))
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'temperature': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp3 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp5 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp8 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp11 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = 0.25
tmp16 = tmp14 * tmp15
tmp17 = tl_math.exp(tmp16)
tl.store(out_ptr0 + x3, tmp17, xmask)
@triton.jit
def triton_poi_fused_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp3 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp5 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp8 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp11 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = 0.25
tmp16 = tmp14 * tmp15
tl.store(out_ptr0 + x3, tmp16, xmask)
@triton.jit
def triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2(in_out_ptr0,
in_ptr0, in_ptr1, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r3 = rindex
r0 = rindex % 16
r2 = rindex // 64
tmp0 = tl.load(in_ptr0 + r3, None)
tmp1 = tl.load(in_ptr0 + (r0 + 64 * r2), None, eviction_policy='evict_last'
)
tmp2 = tl.load(in_ptr0 + (16 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp17 = tl.load(in_ptr1 + r3, None)
tmp18 = tl.load(in_ptr1 + (r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp20 = tl.load(in_ptr1 + (16 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr1 + (32 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp26 = tl.load(in_ptr1 + (48 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tmp9 = libdevice.isnan(tmp8).to(tl.int1)
tmp10 = 0.0
tmp11 = tmp8 == tmp10
tmp12 = tl_math.log(tmp8)
tmp13 = tmp8 * tmp12
tmp14 = tl.where(tmp11, tmp10, tmp13)
tmp15 = float('nan')
tmp16 = tl.where(tmp9, tmp15, tmp14)
tmp19 = tl_math.exp(tmp18)
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tl_math.log(tmp28)
tmp30 = tmp17 - tmp29
tmp31 = tmp8 * tmp30
tmp32 = tmp16 - tmp31
tmp33 = tl.broadcast_to(tmp32, [RBLOCK])
tmp35 = triton_helpers.promote_to_tensor(tl.sum(tmp33, 0))
tmp36 = 256.0
tmp37 = tmp35 / tmp36
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp37, None)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(256)](arg1_1, buf0, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del arg1_1
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_1[grid(256)](arg0_1, buf2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3
del buf3
triton_per_fused__log_softmax__softmax_mean_mul_sub_xlogy_2[grid(1)](
buf4, buf0, buf2, 1, 256, num_warps=2, num_stages=1)
del buf0
del buf2
return buf4,
class InteractiveKLLossNew(nn.Module):
def __init__(self, temperature):
super().__init__()
self.temperature = temperature
self.kl_loss = nn.KLDivLoss()
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| savan77/nni | InteractiveKLLoss | false | 4,273 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, temperature):
super().__init__()
self.temperature = temperature
self.kl_loss = nn.KLDivLoss()
def forward(self, student, teacher):
return self.kl_loss(F.log_softmax(student / self.temperature, dim=1
), F.softmax(teacher / self.temperature, dim=1))
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
GAT | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/rn/crnvqrppxb3ocltksxzamskmx3mpvscqlqbanvhmgkmj5b53kfuf.py
# Topologically Sorted Source Nodes: [add, e], Original ATen: [aten.add, aten.leaky_relu]
# Source node to ATen node mapping:
# add => add
# e => gt
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_1, %permute), kwargs = {})
# %gt : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%add, 0), kwargs = {})
triton_poi_fused_add_leaky_relu_0 = async_compile.triton('triton_poi_fused_add_leaky_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_leaky_relu_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_leaky_relu_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tl.store(out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/k2/ck2qvrsga6qzh3n4zrcqhgn3gcw55gg5hx55rqebdhhoptcty66e.py
# Topologically Sorted Source Nodes: [gt], Original ATen: [aten.gt]
# Source node to ATen node mapping:
# gt => gt_1
# Graph fragment:
# %gt_1 : [num_users=6] = call_function[target=torch.ops.aten.gt.Scalar](args = (%primals_5, 0), kwargs = {})
triton_poi_fused_gt_1 = async_compile.triton('triton_poi_fused_gt_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i1', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_gt_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_gt_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.0
tmp2 = tmp0 > tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/7p/c7pzocdj4z66742c4x73l2mmqga3yo74menkparuudjxrge7crrh.py
# Topologically Sorted Source Nodes: [add, e, zero_vec, attention, attention_1, add_1, e_1, attention_3, attention_4, add_2, e_2, attention_6, attention_7, add_3, e_3, attention_9, attention_10], Original ATen: [aten.add, aten.leaky_relu, aten.mul, aten.where, aten._softmax]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# add_3 => add_3
# attention => where_1
# attention_1 => amax
# attention_10 => amax_3
# attention_3 => where_4
# attention_4 => amax_1
# attention_6 => where_7
# attention_7 => amax_2
# attention_9 => where_10
# e => mul, where
# e_1 => mul_5, where_3
# e_2 => mul_10, where_6
# e_3 => mul_15, where_9
# zero_vec => full_default
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_1, %permute), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, 4), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %add, %mul), kwargs = {})
# %full_default : [num_users=5] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], -8999999815811072.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where, %full_default), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where_1, [1], True), kwargs = {})
# %add_1 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_5, %permute_1), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_1, 4), kwargs = {})
# %where_3 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_3, %add_1, %mul_5), kwargs = {})
# %where_4 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_3, %full_default), kwargs = {})
# %amax_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where_4, [1], True), kwargs = {})
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_9, %permute_2), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_2, 4), kwargs = {})
# %where_6 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_6, %add_2, %mul_10), kwargs = {})
# %where_7 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_6, %full_default), kwargs = {})
# %amax_2 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where_7, [1], True), kwargs = {})
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_13, %permute_3), kwargs = {})
# %mul_15 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_3, 4), kwargs = {})
# %where_9 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_9, %add_3, %mul_15), kwargs = {})
# %where_10 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_9, %full_default), kwargs = {})
# %amax_3 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where_10, [1], True), kwargs = {})
triton_poi_fused__softmax_add_leaky_relu_mul_where_2 = async_compile.triton('triton_poi_fused__softmax_add_leaky_relu_mul_where_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*i1', 1: '*i1', 2: '*fp32', 3: '*fp32', 4: '*i1', 5: '*fp32', 6: '*fp32', 7: '*i1', 8: '*fp32', 9: '*fp32', 10: '*i1', 11: '*fp32', 12: '*fp32', 13: '*fp32', 14: '*fp32', 15: '*fp32', 16: '*fp32', 17: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_leaky_relu_mul_where_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 40, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9, in_ptr10, in_ptr11, in_ptr12, out_ptr0, out_ptr1, out_ptr2, out_ptr3, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x0), xmask)
tmp3 = tl.load(in_ptr3 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp11 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp12 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp13 = tl.load(in_ptr3 + (1))
tmp14 = tl.broadcast_to(tmp13, [XBLOCK])
tmp20 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp21 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp22 = tl.load(in_ptr3 + (2))
tmp23 = tl.broadcast_to(tmp22, [XBLOCK])
tmp29 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp30 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp31 = tl.load(in_ptr3 + (3))
tmp32 = tl.broadcast_to(tmp31, [XBLOCK])
tmp38 = tl.load(in_ptr4 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp39 = tl.load(in_ptr5 + (x0), xmask)
tmp40 = tl.load(in_ptr6 + (0))
tmp41 = tl.broadcast_to(tmp40, [XBLOCK])
tmp46 = tl.load(in_ptr4 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp47 = tl.load(in_ptr6 + (1))
tmp48 = tl.broadcast_to(tmp47, [XBLOCK])
tmp54 = tl.load(in_ptr4 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp55 = tl.load(in_ptr6 + (2))
tmp56 = tl.broadcast_to(tmp55, [XBLOCK])
tmp62 = tl.load(in_ptr4 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp63 = tl.load(in_ptr6 + (3))
tmp64 = tl.broadcast_to(tmp63, [XBLOCK])
tmp70 = tl.load(in_ptr7 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp71 = tl.load(in_ptr8 + (x0), xmask)
tmp72 = tl.load(in_ptr9 + (0))
tmp73 = tl.broadcast_to(tmp72, [XBLOCK])
tmp78 = tl.load(in_ptr7 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp79 = tl.load(in_ptr9 + (1))
tmp80 = tl.broadcast_to(tmp79, [XBLOCK])
tmp86 = tl.load(in_ptr7 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp87 = tl.load(in_ptr9 + (2))
tmp88 = tl.broadcast_to(tmp87, [XBLOCK])
tmp94 = tl.load(in_ptr7 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp95 = tl.load(in_ptr9 + (3))
tmp96 = tl.broadcast_to(tmp95, [XBLOCK])
tmp102 = tl.load(in_ptr10 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp103 = tl.load(in_ptr11 + (x0), xmask)
tmp104 = tl.load(in_ptr12 + (0))
tmp105 = tl.broadcast_to(tmp104, [XBLOCK])
tmp110 = tl.load(in_ptr10 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp111 = tl.load(in_ptr12 + (1))
tmp112 = tl.broadcast_to(tmp111, [XBLOCK])
tmp118 = tl.load(in_ptr10 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp119 = tl.load(in_ptr12 + (2))
tmp120 = tl.broadcast_to(tmp119, [XBLOCK])
tmp126 = tl.load(in_ptr10 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp127 = tl.load(in_ptr12 + (3))
tmp128 = tl.broadcast_to(tmp127, [XBLOCK])
tmp5 = tmp2 + tmp4
tmp6 = 4.0
tmp7 = tmp5 * tmp6
tmp8 = tl.where(tmp1, tmp5, tmp7)
tmp9 = -8999999815811072.0
tmp10 = tl.where(tmp0, tmp8, tmp9)
tmp15 = tmp2 + tmp14
tmp16 = tmp15 * tmp6
tmp17 = tl.where(tmp12, tmp15, tmp16)
tmp18 = tl.where(tmp11, tmp17, tmp9)
tmp19 = triton_helpers.maximum(tmp10, tmp18)
tmp24 = tmp2 + tmp23
tmp25 = tmp24 * tmp6
tmp26 = tl.where(tmp21, tmp24, tmp25)
tmp27 = tl.where(tmp20, tmp26, tmp9)
tmp28 = triton_helpers.maximum(tmp19, tmp27)
tmp33 = tmp2 + tmp32
tmp34 = tmp33 * tmp6
tmp35 = tl.where(tmp30, tmp33, tmp34)
tmp36 = tl.where(tmp29, tmp35, tmp9)
tmp37 = triton_helpers.maximum(tmp28, tmp36)
tmp42 = tmp39 + tmp41
tmp43 = tmp42 * tmp6
tmp44 = tl.where(tmp38, tmp42, tmp43)
tmp45 = tl.where(tmp0, tmp44, tmp9)
tmp49 = tmp39 + tmp48
tmp50 = tmp49 * tmp6
tmp51 = tl.where(tmp46, tmp49, tmp50)
tmp52 = tl.where(tmp11, tmp51, tmp9)
tmp53 = triton_helpers.maximum(tmp45, tmp52)
tmp57 = tmp39 + tmp56
tmp58 = tmp57 * tmp6
tmp59 = tl.where(tmp54, tmp57, tmp58)
tmp60 = tl.where(tmp20, tmp59, tmp9)
tmp61 = triton_helpers.maximum(tmp53, tmp60)
tmp65 = tmp39 + tmp64
tmp66 = tmp65 * tmp6
tmp67 = tl.where(tmp62, tmp65, tmp66)
tmp68 = tl.where(tmp29, tmp67, tmp9)
tmp69 = triton_helpers.maximum(tmp61, tmp68)
tmp74 = tmp71 + tmp73
tmp75 = tmp74 * tmp6
tmp76 = tl.where(tmp70, tmp74, tmp75)
tmp77 = tl.where(tmp0, tmp76, tmp9)
tmp81 = tmp71 + tmp80
tmp82 = tmp81 * tmp6
tmp83 = tl.where(tmp78, tmp81, tmp82)
tmp84 = tl.where(tmp11, tmp83, tmp9)
tmp85 = triton_helpers.maximum(tmp77, tmp84)
tmp89 = tmp71 + tmp88
tmp90 = tmp89 * tmp6
tmp91 = tl.where(tmp86, tmp89, tmp90)
tmp92 = tl.where(tmp20, tmp91, tmp9)
tmp93 = triton_helpers.maximum(tmp85, tmp92)
tmp97 = tmp71 + tmp96
tmp98 = tmp97 * tmp6
tmp99 = tl.where(tmp94, tmp97, tmp98)
tmp100 = tl.where(tmp29, tmp99, tmp9)
tmp101 = triton_helpers.maximum(tmp93, tmp100)
tmp106 = tmp103 + tmp105
tmp107 = tmp106 * tmp6
tmp108 = tl.where(tmp102, tmp106, tmp107)
tmp109 = tl.where(tmp0, tmp108, tmp9)
tmp113 = tmp103 + tmp112
tmp114 = tmp113 * tmp6
tmp115 = tl.where(tmp110, tmp113, tmp114)
tmp116 = tl.where(tmp11, tmp115, tmp9)
tmp117 = triton_helpers.maximum(tmp109, tmp116)
tmp121 = tmp103 + tmp120
tmp122 = tmp121 * tmp6
tmp123 = tl.where(tmp118, tmp121, tmp122)
tmp124 = tl.where(tmp20, tmp123, tmp9)
tmp125 = triton_helpers.maximum(tmp117, tmp124)
tmp129 = tmp103 + tmp128
tmp130 = tmp129 * tmp6
tmp131 = tl.where(tmp126, tmp129, tmp130)
tmp132 = tl.where(tmp29, tmp131, tmp9)
tmp133 = triton_helpers.maximum(tmp125, tmp132)
tl.store(out_ptr0 + (x0), tmp37, xmask)
tl.store(out_ptr1 + (x0), tmp69, xmask)
tl.store(out_ptr2 + (x0), tmp101, xmask)
tl.store(out_ptr3 + (x0), tmp133, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/si/csiextwu44e63m7askimz3girudboqtyp45f2wu2wmll5iovqchv.py
# Topologically Sorted Source Nodes: [add, e, zero_vec, attention, attention_1, add_1, e_1, attention_3, attention_4, add_2, e_2, attention_6, attention_7, add_3, e_3, attention_9, attention_10], Original ATen: [aten.add, aten.leaky_relu, aten.mul, aten.where, aten._softmax]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# add_3 => add_3
# attention => where_1
# attention_1 => exp, sub
# attention_10 => exp_3, sub_3
# attention_3 => where_4
# attention_4 => exp_1, sub_1
# attention_6 => where_7
# attention_7 => exp_2, sub_2
# attention_9 => where_10
# e => mul, where
# e_1 => mul_5, where_3
# e_2 => mul_10, where_6
# e_3 => mul_15, where_9
# zero_vec => full_default
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_1, %permute), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, 4), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %add, %mul), kwargs = {})
# %full_default : [num_users=5] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], -8999999815811072.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where, %full_default), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %add_1 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_5, %permute_1), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_1, 4), kwargs = {})
# %where_3 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_3, %add_1, %mul_5), kwargs = {})
# %where_4 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_3, %full_default), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_4, %amax_1), kwargs = {})
# %exp_1 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_9, %permute_2), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_2, 4), kwargs = {})
# %where_6 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_6, %add_2, %mul_10), kwargs = {})
# %where_7 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_6, %full_default), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_7, %amax_2), kwargs = {})
# %exp_2 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_2,), kwargs = {})
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_13, %permute_3), kwargs = {})
# %mul_15 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_3, 4), kwargs = {})
# %where_9 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_9, %add_3, %mul_15), kwargs = {})
# %where_10 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_9, %full_default), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_10, %amax_3), kwargs = {})
# %exp_3 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_3,), kwargs = {})
triton_poi_fused__softmax_add_leaky_relu_mul_where_3 = async_compile.triton('triton_poi_fused__softmax_add_leaky_relu_mul_where_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*i1', 1: '*i1', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*i1', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*i1', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*i1', 14: '*fp32', 15: '*fp32', 16: '*fp32', 17: '*fp32', 18: '*fp32', 19: '*fp32', 20: '*fp32', 21: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_leaky_relu_mul_where_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 17, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9, in_ptr10, in_ptr11, in_ptr12, in_ptr13, in_ptr14, in_ptr15, in_ptr16, out_ptr0, out_ptr1, out_ptr2, out_ptr3, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x2), xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr5 + (x2), xmask).to(tl.int1)
tmp14 = tl.load(in_ptr6 + (x1), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr7 + (x0), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr8 + (x1), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr9 + (x2), xmask).to(tl.int1)
tmp24 = tl.load(in_ptr10 + (x1), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr11 + (x0), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr12 + (x1), xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr13 + (x2), xmask).to(tl.int1)
tmp34 = tl.load(in_ptr14 + (x1), xmask, eviction_policy='evict_last')
tmp35 = tl.load(in_ptr15 + (x0), xmask, eviction_policy='evict_last')
tmp40 = tl.load(in_ptr16 + (x1), xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp5 = 4.0
tmp6 = tmp4 * tmp5
tmp7 = tl.where(tmp1, tmp4, tmp6)
tmp8 = -8999999815811072.0
tmp9 = tl.where(tmp0, tmp7, tmp8)
tmp11 = tmp9 - tmp10
tmp12 = tl_math.exp(tmp11)
tmp16 = tmp14 + tmp15
tmp17 = tmp16 * tmp5
tmp18 = tl.where(tmp13, tmp16, tmp17)
tmp19 = tl.where(tmp0, tmp18, tmp8)
tmp21 = tmp19 - tmp20
tmp22 = tl_math.exp(tmp21)
tmp26 = tmp24 + tmp25
tmp27 = tmp26 * tmp5
tmp28 = tl.where(tmp23, tmp26, tmp27)
tmp29 = tl.where(tmp0, tmp28, tmp8)
tmp31 = tmp29 - tmp30
tmp32 = tl_math.exp(tmp31)
tmp36 = tmp34 + tmp35
tmp37 = tmp36 * tmp5
tmp38 = tl.where(tmp33, tmp36, tmp37)
tmp39 = tl.where(tmp0, tmp38, tmp8)
tmp41 = tmp39 - tmp40
tmp42 = tl_math.exp(tmp41)
tl.store(out_ptr0 + (x2), tmp12, xmask)
tl.store(out_ptr1 + (x2), tmp22, xmask)
tl.store(out_ptr2 + (x2), tmp32, xmask)
tl.store(out_ptr3 + (x2), tmp42, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/rr/crrmj7r54x5uk325xkhuskxp4m5prz3fpx53yc2st4o5pwbhq32p.py
# Topologically Sorted Source Nodes: [attention_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attention_1 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_4 = async_compile.triton('triton_poi_fused__softmax_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xp/cxpnlviefwxbdj7cbio4oqhkzb74qnjn5guhdplnmdcsr7cnbsyp.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# x_1 => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%where_2, %where_5, %where_8, %where_11], 1), kwargs = {})
triton_poi_fused_cat_5 = async_compile.triton('triton_poi_fused_cat_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_5(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = 0.0
tmp7 = tmp5 > tmp6
tmp8 = 1.0
tmp9 = tmp5 * tmp8
tmp10 = libdevice.expm1(tmp9)
tmp11 = tmp10 * tmp8
tmp12 = tl.where(tmp7, tmp9, tmp11)
tmp13 = tl.full(tmp12.shape, 0.0, tmp12.dtype)
tmp14 = tl.where(tmp4, tmp12, tmp13)
tmp15 = tmp0 >= tmp3
tmp16 = tl.full([1], 8, tl.int64)
tmp17 = tmp0 < tmp16
tmp18 = tmp15 & tmp17
tmp19 = tl.load(in_ptr1 + ((4*x1) + ((-4) + x0)), tmp18 & xmask, eviction_policy='evict_last', other=0.0)
tmp20 = tmp19 > tmp6
tmp21 = tmp19 * tmp8
tmp22 = libdevice.expm1(tmp21)
tmp23 = tmp22 * tmp8
tmp24 = tl.where(tmp20, tmp21, tmp23)
tmp25 = tl.full(tmp24.shape, 0.0, tmp24.dtype)
tmp26 = tl.where(tmp18, tmp24, tmp25)
tmp27 = tmp0 >= tmp16
tmp28 = tl.full([1], 12, tl.int64)
tmp29 = tmp0 < tmp28
tmp30 = tmp27 & tmp29
tmp31 = tl.load(in_ptr2 + ((4*x1) + ((-8) + x0)), tmp30 & xmask, eviction_policy='evict_last', other=0.0)
tmp32 = tmp31 > tmp6
tmp33 = tmp31 * tmp8
tmp34 = libdevice.expm1(tmp33)
tmp35 = tmp34 * tmp8
tmp36 = tl.where(tmp32, tmp33, tmp35)
tmp37 = tl.full(tmp36.shape, 0.0, tmp36.dtype)
tmp38 = tl.where(tmp30, tmp36, tmp37)
tmp39 = tmp0 >= tmp28
tmp40 = tl.full([1], 16, tl.int64)
tmp41 = tmp0 < tmp40
tmp42 = tl.load(in_ptr3 + ((4*x1) + ((-12) + x0)), tmp39 & xmask, eviction_policy='evict_last', other=0.0)
tmp43 = tmp42 > tmp6
tmp44 = tmp42 * tmp8
tmp45 = libdevice.expm1(tmp44)
tmp46 = tmp45 * tmp8
tmp47 = tl.where(tmp43, tmp44, tmp46)
tmp48 = tl.full(tmp47.shape, 0.0, tmp47.dtype)
tmp49 = tl.where(tmp39, tmp47, tmp48)
tmp50 = tl.where(tmp30, tmp38, tmp49)
tmp51 = tl.where(tmp18, tmp26, tmp50)
tmp52 = tl.where(tmp4, tmp14, tmp51)
tl.store(out_ptr0 + (x2), tmp52, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/g2/cg2m5jpgsmsheozlj7op7zoco4x7nplswmyzh66rmfnns22uywtb.py
# Topologically Sorted Source Nodes: [zero_vec, add_4, e_4, attention_12, attention_13], Original ATen: [aten.mul, aten.add, aten.leaky_relu, aten.where, aten._softmax]
# Source node to ATen node mapping:
# add_4 => add_4
# attention_12 => where_13
# attention_13 => amax_4
# e_4 => mul_20, where_12
# zero_vec => full_default
# Graph fragment:
# %full_default : [num_users=5] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], -8999999815811072.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %add_4 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_17, %permute_4), kwargs = {})
# %mul_20 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_4, 4), kwargs = {})
# %where_12 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_12, %add_4, %mul_20), kwargs = {})
# %where_13 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_12, %full_default), kwargs = {})
# %amax_4 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where_13, [1], True), kwargs = {})
triton_poi_fused__softmax_add_leaky_relu_mul_where_6 = async_compile.triton('triton_poi_fused__softmax_add_leaky_relu_mul_where_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*i1', 1: '*i1', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_leaky_relu_mul_where_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 13, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last').to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x0), xmask)
tmp3 = tl.load(in_ptr3 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp11 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp12 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp13 = tl.load(in_ptr3 + (1))
tmp14 = tl.broadcast_to(tmp13, [XBLOCK])
tmp20 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp21 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp22 = tl.load(in_ptr3 + (2))
tmp23 = tl.broadcast_to(tmp22, [XBLOCK])
tmp29 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp30 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp31 = tl.load(in_ptr3 + (3))
tmp32 = tl.broadcast_to(tmp31, [XBLOCK])
tmp5 = tmp2 + tmp4
tmp6 = 4.0
tmp7 = tmp5 * tmp6
tmp8 = tl.where(tmp1, tmp5, tmp7)
tmp9 = -8999999815811072.0
tmp10 = tl.where(tmp0, tmp8, tmp9)
tmp15 = tmp2 + tmp14
tmp16 = tmp15 * tmp6
tmp17 = tl.where(tmp12, tmp15, tmp16)
tmp18 = tl.where(tmp11, tmp17, tmp9)
tmp19 = triton_helpers.maximum(tmp10, tmp18)
tmp24 = tmp2 + tmp23
tmp25 = tmp24 * tmp6
tmp26 = tl.where(tmp21, tmp24, tmp25)
tmp27 = tl.where(tmp20, tmp26, tmp9)
tmp28 = triton_helpers.maximum(tmp19, tmp27)
tmp33 = tmp2 + tmp32
tmp34 = tmp33 * tmp6
tmp35 = tl.where(tmp30, tmp33, tmp34)
tmp36 = tl.where(tmp29, tmp35, tmp9)
tmp37 = triton_helpers.maximum(tmp28, tmp36)
tl.store(out_ptr0 + (x0), tmp37, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/3p/c3pw4zgk6qf2aose2yqbculke53py2ghtcozhat76besgjeo2j6h.py
# Topologically Sorted Source Nodes: [zero_vec, add_4, e_4, attention_12, attention_13], Original ATen: [aten.mul, aten.add, aten.leaky_relu, aten.where, aten._softmax]
# Source node to ATen node mapping:
# add_4 => add_4
# attention_12 => where_13
# attention_13 => exp_4, sub_4
# e_4 => mul_20, where_12
# zero_vec => full_default
# Graph fragment:
# %full_default : [num_users=5] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], -8999999815811072.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %add_4 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_17, %permute_4), kwargs = {})
# %mul_20 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_4, 4), kwargs = {})
# %where_12 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_12, %add_4, %mul_20), kwargs = {})
# %where_13 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %where_12, %full_default), kwargs = {})
# %sub_4 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_13, %amax_4), kwargs = {})
# %exp_4 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_4,), kwargs = {})
triton_poi_fused__softmax_add_leaky_relu_mul_where_7 = async_compile.triton('triton_poi_fused__softmax_add_leaky_relu_mul_where_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*i1', 1: '*i1', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_leaky_relu_mul_where_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_7(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x2), xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp5 = 4.0
tmp6 = tmp4 * tmp5
tmp7 = tl.where(tmp1, tmp4, tmp6)
tmp8 = -8999999815811072.0
tmp9 = tl.where(tmp0, tmp7, tmp8)
tmp11 = tmp9 - tmp10
tmp12 = tl_math.exp(tmp11)
tl.store(out_ptr0 + (x2), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vn/cvnlzux2xi3qwahaidfxslw62nvrvjgyqouwnuxx7bn6vc4gb2jx.py
# Topologically Sorted Source Nodes: [x_3, log_softmax], Original ATen: [aten.elu, aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => amax_5, sub_5
# x_3 => expm1_4, gt_14, mul_22, mul_24, where_14
# Graph fragment:
# %gt_14 : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%mm_19, 0), kwargs = {})
# %mul_22 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mm_19, 1.0), kwargs = {})
# %expm1_4 : [num_users=1] = call_function[target=torch.ops.aten.expm1.default](args = (%mul_22,), kwargs = {})
# %mul_24 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%expm1_4, 1.0), kwargs = {})
# %where_14 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%gt_14, %mul_22, %mul_24), kwargs = {})
# %amax_5 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where_14, [1], True), kwargs = {})
# %sub_5 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where_14, %amax_5), kwargs = {})
triton_poi_fused__log_softmax_elu_8 = async_compile.triton('triton_poi_fused__log_softmax_elu_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_elu_8', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_elu_8(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp8 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = 0.0
tmp2 = tmp0 > tmp1
tmp3 = 1.0
tmp4 = tmp0 * tmp3
tmp5 = libdevice.expm1(tmp4)
tmp6 = tmp5 * tmp3
tmp7 = tl.where(tmp2, tmp4, tmp6)
tmp9 = tmp8 > tmp1
tmp10 = tmp8 * tmp3
tmp11 = libdevice.expm1(tmp10)
tmp12 = tmp11 * tmp3
tmp13 = tl.where(tmp9, tmp10, tmp12)
tmp15 = tmp14 > tmp1
tmp16 = tmp14 * tmp3
tmp17 = libdevice.expm1(tmp16)
tmp18 = tmp17 * tmp3
tmp19 = tl.where(tmp15, tmp16, tmp18)
tmp20 = triton_helpers.maximum(tmp13, tmp19)
tmp22 = tmp21 > tmp1
tmp23 = tmp21 * tmp3
tmp24 = libdevice.expm1(tmp23)
tmp25 = tmp24 * tmp3
tmp26 = tl.where(tmp22, tmp23, tmp25)
tmp27 = triton_helpers.maximum(tmp20, tmp26)
tmp29 = tmp28 > tmp1
tmp30 = tmp28 * tmp3
tmp31 = libdevice.expm1(tmp30)
tmp32 = tmp31 * tmp3
tmp33 = tl.where(tmp29, tmp30, tmp32)
tmp34 = triton_helpers.maximum(tmp27, tmp33)
tmp35 = tmp7 - tmp34
tl.store(out_ptr0 + (x2), tmp35, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/uv/cuvhhdiycantl2koymc5p6rharsyggopa6l4kqtmxc7kcvu52m2v.py
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => exp_5, log, sub_6, sum_6
# Graph fragment:
# %exp_5 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_5,), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_5, [1], True), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_6,), kwargs = {})
# %sub_6 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_5, %log), kwargs = {})
triton_poi_fused__log_softmax_9 = async_compile.triton('triton_poi_fused__log_softmax_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_9', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_9(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp2 = tl_math.exp(tmp1)
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp2 + tmp4
tmp7 = tl_math.exp(tmp6)
tmp8 = tmp5 + tmp7
tmp10 = tl_math.exp(tmp9)
tmp11 = tmp8 + tmp10
tmp12 = tl_math.log(tmp11)
tmp13 = tmp0 - tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 1), (1, 1))
assert_size_stride(primals_4, (4, 1), (1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, 1), (1, 1))
assert_size_stride(primals_8, (4, 1), (1, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, 1), (1, 1))
assert_size_stride(primals_11, (4, 1), (1, 1))
assert_size_stride(primals_12, (4, 4), (4, 1))
assert_size_stride(primals_13, (4, 1), (1, 1))
assert_size_stride(primals_14, (4, 1), (1, 1))
assert_size_stride(primals_15, (16, 4), (4, 1))
assert_size_stride(primals_16, (4, 1), (1, 1))
assert_size_stride(primals_17, (4, 1), (1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.mm]
extern_kernels.mm(primals_1, primals_2, out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_1], Original ATen: [aten.mm]
extern_kernels.mm(buf0, primals_3, out=buf1)
buf2 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_2], Original ATen: [aten.mm]
extern_kernels.mm(buf0, primals_4, out=buf2)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add, e], Original ATen: [aten.add, aten.leaky_relu]
stream0 = get_raw_stream(0)
triton_poi_fused_add_leaky_relu_0.run(buf1, buf2, buf3, 16, grid=grid(16), stream=stream0)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [gt], Original ATen: [aten.gt]
triton_poi_fused_gt_1.run(primals_5, buf4, 16, grid=grid(16), stream=stream0)
del primals_5
buf9 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.mm]
extern_kernels.mm(primals_1, primals_6, out=buf9)
del primals_6
buf10 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_3], Original ATen: [aten.mm]
extern_kernels.mm(buf9, primals_7, out=buf10)
buf11 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_4], Original ATen: [aten.mm]
extern_kernels.mm(buf9, primals_8, out=buf11)
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add_1, e_1], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf10, buf11, buf12, 16, grid=grid(16), stream=stream0)
buf17 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_2], Original ATen: [aten.mm]
extern_kernels.mm(primals_1, primals_9, out=buf17)
del primals_9
buf18 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_5], Original ATen: [aten.mm]
extern_kernels.mm(buf17, primals_10, out=buf18)
buf19 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_6], Original ATen: [aten.mm]
extern_kernels.mm(buf17, primals_11, out=buf19)
buf20 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add_2, e_2], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf18, buf19, buf20, 16, grid=grid(16), stream=stream0)
buf25 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_3], Original ATen: [aten.mm]
extern_kernels.mm(primals_1, primals_12, out=buf25)
del primals_12
buf26 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_7], Original ATen: [aten.mm]
extern_kernels.mm(buf25, primals_13, out=buf26)
buf27 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [f_8], Original ATen: [aten.mm]
extern_kernels.mm(buf25, primals_14, out=buf27)
buf28 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add_3, e_3], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf26, buf27, buf28, 16, grid=grid(16), stream=stream0)
buf5 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf13 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf21 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf29 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [add, e, zero_vec, attention, attention_1, add_1, e_1, attention_3, attention_4, add_2, e_2, attention_6, attention_7, add_3, e_3, attention_9, attention_10], Original ATen: [aten.add, aten.leaky_relu, aten.mul, aten.where, aten._softmax]
triton_poi_fused__softmax_add_leaky_relu_mul_where_2.run(buf4, buf3, buf1, buf2, buf12, buf10, buf11, buf20, buf18, buf19, buf28, buf26, buf27, buf5, buf13, buf21, buf29, 4, grid=grid(4), stream=stream0)
buf6 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf22 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf30 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, e, zero_vec, attention, attention_1, add_1, e_1, attention_3, attention_4, add_2, e_2, attention_6, attention_7, add_3, e_3, attention_9, attention_10], Original ATen: [aten.add, aten.leaky_relu, aten.mul, aten.where, aten._softmax]
triton_poi_fused__softmax_add_leaky_relu_mul_where_3.run(buf4, buf3, buf1, buf2, buf5, buf12, buf10, buf11, buf13, buf20, buf18, buf19, buf21, buf28, buf26, buf27, buf29, buf6, buf14, buf22, buf30, 16, grid=grid(16), stream=stream0)
del buf1
del buf10
del buf11
del buf13
del buf18
del buf19
del buf2
del buf21
del buf26
buf7 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf6, buf7, 16, grid=grid(16), stream=stream0)
buf8 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [h_prime], Original ATen: [aten.mm]
extern_kernels.mm(buf7, buf0, out=buf8)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_4], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf14, buf15, 16, grid=grid(16), stream=stream0)
buf16 = buf14; del buf14 # reuse
# Topologically Sorted Source Nodes: [h_prime_1], Original ATen: [aten.mm]
extern_kernels.mm(buf15, buf9, out=buf16)
buf23 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_7], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf22, buf23, 16, grid=grid(16), stream=stream0)
buf24 = buf22; del buf22 # reuse
# Topologically Sorted Source Nodes: [h_prime_2], Original ATen: [aten.mm]
extern_kernels.mm(buf23, buf17, out=buf24)
buf31 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_10], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf30, buf31, 16, grid=grid(16), stream=stream0)
buf32 = buf30; del buf30 # reuse
# Topologically Sorted Source Nodes: [h_prime_3], Original ATen: [aten.mm]
extern_kernels.mm(buf31, buf25, out=buf32)
buf33 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.cat]
triton_poi_fused_cat_5.run(buf8, buf16, buf24, buf32, buf33, 64, grid=grid(64), stream=stream0)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.mm]
extern_kernels.mm(buf33, primals_15, out=buf34)
buf35 = reinterpret_tensor(buf5, (4, 1), (1, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [f_9], Original ATen: [aten.mm]
extern_kernels.mm(buf34, primals_16, out=buf35)
buf36 = reinterpret_tensor(buf29, (4, 1), (1, 1), 0); del buf29 # reuse
# Topologically Sorted Source Nodes: [f_10], Original ATen: [aten.mm]
extern_kernels.mm(buf34, primals_17, out=buf36)
buf37 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add_4, e_4], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf35, buf36, buf37, 16, grid=grid(16), stream=stream0)
buf38 = reinterpret_tensor(buf27, (4, 1), (1, 4), 0); del buf27 # reuse
# Topologically Sorted Source Nodes: [zero_vec, add_4, e_4, attention_12, attention_13], Original ATen: [aten.mul, aten.add, aten.leaky_relu, aten.where, aten._softmax]
triton_poi_fused__softmax_add_leaky_relu_mul_where_6.run(buf4, buf37, buf35, buf36, buf38, 4, grid=grid(4), stream=stream0)
buf39 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [zero_vec, add_4, e_4, attention_12, attention_13], Original ATen: [aten.mul, aten.add, aten.leaky_relu, aten.where, aten._softmax]
triton_poi_fused__softmax_add_leaky_relu_mul_where_7.run(buf4, buf37, buf35, buf36, buf38, buf39, 16, grid=grid(16), stream=stream0)
del buf35
del buf36
del buf38
buf40 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_13], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf39, buf40, 16, grid=grid(16), stream=stream0)
buf41 = buf39; del buf39 # reuse
# Topologically Sorted Source Nodes: [h_prime_4], Original ATen: [aten.mm]
extern_kernels.mm(buf40, buf34, out=buf41)
buf42 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3, log_softmax], Original ATen: [aten.elu, aten._log_softmax]
triton_poi_fused__log_softmax_elu_8.run(buf41, buf42, 16, grid=grid(16), stream=stream0)
buf43 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_9.run(buf42, buf43, 16, grid=grid(16), stream=stream0)
del buf42
return (buf43, buf3, buf4, buf7, buf8, buf12, buf15, buf16, buf20, buf23, buf24, buf28, buf31, buf32, buf37, buf40, buf41, buf43, reinterpret_tensor(buf34, (4, 4), (1, 4), 0), reinterpret_tensor(primals_17, (1, 4), (1, 1), 0), reinterpret_tensor(primals_16, (1, 4), (1, 1), 0), reinterpret_tensor(buf33, (16, 4), (1, 16), 0), reinterpret_tensor(primals_15, (4, 16), (1, 4), 0), reinterpret_tensor(buf25, (4, 4), (1, 4), 0), reinterpret_tensor(primals_14, (1, 4), (1, 1), 0), reinterpret_tensor(primals_13, (1, 4), (1, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), reinterpret_tensor(buf17, (4, 4), (1, 4), 0), reinterpret_tensor(primals_11, (1, 4), (1, 1), 0), reinterpret_tensor(primals_10, (1, 4), (1, 1), 0), reinterpret_tensor(buf9, (4, 4), (1, 4), 0), reinterpret_tensor(primals_8, (1, 4), (1, 1), 0), reinterpret_tensor(primals_7, (1, 4), (1, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0), reinterpret_tensor(primals_4, (1, 4), (1, 1), 0), reinterpret_tensor(primals_3, (1, 4), (1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
class GraphAttention(nn.Module):
"""
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, concat=True):
super(GraphAttention, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
if torch.cuda.is_available():
param_type = torch.FloatTensor
else:
param_type = torch.FloatTensor
self.W = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
in_features, out_features).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.a1 = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
out_features, 1).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.a2 = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
out_features, 1).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.leaky_relu = nn.LeakyReLU(self.alpha)
def forward(self, input, adj):
h = torch.mm(input, self.W)
h.size()[0]
f_1 = torch.mm(h, self.a1)
f_2 = torch.mm(h, self.a2)
e = self.leaky_relu(f_1 + f_2.transpose(0, 1))
zero_vec = -9000000000000000.0 * torch.ones_like(e)
attention = torch.where(adj > 0, e, zero_vec)
attention = F.softmax(attention, dim=1)
attention = F.dropout(attention, self.dropout, training=self.training)
h_prime = torch.matmul(attention, h)
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class GAT(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, alpha, nheads):
super(GAT, self).__init__()
self.dropout = dropout
self.attentions = [GraphAttention(nfeat, nhid, dropout=dropout,
alpha=alpha, concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
self.out_att = GraphAttention(nhid * nheads, nclass, dropout=
dropout, alpha=alpha, concat=False)
def forward(self, x, adj):
x = F.dropout(x, self.dropout, training=self.training)
x = torch.cat([att(x, adj) for att in self.attentions], dim=1)
x = F.dropout(x, self.dropout, training=self.training)
x = F.elu(self.out_att(x, adj))
return F.log_softmax(x, dim=1)
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'nfeat': 4, 'nhid': 4, 'nclass': 4, 'dropout': 0.5,
'alpha': 4, 'nheads': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_leaky_relu_0(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tl.store(out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_gt_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.0
tmp2 = tmp0 > tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_2(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9,
in_ptr10, in_ptr11, in_ptr12, out_ptr0, out_ptr1, out_ptr2, out_ptr3,
xnumel, XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp2 = tl.load(in_ptr2 + x0, xmask)
tmp3 = tl.load(in_ptr3 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp11 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp12 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp13 = tl.load(in_ptr3 + 1)
tmp14 = tl.broadcast_to(tmp13, [XBLOCK])
tmp20 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp21 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp22 = tl.load(in_ptr3 + 2)
tmp23 = tl.broadcast_to(tmp22, [XBLOCK])
tmp29 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp30 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp31 = tl.load(in_ptr3 + 3)
tmp32 = tl.broadcast_to(tmp31, [XBLOCK])
tmp38 = tl.load(in_ptr4 + 4 * x0, xmask, eviction_policy='evict_last').to(
tl.int1)
tmp39 = tl.load(in_ptr5 + x0, xmask)
tmp40 = tl.load(in_ptr6 + 0)
tmp41 = tl.broadcast_to(tmp40, [XBLOCK])
tmp46 = tl.load(in_ptr4 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp47 = tl.load(in_ptr6 + 1)
tmp48 = tl.broadcast_to(tmp47, [XBLOCK])
tmp54 = tl.load(in_ptr4 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp55 = tl.load(in_ptr6 + 2)
tmp56 = tl.broadcast_to(tmp55, [XBLOCK])
tmp62 = tl.load(in_ptr4 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp63 = tl.load(in_ptr6 + 3)
tmp64 = tl.broadcast_to(tmp63, [XBLOCK])
tmp70 = tl.load(in_ptr7 + 4 * x0, xmask, eviction_policy='evict_last').to(
tl.int1)
tmp71 = tl.load(in_ptr8 + x0, xmask)
tmp72 = tl.load(in_ptr9 + 0)
tmp73 = tl.broadcast_to(tmp72, [XBLOCK])
tmp78 = tl.load(in_ptr7 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp79 = tl.load(in_ptr9 + 1)
tmp80 = tl.broadcast_to(tmp79, [XBLOCK])
tmp86 = tl.load(in_ptr7 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp87 = tl.load(in_ptr9 + 2)
tmp88 = tl.broadcast_to(tmp87, [XBLOCK])
tmp94 = tl.load(in_ptr7 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp95 = tl.load(in_ptr9 + 3)
tmp96 = tl.broadcast_to(tmp95, [XBLOCK])
tmp102 = tl.load(in_ptr10 + 4 * x0, xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp103 = tl.load(in_ptr11 + x0, xmask)
tmp104 = tl.load(in_ptr12 + 0)
tmp105 = tl.broadcast_to(tmp104, [XBLOCK])
tmp110 = tl.load(in_ptr10 + (1 + 4 * x0), xmask, eviction_policy=
'evict_last').to(tl.int1)
tmp111 = tl.load(in_ptr12 + 1)
tmp112 = tl.broadcast_to(tmp111, [XBLOCK])
tmp118 = tl.load(in_ptr10 + (2 + 4 * x0), xmask, eviction_policy=
'evict_last').to(tl.int1)
tmp119 = tl.load(in_ptr12 + 2)
tmp120 = tl.broadcast_to(tmp119, [XBLOCK])
tmp126 = tl.load(in_ptr10 + (3 + 4 * x0), xmask, eviction_policy=
'evict_last').to(tl.int1)
tmp127 = tl.load(in_ptr12 + 3)
tmp128 = tl.broadcast_to(tmp127, [XBLOCK])
tmp5 = tmp2 + tmp4
tmp6 = 4.0
tmp7 = tmp5 * tmp6
tmp8 = tl.where(tmp1, tmp5, tmp7)
tmp9 = -8999999815811072.0
tmp10 = tl.where(tmp0, tmp8, tmp9)
tmp15 = tmp2 + tmp14
tmp16 = tmp15 * tmp6
tmp17 = tl.where(tmp12, tmp15, tmp16)
tmp18 = tl.where(tmp11, tmp17, tmp9)
tmp19 = triton_helpers.maximum(tmp10, tmp18)
tmp24 = tmp2 + tmp23
tmp25 = tmp24 * tmp6
tmp26 = tl.where(tmp21, tmp24, tmp25)
tmp27 = tl.where(tmp20, tmp26, tmp9)
tmp28 = triton_helpers.maximum(tmp19, tmp27)
tmp33 = tmp2 + tmp32
tmp34 = tmp33 * tmp6
tmp35 = tl.where(tmp30, tmp33, tmp34)
tmp36 = tl.where(tmp29, tmp35, tmp9)
tmp37 = triton_helpers.maximum(tmp28, tmp36)
tmp42 = tmp39 + tmp41
tmp43 = tmp42 * tmp6
tmp44 = tl.where(tmp38, tmp42, tmp43)
tmp45 = tl.where(tmp0, tmp44, tmp9)
tmp49 = tmp39 + tmp48
tmp50 = tmp49 * tmp6
tmp51 = tl.where(tmp46, tmp49, tmp50)
tmp52 = tl.where(tmp11, tmp51, tmp9)
tmp53 = triton_helpers.maximum(tmp45, tmp52)
tmp57 = tmp39 + tmp56
tmp58 = tmp57 * tmp6
tmp59 = tl.where(tmp54, tmp57, tmp58)
tmp60 = tl.where(tmp20, tmp59, tmp9)
tmp61 = triton_helpers.maximum(tmp53, tmp60)
tmp65 = tmp39 + tmp64
tmp66 = tmp65 * tmp6
tmp67 = tl.where(tmp62, tmp65, tmp66)
tmp68 = tl.where(tmp29, tmp67, tmp9)
tmp69 = triton_helpers.maximum(tmp61, tmp68)
tmp74 = tmp71 + tmp73
tmp75 = tmp74 * tmp6
tmp76 = tl.where(tmp70, tmp74, tmp75)
tmp77 = tl.where(tmp0, tmp76, tmp9)
tmp81 = tmp71 + tmp80
tmp82 = tmp81 * tmp6
tmp83 = tl.where(tmp78, tmp81, tmp82)
tmp84 = tl.where(tmp11, tmp83, tmp9)
tmp85 = triton_helpers.maximum(tmp77, tmp84)
tmp89 = tmp71 + tmp88
tmp90 = tmp89 * tmp6
tmp91 = tl.where(tmp86, tmp89, tmp90)
tmp92 = tl.where(tmp20, tmp91, tmp9)
tmp93 = triton_helpers.maximum(tmp85, tmp92)
tmp97 = tmp71 + tmp96
tmp98 = tmp97 * tmp6
tmp99 = tl.where(tmp94, tmp97, tmp98)
tmp100 = tl.where(tmp29, tmp99, tmp9)
tmp101 = triton_helpers.maximum(tmp93, tmp100)
tmp106 = tmp103 + tmp105
tmp107 = tmp106 * tmp6
tmp108 = tl.where(tmp102, tmp106, tmp107)
tmp109 = tl.where(tmp0, tmp108, tmp9)
tmp113 = tmp103 + tmp112
tmp114 = tmp113 * tmp6
tmp115 = tl.where(tmp110, tmp113, tmp114)
tmp116 = tl.where(tmp11, tmp115, tmp9)
tmp117 = triton_helpers.maximum(tmp109, tmp116)
tmp121 = tmp103 + tmp120
tmp122 = tmp121 * tmp6
tmp123 = tl.where(tmp118, tmp121, tmp122)
tmp124 = tl.where(tmp20, tmp123, tmp9)
tmp125 = triton_helpers.maximum(tmp117, tmp124)
tmp129 = tmp103 + tmp128
tmp130 = tmp129 * tmp6
tmp131 = tl.where(tmp126, tmp129, tmp130)
tmp132 = tl.where(tmp29, tmp131, tmp9)
tmp133 = triton_helpers.maximum(tmp125, tmp132)
tl.store(out_ptr0 + x0, tmp37, xmask)
tl.store(out_ptr1 + x0, tmp69, xmask)
tl.store(out_ptr2 + x0, tmp101, xmask)
tl.store(out_ptr3 + x0, tmp133, xmask)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_3(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9,
in_ptr10, in_ptr11, in_ptr12, in_ptr13, in_ptr14, in_ptr15, in_ptr16,
out_ptr0, out_ptr1, out_ptr2, out_ptr3, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + x2, xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr5 + x2, xmask).to(tl.int1)
tmp14 = tl.load(in_ptr6 + x1, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr7 + x0, xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr8 + x1, xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr9 + x2, xmask).to(tl.int1)
tmp24 = tl.load(in_ptr10 + x1, xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr11 + x0, xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr12 + x1, xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr13 + x2, xmask).to(tl.int1)
tmp34 = tl.load(in_ptr14 + x1, xmask, eviction_policy='evict_last')
tmp35 = tl.load(in_ptr15 + x0, xmask, eviction_policy='evict_last')
tmp40 = tl.load(in_ptr16 + x1, xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp5 = 4.0
tmp6 = tmp4 * tmp5
tmp7 = tl.where(tmp1, tmp4, tmp6)
tmp8 = -8999999815811072.0
tmp9 = tl.where(tmp0, tmp7, tmp8)
tmp11 = tmp9 - tmp10
tmp12 = tl_math.exp(tmp11)
tmp16 = tmp14 + tmp15
tmp17 = tmp16 * tmp5
tmp18 = tl.where(tmp13, tmp16, tmp17)
tmp19 = tl.where(tmp0, tmp18, tmp8)
tmp21 = tmp19 - tmp20
tmp22 = tl_math.exp(tmp21)
tmp26 = tmp24 + tmp25
tmp27 = tmp26 * tmp5
tmp28 = tl.where(tmp23, tmp26, tmp27)
tmp29 = tl.where(tmp0, tmp28, tmp8)
tmp31 = tmp29 - tmp30
tmp32 = tl_math.exp(tmp31)
tmp36 = tmp34 + tmp35
tmp37 = tmp36 * tmp5
tmp38 = tl.where(tmp33, tmp36, tmp37)
tmp39 = tl.where(tmp0, tmp38, tmp8)
tmp41 = tmp39 - tmp40
tmp42 = tl_math.exp(tmp41)
tl.store(out_ptr0 + x2, tmp12, xmask)
tl.store(out_ptr1 + x2, tmp22, xmask)
tl.store(out_ptr2 + x2, tmp32, xmask)
tl.store(out_ptr3 + x2, tmp42, xmask)
@triton.jit
def triton_poi_fused__softmax_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_cat_5(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = 0.0
tmp7 = tmp5 > tmp6
tmp8 = 1.0
tmp9 = tmp5 * tmp8
tmp10 = libdevice.expm1(tmp9)
tmp11 = tmp10 * tmp8
tmp12 = tl.where(tmp7, tmp9, tmp11)
tmp13 = tl.full(tmp12.shape, 0.0, tmp12.dtype)
tmp14 = tl.where(tmp4, tmp12, tmp13)
tmp15 = tmp0 >= tmp3
tmp16 = tl.full([1], 8, tl.int64)
tmp17 = tmp0 < tmp16
tmp18 = tmp15 & tmp17
tmp19 = tl.load(in_ptr1 + (4 * x1 + (-4 + x0)), tmp18 & xmask,
eviction_policy='evict_last', other=0.0)
tmp20 = tmp19 > tmp6
tmp21 = tmp19 * tmp8
tmp22 = libdevice.expm1(tmp21)
tmp23 = tmp22 * tmp8
tmp24 = tl.where(tmp20, tmp21, tmp23)
tmp25 = tl.full(tmp24.shape, 0.0, tmp24.dtype)
tmp26 = tl.where(tmp18, tmp24, tmp25)
tmp27 = tmp0 >= tmp16
tmp28 = tl.full([1], 12, tl.int64)
tmp29 = tmp0 < tmp28
tmp30 = tmp27 & tmp29
tmp31 = tl.load(in_ptr2 + (4 * x1 + (-8 + x0)), tmp30 & xmask,
eviction_policy='evict_last', other=0.0)
tmp32 = tmp31 > tmp6
tmp33 = tmp31 * tmp8
tmp34 = libdevice.expm1(tmp33)
tmp35 = tmp34 * tmp8
tmp36 = tl.where(tmp32, tmp33, tmp35)
tmp37 = tl.full(tmp36.shape, 0.0, tmp36.dtype)
tmp38 = tl.where(tmp30, tmp36, tmp37)
tmp39 = tmp0 >= tmp28
tl.full([1], 16, tl.int64)
tmp42 = tl.load(in_ptr3 + (4 * x1 + (-12 + x0)), tmp39 & xmask,
eviction_policy='evict_last', other=0.0)
tmp43 = tmp42 > tmp6
tmp44 = tmp42 * tmp8
tmp45 = libdevice.expm1(tmp44)
tmp46 = tmp45 * tmp8
tmp47 = tl.where(tmp43, tmp44, tmp46)
tmp48 = tl.full(tmp47.shape, 0.0, tmp47.dtype)
tmp49 = tl.where(tmp39, tmp47, tmp48)
tmp50 = tl.where(tmp30, tmp38, tmp49)
tmp51 = tl.where(tmp18, tmp26, tmp50)
tmp52 = tl.where(tmp4, tmp14, tmp51)
tl.store(out_ptr0 + x2, tmp52, xmask)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_6(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp2 = tl.load(in_ptr2 + x0, xmask)
tmp3 = tl.load(in_ptr3 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp11 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp12 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp13 = tl.load(in_ptr3 + 1)
tmp14 = tl.broadcast_to(tmp13, [XBLOCK])
tmp20 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp21 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp22 = tl.load(in_ptr3 + 2)
tmp23 = tl.broadcast_to(tmp22, [XBLOCK])
tmp29 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp30 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp31 = tl.load(in_ptr3 + 3)
tmp32 = tl.broadcast_to(tmp31, [XBLOCK])
tmp5 = tmp2 + tmp4
tmp6 = 4.0
tmp7 = tmp5 * tmp6
tmp8 = tl.where(tmp1, tmp5, tmp7)
tmp9 = -8999999815811072.0
tmp10 = tl.where(tmp0, tmp8, tmp9)
tmp15 = tmp2 + tmp14
tmp16 = tmp15 * tmp6
tmp17 = tl.where(tmp12, tmp15, tmp16)
tmp18 = tl.where(tmp11, tmp17, tmp9)
tmp19 = triton_helpers.maximum(tmp10, tmp18)
tmp24 = tmp2 + tmp23
tmp25 = tmp24 * tmp6
tmp26 = tl.where(tmp21, tmp24, tmp25)
tmp27 = tl.where(tmp20, tmp26, tmp9)
tmp28 = triton_helpers.maximum(tmp19, tmp27)
tmp33 = tmp2 + tmp32
tmp34 = tmp33 * tmp6
tmp35 = tl.where(tmp30, tmp33, tmp34)
tmp36 = tl.where(tmp29, tmp35, tmp9)
tmp37 = triton_helpers.maximum(tmp28, tmp36)
tl.store(out_ptr0 + x0, tmp37, xmask)
@triton.jit
def triton_poi_fused__softmax_add_leaky_relu_mul_where_7(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + x2, xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp5 = 4.0
tmp6 = tmp4 * tmp5
tmp7 = tl.where(tmp1, tmp4, tmp6)
tmp8 = -8999999815811072.0
tmp9 = tl.where(tmp0, tmp7, tmp8)
tmp11 = tmp9 - tmp10
tmp12 = tl_math.exp(tmp11)
tl.store(out_ptr0 + x2, tmp12, xmask)
@triton.jit
def triton_poi_fused__log_softmax_elu_8(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp8 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp21 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp28 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = 0.0
tmp2 = tmp0 > tmp1
tmp3 = 1.0
tmp4 = tmp0 * tmp3
tmp5 = libdevice.expm1(tmp4)
tmp6 = tmp5 * tmp3
tmp7 = tl.where(tmp2, tmp4, tmp6)
tmp9 = tmp8 > tmp1
tmp10 = tmp8 * tmp3
tmp11 = libdevice.expm1(tmp10)
tmp12 = tmp11 * tmp3
tmp13 = tl.where(tmp9, tmp10, tmp12)
tmp15 = tmp14 > tmp1
tmp16 = tmp14 * tmp3
tmp17 = libdevice.expm1(tmp16)
tmp18 = tmp17 * tmp3
tmp19 = tl.where(tmp15, tmp16, tmp18)
tmp20 = triton_helpers.maximum(tmp13, tmp19)
tmp22 = tmp21 > tmp1
tmp23 = tmp21 * tmp3
tmp24 = libdevice.expm1(tmp23)
tmp25 = tmp24 * tmp3
tmp26 = tl.where(tmp22, tmp23, tmp25)
tmp27 = triton_helpers.maximum(tmp20, tmp26)
tmp29 = tmp28 > tmp1
tmp30 = tmp28 * tmp3
tmp31 = libdevice.expm1(tmp30)
tmp32 = tmp31 * tmp3
tmp33 = tl.where(tmp29, tmp30, tmp32)
tmp34 = triton_helpers.maximum(tmp27, tmp33)
tmp35 = tmp7 - tmp34
tl.store(out_ptr0 + x2, tmp35, xmask)
@triton.jit
def triton_poi_fused__log_softmax_9(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp2 = tl_math.exp(tmp1)
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp2 + tmp4
tmp7 = tl_math.exp(tmp6)
tmp8 = tmp5 + tmp7
tmp10 = tl_math.exp(tmp9)
tmp11 = tmp8 + tmp10
tmp12 = tl_math.log(tmp11)
tmp13 = tmp0 - tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13, primals_14, primals_15, primals_16, primals_17) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 1), (1, 1))
assert_size_stride(primals_4, (4, 1), (1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, 1), (1, 1))
assert_size_stride(primals_8, (4, 1), (1, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, 1), (1, 1))
assert_size_stride(primals_11, (4, 1), (1, 1))
assert_size_stride(primals_12, (4, 4), (4, 1))
assert_size_stride(primals_13, (4, 1), (1, 1))
assert_size_stride(primals_14, (4, 1), (1, 1))
assert_size_stride(primals_15, (16, 4), (4, 1))
assert_size_stride(primals_16, (4, 1), (1, 1))
assert_size_stride(primals_17, (4, 1), (1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_1, primals_2, out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf0, primals_3, out=buf1)
buf2 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf0, primals_4, out=buf2)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_leaky_relu_0[grid(16)](buf1, buf2, buf3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
triton_poi_fused_gt_1[grid(16)](primals_5, buf4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_5
buf9 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_1, primals_6, out=buf9)
del primals_6
buf10 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf9, primals_7, out=buf10)
buf11 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf9, primals_8, out=buf11)
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(16)](buf10, buf11, buf12, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf17 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_1, primals_9, out=buf17)
del primals_9
buf18 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf17, primals_10, out=buf18)
buf19 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf17, primals_11, out=buf19)
buf20 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(16)](buf18, buf19, buf20, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf25 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_1, primals_12, out=buf25)
del primals_12
buf26 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf25, primals_13, out=buf26)
buf27 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(buf25, primals_14, out=buf27)
buf28 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(16)](buf26, buf27, buf28, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf5 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf13 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf21 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf29 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
triton_poi_fused__softmax_add_leaky_relu_mul_where_2[grid(4)](buf4,
buf3, buf1, buf2, buf12, buf10, buf11, buf20, buf18, buf19,
buf28, buf26, buf27, buf5, buf13, buf21, buf29, 4, XBLOCK=4,
num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf22 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf30 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_add_leaky_relu_mul_where_3[grid(16)](buf4,
buf3, buf1, buf2, buf5, buf12, buf10, buf11, buf13, buf20,
buf18, buf19, buf21, buf28, buf26, buf27, buf29, buf6, buf14,
buf22, buf30, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf1
del buf10
del buf11
del buf13
del buf18
del buf19
del buf2
del buf21
del buf26
buf7 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(16)](buf6, buf7, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf8 = buf6
del buf6
extern_kernels.mm(buf7, buf0, out=buf8)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(16)](buf14, buf15, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf16 = buf14
del buf14
extern_kernels.mm(buf15, buf9, out=buf16)
buf23 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(16)](buf22, buf23, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf24 = buf22
del buf22
extern_kernels.mm(buf23, buf17, out=buf24)
buf31 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(16)](buf30, buf31, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf32 = buf30
del buf30
extern_kernels.mm(buf31, buf25, out=buf32)
buf33 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
triton_poi_fused_cat_5[grid(64)](buf8, buf16, buf24, buf32, buf33,
64, XBLOCK=64, num_warps=1, num_stages=1)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf33, primals_15, out=buf34)
buf35 = reinterpret_tensor(buf5, (4, 1), (1, 1), 0)
del buf5
extern_kernels.mm(buf34, primals_16, out=buf35)
buf36 = reinterpret_tensor(buf29, (4, 1), (1, 1), 0)
del buf29
extern_kernels.mm(buf34, primals_17, out=buf36)
buf37 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(16)](buf35, buf36, buf37, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf38 = reinterpret_tensor(buf27, (4, 1), (1, 4), 0)
del buf27
triton_poi_fused__softmax_add_leaky_relu_mul_where_6[grid(4)](buf4,
buf37, buf35, buf36, buf38, 4, XBLOCK=4, num_warps=1, num_stages=1)
buf39 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_add_leaky_relu_mul_where_7[grid(16)](buf4,
buf37, buf35, buf36, buf38, buf39, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del buf35
del buf36
del buf38
buf40 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(16)](buf39, buf40, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf41 = buf39
del buf39
extern_kernels.mm(buf40, buf34, out=buf41)
buf42 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__log_softmax_elu_8[grid(16)](buf41, buf42, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf43 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__log_softmax_9[grid(16)](buf42, buf43, 16, XBLOCK=
16, num_warps=1, num_stages=1)
del buf42
return (buf43, buf3, buf4, buf7, buf8, buf12, buf15, buf16, buf20,
buf23, buf24, buf28, buf31, buf32, buf37, buf40, buf41, buf43,
reinterpret_tensor(buf34, (4, 4), (1, 4), 0), reinterpret_tensor(
primals_17, (1, 4), (1, 1), 0), reinterpret_tensor(primals_16, (1,
4), (1, 1), 0), reinterpret_tensor(buf33, (16, 4), (1, 16), 0),
reinterpret_tensor(primals_15, (4, 16), (1, 4), 0),
reinterpret_tensor(buf25, (4, 4), (1, 4), 0), reinterpret_tensor(
primals_14, (1, 4), (1, 1), 0), reinterpret_tensor(primals_13, (1,
4), (1, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0),
reinterpret_tensor(buf17, (4, 4), (1, 4), 0), reinterpret_tensor(
primals_11, (1, 4), (1, 1), 0), reinterpret_tensor(primals_10, (1,
4), (1, 1), 0), reinterpret_tensor(buf9, (4, 4), (1, 4), 0),
reinterpret_tensor(primals_8, (1, 4), (1, 1), 0),
reinterpret_tensor(primals_7, (1, 4), (1, 1), 0),
reinterpret_tensor(buf0, (4, 4), (1, 4), 0), reinterpret_tensor(
primals_4, (1, 4), (1, 1), 0), reinterpret_tensor(primals_3, (1, 4),
(1, 1), 0))
class GraphAttention(nn.Module):
"""
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, concat=True):
super(GraphAttention, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
if torch.cuda.is_available():
param_type = torch.FloatTensor
else:
param_type = torch.FloatTensor
self.W = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
in_features, out_features).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.a1 = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
out_features, 1).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.a2 = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
out_features, 1).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.leaky_relu = nn.LeakyReLU(self.alpha)
def forward(self, input, adj):
h = torch.mm(input, self.W)
h.size()[0]
f_1 = torch.mm(h, self.a1)
f_2 = torch.mm(h, self.a2)
e = self.leaky_relu(f_1 + f_2.transpose(0, 1))
zero_vec = -9000000000000000.0 * torch.ones_like(e)
attention = torch.where(adj > 0, e, zero_vec)
attention = F.softmax(attention, dim=1)
attention = F.dropout(attention, self.dropout, training=self.training)
h_prime = torch.matmul(attention, h)
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class GATNew(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, alpha, nheads):
super(GATNew, self).__init__()
self.dropout = dropout
self.attentions = [GraphAttention(nfeat, nhid, dropout=dropout,
alpha=alpha, concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
self.out_att = GraphAttention(nhid * nheads, nclass, dropout=
dropout, alpha=alpha, concat=False)
def forward(self, input_0, input_1):
primals_1 = self.attention_0.W
primals_3 = self.attention_0.a1
primals_4 = self.attention_0.a2
primals_2 = self.attention_1.W
primals_7 = self.attention_1.a1
primals_8 = self.attention_1.a2
primals_5 = self.attention_2.W
primals_10 = self.attention_2.a1
primals_11 = self.attention_2.a2
primals_6 = self.attention_3.W
primals_13 = self.attention_3.a1
primals_14 = self.attention_3.a2
primals_15 = self.out_att.W
primals_16 = self.out_att.a1
primals_17 = self.out_att.a2
primals_9 = input_0
primals_12 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13, primals_14,
primals_15, primals_16, primals_17])
return output[0]
| NightmareNyx/pygcn | GAT | false | 4,274 | [
"MIT"
] | 0 | 3972f167ce7fcc41cb21284d75816dfd9a15f7ef | https://github.com/NightmareNyx/pygcn/tree/3972f167ce7fcc41cb21284d75816dfd9a15f7ef | import torch
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
class GraphAttention(nn.Module):
"""
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, concat=True):
super().__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
if torch.cuda.is_available():
param_type = torch.FloatTensor
else:
param_type = torch.FloatTensor
self.W = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
in_features, out_features).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.a1 = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
out_features, 1).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.a2 = nn.Parameter(nn.init.xavier_normal_(torch.Tensor(
out_features, 1).type(param_type), gain=np.sqrt(2.0)),
requires_grad=True)
self.leaky_relu = nn.LeakyReLU(self.alpha)
def forward(self, input, adj):
h = torch.mm(input, self.W)
h.size()[0]
f_1 = torch.mm(h, self.a1)
f_2 = torch.mm(h, self.a2)
e = self.leaky_relu(f_1 + f_2.transpose(0, 1))
zero_vec = -9000000000000000.0 * torch.ones_like(e)
attention = torch.where(adj > 0, e, zero_vec)
attention = F.softmax(attention, dim=1)
attention = F.dropout(attention, self.dropout, training=self.training)
h_prime = torch.matmul(attention, h)
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class Model(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, alpha, nheads):
super().__init__()
self.dropout = dropout
self.attentions = [GraphAttention(nfeat, nhid, dropout=dropout,
alpha=alpha, concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
self.out_att = GraphAttention(nhid * nheads, nclass, dropout=
dropout, alpha=alpha, concat=False)
def forward(self, x, adj):
x = F.dropout(x, self.dropout, training=self.training)
x = torch.cat([att(x, adj) for att in self.attentions], dim=1)
x = F.dropout(x, self.dropout, training=self.training)
x = F.elu(self.out_att(x, adj))
return F.log_softmax(x, dim=1)
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'nfeat': 4, 'nhid': 4, 'nclass': 4, 'dropout': 0.5,
'alpha': 4, 'nheads': 4}]
|
Auto_Encoder_Model | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ej/cejfrwnzxinkchwn6symdb72fdtj7gix5hy2vuswodhbeh45mrae.py
# Topologically Sorted Source Nodes: [conv2d, output], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d => convolution
# output => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1048576],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1048576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 4096) % 64
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/7z/c7zuih2ysjtir5rh5seep5ijnhokjlgkyjw2edhf257ahvz4iipr.py
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# output_1 => getitem, getitem_1
# Graph fragment:
# %getitem : [num_users=2] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 0), kwargs = {})
# %getitem_1 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_1 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i8', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_max_pool2d_with_indices_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_1(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 262144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 32
x1 = (xindex // 32)
x2 = xindex
tmp0 = tl.load(in_ptr0 + ((2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + (2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (65 + (2*x0) + (128*x1)), None, eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (x2), tmp6, None)
tl.store(out_ptr1 + (x2), tmp16, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/j6/cj6faeofhfnxsh5iuwazughjlau4igyajnmvjequyelq7apzs4qm.py
# Topologically Sorted Source Nodes: [conv2d_1, output_2], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d_1 => convolution_1
# output_2 => relu_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%getitem, %primals_4, %primals_5, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_relu_2 = async_compile.triton('triton_poi_fused_convolution_relu_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[131072],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 131072
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 1024) % 32
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6y/c6yx6oq7oo2cwoaop3iwu5iqfdckg6lycdtu4jjuiv3wdcf2o6p7.py
# Topologically Sorted Source Nodes: [output_3], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# output_3 => getitem_2, getitem_3
# Graph fragment:
# %getitem_2 : [num_users=2] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets_1, 0), kwargs = {})
# %getitem_3 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets_1, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_3 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i8', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_max_pool2d_with_indices_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_3(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 32768
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + ((2*x0) + (64*x1)), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (64*x1)), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (32 + (2*x0) + (64*x1)), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (33 + (2*x0) + (64*x1)), None, eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (x2), tmp6, None)
tl.store(out_ptr1 + (x2), tmp16, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/c3/cc37wvituo2asffgdbn2cnuhsr4nuj5pzt75pvxxxx4t7tdtdkqj.py
# Topologically Sorted Source Nodes: [conv2d_2, output_4], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d_2 => convolution_2
# output_4 => relu_2
# Graph fragment:
# %convolution_2 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%getitem_2, %primals_6, %primals_7, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_2,), kwargs = {})
triton_poi_fused_convolution_relu_4 = async_compile.triton('triton_poi_fused_convolution_relu_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_4', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 256) % 16
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/t7/ct7cf34m4g63ojfteengkc3tcdxkjvs4wde47kna4a7bol6sdtyb.py
# Topologically Sorted Source Nodes: [conv2d_4, output_8], Original ATen: [aten.convolution, aten.sigmoid]
# Source node to ATen node mapping:
# conv2d_4 => convolution_6
# output_8 => sigmoid
# Graph fragment:
# %convolution_6 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu_5, %primals_14, %primals_15, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_6,), kwargs = {})
triton_poi_fused_convolution_sigmoid_5 = async_compile.triton('triton_poi_fused_convolution_sigmoid_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_sigmoid_5', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_sigmoid_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), None)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15 = args
args.clear()
assert_size_stride(primals_1, (64, 1, 3, 3), (9, 9, 3, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
assert_size_stride(primals_4, (32, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_5, (32, ), (1, ))
assert_size_stride(primals_6, (16, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_7, (16, ), (1, ))
assert_size_stride(primals_8, (16, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_9, (32, ), (1, ))
assert_size_stride(primals_10, (32, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_11, (32, ), (1, ))
assert_size_stride(primals_12, (32, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_13, (64, ), (1, ))
assert_size_stride(primals_14, (1, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_15, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv2d, output], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 1048576, grid=grid(1048576), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((4, 64, 32, 32), (65536, 1024, 32, 1), torch.float32)
buf3 = empty_strided_cuda((4, 64, 32, 32), (65536, 1024, 32, 1), torch.int8)
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_1.run(buf1, buf2, buf3, 262144, grid=grid(262144), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 32, 32, 32), (32768, 1024, 32, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [conv2d_1, output_2], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf5, primals_5, 131072, grid=grid(131072), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 32, 16, 16), (8192, 256, 16, 1), torch.float32)
buf7 = empty_strided_cuda((4, 32, 16, 16), (8192, 256, 16, 1), torch.int8)
# Topologically Sorted Source Nodes: [output_3], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_3.run(buf5, buf6, buf7, 32768, grid=grid(32768), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
buf8 = extern_kernels.convolution(buf6, primals_6, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 16, 16, 16), (4096, 256, 16, 1))
buf9 = buf8; del buf8 # reuse
# Topologically Sorted Source Nodes: [conv2d_2, output_4], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_4.run(buf9, primals_7, 16384, grid=grid(16384), stream=stream0)
del primals_7
# Topologically Sorted Source Nodes: [conv_transpose2d], Original ATen: [aten.convolution]
buf10 = extern_kernels.convolution(buf9, primals_8, stride=(2, 2), padding=(1, 1), dilation=(1, 1), transposed=True, output_padding=(1, 1), groups=1, bias=None)
assert_size_stride(buf10, (4, 32, 32, 32), (32768, 1024, 32, 1))
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [conv_transpose2d, output_5], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf11, primals_9, 131072, grid=grid(131072), stream=stream0)
del primals_9
# Topologically Sorted Source Nodes: [conv2d_3], Original ATen: [aten.convolution]
buf12 = extern_kernels.convolution(buf11, primals_10, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 32, 32, 32), (32768, 1024, 32, 1))
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [conv2d_3, output_6], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf13, primals_11, 131072, grid=grid(131072), stream=stream0)
del primals_11
# Topologically Sorted Source Nodes: [conv_transpose2d_1], Original ATen: [aten.convolution]
buf14 = extern_kernels.convolution(buf13, primals_12, stride=(2, 2), padding=(1, 1), dilation=(1, 1), transposed=True, output_padding=(1, 1), groups=1, bias=None)
assert_size_stride(buf14, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf15 = buf14; del buf14 # reuse
# Topologically Sorted Source Nodes: [conv_transpose2d_1, output_7], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_0.run(buf15, primals_13, 1048576, grid=grid(1048576), stream=stream0)
del primals_13
# Topologically Sorted Source Nodes: [conv2d_4], Original ATen: [aten.convolution]
buf16 = extern_kernels.convolution(buf15, primals_14, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf16, (4, 1, 64, 64), (4096, 4096, 64, 1))
buf17 = buf16; del buf16 # reuse
# Topologically Sorted Source Nodes: [conv2d_4, output_8], Original ATen: [aten.convolution, aten.sigmoid]
triton_poi_fused_convolution_sigmoid_5.run(buf17, primals_15, 16384, grid=grid(16384), stream=stream0)
del primals_15
return (buf17, primals_1, primals_3, primals_4, primals_6, primals_8, primals_10, primals_12, primals_14, buf1, buf2, buf3, buf5, buf6, buf7, buf9, buf11, buf13, buf15, buf17, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 1, 3, 3), (9, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 1, 64, 64), (4096, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((32, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((16, 32, 3, 3), (288, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((16, 32, 3, 3), (288, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((32, 32, 3, 3), (288, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((32, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((1, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class Auto_Encoder_Model(nn.Module):
def __init__(self):
super(Auto_Encoder_Model, self).__init__()
self.conv1 = nn.Conv2d(1, 64, padding=1, kernel_size=3)
self.max_pool1 = nn.MaxPool2d(2)
self.conv2 = nn.Conv2d(64, 32, padding=1, kernel_size=3)
self.max_pool2 = nn.MaxPool2d(2)
self.conv3 = nn.Conv2d(32, 16, padding=1, kernel_size=3)
self.tran_conv1 = nn.ConvTranspose2d(16, 32, kernel_size=3, stride=
2, padding=1, output_padding=1)
self.conv4 = nn.Conv2d(32, 32, kernel_size=3, padding=1)
self.tran_conv2 = nn.ConvTranspose2d(32, 64, kernel_size=3, stride=
2, padding=1, output_padding=1)
self.conv5 = nn.Conv2d(64, 1, kernel_size=3, padding=1)
def forward_pass(self, x):
output = F.relu(self.conv1(x))
output = self.max_pool1(output)
output = F.relu(self.conv2(output))
output = self.max_pool2(output)
output = F.relu(self.conv3(output))
return output
def reconstruct_pass(self, x):
output = F.relu(self.tran_conv1(x))
output = F.relu(self.conv4(output))
output = F.relu(self.tran_conv2(output))
output = torch.sigmoid(self.conv5(output))
return output
def forward(self, x):
output = self.forward_pass(x)
output = self.reconstruct_pass(output)
return output
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 4096 % 64
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, None)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_1(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 32
x1 = xindex // 32
x2 = xindex
tmp0 = tl.load(in_ptr0 + (2 * x0 + 128 * x1), None, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 128 * x1), None, eviction_policy
='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + 2 * x0 + 128 * x1), None,
eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (65 + 2 * x0 + 128 * x1), None,
eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + x2, tmp6, None)
tl.store(out_ptr1 + x2, tmp16, None)
@triton.jit
def triton_poi_fused_convolution_relu_2(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 1024 % 32
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, None)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_3(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (2 * x0 + 64 * x1), None, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 64 * x1), None, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (32 + 2 * x0 + 64 * x1), None, eviction_policy
='evict_last')
tmp5 = tl.load(in_ptr0 + (33 + 2 * x0 + 64 * x1), None, eviction_policy
='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + x2, tmp6, None)
tl.store(out_ptr1 + x2, tmp16, None)
@triton.jit
def triton_poi_fused_convolution_relu_4(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 256 % 16
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, None)
@triton.jit
def triton_poi_fused_convolution_sigmoid_5(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, None)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13, primals_14, primals_15) = args
args.clear()
assert_size_stride(primals_1, (64, 1, 3, 3), (9, 9, 3, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
assert_size_stride(primals_4, (32, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_5, (32,), (1,))
assert_size_stride(primals_6, (16, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_7, (16,), (1,))
assert_size_stride(primals_8, (16, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_9, (32,), (1,))
assert_size_stride(primals_10, (32, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_11, (32,), (1,))
assert_size_stride(primals_12, (32, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_13, (64,), (1,))
assert_size_stride(primals_14, (1, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_15, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(1048576)](buf1, primals_2,
1048576, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((4, 64, 32, 32), (65536, 1024, 32, 1),
torch.float32)
buf3 = empty_strided_cuda((4, 64, 32, 32), (65536, 1024, 32, 1),
torch.int8)
triton_poi_fused_max_pool2d_with_indices_1[grid(262144)](buf1, buf2,
buf3, 262144, XBLOCK=512, num_warps=8, num_stages=1)
buf4 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 32, 32, 32), (32768, 1024, 32, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_relu_2[grid(131072)](buf5, primals_5,
131072, XBLOCK=512, num_warps=8, num_stages=1)
del primals_5
buf6 = empty_strided_cuda((4, 32, 16, 16), (8192, 256, 16, 1),
torch.float32)
buf7 = empty_strided_cuda((4, 32, 16, 16), (8192, 256, 16, 1),
torch.int8)
triton_poi_fused_max_pool2d_with_indices_3[grid(32768)](buf5, buf6,
buf7, 32768, XBLOCK=128, num_warps=4, num_stages=1)
buf8 = extern_kernels.convolution(buf6, primals_6, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 16, 16, 16), (4096, 256, 16, 1))
buf9 = buf8
del buf8
triton_poi_fused_convolution_relu_4[grid(16384)](buf9, primals_7,
16384, XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
buf10 = extern_kernels.convolution(buf9, primals_8, stride=(2, 2),
padding=(1, 1), dilation=(1, 1), transposed=True,
output_padding=(1, 1), groups=1, bias=None)
assert_size_stride(buf10, (4, 32, 32, 32), (32768, 1024, 32, 1))
buf11 = buf10
del buf10
triton_poi_fused_convolution_relu_2[grid(131072)](buf11, primals_9,
131072, XBLOCK=512, num_warps=8, num_stages=1)
del primals_9
buf12 = extern_kernels.convolution(buf11, primals_10, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 32, 32, 32), (32768, 1024, 32, 1))
buf13 = buf12
del buf12
triton_poi_fused_convolution_relu_2[grid(131072)](buf13, primals_11,
131072, XBLOCK=512, num_warps=8, num_stages=1)
del primals_11
buf14 = extern_kernels.convolution(buf13, primals_12, stride=(2, 2),
padding=(1, 1), dilation=(1, 1), transposed=True,
output_padding=(1, 1), groups=1, bias=None)
assert_size_stride(buf14, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf15 = buf14
del buf14
triton_poi_fused_convolution_relu_0[grid(1048576)](buf15,
primals_13, 1048576, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_13
buf16 = extern_kernels.convolution(buf15, primals_14, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf16, (4, 1, 64, 64), (4096, 4096, 64, 1))
buf17 = buf16
del buf16
triton_poi_fused_convolution_sigmoid_5[grid(16384)](buf17,
primals_15, 16384, XBLOCK=256, num_warps=4, num_stages=1)
del primals_15
return (buf17, primals_1, primals_3, primals_4, primals_6, primals_8,
primals_10, primals_12, primals_14, buf1, buf2, buf3, buf5, buf6,
buf7, buf9, buf11, buf13, buf15, buf17)
class Auto_Encoder_ModelNew(nn.Module):
def __init__(self):
super(Auto_Encoder_ModelNew, self).__init__()
self.conv1 = nn.Conv2d(1, 64, padding=1, kernel_size=3)
self.max_pool1 = nn.MaxPool2d(2)
self.conv2 = nn.Conv2d(64, 32, padding=1, kernel_size=3)
self.max_pool2 = nn.MaxPool2d(2)
self.conv3 = nn.Conv2d(32, 16, padding=1, kernel_size=3)
self.tran_conv1 = nn.ConvTranspose2d(16, 32, kernel_size=3, stride=
2, padding=1, output_padding=1)
self.conv4 = nn.Conv2d(32, 32, kernel_size=3, padding=1)
self.tran_conv2 = nn.ConvTranspose2d(32, 64, kernel_size=3, stride=
2, padding=1, output_padding=1)
self.conv5 = nn.Conv2d(64, 1, kernel_size=3, padding=1)
def forward_pass(self, x):
output = F.relu(self.conv1(x))
output = self.max_pool1(output)
output = F.relu(self.conv2(output))
output = self.max_pool2(output)
output = F.relu(self.conv3(output))
return output
def reconstruct_pass(self, x):
output = F.relu(self.tran_conv1(x))
output = F.relu(self.conv4(output))
output = F.relu(self.tran_conv2(output))
output = torch.sigmoid(self.conv5(output))
return output
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_6 = self.conv3.weight
primals_7 = self.conv3.bias
primals_8 = self.tran_conv1.weight
primals_9 = self.tran_conv1.bias
primals_10 = self.conv4.weight
primals_11 = self.conv4.bias
primals_12 = self.tran_conv2.weight
primals_13 = self.tran_conv2.bias
primals_14 = self.conv5.weight
primals_15 = self.conv5.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13, primals_14,
primals_15])
return output[0]
| sarahESL/MICCAI19-MedVQA | Auto_Encoder_Model | false | 4,275 | [
"MIT"
] | 0 | aa751cb905f79cd356ad5746f8a0640f1d81b5d2 | https://github.com/sarahESL/MICCAI19-MedVQA/tree/aa751cb905f79cd356ad5746f8a0640f1d81b5d2 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 64, padding=1, kernel_size=3)
self.max_pool1 = nn.MaxPool2d(2)
self.conv2 = nn.Conv2d(64, 32, padding=1, kernel_size=3)
self.max_pool2 = nn.MaxPool2d(2)
self.conv3 = nn.Conv2d(32, 16, padding=1, kernel_size=3)
self.tran_conv1 = nn.ConvTranspose2d(16, 32, kernel_size=3, stride=
2, padding=1, output_padding=1)
self.conv4 = nn.Conv2d(32, 32, kernel_size=3, padding=1)
self.tran_conv2 = nn.ConvTranspose2d(32, 64, kernel_size=3, stride=
2, padding=1, output_padding=1)
self.conv5 = nn.Conv2d(64, 1, kernel_size=3, padding=1)
def forward_pass(self, x):
output = F.relu(self.conv1(x))
output = self.max_pool1(output)
output = F.relu(self.conv2(output))
output = self.max_pool2(output)
output = F.relu(self.conv3(output))
return output
def reconstruct_pass(self, x):
output = F.relu(self.tran_conv1(x))
output = F.relu(self.conv4(output))
output = F.relu(self.tran_conv2(output))
output = torch.sigmoid(self.conv5(output))
return output
def forward(self, x):
output = self.forward_pass(x)
output = self.reconstruct_pass(output)
return output
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return []
|
ZeroLayer | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/zi/cziatn4srpsymxab7n67k7jt34egxdol3kpyktgeck2cxwbklbyh.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 0), kwargs = {})
triton_poi_fused_mul_0 = async_compile.triton('triton_poi_fused_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class ZeroLayer(nn.Module):
def __init__(self, stride):
super(ZeroLayer, self).__init__()
self.stride = stride
def forward(self, x):
"""n, c, h, w = x.size()
h //= self.stride
w //= self.stride
device = x.get_device() if x.is_cuda else torch.device('cpu')
# noinspection PyUnresolvedReferences
padding = torch.zeros(n, c, h, w, device=device, requires_grad=False)
return padding"""
return x * 0
@staticmethod
def is_zero_layer():
return True
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'stride': 1}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class ZeroLayerNew(nn.Module):
def __init__(self, stride):
super(ZeroLayerNew, self).__init__()
self.stride = stride
@staticmethod
def is_zero_layer():
return True
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| savan77/nni | ZeroLayer | false | 4,276 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class Model(nn.Module):
def __init__(self, stride):
super().__init__()
self.stride = stride
def forward(self, x):
"""n, c, h, w = x.size()
h //= self.stride
w //= self.stride
device = x.get_device() if x.is_cuda else torch.device('cpu')
# noinspection PyUnresolvedReferences
padding = torch.zeros(n, c, h, w, device=device, requires_grad=False)
return padding"""
return x * 0
@staticmethod
def is_zero_layer():
return True
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [1]
|
FCNet | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/bm/cbmd63mrouqmm2pha5x6evse3dkbpy5o4xnk5v7quflfkqfdvwck.py
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# output_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 320
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 5
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (5, 4), (4, 1))
assert_size_stride(primals_2, (5, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 5), (5, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 5), (5, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 5), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 5), (80, 20, 5, 1), 0); del buf0 # reuse
buf3 = empty_strided_cuda((4, 4, 4, 5), (80, 20, 5, 1), torch.bool)
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf3, 320, grid=grid(320), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 5), (5, 1), 0), reinterpret_tensor(primals_4, (5, 4), (1, 5), 0), alpha=1, beta=1, out=buf2)
del primals_5
return (reinterpret_tensor(buf2, (256, ), (1, ), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 5), (5, 1), 0), primals_4, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((5, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((5, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 5), (5, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class FCNet(nn.Module):
def __init__(self, input_size, output_size):
super().__init__()
self.l1 = nn.Linear(input_size, 5)
self.relu = nn.ReLU()
self.l2 = nn.Linear(5, output_size)
def forward(self, x):
output = self.l1(x)
output = self.relu(output)
output = self.l2(output)
return output.view(-1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4, 'output_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 320
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 5
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (5, 4), (4, 1))
assert_size_stride(primals_2, (5,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 5), (5, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 5), (5, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 5), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 5), (80, 20, 5, 1), 0)
del buf0
buf3 = empty_strided_cuda((4, 4, 4, 5), (80, 20, 5, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(320)](buf1,
primals_2, buf3, 320, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 5), (
5, 1), 0), reinterpret_tensor(primals_4, (5, 4), (1, 5), 0),
alpha=1, beta=1, out=buf2)
del primals_5
return reinterpret_tensor(buf2, (256,), (1,), 0), reinterpret_tensor(
primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 5), (
5, 1), 0), primals_4, buf3
class FCNetNew(nn.Module):
def __init__(self, input_size, output_size):
super().__init__()
self.l1 = nn.Linear(input_size, 5)
self.relu = nn.ReLU()
self.l2 = nn.Linear(5, output_size)
def forward(self, input_0):
primals_1 = self.l1.weight
primals_2 = self.l1.bias
primals_4 = self.l2.weight
primals_5 = self.l2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| savan77/nni | FCNet | false | 4,277 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class Model(nn.Module):
def __init__(self, input_size, output_size):
super().__init__()
self.l1 = nn.Linear(input_size, 5)
self.relu = nn.ReLU()
self.l2 = nn.Linear(5, output_size)
def forward(self, x):
output = self.l1(x)
output = self.relu(output)
output = self.l2(output)
return output.view(-1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
LinearCombine | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/hu/chuao3goscfvc5gm5ggoerju3pembwo7thvhuzz6h7r3gyxruobd.py
# Topologically Sorted Source Nodes: [nw, seq, seq_1], Original ATen: [aten._softmax, aten.mul, aten.sum]
# Source node to ATen node mapping:
# nw => amax, div, exp, sub, sum_1
# seq => mul
# seq_1 => sum_2
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%primals_1, [0], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [0], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %div), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [0]), kwargs = {})
triton_poi_fused__softmax_mul_sum_0 = async_compile.triton('triton_poi_fused__softmax_mul_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_mul_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mul_sum_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp10 = tl.load(in_ptr0 + (128 + x0), xmask)
tmp13 = tl.load(in_ptr0 + (192 + x0), xmask)
tmp3 = tmp2 - tmp2
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp4 / tmp4
tmp6 = tmp0 * tmp5
tmp8 = tmp7 * tmp5
tmp9 = tmp6 + tmp8
tmp11 = tmp10 * tmp5
tmp12 = tmp9 + tmp11
tmp14 = tmp13 * tmp5
tmp15 = tmp12 + tmp14
tl.store(out_ptr0 + (x0), tmp15, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [nw, seq, seq_1], Original ATen: [aten._softmax, aten.mul, aten.sum]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_mul_sum_0.run(primals_2, primals_1, buf0, 64, grid=grid(64), stream=stream0)
return (buf0, primals_1, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((1, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class LinearCombine(nn.Module):
def __init__(self, layers_num, trainable=True, input_aware=False,
word_level=False):
super(LinearCombine, self).__init__()
self.input_aware = input_aware
self.word_level = word_level
if input_aware:
raise NotImplementedError('Input aware is not supported.')
self.w = nn.Parameter(torch.full((layers_num, 1, 1, 1), 1.0 /
layers_num), requires_grad=trainable)
def forward(self, seq):
nw = F.softmax(self.w, dim=0)
seq = torch.mul(seq, nw)
seq = torch.sum(seq, dim=0)
return seq
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'layers_num': 1}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused__softmax_mul_sum_0(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp10 = tl.load(in_ptr0 + (128 + x0), xmask)
tmp13 = tl.load(in_ptr0 + (192 + x0), xmask)
tmp3 = tmp2 - tmp2
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp4 / tmp4
tmp6 = tmp0 * tmp5
tmp8 = tmp7 * tmp5
tmp9 = tmp6 + tmp8
tmp11 = tmp10 * tmp5
tmp12 = tmp9 + tmp11
tmp14 = tmp13 * tmp5
tmp15 = tmp12 + tmp14
tl.store(out_ptr0 + x0, tmp15, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_mul_sum_0[grid(64)](primals_2, primals_1,
buf0, 64, XBLOCK=64, num_warps=1, num_stages=1)
return buf0, primals_1, primals_2
class LinearCombineNew(nn.Module):
def __init__(self, layers_num, trainable=True, input_aware=False,
word_level=False):
super(LinearCombineNew, self).__init__()
self.input_aware = input_aware
self.word_level = word_level
if input_aware:
raise NotImplementedError('Input aware is not supported.')
self.w = nn.Parameter(torch.full((layers_num, 1, 1, 1), 1.0 /
layers_num), requires_grad=trainable)
def forward(self, input_0):
primals_1 = self.w
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
| savan77/nni | LinearCombine | false | 4,278 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, layers_num, trainable=True, input_aware=False,
word_level=False):
super().__init__()
self.input_aware = input_aware
self.word_level = word_level
if input_aware:
raise NotImplementedError('Input aware is not supported.')
self.w = nn.Parameter(torch.full((layers_num, 1, 1, 1), 1.0 /
layers_num), requires_grad=trainable)
def forward(self, seq):
nw = F.softmax(self.w, dim=0)
seq = torch.mul(seq, nw)
seq = torch.sum(seq, dim=0)
return seq
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [1]
|
TorchAdd | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/gx/cgxrigsvtx4nc75mpdz7qivonc3wkrexg4c7zrh6gk2vmbwc4atl.py
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select, %select_1), kwargs = {})
triton_poi_fused_add_0 = async_compile.triton('triton_poi_fused_add_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(arg0_1, buf0, 64, grid=grid(64), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class TorchAdd(nn.Module):
"""
TorchAdd Module.
"""
def forward(self, input_list):
return input_list[0] + input_list[1]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class TorchAddNew(nn.Module):
"""
TorchAdd Module.
"""
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| savan77/nni | TorchAdd | false | 4,279 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
class Model(nn.Module):
"""
TorchAdd Module.
"""
def forward(self, input_list):
return input_list[0] + input_list[1]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ActorCritic | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/r3/cr3febcwm3t44fuoitsx3ou2p6xg4sk4f7unagmmrvffasxf47te.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xk/cxkugsynlmnyrjhah42fewrhwovuvurnuv2qimo2qhxq27wjmq7q.py
# Topologically Sorted Source Nodes: [policy_dist], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# policy_dist => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/jf/cjfzp64ny4hf7wdw5wptah3hqv5fcsh5rrw4brz7uxcy6ad57n7h.py
# Topologically Sorted Source Nodes: [policy_dist], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# policy_dist => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf7, 256, grid=grid(256), stream=stream0)
del primals_2
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [value], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [policy_dist], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [policy_dist], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
return (reinterpret_tensor(buf3, (4, 4, 4, 1), (16, 4, 1, 1), 0), buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), buf6, primals_6, primals_4, buf7, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class ActorCritic(nn.Module):
def __init__(self, num_states, num_actions, hidden_size):
super(ActorCritic, self).__init__()
self.num_actions = num_actions
self.fc = nn.Linear(num_states, hidden_size)
self.critic_linear2 = nn.Linear(hidden_size, 1)
self.actor_linear2 = nn.Linear(hidden_size, num_actions)
def forward(self, state):
x = F.relu(self.fc(state))
value = self.critic_linear2(x)
policy_dist = F.softmax(self.actor_linear2(x))
return value, policy_dist
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_states': 4, 'num_actions': 4, 'hidden_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf1,
primals_2, buf7, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0),
alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf1, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf4, buf5, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf4
triton_poi_fused__softmax_2[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf5
return reinterpret_tensor(buf3, (4, 4, 4, 1), (16, 4, 1, 1), 0
), buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 4), (4, 1), 0
), buf6, primals_6, primals_4, buf7
class ActorCriticNew(nn.Module):
def __init__(self, num_states, num_actions, hidden_size):
super(ActorCriticNew, self).__init__()
self.num_actions = num_actions
self.fc = nn.Linear(num_states, hidden_size)
self.critic_linear2 = nn.Linear(hidden_size, 1)
self.actor_linear2 = nn.Linear(hidden_size, num_actions)
def forward(self, input_0):
primals_1 = self.fc.weight
primals_2 = self.fc.bias
primals_4 = self.critic_linear2.weight
primals_5 = self.critic_linear2.bias
primals_6 = self.actor_linear2.weight
primals_7 = self.actor_linear2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0], output[1]
| savan77/nni | ActorCritic | false | 4,280 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, num_states, num_actions, hidden_size):
super().__init__()
self.num_actions = num_actions
self.fc = nn.Linear(num_states, hidden_size)
self.critic_linear2 = nn.Linear(hidden_size, 1)
self.actor_linear2 = nn.Linear(hidden_size, num_actions)
def forward(self, state):
x = F.relu(self.fc(state))
value = self.critic_linear2(x)
policy_dist = F.softmax(self.actor_linear2(x))
return value, policy_dist
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
LipschitzCube | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/fh/cfhy3agarxy4nw2x57myddkk4vpnhtk5yjzrljts6cbbw3pyiwgb.py
# Topologically Sorted Source Nodes: [ge, sub, mul, le, add, mul_1, add_1, gt, lt, mul_2, pow_1, mul_3, truediv, add_2], Original ATen: [aten.ge, aten.sub, aten.mul, aten.le, aten.add, aten.gt, aten.lt, aten.pow, aten.div]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# ge => ge
# gt => gt
# le => le
# lt => lt
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# mul_3 => mul_3
# pow_1 => pow_1
# sub => sub
# truediv => div
# Graph fragment:
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%arg0_1, 1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, 0.6666666666666666), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%ge, %sub), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%arg0_1, -1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 0.6666666666666666), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%le, %add), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%arg0_1, -1), kwargs = {})
# %lt : [num_users=1] = call_function[target=torch.ops.aten.lt.Scalar](args = (%arg0_1, 1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%gt, %lt), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%arg0_1, 3), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %pow_1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul_3, 3), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %div), kwargs = {})
triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0 = async_compile.triton('triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 1.0
tmp2 = tmp0 >= tmp1
tmp3 = tmp2.to(tl.float32)
tmp4 = 0.6666666666666666
tmp5 = tmp0 - tmp4
tmp6 = tmp3 * tmp5
tmp7 = -1.0
tmp8 = tmp0 <= tmp7
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp0 + tmp4
tmp11 = tmp9 * tmp10
tmp12 = tmp6 + tmp11
tmp13 = tmp0 > tmp7
tmp14 = tmp0 < tmp1
tmp15 = tmp13 & tmp14
tmp16 = tmp15.to(tl.float32)
tmp17 = tmp0 * tmp0
tmp18 = tmp17 * tmp0
tmp19 = tmp16 * tmp18
tmp20 = 0.3333333333333333
tmp21 = tmp19 * tmp20
tmp22 = tmp12 + tmp21
tl.store(out_ptr0 + (x0), tmp22, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [ge, sub, mul, le, add, mul_1, add_1, gt, lt, mul_2, pow_1, mul_3, truediv, add_2], Original ATen: [aten.ge, aten.sub, aten.mul, aten.le, aten.add, aten.gt, aten.lt, aten.pow, aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.utils.data.distributed
class LipschitzCube(nn.Module):
def forward(self, x):
return (x >= 1) * (x - 2 / 3) + (x <= -1) * (x + 2 / 3) + (x > -1) * (x
< 1) * x ** 3 / 3
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0(in_ptr0, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 1.0
tmp2 = tmp0 >= tmp1
tmp3 = tmp2.to(tl.float32)
tmp4 = 0.6666666666666666
tmp5 = tmp0 - tmp4
tmp6 = tmp3 * tmp5
tmp7 = -1.0
tmp8 = tmp0 <= tmp7
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp0 + tmp4
tmp11 = tmp9 * tmp10
tmp12 = tmp6 + tmp11
tmp13 = tmp0 > tmp7
tmp14 = tmp0 < tmp1
tmp15 = tmp13 & tmp14
tmp16 = tmp15.to(tl.float32)
tmp17 = tmp0 * tmp0
tmp18 = tmp17 * tmp0
tmp19 = tmp16 * tmp18
tmp20 = 0.3333333333333333
tmp21 = tmp19 * tmp20
tmp22 = tmp12 + tmp21
tl.store(out_ptr0 + x0, tmp22, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_div_ge_gt_le_lt_mul_pow_sub_0[grid(256)](arg0_1,
buf0, 256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class LipschitzCubeNew(nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| rh-ia/color-information | LipschitzCube | false | 4,281 | [
"MIT"
] | 0 | e912a1667e4fffb339dbc574c85020ec6cf78b02 | https://github.com/rh-ia/color-information/tree/e912a1667e4fffb339dbc574c85020ec6cf78b02 | import torch
from torch import nn
import torch.utils.data.distributed
class Model(nn.Module):
def forward(self, x):
return (x >= 1) * (x - 2 / 3) + (x <= -1) * (x + 2 / 3) + (x > -1) * (x
< 1) * x ** 3 / 3
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ExtendedModel | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ai/caitnpldotnv4k4oj67wyecd2ig4qcjrnmr35rmx6o2vxx245xs3.py
# Topologically Sorted Source Nodes: [h_relu], Original ATen: [aten.clamp, aten.ge]
# Source node to ATen node mapping:
# h_relu => clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%view_1, 0), kwargs = {})
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%view_1, 0), kwargs = {})
triton_poi_fused_clamp_ge_0 = async_compile.triton('triton_poi_fused_clamp_ge_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_ge_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clamp_ge_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = tmp2 >= tmp3
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/w3/cw3e5u4uwg7zpc5lz2eloe3l5how3uzo3r6hois7qr3hg5trjibq.py
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, 0.0), kwargs = {})
triton_poi_fused_add_1 = async_compile.triton('triton_poi_fused_add_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [h_relu], Original ATen: [aten.clamp, aten.ge]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_ge_0.run(buf0, primals_2, buf1, buf4, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
triton_poi_fused_add_1.run(buf3, primals_5, 256, grid=grid(256), stream=stream0)
del primals_5
return (buf3, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_4, buf4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class ExtendedModel(nn.Module):
def __init__(self, D_in, H, D_out):
"""
In the constructor we instantiate two nn.Linear modules and assign them as
member variables.
"""
super(ExtendedModel, self).__init__()
self.linear1 = nn.Linear(D_in, H)
self.linear2 = nn.Linear(H, D_out)
def forward(self, x, bias=0.0):
"""
In the forward function we accept a Tensor of input data and an optional bias
"""
h_relu = self.linear1(x).clamp(min=0)
y_pred = self.linear2(h_relu)
return y_pred + bias
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'D_in': 4, 'H': 4, 'D_out': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clamp_ge_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = tmp2 >= tmp3
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp5, xmask)
@triton.jit
def triton_poi_fused_add_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_clamp_ge_0[grid(256)](buf0, primals_2, buf1, buf4,
256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = buf0
del buf0
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
triton_poi_fused_add_1[grid(256)](buf3, primals_5, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_5
return buf3, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_4, buf4
class ExtendedModelNew(nn.Module):
def __init__(self, D_in, H, D_out):
"""
In the constructor we instantiate two nn.Linear modules and assign them as
member variables.
"""
super(ExtendedModelNew, self).__init__()
self.linear1 = nn.Linear(D_in, H)
self.linear2 = nn.Linear(H, D_out)
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| sauyon/BentoML | ExtendedModel | false | 4,282 | [
"Apache-2.0"
] | 0 | ff702f1fc1ee7cc4cf7aab2e67d1e27512858fe4 | https://github.com/sauyon/BentoML/tree/ff702f1fc1ee7cc4cf7aab2e67d1e27512858fe4 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, D_in, H, D_out):
"""
In the constructor we instantiate two nn.Linear modules and assign them as
member variables.
"""
super().__init__()
self.linear1 = nn.Linear(D_in, H)
self.linear2 = nn.Linear(H, D_out)
def forward(self, x, bias=0.0):
"""
In the forward function we accept a Tensor of input data and an optional bias
"""
h_relu = self.linear1(x).clamp(min=0)
y_pred = self.linear2(h_relu)
return y_pred + bias
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
FullSort | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/zu/czuykkrx3bycxkuqvpdouv5j3rcauet5ovtjh3346eg2mc7xjxpx.py
# Topologically Sorted Source Nodes: [sort], Original ATen: [aten.sort]
# Source node to ATen node mapping:
# sort => sort
# Graph fragment:
# %sort : [num_users=1] = call_function[target=torch.ops.aten.sort.default](args = (%arg0_1, 1), kwargs = {})
triton_per_fused_sort_0 = async_compile.triton('triton_per_fused_sort_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[64, 4],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_sort_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_sort_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 64
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex
x0 = xindex % 16
x1 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x0 + (16*r2) + (64*x1)), xmask, other=0.0)
tmp1 = r2
tmp2 = tmp1.to(tl.int16)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5, tmp6, = triton_helpers.sort_with_index(tmp3, tmp4, None, 1, stable=False, descending=False)
tl.store(out_ptr0 + (x0 + (16*r2) + (64*x1)), tmp5, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sort], Original ATen: [aten.sort]
stream0 = get_raw_stream(0)
triton_per_fused_sort_0.run(arg0_1, buf0, 64, 4, grid=grid(64), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.utils.data.distributed
class FullSort(nn.Module):
def forward(self, x):
return torch.sort(x, 1)[0]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_per_fused_sort_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK: tl.
constexpr):
xnumel = 64
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex
x0 = xindex % 16
x1 = xindex // 16
tmp0 = tl.load(in_ptr0 + (x0 + 16 * r2 + 64 * x1), xmask, other=0.0)
tmp1 = r2
tmp2 = tmp1.to(tl.int16)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5, _tmp6 = triton_helpers.sort_with_index(tmp3, tmp4, None, 1,
stable=False, descending=False)
tl.store(out_ptr0 + (x0 + 16 * r2 + 64 * x1), tmp5, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_sort_0[grid(64)](arg0_1, buf0, 64, 4, XBLOCK=1,
num_warps=2, num_stages=1)
del arg0_1
return buf0,
class FullSortNew(nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| rh-ia/color-information | FullSort | false | 4,283 | [
"MIT"
] | 0 | e912a1667e4fffb339dbc574c85020ec6cf78b02 | https://github.com/rh-ia/color-information/tree/e912a1667e4fffb339dbc574c85020ec6cf78b02 | import torch
from torch import nn
import torch.utils.data.distributed
class Model(nn.Module):
def forward(self, x):
return torch.sort(x, 1)[0]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Clamp | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/lt/cltgqtfsflyt6yb6tdfryeqzjuvjc6f5yyrsp5cvrn536kh76u7p.py
# Topologically Sorted Source Nodes: [clamp], Original ATen: [aten.clamp]
# Source node to ATen node mapping:
# clamp => clamp_max, clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%arg0_1, -3), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 3), kwargs = {})
triton_poi_fused_clamp_0 = async_compile.triton('triton_poi_fused_clamp_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clamp_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = -3.0
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp3 = 3.0
tmp4 = triton_helpers.minimum(tmp2, tmp3)
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [clamp], Original ATen: [aten.clamp]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.utils.data
class Clamp(nn.Module):
def __init__(self, min_out=-3, max_out=3):
super().__init__()
self.min_out = min_out
self.max_out = max_out
def forward(self, input):
return input.clamp(self.min_out, self.max_out)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_clamp_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = -3.0
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp3 = 3.0
tmp4 = triton_helpers.minimum(tmp2, tmp3)
tl.store(out_ptr0 + x0, tmp4, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clamp_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class ClampNew(nn.Module):
def __init__(self, min_out=-3, max_out=3):
super().__init__()
self.min_out = min_out
self.max_out = max_out
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| sbuschjaeger/Pysembles | Clamp | false | 4,284 | [
"MIT"
] | 0 | 7e69b0975a7d4373242c7026ade6c5fdbad4fe67 | https://github.com/sbuschjaeger/Pysembles/tree/7e69b0975a7d4373242c7026ade6c5fdbad4fe67 | import torch
from torch import nn
import torch.utils.data
class Model(nn.Module):
def __init__(self, min_out=-3, max_out=3):
super().__init__()
self.min_out = min_out
self.max_out = max_out
def forward(self, input):
return input.clamp(self.min_out, self.max_out)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SpatialAttentionGate | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/qv/cqvpm4tidhpw42vquodkna5kubx3c46djnnb2jim63auds7wtadt.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x => convolution
# x_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 16
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/go/cgofqcgduqrtcjakfd7uk3wkcrpwsqxispluihwsstry6ekodk2u.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.convolution, aten.sigmoid]
# Source node to ATen node mapping:
# x_2 => convolution_1
# x_3 => sigmoid
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_sigmoid_1 = async_compile.triton('triton_poi_fused_convolution_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_sigmoid_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 16, 1, 1), (16, 1, 1, 1))
assert_size_stride(primals_5, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 1024, grid=grid(1024), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 4, 4), (16, 16, 4, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.convolution, aten.sigmoid]
triton_poi_fused_convolution_sigmoid_1.run(buf3, primals_5, 64, grid=grid(64), stream=stream0)
del primals_5
return (buf3, primals_1, primals_3, primals_4, buf1, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((16, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 16, 1, 1), (16, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class SpatialAttentionGate(nn.Module):
def __init__(self, channel, reduction=16):
super(SpatialAttentionGate, self).__init__()
self.fc1 = nn.Conv2d(channel, reduction, kernel_size=1, padding=0)
self.fc2 = nn.Conv2d(reduction, 1, kernel_size=1, padding=0)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x, inplace=True)
x = self.fc2(x)
x = torch.sigmoid(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'channel': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 16
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, xmask)
@triton.jit
def triton_poi_fused_convolution_sigmoid_1(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 16, 1, 1), (16, 1, 1, 1))
assert_size_stride(primals_5, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(1024)](buf1, primals_2,
1024, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 4, 4), (16, 16, 4, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_sigmoid_1[grid(64)](buf3, primals_5,
64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_5
return buf3, primals_1, primals_3, primals_4, buf1, buf3
class SpatialAttentionGateNew(nn.Module):
def __init__(self, channel, reduction=16):
super(SpatialAttentionGateNew, self).__init__()
self.fc1 = nn.Conv2d(channel, reduction, kernel_size=1, padding=0)
self.fc2 = nn.Conv2d(reduction, 1, kernel_size=1, padding=0)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| savan77/nni | SpatialAttentionGate | false | 4,285 | [
"MIT"
] | 0 | 510213393d9cae58c5a8cccd21f322f7bba4e0cf | https://github.com/savan77/nni/tree/510213393d9cae58c5a8cccd21f322f7bba4e0cf | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, channel, reduction=16):
super().__init__()
self.fc1 = nn.Conv2d(channel, reduction, kernel_size=1, padding=0)
self.fc2 = nn.Conv2d(reduction, 1, kernel_size=1, padding=0)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x, inplace=True)
x = self.fc2(x)
x = torch.sigmoid(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
FlexibleDropout | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/yl/cylp6sds3y5rpwqc43muaio5wwmfwwsccgosjr3putgu5c5grjy6.py
# Topologically Sorted Source Nodes: [sub, mul], Original ATen: [aten.rsub, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# sub => sub
# Graph fragment:
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %arg1_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %sub), kwargs = {})
triton_poi_fused_mul_rsub_0 = async_compile.triton('triton_poi_fused_mul_rsub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_rsub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_rsub_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp3 = tl.load(in_ptr1 + (x0), xmask)
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp3 * tmp2
tl.store(out_ptr0 + (x0), tmp2, xmask)
tl.store(out_ptr1 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sub, mul], Original ATen: [aten.rsub, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_rsub_0.run(arg1_1, arg0_1, buf0, buf1, 256, grid=grid(256), stream=stream0)
del arg0_1
del arg1_1
return (buf1, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
from torch.distributions import Bernoulli
class FlexibleDropout(nn.Module):
"""FlexibleDropout disconnects the sampling step from the masking step of dropout.
There are two important differences between FlexibleDropout and nn.Dropout. First, FlexibleDropout exposes a
sample_mask and apply_mask function, that allows for the same mask to be used repeatedly. Second, FlexibleDropout
scales the input at test time with p, as opposed to scaling with 1/p at training time. This is convenient when
Dropout is used for uncertainty estimation.
"""
def __init__(self):
super().__init__()
def forward(self, input, p, shape=None):
"""Similar to F.dropout, a mask is sampled and directly applied to the input."""
if shape is None:
self.sample_mask(p, input.shape)
else:
self.sample_mask(p, shape)
return self.apply_mask(input)
def apply_mask(self, input):
"""Applies the sampled mask to the input."""
return input * self._mask
def sample_mask(self, p, shape):
"""Samples a dropout mask from a Bernoulli distribution.
Args:
p(float): the dropout probability [0, 1].
shape(torch.Size): shape of the mask to be sampled.
"""
if self.training:
self._mask = Bernoulli(1.0 - p).sample(shape)
else:
self._mask = 1.0 - p
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
from torch.distributions import Bernoulli
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_mul_rsub_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp3 = tl.load(in_ptr1 + x0, xmask)
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp3 * tmp2
tl.store(out_ptr0 + x0, tmp2, xmask)
tl.store(out_ptr1 + x0, tmp4, xmask)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_rsub_0[grid(256)](arg1_1, arg0_1, buf0, buf1,
256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf1, buf0
class FlexibleDropoutNew(nn.Module):
"""FlexibleDropout disconnects the sampling step from the masking step of dropout.
There are two important differences between FlexibleDropout and nn.Dropout. First, FlexibleDropout exposes a
sample_mask and apply_mask function, that allows for the same mask to be used repeatedly. Second, FlexibleDropout
scales the input at test time with p, as opposed to scaling with 1/p at training time. This is convenient when
Dropout is used for uncertainty estimation.
"""
def __init__(self):
super().__init__()
def apply_mask(self, input):
"""Applies the sampled mask to the input."""
return input * self._mask
def sample_mask(self, p, shape):
"""Samples a dropout mask from a Bernoulli distribution.
Args:
p(float): the dropout probability [0, 1].
shape(torch.Size): shape of the mask to be sampled.
"""
if self.training:
self._mask = Bernoulli(1.0 - p).sample(shape)
else:
self._mask = 1.0 - p
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| scfrank/deep-generative-lm | FlexibleDropout | false | 4,286 | [
"MIT"
] | 0 | 70067fcda82aa035bba805ce6c2709097166a7a4 | https://github.com/scfrank/deep-generative-lm/tree/70067fcda82aa035bba805ce6c2709097166a7a4 | import torch
import torch.nn as nn
from torch.distributions import Bernoulli
class Model(nn.Module):
"""FlexibleDropout disconnects the sampling step from the masking step of dropout.
There are two important differences between FlexibleDropout and nn.Dropout. First, FlexibleDropout exposes a
sample_mask and apply_mask function, that allows for the same mask to be used repeatedly. Second, FlexibleDropout
scales the input at test time with p, as opposed to scaling with 1/p at training time. This is convenient when
Dropout is used for uncertainty estimation.
"""
def __init__(self):
super().__init__()
def forward(self, input, p, shape=None):
"""Similar to F.dropout, a mask is sampled and directly applied to the input."""
if shape is None:
self.sample_mask(p, input.shape)
else:
self.sample_mask(p, shape)
return self.apply_mask(input)
def apply_mask(self, input):
"""Applies the sampled mask to the input."""
return input * self._mask
def sample_mask(self, p, shape):
"""Samples a dropout mask from a Bernoulli distribution.
Args:
p(float): the dropout probability [0, 1].
shape(torch.Size): shape of the mask to be sampled.
"""
if self.training:
self._mask = Bernoulli(1.0 - p).sample(shape)
else:
self._mask = 1.0 - p
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
BertImagePooler | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/yy/cyya3js6wt64vdji3sfisvrqyfvqxwkwqq5mzg5bqjl2crzjs4t3.py
# Topologically Sorted Source Nodes: [pooled_output], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# pooled_output => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%select,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xk/cxkfjvxcrwrocrik25vel4gb2spp4jrbijo33ra4mgkw3hn2qgah.py
# Topologically Sorted Source Nodes: [pooled_output, pooled_output_1], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# pooled_output => add
# pooled_output_1 => relu
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_3), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_add_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pooled_output], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(primals_1, buf0, 64, grid=grid(64), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pooled_output], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf1)
del primals_2
buf2 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0); del buf1 # reuse
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [pooled_output, pooled_output_1], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_1.run(buf2, primals_3, buf3, 64, grid=grid(64), stream=stream0)
del primals_3
return (buf2, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import torch
import torch.nn as nn
import torch.multiprocessing
class BertImagePooler(nn.Module):
def __init__(self, config):
super(BertImagePooler, self).__init__()
self.dense = nn.Linear(config.v_hidden_size, config.bi_hidden_size)
self.activation = nn.ReLU()
def forward(self, hidden_states):
first_token_tensor = hidden_states[:, 0]
pooled_output = self.dense(first_token_tensor)
pooled_output = self.activation(pooled_output)
return pooled_output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'config': _mock_config(v_hidden_size=4, bi_hidden_size=4)}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.multiprocessing
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64)](primals_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf1)
del primals_2
buf2 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0)
del buf1
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_1[grid(64)](buf2,
primals_3, buf3, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_3
return buf2, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), buf3
class BertImagePoolerNew(nn.Module):
def __init__(self, config):
super(BertImagePoolerNew, self).__init__()
self.dense = nn.Linear(config.v_hidden_size, config.bi_hidden_size)
self.activation = nn.ReLU()
def forward(self, input_0):
primals_2 = self.dense.weight
primals_3 = self.dense.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| ayushjain1144/vilbert-multi-task | BertImagePooler | false | 4,287 | [
"MIT"
] | 0 | cf30feee9617dd92bb030f380f8b59388b7054f6 | https://github.com/ayushjain1144/vilbert-multi-task/tree/cf30feee9617dd92bb030f380f8b59388b7054f6 | from _paritybench_helpers import _mock_config
import torch
import torch.nn as nn
import torch.multiprocessing
class Model(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.v_hidden_size, config.bi_hidden_size)
self.activation = nn.ReLU()
def forward(self, hidden_states):
first_token_tensor = hidden_states[:, 0]
pooled_output = self.dense(first_token_tensor)
pooled_output = self.activation(pooled_output)
return pooled_output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
LipNormConv2d | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/in/cinvfeqy6ry6tu24qtajhlnsldmnir4gax6ikx6vnnqb4avr5ien.py
# Topologically Sorted Source Nodes: [w_scale, truediv, sigmoid, weight], Original ATen: [aten.norm, aten.div, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# sigmoid => sigmoid
# truediv => div
# w_scale => abs_1, pow_2, sum_1
# weight => mul
# Graph fragment:
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%view,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%abs_1, [1]), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_1, 1.0), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_1, %view_1), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%primals_2,), kwargs = {})
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %sigmoid), kwargs = {})
triton_per_fused_div_mul_norm_sigmoid_0 = async_compile.triton('triton_per_fused_div_mul_norm_sigmoid_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_div_mul_norm_sigmoid_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_div_mul_norm_sigmoid_0(in_out_ptr0, in_ptr0, in_ptr1, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 64
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (64*x0)), xmask, other=0.0)
tmp7 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl_math.abs(tmp0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.where(xmask, tmp2, 0)
tmp5 = tl.sum(tmp4, 1)[:, None]
tmp6 = tmp0 / tmp5
tmp8 = tl.sigmoid(tmp7)
tmp9 = tmp6 * tmp8
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp5, xmask)
tl.store(out_ptr0 + (r1 + (64*x0)), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/2i/c2ia6clymej2axaxwh5dhlf5hhex6emmkbazo7542zo3gcyaffyw.py
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_4, %mul, %primals_3, [1, 1], [4, 4], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1296
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 81) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf1 = buf0; del buf0 # reuse
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [w_scale, truediv, sigmoid, weight], Original ATen: [aten.norm, aten.div, aten.sigmoid, aten.mul]
stream0 = get_raw_stream(0)
triton_per_fused_div_mul_norm_sigmoid_0.run(buf1, primals_1, primals_2, buf2, 4, 64, grid=grid(4), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf3 = extern_kernels.convolution(primals_4, buf2, stride=(1, 1), padding=(4, 4), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 9, 9), (324, 81, 9, 1))
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf4, primals_3, 1296, grid=grid(1296), stream=stream0)
del primals_3
return (buf4, primals_1, primals_2, primals_4, reinterpret_tensor(buf1, (4, 1, 1, 1), (1, 1, 1, 1), 0), buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.nn.functional as F
import torch.utils.data.distributed
def _max_except_dim(input, dim):
maxed = input
for axis in range(input.ndimension() - 1, dim, -1):
maxed, _ = maxed.max(axis, keepdim=True)
for axis in range(dim - 1, -1, -1):
maxed, _ = maxed.max(axis, keepdim=True)
return maxed
def _norm_except_dim(w, norm_type, dim):
if norm_type == 1 or norm_type == 2:
return torch.norm_except_dim(w, norm_type, dim)
elif norm_type == float('inf'):
return _max_except_dim(w, dim)
def operator_norm_settings(domain, codomain):
if domain == 1 and codomain == 1:
max_across_input_dims = True
norm_type = 1
elif domain == 1 and codomain == 2:
max_across_input_dims = True
norm_type = 2
elif domain == 1 and codomain == float('inf'):
max_across_input_dims = True
norm_type = float('inf')
elif domain == 2 and codomain == float('inf'):
max_across_input_dims = False
norm_type = 2
elif domain == float('inf') and codomain == float('inf'):
max_across_input_dims = False
norm_type = 1
else:
raise ValueError('Unknown combination of domain "{}" and codomain "{}"'
.format(domain, codomain))
return max_across_input_dims, norm_type
def _logit(p):
p = torch.max(torch.ones(1) * 0.1, torch.min(torch.ones(1) * 0.9, p))
return torch.log(p + 1e-10) + torch.log(1 - p + 1e-10)
class LipNormConv2d(nn.Conv2d):
"""Lipschitz constant defined using operator norms."""
def __init__(self, in_channels, out_channels, kernel_size, stride,
padding, bias=True, coeff=0.97, domain=float('inf'), codomain=float
('inf'), local_constraint=True, **unused_kwargs):
del unused_kwargs
super(LipNormConv2d, self).__init__(in_channels, out_channels,
kernel_size, stride, padding, bias)
self.coeff = coeff
self.domain = domain
self.codomain = codomain
self.local_constraint = local_constraint
max_across_input_dims, self.norm_type = operator_norm_settings(self
.domain, self.codomain)
self.max_across_dim = 1 if max_across_input_dims else 0
with torch.no_grad():
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=
self.max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
self.scale = nn.Parameter(_logit(w_scale / self.coeff))
def compute_weight(self):
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=self.
max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
return self.weight / w_scale * torch.sigmoid(self.scale)
def forward(self, input):
weight = self.compute_weight()
return F.conv2d(input, weight, self.bias, self.stride, self.padding,
1, 1)
def extra_repr(self):
s = super(LipNormConv2d, self).extra_repr()
return s + ', coeff={}, domain={}, codomain={}, local={}'.format(self
.coeff, self.domain, self.codomain, self.local_constraint)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4, 'kernel_size': 4,
'stride': 1, 'padding': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_div_mul_norm_sigmoid_0(in_out_ptr0, in_ptr0, in_ptr1,
out_ptr0, xnumel, rnumel, XBLOCK: tl.constexpr):
xnumel = 4
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 64 * x0), xmask, other=0.0)
tmp7 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl_math.abs(tmp0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.where(xmask, tmp2, 0)
tmp5 = tl.sum(tmp4, 1)[:, None]
tmp6 = tmp0 / tmp5
tmp8 = tl.sigmoid(tmp7)
tmp9 = tmp6 * tmp8
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp5, xmask)
tl.store(out_ptr0 + (r1 + 64 * x0), tmp9, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 1296
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 81 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4,), (1,), torch.float32)
buf1 = buf0
del buf0
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_div_mul_norm_sigmoid_0[grid(4)](buf1, primals_1,
primals_2, buf2, 4, 64, XBLOCK=1, num_warps=2, num_stages=1)
buf3 = extern_kernels.convolution(primals_4, buf2, stride=(1, 1),
padding=(4, 4), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 9, 9), (324, 81, 9, 1))
buf4 = buf3
del buf3
triton_poi_fused_convolution_1[grid(1296)](buf4, primals_3, 1296,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_3
return buf4, primals_1, primals_2, primals_4, reinterpret_tensor(buf1,
(4, 1, 1, 1), (1, 1, 1, 1), 0), buf2
def _max_except_dim(input, dim):
maxed = input
for axis in range(input.ndimension() - 1, dim, -1):
maxed, _ = maxed.max(axis, keepdim=True)
for axis in range(dim - 1, -1, -1):
maxed, _ = maxed.max(axis, keepdim=True)
return maxed
def _norm_except_dim(w, norm_type, dim):
if norm_type == 1 or norm_type == 2:
return torch.norm_except_dim(w, norm_type, dim)
elif norm_type == float('inf'):
return _max_except_dim(w, dim)
def operator_norm_settings(domain, codomain):
if domain == 1 and codomain == 1:
max_across_input_dims = True
norm_type = 1
elif domain == 1 and codomain == 2:
max_across_input_dims = True
norm_type = 2
elif domain == 1 and codomain == float('inf'):
max_across_input_dims = True
norm_type = float('inf')
elif domain == 2 and codomain == float('inf'):
max_across_input_dims = False
norm_type = 2
elif domain == float('inf') and codomain == float('inf'):
max_across_input_dims = False
norm_type = 1
else:
raise ValueError('Unknown combination of domain "{}" and codomain "{}"'
.format(domain, codomain))
return max_across_input_dims, norm_type
def _logit(p):
p = torch.max(torch.ones(1) * 0.1, torch.min(torch.ones(1) * 0.9, p))
return torch.log(p + 1e-10) + torch.log(1 - p + 1e-10)
class LipNormConv2dNew(nn.Conv2d):
"""Lipschitz constant defined using operator norms."""
def __init__(self, in_channels, out_channels, kernel_size, stride,
padding, bias=True, coeff=0.97, domain=float('inf'), codomain=float
('inf'), local_constraint=True, **unused_kwargs):
del unused_kwargs
super(LipNormConv2dNew, self).__init__(in_channels, out_channels,
kernel_size, stride, padding, bias)
self.coeff = coeff
self.domain = domain
self.codomain = codomain
self.local_constraint = local_constraint
max_across_input_dims, self.norm_type = operator_norm_settings(self
.domain, self.codomain)
self.max_across_dim = 1 if max_across_input_dims else 0
with torch.no_grad():
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=
self.max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
self.scale = nn.Parameter(_logit(w_scale / self.coeff))
def compute_weight(self):
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=self.
max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
return self.weight / w_scale * torch.sigmoid(self.scale)
def extra_repr(self):
s = super(LipNormConv2dNew, self).extra_repr()
return s + ', coeff={}, domain={}, codomain={}, local={}'.format(self
.coeff, self.domain, self.codomain, self.local_constraint)
def forward(self, input_0):
primals_1 = self.weight
primals_3 = self.bias
primals_2 = self.scale
primals_4 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| rh-ia/color-information | LipNormConv2d | false | 4,288 | [
"MIT"
] | 0 | e912a1667e4fffb339dbc574c85020ec6cf78b02 | https://github.com/rh-ia/color-information/tree/e912a1667e4fffb339dbc574c85020ec6cf78b02 | import torch
from torch import nn
import torch.nn.functional as F
import torch.utils.data.distributed
def _max_except_dim(input, dim):
maxed = input
for axis in range(input.ndimension() - 1, dim, -1):
maxed, _ = maxed.max(axis, keepdim=True)
for axis in range(dim - 1, -1, -1):
maxed, _ = maxed.max(axis, keepdim=True)
return maxed
def _norm_except_dim(w, norm_type, dim):
if norm_type == 1 or norm_type == 2:
return torch.norm_except_dim(w, norm_type, dim)
elif norm_type == float('inf'):
return _max_except_dim(w, dim)
def operator_norm_settings(domain, codomain):
if domain == 1 and codomain == 1:
max_across_input_dims = True
norm_type = 1
elif domain == 1 and codomain == 2:
max_across_input_dims = True
norm_type = 2
elif domain == 1 and codomain == float('inf'):
max_across_input_dims = True
norm_type = float('inf')
elif domain == 2 and codomain == float('inf'):
max_across_input_dims = False
norm_type = 2
elif domain == float('inf') and codomain == float('inf'):
max_across_input_dims = False
norm_type = 1
else:
raise ValueError('Unknown combination of domain "{}" and codomain "{}"'
.format(domain, codomain))
return max_across_input_dims, norm_type
def _logit(p):
p = torch.max(torch.ones(1) * 0.1, torch.min(torch.ones(1) * 0.9, p))
return torch.log(p + 1e-10) + torch.log(1 - p + 1e-10)
class Model(nn.Conv2d):
"""Lipschitz constant defined using operator norms."""
def __init__(self, in_channels, out_channels, kernel_size, stride,
padding, bias=True, coeff=0.97, domain=float('inf'), codomain=float
('inf'), local_constraint=True, **unused_kwargs):
del unused_kwargs
super().__init__(in_channels, out_channels,
kernel_size, stride, padding, bias)
self.coeff = coeff
self.domain = domain
self.codomain = codomain
self.local_constraint = local_constraint
max_across_input_dims, self.norm_type = operator_norm_settings(self
.domain, self.codomain)
self.max_across_dim = 1 if max_across_input_dims else 0
with torch.no_grad():
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=
self.max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
self.scale = nn.Parameter(_logit(w_scale / self.coeff))
def compute_weight(self):
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=self.
max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
return self.weight / w_scale * torch.sigmoid(self.scale)
def forward(self, input):
weight = self.compute_weight()
return F.conv2d(input, weight, self.bias, self.stride, self.padding,
1, 1)
def extra_repr(self):
s = super(LipNormConv2d, self).extra_repr()
return s + ', coeff={}, domain={}, codomain={}, local={}'.format(self
.coeff, self.domain, self.codomain, self.local_constraint)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4, 'kernel_size': 4,
'stride': 1, 'padding': 4}]
|
RelevanceVector | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/fu/cfua5hj4rr7lkakrjc25n3iarq6qizang3sfgcx4u6k2fufmtysu.py
# Topologically Sorted Source Nodes: [rv], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# rv => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%primals_1,), kwargs = {})
triton_poi_fused_sigmoid_0 = async_compile.triton('triton_poi_fused_sigmoid_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tl.store(out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, = args
args.clear()
assert_size_stride(primals_1, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [rv], Original ATen: [aten.sigmoid]
stream0 = get_raw_stream(0)
triton_poi_fused_sigmoid_0.run(primals_1, buf0, 4, grid=grid(4), stream=stream0)
del primals_1
return (buf0, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class RelevanceVector(nn.Module):
def __init__(self, z_dim):
super(RelevanceVector, self).__init__()
self.rvlogit = nn.Parameter(0.001 * torch.randn(z_dim))
def forward(self):
rv = torch.sigmoid(self.rvlogit)
return self.rvlogit, rv
def get_inputs():
return []
def get_init_inputs():
return [[], {'z_dim': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_sigmoid_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tl.store(out_ptr0 + x0, tmp1, xmask)
def call(args):
primals_1, = args
args.clear()
assert_size_stride(primals_1, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4,), (1,), torch.float32)
get_raw_stream(0)
triton_poi_fused_sigmoid_0[grid(4)](primals_1, buf0, 4, XBLOCK=4,
num_warps=1, num_stages=1)
del primals_1
return buf0, buf0
class RelevanceVectorNew(nn.Module):
def __init__(self, z_dim):
super(RelevanceVectorNew, self).__init__()
self.rvlogit = nn.Parameter(0.001 * torch.randn(z_dim))
def forward(self):
primals_1 = self.rvlogit
output = call([primals_1])
return output[0], output[1]
| seqam-lab/rfvae | RelevanceVector | false | 4,289 | [
"MIT"
] | 0 | 07089e2cca6d51f305731750c2c67b83a42df12a | https://github.com/seqam-lab/rfvae/tree/07089e2cca6d51f305731750c2c67b83a42df12a | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, z_dim):
super().__init__()
self.rvlogit = nn.Parameter(0.001 * torch.randn(z_dim))
def forward(self):
rv = torch.sigmoid(self.rvlogit)
return self.rvlogit, rv
def get_inputs():
return []
def get_init_inputs():
return [4]
|
QNetwork | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/nq/cnqjufcqn3ur3s7xvlb2i747nyf24md4zaiatlwgkasynplfjstu.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/dh/cdhj4aozvvzkw7stzrqoauyoij3petwtvi4g4weydesiaurrughd.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_1 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (128, 64), (64, 1))
assert_size_stride(primals_5, (128, ), (1, ))
assert_size_stride(primals_6, (4, 128), (128, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf0 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf6, 4096, grid=grid(4096), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0), reinterpret_tensor(primals_4, (64, 128), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 128), (2048, 512, 128, 1), 0); del buf2 # reuse
buf5 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf5, 8192, grid=grid(8192), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 128), (128, 1), 0), reinterpret_tensor(primals_6, (128, 4), (1, 128), 0), alpha=1, beta=1, out=buf4)
del primals_7
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 64), (64, 1), 0), reinterpret_tensor(buf3, (64, 128), (128, 1), 0), primals_6, buf5, primals_4, buf6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((128, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 128), (128, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn.functional as F
import torch.nn as nn
class QNetwork(nn.Module):
"""Actor (Policy) Model."""
def __init__(self, state_size, action_size, seed, fc1_units=64, fc2=128):
"""Initialize parameters and build model.
Params
======
state_size (int): Dimension of each state
action_size (int): Dimension of each action
seed (int): Random seed
fc1_units (int): Number of nodes in first hidden layer
fc2 (int): Number of nodes in the hidden layers
"""
super(QNetwork, self).__init__()
self.seed = torch.manual_seed(seed)
self.fc1 = nn.Linear(state_size, fc1_units)
self.fc2 = nn.Linear(fc1_units, fc2)
self.fc3 = nn.Linear(fc2, action_size)
def forward(self, state):
"""Build a network that maps state -> action values."""
x = F.relu(self.fc1(state))
x = F.relu(self.fc2(x))
return self.fc3(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_size': 4, 'action_size': 4, 'seed': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (128, 64), (64, 1))
assert_size_stride(primals_5, (128,), (1,))
assert_size_stride(primals_6, (4, 128), (128, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf0
buf6 = empty_strided_cuda((4, 4, 4, 64), (1024, 256, 64, 1), torch.bool
)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(4096)](buf1,
primals_2, buf6, 4096, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0),
reinterpret_tensor(primals_4, (64, 128), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 128), (2048, 512, 128, 1), 0)
del buf2
buf5 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(8192)](buf3,
primals_5, buf5, 8192, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 128),
(128, 1), 0), reinterpret_tensor(primals_6, (128, 4), (1, 128),
0), alpha=1, beta=1, out=buf4)
del primals_7
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 64), (64, 1), 0), reinterpret_tensor(
buf3, (64, 128), (128, 1), 0), primals_6, buf5, primals_4, buf6
class QNetworkNew(nn.Module):
"""Actor (Policy) Model."""
def __init__(self, state_size, action_size, seed, fc1_units=64, fc2=128):
"""Initialize parameters and build model.
Params
======
state_size (int): Dimension of each state
action_size (int): Dimension of each action
seed (int): Random seed
fc1_units (int): Number of nodes in first hidden layer
fc2 (int): Number of nodes in the hidden layers
"""
super(QNetworkNew, self).__init__()
self.seed = torch.manual_seed(seed)
self.fc1 = nn.Linear(state_size, fc1_units)
self.fc2 = nn.Linear(fc1_units, fc2)
self.fc3 = nn.Linear(fc2, action_size)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| schottkey7/deep-reinforcement-learning | QNetwork | false | 4,290 | [
"MIT"
] | 0 | 92c97fadbb5b95caa3fd3813a0757debc2c2747a | https://github.com/schottkey7/deep-reinforcement-learning/tree/92c97fadbb5b95caa3fd3813a0757debc2c2747a | import torch
import torch.nn.functional as F
import torch.nn as nn
class Model(nn.Module):
"""Actor (Policy) Model."""
def __init__(self, state_size, action_size, seed, fc1_units=64, fc2=128):
"""Initialize parameters and build model.
Params
======
state_size (int): Dimension of each state
action_size (int): Dimension of each action
seed (int): Random seed
fc1_units (int): Number of nodes in first hidden layer
fc2 (int): Number of nodes in the hidden layers
"""
super().__init__()
self.seed = torch.manual_seed(seed)
self.fc1 = nn.Linear(state_size, fc1_units)
self.fc2 = nn.Linear(fc1_units, fc2)
self.fc3 = nn.Linear(fc2, action_size)
def forward(self, state):
"""Build a network that maps state -> action values."""
x = F.relu(self.fc1(state))
x = F.relu(self.fc2(x))
return self.fc3(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
LipNormLinear | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/cs/ccs52a7zexgfgg42ljzuxkjmhqfue6bfcp5dftqxbw4sww5mifsx.py
# Topologically Sorted Source Nodes: [truediv, sigmoid, mul, weight], Original ATen: [aten.div, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# sigmoid => sigmoid
# truediv => div
# weight => mul_1
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_1, %view_1), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%primals_2,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %sigmoid), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, 0.97), kwargs = {})
triton_poi_fused_div_mul_sigmoid_0 = async_compile.triton('triton_poi_fused_div_mul_sigmoid_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_mul_sigmoid_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_div_mul_sigmoid_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tl_math.abs(tmp1)
tmp4 = tl_math.abs(tmp3)
tmp5 = tmp2 + tmp4
tmp7 = tl_math.abs(tmp6)
tmp8 = tmp5 + tmp7
tmp10 = tl_math.abs(tmp9)
tmp11 = tmp8 + tmp10
tmp12 = tmp0 / tmp11
tmp14 = tl.sigmoid(tmp13)
tmp15 = tmp12 * tmp14
tmp16 = 0.97
tmp17 = tmp15 * tmp16
tl.store(out_ptr0 + (x2), tmp17, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 1), (1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv, sigmoid, mul, weight], Original ATen: [aten.div, aten.sigmoid, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_div_mul_sigmoid_0.run(primals_1, primals_2, buf0, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del buf0
del primals_3
return (reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_1, primals_2, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.nn.functional as F
import torch.utils.data.distributed
def _max_except_dim(input, dim):
maxed = input
for axis in range(input.ndimension() - 1, dim, -1):
maxed, _ = maxed.max(axis, keepdim=True)
for axis in range(dim - 1, -1, -1):
maxed, _ = maxed.max(axis, keepdim=True)
return maxed
def _norm_except_dim(w, norm_type, dim):
if norm_type == 1 or norm_type == 2:
return torch.norm_except_dim(w, norm_type, dim)
elif norm_type == float('inf'):
return _max_except_dim(w, dim)
def operator_norm_settings(domain, codomain):
if domain == 1 and codomain == 1:
max_across_input_dims = True
norm_type = 1
elif domain == 1 and codomain == 2:
max_across_input_dims = True
norm_type = 2
elif domain == 1 and codomain == float('inf'):
max_across_input_dims = True
norm_type = float('inf')
elif domain == 2 and codomain == float('inf'):
max_across_input_dims = False
norm_type = 2
elif domain == float('inf') and codomain == float('inf'):
max_across_input_dims = False
norm_type = 1
else:
raise ValueError('Unknown combination of domain "{}" and codomain "{}"'
.format(domain, codomain))
return max_across_input_dims, norm_type
def _logit(p):
p = torch.max(torch.ones(1) * 0.1, torch.min(torch.ones(1) * 0.9, p))
return torch.log(p + 1e-10) + torch.log(1 - p + 1e-10)
class LipNormLinear(nn.Linear):
"""Lipschitz constant defined using operator norms."""
def __init__(self, in_features, out_features, bias=True, coeff=0.97,
domain=float('inf'), codomain=float('inf'), local_constraint=True,
**unused_kwargs):
del unused_kwargs
super(LipNormLinear, self).__init__(in_features, out_features, bias)
self.coeff = coeff
self.domain = domain
self.codomain = codomain
self.local_constraint = local_constraint
max_across_input_dims, self.norm_type = operator_norm_settings(self
.domain, self.codomain)
self.max_across_dim = 1 if max_across_input_dims else 0
with torch.no_grad():
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=
self.max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
self.scale = nn.Parameter(_logit(w_scale / self.coeff))
def compute_weight(self):
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=self.
max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
return self.weight / w_scale * torch.sigmoid(self.scale) * self.coeff
def forward(self, input):
weight = self.compute_weight()
return F.linear(input, weight, self.bias)
def extra_repr(self):
s = super(LipNormLinear, self).extra_repr()
return s + ', coeff={}, domain={}, codomain={}, local={}'.format(self
.coeff, self.domain, self.codomain, self.local_constraint)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_features': 4, 'out_features': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_div_mul_sigmoid_0(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tl_math.abs(tmp1)
tmp4 = tl_math.abs(tmp3)
tmp5 = tmp2 + tmp4
tmp7 = tl_math.abs(tmp6)
tmp8 = tmp5 + tmp7
tmp10 = tl_math.abs(tmp9)
tmp11 = tmp8 + tmp10
tmp12 = tmp0 / tmp11
tmp14 = tl.sigmoid(tmp13)
tmp15 = tmp12 * tmp14
tmp16 = 0.97
tmp17 = tmp15 * tmp16
tl.store(out_ptr0 + x2, tmp17, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 1), (1, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_div_mul_sigmoid_0[grid(16)](primals_1, primals_2,
buf0, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_4, (64,
4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf1)
del buf0
del primals_3
return reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0
), primals_1, primals_2, reinterpret_tensor(primals_4, (64, 4), (4,
1), 0)
def _max_except_dim(input, dim):
maxed = input
for axis in range(input.ndimension() - 1, dim, -1):
maxed, _ = maxed.max(axis, keepdim=True)
for axis in range(dim - 1, -1, -1):
maxed, _ = maxed.max(axis, keepdim=True)
return maxed
def _norm_except_dim(w, norm_type, dim):
if norm_type == 1 or norm_type == 2:
return torch.norm_except_dim(w, norm_type, dim)
elif norm_type == float('inf'):
return _max_except_dim(w, dim)
def operator_norm_settings(domain, codomain):
if domain == 1 and codomain == 1:
max_across_input_dims = True
norm_type = 1
elif domain == 1 and codomain == 2:
max_across_input_dims = True
norm_type = 2
elif domain == 1 and codomain == float('inf'):
max_across_input_dims = True
norm_type = float('inf')
elif domain == 2 and codomain == float('inf'):
max_across_input_dims = False
norm_type = 2
elif domain == float('inf') and codomain == float('inf'):
max_across_input_dims = False
norm_type = 1
else:
raise ValueError('Unknown combination of domain "{}" and codomain "{}"'
.format(domain, codomain))
return max_across_input_dims, norm_type
def _logit(p):
p = torch.max(torch.ones(1) * 0.1, torch.min(torch.ones(1) * 0.9, p))
return torch.log(p + 1e-10) + torch.log(1 - p + 1e-10)
class LipNormLinearNew(nn.Linear):
"""Lipschitz constant defined using operator norms."""
def __init__(self, in_features, out_features, bias=True, coeff=0.97,
domain=float('inf'), codomain=float('inf'), local_constraint=True,
**unused_kwargs):
del unused_kwargs
super(LipNormLinearNew, self).__init__(in_features, out_features, bias)
self.coeff = coeff
self.domain = domain
self.codomain = codomain
self.local_constraint = local_constraint
max_across_input_dims, self.norm_type = operator_norm_settings(self
.domain, self.codomain)
self.max_across_dim = 1 if max_across_input_dims else 0
with torch.no_grad():
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=
self.max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
self.scale = nn.Parameter(_logit(w_scale / self.coeff))
def compute_weight(self):
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=self.
max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
return self.weight / w_scale * torch.sigmoid(self.scale) * self.coeff
def extra_repr(self):
s = super(LipNormLinearNew, self).extra_repr()
return s + ', coeff={}, domain={}, codomain={}, local={}'.format(self
.coeff, self.domain, self.codomain, self.local_constraint)
def forward(self, input_0):
primals_1 = self.weight
primals_3 = self.bias
primals_2 = self.scale
primals_4 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| rh-ia/color-information | LipNormLinear | false | 4,291 | [
"MIT"
] | 0 | e912a1667e4fffb339dbc574c85020ec6cf78b02 | https://github.com/rh-ia/color-information/tree/e912a1667e4fffb339dbc574c85020ec6cf78b02 | import torch
from torch import nn
import torch.nn.functional as F
import torch.utils.data.distributed
def _max_except_dim(input, dim):
maxed = input
for axis in range(input.ndimension() - 1, dim, -1):
maxed, _ = maxed.max(axis, keepdim=True)
for axis in range(dim - 1, -1, -1):
maxed, _ = maxed.max(axis, keepdim=True)
return maxed
def _norm_except_dim(w, norm_type, dim):
if norm_type == 1 or norm_type == 2:
return torch.norm_except_dim(w, norm_type, dim)
elif norm_type == float('inf'):
return _max_except_dim(w, dim)
def operator_norm_settings(domain, codomain):
if domain == 1 and codomain == 1:
max_across_input_dims = True
norm_type = 1
elif domain == 1 and codomain == 2:
max_across_input_dims = True
norm_type = 2
elif domain == 1 and codomain == float('inf'):
max_across_input_dims = True
norm_type = float('inf')
elif domain == 2 and codomain == float('inf'):
max_across_input_dims = False
norm_type = 2
elif domain == float('inf') and codomain == float('inf'):
max_across_input_dims = False
norm_type = 1
else:
raise ValueError('Unknown combination of domain "{}" and codomain "{}"'
.format(domain, codomain))
return max_across_input_dims, norm_type
def _logit(p):
p = torch.max(torch.ones(1) * 0.1, torch.min(torch.ones(1) * 0.9, p))
return torch.log(p + 1e-10) + torch.log(1 - p + 1e-10)
class Model(nn.Linear):
"""Lipschitz constant defined using operator norms."""
def __init__(self, in_features, out_features, bias=True, coeff=0.97,
domain=float('inf'), codomain=float('inf'), local_constraint=True,
**unused_kwargs):
del unused_kwargs
super().__init__(in_features, out_features, bias)
self.coeff = coeff
self.domain = domain
self.codomain = codomain
self.local_constraint = local_constraint
max_across_input_dims, self.norm_type = operator_norm_settings(self
.domain, self.codomain)
self.max_across_dim = 1 if max_across_input_dims else 0
with torch.no_grad():
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=
self.max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
self.scale = nn.Parameter(_logit(w_scale / self.coeff))
def compute_weight(self):
w_scale = _norm_except_dim(self.weight, self.norm_type, dim=self.
max_across_dim)
if not self.local_constraint:
w_scale = w_scale.max()
return self.weight / w_scale * torch.sigmoid(self.scale) * self.coeff
def forward(self, input):
weight = self.compute_weight()
return F.linear(input, weight, self.bias)
def extra_repr(self):
s = super(LipNormLinear, self).extra_repr()
return s + ', coeff={}, domain={}, codomain={}, local={}'.format(self
.coeff, self.domain, self.codomain, self.local_constraint)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Net1 | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/lg/clgnkdclmu24adiu7cbx7ybrhsind74uypy3cvuihzwv4hxzyjlf.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x => convolution
# x_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[524288],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 492032
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 3844) % 32
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/s6/cs6oowloxihsa2iqf677aggor4dgxx3q4fq7upsksc3m7rf4j6xb.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x_2 => convolution_1
# x_3 => relu_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_relu_1 = async_compile.triton('triton_poi_fused_convolution_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1048576],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 921600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 3600) % 64
x0 = xindex % 3600
x4 = (xindex // 3600)
tmp0 = tl.load(in_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(out_ptr0 + (x0 + (3616*x4)), tmp4, None)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/6v/c6votzbefolf4fzzmnc4okxekuuh37ajyccnmrjo4ze3cvigljbw.py
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x_4 => _low_memory_max_pool2d_with_offsets, getitem_1
# Graph fragment:
# %_low_memory_max_pool2d_with_offsets : [num_users=2] = call_function[target=torch.ops.prims._low_memory_max_pool2d_with_offsets.default](args = (%relu_1, [2, 2], [2, 2], [0, 0], [1, 1], False), kwargs = {})
# %getitem_1 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_2 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i8', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_max_pool2d_with_indices_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_2(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 230400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 30
x1 = (xindex // 30) % 30
x2 = (xindex // 900)
x3 = xindex
tmp0 = tl.load(in_ptr0 + ((2*x0) + (120*x1) + (3616*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (120*x1) + (3616*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (60 + (2*x0) + (120*x1) + (3616*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (61 + (2*x0) + (120*x1) + (3616*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 > tmp0
tmp3 = tl.full([1], 1, tl.int8)
tmp4 = tl.full([1], 0, tl.int8)
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = triton_helpers.maximum(tmp1, tmp0)
tmp8 = tmp7 > tmp6
tmp9 = tl.full([1], 2, tl.int8)
tmp10 = tl.where(tmp8, tmp9, tmp5)
tmp11 = triton_helpers.maximum(tmp7, tmp6)
tmp13 = tmp12 > tmp11
tmp14 = tl.full([1], 3, tl.int8)
tmp15 = tl.where(tmp13, tmp14, tmp10)
tmp16 = triton_helpers.maximum(tmp12, tmp11)
tl.store(out_ptr0 + (x3), tmp15, xmask)
tl.store(out_ptr1 + (x3), tmp16, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (32, 1, 3, 3), (9, 9, 3, 1))
assert_size_stride(primals_2, (32, ), (1, ))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
assert_size_stride(primals_4, (64, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_5, (64, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 32, 62, 62), (123008, 3844, 62, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 492032, grid=grid(492032), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 64, 60, 60), (230400, 3600, 60, 1))
buf3 = empty_strided_cuda((4, 64, 60, 60), (231424, 3616, 60, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_1.run(buf2, primals_5, buf3, 921600, grid=grid(921600), stream=stream0)
del buf2
del primals_5
buf4 = empty_strided_cuda((4, 64, 30, 30), (57600, 900, 30, 1), torch.int8)
buf5 = empty_strided_cuda((4, 64, 30, 30), (57600, 900, 30, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_2.run(buf3, buf4, buf5, 230400, grid=grid(230400), stream=stream0)
return (reinterpret_tensor(buf5, (4, 57600), (57600, 1), 0), primals_1, primals_3, primals_4, buf1, buf3, buf4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((32, 1, 3, 3), (9, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 1, 64, 64), (4096, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 32, 3, 3), (288, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.utils.data.distributed
import torch.nn.functional as F
class Net1(nn.Module):
def __init__(self):
super(Net1, self).__init__()
self.conv1 = nn.Conv2d(1, 32, 3, 1)
self.conv2 = nn.Conv2d(32, 64, 3, 1)
def forward(self, x):
x = self.conv1(x)
x = F.relu(x)
x = self.conv2(x)
x = F.relu(x)
x = F.max_pool2d(x, 2)
x = torch.flatten(x, 1)
return x
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 492032
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 3844 % 32
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 3600 % 64
x0 = xindex % 3600
x4 = xindex // 3600
tmp0 = tl.load(in_ptr0 + x3, None)
tmp1 = tl.load(in_ptr1 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(out_ptr0 + (x0 + 3616 * x4), tmp4, None)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_2(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 230400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 30
x1 = xindex // 30 % 30
x2 = xindex // 900
x3 = xindex
tmp0 = tl.load(in_ptr0 + (2 * x0 + 120 * x1 + 3616 * x2), xmask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 120 * x1 + 3616 * x2), xmask,
eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (60 + 2 * x0 + 120 * x1 + 3616 * x2), xmask,
eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (61 + 2 * x0 + 120 * x1 + 3616 * x2), xmask,
eviction_policy='evict_last')
tmp2 = tmp1 > tmp0
tmp3 = tl.full([1], 1, tl.int8)
tmp4 = tl.full([1], 0, tl.int8)
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = triton_helpers.maximum(tmp1, tmp0)
tmp8 = tmp7 > tmp6
tmp9 = tl.full([1], 2, tl.int8)
tmp10 = tl.where(tmp8, tmp9, tmp5)
tmp11 = triton_helpers.maximum(tmp7, tmp6)
tmp13 = tmp12 > tmp11
tmp14 = tl.full([1], 3, tl.int8)
tmp15 = tl.where(tmp13, tmp14, tmp10)
tmp16 = triton_helpers.maximum(tmp12, tmp11)
tl.store(out_ptr0 + x3, tmp15, xmask)
tl.store(out_ptr1 + x3, tmp16, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (32, 1, 3, 3), (9, 9, 3, 1))
assert_size_stride(primals_2, (32,), (1,))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
assert_size_stride(primals_4, (64, 32, 3, 3), (288, 9, 3, 1))
assert_size_stride(primals_5, (64,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 32, 62, 62), (123008, 3844, 62, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(492032)](buf1, primals_2,
492032, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 64, 60, 60), (230400, 3600, 60, 1))
buf3 = empty_strided_cuda((4, 64, 60, 60), (231424, 3616, 60, 1),
torch.float32)
triton_poi_fused_convolution_relu_1[grid(921600)](buf2, primals_5,
buf3, 921600, XBLOCK=1024, num_warps=4, num_stages=1)
del buf2
del primals_5
buf4 = empty_strided_cuda((4, 64, 30, 30), (57600, 900, 30, 1),
torch.int8)
buf5 = empty_strided_cuda((4, 64, 30, 30), (57600, 900, 30, 1),
torch.float32)
triton_poi_fused_max_pool2d_with_indices_2[grid(230400)](buf3, buf4,
buf5, 230400, XBLOCK=512, num_warps=8, num_stages=1)
return reinterpret_tensor(buf5, (4, 57600), (57600, 1), 0
), primals_1, primals_3, primals_4, buf1, buf3, buf4
class Net1New(nn.Module):
def __init__(self):
super(Net1New, self).__init__()
self.conv1 = nn.Conv2d(1, 32, 3, 1)
self.conv2 = nn.Conv2d(32, 64, 3, 1)
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| sermolin/amazon-sagemaker-examples | Net1 | false | 4,292 | [
"Apache-2.0"
] | 0 | 3e6083d1b53cb718893a04c46513a9482a17bd6b | https://github.com/sermolin/amazon-sagemaker-examples/tree/3e6083d1b53cb718893a04c46513a9482a17bd6b | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import torch.utils.data.distributed
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 32, 3, 1)
self.conv2 = nn.Conv2d(32, 64, 3, 1)
def forward(self, x):
x = self.conv1(x)
x = F.relu(x)
x = self.conv2(x)
x = F.relu(x)
x = F.max_pool2d(x, 2)
x = torch.flatten(x, 1)
return x
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return []
|
DemodulatedConv2d | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/75/c753nmnallr7352v6kci326ipcp72djogbwitjnsojsxpzcw7rbv.py
# Topologically Sorted Source Nodes: [pow_1, sum_1, add, demod], Original ATen: [aten.pow, aten.sum, aten.add, aten.rsqrt]
# Source node to ATen node mapping:
# add => add
# demod => rsqrt
# pow_1 => pow_1
# sum_1 => sum_1
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%primals_2, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [2, 3, 4]), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_1, 1e-08), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_per_fused_add_pow_rsqrt_sum_0 = async_compile.triton('triton_per_fused_add_pow_rsqrt_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_pow_rsqrt_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_pow_rsqrt_sum_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 36
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (36*x0)), rmask & xmask, other=0.0)
tmp1 = tmp0 * tmp0
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.where(rmask & xmask, tmp2, 0)
tmp5 = tl.sum(tmp4, 1)[:, None]
tmp6 = 1e-08
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/qb/cqbmllyzpe2nhtr4xblvy4f2byhtbttctscjpfu4ltq23mz3g6al.py
# Topologically Sorted Source Nodes: [repeat], Original ATen: [aten.repeat]
# Source node to ATen node mapping:
# repeat => repeat
# Graph fragment:
# %repeat : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%rsqrt, [4, 1]), kwargs = {})
triton_poi_fused_repeat_1 = async_compile.triton('triton_poi_fused_repeat_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_repeat_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_repeat_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/o2/co2fztvn5bb5kaf22skkom7wt43kzeptwatzi2mwgb65eoysciue.py
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# weight => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %view), kwargs = {})
triton_poi_fused_mul_2 = async_compile.triton('triton_poi_fused_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex % 144
x4 = (xindex // 36)
x5 = xindex
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x4), xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x5), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4, 3, 3), (144, 36, 9, 3, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((1, 4), (4, 1), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [pow_1, sum_1, add, demod], Original ATen: [aten.pow, aten.sum, aten.add, aten.rsqrt]
stream0 = get_raw_stream(0)
triton_per_fused_add_pow_rsqrt_sum_0.run(buf1, primals_2, 4, 36, grid=grid(4), stream=stream0)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [repeat], Original ATen: [aten.repeat]
triton_poi_fused_repeat_1.run(buf1, buf2, 16, grid=grid(16), stream=stream0)
buf3 = empty_strided_cuda((4, 4, 4, 3, 3), (144, 36, 9, 3, 1), torch.float32)
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(primals_2, buf2, buf3, 576, grid=grid(576), stream=stream0)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf3, (16, 4, 3, 3), (36, 9, 3, 1), 0), stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=4, bias=None)
assert_size_stride(buf4, (1, 16, 2, 2), (64, 4, 2, 1))
return (reinterpret_tensor(buf4, (4, 4, 2, 2), (16, 4, 2, 1), 0), primals_2, buf1, buf2, reinterpret_tensor(buf3, (16, 4, 3, 3), (36, 9, 3, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4, 4, 3, 3), (144, 36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.utils.data
import torch
from torchvision.transforms import functional as F
import torch.nn as nn
from torch.nn import functional as F
class DemodulatedConv2d(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size=3, stride=1,
padding=0, bias=False, dilation=1):
super().__init__()
self.eps = 1e-08
self.kernel_size = kernel_size
self.in_channel = in_channel
self.out_channel = out_channel
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
kernel_size, kernel_size))
self.bias = None
if bias:
self.bias = nn.Parameter(torch.randn(out_channel))
self.stride = stride
self.padding = padding
self.dilation = dilation
def forward(self, input):
batch, in_channel, height, width = input.shape
demod = torch.rsqrt(self.weight.pow(2).sum([2, 3, 4]) + 1e-08)
weight = self.weight * demod.repeat([batch, 1]).view(batch, self.
out_channel, 1, 1, 1)
weight = weight.view(batch * self.out_channel, in_channel, self.
kernel_size, self.kernel_size)
input = input.view(1, batch * in_channel, height, width)
if self.bias is None:
out = F.conv2d(input, weight, padding=self.padding, groups=
batch, dilation=self.dilation, stride=self.stride)
else:
out = F.conv2d(input, weight, bias=self.bias, padding=self.
padding, groups=batch, dilation=self.dilation, stride=self.
stride)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'out_channel': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.utils.data
import torch
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_add_pow_rsqrt_sum_0(in_out_ptr0, in_ptr0, xnumel,
rnumel, XBLOCK: tl.constexpr):
xnumel = 4
rnumel = 36
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 36 * x0), rmask & xmask, other=0.0)
tmp1 = tmp0 * tmp0
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.where(rmask & xmask, tmp2, 0)
tmp5 = tl.sum(tmp4, 1)[:, None]
tmp6 = 1e-08
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp8, xmask)
@triton.jit
def triton_poi_fused_repeat_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex % 144
x4 = xindex // 36
x5 = xindex
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x4, xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x5, tmp2, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4, 3, 3), (144, 36, 9, 3, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((1, 4), (4, 1), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_pow_rsqrt_sum_0[grid(4)](buf1, primals_2, 4,
36, XBLOCK=1, num_warps=2, num_stages=1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_repeat_1[grid(16)](buf1, buf2, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((4, 4, 4, 3, 3), (144, 36, 9, 3, 1),
torch.float32)
triton_poi_fused_mul_2[grid(576)](primals_2, buf2, buf3, 576,
XBLOCK=256, num_warps=4, num_stages=1)
buf4 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1,
16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf3, (16, 4,
3, 3), (36, 9, 3, 1), 0), stride=(1, 1), padding=(0, 0),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=4, bias=None)
assert_size_stride(buf4, (1, 16, 2, 2), (64, 4, 2, 1))
return reinterpret_tensor(buf4, (4, 4, 2, 2), (16, 4, 2, 1), 0
), primals_2, buf1, buf2, reinterpret_tensor(buf3, (16, 4, 3, 3), (
36, 9, 3, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4, 4), (256,
16, 4, 1), 0)
class DemodulatedConv2dNew(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size=3, stride=1,
padding=0, bias=False, dilation=1):
super().__init__()
self.eps = 1e-08
self.kernel_size = kernel_size
self.in_channel = in_channel
self.out_channel = out_channel
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
kernel_size, kernel_size))
self.bias = None
if bias:
self.bias = nn.Parameter(torch.randn(out_channel))
self.stride = stride
self.padding = padding
self.dilation = dilation
def forward(self, input_0):
primals_2 = self.weight
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
| seawee1/ForkGAN-pytorch | DemodulatedConv2d | false | 4,293 | [
"BSD-3-Clause"
] | 0 | 02d721875d47e4a1e96a14cc4770edcb6b68a5d0 | https://github.com/seawee1/ForkGAN-pytorch/tree/02d721875d47e4a1e96a14cc4770edcb6b68a5d0 | import torch
import torch.utils.data
import torch
from torchvision.transforms import functional as F
import torch.nn as nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size=3, stride=1,
padding=0, bias=False, dilation=1):
super().__init__()
self.eps = 1e-08
self.kernel_size = kernel_size
self.in_channel = in_channel
self.out_channel = out_channel
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
kernel_size, kernel_size))
self.bias = None
if bias:
self.bias = nn.Parameter(torch.randn(out_channel))
self.stride = stride
self.padding = padding
self.dilation = dilation
def forward(self, input):
batch, in_channel, height, width = input.shape
demod = torch.rsqrt(self.weight.pow(2).sum([2, 3, 4]) + 1e-08)
weight = self.weight * demod.repeat([batch, 1]).view(batch, self.
out_channel, 1, 1, 1)
weight = weight.view(batch * self.out_channel, in_channel, self.
kernel_size, self.kernel_size)
input = input.view(1, batch * in_channel, height, width)
if self.bias is None:
out = F.conv2d(input, weight, padding=self.padding, groups=
batch, dilation=self.dilation, stride=self.stride)
else:
out = F.conv2d(input, weight, bias=self.bias, padding=self.
padding, groups=batch, dilation=self.dilation, stride=self.
stride)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ATLoss | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/ft/cftz5ja4wgrkcx5pv7d7j6lunvnnmz6djadit33a2463r447sayy.py
# Topologically Sorted Source Nodes: [setitem_1], Original ATen: [aten.lift_fresh, aten.fill]
# Source node to ATen node mapping:
# setitem_1 => copy_1, full_default_2
# Graph fragment:
# %full_default_2 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_1 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_2, %full_default_2), kwargs = {})
# %copy__default : [num_users=0] = call_function[target=torch.ops.aten.copy_.default](args = (%select_int, %copy_1), kwargs = {})
triton_poi_fused_fill_lift_fresh_0 = async_compile.triton('triton_poi_fused_fill_lift_fresh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_fill_lift_fresh_0', 'mutated_arg_names': ['out_ptr0'], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_fill_lift_fresh_0(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
tmp0 = 0.0
tl.store(out_ptr0 + (x0 + (64*x1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/cn/ccnjm2dqpzcrnbuww2omqd3latcfln56427nk6wxia5xlnc6b5c5.py
# Topologically Sorted Source Nodes: [th_label, setitem, p_mask, sub_1, mul, logit1, log_softmax, n_mask, sub_3, mul_2, logit2, log_softmax_1], Original ATen: [aten.zeros_like, aten.lift_fresh, aten.fill, aten.add, aten.rsub, aten.mul, aten.sub, aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => amax, exp, sub_3, sum_1
# log_softmax_1 => amax_1, exp_1, sub_7, sum_3
# logit1 => sub_2
# logit2 => sub_6
# mul => mul
# mul_2 => mul_2
# n_mask => sub
# p_mask => add
# setitem => copy, full_default_1
# sub_1 => sub_1
# sub_3 => sub_5
# th_label => full_default
# Graph fragment:
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 1.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select, %full_default_1), kwargs = {})
# %select_scatter_default : [num_users=2] = call_function[target=torch.ops.aten.select_scatter.default](args = (%full_default, %copy, 1, 0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, %select_scatter_default), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %add), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, 1e+30), kwargs = {})
# %sub_2 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %mul), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%sub_2, [-1], True), kwargs = {})
# %sub_3 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_2, %amax), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_3,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg0_1), kwargs = {})
# %sub_5 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sub), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_5, 1e+30), kwargs = {})
# %sub_6 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %mul_2), kwargs = {})
# %amax_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%sub_6, [-1], True), kwargs = {})
# %sub_7 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_6, %amax_1), kwargs = {})
# %exp_1 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_7,), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_1, [-1], True), kwargs = {})
triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1 = async_compile.triton('triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, out_ptr3, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 4) % 4
tmp0 = tl.load(in_ptr0 + (4*x3), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x3), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (1 + (4*x3)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr1 + (1 + (4*x3)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr0 + (2 + (4*x3)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr1 + (2 + (4*x3)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (3 + (4*x3)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (3 + (4*x3)), xmask, eviction_policy='evict_last')
tmp2 = x1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp5 = 1.0
tmp6 = 0.0
tmp7 = tl.where(tmp4, tmp5, tmp6)
tmp8 = tmp1 + tmp7
tmp9 = tmp5 - tmp8
tmp10 = 1e+30
tmp11 = tmp9 * tmp10
tmp12 = tmp0 - tmp11
tmp15 = tmp14 + tmp7
tmp16 = tmp5 - tmp15
tmp17 = tmp16 * tmp10
tmp18 = tmp13 - tmp17
tmp19 = triton_helpers.maximum(tmp12, tmp18)
tmp22 = tmp21 + tmp7
tmp23 = tmp5 - tmp22
tmp24 = tmp23 * tmp10
tmp25 = tmp20 - tmp24
tmp26 = triton_helpers.maximum(tmp19, tmp25)
tmp29 = tmp28 + tmp7
tmp30 = tmp5 - tmp29
tmp31 = tmp30 * tmp10
tmp32 = tmp27 - tmp31
tmp33 = triton_helpers.maximum(tmp26, tmp32)
tmp34 = tmp12 - tmp33
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp18 - tmp33
tmp37 = tl_math.exp(tmp36)
tmp38 = tmp35 + tmp37
tmp39 = tmp25 - tmp33
tmp40 = tl_math.exp(tmp39)
tmp41 = tmp38 + tmp40
tmp42 = tmp32 - tmp33
tmp43 = tl_math.exp(tmp42)
tmp44 = tmp41 + tmp43
tmp45 = tmp5 - tmp1
tmp46 = tmp5 - tmp45
tmp47 = tmp46 * tmp10
tmp48 = tmp0 - tmp47
tmp49 = tmp5 - tmp14
tmp50 = tmp5 - tmp49
tmp51 = tmp50 * tmp10
tmp52 = tmp13 - tmp51
tmp53 = triton_helpers.maximum(tmp48, tmp52)
tmp54 = tmp5 - tmp21
tmp55 = tmp5 - tmp54
tmp56 = tmp55 * tmp10
tmp57 = tmp20 - tmp56
tmp58 = triton_helpers.maximum(tmp53, tmp57)
tmp59 = tmp5 - tmp28
tmp60 = tmp5 - tmp59
tmp61 = tmp60 * tmp10
tmp62 = tmp27 - tmp61
tmp63 = triton_helpers.maximum(tmp58, tmp62)
tmp64 = tmp48 - tmp63
tmp65 = tl_math.exp(tmp64)
tmp66 = tmp52 - tmp63
tmp67 = tl_math.exp(tmp66)
tmp68 = tmp65 + tmp67
tmp69 = tmp57 - tmp63
tmp70 = tl_math.exp(tmp69)
tmp71 = tmp68 + tmp70
tmp72 = tmp62 - tmp63
tmp73 = tl_math.exp(tmp72)
tmp74 = tmp71 + tmp73
tl.store(out_ptr0 + (x3), tmp33, xmask)
tl.store(out_ptr1 + (x3), tmp44, xmask)
tl.store(out_ptr2 + (x3), tmp63, xmask)
tl.store(out_ptr3 + (x3), tmp74, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/uj/cuj4drsaoeeafppqq56jc2dqbimn76sbovf5a42dyfiev4rt56n7.py
# Topologically Sorted Source Nodes: [th_label, setitem, p_mask, sub_1, mul, logit1, log_softmax, mul_1, sum_1, loss1, n_mask, sub_3, mul_2, logit2, log_softmax_1, mul_3, sum_2, loss2, loss, loss_1], Original ATen: [aten.zeros_like, aten.lift_fresh, aten.fill, aten.add, aten.rsub, aten.mul, aten.sub, aten._log_softmax, aten.sum, aten.neg, aten.mean]
# Source node to ATen node mapping:
# log_softmax => log, sub_3, sub_4
# log_softmax_1 => amax_1, log_1, sub_7, sub_8
# logit1 => sub_2
# logit2 => sub_6
# loss => add_1
# loss1 => neg
# loss2 => neg_1
# loss_1 => mean
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# mul_3 => mul_3
# n_mask => sub
# p_mask => add
# setitem => copy, full_default_1
# sub_1 => sub_1
# sub_3 => sub_5
# sum_1 => sum_2
# sum_2 => sum_4
# th_label => full_default
# Graph fragment:
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 1.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select, %full_default_1), kwargs = {})
# %select_scatter_default : [num_users=2] = call_function[target=torch.ops.aten.select_scatter.default](args = (%full_default, %copy, 1, 0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, %select_scatter_default), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %add), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, 1e+30), kwargs = {})
# %sub_2 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %mul), kwargs = {})
# %sub_3 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_2, %amax), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %sub_4 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_3, %log), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_4, %arg0_1), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [1]), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%sum_2,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg0_1), kwargs = {})
# %sub_5 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sub), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_5, 1e+30), kwargs = {})
# %sub_6 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %mul_2), kwargs = {})
# %amax_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%sub_6, [-1], True), kwargs = {})
# %sub_7 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_6, %amax_1), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_3,), kwargs = {})
# %sub_8 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_7, %log_1), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_8, %select_scatter_default), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_3, [1]), kwargs = {})
# %neg_1 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%sum_4,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg, %neg_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%add_1,), kwargs = {})
triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2 = async_compile.triton('triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32', 8: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {7: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 8), equal_to_1=(7,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 24, 'num_reduction': 1, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 64
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = (rindex // 16)
r4 = rindex % 16
r1 = (rindex // 4) % 4
r3 = rindex
tmp0 = tl.load(in_ptr0 + (r4 + (64*r2)), None)
tmp1 = tl.load(in_ptr1 + (r4 + (64*r2)), None)
tmp12 = tl.load(in_ptr2 + (r1 + (16*r2)), None, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr3 + (r1 + (16*r2)), None, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (16 + r4 + (64*r2)), None)
tmp19 = tl.load(in_ptr1 + (16 + r4 + (64*r2)), None)
tmp27 = tl.load(in_ptr2 + (4 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr3 + (4 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp34 = tl.load(in_ptr0 + (32 + r4 + (64*r2)), None)
tmp35 = tl.load(in_ptr1 + (32 + r4 + (64*r2)), None)
tmp43 = tl.load(in_ptr2 + (8 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp45 = tl.load(in_ptr3 + (8 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp50 = tl.load(in_ptr0 + (48 + r4 + (64*r2)), None)
tmp51 = tl.load(in_ptr1 + (48 + r4 + (64*r2)), None)
tmp59 = tl.load(in_ptr2 + (12 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp61 = tl.load(in_ptr3 + (12 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp70 = tl.load(in_ptr4 + (r1 + (16*r2)), None, eviction_policy='evict_last')
tmp72 = tl.load(in_ptr5 + (r1 + (16*r2)), None, eviction_policy='evict_last')
tmp80 = tl.load(in_ptr4 + (4 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp82 = tl.load(in_ptr5 + (4 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp91 = tl.load(in_ptr4 + (8 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp93 = tl.load(in_ptr5 + (8 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp102 = tl.load(in_ptr4 + (12 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp104 = tl.load(in_ptr5 + (12 + r1 + (16*r2)), None, eviction_policy='evict_last')
tmp2 = tl.full([1, 1], 0, tl.int32)
tmp3 = tmp2 == tmp2
tmp4 = 1.0
tmp5 = 0.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp1 + tmp6
tmp8 = tmp4 - tmp7
tmp9 = 1e+30
tmp10 = tmp8 * tmp9
tmp11 = tmp0 - tmp10
tmp13 = tmp11 - tmp12
tmp15 = tl_math.log(tmp14)
tmp16 = tmp13 - tmp15
tmp17 = tmp16 * tmp1
tmp20 = tl.full([1, 1], 1, tl.int32)
tmp21 = tmp20 == tmp2
tmp22 = tl.where(tmp21, tmp4, tmp5)
tmp23 = tmp19 + tmp22
tmp24 = tmp4 - tmp23
tmp25 = tmp24 * tmp9
tmp26 = tmp18 - tmp25
tmp28 = tmp26 - tmp27
tmp30 = tl_math.log(tmp29)
tmp31 = tmp28 - tmp30
tmp32 = tmp31 * tmp19
tmp33 = tmp17 + tmp32
tmp36 = tl.full([1, 1], 2, tl.int32)
tmp37 = tmp36 == tmp2
tmp38 = tl.where(tmp37, tmp4, tmp5)
tmp39 = tmp35 + tmp38
tmp40 = tmp4 - tmp39
tmp41 = tmp40 * tmp9
tmp42 = tmp34 - tmp41
tmp44 = tmp42 - tmp43
tmp46 = tl_math.log(tmp45)
tmp47 = tmp44 - tmp46
tmp48 = tmp47 * tmp35
tmp49 = tmp33 + tmp48
tmp52 = tl.full([1, 1], 3, tl.int32)
tmp53 = tmp52 == tmp2
tmp54 = tl.where(tmp53, tmp4, tmp5)
tmp55 = tmp51 + tmp54
tmp56 = tmp4 - tmp55
tmp57 = tmp56 * tmp9
tmp58 = tmp50 - tmp57
tmp60 = tmp58 - tmp59
tmp62 = tl_math.log(tmp61)
tmp63 = tmp60 - tmp62
tmp64 = tmp63 * tmp51
tmp65 = tmp49 + tmp64
tmp66 = tmp4 - tmp1
tmp67 = tmp4 - tmp66
tmp68 = tmp67 * tmp9
tmp69 = tmp0 - tmp68
tmp71 = tmp69 - tmp70
tmp73 = tl_math.log(tmp72)
tmp74 = tmp71 - tmp73
tmp75 = tmp74 * tmp6
tmp76 = tmp4 - tmp19
tmp77 = tmp4 - tmp76
tmp78 = tmp77 * tmp9
tmp79 = tmp18 - tmp78
tmp81 = tmp79 - tmp80
tmp83 = tl_math.log(tmp82)
tmp84 = tmp81 - tmp83
tmp85 = tmp84 * tmp22
tmp86 = tmp75 + tmp85
tmp87 = tmp4 - tmp35
tmp88 = tmp4 - tmp87
tmp89 = tmp88 * tmp9
tmp90 = tmp34 - tmp89
tmp92 = tmp90 - tmp91
tmp94 = tl_math.log(tmp93)
tmp95 = tmp92 - tmp94
tmp96 = tmp95 * tmp38
tmp97 = tmp86 + tmp96
tmp98 = tmp4 - tmp51
tmp99 = tmp4 - tmp98
tmp100 = tmp99 * tmp9
tmp101 = tmp50 - tmp100
tmp103 = tmp101 - tmp102
tmp105 = tl_math.log(tmp104)
tmp106 = tmp103 - tmp105
tmp107 = tmp106 * tmp54
tmp108 = tmp97 + tmp107
tmp109 = -tmp65
tmp110 = -tmp108
tmp111 = tmp109 + tmp110
tmp112 = tl.broadcast_to(tmp111, [XBLOCK, RBLOCK])
tmp114 = tl.sum(tmp112, 1)[:, None]
tmp115 = 64.0
tmp116 = tmp114 / tmp115
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp116, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [setitem_1], Original ATen: [aten.lift_fresh, aten.fill]
stream0 = get_raw_stream(0)
triton_poi_fused_fill_lift_fresh_0.run(arg0_1, 64, grid=grid(64), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf4 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf5 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [th_label, setitem, p_mask, sub_1, mul, logit1, log_softmax, n_mask, sub_3, mul_2, logit2, log_softmax_1], Original ATen: [aten.zeros_like, aten.lift_fresh, aten.fill, aten.add, aten.rsub, aten.mul, aten.sub, aten._log_softmax]
triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1.run(arg1_1, arg0_1, buf1, buf2, buf4, buf5, 64, grid=grid(64), stream=stream0)
buf7 = empty_strided_cuda((), (), torch.float32)
buf8 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [th_label, setitem, p_mask, sub_1, mul, logit1, log_softmax, mul_1, sum_1, loss1, n_mask, sub_3, mul_2, logit2, log_softmax_1, mul_3, sum_2, loss2, loss, loss_1], Original ATen: [aten.zeros_like, aten.lift_fresh, aten.fill, aten.add, aten.rsub, aten.mul, aten.sub, aten._log_softmax, aten.sum, aten.neg, aten.mean]
triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2.run(buf8, arg1_1, arg0_1, buf1, buf2, buf4, buf5, 1, 64, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
del buf1
del buf2
del buf4
del buf5
return (buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class ATLoss(nn.Module):
def __init__(self):
super().__init__()
def forward(self, logits, labels):
th_label = torch.zeros_like(labels, dtype=torch.float)
th_label[:, 0] = 1.0
labels[:, 0] = 0.0
p_mask = labels + th_label
n_mask = 1 - labels
logit1 = logits - (1 - p_mask) * 1e+30
loss1 = -(F.log_softmax(logit1, dim=-1) * labels).sum(1)
logit2 = logits - (1 - n_mask) * 1e+30
loss2 = -(F.log_softmax(logit2, dim=-1) * th_label).sum(1)
loss = loss1 + loss2
loss = loss.mean()
return loss
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_fill_lift_fresh_0(out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
tmp0 = 0.0
tl.store(out_ptr0 + (x0 + 64 * x1), tmp0, xmask)
@triton.jit
def triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1(
in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, out_ptr3, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 4 % 4
tmp0 = tl.load(in_ptr0 + 4 * x3, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x3, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (1 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp14 = tl.load(in_ptr1 + (1 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp20 = tl.load(in_ptr0 + (2 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp21 = tl.load(in_ptr1 + (2 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp27 = tl.load(in_ptr0 + (3 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp28 = tl.load(in_ptr1 + (3 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp2 = x1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp5 = 1.0
tmp6 = 0.0
tmp7 = tl.where(tmp4, tmp5, tmp6)
tmp8 = tmp1 + tmp7
tmp9 = tmp5 - tmp8
tmp10 = 1e+30
tmp11 = tmp9 * tmp10
tmp12 = tmp0 - tmp11
tmp15 = tmp14 + tmp7
tmp16 = tmp5 - tmp15
tmp17 = tmp16 * tmp10
tmp18 = tmp13 - tmp17
tmp19 = triton_helpers.maximum(tmp12, tmp18)
tmp22 = tmp21 + tmp7
tmp23 = tmp5 - tmp22
tmp24 = tmp23 * tmp10
tmp25 = tmp20 - tmp24
tmp26 = triton_helpers.maximum(tmp19, tmp25)
tmp29 = tmp28 + tmp7
tmp30 = tmp5 - tmp29
tmp31 = tmp30 * tmp10
tmp32 = tmp27 - tmp31
tmp33 = triton_helpers.maximum(tmp26, tmp32)
tmp34 = tmp12 - tmp33
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp18 - tmp33
tmp37 = tl_math.exp(tmp36)
tmp38 = tmp35 + tmp37
tmp39 = tmp25 - tmp33
tmp40 = tl_math.exp(tmp39)
tmp41 = tmp38 + tmp40
tmp42 = tmp32 - tmp33
tmp43 = tl_math.exp(tmp42)
tmp44 = tmp41 + tmp43
tmp45 = tmp5 - tmp1
tmp46 = tmp5 - tmp45
tmp47 = tmp46 * tmp10
tmp48 = tmp0 - tmp47
tmp49 = tmp5 - tmp14
tmp50 = tmp5 - tmp49
tmp51 = tmp50 * tmp10
tmp52 = tmp13 - tmp51
tmp53 = triton_helpers.maximum(tmp48, tmp52)
tmp54 = tmp5 - tmp21
tmp55 = tmp5 - tmp54
tmp56 = tmp55 * tmp10
tmp57 = tmp20 - tmp56
tmp58 = triton_helpers.maximum(tmp53, tmp57)
tmp59 = tmp5 - tmp28
tmp60 = tmp5 - tmp59
tmp61 = tmp60 * tmp10
tmp62 = tmp27 - tmp61
tmp63 = triton_helpers.maximum(tmp58, tmp62)
tmp64 = tmp48 - tmp63
tmp65 = tl_math.exp(tmp64)
tmp66 = tmp52 - tmp63
tmp67 = tl_math.exp(tmp66)
tmp68 = tmp65 + tmp67
tmp69 = tmp57 - tmp63
tmp70 = tl_math.exp(tmp69)
tmp71 = tmp68 + tmp70
tmp72 = tmp62 - tmp63
tmp73 = tl_math.exp(tmp72)
tmp74 = tmp71 + tmp73
tl.store(out_ptr0 + x3, tmp33, xmask)
tl.store(out_ptr1 + x3, tmp44, xmask)
tl.store(out_ptr2 + x3, tmp63, xmask)
tl.store(out_ptr3 + x3, tmp74, xmask)
@triton.jit
def triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2(
in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5,
xnumel, rnumel, XBLOCK: tl.constexpr):
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex // 16
r4 = rindex % 16
r1 = rindex // 4 % 4
tmp0 = tl.load(in_ptr0 + (r4 + 64 * r2), None)
tmp1 = tl.load(in_ptr1 + (r4 + 64 * r2), None)
tmp12 = tl.load(in_ptr2 + (r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp14 = tl.load(in_ptr3 + (r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp18 = tl.load(in_ptr0 + (16 + r4 + 64 * r2), None)
tmp19 = tl.load(in_ptr1 + (16 + r4 + 64 * r2), None)
tmp27 = tl.load(in_ptr2 + (4 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr3 + (4 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp34 = tl.load(in_ptr0 + (32 + r4 + 64 * r2), None)
tmp35 = tl.load(in_ptr1 + (32 + r4 + 64 * r2), None)
tmp43 = tl.load(in_ptr2 + (8 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp45 = tl.load(in_ptr3 + (8 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp50 = tl.load(in_ptr0 + (48 + r4 + 64 * r2), None)
tmp51 = tl.load(in_ptr1 + (48 + r4 + 64 * r2), None)
tmp59 = tl.load(in_ptr2 + (12 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp61 = tl.load(in_ptr3 + (12 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp70 = tl.load(in_ptr4 + (r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp72 = tl.load(in_ptr5 + (r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp80 = tl.load(in_ptr4 + (4 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp82 = tl.load(in_ptr5 + (4 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp91 = tl.load(in_ptr4 + (8 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp93 = tl.load(in_ptr5 + (8 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp102 = tl.load(in_ptr4 + (12 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp104 = tl.load(in_ptr5 + (12 + r1 + 16 * r2), None, eviction_policy=
'evict_last')
tmp2 = tl.full([1, 1], 0, tl.int32)
tmp3 = tmp2 == tmp2
tmp4 = 1.0
tmp5 = 0.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp1 + tmp6
tmp8 = tmp4 - tmp7
tmp9 = 1e+30
tmp10 = tmp8 * tmp9
tmp11 = tmp0 - tmp10
tmp13 = tmp11 - tmp12
tmp15 = tl_math.log(tmp14)
tmp16 = tmp13 - tmp15
tmp17 = tmp16 * tmp1
tmp20 = tl.full([1, 1], 1, tl.int32)
tmp21 = tmp20 == tmp2
tmp22 = tl.where(tmp21, tmp4, tmp5)
tmp23 = tmp19 + tmp22
tmp24 = tmp4 - tmp23
tmp25 = tmp24 * tmp9
tmp26 = tmp18 - tmp25
tmp28 = tmp26 - tmp27
tmp30 = tl_math.log(tmp29)
tmp31 = tmp28 - tmp30
tmp32 = tmp31 * tmp19
tmp33 = tmp17 + tmp32
tmp36 = tl.full([1, 1], 2, tl.int32)
tmp37 = tmp36 == tmp2
tmp38 = tl.where(tmp37, tmp4, tmp5)
tmp39 = tmp35 + tmp38
tmp40 = tmp4 - tmp39
tmp41 = tmp40 * tmp9
tmp42 = tmp34 - tmp41
tmp44 = tmp42 - tmp43
tmp46 = tl_math.log(tmp45)
tmp47 = tmp44 - tmp46
tmp48 = tmp47 * tmp35
tmp49 = tmp33 + tmp48
tmp52 = tl.full([1, 1], 3, tl.int32)
tmp53 = tmp52 == tmp2
tmp54 = tl.where(tmp53, tmp4, tmp5)
tmp55 = tmp51 + tmp54
tmp56 = tmp4 - tmp55
tmp57 = tmp56 * tmp9
tmp58 = tmp50 - tmp57
tmp60 = tmp58 - tmp59
tmp62 = tl_math.log(tmp61)
tmp63 = tmp60 - tmp62
tmp64 = tmp63 * tmp51
tmp65 = tmp49 + tmp64
tmp66 = tmp4 - tmp1
tmp67 = tmp4 - tmp66
tmp68 = tmp67 * tmp9
tmp69 = tmp0 - tmp68
tmp71 = tmp69 - tmp70
tmp73 = tl_math.log(tmp72)
tmp74 = tmp71 - tmp73
tmp75 = tmp74 * tmp6
tmp76 = tmp4 - tmp19
tmp77 = tmp4 - tmp76
tmp78 = tmp77 * tmp9
tmp79 = tmp18 - tmp78
tmp81 = tmp79 - tmp80
tmp83 = tl_math.log(tmp82)
tmp84 = tmp81 - tmp83
tmp85 = tmp84 * tmp22
tmp86 = tmp75 + tmp85
tmp87 = tmp4 - tmp35
tmp88 = tmp4 - tmp87
tmp89 = tmp88 * tmp9
tmp90 = tmp34 - tmp89
tmp92 = tmp90 - tmp91
tmp94 = tl_math.log(tmp93)
tmp95 = tmp92 - tmp94
tmp96 = tmp95 * tmp38
tmp97 = tmp86 + tmp96
tmp98 = tmp4 - tmp51
tmp99 = tmp4 - tmp98
tmp100 = tmp99 * tmp9
tmp101 = tmp50 - tmp100
tmp103 = tmp101 - tmp102
tmp105 = tl_math.log(tmp104)
tmp106 = tmp103 - tmp105
tmp107 = tmp106 * tmp54
tmp108 = tmp97 + tmp107
tmp109 = -tmp65
tmp110 = -tmp108
tmp111 = tmp109 + tmp110
tmp112 = tl.broadcast_to(tmp111, [XBLOCK, RBLOCK])
tmp114 = tl.sum(tmp112, 1)[:, None]
tmp115 = 64.0
tmp116 = tmp114 / tmp115
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp116, None)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
get_raw_stream(0)
triton_poi_fused_fill_lift_fresh_0[grid(64)](arg0_1, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf2 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf4 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf5 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused__log_softmax_add_fill_lift_fresh_mul_rsub_sub_zeros_like_1[
grid(64)](arg1_1, arg0_1, buf1, buf2, buf4, buf5, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf7 = empty_strided_cuda((), (), torch.float32)
buf8 = buf7
del buf7
triton_per_fused__log_softmax_add_fill_lift_fresh_mean_mul_neg_rsub_sub_sum_zeros_like_2[
grid(1)](buf8, arg1_1, arg0_1, buf1, buf2, buf4, buf5, 1, 64,
XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
del buf1
del buf2
del buf4
del buf5
return buf8,
class ATLossNew(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| seanswyi/R-BERT | ATLoss | false | 4,294 | [
"Apache-2.0"
] | 0 | 4a4aeab3a9314307ce4458bd2b943d94aaf4a706 | https://github.com/seanswyi/R-BERT/tree/4a4aeab3a9314307ce4458bd2b943d94aaf4a706 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, logits, labels):
th_label = torch.zeros_like(labels, dtype=torch.float)
th_label[:, 0] = 1.0
labels[:, 0] = 0.0
p_mask = labels + th_label
n_mask = 1 - labels
logit1 = logits - (1 - p_mask) * 1e+30
loss1 = -(F.log_softmax(logit1, dim=-1) * labels).sum(1)
logit2 = logits - (1 - n_mask) * 1e+30
loss2 = -(F.log_softmax(logit2, dim=-1) * th_label).sum(1)
loss = loss1 + loss2
loss = loss.mean()
return loss
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Planar | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/j6/cj6hneuwtxuqku4ka6k6f2vxucocdiv6p3b2c4kfrcb5qwbmb6sg.py
# Topologically Sorted Source Nodes: [pow_1, w_norm_sq], Original ATen: [aten.pow, aten.sum]
# Source node to ATen node mapping:
# pow_1 => pow_1
# w_norm_sq => sum_1
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_1, 2), kwargs = {})
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [2], True), kwargs = {})
triton_poi_fused_pow_sum_0 = async_compile.triton('triton_poi_fused_pow_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_pow_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_pow_sum_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp6 = tmp5 * tmp5
tmp7 = tmp4 + tmp6
tmp9 = tmp8 * tmp8
tmp10 = tmp7 + tmp9
tl.store(out_ptr0 + (x0), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/vr/cvryswlc7olf6iscoyyxog3zotceoyz6z6tn3m7r27lzurpsbsjs.py
# Topologically Sorted Source Nodes: [wzb, tanh], Original ATen: [aten.add, aten.tanh]
# Source node to ATen node mapping:
# tanh => tanh
# wzb => add_2
# Graph fragment:
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%bmm_1, %select_2), kwargs = {})
# %tanh : [num_users=3] = call_function[target=torch.ops.aten.tanh.default](args = (%add_2,), kwargs = {})
triton_poi_fused_add_tanh_1 = async_compile.triton('triton_poi_fused_add_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_tanh_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_tanh_1(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_out_ptr0 + (x0), xmask)
tmp2 = tl.load(in_ptr1 + (0))
tmp3 = tl.broadcast_to(tmp2, [XBLOCK])
tmp4 = tmp1 + tmp3
tmp5 = tmp0 + tmp4
tmp6 = libdevice.tanh(tmp5)
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/5z/c5zxdmoivium33w2qam24kojs7lyxgbfb6lulnqtriq4xzdmkzgp.py
# Topologically Sorted Source Nodes: [softplus, m_uw, sub, mul, truediv, u_hat, mul_1, z, pow_2, sub_1, psi], Original ATen: [aten.softplus, aten.add, aten.sub, aten.mul, aten.div, aten.pow, aten.rsub]
# Source node to ATen node mapping:
# m_uw => add
# mul => mul_1
# mul_1 => mul_2
# pow_2 => pow_2
# psi => mul_3
# softplus => div, exp, gt, log1p, mul, where
# sub => sub
# sub_1 => sub_1
# truediv => div_1
# u_hat => add_1
# z => add_3
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%bmm, 1.0), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%mul,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%log1p, 1.0), kwargs = {})
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%mul, 20.0), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %bmm, %div), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%where, -1.0), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %bmm), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %permute_3), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul_1, %sum_1), kwargs = {})
# %add_1 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%select, %div_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_1, %tanh), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%unsqueeze, %mul_2), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%tanh, 2), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %pow_2), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_1, %sub_1), kwargs = {})
triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2 = async_compile.triton('triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7, 8, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr2 + (x2), xmask)
tmp15 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr4 + (x2), xmask)
tmp19 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = 20.0
tmp5 = tmp3 > tmp4
tmp6 = tl_math.exp(tmp3)
tmp7 = libdevice.log1p(tmp6)
tmp8 = tmp7 * tmp2
tmp9 = tl.where(tmp5, tmp1, tmp8)
tmp10 = -1.0
tmp11 = tmp9 + tmp10
tmp12 = tmp11 - tmp1
tmp14 = tmp12 * tmp13
tmp16 = tmp14 / tmp15
tmp17 = tmp0 + tmp16
tmp20 = tmp17 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp19 * tmp19
tmp23 = tmp2 - tmp22
tmp24 = tmp13 * tmp23
tl.store(out_ptr0 + (x2), tmp17, xmask)
tl.store(out_ptr1 + (x2), tmp21, xmask)
tl.store(out_ptr2 + (x2), tmp24, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/as/casa25kbsjcsboxppwpjigmfgz4dqo5yacivs4ovqvekprlytwt6.py
# Topologically Sorted Source Nodes: [logdet_2], Original ATen: [aten.add]
# Source node to ATen node mapping:
# logdet_2 => add_5
# Graph fragment:
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%squeeze_2, 0.0), kwargs = {})
triton_poi_fused_add_3 = async_compile.triton('triton_poi_fused_add_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 1.0
tmp2 = tmp0 + tmp1
tmp3 = tl_math.abs(tmp2)
tmp4 = tl_math.log(tmp3)
tmp5 = 0.0
tmp6 = tmp4 + tmp5
tl.store(out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (1, 4), (4, 1))
assert_size_stride(primals_7, (1, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, primals_3, reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, primals_3, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_4
del primals_5
buf2 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(primals_3, reinterpret_tensor(primals_6, (4, 1), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [uw], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0), reinterpret_tensor(buf0, (4, 4, 1), (4, 1, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [pow_1, w_norm_sq], Original ATen: [aten.pow, aten.sum]
stream0 = get_raw_stream(0)
triton_poi_fused_pow_sum_0.run(buf1, buf4, 4, grid=grid(4), stream=stream0)
buf6 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [bmm_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0), reinterpret_tensor(primals_8, (4, 4, 1), (4, 1, 1), 0), out=buf6)
buf7 = reinterpret_tensor(buf2, (4, 1, 1), (1, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [wzb, tanh], Original ATen: [aten.add, aten.tanh]
triton_poi_fused_add_tanh_1.run(buf7, buf6, primals_7, 4, grid=grid(4), stream=stream0)
del primals_7
buf5 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
buf8 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
buf9 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [softplus, m_uw, sub, mul, truediv, u_hat, mul_1, z, pow_2, sub_1, psi], Original ATen: [aten.softplus, aten.add, aten.sub, aten.mul, aten.div, aten.pow, aten.rsub]
triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2.run(buf0, buf3, buf1, buf4, primals_8, buf7, buf5, buf8, buf9, 16, grid=grid(16), stream=stream0)
buf10 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [bmm_2], Original ATen: [aten.bmm]
extern_kernels.bmm(buf9, buf5, out=buf10)
buf11 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [logdet_2], Original ATen: [aten.add]
triton_poi_fused_add_3.run(buf10, buf11, 4, grid=grid(4), stream=stream0)
return (reinterpret_tensor(buf8, (4, 4), (4, 1), 0), buf11, primals_3, reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0), buf3, buf4, buf5, buf7, buf10, reinterpret_tensor(buf9, (4, 4, 1), (4, 1, 4), 0), reinterpret_tensor(primals_8, (4, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf0, (4, 1, 4), (4, 1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class PlanarStep(nn.Module):
def __init__(self):
super(PlanarStep, self).__init__()
self.h = nn.Tanh()
self.softplus = nn.Softplus()
def _der_h(self, x):
"""Derivative of activation function h."""
return self._der_tanh(x)
def _der_tanh(self, x):
"""Derivative of the Tanh function."""
return 1 - self.h(x) ** 2
def forward(self, zk, u, w, b):
"""
Forward pass. Assumes amortized u, w and b. Conditions on diagonals of u and w for invertibility
will be be satisfied inside this function. Computes the following transformation:
z' = z + u h( w^T z + b)
or actually
z'^T = z^T + h(z^T w + b)u^T
Assumes the following input shapes:
shape u = (batch_size, z_dim, 1)
shape w = (batch_size, 1, z_dim)
shape b = (batch_size, 1, 1)
shape z = (batch_size, z_dim).
"""
zk = zk.unsqueeze(2)
uw = torch.bmm(w, u)
m_uw = -1.0 + self.softplus(uw)
w_norm_sq = torch.sum(w ** 2, dim=2, keepdim=True)
u_hat = u + (m_uw - uw) * w.transpose(2, 1) / w_norm_sq
wzb = torch.bmm(w, zk) + b
z = zk + u_hat * self.h(wzb)
z = z.squeeze(2)
psi = w * self._der_h(wzb)
logdet = torch.log(torch.abs(1 + torch.bmm(psi, u_hat)))
logdet = logdet.squeeze(2).squeeze(1)
return z, logdet
class Error(Exception):
"""Base error class, from which all other errors derive."""
pass
class InvalidArgumentError(Error):
"""This error will be shown when a given argument has an invalid value."""
pass
class NormalizingFlow(nn.Module):
"""Base class for normalizing flows."""
def __init__(self, h_dim, z_dim, flow_depth, hidden_depth):
super(NormalizingFlow, self).__init__()
self.h_dim = h_dim
self.z_dim = z_dim
self.flow_depth = flow_depth
self.hidden_depth = hidden_depth
@property
def flow_depth(self):
return self._flow_depth
@flow_depth.setter
def flow_depth(self, value):
if not isinstance(value, int):
raise InvalidArgumentError('flow_depth should be an integer.')
elif value < 1:
raise InvalidArgumentError(
'flow_depth should be strictly positive.')
else:
self._flow_depth = value
@property
def hidden_depth(self):
return self._hidden_depth
@hidden_depth.setter
def hidden_depth(self, value):
if not isinstance(value, int):
raise InvalidArgumentError('hidden_depth should be an integer.')
elif value < 0:
raise InvalidArgumentError('hidden_depth should be positive.')
else:
self._hidden_depth = value
class Planar(NormalizingFlow):
"""Planar Normalizing flow with single unit bottleneck."""
def __init__(self, h_dim, z_dim, flow_depth):
super(Planar, self).__init__(h_dim, z_dim, flow_depth, 0)
self.flow = PlanarStep()
self.h_to_u = nn.Linear(self.h_dim, self.flow_depth * self.z_dim)
self.h_to_w = nn.Linear(self.h_dim, self.flow_depth * self.z_dim)
self.h_to_b = nn.Linear(self.h_dim, self.flow_depth)
def forward(self, z, h):
u = self.h_to_u(h).view(-1, self.flow_depth, self.z_dim, 1)
w = self.h_to_w(h).view(-1, self.flow_depth, 1, self.z_dim)
b = self.h_to_b(h).view(-1, self.flow_depth, 1, 1)
z_k = z
logdet = 0.0
for k in range(self.flow_depth):
z_k, ldj = self.flow(z_k, u[:, k, :, :], w[:, k, :, :], b[:, k,
:, :])
logdet += ldj
return z_k, logdet
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'h_dim': 4, 'z_dim': 4, 'flow_depth': 1}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_pow_sum_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp6 = tmp5 * tmp5
tmp7 = tmp4 + tmp6
tmp9 = tmp8 * tmp8
tmp10 = tmp7 + tmp9
tl.store(out_ptr0 + x0, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_tanh_1(in_out_ptr0, in_ptr0, in_ptr1, xnumel,
XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_out_ptr0 + x0, xmask)
tmp2 = tl.load(in_ptr1 + 0)
tmp3 = tl.broadcast_to(tmp2, [XBLOCK])
tmp4 = tmp1 + tmp3
tmp5 = tmp0 + tmp4
tmp6 = libdevice.tanh(tmp5)
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, out_ptr1, out_ptr2,
xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr2 + x2, xmask)
tmp15 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr4 + x2, xmask)
tmp19 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = 20.0
tmp5 = tmp3 > tmp4
tmp6 = tl_math.exp(tmp3)
tmp7 = libdevice.log1p(tmp6)
tmp8 = tmp7 * tmp2
tmp9 = tl.where(tmp5, tmp1, tmp8)
tmp10 = -1.0
tmp11 = tmp9 + tmp10
tmp12 = tmp11 - tmp1
tmp14 = tmp12 * tmp13
tmp16 = tmp14 / tmp15
tmp17 = tmp0 + tmp16
tmp20 = tmp17 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp19 * tmp19
tmp23 = tmp2 - tmp22
tmp24 = tmp13 * tmp23
tl.store(out_ptr0 + x2, tmp17, xmask)
tl.store(out_ptr1 + x2, tmp21, xmask)
tl.store(out_ptr2 + x2, tmp24, xmask)
@triton.jit
def triton_poi_fused_add_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 1.0
tmp2 = tmp0 + tmp1
tmp3 = tl_math.abs(tmp2)
tmp4 = tl_math.log(tmp3)
tmp5 = 0.0
tmp6 = tmp4 + tmp5
tl.store(out_ptr0 + x0, tmp6, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (1, 4), (4, 1))
assert_size_stride(primals_7, (1,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, primals_3, reinterpret_tensor(
primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, primals_3, reinterpret_tensor(
primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_4
del primals_5
buf2 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(primals_3, reinterpret_tensor(primals_6, (4, 1),
(1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0
), reinterpret_tensor(buf0, (4, 4, 1), (4, 1, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_pow_sum_0[grid(4)](buf1, buf4, 4, XBLOCK=4,
num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0
), reinterpret_tensor(primals_8, (4, 4, 1), (4, 1, 1), 0), out=buf6
)
buf7 = reinterpret_tensor(buf2, (4, 1, 1), (1, 1, 1), 0)
del buf2
triton_poi_fused_add_tanh_1[grid(4)](buf7, buf6, primals_7, 4,
XBLOCK=4, num_warps=1, num_stages=1)
del primals_7
buf5 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
buf8 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
buf9 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
triton_poi_fused_add_div_mul_pow_rsub_softplus_sub_2[grid(16)](buf0,
buf3, buf1, buf4, primals_8, buf7, buf5, buf8, buf9, 16, XBLOCK
=16, num_warps=1, num_stages=1)
buf10 = buf6
del buf6
extern_kernels.bmm(buf9, buf5, out=buf10)
buf11 = empty_strided_cuda((4,), (1,), torch.float32)
triton_poi_fused_add_3[grid(4)](buf10, buf11, 4, XBLOCK=4,
num_warps=1, num_stages=1)
return reinterpret_tensor(buf8, (4, 4), (4, 1), 0
), buf11, primals_3, reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0
), buf3, buf4, buf5, buf7, buf10, reinterpret_tensor(buf9, (4, 4, 1
), (4, 1, 4), 0), reinterpret_tensor(primals_8, (4, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf0, (4, 1, 4), (4, 1, 1), 0)
class PlanarStep(nn.Module):
def __init__(self):
super(PlanarStep, self).__init__()
self.h = nn.Tanh()
self.softplus = nn.Softplus()
def _der_h(self, x):
"""Derivative of activation function h."""
return self._der_tanh(x)
def _der_tanh(self, x):
"""Derivative of the Tanh function."""
return 1 - self.h(x) ** 2
def forward(self, zk, u, w, b):
"""
Forward pass. Assumes amortized u, w and b. Conditions on diagonals of u and w for invertibility
will be be satisfied inside this function. Computes the following transformation:
z' = z + u h( w^T z + b)
or actually
z'^T = z^T + h(z^T w + b)u^T
Assumes the following input shapes:
shape u = (batch_size, z_dim, 1)
shape w = (batch_size, 1, z_dim)
shape b = (batch_size, 1, 1)
shape z = (batch_size, z_dim).
"""
zk = zk.unsqueeze(2)
uw = torch.bmm(w, u)
m_uw = -1.0 + self.softplus(uw)
w_norm_sq = torch.sum(w ** 2, dim=2, keepdim=True)
u_hat = u + (m_uw - uw) * w.transpose(2, 1) / w_norm_sq
wzb = torch.bmm(w, zk) + b
z = zk + u_hat * self.h(wzb)
z = z.squeeze(2)
psi = w * self._der_h(wzb)
logdet = torch.log(torch.abs(1 + torch.bmm(psi, u_hat)))
logdet = logdet.squeeze(2).squeeze(1)
return z, logdet
class Error(Exception):
"""Base error class, from which all other errors derive."""
pass
class InvalidArgumentError(Error):
"""This error will be shown when a given argument has an invalid value."""
pass
class NormalizingFlow(nn.Module):
"""Base class for normalizing flows."""
def __init__(self, h_dim, z_dim, flow_depth, hidden_depth):
super(NormalizingFlow, self).__init__()
self.h_dim = h_dim
self.z_dim = z_dim
self.flow_depth = flow_depth
self.hidden_depth = hidden_depth
@property
def flow_depth(self):
return self._flow_depth
@flow_depth.setter
def flow_depth(self, value):
if not isinstance(value, int):
raise InvalidArgumentError('flow_depth should be an integer.')
elif value < 1:
raise InvalidArgumentError(
'flow_depth should be strictly positive.')
else:
self._flow_depth = value
@property
def hidden_depth(self):
return self._hidden_depth
@hidden_depth.setter
def hidden_depth(self, value):
if not isinstance(value, int):
raise InvalidArgumentError('hidden_depth should be an integer.')
elif value < 0:
raise InvalidArgumentError('hidden_depth should be positive.')
else:
self._hidden_depth = value
class PlanarNew(NormalizingFlow):
"""Planar Normalizing flow with single unit bottleneck."""
def __init__(self, h_dim, z_dim, flow_depth):
super(PlanarNew, self).__init__(h_dim, z_dim, flow_depth, 0)
self.flow = PlanarStep()
self.h_to_u = nn.Linear(self.h_dim, self.flow_depth * self.z_dim)
self.h_to_w = nn.Linear(self.h_dim, self.flow_depth * self.z_dim)
self.h_to_b = nn.Linear(self.h_dim, self.flow_depth)
def forward(self, input_0, input_1):
primals_1 = self.h_to_u.weight
primals_2 = self.h_to_u.bias
primals_3 = self.h_to_w.weight
primals_5 = self.h_to_w.bias
primals_6 = self.h_to_b.weight
primals_7 = self.h_to_b.bias
primals_4 = input_0
primals_8 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0], output[1]
| scfrank/deep-generative-lm | Planar | false | 4,295 | [
"MIT"
] | 0 | 70067fcda82aa035bba805ce6c2709097166a7a4 | https://github.com/scfrank/deep-generative-lm/tree/70067fcda82aa035bba805ce6c2709097166a7a4 | import torch
import torch.nn as nn
class PlanarStep(nn.Module):
def __init__(self):
super().__init__()
self.h = nn.Tanh()
self.softplus = nn.Softplus()
def _der_h(self, x):
"""Derivative of activation function h."""
return self._der_tanh(x)
def _der_tanh(self, x):
"""Derivative of the Tanh function."""
return 1 - self.h(x) ** 2
def forward(self, zk, u, w, b):
"""
Forward pass. Assumes amortized u, w and b. Conditions on diagonals of u and w for invertibility
will be be satisfied inside this function. Computes the following transformation:
z' = z + u h( w^T z + b)
or actually
z'^T = z^T + h(z^T w + b)u^T
Assumes the following input shapes:
shape u = (batch_size, z_dim, 1)
shape w = (batch_size, 1, z_dim)
shape b = (batch_size, 1, 1)
shape z = (batch_size, z_dim).
"""
zk = zk.unsqueeze(2)
uw = torch.bmm(w, u)
m_uw = -1.0 + self.softplus(uw)
w_norm_sq = torch.sum(w ** 2, dim=2, keepdim=True)
u_hat = u + (m_uw - uw) * w.transpose(2, 1) / w_norm_sq
wzb = torch.bmm(w, zk) + b
z = zk + u_hat * self.h(wzb)
z = z.squeeze(2)
psi = w * self._der_h(wzb)
logdet = torch.log(torch.abs(1 + torch.bmm(psi, u_hat)))
logdet = logdet.squeeze(2).squeeze(1)
return z, logdet
class Error(Exception):
"""Base error class, from which all other errors derive."""
pass
class InvalidArgumentError(Error):
"""This error will be shown when a given argument has an invalid value."""
pass
class NormalizingFlow(nn.Module):
"""Base class for normalizing flows."""
def __init__(self, h_dim, z_dim, flow_depth, hidden_depth):
super().__init__()
self.h_dim = h_dim
self.z_dim = z_dim
self.flow_depth = flow_depth
self.hidden_depth = hidden_depth
@property
def flow_depth(self):
return self._flow_depth
@flow_depth.setter
def flow_depth(self, value):
if not isinstance(value, int):
raise InvalidArgumentError('flow_depth should be an integer.')
elif value < 1:
raise InvalidArgumentError(
'flow_depth should be strictly positive.')
else:
self._flow_depth = value
@property
def hidden_depth(self):
return self._hidden_depth
@hidden_depth.setter
def hidden_depth(self, value):
if not isinstance(value, int):
raise InvalidArgumentError('hidden_depth should be an integer.')
elif value < 0:
raise InvalidArgumentError('hidden_depth should be positive.')
else:
self._hidden_depth = value
class Model(NormalizingFlow):
"""Planar Normalizing flow with single unit bottleneck."""
def __init__(self, h_dim, z_dim, flow_depth):
super().__init__(h_dim, z_dim, flow_depth, 0)
self.flow = PlanarStep()
self.h_to_u = nn.Linear(self.h_dim, self.flow_depth * self.z_dim)
self.h_to_w = nn.Linear(self.h_dim, self.flow_depth * self.z_dim)
self.h_to_b = nn.Linear(self.h_dim, self.flow_depth)
def forward(self, z, h):
u = self.h_to_u(h).view(-1, self.flow_depth, self.z_dim, 1)
w = self.h_to_w(h).view(-1, self.flow_depth, 1, self.z_dim)
b = self.h_to_b(h).view(-1, self.flow_depth, 1, 1)
z_k = z
logdet = 0.0
for k in range(self.flow_depth):
z_k, ldj = self.flow(z_k, u[:, k, :, :], w[:, k, :, :], b[:, k,
:, :])
logdet += ldj
return z_k, logdet
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [4, 4, 1]
|
Decoder_h | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/e7/ce7o5lvrzmyyeuisijxjne3lbkm23muivsnshxghpa4un3wa5g4d.py
# Topologically Sorted Source Nodes: [v, v_1, mul, add], Original ATen: [aten.index, aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# v => index
# v_1 => index_1
# Graph fragment:
# %index : [num_users=1] = call_function[target=torch.ops.aten.index.Tensor](args = (%primals_1, [%primals_2]), kwargs = {})
# %index_1 : [num_users=1] = call_function[target=torch.ops.aten.index.Tensor](args = (%primals_3, [%primals_2]), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%normal_functional, %index_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%index, %mul), kwargs = {})
triton_poi_fused_add_index_mul_0 = async_compile.triton('triton_poi_fused_add_index_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*i64', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_index_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_index_mul_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr2 + (x2), xmask)
tmp1 = tl.full([XBLOCK], 4, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert(((0 <= tmp4) & (tmp4 < 4)) | ~(xmask), "index out of bounds: 0 <= tmp4 < 4")
tmp6 = tl.load(in_ptr1 + (x0 + (4*tmp4)), xmask)
tmp8 = tl.load(in_ptr3 + (x0 + (4*tmp4)), xmask)
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4), (4, 1))
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [eps], Original ATen: [aten.normal_functional]
buf1 = torch.ops.aten.normal_functional.default(buf0)
buf2 = buf1
del buf1
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [v, v_1, mul, add], Original ATen: [aten.index, aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_index_mul_0.run(primals_2, primals_1, buf2, primals_3, buf3, 16, grid=grid(16), stream=stream0)
del primals_1
del primals_3
return (buf3, primals_2, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.int64)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.distributions as dist
import torch.nn as nn
class Decoder_h(nn.Module):
def __init__(self, B, H_dim):
super().__init__()
self.B = B
self.H_dim = H_dim
self.make_parameters()
def make_parameters(self):
self.mu = nn.Linear(self.H_dim, self.B, bias=False)
self.sigma = nn.Linear(self.H_dim, self.B, bias=False)
torch.nn.init.uniform_(self.sigma.weight, a=1.0, b=2.0)
def _log_likelihood(self, h):
"""
h: shape=(BS,N,H_dim)
"""
BS, S, H_dim = h.shape
return dist.Normal(self.mu.weight.view(1, 1, self.B, H_dim), self.
sigma.weight.view(1, 1, self.B, self.H_dim)).log_prob(h.view(BS,
S, 1, H_dim))
def forward(self, z):
"""
z: shape = (BS,N) or (BS,) or (1,)
"""
h_dist = dist.Normal(self.mu.weight[z], self.sigma.weight[z])
return h_dist.rsample()
def get_inputs():
return [torch.ones([4], dtype=torch.int64)]
def get_init_inputs():
return [[], {'B': 4, 'H_dim': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.distributions as dist
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_index_mul_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr2 + x2, xmask)
tmp1 = tl.full([XBLOCK], 4, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert((0 <= tmp4) & (tmp4 < 4) | ~xmask,
'index out of bounds: 0 <= tmp4 < 4')
tmp6 = tl.load(in_ptr1 + (x0 + 4 * tmp4), xmask)
tmp8 = tl.load(in_ptr3 + (x0 + 4 * tmp4), xmask)
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tl.store(out_ptr0 + x2, tmp10, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4), (4, 1))
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf1 = torch.ops.aten.normal_functional.default(buf0)
buf2 = buf1
del buf1
buf3 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_add_index_mul_0[grid(16)](primals_2, primals_1,
buf2, primals_3, buf3, 16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_1
del primals_3
return buf3, primals_2, buf2
class Decoder_hNew(nn.Module):
def __init__(self, B, H_dim):
super().__init__()
self.B = B
self.H_dim = H_dim
self.make_parameters()
def make_parameters(self):
self.mu = nn.Linear(self.H_dim, self.B, bias=False)
self.sigma = nn.Linear(self.H_dim, self.B, bias=False)
torch.nn.init.uniform_(self.sigma.weight, a=1.0, b=2.0)
def _log_likelihood(self, h):
"""
h: shape=(BS,N,H_dim)
"""
BS, S, H_dim = h.shape
return dist.Normal(self.mu.weight.view(1, 1, self.B, H_dim), self.
sigma.weight.view(1, 1, self.B, self.H_dim)).log_prob(h.view(BS,
S, 1, H_dim))
def forward(self, input_0):
primals_1 = self.mu.weight
primals_3 = self.sigma.weight
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| shaabhishek/pp_lvm | Decoder_h | false | 4,296 | [
"Apache-2.0"
] | 0 | 0fcceb7f004ab01da7c5508b576983b9d4af36c8 | https://github.com/shaabhishek/pp_lvm/tree/0fcceb7f004ab01da7c5508b576983b9d4af36c8 | import torch
import torch.distributions as dist
import torch.nn as nn
class Model(nn.Module):
def __init__(self, B, H_dim):
super().__init__()
self.B = B
self.H_dim = H_dim
self.make_parameters()
def make_parameters(self):
self.mu = nn.Linear(self.H_dim, self.B, bias=False)
self.sigma = nn.Linear(self.H_dim, self.B, bias=False)
torch.nn.init.uniform_(self.sigma.weight, a=1.0, b=2.0)
def _log_likelihood(self, h):
"""
h: shape=(BS,N,H_dim)
"""
BS, S, H_dim = h.shape
return dist.Normal(self.mu.weight.view(1, 1, self.B, H_dim), self.
sigma.weight.view(1, 1, self.B, self.H_dim)).log_prob(h.view(BS,
S, 1, H_dim))
def forward(self, z):
"""
z: shape = (BS,N) or (BS,) or (1,)
"""
h_dist = dist.Normal(self.mu.weight[z], self.sigma.weight[z])
return h_dist.rsample()
def get_inputs():
return [torch.ones([4], dtype=torch.int64)]
def get_init_inputs():
return [4, 4]
|
VDB | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_7/inductor_cache/nc/cncwsucylpsg2zmlivjfxu6vbd64ztxjndlsix2ysjtby3xohgk4.py
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# h => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_1,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_tanh_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/id/cid2gh4po4iynry6fjv4xkpr2m27rgu7cpf347kfyghfq4xzksz5.py
# Topologically Sorted Source Nodes: [truediv, std, mul, z], Original ATen: [aten.div, aten.exp, aten.mul, aten.add]
# Source node to ATen node mapping:
# mul => mul
# std => exp
# truediv => div
# z => add
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_5, 2), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%div,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp, %randn), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %mul), kwargs = {})
triton_poi_fused_add_div_exp_mul_1 = async_compile.triton('triton_poi_fused_add_div_exp_mul_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_exp_mul_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_div_exp_mul_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp5 = tl.load(in_ptr2 + (x0), xmask)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tl_math.exp(tmp3)
tmp6 = tmp4 * tmp5
tmp7 = tmp0 + tmp6
tl.store(out_ptr0 + (x0), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_7/inductor_cache/xr/cxrxf4nkydknjv7xhdecpyrprhviagsqwicrk4lpp64qv2hkzaxp.py
# Topologically Sorted Source Nodes: [prob], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# prob => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_9,), kwargs = {})
triton_poi_fused_sigmoid_2 = async_compile.triton('triton_poi_fused_sigmoid_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=108), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (1, 4), (4, 1))
assert_size_stride(primals_11, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_2, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mu], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_5
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [logvar], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_7
# Topologically Sorted Source Nodes: [eps], Original ATen: [aten.randn_like]
buf4 = torch.ops.aten.randn.default([4, 4, 4, 4], dtype=torch.float32, device=device(type='cuda', index=0), pin_memory=False)
buf5 = buf4
del buf4
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv, std, mul, z], Original ATen: [aten.div, aten.exp, aten.mul, aten.add]
triton_poi_fused_add_div_exp_mul_1.run(buf2, buf3, buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf6, (64, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf7)
buf8 = reinterpret_tensor(buf7, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.tanh]
triton_poi_fused_tanh_0.run(buf8, primals_9, 256, grid=grid(256), stream=stream0)
del primals_9
buf9 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf8, (64, 4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 1), (1, 4), 0), out=buf9)
buf10 = reinterpret_tensor(buf9, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [prob], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_2.run(buf10, primals_11, 64, grid=grid(64), stream=stream0)
del primals_11
return (buf10, reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0), buf5, reinterpret_tensor(buf6, (64, 4), (4, 1), 0), buf8, buf10, primals_10, primals_8, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from _paritybench_helpers import _mock_config
import torch
import torch.nn as nn
class VDB(nn.Module):
def __init__(self, num_inputs, args):
super(VDB, self).__init__()
self.fc1 = nn.Linear(num_inputs, args.hidden_size)
self.fc2 = nn.Linear(args.hidden_size, args.z_size)
self.fc3 = nn.Linear(args.hidden_size, args.z_size)
self.fc4 = nn.Linear(args.z_size, args.hidden_size)
self.fc5 = nn.Linear(args.hidden_size, 1)
self.fc5.weight.data.mul_(0.1)
self.fc5.bias.data.mul_(0.0)
def encoder(self, x):
h = torch.tanh(self.fc1(x))
return self.fc2(h), self.fc3(h)
def reparameterize(self, mu, logvar):
std = torch.exp(logvar / 2)
eps = torch.randn_like(std)
return mu + std * eps
def discriminator(self, z):
h = torch.tanh(self.fc4(z))
return torch.sigmoid(self.fc5(h))
def forward(self, x):
mu, logvar = self.encoder(x)
z = self.reparameterize(mu, logvar)
prob = self.discriminator(z)
return prob, mu, logvar
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_inputs': 4, 'args': _mock_config(hidden_size=4,
z_size=4)}]
| import torch
from torch import device
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_add_div_exp_mul_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp5 = tl.load(in_ptr2 + x0, xmask)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tl_math.exp(tmp3)
tmp6 = tmp4 * tmp5
tmp7 = tmp0 + tmp6
tl.store(out_ptr0 + x0, tmp7, xmask)
@triton.jit
def triton_poi_fused_sigmoid_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (1, 4), (4, 1))
assert_size_stride(primals_11, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(256)](buf1, primals_2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf2)
del primals_5
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf1, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf3)
del primals_7
buf4 = torch.ops.aten.randn.default([4, 4, 4, 4], dtype=torch.
float32, device=device(type='cuda', index=0), pin_memory=False)
buf5 = buf4
del buf4
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_div_exp_mul_1[grid(256)](buf2, buf3, buf5,
buf6, 256, XBLOCK=256, num_warps=4, num_stages=1)
buf7 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf6, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf7)
buf8 = reinterpret_tensor(buf7, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf7
triton_poi_fused_tanh_0[grid(256)](buf8, primals_9, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_9
buf9 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf8, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_10, (4, 1), (1, 4), 0), out=buf9)
buf10 = reinterpret_tensor(buf9, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf9
triton_poi_fused_sigmoid_2[grid(64)](buf10, primals_11, 64, XBLOCK=
64, num_warps=1, num_stages=1)
del primals_11
return buf10, reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0
), buf5, reinterpret_tensor(buf6, (64, 4), (4, 1), 0
), buf8, buf10, primals_10, primals_8, primals_6, primals_4
class VDBNew(nn.Module):
def __init__(self, num_inputs, args):
super(VDBNew, self).__init__()
self.fc1 = nn.Linear(num_inputs, args.hidden_size)
self.fc2 = nn.Linear(args.hidden_size, args.z_size)
self.fc3 = nn.Linear(args.hidden_size, args.z_size)
self.fc4 = nn.Linear(args.z_size, args.hidden_size)
self.fc5 = nn.Linear(args.hidden_size, 1)
self.fc5.weight.data.mul_(0.1)
self.fc5.bias.data.mul_(0.0)
def encoder(self, x):
h = torch.tanh(self.fc1(x))
return self.fc2(h), self.fc3(h)
def reparameterize(self, mu, logvar):
std = torch.exp(logvar / 2)
eps = torch.randn_like(std)
return mu + std * eps
def discriminator(self, z):
h = torch.tanh(self.fc4(z))
return torch.sigmoid(self.fc5(h))
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_8 = self.fc4.weight
primals_9 = self.fc4.bias
primals_10 = self.fc5.weight
primals_11 = self.fc5.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0], output[1], output[2]
| sgrimbly/lets-do-irl | VDB | false | 4,297 | [
"MIT"
] | 0 | 4233e238342394feef6a7bd495cc6b700d435b00 | https://github.com/sgrimbly/lets-do-irl/tree/4233e238342394feef6a7bd495cc6b700d435b00 | from _paritybench_helpers import _mock_config
import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, num_inputs, args):
super().__init__()
self.fc1 = nn.Linear(num_inputs, args.hidden_size)
self.fc2 = nn.Linear(args.hidden_size, args.z_size)
self.fc3 = nn.Linear(args.hidden_size, args.z_size)
self.fc4 = nn.Linear(args.z_size, args.hidden_size)
self.fc5 = nn.Linear(args.hidden_size, 1)
self.fc5.weight.data.mul_(0.1)
self.fc5.bias.data.mul_(0.0)
def encoder(self, x):
h = torch.tanh(self.fc1(x))
return self.fc2(h), self.fc3(h)
def reparameterize(self, mu, logvar):
std = torch.exp(logvar / 2)
eps = torch.randn_like(std)
return mu + std * eps
def discriminator(self, z):
h = torch.tanh(self.fc4(z))
return torch.sigmoid(self.fc5(h))
def forward(self, x):
mu, logvar = self.encoder(x)
z = self.reparameterize(mu, logvar)
prob = self.discriminator(z)
return prob, mu, logvar
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_inputs': 4, 'args': _mock_config(hidden_size=4,
z_size=4)}]
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.