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
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
GlobalpoolFC | # 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_6/inductor_cache/l3/cl35tzbhrd24dhunkbb6gjs54aklpyr46oikqhoylcgmkcmhujil.py
# Topologically Sorted Source Nodes: [rep], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# rep => mean
# Graph fragment:
# %mean : [num_users=2] = 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')
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, 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: [rep], 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)
del primals_1
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [y_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(buf1, (4, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_2
del primals_3
return (buf1, buf2, reinterpret_tensor(buf1, (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
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
from torch import nn
class GlobalpoolFC(nn.Module):
def __init__(self, num_in, num_class):
super(GlobalpoolFC, self).__init__()
self.pool = nn.AdaptiveAvgPool2d(output_size=1)
self.fc = nn.Linear(num_in, num_class)
def forward(self, x):
rep = self.pool(x)
y = rep.reshape(rep.shape[0], -1)
y = self.fc(y)
return rep, y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_in': 4, 'num_class': 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
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)
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, 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)
del primals_1
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(buf1, (4, 4), (4,
1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), alpha
=1, beta=1, out=buf2)
del primals_2
del primals_3
return buf1, buf2, reinterpret_tensor(buf1, (4, 4), (4, 1), 0)
class GlobalpoolFCNew(nn.Module):
def __init__(self, num_in, num_class):
super(GlobalpoolFCNew, self).__init__()
self.pool = nn.AdaptiveAvgPool2d(output_size=1)
self.fc = nn.Linear(num_in, num_class)
def forward(self, input_0):
primals_2 = self.fc.weight
primals_3 = self.fc.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0], output[1]
| blackcow/Fair-AT | GlobalpoolFC | false | 1,550 | [
"Apache-2.0"
] | 0 | 62fc269fedd4b63c4b48ae390d494b3832e65fa8 | https://github.com/blackcow/Fair-AT/tree/62fc269fedd4b63c4b48ae390d494b3832e65fa8 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, num_in, num_class):
super().__init__()
self.pool = nn.AdaptiveAvgPool2d(output_size=1)
self.fc = nn.Linear(num_in, num_class)
def forward(self, x):
rep = self.pool(x)
y = rep.reshape(rep.shape[0], -1)
y = self.fc(y)
return rep, y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
FairLoss | # 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_6/inductor_cache/4c/c4cj5eeyzvknv6pm676dxw7lninv5x5m6wurjyeau7os4pjnkq5x.py
# Topologically Sorted Source Nodes: [diag_embed, logits_1, abs_1, logits_2, mul], Original ATen: [aten.diag_embed, aten.sub, aten.abs, aten.sum, aten.mul]
# Source node to ATen node mapping:
# abs_1 => abs_1
# diag_embed => eq, full_default, iota, where
# logits_1 => sub
# logits_2 => sum_1
# mul => mul
# Graph fragment:
# %iota : [num_users=1] = call_function[target=torch.ops.prims.iota.default](args = (4,), kwargs = {start: 0, step: 1, dtype: torch.int64, device: cuda:0, requires_grad: False})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota, %unsqueeze_1), 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})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq, %permute_1, %full_default), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mm, %where), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%abs_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 4), kwargs = {})
triton_per_fused_abs_diag_embed_mul_sub_sum_0 = async_compile.triton('triton_per_fused_abs_diag_embed_mul_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=[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_diag_embed_mul_sub_sum_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_abs_diag_embed_mul_sub_sum_0(in_out_ptr0, in_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)
r2 = rindex
r0 = rindex % 4
r1 = (rindex // 4)
tmp0 = tl.load(in_ptr0 + (r2), None)
tmp4 = tl.load(in_ptr0 + (5*r0), None, eviction_policy='evict_last')
tmp1 = r0
tmp2 = r1
tmp3 = tmp1 == tmp2
tmp5 = 0.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 - tmp6
tmp8 = tl_math.abs(tmp7)
tmp9 = tl.broadcast_to(tmp8, [XBLOCK, RBLOCK])
tmp11 = tl.sum(tmp9, 1)[:, None]
tmp12 = 4.0
tmp13 = tmp11 * tmp12
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp13, None)
''', 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, 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: [logits], Original ATen: [aten.mm]
extern_kernels.mm(arg0_1, reinterpret_tensor(arg0_1, (4, 4), (1, 4), 0), out=buf0)
del arg0_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [diag_embed, logits_1, abs_1, logits_2, mul], Original ATen: [aten.diag_embed, aten.sub, aten.abs, aten.sum, aten.mul]
stream0 = get_raw_stream(0)
triton_per_fused_abs_diag_embed_mul_sub_sum_0.run(buf2, buf0, 1, 16, grid=grid(1), stream=stream0)
del buf0
return (buf2, )
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)
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
class FairLoss(nn.Module):
def __init__(self, lamda):
super(FairLoss, self).__init__()
self.lamda = lamda
def forward(self, rep):
logits = torch.mm(rep, torch.transpose(rep, 0, 1))
logits = logits - torch.diag_embed(torch.diag(logits))
logits = logits.abs().sum()
return logits * self.lamda
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'lamda': 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
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_diag_embed_mul_sub_sum_0(in_out_ptr0, in_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)
r2 = rindex
r0 = rindex % 4
r1 = rindex // 4
tmp0 = tl.load(in_ptr0 + r2, None)
tmp4 = tl.load(in_ptr0 + 5 * r0, None, eviction_policy='evict_last')
tmp1 = r0
tmp2 = r1
tmp3 = tmp1 == tmp2
tmp5 = 0.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 - tmp6
tmp8 = tl_math.abs(tmp7)
tmp9 = tl.broadcast_to(tmp8, [XBLOCK, RBLOCK])
tmp11 = tl.sum(tmp9, 1)[:, None]
tmp12 = 4.0
tmp13 = tmp11 * tmp12
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp13, None)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (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.mm(arg0_1, reinterpret_tensor(arg0_1, (4, 4), (1, 4),
0), out=buf0)
del arg0_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
get_raw_stream(0)
triton_per_fused_abs_diag_embed_mul_sub_sum_0[grid(1)](buf2, buf0,
1, 16, XBLOCK=1, num_warps=2, num_stages=1)
del buf0
return buf2,
class FairLossNew(nn.Module):
def __init__(self, lamda):
super(FairLossNew, self).__init__()
self.lamda = lamda
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| blackcow/Fair-AT | FairLoss | false | 1,551 | [
"Apache-2.0"
] | 0 | 62fc269fedd4b63c4b48ae390d494b3832e65fa8 | https://github.com/blackcow/Fair-AT/tree/62fc269fedd4b63c4b48ae390d494b3832e65fa8 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, lamda):
super().__init__()
self.lamda = lamda
def forward(self, rep):
logits = torch.mm(rep, torch.transpose(rep, 0, 1))
logits = logits - torch.diag_embed(torch.diag(logits))
logits = logits.abs().sum()
return logits * self.lamda
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [4]
|
MockOpposite | # 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_6/inductor_cache/u7/cu7c6x6yq6elqkgwbbyxjiwcikve4tpl62dm2f4ef2eiiujptbve.py
# Topologically Sorted Source Nodes: [setitem, setitem_1], Original ATen: [aten.lift_fresh, aten.index_put]
# Source node to ATen node mapping:
# setitem => full_default, index_put
# setitem_1 => full_default_1, index_put_1
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %index_put : [num_users=1] = call_function[target=torch.ops.aten.index_put.default](args = (%arg0_1, [%eq], %full_default), kwargs = {})
# %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: cpu, pin_memory: False})
# %index_put_1 : [num_users=1] = call_function[target=torch.ops.aten.index_put_.default](args = (%index_put, [%eq_1], %full_default_1), kwargs = {})
triton_poi_fused_index_put_lift_fresh_0 = async_compile.triton('triton_poi_fused_index_put_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=[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_index_put_lift_fresh_0', 'mutated_arg_names': ['in_out_ptr0'], '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_index_put_lift_fresh_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
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 1.0
tmp2 = tmp0 == tmp1
tmp3 = 0.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tmp5 = tmp0 == tmp3
tmp6 = tl.where(tmp5, tmp1, tmp4)
tl.store(in_out_ptr0 + (x0), tmp6, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [setitem, setitem_1], Original ATen: [aten.lift_fresh, aten.index_put]
stream0 = get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_0.run(buf1, arg0_1, 256, grid=grid(256), stream=stream0)
del arg0_1
return (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
class MockOpposite(nn.Module):
def __init__(self):
"""Initialize the model"""
super(MockOpposite, self).__init__()
def forward(self, input):
"""A single forward pass of the model. Returns input - 1.
Args:
input: The input to the model as a tensor of
batch_size X input_size
"""
output = input.clone().detach()
output[input == 1] = 0
output[input == 0] = 1
return output
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
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_index_put_lift_fresh_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
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 1.0
tmp2 = tmp0 == tmp1
tmp3 = 0.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tmp5 = tmp0 == tmp3
tmp6 = tl.where(tmp5, tmp1, tmp4)
tl.store(in_out_ptr0 + x0, tmp6, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_0[grid(256)](buf1, arg0_1,
256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf1,
class MockOppositeNew(nn.Module):
def __init__(self):
"""Initialize the model"""
super(MockOppositeNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| blagojce95/ai-research-mamo-framework | MockOpposite | false | 1,552 | [
"Apache-2.0"
] | 0 | 7f3b5a5a9fb8b19c9eef453b81b03b6046a33bf2 | https://github.com/blagojce95/ai-research-mamo-framework/tree/7f3b5a5a9fb8b19c9eef453b81b03b6046a33bf2 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
"""Initialize the model"""
super().__init__()
def forward(self, input):
"""A single forward pass of the model. Returns input - 1.
Args:
input: The input to the model as a tensor of
batch_size X input_size
"""
output = input.clone().detach()
output[input == 1] = 0
output[input == 0] = 1
return output
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
PatchMerging | # 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_6/inductor_cache/27/c27w2vsk7hohsacusbwl2tb7dxz6zo6ythuojm6gntenf734qgqj.py
# Topologically Sorted Source Nodes: [x_1, x_3], Original ATen: [aten.cat, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_1 => cat
# x_3 => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%slice_3, %slice_7, %slice_11, %slice_15], -1), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_1, [2]), 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 = (%view_1, %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_2), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_3), kwargs = {})
triton_per_fused_cat_native_layer_norm_0 = async_compile.triton('triton_per_fused_cat_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=[4, 16],
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, 8), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_cat_native_layer_norm_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 6, '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_cat_native_layer_norm_0(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, 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
tmp46 = tl.load(in_ptr1 + (r1), None, eviction_policy='evict_last')
tmp48 = tl.load(in_ptr2 + (r1), None, eviction_policy='evict_last')
tmp0 = r1
tmp1 = tl.full([1, 1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1, 1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((16*x0) + r1), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1, 1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr0 + (8 + (16*x0) + ((-4) + r1)), tmp9 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1, 1], 12, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tmp11 & tmp13
tmp15 = tl.load(in_ptr0 + (4 + (16*x0) + ((-8) + r1)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp16 = tmp0 >= tmp12
tmp17 = tl.full([1, 1], 16, tl.int64)
tmp18 = tmp0 < tmp17
tmp19 = tl.load(in_ptr0 + (12 + (16*x0) + ((-12) + r1)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp20 = tl.where(tmp14, tmp15, tmp19)
tmp21 = tl.where(tmp9, tmp10, tmp20)
tmp22 = tl.where(tmp4, tmp5, tmp21)
tmp23 = tl.broadcast_to(tmp22, [XBLOCK, RBLOCK])
tmp25 = tl.where(xmask, tmp23, 0)
tmp26 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp28 = tl.where(xmask, tmp26, 0)
tmp29 = tl.sum(tmp28, 1)[:, None]
tmp30 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp31 = tmp30.to(tl.float32)
tmp32 = tmp29 / tmp31
tmp33 = tmp23 - tmp32
tmp34 = tmp33 * tmp33
tmp35 = tl.broadcast_to(tmp34, [XBLOCK, RBLOCK])
tmp37 = tl.where(xmask, tmp35, 0)
tmp38 = tl.sum(tmp37, 1)[:, None]
tmp39 = 16.0
tmp40 = tmp38 / tmp39
tmp41 = 1e-05
tmp42 = tmp40 + tmp41
tmp43 = libdevice.rsqrt(tmp42)
tmp44 = tmp22 - tmp32
tmp45 = tmp44 * tmp43
tmp47 = tmp45 * tmp46
tmp49 = tmp47 + tmp48
tl.store(out_ptr0 + (r1 + (16*x0)), tmp22, xmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp43, xmask)
tl.store(out_ptr2 + (r1 + (16*x0)), tmp49, xmask)
tl.store(out_ptr1 + (x0), tmp32, 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), (16, 4, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (16, ), (1, ))
assert_size_stride(primals_4, (8, 16), (16, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 1, 16), (16, 16, 16, 1), torch.float32)
buf1 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
buf2 = empty_strided_cuda((4, 1, 1), (1, 4, 4), torch.float32)
buf4 = reinterpret_tensor(buf2, (4, 1, 1), (1, 1, 1), 0); del buf2 # reuse
buf5 = empty_strided_cuda((4, 1, 16), (16, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1, x_3], Original ATen: [aten.cat, aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_per_fused_cat_native_layer_norm_0.run(buf4, primals_1, primals_2, primals_3, buf0, buf1, buf5, 4, 16, grid=grid(4), stream=stream0)
del primals_1
del primals_2
del primals_3
buf6 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf5, (4, 16), (16, 1), 0), reinterpret_tensor(primals_4, (16, 8), (1, 16), 0), out=buf6)
return (reinterpret_tensor(buf6, (4, 1, 8), (8, 8, 1), 0), buf0, buf1, buf4, reinterpret_tensor(buf5, (4, 16), (16, 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, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((8, 16), (16, 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 torch.nn as nn
import torch.nn.functional as F
from math import sqrt
from torch import optim as optim
class PatchMerging(nn.Module):
"""Patch Merging Layer.
Args:
input_resolution (tuple[int]): Resolution of input feature.
dim (int): Number of input channels.
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
"""
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
self.norm = norm_layer(4 * dim)
def forward(self, x):
""" Forward function.
Args:
x: Input feature, tensor size (B, H*W, C).
H, W: Spatial resolution of the input feature.
"""
B, L, C = x.shape
H = int(sqrt(L))
W = H
x = x.view(B, H, W, C)
pad_input = H % 2 == 1 or W % 2 == 1
if pad_input:
x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))
x0 = x[:, 0::2, 0::2, :]
x1 = x[:, 1::2, 0::2, :]
x2 = x[:, 0::2, 1::2, :]
x3 = x[:, 1::2, 1::2, :]
x = torch.cat([x0, x1, x2, x3], -1)
x = x.view(B, -1, 4 * C)
x = self.norm(x)
x = self.reduction(x)
return x
def extra_repr(self) ->str:
return f'input_resolution={self.input_resolution}, dim={self.dim}'
def flops(self):
H, W = self.input_resolution
flops = H * W * self.dim
flops += H // 2 * (W // 2) * 4 * self.dim * 2 * self.dim
return flops
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_resolution': 4, '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
from torch import optim as optim
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_cat_native_layer_norm_0(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, out_ptr0, out_ptr1, out_ptr2, 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
tmp46 = tl.load(in_ptr1 + r1, None, eviction_policy='evict_last')
tmp48 = tl.load(in_ptr2 + r1, None, eviction_policy='evict_last')
tmp0 = r1
tl.full([1, 1], 0, tl.int64)
tmp3 = tl.full([1, 1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (16 * x0 + r1), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1, 1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr0 + (8 + 16 * x0 + (-4 + r1)), tmp9 & xmask,
eviction_policy='evict_last', other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1, 1], 12, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tmp11 & tmp13
tmp15 = tl.load(in_ptr0 + (4 + 16 * x0 + (-8 + r1)), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp16 = tmp0 >= tmp12
tl.full([1, 1], 16, tl.int64)
tmp19 = tl.load(in_ptr0 + (12 + 16 * x0 + (-12 + r1)), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp20 = tl.where(tmp14, tmp15, tmp19)
tmp21 = tl.where(tmp9, tmp10, tmp20)
tmp22 = tl.where(tmp4, tmp5, tmp21)
tmp23 = tl.broadcast_to(tmp22, [XBLOCK, RBLOCK])
tl.where(xmask, tmp23, 0)
tmp26 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp28 = tl.where(xmask, tmp26, 0)
tmp29 = tl.sum(tmp28, 1)[:, None]
tmp30 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp31 = tmp30.to(tl.float32)
tmp32 = tmp29 / tmp31
tmp33 = tmp23 - tmp32
tmp34 = tmp33 * tmp33
tmp35 = tl.broadcast_to(tmp34, [XBLOCK, RBLOCK])
tmp37 = tl.where(xmask, tmp35, 0)
tmp38 = tl.sum(tmp37, 1)[:, None]
tmp39 = 16.0
tmp40 = tmp38 / tmp39
tmp41 = 1e-05
tmp42 = tmp40 + tmp41
tmp43 = libdevice.rsqrt(tmp42)
tmp44 = tmp22 - tmp32
tmp45 = tmp44 * tmp43
tmp47 = tmp45 * tmp46
tmp49 = tmp47 + tmp48
tl.store(out_ptr0 + (r1 + 16 * x0), tmp22, xmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp43, xmask)
tl.store(out_ptr2 + (r1 + 16 * x0), tmp49, xmask)
tl.store(out_ptr1 + x0, tmp32, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (16,), (1,))
assert_size_stride(primals_4, (8, 16), (16, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 1, 16), (16, 16, 16, 1), torch.float32
)
buf1 = empty_strided_cuda((4, 1, 1), (1, 1, 1), torch.float32)
buf2 = empty_strided_cuda((4, 1, 1), (1, 4, 4), torch.float32)
buf4 = reinterpret_tensor(buf2, (4, 1, 1), (1, 1, 1), 0)
del buf2
buf5 = empty_strided_cuda((4, 1, 16), (16, 16, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_cat_native_layer_norm_0[grid(4)](buf4, primals_1,
primals_2, primals_3, buf0, buf1, buf5, 4, 16, XBLOCK=1,
num_warps=2, num_stages=1)
del primals_1
del primals_2
del primals_3
buf6 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf5, (4, 16), (16, 1), 0),
reinterpret_tensor(primals_4, (16, 8), (1, 16), 0), out=buf6)
return reinterpret_tensor(buf6, (4, 1, 8), (8, 8, 1), 0
), buf0, buf1, buf4, reinterpret_tensor(buf5, (4, 16), (16, 1), 0
), primals_4
class PatchMergingNew(nn.Module):
"""Patch Merging Layer.
Args:
input_resolution (tuple[int]): Resolution of input feature.
dim (int): Number of input channels.
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
"""
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
self.norm = norm_layer(4 * dim)
def extra_repr(self) ->str:
return f'input_resolution={self.input_resolution}, dim={self.dim}'
def flops(self):
H, W = self.input_resolution
flops = H * W * self.dim
flops += H // 2 * (W // 2) * 4 * self.dim * 2 * self.dim
return flops
def forward(self, input_0):
primals_4 = self.reduction.weight
primals_2 = self.norm.weight
primals_3 = self.norm.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| billpsomas/ibot | PatchMerging | false | 1,553 | [
"Apache-2.0"
] | 0 | c6fbce7e2a59780f39ad7304ed9a8b1acf038d2d | https://github.com/billpsomas/ibot/tree/c6fbce7e2a59780f39ad7304ed9a8b1acf038d2d | import torch
import torch.nn as nn
import torch.nn.functional as F
from math import sqrt
from torch import optim as optim
class Model(nn.Module):
"""Patch Merging Layer.
Args:
input_resolution (tuple[int]): Resolution of input feature.
dim (int): Number of input channels.
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
"""
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
self.norm = norm_layer(4 * dim)
def forward(self, x):
""" Forward function.
Args:
x: Input feature, tensor size (B, H*W, C).
H, W: Spatial resolution of the input feature.
"""
B, L, C = x.shape
H = int(sqrt(L))
W = H
x = x.view(B, H, W, C)
pad_input = H % 2 == 1 or W % 2 == 1
if pad_input:
x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))
x0 = x[:, 0::2, 0::2, :]
x1 = x[:, 1::2, 0::2, :]
x2 = x[:, 0::2, 1::2, :]
x3 = x[:, 1::2, 1::2, :]
x = torch.cat([x0, x1, x2, x3], -1)
x = x.view(B, -1, 4 * C)
x = self.norm(x)
x = self.reduction(x)
return x
def extra_repr(self) ->str:
return f'input_resolution={self.input_resolution}, dim={self.dim}'
def flops(self):
H, W = self.input_resolution
flops = H * W * self.dim
flops += H // 2 * (W // 2) * 4 * self.dim * 2 * self.dim
return flops
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
MyModel | # 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_6/inductor_cache/vq/cvq3k2pudyocnklorpznpn2x4uwcd75kitqtmbswq44pxwjzv5lf.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 = (%arg0_1, 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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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': 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_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
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: [add], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_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
class MyModel(nn.Module):
def __init__(self):
super(MyModel, self).__init__()
def forward(self, x):
return x + 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
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 = 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)
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_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class MyModelNew(nn.Module):
def __init__(self):
super(MyModelNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| blagojce95/ai-research-mamo-framework | MyModel | false | 1,554 | [
"Apache-2.0"
] | 0 | 7f3b5a5a9fb8b19c9eef453b81b03b6046a33bf2 | https://github.com/blagojce95/ai-research-mamo-framework/tree/7f3b5a5a9fb8b19c9eef453b81b03b6046a33bf2 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return x + 1
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GreedyTop1 | # 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_6/inductor_cache/e7/ce7n4frmk2ex3wtp7kpbj5crm4am7bpa7fg7qdg4e66p4bwxptb4.py
# Topologically Sorted Source Nodes: [best_word_index], Original ATen: [aten.argmin]
# Source node to ATen node mapping:
# best_word_index => argmin
# Graph fragment:
# %argmin : [num_users=1] = call_function[target=torch.ops.aten.argmin.default](args = (%arg0_1, -1, True), kwargs = {})
triton_poi_fused_argmin_0 = async_compile.triton('triton_poi_fused_argmin_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: '*i64', 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_argmin_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_argmin_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 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp32 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 < tmp1
tmp3 = tmp0 == tmp1
tmp4 = tmp0 != tmp0
tmp5 = tmp1 != tmp1
tmp6 = tmp4 > tmp5
tmp7 = tmp2 | tmp6
tmp8 = tmp4 & tmp5
tmp9 = tmp3 | tmp8
tmp10 = tl.full([1], 0, tl.int64)
tmp11 = tl.full([1], 1, tl.int64)
tmp12 = tmp10 < tmp11
tmp13 = tmp9 & tmp12
tmp14 = tmp7 | tmp13
tmp15 = tl.where(tmp14, tmp0, tmp1)
tmp16 = tl.where(tmp14, tmp10, tmp11)
tmp18 = tmp15 < tmp17
tmp19 = tmp15 == tmp17
tmp20 = tmp15 != tmp15
tmp21 = tmp17 != tmp17
tmp22 = tmp20 > tmp21
tmp23 = tmp18 | tmp22
tmp24 = tmp20 & tmp21
tmp25 = tmp19 | tmp24
tmp26 = tl.full([1], 2, tl.int64)
tmp27 = tmp16 < tmp26
tmp28 = tmp25 & tmp27
tmp29 = tmp23 | tmp28
tmp30 = tl.where(tmp29, tmp15, tmp17)
tmp31 = tl.where(tmp29, tmp16, tmp26)
tmp33 = tmp30 < tmp32
tmp34 = tmp30 == tmp32
tmp35 = tmp30 != tmp30
tmp36 = tmp32 != tmp32
tmp37 = tmp35 > tmp36
tmp38 = tmp33 | tmp37
tmp39 = tmp35 & tmp36
tmp40 = tmp34 | tmp39
tmp41 = tl.full([1], 3, tl.int64)
tmp42 = tmp31 < tmp41
tmp43 = tmp40 & tmp42
tmp44 = tmp38 | tmp43
tmp45 = tl.where(tmp44, tmp30, tmp32)
tmp46 = tl.where(tmp44, tmp31, tmp41)
tl.store(out_ptr0 + (x0), tmp46, 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, 1), (16, 4, 1, 1), torch.int64)
# Topologically Sorted Source Nodes: [best_word_index], Original ATen: [aten.argmin]
stream0 = get_raw_stream(0)
triton_poi_fused_argmin_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
from typing import Optional
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
class GreedyTop1(pt.nn.Module):
"""
Implements picking the highest scoring next word with support for vocabulary selection and target factors.
"""
def forward(self, scores: 'pt.Tensor', vocab_slice_ids:
'Optional[pt.Tensor]'=None, target_factors: 'Optional[pt.Tensor]'=None
) ->pt.Tensor:
best_word_index = pt.argmin(scores, dim=-1, keepdim=True)
if vocab_slice_ids is not None:
best_word_index = vocab_slice_ids.index_select(0,
best_word_index.squeeze(1)).unsqueeze(1)
if target_factors is not None:
factor_index = target_factors[:, :, 1].int()
best_word_index = pt.cat((best_word_index, factor_index), dim=1)
return best_word_index
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 as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
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_argmin_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 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp32 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 < tmp1
tmp3 = tmp0 == tmp1
tmp4 = tmp0 != tmp0
tmp5 = tmp1 != tmp1
tmp6 = tmp4 > tmp5
tmp7 = tmp2 | tmp6
tmp8 = tmp4 & tmp5
tmp9 = tmp3 | tmp8
tmp10 = tl.full([1], 0, tl.int64)
tmp11 = tl.full([1], 1, tl.int64)
tmp12 = tmp10 < tmp11
tmp13 = tmp9 & tmp12
tmp14 = tmp7 | tmp13
tmp15 = tl.where(tmp14, tmp0, tmp1)
tmp16 = tl.where(tmp14, tmp10, tmp11)
tmp18 = tmp15 < tmp17
tmp19 = tmp15 == tmp17
tmp20 = tmp15 != tmp15
tmp21 = tmp17 != tmp17
tmp22 = tmp20 > tmp21
tmp23 = tmp18 | tmp22
tmp24 = tmp20 & tmp21
tmp25 = tmp19 | tmp24
tmp26 = tl.full([1], 2, tl.int64)
tmp27 = tmp16 < tmp26
tmp28 = tmp25 & tmp27
tmp29 = tmp23 | tmp28
tmp30 = tl.where(tmp29, tmp15, tmp17)
tmp31 = tl.where(tmp29, tmp16, tmp26)
tmp33 = tmp30 < tmp32
tmp34 = tmp30 == tmp32
tmp35 = tmp30 != tmp30
tmp36 = tmp32 != tmp32
tmp37 = tmp35 > tmp36
tmp38 = tmp33 | tmp37
tmp39 = tmp35 & tmp36
tmp40 = tmp34 | tmp39
tmp41 = tl.full([1], 3, tl.int64)
tmp42 = tmp31 < tmp41
tmp43 = tmp40 & tmp42
tmp44 = tmp38 | tmp43
tl.where(tmp44, tmp30, tmp32)
tmp46 = tl.where(tmp44, tmp31, tmp41)
tl.store(out_ptr0 + x0, tmp46, 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, 1), (16, 4, 1, 1), torch.int64)
get_raw_stream(0)
triton_poi_fused_argmin_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class GreedyTop1New(pt.nn.Module):
"""
Implements picking the highest scoring next word with support for vocabulary selection and target factors.
"""
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| blchu/sockeye | GreedyTop1 | false | 1,555 | [
"Apache-2.0"
] | 0 | 28044a44ee409c9b3df1711c0b16bdebdd463b2e | https://github.com/blchu/sockeye/tree/28044a44ee409c9b3df1711c0b16bdebdd463b2e | import torch
from typing import Optional
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
class Model(pt.nn.Module):
"""
Implements picking the highest scoring next word with support for vocabulary selection and target factors.
"""
def forward(self, scores: 'pt.Tensor', vocab_slice_ids:
'Optional[pt.Tensor]'=None, target_factors: 'Optional[pt.Tensor]'=None
) ->pt.Tensor:
best_word_index = pt.argmin(scores, dim=-1, keepdim=True)
if vocab_slice_ids is not None:
best_word_index = vocab_slice_ids.index_select(0,
best_word_index.squeeze(1)).unsqueeze(1)
if target_factors is not None:
factor_index = target_factors[:, :, 1].int()
best_word_index = pt.cat((best_word_index, factor_index), dim=1)
return best_word_index
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
PyTorchLHUC | # 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_6/inductor_cache/do/cdosk3mmczannzv2vxd73dkanxm666a44wldiwrtguo3x4uzu2w5.py
# Topologically Sorted Source Nodes: [sigmoid, weight, mul_1], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul_1 => mul_1
# sigmoid => sigmoid
# weight => mul
# 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, 2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_2), kwargs = {})
triton_poi_fused_mul_sigmoid_0 = async_compile.triton('triton_poi_fused_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=[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_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_sigmoid_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
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (x2), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp2 = 2.0
tmp3 = tmp1 * tmp2
tmp5 = tmp3 * tmp4
tl.store(out_ptr0 + (x2), tmp5, 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, ), (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, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sigmoid, weight, mul_1], Original ATen: [aten.sigmoid, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_sigmoid_0.run(primals_1, primals_2, buf0, 256, grid=grid(256), 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((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)
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 as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
class PyTorchLHUC(pt.nn.Module):
"""
Learning Hidden Unit Contribution
David Vilar. "Learning Hidden Unit Contribution for Adapting Neural
Machine Translation Models" NAACL 2018
:param num_hidden: Number of hidden units of the layer to be modified.
"""
def __init__(self, num_hidden: 'int') ->None:
super().__init__()
self.weight = pt.nn.Parameter(pt.Tensor(num_hidden))
def forward(self, data: 'pt.Tensor') ->pt.Tensor:
weight = 2 * pt.sigmoid(self.weight)
return weight * data
def weights_from_mxnet_block(self, block_mx: "'LHUC'"):
self.weight.data[:] = pt.as_tensor(block_mx.weight.data().asnumpy())
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_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
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
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_sigmoid_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
x2 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + x2, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp2 = 2.0
tmp3 = tmp1 * tmp2
tmp5 = tmp3 * tmp4
tl.store(out_ptr0 + x2, tmp5, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4,), (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, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_sigmoid_0[grid(256)](primals_1, primals_2,
buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
return buf0, primals_1, primals_2
class PyTorchLHUCNew(pt.nn.Module):
"""
Learning Hidden Unit Contribution
David Vilar. "Learning Hidden Unit Contribution for Adapting Neural
Machine Translation Models" NAACL 2018
:param num_hidden: Number of hidden units of the layer to be modified.
"""
def __init__(self, num_hidden: 'int') ->None:
super().__init__()
self.weight = pt.nn.Parameter(pt.Tensor(num_hidden))
def weights_from_mxnet_block(self, block_mx: "'LHUC'"):
self.weight.data[:] = pt.as_tensor(block_mx.weight.data().asnumpy())
def forward(self, input_0):
primals_1 = self.weight
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
| blchu/sockeye | PyTorchLHUC | false | 1,556 | [
"Apache-2.0"
] | 0 | 28044a44ee409c9b3df1711c0b16bdebdd463b2e | https://github.com/blchu/sockeye/tree/28044a44ee409c9b3df1711c0b16bdebdd463b2e | import torch
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
class Model(pt.nn.Module):
"""
Learning Hidden Unit Contribution
David Vilar. "Learning Hidden Unit Contribution for Adapting Neural
Machine Translation Models" NAACL 2018
:param num_hidden: Number of hidden units of the layer to be modified.
"""
def __init__(self, num_hidden: 'int') ->None:
super().__init__()
self.weight = pt.nn.Parameter(pt.Tensor(num_hidden))
def forward(self, data: 'pt.Tensor') ->pt.Tensor:
weight = 2 * pt.sigmoid(self.weight)
return weight * data
def weights_from_mxnet_block(self, block_mx: "'LHUC'"):
self.weight.data[:] = pt.as_tensor(block_mx.weight.data().asnumpy())
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
SE | # 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_6/inductor_cache/tu/ctuej2j6f3oxr5p43q7juhagc3r3ncgs2ikvxemutunlnxlnvl24.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# x => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [-1, -2, -3], 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=[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, 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 = 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)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 64.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/cv/ccvicmhiupo7cb3dwu3rzvk4zvi24nzhtwz7a5c4ke3qpmz3ofpe.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_2 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%squeeze,), 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 + (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_6/inductor_cache/az/cazousalzuqn73ciahz5izvogzu4ekcsktal4tthjvwjd3cqdayz.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_3 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%unsqueeze_1, %primals_4, %primals_5, [1, 1, 1], [0, 0, 0], [1, 1, 1], False, [0, 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=[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_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 = 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 + (x0), xmask)
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/72/c72kehajbo6zfhkuwjl3g6t24haqfzxumia5abs5c2hzebjb6ubo.py
# Topologically Sorted Source Nodes: [x_4, mul], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# x_4 => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%squeeze_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %sigmoid), kwargs = {})
triton_poi_fused_mul_sigmoid_3 = async_compile.triton('triton_poi_fused_mul_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: '*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_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_mul_sigmoid_3(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 // 64)
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, 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, (16, 4, 1, 1, 1), (4, 1, 1, 1, 1))
assert_size_stride(primals_3, (16, ), (1, ))
assert_size_stride(primals_4, (4, 16, 1, 1, 1), (16, 1, 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, 1, 1, 1), (1, 4, 4, 4), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 4, 64, grid=grid(4), stream=stream0)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 1, 1, 1), (0, 1, 0, 0, 0), 0), primals_2, stride=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False, output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf2, (1, 16, 1, 1, 1), (16, 1, 1, 1, 1))
buf3 = reinterpret_tensor(buf2, (16, 1, 1, 1), (1, 16, 16, 16), 0); del buf2 # reuse
buf7 = empty_strided_cuda((16, 1, 1, 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(buf3, primals_3, buf7, 16, grid=grid(16), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(reinterpret_tensor(buf3, (1, 16, 1, 1, 1), (0, 1, 0, 0, 0), 0), primals_4, stride=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False, output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf4, (1, 4, 1, 1, 1), (4, 1, 1, 1, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf5, primals_5, 4, grid=grid(4), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_4, mul], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_3.run(primals_1, buf5, buf6, 256, grid=grid(256), stream=stream0)
return (buf6, primals_1, primals_2, primals_4, reinterpret_tensor(buf1, (1, 4, 1, 1, 1), (4, 1, 1, 1, 1), 0), reinterpret_tensor(buf3, (1, 16, 1, 1, 1), (16, 1, 1, 1, 1), 0), buf5, 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, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, 4, 1, 1, 1), (4, 1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 16, 1, 1, 1), (16, 1, 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
from itertools import chain as chain
import torch.utils.data
import torch.nn as nn
class SwishEfficient(torch.autograd.Function):
"""Swish activation function: x * sigmoid(x)."""
@staticmethod
def forward(ctx, x):
result = x * torch.sigmoid(x)
ctx.save_for_backward(x)
return result
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_variables[0]
sigmoid_x = torch.sigmoid(x)
return grad_output * (sigmoid_x * (1 + x * (1 - sigmoid_x)))
class Swish(nn.Module):
"""Swish activation function: x * sigmoid(x)."""
def __init__(self):
super(Swish, self).__init__()
def forward(self, x):
return SwishEfficient.apply(x)
class SE(nn.Module):
"""Squeeze-and-Excitation (SE) block w/ Swish: AvgPool, FC, Swish, FC, Sigmoid."""
def _round_width(self, width, multiplier, min_width=8, divisor=8):
"""
Round width of filters based on width multiplier
Args:
width (int): the channel dimensions of the input.
multiplier (float): the multiplication factor.
min_width (int): the minimum width after multiplication.
divisor (int): the new width should be dividable by divisor.
"""
if not multiplier:
return width
width *= multiplier
min_width = min_width or divisor
width_out = max(min_width, int(width + divisor / 2) // divisor *
divisor)
if width_out < 0.9 * width:
width_out += divisor
return int(width_out)
def __init__(self, dim_in, ratio, relu_act=True):
"""
Args:
dim_in (int): the channel dimensions of the input.
ratio (float): the channel reduction ratio for squeeze.
relu_act (bool): whether to use ReLU activation instead
of Swish (default).
divisor (int): the new width should be dividable by divisor.
"""
super(SE, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
dim_fc = self._round_width(dim_in, ratio)
self.fc1 = nn.Conv3d(dim_in, dim_fc, 1, bias=True)
self.fc1_act = nn.ReLU() if relu_act else Swish()
self.fc2 = nn.Conv3d(dim_fc, dim_in, 1, bias=True)
self.fc2_sig = nn.Sigmoid()
def forward(self, x):
x_in = x
for module in self.children():
x = module(x)
return x_in * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'dim_in': 4, 'ratio': 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 itertools import chain as chain
import torch.utils.data
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_mean_0(in_out_ptr0, in_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)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 64.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_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 = 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 + 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_convolution_2(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 + x0, xmask)
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_mul_sigmoid_3(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 // 64
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, 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, (16, 4, 1, 1, 1), (4, 1, 1, 1, 1))
assert_size_stride(primals_3, (16,), (1,))
assert_size_stride(primals_4, (4, 16, 1, 1, 1), (16, 1, 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, 1, 1, 1), (1, 4, 4, 4), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(4)](buf1, primals_1, 4, 64, XBLOCK=1,
num_warps=2, num_stages=1)
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 1,
1, 1), (0, 1, 0, 0, 0), 0), primals_2, stride=(1, 1, 1),
padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False,
output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf2, (1, 16, 1, 1, 1), (16, 1, 1, 1, 1))
buf3 = reinterpret_tensor(buf2, (16, 1, 1, 1), (1, 16, 16, 16), 0)
del buf2
buf7 = empty_strided_cuda((16, 1, 1, 1), (1, 1, 1, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(16)](buf3,
primals_3, buf7, 16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf4 = extern_kernels.convolution(reinterpret_tensor(buf3, (1, 16,
1, 1, 1), (0, 1, 0, 0, 0), 0), primals_4, stride=(1, 1, 1),
padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False,
output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf4, (1, 4, 1, 1, 1), (4, 1, 1, 1, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_2[grid(4)](buf5, primals_5, 4, XBLOCK=
4, num_warps=1, num_stages=1)
del primals_5
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_3[grid(256)](primals_1, buf5, buf6,
256, XBLOCK=256, num_warps=4, num_stages=1)
return buf6, primals_1, primals_2, primals_4, reinterpret_tensor(buf1,
(1, 4, 1, 1, 1), (4, 1, 1, 1, 1), 0), reinterpret_tensor(buf3, (1,
16, 1, 1, 1), (16, 1, 1, 1, 1), 0), buf5, buf7
class SwishEfficient(torch.autograd.Function):
"""Swish activation function: x * sigmoid(x)."""
@staticmethod
def forward(ctx, x):
result = x * torch.sigmoid(x)
ctx.save_for_backward(x)
return result
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_variables[0]
sigmoid_x = torch.sigmoid(x)
return grad_output * (sigmoid_x * (1 + x * (1 - sigmoid_x)))
class Swish(nn.Module):
"""Swish activation function: x * sigmoid(x)."""
def __init__(self):
super(Swish, self).__init__()
def forward(self, x):
return SwishEfficient.apply(x)
class SENew(nn.Module):
"""Squeeze-and-Excitation (SE) block w/ Swish: AvgPool, FC, Swish, FC, Sigmoid."""
def _round_width(self, width, multiplier, min_width=8, divisor=8):
"""
Round width of filters based on width multiplier
Args:
width (int): the channel dimensions of the input.
multiplier (float): the multiplication factor.
min_width (int): the minimum width after multiplication.
divisor (int): the new width should be dividable by divisor.
"""
if not multiplier:
return width
width *= multiplier
min_width = min_width or divisor
width_out = max(min_width, int(width + divisor / 2) // divisor *
divisor)
if width_out < 0.9 * width:
width_out += divisor
return int(width_out)
def __init__(self, dim_in, ratio, relu_act=True):
"""
Args:
dim_in (int): the channel dimensions of the input.
ratio (float): the channel reduction ratio for squeeze.
relu_act (bool): whether to use ReLU activation instead
of Swish (default).
divisor (int): the new width should be dividable by divisor.
"""
super(SENew, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
dim_fc = self._round_width(dim_in, ratio)
self.fc1 = nn.Conv3d(dim_in, dim_fc, 1, bias=True)
self.fc1_act = nn.ReLU() if relu_act else Swish()
self.fc2 = nn.Conv3d(dim_fc, dim_in, 1, bias=True)
self.fc2_sig = nn.Sigmoid()
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]
| billcai/SlowFast | SE | false | 1,557 | [
"Apache-2.0"
] | 0 | 778888e63351e55861801996b37c7ff9a3746587 | https://github.com/billcai/SlowFast/tree/778888e63351e55861801996b37c7ff9a3746587 | import torch
from itertools import chain as chain
import torch.utils.data
import torch.nn as nn
class SwishEfficient(torch.autograd.Function):
"""Swish activation function: x * sigmoid(x)."""
@staticmethod
def forward(ctx, x):
result = x * torch.sigmoid(x)
ctx.save_for_backward(x)
return result
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_variables[0]
sigmoid_x = torch.sigmoid(x)
return grad_output * (sigmoid_x * (1 + x * (1 - sigmoid_x)))
class Swish(nn.Module):
"""Swish activation function: x * sigmoid(x)."""
def __init__(self):
super().__init__()
def forward(self, x):
return SwishEfficient.apply(x)
class Model(nn.Module):
"""Squeeze-and-Excitation (SE) block w/ Swish: AvgPool, FC, Swish, FC, Sigmoid."""
def _round_width(self, width, multiplier, min_width=8, divisor=8):
"""
Round width of filters based on width multiplier
Args:
width (int): the channel dimensions of the input.
multiplier (float): the multiplication factor.
min_width (int): the minimum width after multiplication.
divisor (int): the new width should be dividable by divisor.
"""
if not multiplier:
return width
width *= multiplier
min_width = min_width or divisor
width_out = max(min_width, int(width + divisor / 2) // divisor *
divisor)
if width_out < 0.9 * width:
width_out += divisor
return int(width_out)
def __init__(self, dim_in, ratio, relu_act=True):
"""
Args:
dim_in (int): the channel dimensions of the input.
ratio (float): the channel reduction ratio for squeeze.
relu_act (bool): whether to use ReLU activation instead
of Swish (default).
divisor (int): the new width should be dividable by divisor.
"""
super().__init__()
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
dim_fc = self._round_width(dim_in, ratio)
self.fc1 = nn.Conv3d(dim_in, dim_fc, 1, bias=True)
self.fc1_act = nn.ReLU() if relu_act else Swish()
self.fc2 = nn.Conv3d(dim_fc, dim_in, 1, bias=True)
self.fc2_sig = nn.Sigmoid()
def forward(self, x):
x_in = x
for module in self.children():
x = module(x)
return x_in * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
SimpleNormLayer | # 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_6/inductor_cache/db/cdbvpwij3es3hvx6e56eufnhc2ark7ffetlr353jjxbefbpe6ws4.py
# Topologically Sorted Source Nodes: [truediv], Original ATen: [aten.div]
# Source node to ATen node mapping:
# truediv => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, 4), kwargs = {})
triton_poi_fused_div_0 = async_compile.triton('triton_poi_fused_div_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_div_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_div_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 = 0.25
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: [truediv], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_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
class SimpleNormLayer(nn.Module):
"""Simple normalization layer that divides the output of a
preceding layer by a specified number
Parameters
----------
normalization_strength: float
The number with which input is normalized/dived by
"""
def __init__(self, normalization_strength):
super(SimpleNormLayer, self).__init__()
self.normalization_strength = normalization_strength
def forward(self, input_features):
"""Computes normalized output
Parameters
----------
input_features: torch.Tensor
Input tensor of featuers of any shape
Returns
-------
normalized_features: torch.Tensor
Normalized input features
"""
return input_features / self.normalization_strength
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'normalization_strength': 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_div_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 = 0.25
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_div_0[grid(256)](arg0_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class SimpleNormLayerNew(nn.Module):
"""Simple normalization layer that divides the output of a
preceding layer by a specified number
Parameters
----------
normalization_strength: float
The number with which input is normalized/dived by
"""
def __init__(self, normalization_strength):
super(SimpleNormLayerNew, self).__init__()
self.normalization_strength = normalization_strength
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bokutotu/cgnet | SimpleNormLayer | false | 1,558 | [
"BSD-3-Clause"
] | 0 | a35170001d969d51548dd01522b1ab93e43741b4 | https://github.com/bokutotu/cgnet/tree/a35170001d969d51548dd01522b1ab93e43741b4 | import torch
import torch.nn as nn
class Model(nn.Module):
"""Simple normalization layer that divides the output of a
preceding layer by a specified number
Parameters
----------
normalization_strength: float
The number with which input is normalized/dived by
"""
def __init__(self, normalization_strength):
super().__init__()
self.normalization_strength = normalization_strength
def forward(self, input_features):
"""Computes normalized output
Parameters
----------
input_features: torch.Tensor
Input tensor of featuers of any shape
Returns
-------
normalized_features: torch.Tensor
Normalized input features
"""
return input_features / self.normalization_strength
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
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_6/inductor_cache/cc/ccc3rfqzulhvn4nmibqhwjsomjzanxwwc34nr7d7enltyb7czw67.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=[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_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_relu_threshold_backward_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 32000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 500
x2 = (xindex // 2000)
x3 = xindex % 2000
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 + (2016*x2)), tmp4, xmask)
tl.store(out_ptr1 + (x3 + (2048*x2)), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/od/cod4chcodxhfkpwgvalpo3wjcotr3qr3momwpd2myrvazuq2fy67.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.relu, aten.view]
# Source node to ATen node mapping:
# x => relu
# x_1 => view_2
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %view_2 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%relu, [64, 500]), kwargs = {})
triton_poi_fused_relu_view_1 = async_compile.triton('triton_poi_fused_relu_view_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_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_relu_view_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 500
x1 = (xindex // 500)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (500*(x1 % 4)) + (2016*(x1 // 4))), xmask)
tl.store(out_ptr0 + (x2), 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 = args
args.clear()
assert_size_stride(primals_1, (500, 4), (4, 1))
assert_size_stride(primals_2, (500, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (20, 500), (500, 1))
assert_size_stride(primals_5, (20, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 500), (500, 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, 500), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 500), (8064, 2016, 500, 1), torch.float32)
buf4 = empty_strided_cuda((4, 4, 4, 500), (8192, 2048, 500, 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(buf0, primals_2, buf1, buf4, 32000, grid=grid(32000), stream=stream0)
del primals_2
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.relu, aten.view]
triton_poi_fused_relu_view_1.run(buf1, buf2, 32000, grid=grid(32000), stream=stream0)
del buf1
buf3 = empty_strided_cuda((64, 20), (20, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, buf2, reinterpret_tensor(primals_4, (500, 20), (1, 500), 0), alpha=1, beta=1, out=buf3)
del primals_5
return (reinterpret_tensor(buf3, (4, 4, 4, 20), (320, 80, 20, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf2, 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((500, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((500, ), (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((20, 500), (500, 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
import torch.nn as nn
import torch.nn.functional as F
class Net(nn.Module):
def __init__(self, feature_num):
super(Net, self).__init__()
self.layer_1 = nn.Linear(feature_num, 500)
self.layer_2 = nn.Linear(500, 20)
def forward(self, x):
x = F.relu(self.layer_1(x))
x = self.layer_2(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'feature_num': 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_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 32000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 500
x2 = xindex // 2000
x3 = xindex % 2000
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 + 2016 * x2), tmp4, xmask)
tl.store(out_ptr1 + (x3 + 2048 * x2), tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_view_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 32000
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 500
x1 = xindex // 500
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 500 * (x1 % 4) + 2016 * (x1 // 4)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (500, 4), (4, 1))
assert_size_stride(primals_2, (500,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (20, 500), (500, 1))
assert_size_stride(primals_5, (20,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 500), (500, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 500), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 500), (8064, 2016, 500, 1),
torch.float32)
buf4 = empty_strided_cuda((4, 4, 4, 500), (8192, 2048, 500, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(32000)](buf0,
primals_2, buf1, buf4, 32000, XBLOCK=128, num_warps=4, num_stages=1
)
del primals_2
buf2 = buf0
del buf0
triton_poi_fused_relu_view_1[grid(32000)](buf1, buf2, 32000, XBLOCK
=256, num_warps=4, num_stages=1)
del buf1
buf3 = empty_strided_cuda((64, 20), (20, 1), torch.float32)
extern_kernels.addmm(primals_5, buf2, reinterpret_tensor(primals_4,
(500, 20), (1, 500), 0), alpha=1, beta=1, out=buf3)
del primals_5
return reinterpret_tensor(buf3, (4, 4, 4, 20), (320, 80, 20, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf2, primals_4, buf4
class NetNew(nn.Module):
def __init__(self, feature_num):
super(NetNew, self).__init__()
self.layer_1 = nn.Linear(feature_num, 500)
self.layer_2 = nn.Linear(500, 20)
def forward(self, input_0):
primals_1 = self.layer_1.weight
primals_2 = self.layer_1.bias
primals_4 = self.layer_2.weight
primals_5 = self.layer_2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| bm2-lab/MDML | Net | false | 1,559 | [
"MIT"
] | 0 | 222fb22b2ee53dd3c1a6f2e99a88f71e9635e3a0 | https://github.com/bm2-lab/MDML/tree/222fb22b2ee53dd3c1a6f2e99a88f71e9635e3a0 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, feature_num):
super().__init__()
self.layer_1 = nn.Linear(feature_num, 500)
self.layer_2 = nn.Linear(500, 20)
def forward(self, x):
x = F.relu(self.layer_1(x))
x = self.layer_2(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
PyTorchSSRU | # 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_6/inductor_cache/x2/cx2opldmv23p224bzsdeqf4rnb3hnj2sia22rfsabdb42yyrrphc.py
# Topologically Sorted Source Nodes: [forget_rates, sub, weighted_inputs, mul_1, new_step_state, relu], Original ATen: [aten.sigmoid, aten.rsub, aten.mul, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# forget_rates => sigmoid
# mul_1 => mul_1
# new_step_state => add
# relu => relu
# sub => sub
# weighted_inputs => mul
# Graph fragment:
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_1,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %view_3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %primals_5), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %mul), 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_mul_relu_rsub_sigmoid_threshold_backward_0 = async_compile.triton('triton_poi_fused_add_mul_relu_rsub_sigmoid_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: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*i1', 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_relu_rsub_sigmoid_threshold_backward_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_relu_rsub_sigmoid_threshold_backward_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, 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)
tmp2 = tl.load(in_ptr1 + (x0), xmask)
tmp6 = tl.load(in_ptr2 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = 1.0
tmp5 = tmp4 - tmp1
tmp7 = tmp5 * tmp6
tmp8 = tmp3 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tmp11 = 0.0
tmp12 = tmp10 <= tmp11
tl.store(out_ptr0 + (x0), tmp8, xmask)
tl.store(out_ptr1 + (x0), tmp10, xmask)
tl.store(out_ptr2 + (x0), tmp12, 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, 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: [linear], 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((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf3 = 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: [forget_rates, sub, weighted_inputs, mul_1, new_step_state, relu], Original ATen: [aten.sigmoid, aten.rsub, aten.mul, aten.add, aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_relu_rsub_sigmoid_threshold_backward_0.run(buf0, primals_5, buf1, buf2, buf3, buf4, 256, grid=grid(256), stream=stream0)
return (buf3, buf2, primals_5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf0, buf1, 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, 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])
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 typing import Tuple
from abc import abstractmethod
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
class AutoregressiveLayer(pt.nn.Module):
@property
@abstractmethod
def num_state_tensors(self) ->int:
""" Number of state tensors returned by the layer """
raise NotImplementedError
@property
@abstractmethod
def needs_mask(self) ->bool:
""" Whether the layer makes use of a mask tensor or not """
raise NotImplementedError
@abstractmethod
def get_state_shape(self, batch_size) ->Tuple:
"""
:param batch_size: current batch size
:return: dimensions of each output state (assuming all of them have the same shape)
"""
raise NotImplementedError
@abstractmethod
def forward(self, inputs: 'pt.Tensor', previous_states: 'pt.Tensor', *args
) ->Tuple:
"""
:param inputs: layer input
:param previous_states: Previous states array or list of arrays
:param args: layer-specific arguments and/or arguments to be ignored
:return: layer output and new states
"""
raise NotImplementedError
class PyTorchSSRU(AutoregressiveLayer):
"""
Simpler Simple Recurrent Unit
Kim et al, "From Research to Production and Back: Ludicrously Fast Neural Machine Translation" WNGT 2019
Variant of an LSTM cell aimed at reducing computational dependency across time steps.
Formally described as:
(1) f[t] = sigmoid(W1[t] * x[t] + b[t])
(2) c[t] = f[t] . c[t-1] + (1 - f[t]) . W2[t] * x[t]
(3) h = ReLU(c[t])
where:
. represents elementwise multiplication;
x[t] is the input at time step t;
f[t] is the output of the forget gate at time step t;
c[t] is the cell state at time step t;
h is the output of the unit.
:param model_size: number of hidden units
:param inference_only: flag used to indicate execution at inference time
"""
def __init__(self, model_size: 'int', inference_only: 'bool') ->None:
super().__init__()
self.model_size = model_size
self.inference_only = inference_only
self.cell_state_transform = (self._inference_cell_state_transform if
inference_only else self._training_cell_state_transform)
self.forget_gate = pt.nn.Linear(in_features=model_size,
out_features=model_size, bias=True)
self.forget_gate_act = pt.nn.Sigmoid()
self.linear = pt.nn.Linear(in_features=model_size, out_features=
model_size, bias=False)
self.relu = pt.nn.ReLU(inplace=False)
@property
def num_state_tensors(self) ->int:
""" Number of state tensors returned by the layer """
return 1
@property
def needs_mask(self) ->bool:
""" Whether the layer makes use of a mask tensor or not """
return False
def get_state_shape(self, batch_size: 'int') ->Tuple:
"""
:param batch_size: current batch size
:return: dimensions of each output state (assuming all of them have the same shape)
"""
return 1, batch_size, self.model_size
@staticmethod
@pt.jit.script_if_tracing
def _training_cell_state_transform(previous_cell_state, weighted_inputs,
forget_rates) ->Tuple[pt.Tensor, pt.Tensor]:
"""Update SSRU cell at training time"""
steps = weighted_inputs.size()[0]
cell_state = previous_cell_state.squeeze(0)
states = []
for t in range(steps):
cell_state = forget_rates[t, :, :] * cell_state + weighted_inputs[
t, :, :]
states.append(cell_state)
states = pt.stack(states, dim=0)
return states, cell_state.unsqueeze(0)
@staticmethod
def _inference_cell_state_transform(previous_cell_state,
weighted_inputs, forget_rates) ->Tuple[pt.Tensor, pt.Tensor]:
"""Update SSRU cell at inference time"""
new_step_state = forget_rates * previous_cell_state + weighted_inputs
return new_step_state, new_step_state
def forward(self, inputs: 'pt.Tensor', previous_states: 'pt.Tensor', **args
) ->Tuple[pt.Tensor, pt.Tensor]:
"""
:param inputs: input data. Shape: (max_length, batch, input_depth).
:param previous_states: previous cell states. Shape: (max_length, batch, input_depth)
:return: cell output and new cell states. Both with shape (max_length, batch, input_depth).
"""
forget_rates = self.forget_gate_act(self.forget_gate(inputs))
weighted_inputs = (1 - forget_rates) * self.linear(inputs)
cell_state, last_step_state = self.cell_state_transform(previous_states
, weighted_inputs, forget_rates)
return self.relu(cell_state), last_step_state
def weights_from_mxnet_block(self, block_mx: "'SSRU'"):
self.forget_gate.weight.data[:] = pt.as_tensor(block_mx.forget_gate
.weight.data().asnumpy())
self.forget_gate.bias.data[:] = pt.as_tensor(block_mx.forget_gate.
bias.data().asnumpy())
self.linear.weight.data[:] = pt.as_tensor(block_mx.linear.weight.
data().asnumpy())
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'model_size': 4, 'inference_only': 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 typing import Tuple
from abc import abstractmethod
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
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_relu_rsub_sigmoid_threshold_backward_0(in_ptr0,
in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, 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)
tmp2 = tl.load(in_ptr1 + x0, xmask)
tmp6 = tl.load(in_ptr2 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = 1.0
tmp5 = tmp4 - tmp1
tmp7 = tmp5 * tmp6
tmp8 = tmp3 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tmp11 = 0.0
tmp12 = tmp10 <= tmp11
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp10, xmask)
tl.store(out_ptr2 + x0, tmp12, 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, 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_1
del primals_2
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf3 = 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_add_mul_relu_rsub_sigmoid_threshold_backward_0[grid
(256)](buf0, primals_5, buf1, buf2, buf3, buf4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
return buf3, buf2, primals_5, reinterpret_tensor(primals_3, (64, 4), (4,
1), 0), buf0, buf1, buf4
class AutoregressiveLayer(pt.nn.Module):
@property
@abstractmethod
def num_state_tensors(self) ->int:
""" Number of state tensors returned by the layer """
raise NotImplementedError
@property
@abstractmethod
def needs_mask(self) ->bool:
""" Whether the layer makes use of a mask tensor or not """
raise NotImplementedError
@abstractmethod
def get_state_shape(self, batch_size) ->Tuple:
"""
:param batch_size: current batch size
:return: dimensions of each output state (assuming all of them have the same shape)
"""
raise NotImplementedError
@abstractmethod
def forward(self, inputs: 'pt.Tensor', previous_states: 'pt.Tensor', *args
) ->Tuple:
"""
:param inputs: layer input
:param previous_states: Previous states array or list of arrays
:param args: layer-specific arguments and/or arguments to be ignored
:return: layer output and new states
"""
raise NotImplementedError
class PyTorchSSRUNew(AutoregressiveLayer):
"""
Simpler Simple Recurrent Unit
Kim et al, "From Research to Production and Back: Ludicrously Fast Neural Machine Translation" WNGT 2019
Variant of an LSTM cell aimed at reducing computational dependency across time steps.
Formally described as:
(1) f[t] = sigmoid(W1[t] * x[t] + b[t])
(2) c[t] = f[t] . c[t-1] + (1 - f[t]) . W2[t] * x[t]
(3) h = ReLU(c[t])
where:
. represents elementwise multiplication;
x[t] is the input at time step t;
f[t] is the output of the forget gate at time step t;
c[t] is the cell state at time step t;
h is the output of the unit.
:param model_size: number of hidden units
:param inference_only: flag used to indicate execution at inference time
"""
def __init__(self, model_size: 'int', inference_only: 'bool') ->None:
super().__init__()
self.model_size = model_size
self.inference_only = inference_only
self.cell_state_transform = (self._inference_cell_state_transform if
inference_only else self._training_cell_state_transform)
self.forget_gate = pt.nn.Linear(in_features=model_size,
out_features=model_size, bias=True)
self.forget_gate_act = pt.nn.Sigmoid()
self.linear = pt.nn.Linear(in_features=model_size, out_features=
model_size, bias=False)
self.relu = pt.nn.ReLU(inplace=False)
@property
def num_state_tensors(self) ->int:
""" Number of state tensors returned by the layer """
return 1
@property
def needs_mask(self) ->bool:
""" Whether the layer makes use of a mask tensor or not """
return False
def get_state_shape(self, batch_size: 'int') ->Tuple:
"""
:param batch_size: current batch size
:return: dimensions of each output state (assuming all of them have the same shape)
"""
return 1, batch_size, self.model_size
@staticmethod
@pt.jit.script_if_tracing
def _training_cell_state_transform(previous_cell_state, weighted_inputs,
forget_rates) ->Tuple[pt.Tensor, pt.Tensor]:
"""Update SSRU cell at training time"""
steps = weighted_inputs.size()[0]
cell_state = previous_cell_state.squeeze(0)
states = []
for t in range(steps):
cell_state = forget_rates[t, :, :] * cell_state + weighted_inputs[
t, :, :]
states.append(cell_state)
states = pt.stack(states, dim=0)
return states, cell_state.unsqueeze(0)
@staticmethod
def _inference_cell_state_transform(previous_cell_state,
weighted_inputs, forget_rates) ->Tuple[pt.Tensor, pt.Tensor]:
"""Update SSRU cell at inference time"""
new_step_state = forget_rates * previous_cell_state + weighted_inputs
return new_step_state, new_step_state
def weights_from_mxnet_block(self, block_mx: "'SSRU'"):
self.forget_gate.weight.data[:] = pt.as_tensor(block_mx.forget_gate
.weight.data().asnumpy())
self.forget_gate.bias.data[:] = pt.as_tensor(block_mx.forget_gate.
bias.data().asnumpy())
self.linear.weight.data[:] = pt.as_tensor(block_mx.linear.weight.
data().asnumpy())
def forward(self, input_0, input_1):
primals_1 = self.forget_gate.weight
primals_2 = self.forget_gate.bias
primals_4 = self.linear.weight
primals_3 = input_0
primals_5 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0], output[1]
| blchu/sockeye | PyTorchSSRU | false | 1,560 | [
"Apache-2.0"
] | 0 | 28044a44ee409c9b3df1711c0b16bdebdd463b2e | https://github.com/blchu/sockeye/tree/28044a44ee409c9b3df1711c0b16bdebdd463b2e | import torch
from typing import Tuple
from abc import abstractmethod
import torch as pt
import torch.distributed
import torch.distributed.elastic.multiprocessing.errors
class AutoregressiveLayer(pt.nn.Module):
@property
@abstractmethod
def num_state_tensors(self) ->int:
""" Number of state tensors returned by the layer """
raise NotImplementedError
@property
@abstractmethod
def needs_mask(self) ->bool:
""" Whether the layer makes use of a mask tensor or not """
raise NotImplementedError
@abstractmethod
def get_state_shape(self, batch_size) ->Tuple:
"""
:param batch_size: current batch size
:return: dimensions of each output state (assuming all of them have the same shape)
"""
raise NotImplementedError
@abstractmethod
def forward(self, inputs: 'pt.Tensor', previous_states: 'pt.Tensor', *args
) ->Tuple:
"""
:param inputs: layer input
:param previous_states: Previous states array or list of arrays
:param args: layer-specific arguments and/or arguments to be ignored
:return: layer output and new states
"""
raise NotImplementedError
class Model(AutoregressiveLayer):
"""
Simpler Simple Recurrent Unit
Kim et al, "From Research to Production and Back: Ludicrously Fast Neural Machine Translation" WNGT 2019
Variant of an LSTM cell aimed at reducing computational dependency across time steps.
Formally described as:
(1) f[t] = sigmoid(W1[t] * x[t] + b[t])
(2) c[t] = f[t] . c[t-1] + (1 - f[t]) . W2[t] * x[t]
(3) h = ReLU(c[t])
where:
. represents elementwise multiplication;
x[t] is the input at time step t;
f[t] is the output of the forget gate at time step t;
c[t] is the cell state at time step t;
h is the output of the unit.
:param model_size: number of hidden units
:param inference_only: flag used to indicate execution at inference time
"""
def __init__(self, model_size: 'int', inference_only: 'bool') ->None:
super().__init__()
self.model_size = model_size
self.inference_only = inference_only
self.cell_state_transform = (self._inference_cell_state_transform if
inference_only else self._training_cell_state_transform)
self.forget_gate = pt.nn.Linear(in_features=model_size,
out_features=model_size, bias=True)
self.forget_gate_act = pt.nn.Sigmoid()
self.linear = pt.nn.Linear(in_features=model_size, out_features=
model_size, bias=False)
self.relu = pt.nn.ReLU(inplace=False)
@property
def num_state_tensors(self) ->int:
""" Number of state tensors returned by the layer """
return 1
@property
def needs_mask(self) ->bool:
""" Whether the layer makes use of a mask tensor or not """
return False
def get_state_shape(self, batch_size: 'int') ->Tuple:
"""
:param batch_size: current batch size
:return: dimensions of each output state (assuming all of them have the same shape)
"""
return 1, batch_size, self.model_size
@staticmethod
@pt.jit.script_if_tracing
def _training_cell_state_transform(previous_cell_state, weighted_inputs,
forget_rates) ->Tuple[pt.Tensor, pt.Tensor]:
"""Update SSRU cell at training time"""
steps = weighted_inputs.size()[0]
cell_state = previous_cell_state.squeeze(0)
states = []
for t in range(steps):
cell_state = forget_rates[t, :, :] * cell_state + weighted_inputs[
t, :, :]
states.append(cell_state)
states = pt.stack(states, dim=0)
return states, cell_state.unsqueeze(0)
@staticmethod
def _inference_cell_state_transform(previous_cell_state,
weighted_inputs, forget_rates) ->Tuple[pt.Tensor, pt.Tensor]:
"""Update SSRU cell a
# ... truncated (>4000 chars) for memory efficiency |
ShiftedSoftplus | # 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_6/inductor_cache/im/cimgqeuv3wxx43tl5inlewtmlhgsssqq5ly4qz44xnlpnnfsiltd.py
# Topologically Sorted Source Nodes: [softplus, wrapped_log, sub], Original ATen: [aten.softplus, aten.log, aten.sub]
# Source node to ATen node mapping:
# softplus => exp, gt, log1p, where
# sub => sub
# wrapped_log => full_default
# 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 = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.6931471805599453), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where, %full_default), kwargs = {})
triton_poi_fused_log_softplus_sub_0 = async_compile.triton('triton_poi_fused_log_softplus_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_log_softplus_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_log_softplus_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 = 20.0
tmp2 = tmp0 > tmp1
tmp3 = tl_math.exp(tmp0)
tmp4 = libdevice.log1p(tmp3)
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = 0.6931471805599453
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, wrapped_log, sub], Original ATen: [aten.softplus, aten.log, aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_log_softplus_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
import numpy as np
import torch.nn as nn
class ShiftedSoftplus(nn.Module):
""" Shifted softplus (SSP) activation function
SSP originates from the softplus function:
y = \\ln\\left(1 + e^{-x}\\right)
Schütt et al. (2018) introduced a shifting factor to the function in order
to ensure that SSP(0) = 0 while having infinite order of continuity:
y = \\ln\\left(1 + e^{-x}\\right) - \\ln(2)
SSP allows to obtain smooth potential energy surfaces and second derivatives
that are required for training with forces as well as the calculation of
vibrational modes (Schütt et al. 2018).
References
----------
K.T. Schütt. P.-J. Kindermans, H. E. Sauceda, S. Chmiela,
A. Tkatchenko, K.-R. Müller. (2018)
SchNet - a deep learning architecture for molecules and materials.
The Journal of Chemical Physics.
https://doi.org/10.1063/1.5019779
"""
def __init__(self):
super(ShiftedSoftplus, self).__init__()
def forward(self, input_tensor):
""" Applies the shifted softplus function element-wise
Parameters
----------
input_tensor: torch.Tensor
Input tensor of size (n_examples, *) where `*` means, any number of
additional dimensions.
Returns
-------
Output: torch.Tensor
Same size (n_examples, *) as the input.
"""
return nn.functional.softplus(input_tensor) - np.log(2.0)
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.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_log_softplus_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 = 20.0
tmp2 = tmp0 > tmp1
tmp3 = tl_math.exp(tmp0)
tmp4 = libdevice.log1p(tmp3)
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = 0.6931471805599453
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_log_softplus_sub_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class ShiftedSoftplusNew(nn.Module):
""" Shifted softplus (SSP) activation function
SSP originates from the softplus function:
y = \\ln\\left(1 + e^{-x}\\right)
Schütt et al. (2018) introduced a shifting factor to the function in order
to ensure that SSP(0) = 0 while having infinite order of continuity:
y = \\ln\\left(1 + e^{-x}\\right) - \\ln(2)
SSP allows to obtain smooth potential energy surfaces and second derivatives
that are required for training with forces as well as the calculation of
vibrational modes (Schütt et al. 2018).
References
----------
K.T. Schütt. P.-J. Kindermans, H. E. Sauceda, S. Chmiela,
A. Tkatchenko, K.-R. Müller. (2018)
SchNet - a deep learning architecture for molecules and materials.
The Journal of Chemical Physics.
https://doi.org/10.1063/1.5019779
"""
def __init__(self):
super(ShiftedSoftplusNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bokutotu/cgnet | ShiftedSoftplus | false | 1,561 | [
"BSD-3-Clause"
] | 0 | a35170001d969d51548dd01522b1ab93e43741b4 | https://github.com/bokutotu/cgnet/tree/a35170001d969d51548dd01522b1ab93e43741b4 | import torch
import numpy as np
import torch.nn as nn
class Model(nn.Module):
""" Shifted softplus (SSP) activation function
SSP originates from the softplus function:
y = \\ln\\left(1 + e^{-x}\\right)
Schütt et al. (2018) introduced a shifting factor to the function in order
to ensure that SSP(0) = 0 while having infinite order of continuity:
y = \\ln\\left(1 + e^{-x}\\right) - \\ln(2)
SSP allows to obtain smooth potential energy surfaces and second derivatives
that are required for training with forces as well as the calculation of
vibrational modes (Schütt et al. 2018).
References
----------
K.T. Schütt. P.-J. Kindermans, H. E. Sauceda, S. Chmiela,
A. Tkatchenko, K.-R. Müller. (2018)
SchNet - a deep learning architecture for molecules and materials.
The Journal of Chemical Physics.
https://doi.org/10.1063/1.5019779
"""
def __init__(self):
super().__init__()
def forward(self, input_tensor):
""" Applies the shifted softplus function element-wise
Parameters
----------
input_tensor: torch.Tensor
Input tensor of size (n_examples, *) where `*` means, any number of
additional dimensions.
Returns
-------
Output: torch.Tensor
Same size (n_examples, *) as the input.
"""
return nn.functional.softplus(input_tensor) - np.log(2.0)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
PatchEmbed | # 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_6/inductor_cache/jg/cjgafsignr6eltwpgfdtyyamm7z2oofx6jlesakdp45oail3wyp7.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 4, 4], [1, 7, 7], [1, 1, 1], False, [0, 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=[67108864],
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 = 51904512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 16896) % 768
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 = args
args.clear()
assert_size_stride(primals_1, (768, 3, 1, 16, 16), (768, 256, 256, 16, 1))
assert_size_stride(primals_2, (768, ), (1, ))
assert_size_stride(primals_3, (4, 3, 64, 64, 64), (786432, 262144, 4096, 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, 4, 4), padding=(1, 7, 7), dilation=(1, 1, 1), transposed=False, output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 768, 66, 16, 16), (12976128, 16896, 256, 16, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_2, 51904512, grid=grid(51904512), stream=stream0)
del primals_2
return (reinterpret_tensor(buf1, (4, 16896, 768), (12976128, 1, 16896), 0), 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((768, 3, 1, 16, 16), (768, 256, 256, 16, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((768, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 64, 64, 64), (786432, 262144, 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
from itertools import chain as chain
import torch.utils.data
import torch.nn as nn
class PatchEmbed(nn.Module):
"""
PatchEmbed.
"""
def __init__(self, dim_in=3, dim_out=768, kernel=(1, 16, 16), stride=(1,
4, 4), padding=(1, 7, 7), conv_2d=False):
super().__init__()
if conv_2d:
conv = nn.Conv2d
else:
conv = nn.Conv3d
self.proj = conv(dim_in, dim_out, kernel_size=kernel, stride=stride,
padding=padding)
def forward(self, x):
x = self.proj(x)
return x.flatten(2).transpose(1, 2)
def get_inputs():
return [torch.rand([4, 3, 64, 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 itertools import chain as chain
import torch.utils.data
import torch.nn as 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):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 16896 % 768
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 = args
args.clear()
assert_size_stride(primals_1, (768, 3, 1, 16, 16), (768, 256, 256, 16, 1))
assert_size_stride(primals_2, (768,), (1,))
assert_size_stride(primals_3, (4, 3, 64, 64, 64), (786432, 262144, 4096,
64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
4, 4), padding=(1, 7, 7), dilation=(1, 1, 1), transposed=False,
output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 768, 66, 16, 16), (12976128, 16896,
256, 16, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(51904512)](buf1, primals_2,
51904512, XBLOCK=512, num_warps=8, num_stages=1)
del primals_2
return reinterpret_tensor(buf1, (4, 16896, 768), (12976128, 1, 16896), 0
), primals_1, primals_3
class PatchEmbedNew(nn.Module):
"""
PatchEmbed.
"""
def __init__(self, dim_in=3, dim_out=768, kernel=(1, 16, 16), stride=(1,
4, 4), padding=(1, 7, 7), conv_2d=False):
super().__init__()
if conv_2d:
conv = nn.Conv2d
else:
conv = nn.Conv3d
self.proj = conv(dim_in, dim_out, kernel_size=kernel, stride=stride,
padding=padding)
def forward(self, input_0):
primals_1 = self.proj.weight
primals_2 = self.proj.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| billcai/SlowFast | PatchEmbed | false | 1,562 | [
"Apache-2.0"
] | 0 | 778888e63351e55861801996b37c7ff9a3746587 | https://github.com/billcai/SlowFast/tree/778888e63351e55861801996b37c7ff9a3746587 | import torch
from itertools import chain as chain
import torch.utils.data
import torch.nn as nn
class Model(nn.Module):
"""
PatchEmbed.
"""
def __init__(self, dim_in=3, dim_out=768, kernel=(1, 16, 16), stride=(1,
4, 4), padding=(1, 7, 7), conv_2d=False):
super().__init__()
if conv_2d:
conv = nn.Conv2d
else:
conv = nn.Conv3d
self.proj = conv(dim_in, dim_out, kernel_size=kernel, stride=stride,
padding=padding)
def forward(self, x):
x = self.proj(x)
return x.flatten(2).transpose(1, 2)
def get_inputs():
return [torch.rand([4, 3, 64, 64, 64])]
def get_init_inputs():
return []
|
NeighborNormLayer | # 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_6/inductor_cache/i3/ci3wf6sq5envi5qqy36dg5xyilb5fmbazlijbvgfe3ony6kt56ir.py
# Topologically Sorted Source Nodes: [truediv], Original ATen: [aten.div]
# Source node to ATen node mapping:
# truediv => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
triton_poi_fused_div_0 = async_compile.triton('triton_poi_fused_div_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_div_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_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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: [truediv], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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 NeighborNormLayer(nn.Module):
"""Normalization layer that divides the output of a
preceding layer by the number of neighbor features.
Unlike the SimpleNormLayer, this layer allows for
dynamically changing number of neighbors during training.
"""
def __init__(self):
super(NeighborNormLayer, self).__init__()
def forward(self, input_features, n_neighbors):
"""Computes normalized output
Parameters
----------
input_features: torch.Tensor
Input tensor of featuers of shape
(n_frames, n_beads, n_feats)
n_neighbors: int
the number of neighbors
Returns
-------
normalized_features: torch.Tensor
Normalized input features
"""
return input_features / n_neighbors
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
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_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 / tmp1
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_div_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK
=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class NeighborNormLayerNew(nn.Module):
"""Normalization layer that divides the output of a
preceding layer by the number of neighbor features.
Unlike the SimpleNormLayer, this layer allows for
dynamically changing number of neighbors during training.
"""
def __init__(self):
super(NeighborNormLayerNew, 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]
| bokutotu/cgnet | NeighborNormLayer | false | 1,563 | [
"BSD-3-Clause"
] | 0 | a35170001d969d51548dd01522b1ab93e43741b4 | https://github.com/bokutotu/cgnet/tree/a35170001d969d51548dd01522b1ab93e43741b4 | import torch
import torch.nn as nn
class Model(nn.Module):
"""Normalization layer that divides the output of a
preceding layer by the number of neighbor features.
Unlike the SimpleNormLayer, this layer allows for
dynamically changing number of neighbors during training.
"""
def __init__(self):
super().__init__()
def forward(self, input_features, n_neighbors):
"""Computes normalized output
Parameters
----------
input_features: torch.Tensor
Input tensor of featuers of shape
(n_frames, n_beads, n_feats)
n_neighbors: int
the number of neighbors
Returns
-------
normalized_features: torch.Tensor
Normalized input features
"""
return input_features / n_neighbors
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MultiheadSimilarity | # 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_6/inductor_cache/7o/c7o6kpxam7ua7xq5mlr7dszko5bkolv2zeooi3cve5lrg5vaqdfs.py
# Topologically Sorted Source Nodes: [kv, k], Original ATen: [aten.add, aten.clone]
# Source node to ATen node mapping:
# k => clone
# kv => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%permute_1, %primals_4), kwargs = {})
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%add,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_add_clone_0 = async_compile.triton('triton_poi_fused_add_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: '*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_clone_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_clone_0(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 % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16)
x3 = (xindex // 4)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(out_ptr0 + (x4), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/7b/c7bf34fgn2dhohe7ejneqlees25vyq6sbe4c5lfvoehzliak2nz6.py
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.add]
# Source node to ATen node mapping:
# k => add_1
# Graph fragment:
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_8), 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=[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_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 = 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
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/la/clahfcfxag6dxo4xoucuhnybhz4wwfcawa36iswhy7nt6vkbs2gq.py
# Topologically Sorted Source Nodes: [flatten], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# flatten => clone_2
# Graph fragment:
# %clone_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%view_8,), 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=[16, 4], 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, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_2', '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_clone_2(in_ptr0, in_ptr1, in_ptr2, 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
y3 = yindex
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (y3 + (16*x2)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x2 + (4*y3)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = 1.0
tmp5 = tmp3 * tmp4
tmp6 = tmp2 * tmp5
tl.store(out_ptr0 + (x2 + (4*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 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (16, 4), (4, 1))
assert_size_stride(primals_3, (16, ), (1, ))
assert_size_stride(primals_4, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (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, 16), (16, 1))
assert_size_stride(primals_12, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(primals_1, reinterpret_tensor(primals_2, (4, 16), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [q], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_6, primals_1, reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_5
del primals_6
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [kv, k], Original ATen: [aten.add, aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_add_clone_0.run(buf0, primals_3, primals_4, buf2, 64, grid=grid(64), stream=stream0)
del primals_3
del primals_4
buf3 = reinterpret_tensor(buf0, (16, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf3)
buf4 = reinterpret_tensor(buf3, (4, 4, 4), (16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.add]
triton_poi_fused_add_1.run(buf4, primals_8, 64, grid=grid(64), stream=stream0)
del primals_8
buf5 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [bmm], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 1), (1, 16, 0), 0), reinterpret_tensor(buf1, (16, 1, 1), (1, 1, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [v], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf6)
buf7 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [flatten], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf6, primals_10, buf5, buf7, 16, 4, grid=grid(16, 4), stream=stream0)
buf8 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_12, reinterpret_tensor(buf7, (4, 16), (16, 1), 0), reinterpret_tensor(primals_11, (16, 4), (1, 16), 0), alpha=1, beta=1, out=buf8)
del primals_12
return (buf8, primals_1, primals_10, reinterpret_tensor(buf2, (16, 4), (4, 1), 0), buf5, buf6, reinterpret_tensor(buf7, (4, 16), (16, 1), 0), primals_11, primals_9, reinterpret_tensor(buf4, (16, 1, 4), (1, 1, 16), 0), reinterpret_tensor(buf1, (16, 1, 1), (1, 1, 1), 0), primals_7, )
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((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, ), (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, 4), (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, 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, 16), (16, 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)
| import torch
import torch.utils.data
from torch import nn
class MultiheadSimilarity(nn.Module):
def __init__(self, d_model, num_head, seq_len, in_proj=True):
super().__init__()
self.num_head = num_head
self.seq_len = seq_len
self.d_head = d_model // num_head
self.in_proj = in_proj
if self.in_proj:
self.q_in_proj = nn.Linear(d_model, seq_len * d_model, bias=True)
self.q_proj = nn.Linear(d_model, d_model, bias=True)
self.k_proj = nn.Linear(d_model, d_model, bias=True)
self.v_proj = nn.Linear(d_model, d_model, bias=True)
self.out_proj = nn.Linear(seq_len * d_model, d_model, bias=True)
def forward(self, q, kv):
bs, d_model = q.shape
nbs = bs * self.num_head
if self.in_proj:
q_ = self.q_in_proj(q)
q_ = q_.contiguous().view(bs, self.seq_len, d_model).transpose(0, 1
)
kv = q_ + kv
q = self.q_proj(q)
q = q.contiguous().view(nbs, self.d_head).unsqueeze(-1)
k = self.k_proj(kv)
k = k.contiguous().view(self.seq_len, nbs, self.d_head).transpose(0, 1)
similarity = torch.bmm(k, q) * float(self.d_head) ** -0.5
v = self.v_proj(kv)
v = v.contiguous().view(self.seq_len, nbs, self.d_head).transpose(0, 1)
v = (v * similarity).view(bs, self.num_head, self.seq_len, self.d_head)
output = self.out_proj(v.flatten(1))
return output
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4, 1])]
def get_init_inputs():
return [[], {'d_model': 4, 'num_head': 4, 'seq_len': 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.utils.data
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_clone_0(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 % 4
x1 = xindex // 4 % 4
x2 = xindex // 16
x3 = xindex // 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(out_ptr0 + x4, tmp4, xmask)
@triton.jit
def triton_poi_fused_add_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
tl.store(in_out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_poi_fused_clone_2(in_ptr0, in_ptr1, in_ptr2, 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
y3 = yindex
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (y3 + 16 * x2), xmask & ymask, eviction_policy
='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x2 + 4 * y3), xmask & ymask, eviction_policy=
'evict_last')
tmp2 = tmp0 + tmp1
tmp4 = 1.0
tmp5 = tmp3 * tmp4
tmp6 = tmp2 * tmp5
tl.store(out_ptr0 + (x2 + 4 * 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
) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (16, 4), (4, 1))
assert_size_stride(primals_3, (16,), (1,))
assert_size_stride(primals_4, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (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, 16), (16, 1))
assert_size_stride(primals_12, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
extern_kernels.mm(primals_1, reinterpret_tensor(primals_2, (4, 16),
(1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_6, primals_1, reinterpret_tensor(
primals_5, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_5
del primals_6
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_clone_0[grid(64)](buf0, primals_3, primals_4,
buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_3
del primals_4
buf3 = reinterpret_tensor(buf0, (16, 4), (4, 1), 0)
del buf0
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf3)
buf4 = reinterpret_tensor(buf3, (4, 4, 4), (16, 4, 1), 0)
del buf3
triton_poi_fused_add_1[grid(64)](buf4, primals_8, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_8
buf5 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 1), (1, 16, 0),
0), reinterpret_tensor(buf1, (16, 1, 1), (1, 1, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf6)
buf7 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_2[grid(16, 4)](buf6, primals_10, buf5, buf7,
16, 4, XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf8 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_12, reinterpret_tensor(buf7, (4, 16),
(16, 1), 0), reinterpret_tensor(primals_11, (16, 4), (1, 16), 0
), alpha=1, beta=1, out=buf8)
del primals_12
return buf8, primals_1, primals_10, reinterpret_tensor(buf2, (16, 4), (
4, 1), 0), buf5, buf6, reinterpret_tensor(buf7, (4, 16), (16, 1), 0
), primals_11, primals_9, reinterpret_tensor(buf4, (16, 1, 4), (1,
1, 16), 0), reinterpret_tensor(buf1, (16, 1, 1), (1, 1, 1), 0
), primals_7
class MultiheadSimilarityNew(nn.Module):
def __init__(self, d_model, num_head, seq_len, in_proj=True):
super().__init__()
self.num_head = num_head
self.seq_len = seq_len
self.d_head = d_model // num_head
self.in_proj = in_proj
if self.in_proj:
self.q_in_proj = nn.Linear(d_model, seq_len * d_model, bias=True)
self.q_proj = nn.Linear(d_model, d_model, bias=True)
self.k_proj = nn.Linear(d_model, d_model, bias=True)
self.v_proj = nn.Linear(d_model, d_model, bias=True)
self.out_proj = nn.Linear(seq_len * d_model, d_model, bias=True)
def forward(self, input_0, input_1):
primals_2 = self.q_in_proj.weight
primals_3 = self.q_in_proj.bias
primals_1 = self.q_proj.weight
primals_6 = self.q_proj.bias
primals_5 = self.k_proj.weight
primals_8 = self.k_proj.bias
primals_7 = self.v_proj.weight
primals_10 = self.v_proj.bias
primals_11 = self.out_proj.weight
primals_12 = self.out_proj.bias
primals_9 = input_0
primals_4 = 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]
| bladewaltz1/clipvid-tmp | MultiheadSimilarity | false | 1,564 | [
"MIT"
] | 0 | 8a4a990c318fdfbf6dac443abd3bc16637abba3d | https://github.com/bladewaltz1/clipvid-tmp/tree/8a4a990c318fdfbf6dac443abd3bc16637abba3d | import torch
import torch.utils.data
from torch import nn
class Model(nn.Module):
def __init__(self, d_model, num_head, seq_len, in_proj=True):
super().__init__()
self.num_head = num_head
self.seq_len = seq_len
self.d_head = d_model // num_head
self.in_proj = in_proj
if self.in_proj:
self.q_in_proj = nn.Linear(d_model, seq_len * d_model, bias=True)
self.q_proj = nn.Linear(d_model, d_model, bias=True)
self.k_proj = nn.Linear(d_model, d_model, bias=True)
self.v_proj = nn.Linear(d_model, d_model, bias=True)
self.out_proj = nn.Linear(seq_len * d_model, d_model, bias=True)
def forward(self, q, kv):
bs, d_model = q.shape
nbs = bs * self.num_head
if self.in_proj:
q_ = self.q_in_proj(q)
q_ = q_.contiguous().view(bs, self.seq_len, d_model).transpose(0, 1
)
kv = q_ + kv
q = self.q_proj(q)
q = q.contiguous().view(nbs, self.d_head).unsqueeze(-1)
k = self.k_proj(kv)
k = k.contiguous().view(self.seq_len, nbs, self.d_head).transpose(0, 1)
similarity = torch.bmm(k, q) * float(self.d_head) ** -0.5
v = self.v_proj(kv)
v = v.contiguous().view(self.seq_len, nbs, self.d_head).transpose(0, 1)
v = (v * similarity).view(bs, self.num_head, self.seq_len, self.d_head)
output = self.out_proj(v.flatten(1))
return output
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4, 1])]
def get_init_inputs():
return [4, 4, 4]
|
ResidualBlock | # 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_6/inductor_cache/ba/cba3dn7beshdzv7hwacmgs5xk75nqsfksfqjj6tajh7uagenh3e5.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, %arg0_1), kwargs = {})
# %copy_ : [num_users=1] = call_function[target=torch.ops.aten.copy_.default](args = (%arg0_1, %add), 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_ptr0', 'out_ptr1'], '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_0(in_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)
tmp1 = tmp0 + tmp0
tl.store(out_ptr1 + (x0), tmp1, 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)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(arg0_1, arg0_1, 256, grid=grid(256), stream=stream0)
return (arg0_1, )
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
class ResidualBlock(nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.in_channels, self.out_channels = in_channels, out_channels
self.blocks = nn.Identity()
self.shortcut = nn.Identity()
self.should_apply_shortcut = self.in_channels != self.out_channels
def forward(self, x):
if self.should_apply_shortcut:
residual = self.shortcut(x)
else:
residual = x
x = self.blocks(x)
x += residual
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 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
@triton.jit
def triton_poi_fused_add_0(in_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)
tmp1 = tmp0 + tmp0
tl.store(out_ptr1 + x0, tmp1, 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)
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](arg0_1, arg0_1, 256, XBLOCK=256,
num_warps=4, num_stages=1)
return arg0_1,
class ResidualBlockNew(nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.in_channels, self.out_channels = in_channels, out_channels
self.blocks = nn.Identity()
self.shortcut = nn.Identity()
self.should_apply_shortcut = self.in_channels != self.out_channels
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| boxiXia/pytorch_sac | ResidualBlock | false | 1,565 | [
"MIT"
] | 0 | ad570845c482498769217b398c22fafaff2ff2f1 | https://github.com/boxiXia/pytorch_sac/tree/ad570845c482498769217b398c22fafaff2ff2f1 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.in_channels, self.out_channels = in_channels, out_channels
self.blocks = nn.Identity()
self.shortcut = nn.Identity()
self.should_apply_shortcut = self.in_channels != self.out_channels
def forward(self, x):
if self.should_apply_shortcut:
residual = self.shortcut(x)
else:
residual = x
x = self.blocks(x)
x += residual
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ChannelAttention | # 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_6/inductor_cache/ui/cuic4afqa62ssfcdm5hmkrz53ddvdsm6siush4qsvbllc4tnqhgl.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 = (%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=[512, 16],
reduction_hint=ReductionHint.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, 4), 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, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 384
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)
r2 = rindex
x3 = xindex
x0 = xindex % 96
x1 = (xindex // 96)
tmp0 = tl.load(in_ptr0 + (r2 + (16*x3)), 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.store(out_ptr0 + (x0 + (96*r2) + (1536*x1)), tmp0, xmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + (x3), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/7e/c7ei6d6v36vmsdqtts7nemmrhwqny4y72lxi34r2yyspu4xvbvcs.py
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# relu => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_relu_1 = async_compile.triton('triton_poi_fused_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=[32],
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,), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_1', 'mutated_arg_names': ['in_out_ptr0'], '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_1(in_out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 24
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.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/zj/czjpxgv5sh6tvpcjc67uw3f4yhvtlkpa5mk2k4jfjqcadauphpnd.py
# Topologically Sorted Source Nodes: [adaptive_max_pool2d], Original ATen: [aten.adaptive_max_pool2d]
# Source node to ATen node mapping:
# adaptive_max_pool2d => getitem
# Graph fragment:
# %getitem : [num_users=2] = call_function[target=operator.getitem](args = (%adaptive_max_pool2d, 0), kwargs = {})
triton_poi_fused_adaptive_max_pool2d_2 = async_compile.triton('triton_poi_fused_adaptive_max_pool2d_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=[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_adaptive_max_pool2d_2', '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_adaptive_max_pool2d_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 96
x1 = (xindex // 96)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (1536*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (96 + x0 + (1536*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (192 + x0 + (1536*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (288 + x0 + (1536*x1)), xmask)
tmp7 = tl.load(in_ptr0 + (384 + x0 + (1536*x1)), xmask)
tmp9 = tl.load(in_ptr0 + (480 + x0 + (1536*x1)), xmask)
tmp11 = tl.load(in_ptr0 + (576 + x0 + (1536*x1)), xmask)
tmp13 = tl.load(in_ptr0 + (672 + x0 + (1536*x1)), xmask)
tmp15 = tl.load(in_ptr0 + (768 + x0 + (1536*x1)), xmask)
tmp17 = tl.load(in_ptr0 + (864 + x0 + (1536*x1)), xmask)
tmp19 = tl.load(in_ptr0 + (960 + x0 + (1536*x1)), xmask)
tmp21 = tl.load(in_ptr0 + (1056 + x0 + (1536*x1)), xmask)
tmp23 = tl.load(in_ptr0 + (1152 + x0 + (1536*x1)), xmask)
tmp25 = tl.load(in_ptr0 + (1248 + x0 + (1536*x1)), xmask)
tmp27 = tl.load(in_ptr0 + (1344 + x0 + (1536*x1)), xmask)
tmp29 = tl.load(in_ptr0 + (1440 + x0 + (1536*x1)), xmask)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp8 = triton_helpers.maximum(tmp7, tmp6)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tmp12 = triton_helpers.maximum(tmp11, tmp10)
tmp14 = triton_helpers.maximum(tmp13, tmp12)
tmp16 = triton_helpers.maximum(tmp15, tmp14)
tmp18 = triton_helpers.maximum(tmp17, tmp16)
tmp20 = triton_helpers.maximum(tmp19, tmp18)
tmp22 = triton_helpers.maximum(tmp21, tmp20)
tmp24 = triton_helpers.maximum(tmp23, tmp22)
tmp26 = triton_helpers.maximum(tmp25, tmp24)
tmp28 = triton_helpers.maximum(tmp27, tmp26)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tl.store(out_ptr0 + (x2), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/gs/cgsacshswhsjmlgcdzis3dgyumj23frn6sijgme5fpv4cof6pdba.py
# Topologically Sorted Source Nodes: [out, out_1, mul], Original ATen: [aten.add, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# out => add
# out_1 => sigmoid
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_1, %convolution_3), kwargs = {})
# %sigmoid : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %sigmoid), kwargs = {})
triton_poi_fused_add_mul_sigmoid_3 = async_compile.triton('triton_poi_fused_add_mul_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=[512, 16], 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, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_sigmoid_3', '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_sigmoid_3(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 384
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
x2 = xindex
y0 = yindex % 96
y1 = (yindex // 96)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (96*x2) + (1536*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y3), ymask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + (y3), ymask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.sigmoid(tmp3)
tmp5 = tmp0 * tmp4
tl.store(out_ptr0 + (x2 + (16*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/dl/cdlmneiczcqp6wwsc6yksfonx6k5n3zq6g3dfv46ok3j7zv7btlu.py
# Topologically Sorted Source Nodes: [out, out_1], Original ATen: [aten.add, aten.sigmoid, aten.sigmoid_backward]
# Source node to ATen node mapping:
# out => add
# out_1 => sigmoid
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_1, %convolution_3), kwargs = {})
# %sigmoid : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %sub), kwargs = {})
triton_poi_fused_add_sigmoid_sigmoid_backward_4 = async_compile.triton('triton_poi_fused_add_sigmoid_sigmoid_backward_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=[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_add_sigmoid_sigmoid_backward_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_add_sigmoid_sigmoid_backward_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 384
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.sigmoid(tmp2)
tmp4 = 1.0
tmp5 = tmp4 - tmp3
tmp6 = tmp3 * tmp5
tl.store(in_out_ptr0 + (x0), 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, 96, 4, 4), (1536, 16, 4, 1))
assert_size_stride(primals_2, (6, 96, 1, 1), (96, 1, 1, 1))
assert_size_stride(primals_3, (96, 6, 1, 1), (6, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 96, 4, 4), (1536, 1, 384, 96), torch.float32)
buf1 = empty_strided_cuda((4, 96, 1, 1), (96, 1, 384, 384), torch.float32)
buf2 = reinterpret_tensor(buf1, (4, 96, 1, 1), (96, 1, 96, 96), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [adaptive_avg_pool2d], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf2, primals_1, buf0, 384, 16, grid=grid(384), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf3 = extern_kernels.convolution(buf2, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 6, 1, 1), (6, 1, 6, 6))
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu]
triton_poi_fused_relu_1.run(buf4, 24, grid=grid(24), stream=stream0)
# Topologically Sorted Source Nodes: [avg_out], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(buf4, primals_3, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 96, 1, 1), (96, 1, 96, 96))
buf6 = empty_strided_cuda((4, 96, 1, 1), (96, 1, 96, 96), torch.float32)
# Topologically Sorted Source Nodes: [adaptive_max_pool2d], Original ATen: [aten.adaptive_max_pool2d]
triton_poi_fused_adaptive_max_pool2d_2.run(buf0, buf6, 384, grid=grid(384), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
buf7 = extern_kernels.convolution(buf6, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf7, (4, 6, 1, 1), (6, 1, 6, 6))
buf8 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [relu_1], Original ATen: [aten.relu]
triton_poi_fused_relu_1.run(buf8, 24, grid=grid(24), stream=stream0)
# Topologically Sorted Source Nodes: [max_out], Original ATen: [aten.convolution]
buf9 = extern_kernels.convolution(buf8, primals_3, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf9, (4, 96, 1, 1), (96, 1, 96, 96))
buf10 = empty_strided_cuda((4, 96, 4, 4), (1536, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out, out_1, mul], Original ATen: [aten.add, aten.sigmoid, aten.mul]
triton_poi_fused_add_mul_sigmoid_3.run(buf0, buf5, buf9, buf10, 384, 16, grid=grid(384, 16), stream=stream0)
buf11 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [out, out_1], Original ATen: [aten.add, aten.sigmoid, aten.sigmoid_backward]
triton_poi_fused_add_sigmoid_sigmoid_backward_4.run(buf11, buf9, 384, grid=grid(384), stream=stream0)
del buf9
return (buf10, buf0, primals_2, primals_3, buf2, buf4, buf6, buf8, 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, 96, 4, 4), (1536, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((6, 96, 1, 1), (96, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((96, 6, 1, 1), (6, 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
import torch.nn as nn
class ChannelAttention(nn.Module):
def __init__(self, in_planes=96, ratio=16):
super(ChannelAttention, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.max_pool = nn.AdaptiveMaxPool2d(1)
self.fc1 = nn.Conv2d(in_planes, in_planes // ratio, 1, bias=False)
self.relu1 = nn.ReLU()
self.fc2 = nn.Conv2d(in_planes // ratio, in_planes, 1, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
avg_out = self.fc2(self.relu1(self.fc1(self.avg_pool(x))))
max_out = self.fc2(self.relu1(self.fc1(self.max_pool(x))))
out = avg_out + max_out
out = self.sigmoid(out)
return x * out
def get_inputs():
return [torch.rand([4, 96, 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.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_mean_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 384
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)
r2 = rindex
x3 = xindex
x0 = xindex % 96
x1 = xindex // 96
tmp0 = tl.load(in_ptr0 + (r2 + 16 * x3), 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.store(out_ptr0 + (x0 + 96 * r2 + 1536 * x1), tmp0, xmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + x3, tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_1(in_out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 24
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.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(in_out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_adaptive_max_pool2d_2(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 96
x1 = xindex // 96
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 1536 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (96 + x0 + 1536 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (192 + x0 + 1536 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (288 + x0 + 1536 * x1), xmask)
tmp7 = tl.load(in_ptr0 + (384 + x0 + 1536 * x1), xmask)
tmp9 = tl.load(in_ptr0 + (480 + x0 + 1536 * x1), xmask)
tmp11 = tl.load(in_ptr0 + (576 + x0 + 1536 * x1), xmask)
tmp13 = tl.load(in_ptr0 + (672 + x0 + 1536 * x1), xmask)
tmp15 = tl.load(in_ptr0 + (768 + x0 + 1536 * x1), xmask)
tmp17 = tl.load(in_ptr0 + (864 + x0 + 1536 * x1), xmask)
tmp19 = tl.load(in_ptr0 + (960 + x0 + 1536 * x1), xmask)
tmp21 = tl.load(in_ptr0 + (1056 + x0 + 1536 * x1), xmask)
tmp23 = tl.load(in_ptr0 + (1152 + x0 + 1536 * x1), xmask)
tmp25 = tl.load(in_ptr0 + (1248 + x0 + 1536 * x1), xmask)
tmp27 = tl.load(in_ptr0 + (1344 + x0 + 1536 * x1), xmask)
tmp29 = tl.load(in_ptr0 + (1440 + x0 + 1536 * x1), xmask)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp8 = triton_helpers.maximum(tmp7, tmp6)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tmp12 = triton_helpers.maximum(tmp11, tmp10)
tmp14 = triton_helpers.maximum(tmp13, tmp12)
tmp16 = triton_helpers.maximum(tmp15, tmp14)
tmp18 = triton_helpers.maximum(tmp17, tmp16)
tmp20 = triton_helpers.maximum(tmp19, tmp18)
tmp22 = triton_helpers.maximum(tmp21, tmp20)
tmp24 = triton_helpers.maximum(tmp23, tmp22)
tmp26 = triton_helpers.maximum(tmp25, tmp24)
tmp28 = triton_helpers.maximum(tmp27, tmp26)
tmp30 = triton_helpers.maximum(tmp29, tmp28)
tl.store(out_ptr0 + x2, tmp30, xmask)
@triton.jit
def triton_poi_fused_add_mul_sigmoid_3(in_ptr0, in_ptr1, in_ptr2, out_ptr0,
ynumel, xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 384
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
x2 = xindex
y0 = yindex % 96
y1 = yindex // 96
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 96 * x2 + 1536 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y3, ymask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + y3, ymask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.sigmoid(tmp3)
tmp5 = tmp0 * tmp4
tl.store(out_ptr0 + (x2 + 16 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_add_sigmoid_sigmoid_backward_4(in_out_ptr0, in_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 384
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.sigmoid(tmp2)
tmp4 = 1.0
tmp5 = tmp4 - tmp3
tmp6 = tmp3 * tmp5
tl.store(in_out_ptr0 + x0, tmp6, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 96, 4, 4), (1536, 16, 4, 1))
assert_size_stride(primals_2, (6, 96, 1, 1), (96, 1, 1, 1))
assert_size_stride(primals_3, (96, 6, 1, 1), (6, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 96, 4, 4), (1536, 1, 384, 96), torch.
float32)
buf1 = empty_strided_cuda((4, 96, 1, 1), (96, 1, 384, 384), torch.
float32)
buf2 = reinterpret_tensor(buf1, (4, 96, 1, 1), (96, 1, 96, 96), 0)
del buf1
get_raw_stream(0)
triton_per_fused_mean_0[grid(384)](buf2, primals_1, buf0, 384, 16,
XBLOCK=32, num_warps=4, num_stages=1)
del primals_1
buf3 = extern_kernels.convolution(buf2, primals_2, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 6, 1, 1), (6, 1, 6, 6))
buf4 = buf3
del buf3
triton_poi_fused_relu_1[grid(24)](buf4, 24, XBLOCK=32, num_warps=1,
num_stages=1)
buf5 = extern_kernels.convolution(buf4, primals_3, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 96, 1, 1), (96, 1, 96, 96))
buf6 = empty_strided_cuda((4, 96, 1, 1), (96, 1, 96, 96), torch.float32
)
triton_poi_fused_adaptive_max_pool2d_2[grid(384)](buf0, buf6, 384,
XBLOCK=128, num_warps=4, num_stages=1)
buf7 = extern_kernels.convolution(buf6, primals_2, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf7, (4, 6, 1, 1), (6, 1, 6, 6))
buf8 = buf7
del buf7
triton_poi_fused_relu_1[grid(24)](buf8, 24, XBLOCK=32, num_warps=1,
num_stages=1)
buf9 = extern_kernels.convolution(buf8, primals_3, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf9, (4, 96, 1, 1), (96, 1, 96, 96))
buf10 = empty_strided_cuda((4, 96, 4, 4), (1536, 16, 4, 1), torch.
float32)
triton_poi_fused_add_mul_sigmoid_3[grid(384, 16)](buf0, buf5, buf9,
buf10, 384, 16, XBLOCK=1, YBLOCK=256, num_warps=4, num_stages=1)
buf11 = buf5
del buf5
triton_poi_fused_add_sigmoid_sigmoid_backward_4[grid(384)](buf11,
buf9, 384, XBLOCK=256, num_warps=4, num_stages=1)
del buf9
return buf10, buf0, primals_2, primals_3, buf2, buf4, buf6, buf8, buf11
class ChannelAttentionNew(nn.Module):
def __init__(self, in_planes=96, ratio=16):
super(ChannelAttentionNew, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.max_pool = nn.AdaptiveMaxPool2d(1)
self.fc1 = nn.Conv2d(in_planes, in_planes // ratio, 1, bias=False)
self.relu1 = nn.ReLU()
self.fc2 = nn.Conv2d(in_planes // ratio, in_planes, 1, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.fc1.weight
primals_3 = self.fc2.weight
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| biolee3/SAMDNet | ChannelAttention | false | 1,566 | [
"MIT"
] | 0 | 9a0d70f976e22d512046b4aa5727dd26422d0aff | https://github.com/biolee3/SAMDNet/tree/9a0d70f976e22d512046b4aa5727dd26422d0aff | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, in_planes=96, ratio=16):
super().__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.max_pool = nn.AdaptiveMaxPool2d(1)
self.fc1 = nn.Conv2d(in_planes, in_planes // ratio, 1, bias=False)
self.relu1 = nn.ReLU()
self.fc2 = nn.Conv2d(in_planes // ratio, in_planes, 1, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
avg_out = self.fc2(self.relu1(self.fc1(self.avg_pool(x))))
max_out = self.fc2(self.relu1(self.fc1(self.max_pool(x))))
out = avg_out + max_out
out = self.sigmoid(out)
return x * out
def get_inputs():
return [torch.rand([4, 96, 4, 4])]
def get_init_inputs():
return []
|
MiniBatchStddevLayer | # 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_6/inductor_cache/64/c64ilbvzmdnnrmalq32ejoehyuog4qhx7ikcg5kbwgjcv6pvdhit.py
# Topologically Sorted Source Nodes: [mean, y_1, pow_1, y_2, add, y_3, y_4, y_5], Original ATen: [aten.mean, aten.sub, aten.pow, aten.add, aten.sqrt, aten.repeat]
# Source node to ATen node mapping:
# add => add
# mean => mean
# pow_1 => pow_1
# y_1 => sub
# y_2 => mean_1
# y_3 => sqrt
# y_4 => mean_2
# y_5 => repeat
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view, [0], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view, %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, [0]), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean_1, 1e-08), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add,), kwargs = {})
# %mean_2 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%sqrt, [1, 2, 3], True), kwargs = {})
# %repeat : [num_users=1] = call_function[target=torch.ops.aten.repeat.default](args = (%mean_2, [4, 1, 4, 4]), kwargs = {})
triton_per_fused_add_mean_pow_repeat_sqrt_sub_0 = async_compile.triton('triton_per_fused_add_mean_pow_repeat_sqrt_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.persistent_reduction(
size_hints=[1, 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': {2: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_mean_pow_repeat_sqrt_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, '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_mean_pow_repeat_sqrt_sub_0(in_ptr0, out_ptr1, 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)
r0 = rindex
r1 = rindex % 16
r2 = (rindex // 16)
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = tl.load(in_ptr0 + (64 + r0), None)
tmp3 = tl.load(in_ptr0 + (128 + r0), None)
tmp5 = tl.load(in_ptr0 + (192 + r0), None)
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-08
tmp22 = tmp20 + tmp21
tmp23 = libdevice.sqrt(tmp22)
tmp24 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp26 = tl.sum(tmp24, 1)[:, None]
tmp27 = 64.0
tmp28 = tmp26 / tmp27
tl.store(out_ptr1 + (tl.broadcast_to(r1 + (80*r2), [XBLOCK, RBLOCK])), tmp28, None)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/yi/cyidf2yj3fms5jdxlfe7fdijzfj6p5a5q2qxo4llkuxnpqh6fj5o.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 = ([%arg0_1, %repeat], 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=[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_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_cat_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
x0 = xindex % 64
x1 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tl.store(out_ptr0 + (x0 + (80*x1)), tmp0, 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)
buf3 = empty_strided_cuda((4, 5, 4, 4), (80, 16, 4, 1), torch.float32)
buf2 = reinterpret_tensor(buf3, (4, 1, 4, 4), (80, 16, 4, 1), 64) # alias
# Topologically Sorted Source Nodes: [mean, y_1, pow_1, y_2, add, y_3, y_4, y_5], Original ATen: [aten.mean, aten.sub, aten.pow, aten.add, aten.sqrt, aten.repeat]
stream0 = get_raw_stream(0)
triton_per_fused_add_mean_pow_repeat_sqrt_sub_0.run(arg0_1, buf2, 1, 64, grid=grid(1), stream=stream0)
buf1 = reinterpret_tensor(buf3, (4, 4, 4, 4), (80, 16, 4, 1), 0) # alias
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
triton_poi_fused_cat_1.run(arg0_1, buf1, 256, grid=grid(256), stream=stream0)
del arg0_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)
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.distributed as dist
import torch.autograd as autograd
import torch.utils.cpp_extension
class AllGatherLayer(autograd.Function):
"""All gather layer with backward propagation path.
Indeed, this module is to make ``dist.all_gather()`` in the backward graph.
Such kind of operation has been widely used in Moco and other contrastive
learning algorithms.
"""
@staticmethod
def forward(ctx, x):
"""Forward function."""
ctx.save_for_backward(x)
output = [torch.zeros_like(x) for _ in range(dist.get_world_size())]
dist.all_gather(output, x)
return tuple(output)
@staticmethod
def backward(ctx, *grad_outputs):
"""Backward function."""
x, = ctx.saved_tensors
grad_out = torch.zeros_like(x)
grad_out = grad_outputs[dist.get_rank()]
return grad_out
class MiniBatchStddevLayer(nn.Module):
"""Minibatch standard deviation.
Args:
group_size (int, optional): The size of groups in batch dimension.
Defaults to 4.
eps (float, optional): Epsilon value to avoid computation error.
Defaults to 1e-8.
gather_all_batch (bool, optional): Whether gather batch from all GPUs.
Defaults to False.
"""
def __init__(self, group_size=4, eps=1e-08, gather_all_batch=False):
super().__init__()
self.group_size = group_size
self.eps = eps
self.gather_all_batch = gather_all_batch
if self.gather_all_batch:
assert torch.distributed.is_initialized(
), 'Only in distributed training can the tensors be all gathered.'
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input tensor with shape (n, c, h, w).
Returns:
Tensor: Forward results.
"""
if self.gather_all_batch:
x = torch.cat(AllGatherLayer.apply(x), dim=0)
assert x.shape[0] <= self.group_size or x.shape[0
] % self.group_size == 0, f'Batch size be smaller than or equal to group size. Otherwise, batch size should be divisible by the group size.But got batch size {x.shape[0]}, group size {self.group_size}'
n, c, h, w = x.shape
group_size = min(n, self.group_size)
y = torch.reshape(x, (group_size, -1, c, h, w))
y = y - y.mean(dim=0, keepdim=True)
y = y.pow(2).mean(dim=0, keepdim=False)
y = torch.sqrt(y + self.eps)
y = y.mean(dim=(1, 2, 3), keepdim=True)
y = y.repeat(group_size, 1, h, w)
return torch.cat([x, y], dim=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
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
import torch.distributed as dist
import torch.autograd as autograd
import torch.utils.cpp_extension
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_mean_pow_repeat_sqrt_sub_0(in_ptr0, out_ptr1,
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)
r0 = rindex
r1 = rindex % 16
r2 = rindex // 16
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = tl.load(in_ptr0 + (64 + r0), None)
tmp3 = tl.load(in_ptr0 + (128 + r0), None)
tmp5 = tl.load(in_ptr0 + (192 + r0), None)
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-08
tmp22 = tmp20 + tmp21
tmp23 = libdevice.sqrt(tmp22)
tmp24 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp26 = tl.sum(tmp24, 1)[:, None]
tmp27 = 64.0
tmp28 = tmp26 / tmp27
tl.store(out_ptr1 + tl.broadcast_to(r1 + 80 * r2, [XBLOCK, RBLOCK]),
tmp28, None)
@triton.jit
def triton_poi_fused_cat_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
x0 = xindex % 64
x1 = xindex // 64
tmp0 = tl.load(in_ptr0 + x2, xmask)
tl.store(out_ptr0 + (x0 + 80 * x1), tmp0, 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)
buf3 = empty_strided_cuda((4, 5, 4, 4), (80, 16, 4, 1), torch.float32)
buf2 = reinterpret_tensor(buf3, (4, 1, 4, 4), (80, 16, 4, 1), 64)
get_raw_stream(0)
triton_per_fused_add_mean_pow_repeat_sqrt_sub_0[grid(1)](arg0_1,
buf2, 1, 64, XBLOCK=1, num_warps=2, num_stages=1)
buf1 = reinterpret_tensor(buf3, (4, 4, 4, 4), (80, 16, 4, 1), 0)
triton_poi_fused_cat_1[grid(256)](arg0_1, buf1, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return buf3,
class AllGatherLayer(autograd.Function):
"""All gather layer with backward propagation path.
Indeed, this module is to make ``dist.all_gather()`` in the backward graph.
Such kind of operation has been widely used in Moco and other contrastive
learning algorithms.
"""
@staticmethod
def forward(ctx, x):
"""Forward function."""
ctx.save_for_backward(x)
output = [torch.zeros_like(x) for _ in range(dist.get_world_size())]
dist.all_gather(output, x)
return tuple(output)
@staticmethod
def backward(ctx, *grad_outputs):
"""Backward function."""
x, = ctx.saved_tensors
grad_out = torch.zeros_like(x)
grad_out = grad_outputs[dist.get_rank()]
return grad_out
class MiniBatchStddevLayerNew(nn.Module):
"""Minibatch standard deviation.
Args:
group_size (int, optional): The size of groups in batch dimension.
Defaults to 4.
eps (float, optional): Epsilon value to avoid computation error.
Defaults to 1e-8.
gather_all_batch (bool, optional): Whether gather batch from all GPUs.
Defaults to False.
"""
def __init__(self, group_size=4, eps=1e-08, gather_all_batch=False):
super().__init__()
self.group_size = group_size
self.eps = eps
self.gather_all_batch = gather_all_batch
if self.gather_all_batch:
assert torch.distributed.is_initialized(
), 'Only in distributed training can the tensors be all gathered.'
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bladesaber/mmgeneration | MiniBatchStddevLayer | false | 1,567 | [
"Apache-2.0"
] | 0 | 158b49f7efd8028f231f6e9ca758ae0e20dd72ae | https://github.com/bladesaber/mmgeneration/tree/158b49f7efd8028f231f6e9ca758ae0e20dd72ae | import torch
import torch.nn as nn
import torch.distributed as dist
import torch.autograd as autograd
import torch.utils.cpp_extension
class AllGatherLayer(autograd.Function):
"""All gather layer with backward propagation path.
Indeed, this module is to make ``dist.all_gather()`` in the backward graph.
Such kind of operation has been widely used in Moco and other contrastive
learning algorithms.
"""
@staticmethod
def forward(ctx, x):
"""Forward function."""
ctx.save_for_backward(x)
output = [torch.zeros_like(x) for _ in range(dist.get_world_size())]
dist.all_gather(output, x)
return tuple(output)
@staticmethod
def backward(ctx, *grad_outputs):
"""Backward function."""
x, = ctx.saved_tensors
grad_out = torch.zeros_like(x)
grad_out = grad_outputs[dist.get_rank()]
return grad_out
class Model(nn.Module):
"""Minibatch standard deviation.
Args:
group_size (int, optional): The size of groups in batch dimension.
Defaults to 4.
eps (float, optional): Epsilon value to avoid computation error.
Defaults to 1e-8.
gather_all_batch (bool, optional): Whether gather batch from all GPUs.
Defaults to False.
"""
def __init__(self, group_size=4, eps=1e-08, gather_all_batch=False):
super().__init__()
self.group_size = group_size
self.eps = eps
self.gather_all_batch = gather_all_batch
if self.gather_all_batch:
assert torch.distributed.is_initialized(
), 'Only in distributed training can the tensors be all gathered.'
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input tensor with shape (n, c, h, w).
Returns:
Tensor: Forward results.
"""
if self.gather_all_batch:
x = torch.cat(AllGatherLayer.apply(x), dim=0)
assert x.shape[0] <= self.group_size or x.shape[0
] % self.group_size == 0, f'Batch size be smaller than or equal to group size. Otherwise, batch size should be divisible by the group size.But got batch size {x.shape[0]}, group size {self.group_size}'
n, c, h, w = x.shape
group_size = min(n, self.group_size)
y = torch.reshape(x, (group_size, -1, c, h, w))
y = y - y.mean(dim=0, keepdim=True)
y = y.pow(2).mean(dim=0, keepdim=False)
y = torch.sqrt(y + self.eps)
y = y.mean(dim=(1, 2, 3), keepdim=True)
y = y.repeat(group_size, 1, h, w)
return torch.cat([x, y], dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SigmoidFocalLoss | # 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_6/inductor_cache/sw/cswigq6welcg2ea6b2om22mf7opqse7wgus3uvcqvwzvge7pz3qd.py
# Topologically Sorted Source Nodes: [mul_3, sub_3, mul_4, alpha_t, ce_loss, p, mul, sub, sub_1, mul_1, p_t, sub_2, pow_1, loss, loss_1, loss_2], Original ATen: [aten.mul, aten.rsub, aten.add, aten.binary_cross_entropy_with_logits, aten.sigmoid, aten.pow, aten.mean]
# Source node to ATen node mapping:
# alpha_t => add_1
# ce_loss => abs_1, exp, full_default, log1p, minimum, mul, neg, sub, sub_1, sub_2
# loss => mul_3
# loss_1 => mul_6
# loss_2 => mean
# mul => mul_1
# mul_1 => mul_2
# mul_3 => mul_4
# mul_4 => mul_5
# p => sigmoid
# p_t => add
# pow_1 => pow_1
# sub => sub_3
# sub_1 => sub_4
# sub_2 => sub_5
# sub_3 => sub_6
# Graph fragment:
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, 0.25), kwargs = {})
# %sub_6 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_6, 0.75), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %mul_5), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %arg0_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), 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 = (%full_default, %arg0_1), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%arg0_1,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %sub_1), kwargs = {})
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%arg0_1,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %arg1_1), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %sub_4 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_3, %sub_4), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %mul_2), kwargs = {})
# %sub_5 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %add), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_5, 2.0), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, %pow_1), kwargs = {})
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_1, %mul_3), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul_6,), kwargs = {})
triton_per_fused_add_binary_cross_entropy_with_logits_mean_mul_pow_rsub_sigmoid_0 = async_compile.triton('triton_per_fused_add_binary_cross_entropy_with_logits_mean_mul_pow_rsub_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=[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_binary_cross_entropy_with_logits_mean_mul_pow_rsub_sigmoid_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_add_binary_cross_entropy_with_logits_mean_mul_pow_rsub_sigmoid_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)
tmp8 = tl.load(in_ptr1 + (r0), None)
tmp1 = 0.25
tmp2 = tmp0 * tmp1
tmp3 = 1.0
tmp4 = tmp3 - tmp0
tmp5 = 0.75
tmp6 = tmp4 * tmp5
tmp7 = tmp2 + tmp6
tmp9 = tmp4 * tmp8
tmp10 = 0.0
tmp11 = triton_helpers.minimum(tmp10, tmp8)
tmp12 = tl_math.abs(tmp8)
tmp13 = -tmp12
tmp14 = tl_math.exp(tmp13)
tmp15 = libdevice.log1p(tmp14)
tmp16 = tmp11 - tmp15
tmp17 = tmp9 - tmp16
tmp18 = tl.sigmoid(tmp8)
tmp19 = tmp18 * tmp0
tmp20 = tmp3 - tmp18
tmp21 = tmp20 * tmp4
tmp22 = tmp19 + tmp21
tmp23 = tmp3 - tmp22
tmp24 = tmp23 * tmp23
tmp25 = tmp17 * tmp24
tmp26 = tmp7 * tmp25
tmp27 = tl.broadcast_to(tmp26, [RBLOCK])
tmp29 = triton_helpers.promote_to_tensor(tl.sum(tmp27, 0))
tmp30 = 256.0
tmp31 = tmp29 / tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp31, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul_3, sub_3, mul_4, alpha_t, ce_loss, p, mul, sub, sub_1, mul_1, p_t, sub_2, pow_1, loss, loss_1, loss_2], Original ATen: [aten.mul, aten.rsub, aten.add, aten.binary_cross_entropy_with_logits, aten.sigmoid, aten.pow, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_add_binary_cross_entropy_with_logits_mean_mul_pow_rsub_sigmoid_0.run(buf1, arg1_1, arg0_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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.nn import functional as F
def sigmoid_focal_loss(inputs: 'torch.Tensor', targets: 'torch.Tensor',
alpha: 'float'=-1, gamma: 'float'=2, reduction: 'str'='none'
) ->torch.Tensor:
"""
Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002.
Args:
inputs: A float tensor of arbitrary shape.
The predictions for each example.
targets: A float tensor with the same shape as inputs. Stores the binary
classification label for each element in inputs
(0 for the negative class and 1 for the positive class).
alpha: (optional) Weighting factor in range (0,1) to balance
positive vs negative examples. Default = -1 (no weighting).
gamma: Exponent of the modulating factor (1 - p_t) to
balance easy vs hard examples.
reduction: 'none' | 'mean' | 'sum'
'none': No reduction will be applied to the output.
'mean': The output will be averaged.
'sum': The output will be summed.
Returns:
Loss tensor with the reduction option applied.
"""
p = torch.sigmoid(inputs)
ce_loss = F.binary_cross_entropy_with_logits(inputs, targets, reduction
='none')
p_t = p * targets + (1 - p) * (1 - targets)
loss = ce_loss * (1 - p_t) ** gamma
if alpha >= 0:
alpha_t = alpha * targets + (1 - alpha) * (1 - targets)
loss = alpha_t * loss
if reduction == 'mean':
loss = loss.mean()
elif reduction == 'sum':
loss = loss.sum()
return loss
class SigmoidFocalLoss(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, preds, targets):
return sigmoid_focal_loss(preds, targets, self.alpha, self.gamma,
self.reduction)
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 libdevice, math as tl_math
import torch.nn as nn
from torch.nn import 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_per_fused_add_binary_cross_entropy_with_logits_mean_mul_pow_rsub_sigmoid_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)
tmp8 = tl.load(in_ptr1 + r0, None)
tmp1 = 0.25
tmp2 = tmp0 * tmp1
tmp3 = 1.0
tmp4 = tmp3 - tmp0
tmp5 = 0.75
tmp6 = tmp4 * tmp5
tmp7 = tmp2 + tmp6
tmp9 = tmp4 * tmp8
tmp10 = 0.0
tmp11 = triton_helpers.minimum(tmp10, tmp8)
tmp12 = tl_math.abs(tmp8)
tmp13 = -tmp12
tmp14 = tl_math.exp(tmp13)
tmp15 = libdevice.log1p(tmp14)
tmp16 = tmp11 - tmp15
tmp17 = tmp9 - tmp16
tmp18 = tl.sigmoid(tmp8)
tmp19 = tmp18 * tmp0
tmp20 = tmp3 - tmp18
tmp21 = tmp20 * tmp4
tmp22 = tmp19 + tmp21
tmp23 = tmp3 - tmp22
tmp24 = tmp23 * tmp23
tmp25 = tmp17 * tmp24
tmp26 = tmp7 * tmp25
tmp27 = tl.broadcast_to(tmp26, [RBLOCK])
tmp29 = triton_helpers.promote_to_tensor(tl.sum(tmp27, 0))
tmp30 = 256.0
tmp31 = tmp29 / tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp31, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_binary_cross_entropy_with_logits_mean_mul_pow_rsub_sigmoid_0[
grid(1)](buf1, arg1_1, arg0_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
def sigmoid_focal_loss(inputs: 'torch.Tensor', targets: 'torch.Tensor',
alpha: 'float'=-1, gamma: 'float'=2, reduction: 'str'='none'
) ->torch.Tensor:
"""
Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002.
Args:
inputs: A float tensor of arbitrary shape.
The predictions for each example.
targets: A float tensor with the same shape as inputs. Stores the binary
classification label for each element in inputs
(0 for the negative class and 1 for the positive class).
alpha: (optional) Weighting factor in range (0,1) to balance
positive vs negative examples. Default = -1 (no weighting).
gamma: Exponent of the modulating factor (1 - p_t) to
balance easy vs hard examples.
reduction: 'none' | 'mean' | 'sum'
'none': No reduction will be applied to the output.
'mean': The output will be averaged.
'sum': The output will be summed.
Returns:
Loss tensor with the reduction option applied.
"""
p = torch.sigmoid(inputs)
ce_loss = F.binary_cross_entropy_with_logits(inputs, targets, reduction
='none')
p_t = p * targets + (1 - p) * (1 - targets)
loss = ce_loss * (1 - p_t) ** gamma
if alpha >= 0:
alpha_t = alpha * targets + (1 - alpha) * (1 - targets)
loss = alpha_t * loss
if reduction == 'mean':
loss = loss.mean()
elif reduction == 'sum':
loss = loss.sum()
return loss
class SigmoidFocalLossNew(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| botkop/lark | SigmoidFocalLoss | false | 1,568 | [
"Apache-2.0"
] | 0 | edb2defdb514213fc121418578b0d9006a55f3a0 | https://github.com/botkop/lark/tree/edb2defdb514213fc121418578b0d9006a55f3a0 | import torch
import torch.nn as nn
from torch.nn import functional as F
def sigmoid_focal_loss(inputs: 'torch.Tensor', targets: 'torch.Tensor',
alpha: 'float'=-1, gamma: 'float'=2, reduction: 'str'='none'
) ->torch.Tensor:
"""
Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002.
Args:
inputs: A float tensor of arbitrary shape.
The predictions for each example.
targets: A float tensor with the same shape as inputs. Stores the binary
classification label for each element in inputs
(0 for the negative class and 1 for the positive class).
alpha: (optional) Weighting factor in range (0,1) to balance
positive vs negative examples. Default = -1 (no weighting).
gamma: Exponent of the modulating factor (1 - p_t) to
balance easy vs hard examples.
reduction: 'none' | 'mean' | 'sum'
'none': No reduction will be applied to the output.
'mean': The output will be averaged.
'sum': The output will be summed.
Returns:
Loss tensor with the reduction option applied.
"""
p = torch.sigmoid(inputs)
ce_loss = F.binary_cross_entropy_with_logits(inputs, targets, reduction
='none')
p_t = p * targets + (1 - p) * (1 - targets)
loss = ce_loss * (1 - p_t) ** gamma
if alpha >= 0:
alpha_t = alpha * targets + (1 - alpha) * (1 - targets)
loss = alpha_t * loss
if reduction == 'mean':
loss = loss.mean()
elif reduction == 'sum':
loss = loss.sum()
return loss
class Model(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, preds, targets):
return sigmoid_focal_loss(preds, targets, self.alpha, self.gamma,
self.reduction)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
LSTM | # 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_6/inductor_cache/yq/cyqwuw423ez77uwf6w3fslnus3cyx2u725lvad5i2ijxlswjt456.py
# Topologically Sorted Source Nodes: [matmul, matmul_2, matmul_4, matmul_6, matmul_8, matmul_10, matmul_12, matmul_14, matmul_16, matmul_18, matmul_20, matmul_22, matmul_24, matmul_26, matmul_28, matmul_30], Original ATen: [aten.mv]
# Source node to ATen node mapping:
# matmul => mul, sum_1
# matmul_10 => mul_8, sum_6
# matmul_12 => mul_9, sum_7
# matmul_14 => mul_10, sum_8
# matmul_16 => mul_14, sum_9
# matmul_18 => mul_15, sum_10
# matmul_2 => mul_1, sum_2
# matmul_20 => mul_16, sum_11
# matmul_22 => mul_17, sum_12
# matmul_24 => mul_21, sum_13
# matmul_26 => mul_22, sum_14
# matmul_28 => mul_23, sum_15
# matmul_30 => mul_24, sum_16
# matmul_4 => mul_2, sum_3
# matmul_6 => mul_3, sum_4
# matmul_8 => mul_7, sum_5
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1]), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [1]), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_2, [1]), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_3, [1]), kwargs = {})
# %mul_7 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select_4), kwargs = {})
# %sum_5 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_7, [1]), kwargs = {})
# %mul_8 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select_4), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_8, [1]), kwargs = {})
# %mul_9 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select_4), kwargs = {})
# %sum_7 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_9, [1]), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select_4), kwargs = {})
# %sum_8 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_10, [1]), kwargs = {})
# %mul_14 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select_8), kwargs = {})
# %sum_9 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_14, [1]), kwargs = {})
# %mul_15 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select_8), kwargs = {})
# %sum_10 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_15, [1]), kwargs = {})
# %mul_16 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select_8), kwargs = {})
# %sum_11 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_16, [1]), kwargs = {})
# %mul_17 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select_8), kwargs = {})
# %sum_12 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_17, [1]), kwargs = {})
# %mul_21 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select_12), kwargs = {})
# %sum_13 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_21, [1]), kwargs = {})
# %mul_22 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select_12), kwargs = {})
# %sum_14 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_22, [1]), kwargs = {})
# %mul_23 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select_12), kwargs = {})
# %sum_15 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_23, [1]), kwargs = {})
# %mul_24 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select_12), kwargs = {})
# %sum_16 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_24, [1]), kwargs = {})
triton_poi_fused_mv_0 = async_compile.triton('triton_poi_fused_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.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*fp32', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*fp32', 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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mv_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 32, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, out_ptr1, out_ptr2, out_ptr3, out_ptr4, out_ptr5, out_ptr6, out_ptr7, out_ptr8, out_ptr9, out_ptr10, out_ptr11, out_ptr12, out_ptr13, out_ptr14, out_ptr15, 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')
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (4))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp9 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (8))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp14 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (12))
tmp16 = tl.broadcast_to(tmp15, [XBLOCK])
tmp19 = tl.load(in_ptr1 + (1))
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp22 = tl.load(in_ptr1 + (5))
tmp23 = tl.broadcast_to(tmp22, [XBLOCK])
tmp26 = tl.load(in_ptr1 + (9))
tmp27 = tl.broadcast_to(tmp26, [XBLOCK])
tmp30 = tl.load(in_ptr1 + (13))
tmp31 = tl.broadcast_to(tmp30, [XBLOCK])
tmp34 = tl.load(in_ptr1 + (2))
tmp35 = tl.broadcast_to(tmp34, [XBLOCK])
tmp37 = tl.load(in_ptr1 + (6))
tmp38 = tl.broadcast_to(tmp37, [XBLOCK])
tmp41 = tl.load(in_ptr1 + (10))
tmp42 = tl.broadcast_to(tmp41, [XBLOCK])
tmp45 = tl.load(in_ptr1 + (14))
tmp46 = tl.broadcast_to(tmp45, [XBLOCK])
tmp49 = tl.load(in_ptr1 + (3))
tmp50 = tl.broadcast_to(tmp49, [XBLOCK])
tmp52 = tl.load(in_ptr1 + (7))
tmp53 = tl.broadcast_to(tmp52, [XBLOCK])
tmp56 = tl.load(in_ptr1 + (11))
tmp57 = tl.broadcast_to(tmp56, [XBLOCK])
tmp60 = tl.load(in_ptr1 + (15))
tmp61 = tl.broadcast_to(tmp60, [XBLOCK])
tmp64 = tl.load(in_ptr2 + (4*x0), xmask, eviction_policy='evict_last')
tmp66 = tl.load(in_ptr2 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp69 = tl.load(in_ptr2 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp72 = tl.load(in_ptr2 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp96 = tl.load(in_ptr3 + (4*x0), xmask, eviction_policy='evict_last')
tmp98 = tl.load(in_ptr3 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp101 = tl.load(in_ptr3 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp104 = tl.load(in_ptr3 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp128 = tl.load(in_ptr4 + (4*x0), xmask, eviction_policy='evict_last')
tmp130 = tl.load(in_ptr4 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp133 = tl.load(in_ptr4 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp136 = tl.load(in_ptr4 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp21 = tmp0 * tmp20
tmp24 = tmp4 * tmp23
tmp25 = tmp21 + tmp24
tmp28 = tmp9 * tmp27
tmp29 = tmp25 + tmp28
tmp32 = tmp14 * tmp31
tmp33 = tmp29 + tmp32
tmp36 = tmp0 * tmp35
tmp39 = tmp4 * tmp38
tmp40 = tmp36 + tmp39
tmp43 = tmp9 * tmp42
tmp44 = tmp40 + tmp43
tmp47 = tmp14 * tmp46
tmp48 = tmp44 + tmp47
tmp51 = tmp0 * tmp50
tmp54 = tmp4 * tmp53
tmp55 = tmp51 + tmp54
tmp58 = tmp9 * tmp57
tmp59 = tmp55 + tmp58
tmp62 = tmp14 * tmp61
tmp63 = tmp59 + tmp62
tmp65 = tmp64 * tmp2
tmp67 = tmp66 * tmp6
tmp68 = tmp65 + tmp67
tmp70 = tmp69 * tmp11
tmp71 = tmp68 + tmp70
tmp73 = tmp72 * tmp16
tmp74 = tmp71 + tmp73
tmp75 = tmp64 * tmp20
tmp76 = tmp66 * tmp23
tmp77 = tmp75 + tmp76
tmp78 = tmp69 * tmp27
tmp79 = tmp77 + tmp78
tmp80 = tmp72 * tmp31
tmp81 = tmp79 + tmp80
tmp82 = tmp64 * tmp35
tmp83 = tmp66 * tmp38
tmp84 = tmp82 + tmp83
tmp85 = tmp69 * tmp42
tmp86 = tmp84 + tmp85
tmp87 = tmp72 * tmp46
tmp88 = tmp86 + tmp87
tmp89 = tmp64 * tmp50
tmp90 = tmp66 * tmp53
tmp91 = tmp89 + tmp90
tmp92 = tmp69 * tmp57
tmp93 = tmp91 + tmp92
tmp94 = tmp72 * tmp61
tmp95 = tmp93 + tmp94
tmp97 = tmp96 * tmp2
tmp99 = tmp98 * tmp6
tmp100 = tmp97 + tmp99
tmp102 = tmp101 * tmp11
tmp103 = tmp100 + tmp102
tmp105 = tmp104 * tmp16
tmp106 = tmp103 + tmp105
tmp107 = tmp96 * tmp20
tmp108 = tmp98 * tmp23
tmp109 = tmp107 + tmp108
tmp110 = tmp101 * tmp27
tmp111 = tmp109 + tmp110
tmp112 = tmp104 * tmp31
tmp113 = tmp111 + tmp112
tmp114 = tmp96 * tmp35
tmp115 = tmp98 * tmp38
tmp116 = tmp114 + tmp115
tmp117 = tmp101 * tmp42
tmp118 = tmp116 + tmp117
tmp119 = tmp104 * tmp46
tmp120 = tmp118 + tmp119
tmp121 = tmp96 * tmp50
tmp122 = tmp98 * tmp53
tmp123 = tmp121 + tmp122
tmp124 = tmp101 * tmp57
tmp125 = tmp123 + tmp124
tmp126 = tmp104 * tmp61
tmp127 = tmp125 + tmp126
tmp129 = tmp128 * tmp2
tmp131 = tmp130 * tmp6
tmp132 = tmp129 + tmp131
tmp134 = tmp133 * tmp11
tmp135 = tmp132 + tmp134
tmp137 = tmp136 * tmp16
tmp138 = tmp135 + tmp137
tmp139 = tmp128 * tmp20
tmp140 = tmp130 * tmp23
tmp141 = tmp139 + tmp140
tmp142 = tmp133 * tmp27
tmp143 = tmp141 + tmp142
tmp144 = tmp136 * tmp31
tmp145 = tmp143 + tmp144
tmp146 = tmp128 * tmp35
tmp147 = tmp130 * tmp38
tmp148 = tmp146 + tmp147
tmp149 = tmp133 * tmp42
tmp150 = tmp148 + tmp149
tmp151 = tmp136 * tmp46
tmp152 = tmp150 + tmp151
tmp153 = tmp128 * tmp50
tmp154 = tmp130 * tmp53
tmp155 = tmp153 + tmp154
tmp156 = tmp133 * tmp57
tmp157 = tmp155 + tmp156
tmp158 = tmp136 * tmp61
tmp159 = tmp157 + tmp158
tl.store(out_ptr0 + (x0), tmp18, xmask)
tl.store(out_ptr1 + (x0), tmp33, xmask)
tl.store(out_ptr2 + (x0), tmp48, xmask)
tl.store(out_ptr3 + (x0), tmp63, xmask)
tl.store(out_ptr4 + (x0), tmp74, xmask)
tl.store(out_ptr5 + (x0), tmp81, xmask)
tl.store(out_ptr6 + (x0), tmp88, xmask)
tl.store(out_ptr7 + (x0), tmp95, xmask)
tl.store(out_ptr8 + (x0), tmp106, xmask)
tl.store(out_ptr9 + (x0), tmp113, xmask)
tl.store(out_ptr10 + (x0), tmp120, xmask)
tl.store(out_ptr11 + (x0), tmp127, xmask)
tl.store(out_ptr12 + (x0), tmp138, xmask)
tl.store(out_ptr13 + (x0), tmp145, xmask)
tl.store(out_ptr14 + (x0), tmp152, xmask)
tl.store(out_ptr15 + (x0), tmp159, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/cj/ccj7ibskau5v7hw4pezkintmudm3aq3f4rk77cxss22osizkdxmi.py
# Topologically Sorted Source Nodes: [matmul, add, add_1, g_t, matmul_2, add_2, add_3, i_t, matmul_4, add_4, add_5, f_t, matmul_6, add_6, add_7, o_t, mul, mul_1, c_t, tanh_1, h_t], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# add_3 => add_3
# add_4 => add_4
# add_5 => add_5
# add_6 => add_6
# add_7 => add_7
# c_t => add_8
# f_t => sigmoid_1
# g_t => tanh
# h_t => mul_6
# i_t => sigmoid
# matmul => mul, sum_1
# matmul_2 => mul_1, sum_2
# matmul_4 => mul_2, sum_3
# matmul_6 => mul_3, sum_4
# mul => mul_4
# mul_1 => mul_5
# o_t => sigmoid_2
# tanh_1 => tanh_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1]), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_1, %mm), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %primals_6), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_1,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [1]), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, %mm_1), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %primals_9), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_3,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_2, [1]), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_3, %mm_2), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_4, %primals_12), kwargs = {})
# %sigmoid_1 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_5,), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_3, [1]), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_4, %mm_3), kwargs = {})
# %add_7 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_6, %primals_15), kwargs = {})
# %sigmoid_2 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_7,), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh, %sigmoid), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %sigmoid_1), kwargs = {})
# %add_8 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %mul_5), kwargs = {})
# %tanh_1 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_8,), kwargs = {})
# %mul_6 : [num_users=5] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_1, %sigmoid_2), kwargs = {})
triton_poi_fused_add_mul_mv_sigmoid_tanh_1 = async_compile.triton('triton_poi_fused_add_mul_mv_sigmoid_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=[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: '*fp32', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*fp32', 14: '*fp32', 15: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_mv_sigmoid_tanh_1', '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_add_mul_mv_sigmoid_tanh_1(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, 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), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr6 + (x1), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr7 + (x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr8 + (x1), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr9 + (x1), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr10 + (x0), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr11 + (x1), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr12 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(out_ptr0 + (x2), tmp21, xmask)
tl.store(out_ptr1 + (x2), tmp29, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ky/ckytpgyuy3thu3tnd4vv6dduuwd72dfjt6nprs3fhrpj2mxtjdbm.py
# Topologically Sorted Source Nodes: [matmul_8, add_9, add_10, g_t_1, matmul_10, add_11, add_12, i_t_1, matmul_12, add_13, add_14, f_t_1, matmul_14, add_15, add_16, o_t_1, mul_3, mul_4, c_t_1, tanh_3, h_t_1], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# add_10 => add_10
# add_11 => add_11
# add_12 => add_12
# add_13 => add_13
# add_14 => add_14
# add_15 => add_15
# add_16 => add_16
# add_9 => add_9
# c_t_1 => add_17
# f_t_1 => sigmoid_4
# g_t_1 => tanh_2
# h_t_1 => mul_13
# i_t_1 => sigmoid_3
# matmul_10 => mul_8, sum_6
# matmul_12 => mul_9, sum_7
# matmul_14 => mul_10, sum_8
# matmul_8 => mul_7, sum_5
# mul_3 => mul_11
# mul_4 => mul_12
# o_t_1 => sigmoid_5
# tanh_3 => tanh_3
# Graph fragment:
# %mul_7 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select_4), kwargs = {})
# %sum_5 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_7, [1]), kwargs = {})
# %add_9 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_5, %mm_4), kwargs = {})
# %add_10 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_9, %primals_6), kwargs = {})
# %tanh_2 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_10,), kwargs = {})
# %mul_8 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select_4), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_8, [1]), kwargs = {})
# %add_11 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_6, %mm_5), kwargs = {})
# %add_12 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_11, %primals_9), kwargs = {})
# %sigmoid_3 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_12,), kwargs = {})
# %mul_9 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select_4), kwargs = {})
# %sum_7 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_9, [1]), kwargs = {})
# %add_13 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_7, %mm_6), kwargs = {})
# %add_14 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_13, %primals_12), kwargs = {})
# %sigmoid_4 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_14,), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select_4), kwargs = {})
# %sum_8 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_10, [1]), kwargs = {})
# %add_15 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_8, %mm_7), kwargs = {})
# %add_16 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_15, %primals_15), kwargs = {})
# %sigmoid_5 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_16,), kwargs = {})
# %mul_11 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_2, %sigmoid_3), kwargs = {})
# %mul_12 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_8, %sigmoid_4), kwargs = {})
# %add_17 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_11, %mul_12), kwargs = {})
# %tanh_3 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_17,), kwargs = {})
# %mul_13 : [num_users=5] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_3, %sigmoid_5), kwargs = {})
triton_poi_fused_add_mul_mv_sigmoid_tanh_2 = async_compile.triton('triton_poi_fused_add_mul_mv_sigmoid_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=[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: '*fp32', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*fp32', 14: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_mv_sigmoid_tanh_2', 'mutated_arg_names': ['in_out_ptr0'], '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_add_mul_mv_sigmoid_tanh_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9, in_ptr10, in_ptr11, 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
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), xmask)
tmp9 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_out_ptr0 + (x2), xmask)
tmp14 = tl.load(in_ptr6 + (x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr7 + (x2), xmask)
tmp17 = tl.load(in_ptr8 + (x1), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr9 + (x0), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr10 + (x2), xmask)
tmp26 = tl.load(in_ptr11 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(in_out_ptr0 + (x2), tmp21, xmask)
tl.store(out_ptr0 + (x2), tmp29, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/4n/c4n4xwkgnc6bupibp6fgulwliufexzfbh475qho4dizfkudweotu.py
# Topologically Sorted Source Nodes: [matmul_16, add_18, add_19, g_t_2, matmul_18, add_20, add_21, i_t_2, matmul_20, add_22, add_23, f_t_2, matmul_22, add_24, add_25, o_t_2, mul_6, mul_7, c_t_2, tanh_5, h_t_2], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# add_18 => add_18
# add_19 => add_19
# add_20 => add_20
# add_21 => add_21
# add_22 => add_22
# add_23 => add_23
# add_24 => add_24
# add_25 => add_25
# c_t_2 => add_26
# f_t_2 => sigmoid_7
# g_t_2 => tanh_4
# h_t_2 => mul_20
# i_t_2 => sigmoid_6
# matmul_16 => mul_14, sum_9
# matmul_18 => mul_15, sum_10
# matmul_20 => mul_16, sum_11
# matmul_22 => mul_17, sum_12
# mul_6 => mul_18
# mul_7 => mul_19
# o_t_2 => sigmoid_8
# tanh_5 => tanh_5
# Graph fragment:
# %mul_14 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select_8), kwargs = {})
# %sum_9 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_14, [1]), kwargs = {})
# %add_18 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_9, %mm_8), kwargs = {})
# %add_19 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_18, %primals_6), kwargs = {})
# %tanh_4 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_19,), kwargs = {})
# %mul_15 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select_8), kwargs = {})
# %sum_10 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_15, [1]), kwargs = {})
# %add_20 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_10, %mm_9), kwargs = {})
# %add_21 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_20, %primals_9), kwargs = {})
# %sigmoid_6 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_21,), kwargs = {})
# %mul_16 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select_8), kwargs = {})
# %sum_11 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_16, [1]), kwargs = {})
# %add_22 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_11, %mm_10), kwargs = {})
# %add_23 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_22, %primals_12), kwargs = {})
# %sigmoid_7 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_23,), kwargs = {})
# %mul_17 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select_8), kwargs = {})
# %sum_12 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_17, [1]), kwargs = {})
# %add_24 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_12, %mm_11), kwargs = {})
# %add_25 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_24, %primals_15), kwargs = {})
# %sigmoid_8 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_25,), kwargs = {})
# %mul_18 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_4, %sigmoid_6), kwargs = {})
# %mul_19 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_17, %sigmoid_7), kwargs = {})
# %add_26 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_18, %mul_19), kwargs = {})
# %tanh_5 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_26,), kwargs = {})
# %mul_20 : [num_users=5] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_5, %sigmoid_8), kwargs = {})
triton_poi_fused_add_mul_mv_sigmoid_tanh_3 = async_compile.triton('triton_poi_fused_add_mul_mv_sigmoid_tanh_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: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*fp32', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*fp32', 14: '*fp32', 15: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_mv_sigmoid_tanh_3', '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_add_mul_mv_sigmoid_tanh_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, 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
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), xmask)
tmp9 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr6 + (x2), xmask)
tmp14 = tl.load(in_ptr7 + (x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr8 + (x2), xmask)
tmp17 = tl.load(in_ptr9 + (x1), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr10 + (x0), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr11 + (x2), xmask)
tmp26 = tl.load(in_ptr12 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(out_ptr0 + (x2), tmp21, xmask)
tl.store(out_ptr1 + (x2), tmp29, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/4z/c4zbjn2uiqgzzxararkrcjsoluzfrhtjkkp6borjpdjra3jzima5.py
# Topologically Sorted Source Nodes: [matmul_24, add_27, add_28, g_t_3, matmul_26, add_29, add_30, i_t_3, matmul_28, add_31, add_32, f_t_3, matmul_30, add_33, add_34, o_t_3, mul_9, mul_10, c_t_3, tanh_7, h_t_3], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# add_27 => add_27
# add_28 => add_28
# add_29 => add_29
# add_30 => add_30
# add_31 => add_31
# add_32 => add_32
# add_33 => add_33
# add_34 => add_34
# c_t_3 => add_35
# f_t_3 => sigmoid_10
# g_t_3 => tanh_6
# h_t_3 => mul_27
# i_t_3 => sigmoid_9
# matmul_24 => mul_21, sum_13
# matmul_26 => mul_22, sum_14
# matmul_28 => mul_23, sum_15
# matmul_30 => mul_24, sum_16
# mul_10 => mul_26
# mul_9 => mul_25
# o_t_3 => sigmoid_11
# tanh_7 => tanh_7
# Graph fragment:
# %mul_21 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %select_12), kwargs = {})
# %sum_13 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_21, [1]), kwargs = {})
# %add_27 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_13, %mm_12), kwargs = {})
# %add_28 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_27, %primals_6), kwargs = {})
# %tanh_6 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_28,), kwargs = {})
# %mul_22 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_7, %select_12), kwargs = {})
# %sum_14 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_22, [1]), kwargs = {})
# %add_29 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_14, %mm_13), kwargs = {})
# %add_30 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_29, %primals_9), kwargs = {})
# %sigmoid_9 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_30,), kwargs = {})
# %mul_23 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_10, %select_12), kwargs = {})
# %sum_15 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_23, [1]), kwargs = {})
# %add_31 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_15, %mm_14), kwargs = {})
# %add_32 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_31, %primals_12), kwargs = {})
# %sigmoid_10 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_32,), kwargs = {})
# %mul_24 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_13, %select_12), kwargs = {})
# %sum_16 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_24, [1]), kwargs = {})
# %add_33 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_16, %mm_15), kwargs = {})
# %add_34 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_33, %primals_15), kwargs = {})
# %sigmoid_11 : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add_34,), kwargs = {})
# %mul_25 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_6, %sigmoid_9), kwargs = {})
# %mul_26 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_26, %sigmoid_10), kwargs = {})
# %add_35 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_25, %mul_26), kwargs = {})
# %tanh_7 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_35,), kwargs = {})
# %mul_27 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_7, %sigmoid_11), kwargs = {})
triton_poi_fused_add_mul_mv_sigmoid_tanh_4 = async_compile.triton('triton_poi_fused_add_mul_mv_sigmoid_tanh_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: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*fp32', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*fp32', 14: '*fp32', 15: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_mv_sigmoid_tanh_4', 'mutated_arg_names': ['in_out_ptr0'], '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_add_mul_mv_sigmoid_tanh_4(in_out_ptr0, 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, 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
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), xmask)
tmp9 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr6 + (x2), xmask)
tmp14 = tl.load(in_ptr7 + (x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr8 + (x2), xmask)
tmp17 = tl.load(in_ptr9 + (x1), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr10 + (x0), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr11 + (x2), xmask)
tmp26 = tl.load(in_ptr12 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(in_out_ptr0 + (x2), tmp22, xmask)
tl.store(out_ptr0 + (x2), tmp29, 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, 1), (1, 1))
assert_size_stride(primals_2, (4, 1), (1, 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, 4), (4, 1))
assert_size_stride(primals_6, (4, 1), (1, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, 1), (1, 1))
assert_size_stride(primals_10, (4, 4), (4, 1))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4, 1), (1, 1))
assert_size_stride(primals_13, (4, 4), (4, 1))
assert_size_stride(primals_14, (4, 4), (4, 1))
assert_size_stride(primals_15, (4, 1), (1, 1))
assert_size_stride(primals_16, (4, 4), (4, 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, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, primals_1, out=buf0)
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf11 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf21 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf31 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf3 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf13 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf23 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf33 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf5 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf15 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf25 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf35 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf7 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf17 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf27 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf37 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [matmul, matmul_2, matmul_4, matmul_6, matmul_8, matmul_10, matmul_12, matmul_14, matmul_16, matmul_18, matmul_20, matmul_22, matmul_24, matmul_26, matmul_28, matmul_30], Original ATen: [aten.mv]
stream0 = get_raw_stream(0)
triton_poi_fused_mv_0.run(primals_3, primals_4, primals_7, primals_10, primals_13, buf1, buf11, buf21, buf31, buf3, buf13, buf23, buf33, buf5, buf15, buf25, buf35, buf7, buf17, buf27, buf37, 4, grid=grid(4), stream=stream0)
buf2 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_3], Original ATen: [aten.mm]
extern_kernels.mm(primals_8, primals_1, out=buf2)
buf4 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_5], Original ATen: [aten.mm]
extern_kernels.mm(primals_11, primals_1, out=buf4)
buf6 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_7], Original ATen: [aten.mm]
extern_kernels.mm(primals_14, primals_1, out=buf6)
buf8 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf9 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul, add, add_1, g_t, matmul_2, add_2, add_3, i_t, matmul_4, add_4, add_5, f_t, matmul_6, add_6, add_7, o_t, mul, mul_1, c_t, tanh_1, h_t], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
triton_poi_fused_add_mul_mv_sigmoid_tanh_1.run(buf1, buf0, primals_6, buf3, buf2, primals_9, primals_2, buf5, buf4, primals_12, buf7, buf6, primals_15, buf8, buf9, 16, grid=grid(16), stream=stream0)
del buf1
del buf3
del buf5
del buf7
buf10 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_9], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, buf9, out=buf10)
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_11], Original ATen: [aten.mm]
extern_kernels.mm(primals_8, buf9, out=buf12)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_13], Original ATen: [aten.mm]
extern_kernels.mm(primals_11, buf9, out=buf14)
buf16 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_15], Original ATen: [aten.mm]
extern_kernels.mm(primals_14, buf9, out=buf16)
buf18 = buf8; del buf8 # reuse
buf19 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_8, add_9, add_10, g_t_1, matmul_10, add_11, add_12, i_t_1, matmul_12, add_13, add_14, f_t_1, matmul_14, add_15, add_16, o_t_1, mul_3, mul_4, c_t_1, tanh_3, h_t_1], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
triton_poi_fused_add_mul_mv_sigmoid_tanh_2.run(buf18, buf11, buf10, primals_6, buf13, buf12, primals_9, buf15, buf14, primals_12, buf17, buf16, primals_15, buf19, 16, grid=grid(16), stream=stream0)
del buf11
del buf13
del buf15
del buf17
buf20 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_17], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, buf19, out=buf20)
buf22 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_19], Original ATen: [aten.mm]
extern_kernels.mm(primals_8, buf19, out=buf22)
buf24 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_21], Original ATen: [aten.mm]
extern_kernels.mm(primals_11, buf19, out=buf24)
buf26 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_23], Original ATen: [aten.mm]
extern_kernels.mm(primals_14, buf19, out=buf26)
buf28 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf29 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_16, add_18, add_19, g_t_2, matmul_18, add_20, add_21, i_t_2, matmul_20, add_22, add_23, f_t_2, matmul_22, add_24, add_25, o_t_2, mul_6, mul_7, c_t_2, tanh_5, h_t_2], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
triton_poi_fused_add_mul_mv_sigmoid_tanh_3.run(buf21, buf20, primals_6, buf23, buf22, primals_9, buf18, buf25, buf24, primals_12, buf27, buf26, primals_15, buf28, buf29, 16, grid=grid(16), stream=stream0)
del buf21
del buf23
del buf25
del buf27
buf30 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_25], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, buf29, out=buf30)
buf32 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_27], Original ATen: [aten.mm]
extern_kernels.mm(primals_8, buf29, out=buf32)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_29], Original ATen: [aten.mm]
extern_kernels.mm(primals_11, buf29, out=buf34)
buf36 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_31], Original ATen: [aten.mm]
extern_kernels.mm(primals_14, buf29, out=buf36)
buf38 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf39 = buf38; del buf38 # reuse
buf40 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_24, add_27, add_28, g_t_3, matmul_26, add_29, add_30, i_t_3, matmul_28, add_31, add_32, f_t_3, matmul_30, add_33, add_34, o_t_3, mul_9, mul_10, c_t_3, tanh_7, h_t_3], Original ATen: [aten.mv, aten.add, aten.tanh, aten.sigmoid, aten.mul]
triton_poi_fused_add_mul_mv_sigmoid_tanh_4.run(buf39, buf31, buf30, primals_6, buf33, buf32, primals_9, buf28, buf35, buf34, primals_12, buf37, buf36, primals_15, buf40, 16, grid=grid(16), stream=stream0)
del buf31
del buf33
del buf35
del buf37
buf41 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_30, add_33, add_34, o_t_3, h_t_3], Original ATen: [aten.mv, aten.add, aten.sigmoid, aten.mul]
extern_kernels.addmm(primals_17, primals_16, buf40, alpha=1, beta=1, out=buf41)
del primals_17
return (reinterpret_tensor(buf41, (4, 4), (1, 4), 0), primals_2, primals_3, primals_4, primals_6, primals_7, primals_9, primals_10, primals_12, primals_13, primals_15, buf0, buf2, buf4, buf6, buf10, buf12, buf14, buf16, buf18, buf20, buf22, buf24, buf26, buf28, buf30, buf32, buf34, buf36, buf39, reinterpret_tensor(primals_16, (4, 4), (1, 4), 0), reinterpret_tensor(buf40, (4, 4), (1, 4), 0), reinterpret_tensor(primals_14, (4, 4), (1, 4), 0), reinterpret_tensor(buf29, (4, 4), (1, 4), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), reinterpret_tensor(buf19, (4, 4), (1, 4), 0), reinterpret_tensor(buf9, (4, 4), (1, 4), 0), reinterpret_tensor(primals_1, (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, 1), (1, 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, 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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 1), (1, 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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((4, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((4, 4), (4, 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 torch.nn as nn
class LSTM(nn.Module):
def __init__(self, seq_length, input_dim, num_hidden, num_classes,
batch_size, device='cpu'):
super(LSTM, self).__init__()
self.seq_length = seq_length
self.h_init = nn.Parameter(torch.zeros(num_hidden, 1),
requires_grad=False)
self.c_init = nn.Parameter(torch.zeros(num_hidden, 1),
requires_grad=False)
self.w_gx = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_gh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_g = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ix = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_ih = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_i = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_fx = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_fh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_f = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ox = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_oh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_o = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ph = nn.Parameter(torch.Tensor(num_classes, num_hidden).
normal_(mean=0, std=0.0001))
self.b_p = nn.Parameter(torch.Tensor(num_classes, 1).zero_())
def forward(self, x):
h_t = self.h_init
c_t = self.c_init
tanh = nn.Tanh()
sigmoid = nn.Sigmoid()
for step in range(self.seq_length):
g_t = tanh(self.w_gx @ x[:, step].t() + self.w_gh @ h_t + self.b_g)
i_t = sigmoid(self.w_ix @ x[:, step].t() + self.w_ih @ h_t +
self.b_i)
f_t = sigmoid(self.w_fx @ x[:, step].t() + self.w_fh @ h_t +
self.b_f)
o_t = sigmoid(self.w_ox @ x[:, step].t() + self.w_oh @ h_t +
self.b_o)
c_t = g_t * i_t + c_t * f_t
h_t = tanh(c_t) * o_t
p_t = self.w_ph @ h_t + self.b_p
return p_t.t()
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'seq_length': 4, 'input_dim': 4, 'num_hidden': 4,
'num_classes': 4, 'batch_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 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_mv_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4,
out_ptr0, out_ptr1, out_ptr2, out_ptr3, out_ptr4, out_ptr5, out_ptr6,
out_ptr7, out_ptr8, out_ptr9, out_ptr10, out_ptr11, out_ptr12,
out_ptr13, out_ptr14, out_ptr15, 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')
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + 4)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp9 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + 8)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp14 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr1 + 12)
tmp16 = tl.broadcast_to(tmp15, [XBLOCK])
tmp19 = tl.load(in_ptr1 + 1)
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp22 = tl.load(in_ptr1 + 5)
tmp23 = tl.broadcast_to(tmp22, [XBLOCK])
tmp26 = tl.load(in_ptr1 + 9)
tmp27 = tl.broadcast_to(tmp26, [XBLOCK])
tmp30 = tl.load(in_ptr1 + 13)
tmp31 = tl.broadcast_to(tmp30, [XBLOCK])
tmp34 = tl.load(in_ptr1 + 2)
tmp35 = tl.broadcast_to(tmp34, [XBLOCK])
tmp37 = tl.load(in_ptr1 + 6)
tmp38 = tl.broadcast_to(tmp37, [XBLOCK])
tmp41 = tl.load(in_ptr1 + 10)
tmp42 = tl.broadcast_to(tmp41, [XBLOCK])
tmp45 = tl.load(in_ptr1 + 14)
tmp46 = tl.broadcast_to(tmp45, [XBLOCK])
tmp49 = tl.load(in_ptr1 + 3)
tmp50 = tl.broadcast_to(tmp49, [XBLOCK])
tmp52 = tl.load(in_ptr1 + 7)
tmp53 = tl.broadcast_to(tmp52, [XBLOCK])
tmp56 = tl.load(in_ptr1 + 11)
tmp57 = tl.broadcast_to(tmp56, [XBLOCK])
tmp60 = tl.load(in_ptr1 + 15)
tmp61 = tl.broadcast_to(tmp60, [XBLOCK])
tmp64 = tl.load(in_ptr2 + 4 * x0, xmask, eviction_policy='evict_last')
tmp66 = tl.load(in_ptr2 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp69 = tl.load(in_ptr2 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp72 = tl.load(in_ptr2 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp96 = tl.load(in_ptr3 + 4 * x0, xmask, eviction_policy='evict_last')
tmp98 = tl.load(in_ptr3 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp101 = tl.load(in_ptr3 + (2 + 4 * x0), xmask, eviction_policy=
'evict_last')
tmp104 = tl.load(in_ptr3 + (3 + 4 * x0), xmask, eviction_policy=
'evict_last')
tmp128 = tl.load(in_ptr4 + 4 * x0, xmask, eviction_policy='evict_last')
tmp130 = tl.load(in_ptr4 + (1 + 4 * x0), xmask, eviction_policy=
'evict_last')
tmp133 = tl.load(in_ptr4 + (2 + 4 * x0), xmask, eviction_policy=
'evict_last')
tmp136 = tl.load(in_ptr4 + (3 + 4 * x0), xmask, eviction_policy=
'evict_last')
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp21 = tmp0 * tmp20
tmp24 = tmp4 * tmp23
tmp25 = tmp21 + tmp24
tmp28 = tmp9 * tmp27
tmp29 = tmp25 + tmp28
tmp32 = tmp14 * tmp31
tmp33 = tmp29 + tmp32
tmp36 = tmp0 * tmp35
tmp39 = tmp4 * tmp38
tmp40 = tmp36 + tmp39
tmp43 = tmp9 * tmp42
tmp44 = tmp40 + tmp43
tmp47 = tmp14 * tmp46
tmp48 = tmp44 + tmp47
tmp51 = tmp0 * tmp50
tmp54 = tmp4 * tmp53
tmp55 = tmp51 + tmp54
tmp58 = tmp9 * tmp57
tmp59 = tmp55 + tmp58
tmp62 = tmp14 * tmp61
tmp63 = tmp59 + tmp62
tmp65 = tmp64 * tmp2
tmp67 = tmp66 * tmp6
tmp68 = tmp65 + tmp67
tmp70 = tmp69 * tmp11
tmp71 = tmp68 + tmp70
tmp73 = tmp72 * tmp16
tmp74 = tmp71 + tmp73
tmp75 = tmp64 * tmp20
tmp76 = tmp66 * tmp23
tmp77 = tmp75 + tmp76
tmp78 = tmp69 * tmp27
tmp79 = tmp77 + tmp78
tmp80 = tmp72 * tmp31
tmp81 = tmp79 + tmp80
tmp82 = tmp64 * tmp35
tmp83 = tmp66 * tmp38
tmp84 = tmp82 + tmp83
tmp85 = tmp69 * tmp42
tmp86 = tmp84 + tmp85
tmp87 = tmp72 * tmp46
tmp88 = tmp86 + tmp87
tmp89 = tmp64 * tmp50
tmp90 = tmp66 * tmp53
tmp91 = tmp89 + tmp90
tmp92 = tmp69 * tmp57
tmp93 = tmp91 + tmp92
tmp94 = tmp72 * tmp61
tmp95 = tmp93 + tmp94
tmp97 = tmp96 * tmp2
tmp99 = tmp98 * tmp6
tmp100 = tmp97 + tmp99
tmp102 = tmp101 * tmp11
tmp103 = tmp100 + tmp102
tmp105 = tmp104 * tmp16
tmp106 = tmp103 + tmp105
tmp107 = tmp96 * tmp20
tmp108 = tmp98 * tmp23
tmp109 = tmp107 + tmp108
tmp110 = tmp101 * tmp27
tmp111 = tmp109 + tmp110
tmp112 = tmp104 * tmp31
tmp113 = tmp111 + tmp112
tmp114 = tmp96 * tmp35
tmp115 = tmp98 * tmp38
tmp116 = tmp114 + tmp115
tmp117 = tmp101 * tmp42
tmp118 = tmp116 + tmp117
tmp119 = tmp104 * tmp46
tmp120 = tmp118 + tmp119
tmp121 = tmp96 * tmp50
tmp122 = tmp98 * tmp53
tmp123 = tmp121 + tmp122
tmp124 = tmp101 * tmp57
tmp125 = tmp123 + tmp124
tmp126 = tmp104 * tmp61
tmp127 = tmp125 + tmp126
tmp129 = tmp128 * tmp2
tmp131 = tmp130 * tmp6
tmp132 = tmp129 + tmp131
tmp134 = tmp133 * tmp11
tmp135 = tmp132 + tmp134
tmp137 = tmp136 * tmp16
tmp138 = tmp135 + tmp137
tmp139 = tmp128 * tmp20
tmp140 = tmp130 * tmp23
tmp141 = tmp139 + tmp140
tmp142 = tmp133 * tmp27
tmp143 = tmp141 + tmp142
tmp144 = tmp136 * tmp31
tmp145 = tmp143 + tmp144
tmp146 = tmp128 * tmp35
tmp147 = tmp130 * tmp38
tmp148 = tmp146 + tmp147
tmp149 = tmp133 * tmp42
tmp150 = tmp148 + tmp149
tmp151 = tmp136 * tmp46
tmp152 = tmp150 + tmp151
tmp153 = tmp128 * tmp50
tmp154 = tmp130 * tmp53
tmp155 = tmp153 + tmp154
tmp156 = tmp133 * tmp57
tmp157 = tmp155 + tmp156
tmp158 = tmp136 * tmp61
tmp159 = tmp157 + tmp158
tl.store(out_ptr0 + x0, tmp18, xmask)
tl.store(out_ptr1 + x0, tmp33, xmask)
tl.store(out_ptr2 + x0, tmp48, xmask)
tl.store(out_ptr3 + x0, tmp63, xmask)
tl.store(out_ptr4 + x0, tmp74, xmask)
tl.store(out_ptr5 + x0, tmp81, xmask)
tl.store(out_ptr6 + x0, tmp88, xmask)
tl.store(out_ptr7 + x0, tmp95, xmask)
tl.store(out_ptr8 + x0, tmp106, xmask)
tl.store(out_ptr9 + x0, tmp113, xmask)
tl.store(out_ptr10 + x0, tmp120, xmask)
tl.store(out_ptr11 + x0, tmp127, xmask)
tl.store(out_ptr12 + x0, tmp138, xmask)
tl.store(out_ptr13 + x0, tmp145, xmask)
tl.store(out_ptr14 + x0, tmp152, xmask)
tl.store(out_ptr15 + x0, tmp159, xmask)
@triton.jit
def triton_poi_fused_add_mul_mv_sigmoid_tanh_1(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, 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, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr6 + x1, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr7 + x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr8 + x1, xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr9 + x1, xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr10 + x0, xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr11 + x1, xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr12 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(out_ptr0 + x2, tmp21, xmask)
tl.store(out_ptr1 + x2, tmp29, xmask)
@triton.jit
def triton_poi_fused_add_mul_mv_sigmoid_tanh_2(in_out_ptr0, in_ptr0,
in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8,
in_ptr9, in_ptr10, in_ptr11, 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
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, xmask)
tmp9 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_out_ptr0 + x2, xmask)
tmp14 = tl.load(in_ptr6 + x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr7 + x2, xmask)
tmp17 = tl.load(in_ptr8 + x1, xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr9 + x0, xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr10 + x2, xmask)
tmp26 = tl.load(in_ptr11 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(in_out_ptr0 + x2, tmp21, xmask)
tl.store(out_ptr0 + x2, tmp29, xmask)
@triton.jit
def triton_poi_fused_add_mul_mv_sigmoid_tanh_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, 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
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, xmask)
tmp9 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr6 + x2, xmask)
tmp14 = tl.load(in_ptr7 + x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr8 + x2, xmask)
tmp17 = tl.load(in_ptr9 + x1, xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr10 + x0, xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr11 + x2, xmask)
tmp26 = tl.load(in_ptr12 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(out_ptr0 + x2, tmp21, xmask)
tl.store(out_ptr1 + x2, tmp29, xmask)
@triton.jit
def triton_poi_fused_add_mul_mv_sigmoid_tanh_4(in_out_ptr0, 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, 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
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, xmask)
tmp9 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr6 + x2, xmask)
tmp14 = tl.load(in_ptr7 + x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr8 + x2, xmask)
tmp17 = tl.load(in_ptr9 + x1, xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr10 + x0, xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr11 + x2, xmask)
tmp26 = tl.load(in_ptr12 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.sigmoid(tmp10)
tmp12 = tmp5 * tmp11
tmp16 = tmp14 + tmp15
tmp18 = tmp16 + tmp17
tmp19 = tl.sigmoid(tmp18)
tmp20 = tmp13 * tmp19
tmp21 = tmp12 + tmp20
tmp22 = libdevice.tanh(tmp21)
tmp25 = tmp23 + tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.sigmoid(tmp27)
tmp29 = tmp22 * tmp28
tl.store(in_out_ptr0 + x2, tmp22, xmask)
tl.store(out_ptr0 + x2, tmp29, 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, 1), (1, 1))
assert_size_stride(primals_2, (4, 1), (1, 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, 4), (4, 1))
assert_size_stride(primals_6, (4, 1), (1, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, 1), (1, 1))
assert_size_stride(primals_10, (4, 4), (4, 1))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4, 1), (1, 1))
assert_size_stride(primals_13, (4, 4), (4, 1))
assert_size_stride(primals_14, (4, 4), (4, 1))
assert_size_stride(primals_15, (4, 1), (1, 1))
assert_size_stride(primals_16, (4, 4), (4, 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, 1), (1, 1), torch.float32)
extern_kernels.mm(primals_5, primals_1, out=buf0)
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
buf11 = empty_strided_cuda((4,), (1,), torch.float32)
buf21 = empty_strided_cuda((4,), (1,), torch.float32)
buf31 = empty_strided_cuda((4,), (1,), torch.float32)
buf3 = empty_strided_cuda((4,), (1,), torch.float32)
buf13 = empty_strided_cuda((4,), (1,), torch.float32)
buf23 = empty_strided_cuda((4,), (1,), torch.float32)
buf33 = empty_strided_cuda((4,), (1,), torch.float32)
buf5 = empty_strided_cuda((4,), (1,), torch.float32)
buf15 = empty_strided_cuda((4,), (1,), torch.float32)
buf25 = empty_strided_cuda((4,), (1,), torch.float32)
buf35 = empty_strided_cuda((4,), (1,), torch.float32)
buf7 = empty_strided_cuda((4,), (1,), torch.float32)
buf17 = empty_strided_cuda((4,), (1,), torch.float32)
buf27 = empty_strided_cuda((4,), (1,), torch.float32)
buf37 = empty_strided_cuda((4,), (1,), torch.float32)
get_raw_stream(0)
triton_poi_fused_mv_0[grid(4)](primals_3, primals_4, primals_7,
primals_10, primals_13, buf1, buf11, buf21, buf31, buf3, buf13,
buf23, buf33, buf5, buf15, buf25, buf35, buf7, buf17, buf27,
buf37, 4, XBLOCK=4, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(primals_8, primals_1, out=buf2)
buf4 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(primals_11, primals_1, out=buf4)
buf6 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.mm(primals_14, primals_1, out=buf6)
buf8 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf9 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_mul_mv_sigmoid_tanh_1[grid(16)](buf1, buf0,
primals_6, buf3, buf2, primals_9, primals_2, buf5, buf4,
primals_12, buf7, buf6, primals_15, buf8, buf9, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf1
del buf3
del buf5
del buf7
buf10 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_5, buf9, out=buf10)
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_8, buf9, out=buf12)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_11, buf9, out=buf14)
buf16 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_14, buf9, out=buf16)
buf18 = buf8
del buf8
buf19 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_mul_mv_sigmoid_tanh_2[grid(16)](buf18, buf11,
buf10, primals_6, buf13, buf12, primals_9, buf15, buf14,
primals_12, buf17, buf16, primals_15, buf19, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf11
del buf13
del buf15
del buf17
buf20 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_5, buf19, out=buf20)
buf22 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_8, buf19, out=buf22)
buf24 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_11, buf19, out=buf24)
buf26 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_14, buf19, out=buf26)
buf28 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf29 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_mul_mv_sigmoid_tanh_3[grid(16)](buf21, buf20,
primals_6, buf23, buf22, primals_9, buf18, buf25, buf24,
primals_12, buf27, buf26, primals_15, buf28, buf29, 16, XBLOCK=
16, num_warps=1, num_stages=1)
del buf21
del buf23
del buf25
del buf27
buf30 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_5, buf29, out=buf30)
buf32 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_8, buf29, out=buf32)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_11, buf29, out=buf34)
buf36 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_14, buf29, out=buf36)
buf38 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf39 = buf38
del buf38
buf40 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_mul_mv_sigmoid_tanh_4[grid(16)](buf39, buf31,
buf30, primals_6, buf33, buf32, primals_9, buf28, buf35, buf34,
primals_12, buf37, buf36, primals_15, buf40, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf31
del buf33
del buf35
del buf37
buf41 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_17, primals_16, buf40, alpha=1, beta=1,
out=buf41)
del primals_17
return (reinterpret_tensor(buf41, (4, 4), (1, 4), 0), primals_2,
primals_3, primals_4, primals_6, primals_7, primals_9, primals_10,
primals_12, primals_13, primals_15, buf0, buf2, buf4, buf6, buf10,
buf12, buf14, buf16, buf18, buf20, buf22, buf24, buf26, buf28,
buf30, buf32, buf34, buf36, buf39, reinterpret_tensor(primals_16, (
4, 4), (1, 4), 0), reinterpret_tensor(buf40, (4, 4), (1, 4), 0),
reinterpret_tensor(primals_14, (4, 4), (1, 4), 0),
reinterpret_tensor(buf29, (4, 4), (1, 4), 0), reinterpret_tensor(
primals_11, (4, 4), (1, 4), 0), reinterpret_tensor(primals_8, (4, 4
), (1, 4), 0), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0),
reinterpret_tensor(buf19, (4, 4), (1, 4), 0), reinterpret_tensor(
buf9, (4, 4), (1, 4), 0), reinterpret_tensor(primals_1, (1, 4), (1,
1), 0))
class LSTMNew(nn.Module):
def __init__(self, seq_length, input_dim, num_hidden, num_classes,
batch_size, device='cpu'):
super(LSTMNew, self).__init__()
self.seq_length = seq_length
self.h_init = nn.Parameter(torch.zeros(num_hidden, 1),
requires_grad=False)
self.c_init = nn.Parameter(torch.zeros(num_hidden, 1),
requires_grad=False)
self.w_gx = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_gh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_g = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ix = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_ih = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_i = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_fx = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_fh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_f = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ox = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_oh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_o = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ph = nn.Parameter(torch.Tensor(num_classes, num_hidden).
normal_(mean=0, std=0.0001))
self.b_p = nn.Parameter(torch.Tensor(num_classes, 1).zero_())
def forward(self, input_0):
primals_1 = self.h_init
primals_2 = self.c_init
primals_3 = self.w_gx
primals_4 = self.w_gh
primals_6 = self.b_g
primals_5 = self.w_ix
primals_7 = self.w_ih
primals_9 = self.b_i
primals_8 = self.w_fx
primals_10 = self.w_fh
primals_12 = self.b_f
primals_11 = self.w_ox
primals_13 = self.w_oh
primals_15 = self.b_o
primals_14 = self.w_ph
primals_17 = self.b_p
primals_16 = 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])
return output[0]
| askliar/deep_learning | LSTM | false | 1,569 | [
"MIT"
] | 0 | e61b2391a3258d18719bf12d9ed1404620ce6c02 | https://github.com/askliar/deep_learning/tree/e61b2391a3258d18719bf12d9ed1404620ce6c02 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, seq_length, input_dim, num_hidden, num_classes,
batch_size, device='cpu'):
super().__init__()
self.seq_length = seq_length
self.h_init = nn.Parameter(torch.zeros(num_hidden, 1),
requires_grad=False)
self.c_init = nn.Parameter(torch.zeros(num_hidden, 1),
requires_grad=False)
self.w_gx = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_gh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_g = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ix = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_ih = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_i = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_fx = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_fh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_f = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ox = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, input_dim).normal_(mean=0, std=0.0001)))
self.w_oh = nn.Parameter(nn.init.orthogonal_(torch.Tensor(
num_hidden, num_hidden).normal_(mean=0, std=0.0001)))
self.b_o = nn.Parameter(torch.Tensor(num_hidden, 1).zero_())
self.w_ph = nn.Parameter(torch.Tensor(num_classes, num_hidden).
normal_(mean=0, std=0.0001))
self.b_p = nn.Parameter(torch.Tensor(num_classes, 1).zero_())
def forward(self, x):
h_t = self.h_init
c_t = self.c_init
tanh = nn.Tanh()
sigmoid = nn.Sigmoid()
for step in range(self.seq_length):
g_t = tanh(self.w_gx @ x[:, step].t() + self.w_gh @ h_t + self.b_g)
i_t = sigmoid(self.w_ix @ x[:, step].t() + self.w_ih @ h_t +
self.b_i)
f_t = sigmoid(self.w_fx @ x[:, step].t() + self.w_fh @ h_t +
self.b_f)
o_t = sigmoid(self.w_ox @ x[:, step].t() + self.w_oh @ h_t +
self.b_o)
c_t = g_t * i_t + c_t * f_t
h_t = tanh(c_t) * o_t
p_t = self.w_ph @ h_t + self.b_p
return p_t.t()
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'seq_length': 4, 'input_dim': 4, 'num_hidden': 4,
'num_classes': 4, 'batch_size': 4}]
|
PatchEmbed | # 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_6/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_6/inductor_cache/qy/cqyu5l2p6xh633a7thd2tte3bszrg4ugscf2y523iookhmpheqal.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=[4096, 256], 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, 3), 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 = 2304
xnumel = 256
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 + (256*y3)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (3*x2) + (768*y1)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/4c/c4ckui43udehobca2kb3vy5stpaqfztmtjwrdinx2dhmcmh73fmo.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_1, %primals_2, %primals_3, [16, 16], [0, 0], [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=[4096, 16], 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, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_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_2(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 3072
xnumel = 16
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
y0 = yindex % 768
y1 = (yindex // 768)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (768*x2) + (12288*y1)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (16*y3)), 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, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_2, (768, 3, 16, 16), (768, 256, 16, 1))
assert_size_stride(primals_3, (768, ), (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((768, 3, 16, 16), (768, 1, 48, 3), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(primals_2, buf1, 2304, 256, grid=grid(2304, 256), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf0, buf1, stride=(16, 16), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 768, 4, 4), (12288, 1, 3072, 768))
buf3 = empty_strided_cuda((4, 768, 4, 4), (12288, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf2, primals_3, buf3, 3072, 16, grid=grid(3072, 16), stream=stream0)
del buf2
del primals_3
return (buf3, 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, 3, 64, 64), (12288, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((768, 3, 16, 16), (768, 256, 16, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((768, ), (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
from torch import optim as optim
class PatchEmbed(nn.Module):
""" Image to Patch Embedding
"""
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
super().__init__()
num_patches = img_size // patch_size * (img_size // patch_size)
self.img_size = img_size
self.patch_size = patch_size
self.num_patches = num_patches
self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size,
stride=patch_size)
def forward(self, x):
_B, _C, _H, _W = x.shape
return self.proj(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
import torch.nn as nn
from torch import optim as optim
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_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 = 2304
xnumel = 256
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 + 256 * y3), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + 3 * x2 + 768 * y1), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_convolution_2(in_ptr0, in_ptr1, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
xnumel = 16
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
y0 = yindex % 768
y1 = yindex // 768
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 768 * x2 + 12288 * y1), xmask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 16 * y3), tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_2, (768, 3, 16, 16), (768, 256, 16, 1))
assert_size_stride(primals_3, (768,), (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((768, 3, 16, 16), (768, 1, 48, 3), torch.
float32)
triton_poi_fused_1[grid(2304, 256)](primals_2, buf1, 2304, 256,
XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf0, buf1, stride=(16, 16),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 768, 4, 4), (12288, 1, 3072, 768))
buf3 = empty_strided_cuda((4, 768, 4, 4), (12288, 16, 4, 1), torch.
float32)
triton_poi_fused_convolution_2[grid(3072, 16)](buf2, primals_3,
buf3, 3072, 16, XBLOCK=16, YBLOCK=32, num_warps=4, num_stages=1)
del buf2
del primals_3
return buf3, buf0, buf1
class PatchEmbedNew(nn.Module):
""" Image to Patch Embedding
"""
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
super().__init__()
num_patches = img_size // patch_size * (img_size // patch_size)
self.img_size = img_size
self.patch_size = patch_size
self.num_patches = num_patches
self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size,
stride=patch_size)
def forward(self, input_0):
primals_2 = self.proj.weight
primals_3 = self.proj.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| billpsomas/ibot | PatchEmbed | false | 1,570 | [
"Apache-2.0"
] | 0 | c6fbce7e2a59780f39ad7304ed9a8b1acf038d2d | https://github.com/billpsomas/ibot/tree/c6fbce7e2a59780f39ad7304ed9a8b1acf038d2d | import torch
import torch.nn as nn
from torch import optim as optim
class Model(nn.Module):
""" Image to Patch Embedding
"""
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
super().__init__()
num_patches = img_size // patch_size * (img_size // patch_size)
self.img_size = img_size
self.patch_size = patch_size
self.num_patches = num_patches
self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size,
stride=patch_size)
def forward(self, x):
_B, _C, _H, _W = x.shape
return self.proj(x)
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return []
|
FocalLoss | # 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_6/inductor_cache/2z/c2zkgo2rcqzti4uqgkziitlj7cnt42um35gbg4hoe26wdw75g6ug.py
# Topologically Sorted Source Nodes: [ge, probas, sub, pow_1, mul, bce_loss, mul_1, pow_2, mul_2, loss, loss_1], Original ATen: [aten.ge, aten.sigmoid, aten.rsub, aten.pow, aten.mul, aten.binary_cross_entropy_with_logits, aten.where, aten.mean]
# Source node to ATen node mapping:
# bce_loss => abs_1, exp, full_default, log1p, minimum, mul, neg, sub, sub_1, sub_2
# ge => ge
# loss => where
# loss_1 => mean
# mul => mul_1
# mul_1 => mul_2
# mul_2 => mul_3
# pow_1 => pow_1
# pow_2 => pow_2
# probas => sigmoid
# sub => sub_3
# Graph fragment:
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%arg0_1, 0.5), kwargs = {})
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%arg1_1,), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %sigmoid), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_3, 2.0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_1, 0.25), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg0_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %arg1_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), 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 = (%full_default, %arg1_1), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%arg1_1,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %sub_2 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %sub_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, %sub_2), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sigmoid, 2.0), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_2, %sub_2), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%ge, %mul_2, %mul_3), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%where,), kwargs = {})
triton_per_fused_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_0 = async_compile.triton('triton_per_fused_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_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_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_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)
tmp3 = tl.load(in_ptr1 + (r0), None)
tmp1 = 0.5
tmp2 = tmp0 >= tmp1
tmp4 = tl.sigmoid(tmp3)
tmp5 = 1.0
tmp6 = tmp5 - tmp4
tmp7 = tmp6 * tmp6
tmp8 = 0.25
tmp9 = tmp7 * tmp8
tmp10 = tmp5 - tmp0
tmp11 = tmp10 * tmp3
tmp12 = 0.0
tmp13 = triton_helpers.minimum(tmp12, tmp3)
tmp14 = tl_math.abs(tmp3)
tmp15 = -tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = libdevice.log1p(tmp16)
tmp18 = tmp13 - tmp17
tmp19 = tmp11 - tmp18
tmp20 = tmp9 * tmp19
tmp21 = tmp4 * tmp4
tmp22 = tmp21 * tmp19
tmp23 = tl.where(tmp2, tmp20, tmp22)
tmp24 = tl.broadcast_to(tmp23, [RBLOCK])
tmp26 = triton_helpers.promote_to_tensor(tl.sum(tmp24, 0))
tmp27 = 256.0
tmp28 = tmp26 / tmp27
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp28, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [ge, probas, sub, pow_1, mul, bce_loss, mul_1, pow_2, mul_2, loss, loss_1], Original ATen: [aten.ge, aten.sigmoid, aten.rsub, aten.pow, aten.mul, aten.binary_cross_entropy_with_logits, aten.where, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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 FocalLoss(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.bce = nn.BCEWithLogitsLoss(reduction='none')
self.alpha = alpha
self.gamma = gamma
if reduction == 'mean':
self.reduction = torch.mean
else:
self.reduction = torch.sum
def forward(self, preds, targets):
bce_loss = self.bce(preds, targets)
probas = torch.sigmoid(preds)
loss = torch.where(targets >= 0.5, self.alpha * (1.0 - probas) **
self.gamma * bce_loss, probas ** self.gamma * bce_loss)
loss = self.reduction(loss)
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 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_per_fused_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_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)
tmp3 = tl.load(in_ptr1 + r0, None)
tmp1 = 0.5
tmp2 = tmp0 >= tmp1
tmp4 = tl.sigmoid(tmp3)
tmp5 = 1.0
tmp6 = tmp5 - tmp4
tmp7 = tmp6 * tmp6
tmp8 = 0.25
tmp9 = tmp7 * tmp8
tmp10 = tmp5 - tmp0
tmp11 = tmp10 * tmp3
tmp12 = 0.0
tmp13 = triton_helpers.minimum(tmp12, tmp3)
tmp14 = tl_math.abs(tmp3)
tmp15 = -tmp14
tmp16 = tl_math.exp(tmp15)
tmp17 = libdevice.log1p(tmp16)
tmp18 = tmp13 - tmp17
tmp19 = tmp11 - tmp18
tmp20 = tmp9 * tmp19
tmp21 = tmp4 * tmp4
tmp22 = tmp21 * tmp19
tmp23 = tl.where(tmp2, tmp20, tmp22)
tmp24 = tl.broadcast_to(tmp23, [RBLOCK])
tmp26 = triton_helpers.promote_to_tensor(tl.sum(tmp24, 0))
tmp27 = 256.0
tmp28 = tmp26 / tmp27
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp28, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_ge_mean_mul_pow_rsub_sigmoid_where_0[
grid(1)](buf1, arg0_1, arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class FocalLossNew(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.bce = nn.BCEWithLogitsLoss(reduction='none')
self.alpha = alpha
self.gamma = gamma
if reduction == 'mean':
self.reduction = torch.mean
else:
self.reduction = torch.sum
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| botkop/lark | FocalLoss | false | 1,571 | [
"Apache-2.0"
] | 0 | edb2defdb514213fc121418578b0d9006a55f3a0 | https://github.com/botkop/lark/tree/edb2defdb514213fc121418578b0d9006a55f3a0 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.bce = nn.BCEWithLogitsLoss(reduction='none')
self.alpha = alpha
self.gamma = gamma
if reduction == 'mean':
self.reduction = torch.mean
else:
self.reduction = torch.sum
def forward(self, preds, targets):
bce_loss = self.bce(preds, targets)
probas = torch.sigmoid(preds)
loss = torch.where(targets >= 0.5, self.alpha * (1.0 - probas) **
self.gamma * bce_loss, probas ** self.gamma * bce_loss)
loss = self.reduction(loss)
return loss
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SReLU | # 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_6/inductor_cache/wt/cwtbm7ruscnyd5hf7etfdfwptqhgjo73ir43mntx3mmz3qyymrrg.py
# Topologically Sorted Source Nodes: [sub, relu, add], Original ATen: [aten.sub, aten.relu, aten.add, aten.threshold_backward]
# Source node to ATen node mapping:
# add => add
# relu => relu
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_2, %primals_1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%sub,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%relu, %primals_1), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_sub_threshold_backward_0 = async_compile.triton('triton_poi_fused_add_relu_sub_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: '*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_add_relu_sub_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_add_relu_sub_threshold_backward_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
x3 = xindex
x1 = (xindex // 16) % 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 = tmp4 + tmp1
tmp6 = 0.0
tmp7 = tmp4 <= tmp6
tl.store(out_ptr0 + (x3), tmp5, 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 = args
args.clear()
assert_size_stride(primals_1, (1, 4, 1, 1), (4, 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, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [sub, relu, add], Original ATen: [aten.sub, aten.relu, aten.add, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_add_relu_sub_threshold_backward_0.run(primals_2, primals_1, buf0, buf1, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_2
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
primals_1 = rand_strided((1, 4, 1, 1), (4, 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
class SReLU(nn.Module):
"""Shifted ReLU"""
def __init__(self, nc):
super(SReLU, self).__init__()
self.srelu_bias = nn.Parameter(torch.Tensor(1, nc, 1, 1))
self.srelu_relu = nn.ReLU(inplace=True)
nn.init.constant_(self.srelu_bias, -1.0)
def forward(self, x):
return self.srelu_relu(x - self.srelu_bias) + self.srelu_bias
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nc': 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
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_relu_sub_threshold_backward_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
x3 = xindex
x1 = xindex // 16 % 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 = tmp4 + tmp1
tmp6 = 0.0
tmp7 = tmp4 <= tmp6
tl.store(out_ptr0 + x3, tmp5, xmask)
tl.store(out_ptr1 + x3, tmp7, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (1, 4, 1, 1), (4, 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, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_relu_sub_threshold_backward_0[grid(256)](primals_2
, primals_1, buf0, buf1, 256, XBLOCK=256, num_warps=4, num_stages=1
)
del primals_1
del primals_2
return buf0, buf1
class SReLUNew(nn.Module):
"""Shifted ReLU"""
def __init__(self, nc):
super(SReLUNew, self).__init__()
self.srelu_bias = nn.Parameter(torch.Tensor(1, nc, 1, 1))
self.srelu_relu = nn.ReLU(inplace=True)
nn.init.constant_(self.srelu_bias, -1.0)
def forward(self, input_0):
primals_1 = self.srelu_bias
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
| bpicnbnk/ISONet | SReLU | false | 1,572 | [
"MIT"
] | 0 | 993150fa215dbc92460142ade59a3a5ec4a0efd9 | https://github.com/bpicnbnk/ISONet/tree/993150fa215dbc92460142ade59a3a5ec4a0efd9 | import torch
import torch.nn as nn
class Model(nn.Module):
"""Shifted ReLU"""
def __init__(self, nc):
super().__init__()
self.srelu_bias = nn.Parameter(torch.Tensor(1, nc, 1, 1))
self.srelu_relu = nn.ReLU(inplace=True)
nn.init.constant_(self.srelu_bias, -1.0)
def forward(self, x):
return self.srelu_relu(x - self.srelu_bias) + self.srelu_bias
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
AdaptiveInstanceNorm | # 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_6/inductor_cache/or/corr3tkjnigsk2o6euqdjngfns555swzfvqhbtbmyxsccgvul5e6.py
# Topologically Sorted Source Nodes: [tensor, mul, sqrt, mul_1, weight], Original ATen: [aten.lift_fresh, aten.mul, aten.sqrt]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# sqrt => full_default_1
# tensor => full_default
# weight => mul_2
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 1.4142135381698608), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %full_default), kwargs = {})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.5), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %full_default_1), kwargs = {})
# %mul_2 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, 1.0), kwargs = {})
triton_poi_fused_lift_fresh_mul_sqrt_0 = async_compile.triton('triton_poi_fused_lift_fresh_mul_sqrt_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, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_lift_fresh_mul_sqrt_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_lift_fresh_mul_sqrt_0(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_ptr0 + (x0), xmask)
tmp1 = 1.4142135381698608
tmp2 = tmp0 * tmp1
tmp3 = 0.5
tmp4 = tmp2 * tmp3
tmp5 = 1.0
tmp6 = tmp4 * tmp5
tl.store(out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/jo/cjo3wxmtawsvu7opemz2xwvsknw4nxv74xivifhgb7csue6qqjbi.py
# Topologically Sorted Source Nodes: [out, mul_3, out_1], Original ATen: [aten._native_batch_norm_legit, aten.mul, aten.add]
# Source node to ATen node mapping:
# mul_3 => mul_4
# out => add, rsqrt, var_mean
# out_1 => add_1
# Graph fragment:
# %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_2, 1e-05), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%getitem, %view_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %getitem_1), kwargs = {})
triton_per_fused__native_batch_norm_legit_add_mul_1 = async_compile.triton('triton_per_fused__native_batch_norm_legit_add_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.persistent_reduction(
size_hints=[16, 16],
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_batch_norm_legit_add_mul_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, '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_mul_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, 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
x2 = xindex % 4
x3 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp22 = tl.load(in_ptr1 + (x2 + (8*x3)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (4 + x2 + (8*x3)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr2 + (4 + x2), xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = 16.0
tmp18 = tmp16 / tmp17
tmp19 = 1e-05
tmp20 = tmp18 + tmp19
tmp21 = libdevice.rsqrt(tmp20)
tmp24 = tmp22 + tmp23
tmp25 = tmp0 - tmp10
tmp26 = tmp25 * tmp21
tmp27 = tmp24 * tmp26
tmp30 = tmp28 + tmp29
tmp31 = tmp27 + tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp21, xmask)
tl.store(out_ptr1 + (r1 + (16*x0)), tmp31, xmask)
tl.store(out_ptr0 + (x0), tmp10, 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, (8, 4), (4, 1))
assert_size_stride(primals_2, (8, ), (1, ))
assert_size_stride(primals_3, (4, 4), (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((8, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [tensor, mul, sqrt, mul_1, weight], Original ATen: [aten.lift_fresh, aten.mul, aten.sqrt]
stream0 = get_raw_stream(0)
triton_poi_fused_lift_fresh_mul_sqrt_0.run(primals_1, buf0, 32, grid=grid(32), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(primals_3, reinterpret_tensor(buf0, (4, 8), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 1, 1), torch.float32)
buf3 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 16, 16), torch.float32)
buf5 = reinterpret_tensor(buf3, (1, 16, 1, 1), (16, 1, 1, 1), 0); del buf3 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out, mul_3, out_1], Original ATen: [aten._native_batch_norm_legit, aten.mul, aten.add]
triton_per_fused__native_batch_norm_legit_add_mul_1.run(buf5, primals_4, buf1, primals_2, buf2, buf6, 16, 16, grid=grid(16), stream=stream0)
del buf1
del primals_2
return (buf6, buf0, primals_3, primals_4, buf2, 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((8, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((8, ), (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, 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 torch.nn as nn
from torch.nn.init import _calculate_correct_fan
import torch.utils.cpp_extension
def equalized_lr(module, name='weight', gain=2 ** 0.5, mode='fan_in',
lr_mul=1.0):
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
EqualizedLR.apply(module, name, gain=gain, mode=mode, lr_mul=lr_mul)
return module
class EqualizedLR:
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
"""
def __init__(self, name='weight', gain=2 ** 0.5, mode='fan_in', lr_mul=1.0
):
self.name = name
self.mode = mode
self.gain = gain
self.lr_mul = lr_mul
def compute_weight(self, module):
"""Compute weight with equalized learning rate.
Args:
module (nn.Module): A module that is wrapped with equalized lr.
Returns:
torch.Tensor: Updated weight.
"""
weight = getattr(module, self.name + '_orig')
if weight.ndim == 5:
fan = _calculate_correct_fan(weight[0], self.mode)
else:
assert weight.ndim <= 4
fan = _calculate_correct_fan(weight, self.mode)
weight = weight * torch.tensor(self.gain, device=weight.device
) * torch.sqrt(torch.tensor(1.0 / fan, device=weight.device)
) * self.lr_mul
return weight
def __call__(self, module, inputs):
"""Standard interface for forward pre hooks."""
setattr(module, self.name, self.compute_weight(module))
@staticmethod
def apply(module, name, gain=2 ** 0.5, mode='fan_in', lr_mul=1.0):
"""Apply function.
This function is to register an equalized learning rate hook in an
``nn.Module``.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
for _, hook in module._forward_pre_hooks.items():
if isinstance(hook, EqualizedLR):
raise RuntimeError(
f'Cannot register two equalized_lr hooks on the same parameter {name} in {module} module.'
)
fn = EqualizedLR(name, gain=gain, mode=mode, lr_mul=lr_mul)
weight = module._parameters[name]
delattr(module, name)
module.register_parameter(name + '_orig', weight)
setattr(module, name, weight.data)
module.register_forward_pre_hook(fn)
return fn
class EqualizedLRLinearModule(nn.Linear):
"""Equalized LR LinearModule.
In this module, we adopt equalized lr in ``nn.Linear``. The equalized
learning rate is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
Note that, the initialization of ``self.weight`` will be overwritten as
:math:`\\mathcal{N}(0, 1)`.
Args:
equalized_lr_cfg (dict | None, optional): Config for ``EqualizedLR``.
If ``None``, equalized learning rate is ignored. Defaults to
dict(mode='fan_in').
"""
def __init__(self, *args, equalized_lr_cfg=dict(mode='fan_in'), **kwargs):
super().__init__(*args, **kwargs)
self.with_equalized_lr = equalized_lr_cfg is not None
if self.with_equalized_lr:
self.lr_mul = equalized_lr_cfg.get('lr_mul', 1.0)
else:
self.lr_mul = 1.0
if self.with_equalized_lr:
equalized_lr(self, **equalized_lr_cfg)
self._init_linear_weights()
def _init_linear_weights(self):
"""Initialize linear weights as described in PGGAN."""
nn.init.normal_(self.weight, 0, 1.0 / self.lr_mul)
if self.bias is not None:
nn.init.constant_(self.bias, 0.0)
class AdaptiveInstanceNorm(nn.Module):
"""Adaptive Instance Normalization Module.
Ref: https://github.com/rosinality/style-based-gan-pytorch/blob/master/model.py # noqa
Args:
in_channel (int): The number of input's channel.
style_dim (int): Style latent dimension.
"""
def __init__(self, in_channel, style_dim):
super().__init__()
self.norm = nn.InstanceNorm2d(in_channel)
self.affine = EqualizedLRLinearModule(style_dim, in_channel * 2)
self.affine.bias.data[:in_channel] = 1
self.affine.bias.data[in_channel:] = 0
def forward(self, input, style):
"""Forward function.
Args:
input (Tensor): Input tensor with shape (n, c, h, w).
style (Tensor): Input style tensor with shape (n, c).
Returns:
Tensor: Forward results.
"""
style = self.affine(style).unsqueeze(2).unsqueeze(3)
gamma, beta = style.chunk(2, 1)
out = self.norm(input)
out = gamma * out + beta
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'style_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
from torch.nn.init import _calculate_correct_fan
import torch.utils.cpp_extension
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_lift_fresh_mul_sqrt_0(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_ptr0 + x0, xmask)
tmp1 = 1.4142135381698608
tmp2 = tmp0 * tmp1
tmp3 = 0.5
tmp4 = tmp2 * tmp3
tmp5 = 1.0
tmp6 = tmp4 * tmp5
tl.store(out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_mul_1(in_out_ptr0,
in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, 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
x2 = xindex % 4
x3 = xindex // 4
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp22 = tl.load(in_ptr1 + (x2 + 8 * x3), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (4 + x2 + 8 * x3), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr2 + (4 + x2), xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tl.where(xmask, tmp1, 0)
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = 16.0
tmp18 = tmp16 / tmp17
tmp19 = 1e-05
tmp20 = tmp18 + tmp19
tmp21 = libdevice.rsqrt(tmp20)
tmp24 = tmp22 + tmp23
tmp25 = tmp0 - tmp10
tmp26 = tmp25 * tmp21
tmp27 = tmp24 * tmp26
tmp30 = tmp28 + tmp29
tmp31 = tmp27 + tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp21, xmask)
tl.store(out_ptr1 + (r1 + 16 * x0), tmp31, xmask)
tl.store(out_ptr0 + x0, tmp10, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (8, 4), (4, 1))
assert_size_stride(primals_2, (8,), (1,))
assert_size_stride(primals_3, (4, 4), (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((8, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_lift_fresh_mul_sqrt_0[grid(32)](primals_1, buf0,
32, XBLOCK=32, num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
extern_kernels.mm(primals_3, reinterpret_tensor(buf0, (4, 8), (1, 4
), 0), out=buf1)
buf2 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 1, 1), torch.float32)
buf3 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 16, 16), torch.float32
)
buf5 = reinterpret_tensor(buf3, (1, 16, 1, 1), (16, 1, 1, 1), 0)
del buf3
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_per_fused__native_batch_norm_legit_add_mul_1[grid(16)](buf5,
primals_4, buf1, primals_2, buf2, buf6, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
del buf1
del primals_2
return buf6, buf0, primals_3, primals_4, buf2, buf5
def equalized_lr(module, name='weight', gain=2 ** 0.5, mode='fan_in',
lr_mul=1.0):
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
EqualizedLR.apply(module, name, gain=gain, mode=mode, lr_mul=lr_mul)
return module
class EqualizedLR:
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
"""
def __init__(self, name='weight', gain=2 ** 0.5, mode='fan_in', lr_mul=1.0
):
self.name = name
self.mode = mode
self.gain = gain
self.lr_mul = lr_mul
def compute_weight(self, module):
"""Compute weight with equalized learning rate.
Args:
module (nn.Module): A module that is wrapped with equalized lr.
Returns:
torch.Tensor: Updated weight.
"""
weight = getattr(module, self.name + '_orig')
if weight.ndim == 5:
fan = _calculate_correct_fan(weight[0], self.mode)
else:
assert weight.ndim <= 4
fan = _calculate_correct_fan(weight, self.mode)
weight = weight * torch.tensor(self.gain, device=weight.device
) * torch.sqrt(torch.tensor(1.0 / fan, device=weight.device)
) * self.lr_mul
return weight
def __call__(self, module, inputs):
"""Standard interface for forward pre hooks."""
setattr(module, self.name, self.compute_weight(module))
@staticmethod
def apply(module, name, gain=2 ** 0.5, mode='fan_in', lr_mul=1.0):
"""Apply function.
This function is to register an equalized learning rate hook in an
``nn.Module``.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
for _, hook in module._forward_pre_hooks.items():
if isinstance(hook, EqualizedLR):
raise RuntimeError(
f'Cannot register two equalized_lr hooks on the same parameter {name} in {module} module.'
)
fn = EqualizedLR(name, gain=gain, mode=mode, lr_mul=lr_mul)
weight = module._parameters[name]
delattr(module, name)
module.register_parameter(name + '_orig', weight)
setattr(module, name, weight.data)
module.register_forward_pre_hook(fn)
return fn
class EqualizedLRLinearModule(nn.Linear):
"""Equalized LR LinearModule.
In this module, we adopt equalized lr in ``nn.Linear``. The equalized
learning rate is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
Note that, the initialization of ``self.weight`` will be overwritten as
:math:`\\mathcal{N}(0, 1)`.
Args:
equalized_lr_cfg (dict | None, optional): Config for ``EqualizedLR``.
If ``None``, equalized learning rate is ignored. Defaults to
dict(mode='fan_in').
"""
def __init__(self, *args, equalized_lr_cfg=dict(mode='fan_in'), **kwargs):
super().__init__(*args, **kwargs)
self.with_equalized_lr = equalized_lr_cfg is not None
if self.with_equalized_lr:
self.lr_mul = equalized_lr_cfg.get('lr_mul', 1.0)
else:
self.lr_mul = 1.0
if self.with_equalized_lr:
equalized_lr(self, **equalized_lr_cfg)
self._init_linear_weights()
def _init_linear_weights(self):
"""Initialize linear weights as described in PGGAN."""
nn.init.normal_(self.weight, 0, 1.0 / self.lr_mul)
if self.bias is not None:
nn.init.constant_(self.bias, 0.0)
class AdaptiveInstanceNormNew(nn.Module):
"""Adaptive Instance Normalization Module.
Ref: https://github.com/rosinality/style-based-gan-pytorch/blob/master/model.py # noqa
Args:
in_channel (int): The number of input's channel.
style_dim (int): Style latent dimension.
"""
def __init__(self, in_channel, style_dim):
super().__init__()
self.norm = nn.InstanceNorm2d(in_channel)
self.affine = EqualizedLRLinearModule(style_dim, in_channel * 2)
self.affine.bias.data[:in_channel] = 1
self.affine.bias.data[in_channel:] = 0
def forward(self, input_0, input_1):
primals_2 = self.affine.bias
primals_1 = self.affine.weight_orig
primals_4 = input_0
primals_3 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| bladesaber/mmgeneration | AdaptiveInstanceNorm | false | 1,573 | [
"Apache-2.0"
] | 0 | 158b49f7efd8028f231f6e9ca758ae0e20dd72ae | https://github.com/bladesaber/mmgeneration/tree/158b49f7efd8028f231f6e9ca758ae0e20dd72ae | import torch
import torch.nn as nn
from torch.nn.init import _calculate_correct_fan
import torch.utils.cpp_extension
def equalized_lr(module, name='weight', gain=2 ** 0.5, mode='fan_in',
lr_mul=1.0):
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
EqualizedLR.apply(module, name, gain=gain, mode=mode, lr_mul=lr_mul)
return module
class EqualizedLR:
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
"""
def __init__(self, name='weight', gain=2 ** 0.5, mode='fan_in', lr_mul=1.0
):
self.name = name
self.mode = mode
self.gain = gain
self.lr_mul = lr_mul
def compute_weight(self, module):
"""Compute weight with equalized learning rate.
Args:
module (nn.Module): A module that is wrapped with equalized lr.
Returns:
torch.Tensor: Updated weight.
"""
weight = getattr(module, self.name + '_orig')
if weight.ndim == 5:
fan = _calculate_correct_fan(weight[0], self.mode)
else:
assert weight.ndim <= 4
fan = _calculate_correct_fan(weight, self.mode)
weight = weight * torch.tensor(self.gain, device=weight.device
) * torch.sqrt(torch.tensor(1.0 / fan, device=weight.device)
) * self.lr_mul
return weight
def __call__(self, module, inputs):
"""Standard interface for forward pre hooks."""
setattr(module, self.name, self.compute_weight(module))
@staticmethod
def apply(module, name, gain=2 ** 0.5, mode='fan_in', lr_mul=1.0):
"""Apply function.
This function is to register an equalized learning rate hook in an
``nn.Module``.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
for _, hook in module._forward_pre_hooks.items():
if isinstance(hook, EqualizedLR):
raise RuntimeError(
f'Cannot register two equalized_lr hooks on the same parameter {name} in {module} module.'
# ... truncated (>4000 chars) for memory efficiency |
PANNsLoss | # 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_6/inductor_cache/jp/cjpssbf7mynqo577agfjklqolutavz545rkv3rtwl6os6mtztz73.py
# Topologically Sorted Source Nodes: [binary_cross_entropy, isnan, zeros_like, cwo, isinf, zeros_like_1, cwo_1, cwo_2], Original ATen: [aten.binary_cross_entropy, aten.isnan, aten.zeros_like, aten.where, aten.isinf, aten.clamp]
# Source node to ATen node mapping:
# binary_cross_entropy => full_default_2, full_default_3, log, log1p, maximum, maximum_1, mean, mul, mul_1, neg, sub, sub_1
# cwo => where
# cwo_1 => where_1
# cwo_2 => clamp_max, clamp_min
# isinf => isinf
# isnan => isnan
# zeros_like => full_default
# zeros_like_1 => full_default_1
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, 1), kwargs = {})
# %isnan : [num_users=1] = call_function[target=torch.ops.aten.isnan.default](args = (%arg0_1,), 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 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%isnan, %full_default, %arg0_1), kwargs = {})
# %isinf : [num_users=1] = call_function[target=torch.ops.aten.isinf.default](args = (%where,), kwargs = {})
# %full_default_1 : [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_1 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%isinf, %full_default_1, %where), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%where_1, 0), kwargs = {})
# %clamp_max : [num_users=2] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 1), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%clamp_max,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%neg,), kwargs = {})
# %full_default_2 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], -100), 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 = (%log1p, %full_default_2), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %maximum), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%clamp_max,), kwargs = {})
# %full_default_3 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], -100), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %maximum_1 : [num_users=1] = call_function[target=torch.ops.aten.maximum.default](args = (%log, %full_default_3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, %maximum_1), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %mul_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sub_1,), kwargs = {})
triton_per_fused_binary_cross_entropy_clamp_isinf_isnan_where_zeros_like_0 = async_compile.triton('triton_per_fused_binary_cross_entropy_clamp_isinf_isnan_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.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_binary_cross_entropy_clamp_isinf_isnan_where_zeros_like_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_binary_cross_entropy_clamp_isinf_isnan_where_zeros_like_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)
tmp3 = tl.load(in_ptr1 + (r0), None)
tmp1 = 1.0
tmp2 = tmp0 - tmp1
tmp4 = libdevice.isnan(tmp3).to(tl.int1)
tmp5 = 0.0
tmp6 = tl.where(tmp4, tmp5, tmp3)
tmp7 = libdevice.isinf(tmp6).to(tl.int1)
tmp8 = tl.where(tmp7, tmp5, tmp6)
tmp9 = triton_helpers.maximum(tmp8, tmp5)
tmp10 = triton_helpers.minimum(tmp9, tmp1)
tmp11 = -tmp10
tmp12 = libdevice.log1p(tmp11)
tmp13 = -100.0
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp2 * tmp14
tmp16 = tl_math.log(tmp10)
tmp17 = triton_helpers.maximum(tmp16, tmp13)
tmp18 = tmp0 * tmp17
tmp19 = tmp15 - tmp18
tmp20 = tl.broadcast_to(tmp19, [RBLOCK])
tmp22 = triton_helpers.promote_to_tensor(tl.sum(tmp20, 0))
tmp23 = 256.0
tmp24 = tmp22 / tmp23
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp24, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [binary_cross_entropy, isnan, zeros_like, cwo, isinf, zeros_like_1, cwo_1, cwo_2], Original ATen: [aten.binary_cross_entropy, aten.isnan, aten.zeros_like, aten.where, aten.isinf, aten.clamp]
stream0 = get_raw_stream(0)
triton_per_fused_binary_cross_entropy_clamp_isinf_isnan_where_zeros_like_0.run(buf1, arg1_1, arg0_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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.nn import functional as F
class PANNsLoss(nn.Module):
def __init__(self):
super().__init__()
@staticmethod
def stabilize_input(cwo):
cwo = torch.where(torch.isnan(cwo), torch.zeros_like(cwo), cwo)
cwo = torch.where(torch.isinf(cwo), torch.zeros_like(cwo), cwo)
cwo = cwo.clamp(0, 1)
return cwo
def forward(self, input, target):
cwo = self.stabilize_input(input)
return F.binary_cross_entropy(cwo, target)
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 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_per_fused_binary_cross_entropy_clamp_isinf_isnan_where_zeros_like_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)
tmp3 = tl.load(in_ptr1 + r0, None)
tmp1 = 1.0
tmp2 = tmp0 - tmp1
tmp4 = libdevice.isnan(tmp3).to(tl.int1)
tmp5 = 0.0
tmp6 = tl.where(tmp4, tmp5, tmp3)
tmp7 = libdevice.isinf(tmp6).to(tl.int1)
tmp8 = tl.where(tmp7, tmp5, tmp6)
tmp9 = triton_helpers.maximum(tmp8, tmp5)
tmp10 = triton_helpers.minimum(tmp9, tmp1)
tmp11 = -tmp10
tmp12 = libdevice.log1p(tmp11)
tmp13 = -100.0
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp2 * tmp14
tmp16 = tl_math.log(tmp10)
tmp17 = triton_helpers.maximum(tmp16, tmp13)
tmp18 = tmp0 * tmp17
tmp19 = tmp15 - tmp18
tmp20 = tl.broadcast_to(tmp19, [RBLOCK])
tmp22 = triton_helpers.promote_to_tensor(tl.sum(tmp20, 0))
tmp23 = 256.0
tmp24 = tmp22 / tmp23
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp24, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_binary_cross_entropy_clamp_isinf_isnan_where_zeros_like_0[
grid(1)](buf1, arg1_1, arg0_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class PANNsLossNew(nn.Module):
def __init__(self):
super().__init__()
@staticmethod
def stabilize_input(cwo):
cwo = torch.where(torch.isnan(cwo), torch.zeros_like(cwo), cwo)
cwo = torch.where(torch.isinf(cwo), torch.zeros_like(cwo), cwo)
cwo = cwo.clamp(0, 1)
return cwo
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| botkop/lark | PANNsLoss | false | 1,574 | [
"Apache-2.0"
] | 0 | edb2defdb514213fc121418578b0d9006a55f3a0 | https://github.com/botkop/lark/tree/edb2defdb514213fc121418578b0d9006a55f3a0 | import torch
import torch.nn as nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
@staticmethod
def stabilize_input(cwo):
cwo = torch.where(torch.isnan(cwo), torch.zeros_like(cwo), cwo)
cwo = torch.where(torch.isinf(cwo), torch.zeros_like(cwo), cwo)
cwo = cwo.clamp(0, 1)
return cwo
def forward(self, input, target):
cwo = self.stabilize_input(input)
return F.binary_cross_entropy(cwo, target)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SigmoidFocalLossStar | # 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_6/inductor_cache/w5/cw5om32ckvdnbudtseqw3tv3lev3pmwy7qdhh2z6ekjwcptr3szu.py
# Topologically Sorted Source Nodes: [log_sigmoid, mul, sub, mul_1, shifted_inputs, neg, loss, mul_3, sub_1, mul_4, alpha_t, loss_1, loss_2], Original ATen: [aten.log_sigmoid_forward, aten.mul, aten.sub, aten.neg, aten.div, aten.rsub, aten.add, aten.mean]
# Source node to ATen node mapping:
# alpha_t => add
# log_sigmoid => abs_1, exp, full_default, log1p, minimum, neg, sub_1
# loss => div
# loss_1 => mul_5
# loss_2 => mean
# mul => mul
# mul_1 => mul_1
# mul_3 => mul_3
# mul_4 => mul_4
# neg => neg_1
# shifted_inputs => mul_2
# sub => sub
# sub_1 => sub_2
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, 2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, 1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %sub), kwargs = {})
# %mul_2 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, 2.0), kwargs = {})
# %minimum : [num_users=1] = call_function[target=torch.ops.aten.minimum.default](args = (%full_default, %mul_2), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%mul_2,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %neg_1 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%sub_1,), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%neg_1, 2.0), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, 0.25), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, 0.75), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_3, %mul_4), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %add), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul_5,), kwargs = {})
triton_per_fused_add_div_log_sigmoid_forward_mean_mul_neg_rsub_sub_0 = async_compile.triton('triton_per_fused_add_div_log_sigmoid_forward_mean_mul_neg_rsub_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.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_log_sigmoid_forward_mean_mul_neg_rsub_sub_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_add_div_log_sigmoid_forward_mean_mul_neg_rsub_sub_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)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = 2.0
tmp3 = tmp1 * tmp2
tmp4 = 1.0
tmp5 = tmp3 - tmp4
tmp6 = tmp0 * tmp5
tmp7 = tmp6 * tmp2
tmp8 = 0.0
tmp9 = triton_helpers.minimum(tmp8, tmp7)
tmp10 = tl_math.abs(tmp7)
tmp11 = -tmp10
tmp12 = tl_math.exp(tmp11)
tmp13 = libdevice.log1p(tmp12)
tmp14 = tmp9 - tmp13
tmp15 = -tmp14
tmp16 = 0.5
tmp17 = tmp15 * tmp16
tmp18 = 0.25
tmp19 = tmp1 * tmp18
tmp20 = tmp4 - tmp1
tmp21 = 0.75
tmp22 = tmp20 * tmp21
tmp23 = tmp19 + tmp22
tmp24 = tmp17 * tmp23
tmp25 = tl.broadcast_to(tmp24, [RBLOCK])
tmp27 = triton_helpers.promote_to_tensor(tl.sum(tmp25, 0))
tmp28 = 256.0
tmp29 = tmp27 / tmp28
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp29, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [log_sigmoid, mul, sub, mul_1, shifted_inputs, neg, loss, mul_3, sub_1, mul_4, alpha_t, loss_1, loss_2], Original ATen: [aten.log_sigmoid_forward, aten.mul, aten.sub, aten.neg, aten.div, aten.rsub, aten.add, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_add_div_log_sigmoid_forward_mean_mul_neg_rsub_sub_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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.nn import functional as F
def sigmoid_focal_loss_star(inputs: 'torch.Tensor', targets: 'torch.Tensor',
alpha: 'float'=-1, gamma: 'float'=1, reduction: 'str'='none'
) ->torch.Tensor:
"""
FL* described in RetinaNet paper Appendix: https://arxiv.org/abs/1708.02002.
Args:
inputs: A float tensor of arbitrary shape.
The predictions for each example.
targets: A float tensor with the same shape as inputs. Stores the binary
classification label for each element in inputs
(0 for the negative class and 1 for the positive class).
alpha: (optional) Weighting factor in range (0,1) to balance
positive vs negative examples. Default = -1 (no weighting).
gamma: Gamma parameter described in FL*. Default = 1 (no weighting).
reduction: 'none' | 'mean' | 'sum'
'none': No reduction will be applied to the output.
'mean': The output will be averaged.
'sum': The output will be summed.
Returns:
Loss tensor with the reduction option applied.
"""
shifted_inputs = gamma * (inputs * (2 * targets - 1))
loss = -F.logsigmoid(shifted_inputs) / gamma
if alpha >= 0:
alpha_t = alpha * targets + (1 - alpha) * (1 - targets)
loss *= alpha_t
if reduction == 'mean':
loss = loss.mean()
elif reduction == 'sum':
loss = loss.sum()
return loss
class SigmoidFocalLossStar(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, preds, targets):
return sigmoid_focal_loss_star(preds, targets, self.alpha, self.
gamma, self.reduction)
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 libdevice, math as tl_math
import torch.nn as nn
from torch.nn import 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_per_fused_add_div_log_sigmoid_forward_mean_mul_neg_rsub_sub_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)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = 2.0
tmp3 = tmp1 * tmp2
tmp4 = 1.0
tmp5 = tmp3 - tmp4
tmp6 = tmp0 * tmp5
tmp7 = tmp6 * tmp2
tmp8 = 0.0
tmp9 = triton_helpers.minimum(tmp8, tmp7)
tmp10 = tl_math.abs(tmp7)
tmp11 = -tmp10
tmp12 = tl_math.exp(tmp11)
tmp13 = libdevice.log1p(tmp12)
tmp14 = tmp9 - tmp13
tmp15 = -tmp14
tmp16 = 0.5
tmp17 = tmp15 * tmp16
tmp18 = 0.25
tmp19 = tmp1 * tmp18
tmp20 = tmp4 - tmp1
tmp21 = 0.75
tmp22 = tmp20 * tmp21
tmp23 = tmp19 + tmp22
tmp24 = tmp17 * tmp23
tmp25 = tl.broadcast_to(tmp24, [RBLOCK])
tmp27 = triton_helpers.promote_to_tensor(tl.sum(tmp25, 0))
tmp28 = 256.0
tmp29 = tmp27 / tmp28
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp29, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_div_log_sigmoid_forward_mean_mul_neg_rsub_sub_0[
grid(1)](buf1, arg0_1, arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
def sigmoid_focal_loss_star(inputs: 'torch.Tensor', targets: 'torch.Tensor',
alpha: 'float'=-1, gamma: 'float'=1, reduction: 'str'='none'
) ->torch.Tensor:
"""
FL* described in RetinaNet paper Appendix: https://arxiv.org/abs/1708.02002.
Args:
inputs: A float tensor of arbitrary shape.
The predictions for each example.
targets: A float tensor with the same shape as inputs. Stores the binary
classification label for each element in inputs
(0 for the negative class and 1 for the positive class).
alpha: (optional) Weighting factor in range (0,1) to balance
positive vs negative examples. Default = -1 (no weighting).
gamma: Gamma parameter described in FL*. Default = 1 (no weighting).
reduction: 'none' | 'mean' | 'sum'
'none': No reduction will be applied to the output.
'mean': The output will be averaged.
'sum': The output will be summed.
Returns:
Loss tensor with the reduction option applied.
"""
shifted_inputs = gamma * (inputs * (2 * targets - 1))
loss = -F.logsigmoid(shifted_inputs) / gamma
if alpha >= 0:
alpha_t = alpha * targets + (1 - alpha) * (1 - targets)
loss *= alpha_t
if reduction == 'mean':
loss = loss.mean()
elif reduction == 'sum':
loss = loss.sum()
return loss
class SigmoidFocalLossStarNew(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| botkop/lark | SigmoidFocalLossStar | false | 1,575 | [
"Apache-2.0"
] | 0 | edb2defdb514213fc121418578b0d9006a55f3a0 | https://github.com/botkop/lark/tree/edb2defdb514213fc121418578b0d9006a55f3a0 | import torch
import torch.nn as nn
from torch.nn import functional as F
def sigmoid_focal_loss_star(inputs: 'torch.Tensor', targets: 'torch.Tensor',
alpha: 'float'=-1, gamma: 'float'=1, reduction: 'str'='none'
) ->torch.Tensor:
"""
FL* described in RetinaNet paper Appendix: https://arxiv.org/abs/1708.02002.
Args:
inputs: A float tensor of arbitrary shape.
The predictions for each example.
targets: A float tensor with the same shape as inputs. Stores the binary
classification label for each element in inputs
(0 for the negative class and 1 for the positive class).
alpha: (optional) Weighting factor in range (0,1) to balance
positive vs negative examples. Default = -1 (no weighting).
gamma: Gamma parameter described in FL*. Default = 1 (no weighting).
reduction: 'none' | 'mean' | 'sum'
'none': No reduction will be applied to the output.
'mean': The output will be averaged.
'sum': The output will be summed.
Returns:
Loss tensor with the reduction option applied.
"""
shifted_inputs = gamma * (inputs * (2 * targets - 1))
loss = -F.logsigmoid(shifted_inputs) / gamma
if alpha >= 0:
alpha_t = alpha * targets + (1 - alpha) * (1 - targets)
loss *= alpha_t
if reduction == 'mean':
loss = loss.mean()
elif reduction == 'sum':
loss = loss.sum()
return loss
class Model(nn.Module):
def __init__(self, alpha: 'float'=0.25, gamma: 'float'=2.0, reduction:
'str'='mean'):
super().__init__()
self.alpha = alpha
self.gamma = gamma
self.reduction = reduction
def forward(self, preds, targets):
return sigmoid_focal_loss_star(preds, targets, self.alpha, self.
gamma, self.reduction)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ModMBStddevLayer | # 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_6/inductor_cache/mi/cmi24qb25fs7ygfa5sk324jr6btb27kyh56k3h4duigwphd56khs.py
# Topologically Sorted Source Nodes: [y_1, add, y_2, mean, y_4], Original ATen: [aten.var, aten.add, aten.sqrt, aten.mean, aten.repeat]
# Source node to ATen node mapping:
# add => add
# mean => mean
# y_1 => var
# y_2 => sqrt
# y_4 => repeat
# Graph fragment:
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%view, [0]), kwargs = {correction: 0})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%var, 1e-08), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add,), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%sqrt, [2, 3, 4], True), kwargs = {})
# %repeat : [num_users=1] = call_function[target=torch.ops.aten.repeat.default](args = (%squeeze, [4, 1, 4, 4]), kwargs = {})
triton_per_fused_add_mean_repeat_sqrt_var_0 = async_compile.triton('triton_per_fused_add_mean_repeat_sqrt_var_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, 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': {2: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 3), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_mean_repeat_sqrt_var_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, '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_mean_repeat_sqrt_var_0(in_ptr0, out_ptr1, 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)
r0 = rindex
r1 = rindex % 16
r2 = (rindex // 16)
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = tl.load(in_ptr0 + (64 + r0), None)
tmp3 = tl.load(in_ptr0 + (128 + r0), None)
tmp5 = tl.load(in_ptr0 + (192 + r0), None)
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-08
tmp22 = tmp20 + tmp21
tmp23 = libdevice.sqrt(tmp22)
tmp24 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp26 = tl.sum(tmp24, 1)[:, None]
tmp27 = 64.0
tmp28 = tmp26 / tmp27
tl.store(out_ptr1 + (tl.broadcast_to(r1 + (80*r2), [XBLOCK, RBLOCK])), tmp28, None)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/yi/cyidf2yj3fms5jdxlfe7fdijzfj6p5a5q2qxo4llkuxnpqh6fj5o.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 = ([%arg0_1, %repeat], 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=[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_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_cat_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
x0 = xindex % 64
x1 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tl.store(out_ptr0 + (x0 + (80*x1)), tmp0, 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)
buf3 = empty_strided_cuda((4, 5, 4, 4), (80, 16, 4, 1), torch.float32)
buf2 = reinterpret_tensor(buf3, (4, 1, 4, 4), (80, 16, 4, 1), 64) # alias
# Topologically Sorted Source Nodes: [y_1, add, y_2, mean, y_4], Original ATen: [aten.var, aten.add, aten.sqrt, aten.mean, aten.repeat]
stream0 = get_raw_stream(0)
triton_per_fused_add_mean_repeat_sqrt_var_0.run(arg0_1, buf2, 1, 64, grid=grid(1), stream=stream0)
buf1 = reinterpret_tensor(buf3, (4, 4, 4, 4), (80, 16, 4, 1), 0) # alias
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
triton_poi_fused_cat_1.run(arg0_1, buf1, 256, grid=grid(256), stream=stream0)
del arg0_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)
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.distributed as dist
import torch.autograd as autograd
import torch.utils.cpp_extension
class AllGatherLayer(autograd.Function):
"""All gather layer with backward propagation path.
Indeed, this module is to make ``dist.all_gather()`` in the backward graph.
Such kind of operation has been widely used in Moco and other contrastive
learning algorithms.
"""
@staticmethod
def forward(ctx, x):
"""Forward function."""
ctx.save_for_backward(x)
output = [torch.zeros_like(x) for _ in range(dist.get_world_size())]
dist.all_gather(output, x)
return tuple(output)
@staticmethod
def backward(ctx, *grad_outputs):
"""Backward function."""
x, = ctx.saved_tensors
grad_out = torch.zeros_like(x)
grad_out = grad_outputs[dist.get_rank()]
return grad_out
class ModMBStddevLayer(nn.Module):
"""Modified MiniBatch Stddev Layer.
This layer is modified from ``MiniBatchStddevLayer`` used in PGGAN. In
StyleGAN2, the authors add a new feature, `channel_groups`, into this
layer.
Note that to accelerate the training procedure, we also add a new feature
of ``sync_std`` to achieve multi-nodes/machine training. This feature is
still in beta version and we have tested it on 256 scales.
Args:
group_size (int, optional): The size of groups in batch dimension.
Defaults to 4.
channel_groups (int, optional): The size of groups in channel
dimension. Defaults to 1.
sync_std (bool, optional): Whether to use synchronized std feature.
Defaults to False.
sync_groups (int | None, optional): The size of groups in node
dimension. Defaults to None.
eps (float, optional): Epsilon value to avoid computation error.
Defaults to 1e-8.
"""
def __init__(self, group_size=4, channel_groups=1, sync_std=False,
sync_groups=None, eps=1e-08):
super().__init__()
self.group_size = group_size
self.eps = eps
self.channel_groups = channel_groups
self.sync_std = sync_std
self.sync_groups = group_size if sync_groups is None else sync_groups
if self.sync_std:
assert torch.distributed.is_initialized(
), 'Only in distributed training can the sync_std be activated.'
mmcv.print_log('Adopt synced minibatch stddev layer', 'mmgen')
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input feature map with shape of (N, C, H, W).
Returns:
Tensor: Output feature map with shape of (N, C+1, H, W).
"""
if self.sync_std:
all_features = torch.cat(AllGatherLayer.apply(x), dim=0)
rank, ws = get_dist_info()
local_bs = all_features.shape[0] // ws
start_idx = local_bs * rank
if start_idx + self.sync_groups > all_features.shape[0]:
start_idx = all_features.shape[0] - self.sync_groups
end_idx = min(local_bs * rank + self.sync_groups, all_features.
shape[0])
x = all_features[start_idx:end_idx]
assert x.shape[0] <= self.group_size or x.shape[0
] % self.group_size == 0, f'Batch size be smaller than or equal to group size. Otherwise, batch size should be divisible by the group size.But got batch size {x.shape[0]}, group size {self.group_size}'
assert x.shape[1
] % self.channel_groups == 0, f'"channel_groups" must be divided by the feature channels. channel_groups: {self.channel_groups}, feature channels: {x.shape[1]}'
n, c, h, w = x.shape
group_size = min(n, self.group_size)
y = torch.reshape(x, (group_size, -1, self.channel_groups, c //
self.channel_groups, h, w))
y = torch.var(y, dim=0, unbiased=False)
y = torch.sqrt(y + self.eps)
y = y.mean(dim=(2, 3, 4), keepdim=True).squeeze(2)
y = y.repeat(group_size, 1, h, w)
return torch.cat([x, y], dim=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
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
import torch.distributed as dist
import torch.autograd as autograd
import torch.utils.cpp_extension
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_mean_repeat_sqrt_var_0(in_ptr0, out_ptr1, 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)
r0 = rindex
r1 = rindex % 16
r2 = rindex // 16
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = tl.load(in_ptr0 + (64 + r0), None)
tmp3 = tl.load(in_ptr0 + (128 + r0), None)
tmp5 = tl.load(in_ptr0 + (192 + r0), None)
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-08
tmp22 = tmp20 + tmp21
tmp23 = libdevice.sqrt(tmp22)
tmp24 = tl.broadcast_to(tmp23, [XBLOCK, RBLOCK])
tmp26 = tl.sum(tmp24, 1)[:, None]
tmp27 = 64.0
tmp28 = tmp26 / tmp27
tl.store(out_ptr1 + tl.broadcast_to(r1 + 80 * r2, [XBLOCK, RBLOCK]),
tmp28, None)
@triton.jit
def triton_poi_fused_cat_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
x0 = xindex % 64
x1 = xindex // 64
tmp0 = tl.load(in_ptr0 + x2, xmask)
tl.store(out_ptr0 + (x0 + 80 * x1), tmp0, 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)
buf3 = empty_strided_cuda((4, 5, 4, 4), (80, 16, 4, 1), torch.float32)
buf2 = reinterpret_tensor(buf3, (4, 1, 4, 4), (80, 16, 4, 1), 64)
get_raw_stream(0)
triton_per_fused_add_mean_repeat_sqrt_var_0[grid(1)](arg0_1, buf2,
1, 64, XBLOCK=1, num_warps=2, num_stages=1)
buf1 = reinterpret_tensor(buf3, (4, 4, 4, 4), (80, 16, 4, 1), 0)
triton_poi_fused_cat_1[grid(256)](arg0_1, buf1, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return buf3,
class AllGatherLayer(autograd.Function):
"""All gather layer with backward propagation path.
Indeed, this module is to make ``dist.all_gather()`` in the backward graph.
Such kind of operation has been widely used in Moco and other contrastive
learning algorithms.
"""
@staticmethod
def forward(ctx, x):
"""Forward function."""
ctx.save_for_backward(x)
output = [torch.zeros_like(x) for _ in range(dist.get_world_size())]
dist.all_gather(output, x)
return tuple(output)
@staticmethod
def backward(ctx, *grad_outputs):
"""Backward function."""
x, = ctx.saved_tensors
grad_out = torch.zeros_like(x)
grad_out = grad_outputs[dist.get_rank()]
return grad_out
class ModMBStddevLayerNew(nn.Module):
"""Modified MiniBatch Stddev Layer.
This layer is modified from ``MiniBatchStddevLayer`` used in PGGAN. In
StyleGAN2, the authors add a new feature, `channel_groups`, into this
layer.
Note that to accelerate the training procedure, we also add a new feature
of ``sync_std`` to achieve multi-nodes/machine training. This feature is
still in beta version and we have tested it on 256 scales.
Args:
group_size (int, optional): The size of groups in batch dimension.
Defaults to 4.
channel_groups (int, optional): The size of groups in channel
dimension. Defaults to 1.
sync_std (bool, optional): Whether to use synchronized std feature.
Defaults to False.
sync_groups (int | None, optional): The size of groups in node
dimension. Defaults to None.
eps (float, optional): Epsilon value to avoid computation error.
Defaults to 1e-8.
"""
def __init__(self, group_size=4, channel_groups=1, sync_std=False,
sync_groups=None, eps=1e-08):
super().__init__()
self.group_size = group_size
self.eps = eps
self.channel_groups = channel_groups
self.sync_std = sync_std
self.sync_groups = group_size if sync_groups is None else sync_groups
if self.sync_std:
assert torch.distributed.is_initialized(
), 'Only in distributed training can the sync_std be activated.'
mmcv.print_log('Adopt synced minibatch stddev layer', 'mmgen')
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bladesaber/mmgeneration | ModMBStddevLayer | false | 1,576 | [
"Apache-2.0"
] | 0 | 158b49f7efd8028f231f6e9ca758ae0e20dd72ae | https://github.com/bladesaber/mmgeneration/tree/158b49f7efd8028f231f6e9ca758ae0e20dd72ae | import torch
import torch.nn as nn
import torch.distributed as dist
import torch.autograd as autograd
import torch.utils.cpp_extension
class AllGatherLayer(autograd.Function):
"""All gather layer with backward propagation path.
Indeed, this module is to make ``dist.all_gather()`` in the backward graph.
Such kind of operation has been widely used in Moco and other contrastive
learning algorithms.
"""
@staticmethod
def forward(ctx, x):
"""Forward function."""
ctx.save_for_backward(x)
output = [torch.zeros_like(x) for _ in range(dist.get_world_size())]
dist.all_gather(output, x)
return tuple(output)
@staticmethod
def backward(ctx, *grad_outputs):
"""Backward function."""
x, = ctx.saved_tensors
grad_out = torch.zeros_like(x)
grad_out = grad_outputs[dist.get_rank()]
return grad_out
class Model(nn.Module):
"""Modified MiniBatch Stddev Layer.
This layer is modified from ``MiniBatchStddevLayer`` used in PGGAN. In
StyleGAN2, the authors add a new feature, `channel_groups`, into this
layer.
Note that to accelerate the training procedure, we also add a new feature
of ``sync_std`` to achieve multi-nodes/machine training. This feature is
still in beta version and we have tested it on 256 scales.
Args:
group_size (int, optional): The size of groups in batch dimension.
Defaults to 4.
channel_groups (int, optional): The size of groups in channel
dimension. Defaults to 1.
sync_std (bool, optional): Whether to use synchronized std feature.
Defaults to False.
sync_groups (int | None, optional): The size of groups in node
dimension. Defaults to None.
eps (float, optional): Epsilon value to avoid computation error.
Defaults to 1e-8.
"""
def __init__(self, group_size=4, channel_groups=1, sync_std=False,
sync_groups=None, eps=1e-08):
super().__init__()
self.group_size = group_size
self.eps = eps
self.channel_groups = channel_groups
self.sync_std = sync_std
self.sync_groups = group_size if sync_groups is None else sync_groups
if self.sync_std:
assert torch.distributed.is_initialized(
), 'Only in distributed training can the sync_std be activated.'
mmcv.print_log('Adopt synced minibatch stddev layer', 'mmgen')
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input feature map with shape of (N, C, H, W).
Returns:
Tensor: Output feature map with shape of (N, C+1, H, W).
"""
if self.sync_std:
all_features = torch.cat(AllGatherLayer.apply(x), dim=0)
rank, ws = get_dist_info()
local_bs = all_features.shape[0] // ws
start_idx = local_bs * rank
if start_idx + self.sync_groups > all_features.shape[0]:
start_idx = all_features.shape[0] - self.sync_groups
end_idx = min(local_bs * rank + self.sync_groups, all_features.
shape[0])
x = all_features[start_idx:end_idx]
assert x.shape[0] <= self.group_size or x.shape[0
] % self.group_size == 0, f'Batch size be smaller than or equal to group size. Otherwise, batch size should be divisible by the group size.But got batch size {x.shape[0]}, group size {self.group_size}'
assert x.shape[1
] % self.channel_groups == 0, f'"channel_groups" must be divided by the feature channels. channel_groups: {self.channel_groups}, feature channels: {x.shape[1]}'
n, c, h, w = x.shape
group_size = min(n, self.group_size)
y = torch.reshape(x, (group_size, -1, self.channel_groups, c //
self.channel_groups, h, w))
y = torch.var(y, dim=0, unbiased=False)
y = torch.sqrt(y + self.eps)
y =
# ... truncated (>4000 chars) for memory efficiency |
CriticArchitecture | # 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_6/inductor_cache/lg/clgg7lidrsfgaud7ua7lulq4csdjjfhoa5iu6ldcldqvnqeoprch.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 = ([%relu, %primals_4], 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: '*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_cat_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_cat_0(in_ptr0, in_ptr1, in_ptr2, 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 = tl.load(in_ptr1 + (x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.full([1], 0, tl.int32)
tmp9 = triton_helpers.maximum(tmp8, tmp7)
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp4, tmp9, tmp10)
tmp12 = tmp0 >= tmp3
tmp13 = tl.full([1], 8, tl.int64)
tmp14 = tmp0 < tmp13
tmp15 = tl.load(in_ptr2 + ((4*x1) + ((-4) + x0)), tmp12 & xmask, eviction_policy='evict_last', other=0.0)
tmp16 = tl.where(tmp4, tmp11, tmp15)
tl.store(out_ptr0 + (x2), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/5b/c5br3r4gpi7zzaygqfdgcqeerwiekt2d2t2wkw4sj54lam6radgq.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_2 => relu_1
# Graph fragment:
# %add_tensor : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default, %primals_6), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor,), kwargs = {})
triton_poi_fused_relu_1 = async_compile.triton('triton_poi_fused_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=[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_relu_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_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
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)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/mj/cmj6mwbx6q5kpdkkge737ye2m2mv7kbiyvla5eopzuo3u6c56zon.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_2), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_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_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=[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_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_relu_threshold_backward_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
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 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
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, 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, 8), (8, 1))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (1, 4), (4, 1))
assert_size_stride(primals_8, (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: [], Original ATen: []
extern_kernels.mm(primals_3, reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(buf0, primals_2, primals_4, buf1, 32, grid=grid(32), stream=stream0)
del primals_4
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf1, reinterpret_tensor(primals_5, (8, 4), (1, 8), 0), out=buf2)
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu]
triton_poi_fused_relu_1.run(buf3, primals_6, 16, grid=grid(16), stream=stream0)
del primals_6
buf5 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, buf3, reinterpret_tensor(primals_7, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf5)
del primals_8
buf6 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_2.run(buf0, primals_2, buf6, 16, grid=grid(16), stream=stream0)
del buf0
del primals_2
return (buf5, primals_3, buf1, buf3, primals_7, primals_5, 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((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, 8), (8, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = 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])
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 abc import ABC
import torch.nn.functional as F
from torch import nn
from torch.nn import init
def fan_in_init(tensor):
fan_in = tensor.size(1)
v = 1.0 / np.sqrt(fan_in)
init.uniform_(tensor, -v, v)
class Architecture(nn.Module, ABC):
def __init__(self):
super().__init__()
class CriticArchitecture(Architecture):
def __init__(self, input_size, hidden_layers, output_size,
output_activation):
"""
Initialize a Critic for low dimensional environment.
num_feature: number of features of input.
"""
super().__init__()
self._input_size = input_size
self._hidden_layers = hidden_layers
self._output_size = output_size
self.fc1 = nn.Linear(self._input_size[0], self._hidden_layers[0])
fan_in_init(self.fc1.weight)
self.fc2 = nn.Linear(self._hidden_layers[0] + self._output_size[0],
self._hidden_layers[1])
fan_in_init(self.fc2.weight)
self.head = nn.Linear(self._hidden_layers[1], 1)
init.uniform_(self.head.weight, -0.003, 0.003)
init.uniform_(self.head.bias, -0.003, 0.003)
def forward(self, states, actions):
x = F.relu(self.fc1(states))
x = torch.cat((x, actions), 1)
x = F.relu(self.fc2(x))
x = self.head(x)
return x
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'input_size': [4, 4], 'hidden_layers': [4, 4],
'output_size': [4, 4], 'output_activation': 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
from abc import ABC
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_cat_0(in_ptr0, in_ptr1, in_ptr2, 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 = tl.load(in_ptr1 + x0, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.full([1], 0, tl.int32)
tmp9 = triton_helpers.maximum(tmp8, tmp7)
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp4, tmp9, tmp10)
tmp12 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp15 = tl.load(in_ptr2 + (4 * x1 + (-4 + x0)), tmp12 & xmask,
eviction_policy='evict_last', other=0.0)
tmp16 = tl.where(tmp4, tmp11, tmp15)
tl.store(out_ptr0 + x2, tmp16, xmask)
@triton.jit
def triton_poi_fused_relu_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
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)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_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
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 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(out_ptr0 + x2, 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, 8), (8, 1))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (1, 4), (4, 1))
assert_size_stride(primals_8, (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_3, reinterpret_tensor(primals_1, (4, 4),
(1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(32)](buf0, primals_2, primals_4, buf1,
32, XBLOCK=32, num_warps=1, num_stages=1)
del primals_4
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf1, reinterpret_tensor(primals_5, (8, 4), (1, 8
), 0), out=buf2)
buf3 = buf2
del buf2
triton_poi_fused_relu_1[grid(16)](buf3, primals_6, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_6
buf5 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_8, buf3, reinterpret_tensor(primals_7,
(4, 1), (1, 4), 0), alpha=1, beta=1, out=buf5)
del primals_8
buf6 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_2[grid(16)](buf0,
primals_2, buf6, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf0
del primals_2
return buf5, primals_3, buf1, buf3, primals_7, primals_5, buf6
def fan_in_init(tensor):
fan_in = tensor.size(1)
v = 1.0 / np.sqrt(fan_in)
init.uniform_(tensor, -v, v)
class Architecture(nn.Module, ABC):
def __init__(self):
super().__init__()
class CriticArchitectureNew(Architecture):
def __init__(self, input_size, hidden_layers, output_size,
output_activation):
"""
Initialize a Critic for low dimensional environment.
num_feature: number of features of input.
"""
super().__init__()
self._input_size = input_size
self._hidden_layers = hidden_layers
self._output_size = output_size
self.fc1 = nn.Linear(self._input_size[0], self._hidden_layers[0])
fan_in_init(self.fc1.weight)
self.fc2 = nn.Linear(self._hidden_layers[0] + self._output_size[0],
self._hidden_layers[1])
fan_in_init(self.fc2.weight)
self.head = nn.Linear(self._hidden_layers[1], 1)
init.uniform_(self.head.weight, -0.003, 0.003)
init.uniform_(self.head.bias, -0.003, 0.003)
def forward(self, input_0, input_1):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_5 = self.fc2.weight
primals_6 = self.fc2.bias
primals_7 = self.head.weight
primals_8 = self.head.bias
primals_3 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0]
| bootml/agent | CriticArchitecture | false | 1,577 | [
"Apache-2.0"
] | 0 | 84235db931d6e4ef956962961c619994898ebdd5 | https://github.com/bootml/agent/tree/84235db931d6e4ef956962961c619994898ebdd5 | import torch
import numpy as np
from abc import ABC
import torch.nn.functional as F
from torch import nn
from torch.nn import init
def fan_in_init(tensor):
fan_in = tensor.size(1)
v = 1.0 / np.sqrt(fan_in)
init.uniform_(tensor, -v, v)
class Architecture(nn.Module, ABC):
def __init__(self):
super().__init__()
class Model(Architecture):
def __init__(self, input_size, hidden_layers, output_size,
output_activation):
"""
Initialize a Critic for low dimensional environment.
num_feature: number of features of input.
"""
super().__init__()
self._input_size = input_size
self._hidden_layers = hidden_layers
self._output_size = output_size
self.fc1 = nn.Linear(self._input_size[0], self._hidden_layers[0])
fan_in_init(self.fc1.weight)
self.fc2 = nn.Linear(self._hidden_layers[0] + self._output_size[0],
self._hidden_layers[1])
fan_in_init(self.fc2.weight)
self.head = nn.Linear(self._hidden_layers[1], 1)
init.uniform_(self.head.weight, -0.003, 0.003)
init.uniform_(self.head.bias, -0.003, 0.003)
def forward(self, states, actions):
x = F.relu(self.fc1(states))
x = torch.cat((x, actions), 1)
x = F.relu(self.fc2(x))
x = self.head(x)
return x
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'input_size': [4, 4], 'hidden_layers': [4, 4],
'output_size': [4, 4], 'output_activation': 4}]
|
SelfGating | # 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_6/inductor_cache/32/c32gpnu7y6kwawwiknabqcyafcipv27fjg22cpx6wzdxmd52bm4o.py
# Topologically Sorted Source Nodes: [spatiotemporal_average], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# spatiotemporal_average => mean
# Graph fragment:
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [2, 3, 4]), 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, 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, 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 = 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)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 64.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/c2/cc26hd4oq32vrxgrqqakqqnzwxv6d6ovzdl26r33utpsmw442x3p.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 = (%unsqueeze_2, %primals_1), 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=[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_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 = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 64)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x1), 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 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4, 4), (256, 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, 1), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [spatiotemporal_average], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 16, 64, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [weights], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, buf1, reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_2
del primals_3
buf3 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
triton_poi_fused_mul_1.run(buf2, primals_1, buf3, 1024, grid=grid(1024), stream=stream0)
return (buf3, primals_1, 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, 4, 4), (256, 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.utils.data
import torch as th
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class SelfGating(nn.Module):
def __init__(self, input_dim):
super(SelfGating, self).__init__()
self.fc = nn.Linear(input_dim, input_dim)
def forward(self, input_tensor):
"""Feature gating as used in S3D-G.
"""
spatiotemporal_average = th.mean(input_tensor, dim=[2, 3, 4])
weights = self.fc(spatiotemporal_average)
weights = th.sigmoid(weights)
return weights[:, :, None, None, None] * input_tensor
def get_inputs():
return [torch.rand([4, 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
import torch.utils.data
import torch.nn as nn
import torch.nn.parallel
import torch.optim
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_mean_0(in_out_ptr0, in_ptr0, 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)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 64 * 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 = 64.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_mul_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 64
x2 = xindex
tmp0 = tl.load(in_ptr0 + x1, 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 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4, 4), (256, 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, 1), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(16)](buf1, primals_1, 16, 64, XBLOCK=8,
num_warps=4, num_stages=1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_3, buf1, reinterpret_tensor(primals_2,
(4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_2
del primals_3
buf3 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1),
torch.float32)
triton_poi_fused_mul_1[grid(1024)](buf2, primals_1, buf3, 1024,
XBLOCK=128, num_warps=4, num_stages=1)
return buf3, primals_1, buf1, buf2
class SelfGatingNew(nn.Module):
def __init__(self, input_dim):
super(SelfGatingNew, self).__init__()
self.fc = nn.Linear(input_dim, input_dim)
def forward(self, input_0):
primals_2 = self.fc.weight
primals_3 = self.fc.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| bryant1410/MIL-NCE_HowTo100M | SelfGating | false | 1,578 | [
"Apache-2.0"
] | 0 | 9ba876bd67160e24a5ce379a07d18a8036be0d36 | https://github.com/bryant1410/MIL-NCE_HowTo100M/tree/9ba876bd67160e24a5ce379a07d18a8036be0d36 | import torch
import torch.utils.data
import torch as th
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class Model(nn.Module):
def __init__(self, input_dim):
super().__init__()
self.fc = nn.Linear(input_dim, input_dim)
def forward(self, input_tensor):
"""Feature gating as used in S3D-G.
"""
spatiotemporal_average = th.mean(input_tensor, dim=[2, 3, 4])
weights = self.fc(spatiotemporal_average)
weights = th.sigmoid(weights)
return weights[:, :, None, None, None] * input_tensor
def get_inputs():
return [torch.rand([4, 4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
LearnedPositionalEmbedding | # 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_6/inductor_cache/5i/c5iybmnijeaxq3pumkl5crtkns462pwdrh72bxy4lcvnlh3r4364.py
# Topologically Sorted Source Nodes: [ne, mask, cumsum], Original ATen: [aten.ne, aten._to_copy, aten.cumsum]
# Source node to ATen node mapping:
# cumsum => cumsum
# mask => convert_element_type
# ne => ne
# Graph fragment:
# %ne : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%primals_1, 4), kwargs = {})
# %convert_element_type : [num_users=2] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%ne, torch.int32), kwargs = {})
# %cumsum : [num_users=1] = call_function[target=torch.ops.aten.cumsum.default](args = (%convert_element_type, 1), kwargs = {})
triton_per_fused__to_copy_cumsum_ne_0 = async_compile.triton('triton_per_fused__to_copy_cumsum_ne_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.jit
def _triton_helper_fn_add0(arg0_0, arg1_0):
tmp0 = arg0_0 + arg1_0
return tmp0
@triton_heuristics.persistent_reduction(
size_hints=[64, 4],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i64', 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__to_copy_cumsum_ne_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__to_copy_cumsum_ne_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 = 4.0
tmp2 = tmp0 != tmp1
tmp3 = tmp2.to(tl.int32)
tmp4 = tmp3.to(tl.int64)
tmp5 = tmp4.to(tl.int64)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp7, = tl.associative_scan((tmp6,), 1, _triton_helper_fn_add0)
tl.store(out_ptr0 + (x0 + (16*r2) + (64*x1)), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ft/cftxaavy7b7scxgnrhfsvfnicvimxnf3kpckow5nzkyed3meyoli.py
# Topologically Sorted Source Nodes: [ne, mask, type_as, mul, long, positions], Original ATen: [aten.ne, aten._to_copy, aten.mul, aten.add]
# Source node to ATen node mapping:
# long => convert_element_type_2
# mask => convert_element_type
# mul => mul
# ne => ne
# positions => add
# type_as => convert_element_type_1
# Graph fragment:
# %ne : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%primals_1, 4), kwargs = {})
# %convert_element_type : [num_users=2] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%ne, torch.int32), kwargs = {})
# %convert_element_type_1 : [num_users=1] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%cumsum, torch.int32), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convert_element_type_1, %convert_element_type), kwargs = {})
# %convert_element_type_2 : [num_users=1] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%mul, torch.int64), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%convert_element_type_2, 4), kwargs = {})
triton_poi_fused__to_copy_add_mul_ne_1 = async_compile.triton('triton_poi_fused__to_copy_add_mul_ne_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: '*i64', 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__to_copy_add_mul_ne_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__to_copy_add_mul_ne_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
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp2 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tmp0.to(tl.int32)
tmp3 = 4.0
tmp4 = tmp2 != tmp3
tmp5 = tmp4.to(tl.int32)
tmp6 = tmp1 * tmp5
tmp7 = tmp6.to(tl.int64)
tmp8 = tl.full([1], 4, tl.int64)
tmp9 = tmp7 + tmp8
tl.store(in_out_ptr0 + (x0), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/in/cinglqnf6mtochspmiolvr3bqay6yiivzgqlihpkdlbd5p4ccw54.py
# Topologically Sorted Source Nodes: [embedding], Original ATen: [aten.embedding]
# Source node to ATen node mapping:
# embedding => embedding
# Graph fragment:
# %embedding : [num_users=1] = call_function[target=torch.ops.aten.embedding.default](args = (%primals_2, %add, 4), kwargs = {})
triton_poi_fused_embedding_2 = async_compile.triton('triton_poi_fused_embedding_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: '*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_embedding_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_embedding_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
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.full([XBLOCK], 9, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert(((0 <= tmp4) & (tmp4 < 9)) | ~(xmask), "index out of bounds: 0 <= tmp4 < 9")
tmp6 = tl.load(in_ptr1 + (x0 + (4*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 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (9, 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.int64)
# Topologically Sorted Source Nodes: [ne, mask, cumsum], Original ATen: [aten.ne, aten._to_copy, aten.cumsum]
stream0 = get_raw_stream(0)
triton_per_fused__to_copy_cumsum_ne_0.run(primals_1, buf0, 64, 4, grid=grid(64), stream=stream0)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [ne, mask, type_as, mul, long, positions], Original ATen: [aten.ne, aten._to_copy, aten.mul, aten.add]
triton_poi_fused__to_copy_add_mul_ne_1.run(buf1, primals_1, 256, grid=grid(256), stream=stream0)
del primals_1
buf2 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [embedding], Original ATen: [aten.embedding]
triton_poi_fused_embedding_2.run(buf1, primals_2, buf2, 1024, grid=grid(1024), stream=stream0)
del primals_2
return (buf2, 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((9, 4), (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.functional as F
class LearnedPositionalEmbedding(nn.Embedding):
"""
This module learns positional embeddings up to a fixed maximum size.
Padding ids are ignored by either offsetting based on padding_idx
or by setting padding_idx to None and ensuring that the appropriate
position ids are passed to the forward function.
"""
def __init__(self, num_embeddings: 'int', embedding_dim: 'int',
padding_idx: 'int'):
if padding_idx is not None:
num_embeddings_ = num_embeddings + padding_idx + 1
else:
num_embeddings_ = num_embeddings
super().__init__(num_embeddings_, embedding_dim, padding_idx)
self.max_positions = num_embeddings
def forward(self, input: 'torch.Tensor'):
"""Input is expected to be of size [bsz x seqlen]."""
if input.size(1) > self.max_positions:
raise ValueError(
f'Sequence length {input.size(1)} above maximum sequence length of {self.max_positions}'
)
mask = input.ne(self.padding_idx).int()
positions = (torch.cumsum(mask, dim=1).type_as(mask) * mask).long(
) + self.padding_idx
return F.embedding(positions, self.weight, self.padding_idx, self.
max_norm, self.norm_type, self.scale_grad_by_freq, self.sparse)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_embeddings': 4, 'embedding_dim': 4, 'padding_idx': 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_helper_fn_add0(arg0_0, arg1_0):
tmp0 = arg0_0 + arg1_0
return tmp0
@triton.jit
def triton_per_fused__to_copy_cumsum_ne_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 = 4.0
tmp2 = tmp0 != tmp1
tmp3 = tmp2.to(tl.int32)
tmp4 = tmp3.to(tl.int64)
tmp5 = tmp4.to(tl.int64)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp7, = tl.associative_scan((tmp6,), 1, _triton_helper_fn_add0)
tl.store(out_ptr0 + (x0 + 16 * r2 + 64 * x1), tmp7, xmask)
@triton.jit
def triton_poi_fused__to_copy_add_mul_ne_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
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp2 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tmp0.to(tl.int32)
tmp3 = 4.0
tmp4 = tmp2 != tmp3
tmp5 = tmp4.to(tl.int32)
tmp6 = tmp1 * tmp5
tmp7 = tmp6.to(tl.int64)
tmp8 = tl.full([1], 4, tl.int64)
tmp9 = tmp7 + tmp8
tl.store(in_out_ptr0 + x0, tmp9, xmask)
@triton.jit
def triton_poi_fused_embedding_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 1024
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.full([XBLOCK], 9, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert((0 <= tmp4) & (tmp4 < 9) | ~xmask,
'index out of bounds: 0 <= tmp4 < 9')
tmp6 = tl.load(in_ptr1 + (x0 + 4 * tmp4), xmask)
tl.store(out_ptr0 + x2, tmp6, 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, (9, 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.int64)
get_raw_stream(0)
triton_per_fused__to_copy_cumsum_ne_0[grid(64)](primals_1, buf0, 64,
4, XBLOCK=8, num_warps=2, num_stages=1)
buf1 = buf0
del buf0
triton_poi_fused__to_copy_add_mul_ne_1[grid(256)](buf1, primals_1,
256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_1
buf2 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1),
torch.float32)
triton_poi_fused_embedding_2[grid(1024)](buf1, primals_2, buf2,
1024, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
return buf2, buf1
class LearnedPositionalEmbeddingNew(nn.Embedding):
"""
This module learns positional embeddings up to a fixed maximum size.
Padding ids are ignored by either offsetting based on padding_idx
or by setting padding_idx to None and ensuring that the appropriate
position ids are passed to the forward function.
"""
def __init__(self, num_embeddings: 'int', embedding_dim: 'int',
padding_idx: 'int'):
if padding_idx is not None:
num_embeddings_ = num_embeddings + padding_idx + 1
else:
num_embeddings_ = num_embeddings
super().__init__(num_embeddings_, embedding_dim, padding_idx)
self.max_positions = num_embeddings
def forward(self, input_0):
primals_2 = self.weight
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
| boxiangliu/esm | LearnedPositionalEmbedding | false | 1,579 | [
"MIT"
] | 0 | 3c143d99103e0ea38a9455f30a73cd9c87376606 | https://github.com/boxiangliu/esm/tree/3c143d99103e0ea38a9455f30a73cd9c87376606 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Embedding):
"""
This module learns positional embeddings up to a fixed maximum size.
Padding ids are ignored by either offsetting based on padding_idx
or by setting padding_idx to None and ensuring that the appropriate
position ids are passed to the forward function.
"""
def __init__(self, num_embeddings: 'int', embedding_dim: 'int',
padding_idx: 'int'):
if padding_idx is not None:
num_embeddings_ = num_embeddings + padding_idx + 1
else:
num_embeddings_ = num_embeddings
super().__init__(num_embeddings_, embedding_dim, padding_idx)
self.max_positions = num_embeddings
def forward(self, input: 'torch.Tensor'):
"""Input is expected to be of size [bsz x seqlen]."""
if input.size(1) > self.max_positions:
raise ValueError(
f'Sequence length {input.size(1)} above maximum sequence length of {self.max_positions}'
)
mask = input.ne(self.padding_idx).int()
positions = (torch.cumsum(mask, dim=1).type_as(mask) * mask).long(
) + self.padding_idx
return F.embedding(positions, self.weight, self.padding_idx, self.
max_norm, self.norm_type, self.scale_grad_by_freq, self.sparse)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
simple_mlp | # 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_6/inductor_cache/au/cau6qypw2vz4drppp6yr6chutchyhnniousxhhlq2y5r3yu3gep5.py
# Topologically Sorted Source Nodes: [hidden], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# hidden => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_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': ['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_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 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
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, (1, 4), (4, 1))
assert_size_stride(primals_5, (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: [hidden], Original ATen: [aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_0.run(buf1, primals_2, 256, grid=grid(256), stream=stream0)
del primals_2
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], 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
return (reinterpret_tensor(buf3, (4, 4, 4, 1), (16, 4, 1, 1), 0), buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, 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((1, 4), (4, 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
from torch.nn import functional as F
class simple_mlp(nn.Module):
def __init__(self, feature_dim, layer, hidden):
super(simple_mlp, self).__init__()
self.layer = layer
self.linear1 = nn.Linear(feature_dim, hidden)
if layer == 2:
self.linear2 = nn.Linear(hidden, hidden)
self.linear3 = nn.Linear(hidden, 1)
def forward(self, x, weights=None):
hidden = F.relu(self.linear1(x))
if self.layer == 2:
hidden = F.relu(self.linear2(hidden))
out = self.linear3(hidden)
return out, hidden
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'feature_dim': 4, 'layer': 1, '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 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_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.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
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, (1, 4), (4, 1))
assert_size_stride(primals_5, (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_relu_0[grid(256)](buf1, primals_2, 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
return reinterpret_tensor(buf3, (4, 4, 4, 1), (16, 4, 1, 1), 0
), buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, primals_4
class simple_mlpNew(nn.Module):
def __init__(self, feature_dim, layer, hidden):
super(simple_mlpNew, self).__init__()
self.layer = layer
self.linear1 = nn.Linear(feature_dim, hidden)
if layer == 2:
self.linear2 = nn.Linear(hidden, hidden)
self.linear3 = nn.Linear(hidden, 1)
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear3.weight
primals_5 = self.linear3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0], output[1]
| broadinstitute/TCRP | simple_mlp | false | 1,580 | [
"MIT"
] | 0 | 9e580dbf0c9d0ec5e5b1a949087df5a3724fa35b | https://github.com/broadinstitute/TCRP/tree/9e580dbf0c9d0ec5e5b1a949087df5a3724fa35b | import torch
import torch.nn as nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self, feature_dim, layer, hidden):
super().__init__()
self.layer = layer
self.linear1 = nn.Linear(feature_dim, hidden)
if layer == 2:
self.linear2 = nn.Linear(hidden, hidden)
self.linear3 = nn.Linear(hidden, 1)
def forward(self, x, weights=None):
hidden = F.relu(self.linear1(x))
if self.layer == 2:
hidden = F.relu(self.linear2(hidden))
out = self.linear3(hidden)
return out, hidden
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 1, 4]
|
MaskedInstanceNorm1d | # 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_6/inductor_cache/hv/chvlig3vkflhzd7ivdbf6ojzalx44fop7l6z3mhozziz64jmw5ck.py
# Topologically Sorted Source Nodes: [mul, sum_2, cnt, cnt_for_mu, mu, sub, sigma, mul_1, sum_3, sub_1, cnt_fot_sigma, sigma_1, add, sigma_2], Original ATen: [aten.mul, aten.sum, aten.clamp, aten.div, aten.sub, aten.pow, aten.add, aten.sqrt]
# Source node to ATen node mapping:
# add => add
# cnt => sum_1
# cnt_for_mu => clamp_max, clamp_min
# cnt_fot_sigma => clamp_max_1, clamp_min_1
# mu => div
# mul => mul
# mul_1 => mul_1
# sigma => pow_1
# sigma_1 => div_1
# sigma_2 => sqrt
# sub => sub
# sub_1 => sub_1
# sum_2 => sum_2
# sum_3 => sum_3
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, %unsqueeze), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [-1], True), kwargs = {})
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%unsqueeze, [-1], True), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%sum_1, 1.0), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 100000.0), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_2, %clamp_max), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %div), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_1, %unsqueeze), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [-1], True), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sum_1, 1), kwargs = {})
# %clamp_min_1 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%sub_1, 1.0), kwargs = {})
# %clamp_max_1 : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min_1, 100000.0), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_3, %clamp_max_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_1, 1e-08), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_add_clamp_div_mul_pow_sqrt_sub_sum_0 = async_compile.triton('triton_poi_fused_add_clamp_div_mul_pow_sqrt_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.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_clamp_div_mul_pow_sqrt_sub_sum_0', 'mutated_arg_names': ['in_out_ptr0'], '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_clamp_div_mul_pow_sqrt_sub_sum_0(in_out_ptr0, 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 % 64
x0 = xindex % 16
x2 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (4*x3), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + ((4*x0) + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x3)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0) + (64*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x3)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0) + (64*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x3)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0) + (64*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 + tmp4
tmp16 = tmp15 + tmp8
tmp17 = tmp16 + tmp12
tmp18 = 1.0
tmp19 = triton_helpers.maximum(tmp17, tmp18)
tmp20 = 100000.0
tmp21 = triton_helpers.minimum(tmp19, tmp20)
tmp22 = tmp14 / tmp21
tmp23 = tmp0 - tmp22
tmp24 = tmp23 * tmp23
tmp25 = tmp24 * tmp1
tmp26 = tmp3 - tmp22
tmp27 = tmp26 * tmp26
tmp28 = tmp27 * tmp4
tmp29 = tmp25 + tmp28
tmp30 = tmp7 - tmp22
tmp31 = tmp30 * tmp30
tmp32 = tmp31 * tmp8
tmp33 = tmp29 + tmp32
tmp34 = tmp11 - tmp22
tmp35 = tmp34 * tmp34
tmp36 = tmp35 * tmp12
tmp37 = tmp33 + tmp36
tmp38 = tmp17 - tmp18
tmp39 = triton_helpers.maximum(tmp38, tmp18)
tmp40 = triton_helpers.minimum(tmp39, tmp20)
tmp41 = tmp37 / tmp40
tmp42 = 1e-08
tmp43 = tmp41 + tmp42
tmp44 = libdevice.sqrt(tmp43)
tl.store(out_ptr0 + (x4), tmp22, xmask)
tl.store(in_out_ptr0 + (x4), tmp44, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/x7/cx7htd3xitvxaxetefd5p23n7ecwnggzfyxx3t47rzzxl37p6cg2.py
# Topologically Sorted Source Nodes: [cnt, sub_2, sub_1, cnt_fot_sigma, sigma_1, add, sigma_2, y], Original ATen: [aten.sum, aten.sub, aten.clamp, aten.div, aten.add, aten.sqrt]
# Source node to ATen node mapping:
# add => add
# cnt => sum_1
# cnt_fot_sigma => clamp_max_1, clamp_min_1
# sigma_1 => div_1
# sigma_2 => sqrt
# sub_1 => sub_1
# sub_2 => sub_2
# y => div_2
# Graph fragment:
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%unsqueeze, [-1], True), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %div), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sum_1, 1), kwargs = {})
# %clamp_min_1 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%sub_1, 1.0), kwargs = {})
# %clamp_max_1 : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min_1, 100000.0), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_3, %clamp_max_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_1, 1e-08), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add,), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_2, %sqrt), kwargs = {})
triton_poi_fused_add_clamp_div_sqrt_sub_sum_1 = async_compile.triton('triton_poi_fused_add_clamp_div_sqrt_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.pointwise(
size_hints=[1024],
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_clamp_div_sqrt_sub_sum_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_clamp_div_sqrt_sub_sum_1(in_ptr0, in_ptr1, in_ptr2, 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 % 256
x4 = (xindex // 4)
x5 = xindex
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x4), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 / tmp3
tl.store(out_ptr0 + (x5), 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, 1), (64, 16, 4, 1, 256), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4, 1), (64, 16, 4, 1, 256), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [mul, sum_2, cnt, cnt_for_mu, mu, sub, sigma, mul_1, sum_3, sub_1, cnt_fot_sigma, sigma_1, add, sigma_2], Original ATen: [aten.mul, aten.sum, aten.clamp, aten.div, aten.sub, aten.pow, aten.add, aten.sqrt]
stream0 = get_raw_stream(0)
triton_poi_fused_add_clamp_div_mul_pow_sqrt_sub_sum_0.run(buf2, arg1_1, arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
buf3 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cnt, sub_2, sub_1, cnt_fot_sigma, sigma_1, add, sigma_2, y], Original ATen: [aten.sum, aten.sub, aten.clamp, aten.div, aten.add, aten.sqrt]
triton_poi_fused_add_clamp_div_sqrt_sub_sum_1.run(arg1_1, buf0, buf2, buf3, 1024, grid=grid(1024), stream=stream0)
del arg1_1
del buf0
del buf2
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.cuda
from torch import nn
import torch.distributed
import torch.utils.data
import torch.optim
class MaskedInstanceNorm1d(nn.Module):
"""Instance norm + masking."""
MAX_CNT = 100000.0
def __init__(self, d_channel: 'int', unbiased: 'bool'=True, affine:
'bool'=False):
super().__init__()
self.d_channel = d_channel
self.unbiased = unbiased
self.affine = affine
if self.affine:
gamma = torch.ones(d_channel, dtype=torch.float)
beta = torch.zeros_like(gamma)
self.register_parameter('gamma', nn.Parameter(gamma))
self.register_parameter('beta', nn.Parameter(beta))
def forward(self, x: 'torch.Tensor', x_mask: 'torch.Tensor'
) ->torch.Tensor:
"""`x`: [B,C,T], `x_mask`: [B,T] => [B,C,T]."""
x_mask = x_mask.unsqueeze(1).type_as(x)
cnt = x_mask.sum(dim=-1, keepdim=True)
cnt_for_mu = cnt.clamp(1.0, self.MAX_CNT)
mu = (x * x_mask).sum(dim=-1, keepdim=True) / cnt_for_mu
sigma = (x - mu) ** 2
cnt_fot_sigma = (cnt - int(self.unbiased)).clamp(1.0, self.MAX_CNT)
sigma = (sigma * x_mask).sum(dim=-1, keepdim=True) / cnt_fot_sigma
sigma = (sigma + 1e-08).sqrt()
y = (x - mu) / sigma
if self.affine:
gamma = self.gamma.unsqueeze(0).unsqueeze(-1)
beta = self.beta.unsqueeze(0).unsqueeze(-1)
y = y * gamma + beta
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_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._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import torch.cuda
from torch import nn
import torch.distributed
import torch.utils.data
import torch.optim
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_div_mul_pow_sqrt_sub_sum_0(in_out_ptr0,
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 % 64
x0 = xindex % 16
x2 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x3, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4 * x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x3), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0 + 64 * x2), xmask, eviction_policy
='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x3), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0 + 64 * x2), xmask, eviction_policy
='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0 + 64 * 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 + tmp4
tmp16 = tmp15 + tmp8
tmp17 = tmp16 + tmp12
tmp18 = 1.0
tmp19 = triton_helpers.maximum(tmp17, tmp18)
tmp20 = 100000.0
tmp21 = triton_helpers.minimum(tmp19, tmp20)
tmp22 = tmp14 / tmp21
tmp23 = tmp0 - tmp22
tmp24 = tmp23 * tmp23
tmp25 = tmp24 * tmp1
tmp26 = tmp3 - tmp22
tmp27 = tmp26 * tmp26
tmp28 = tmp27 * tmp4
tmp29 = tmp25 + tmp28
tmp30 = tmp7 - tmp22
tmp31 = tmp30 * tmp30
tmp32 = tmp31 * tmp8
tmp33 = tmp29 + tmp32
tmp34 = tmp11 - tmp22
tmp35 = tmp34 * tmp34
tmp36 = tmp35 * tmp12
tmp37 = tmp33 + tmp36
tmp38 = tmp17 - tmp18
tmp39 = triton_helpers.maximum(tmp38, tmp18)
tmp40 = triton_helpers.minimum(tmp39, tmp20)
tmp41 = tmp37 / tmp40
tmp42 = 1e-08
tmp43 = tmp41 + tmp42
tmp44 = libdevice.sqrt(tmp43)
tl.store(out_ptr0 + x4, tmp22, xmask)
tl.store(in_out_ptr0 + x4, tmp44, xmask)
@triton.jit
def triton_poi_fused_add_clamp_div_sqrt_sub_sum_1(in_ptr0, in_ptr1, in_ptr2,
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 % 256
x4 = xindex // 4
x5 = xindex
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x4, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 / tmp3
tl.store(out_ptr0 + x5, 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, 1), (64, 16, 4, 1, 256),
torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4, 1), (64, 16, 4, 1, 256),
torch.float32)
buf2 = buf1
del buf1
get_raw_stream(0)
triton_poi_fused_add_clamp_div_mul_pow_sqrt_sub_sum_0[grid(256)](buf2,
arg1_1, arg0_1, buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
buf3 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1),
torch.float32)
triton_poi_fused_add_clamp_div_sqrt_sub_sum_1[grid(1024)](arg1_1,
buf0, buf2, buf3, 1024, XBLOCK=128, num_warps=4, num_stages=1)
del arg1_1
del buf0
del buf2
return buf3,
class MaskedInstanceNorm1dNew(nn.Module):
"""Instance norm + masking."""
MAX_CNT = 100000.0
def __init__(self, d_channel: 'int', unbiased: 'bool'=True, affine:
'bool'=False):
super().__init__()
self.d_channel = d_channel
self.unbiased = unbiased
self.affine = affine
if self.affine:
gamma = torch.ones(d_channel, dtype=torch.float)
beta = torch.zeros_like(gamma)
self.register_parameter('gamma', nn.Parameter(gamma))
self.register_parameter('beta', nn.Parameter(beta))
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| bugface/NeMo | MaskedInstanceNorm1d | false | 1,581 | [
"Apache-2.0"
] | 0 | 431c561380a120e9e164a4c9deed8f1ca9acace5 | https://github.com/bugface/NeMo/tree/431c561380a120e9e164a4c9deed8f1ca9acace5 | import torch
import torch.cuda
from torch import nn
import torch.distributed
import torch.utils.data
import torch.optim
class Model(nn.Module):
"""Instance norm + masking."""
MAX_CNT = 100000.0
def __init__(self, d_channel: 'int', unbiased: 'bool'=True, affine:
'bool'=False):
super().__init__()
self.d_channel = d_channel
self.unbiased = unbiased
self.affine = affine
if self.affine:
gamma = torch.ones(d_channel, dtype=torch.float)
beta = torch.zeros_like(gamma)
self.register_parameter('gamma', nn.Parameter(gamma))
self.register_parameter('beta', nn.Parameter(beta))
def forward(self, x: 'torch.Tensor', x_mask: 'torch.Tensor'
) ->torch.Tensor:
"""`x`: [B,C,T], `x_mask`: [B,T] => [B,C,T]."""
x_mask = x_mask.unsqueeze(1).type_as(x)
cnt = x_mask.sum(dim=-1, keepdim=True)
cnt_for_mu = cnt.clamp(1.0, self.MAX_CNT)
mu = (x * x_mask).sum(dim=-1, keepdim=True) / cnt_for_mu
sigma = (x - mu) ** 2
cnt_fot_sigma = (cnt - int(self.unbiased)).clamp(1.0, self.MAX_CNT)
sigma = (sigma * x_mask).sum(dim=-1, keepdim=True) / cnt_fot_sigma
sigma = (sigma + 1e-08).sqrt()
y = (x - mu) / sigma
if self.affine:
gamma = self.gamma.unsqueeze(0).unsqueeze(-1)
beta = self.beta.unsqueeze(0).unsqueeze(-1)
y = y * gamma + beta
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
LayerNorm | # 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_6/inductor_cache/3n/c3nvixp65c5in7rfhk46ir3grsaboyrs4my6jyzv5mwmau3tkh7l.py
# Topologically Sorted Source Nodes: [mean, std, sub, add, out, mul, out_1], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
# Source node to ATen node mapping:
# add => add
# mean => mean
# mul => mul
# out => div
# out_1 => add_1
# std => sqrt, var
# sub => sub
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [-1], True), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%primals_1, [-1]), kwargs = {correction: 1.0, keepdim: True})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_1, %mean), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt, 1e-12), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %add), 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_std_sub_0 = async_compile.triton('triton_poi_fused_add_div_mean_mul_std_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_add_div_mean_mul_std_sub_0', '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_std_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 % 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')
tmp3 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp2 - tmp10
tmp13 = tmp12 * tmp12
tmp14 = tmp3 - tmp10
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp10
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp10
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = 3.0
tmp24 = tmp22 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = 1e-12
tmp27 = tmp25 + tmp26
tmp28 = tmp11 / tmp27
tmp29 = tmp0 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + (x2), tmp31, 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: [mean, std, sub, add, out, mul, out_1], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_mean_mul_std_sub_0.run(primals_2, primals_1, primals_3, buf0, 256, grid=grid(256), stream=stream0)
del primals_2
del primals_3
return (buf0, 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
from torch import nn
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super(LayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_model': 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
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_div_mean_mul_std_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 % 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')
tmp3 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp2 - tmp10
tmp13 = tmp12 * tmp12
tmp14 = tmp3 - tmp10
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp10
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp10
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = 3.0
tmp24 = tmp22 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = 1e-12
tmp27 = tmp25 + tmp26
tmp28 = tmp11 / tmp27
tmp29 = tmp0 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + x2, tmp31, 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_add_div_mean_mul_std_sub_0[grid(256)](primals_2,
primals_1, primals_3, buf0, 256, XBLOCK=256, num_warps=4,
num_stages=1)
del primals_2
del primals_3
return buf0, primals_1
class LayerNormNew(nn.Module):
def __init__(self, d_model, eps=1e-12):
super(LayerNormNew, self).__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
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]
| bsgiovanini/transformer | LayerNorm | false | 1,582 | [
"Apache-2.0"
] | 0 | e128fa862f1b3d17d7b92df169a2bbee3f08366f | https://github.com/bsgiovanini/transformer/tree/e128fa862f1b3d17d7b92df169a2bbee3f08366f | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, d_model, eps=1e-12):
super().__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
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_6/inductor_cache/nw/cnwmrai6tuqz753kpnkczjc73mugn3chorrbvpqrqbq4xr2swbvn.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x_1 => 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=[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_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 = 25600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 400
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_6/inductor_cache/op/coptu6xep3awc4lajb4xivopppqmjtx3zy7ebtazm45rqvyeknds.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_3 => relu
# Graph fragment:
# %relu : [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, 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_6/inductor_cache/as/casrc7bf7ghsendgi7tkqxk3hj4ic6aqb4rmkxzuk5dhbidznia7.py
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.relu, aten.view]
# Source node to ATen node mapping:
# x_3 => relu
# x_4 => view_4
# Graph fragment:
# %relu : [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, [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_6/inductor_cache/sb/csbbn2sf7v6eutogmmhksp5x7rwd7z3vppgfx2mbu7v3njic7bqh.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x_5 => tanh_1
# Graph fragment:
# %tanh_1 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%view_5,), kwargs = {})
triton_poi_fused_tanh_3 = async_compile.triton('triton_poi_fused_tanh_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_tanh_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_tanh_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 = 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, 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
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_2, 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: [x_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: [x_3, x_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: [x_5], Original ATen: [aten.tanh]
triton_poi_fused_tanh_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), buf1, buf4, buf6, primals_6, buf7, 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((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 numpy as np
import torch.nn as nn
def fanin_init(size, fanin=None):
fanin = fanin or size[0]
v = 1.0 / np.sqrt(fanin)
return torch.Tensor(size).uniform_(-v, v)
class Actor(nn.Module):
def __init__(self, s_dim, a_dim):
super(Actor, self).__init__()
self.forward1 = nn.Linear(s_dim, 400)
self.Relu = nn.ReLU()
self.forward2 = nn.Linear(400, 300)
self.forward3 = nn.Linear(300, a_dim)
self.tanh = nn.Tanh()
for m in self.modules():
if isinstance(m, nn.Linear):
m.weight.data = fanin_init(m.weight.data.size())
self.forward2.weight.data.uniform_(-0.003, 0.003)
def forward(self, x):
x = self.forward1(x)
x = self.tanh(x)
x = self.forward2(x)
x = self.Relu(x)
x = self.forward3(x)
x = self.tanh(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'s_dim': 4, 'a_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
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
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 = 25600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 400
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_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_tanh_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 = libdevice.tanh(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
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(25600)](buf1, primals_2, 25600, XBLOCK
=128, 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=256, 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_tanh_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
), buf1, buf4, buf6, primals_6, buf7, primals_4
def fanin_init(size, fanin=None):
fanin = fanin or size[0]
v = 1.0 / np.sqrt(fanin)
return torch.Tensor(size).uniform_(-v, v)
class ActorNew(nn.Module):
def __init__(self, s_dim, a_dim):
super(ActorNew, self).__init__()
self.forward1 = nn.Linear(s_dim, 400)
self.Relu = nn.ReLU()
self.forward2 = nn.Linear(400, 300)
self.forward3 = nn.Linear(300, a_dim)
self.tanh = nn.Tanh()
for m in self.modules():
if isinstance(m, nn.Linear):
m.weight.data = fanin_init(m.weight.data.size())
self.forward2.weight.data.uniform_(-0.003, 0.003)
def forward(self, input_0):
primals_1 = self.forward1.weight
primals_2 = self.forward1.bias
primals_4 = self.forward2.weight
primals_5 = self.forward2.bias
primals_6 = self.forward3.weight
primals_7 = self.forward3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| brooky56/DeepRL-UR-motion-planning | Actor | false | 1,583 | [
"MIT"
] | 0 | 0cc523da6d8a55896773f1f57feed1f0c77fea78 | https://github.com/brooky56/DeepRL-UR-motion-planning/tree/0cc523da6d8a55896773f1f57feed1f0c77fea78 | import torch
import numpy as np
import torch.nn as nn
def fanin_init(size, fanin=None):
fanin = fanin or size[0]
v = 1.0 / np.sqrt(fanin)
return torch.Tensor(size).uniform_(-v, v)
class Model(nn.Module):
def __init__(self, s_dim, a_dim):
super().__init__()
self.forward1 = nn.Linear(s_dim, 400)
self.Relu = nn.ReLU()
self.forward2 = nn.Linear(400, 300)
self.forward3 = nn.Linear(300, a_dim)
self.tanh = nn.Tanh()
for m in self.modules():
if isinstance(m, nn.Linear):
m.weight.data = fanin_init(m.weight.data.size())
self.forward2.weight.data.uniform_(-0.003, 0.003)
def forward(self, x):
x = self.forward1(x)
x = self.tanh(x)
x = self.forward2(x)
x = self.Relu(x)
x = self.forward3(x)
x = self.tanh(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ScaleDotProductAttention | # 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_6/inductor_cache/hs/chsaatbgca2i2b4wr7ztvbkvqkcpp7gr4vndrvwjiwnbygf3hqzc.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# score_1 => exp
# Graph fragment:
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_3, 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=[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 = 0.5
tmp16 = tmp14 * tmp15
tmp17 = tl_math.exp(tmp16)
tl.store(out_ptr0 + (x3), tmp17, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/v4/cv4nyn2kde7dd2c53ddahw4vtxyldln6pqt62jrliqindkf3sj5m.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# score_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_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 = 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):
arg0_1, arg1_1, arg2_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))
assert_size_stride(arg2_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = 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(arg1_1, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(arg0_1, (16, 4, 4), (16, 4, 1), 0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf0, buf1, 256, grid=grid(256), stream=stream0)
buf2 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (16, 4, 4), (16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [v], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(arg2_1, (16, 4, 4), (16, 4, 1), 0), out=buf3)
del arg2_1
return (reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 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
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)
arg2_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1, arg2_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
from torch import nn
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4]), 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._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__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 = 0.5
tmp16 = tmp14 * tmp15
tmp17 = tl_math.exp(tmp16)
tl.store(out_ptr0 + x3, tmp17, 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 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
arg0_1, arg1_1, arg2_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))
assert_size_stride(arg2_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(arg1_1, (16, 4, 4), (16, 4, 1
), 0), reinterpret_tensor(arg0_1, (16, 4, 4), (16, 4, 1), 0),
out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(256)](buf0, buf1, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf2 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
triton_poi_fused__softmax_1[grid(256)](buf1, buf2, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (16, 4, 4), (16, 4, 1), 0)
del buf1
extern_kernels.bmm(reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(arg2_1, (16, 4, 4), (16, 4, 1), 0), out=buf3
)
del arg2_1
return reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0), buf2
class ScaleDotProductAttentionNew(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttentionNew, self).__init__()
self.softmax = nn.Softmax()
def forward(self, input_0, input_1, input_2):
arg0_1 = input_0
arg1_1 = input_1
arg2_1 = input_2
output = call([arg0_1, arg1_1, arg2_1])
return output[0], output[1]
| bsgiovanini/transformer | ScaleDotProductAttention | false | 1,584 | [
"Apache-2.0"
] | 0 | e128fa862f1b3d17d7b92df169a2bbee3f08366f | https://github.com/bsgiovanini/transformer/tree/e128fa862f1b3d17d7b92df169a2bbee3f08366f | import math
import torch
from torch import nn
class Model(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super().__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4]), torch.rand(
[4, 4, 4, 4])]
def get_init_inputs():
return []
|
EqualLinearActModule | # 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_6/inductor_cache/4q/c4qwjg3oiykrck3lufgrneimjnjyzxvb3c3h6obmpachvwty6igj.py
# Topologically Sorted Source Nodes: [sqrt, mul_1, weight], Original ATen: [aten.sqrt, aten.mul]
# Source node to ATen node mapping:
# mul_1 => mul_1
# sqrt => full_default_1
# weight => mul_2
# Graph fragment:
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.5), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %full_default_1), kwargs = {})
# %mul_2 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, 1.0), kwargs = {})
triton_poi_fused_mul_sqrt_0 = async_compile.triton('triton_poi_fused_mul_sqrt_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_mul_sqrt_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_sqrt_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 + (x0), xmask)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/2o/c2oqkq7zaubqmw7vuixxlseb2ff5jzqqbyczicxlmsahuxwdpdyp.py
# Topologically Sorted Source Nodes: [mul_3], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_3 => mul_3
# Graph fragment:
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, 1.0), 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=[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_mul_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_mul_1(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
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, 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, 1), torch.float32)
# Topologically Sorted Source Nodes: [sqrt, mul_1, weight], Original ATen: [aten.sqrt, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_sqrt_0.run(primals_2, buf0, 16, grid=grid(16), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mul_3], Original ATen: [aten.mul]
triton_poi_fused_mul_1.run(primals_3, buf1, 4, grid=grid(4), stream=stream0)
del primals_3
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_3], Original ATen: [aten.mul]
extern_kernels.addmm(buf1, primals_1, reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
return (buf2, buf0, 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, 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 as nn
from copy import deepcopy
from functools import partial
from torch.nn.init import _calculate_correct_fan
import torch.utils.cpp_extension
def equalized_lr(module, name='weight', gain=2 ** 0.5, mode='fan_in',
lr_mul=1.0):
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
EqualizedLR.apply(module, name, gain=gain, mode=mode, lr_mul=lr_mul)
return module
class EqualizedLR:
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
"""
def __init__(self, name='weight', gain=2 ** 0.5, mode='fan_in', lr_mul=1.0
):
self.name = name
self.mode = mode
self.gain = gain
self.lr_mul = lr_mul
def compute_weight(self, module):
"""Compute weight with equalized learning rate.
Args:
module (nn.Module): A module that is wrapped with equalized lr.
Returns:
torch.Tensor: Updated weight.
"""
weight = getattr(module, self.name + '_orig')
if weight.ndim == 5:
fan = _calculate_correct_fan(weight[0], self.mode)
else:
assert weight.ndim <= 4
fan = _calculate_correct_fan(weight, self.mode)
weight = weight * torch.tensor(self.gain, device=weight.device
) * torch.sqrt(torch.tensor(1.0 / fan, device=weight.device)
) * self.lr_mul
return weight
def __call__(self, module, inputs):
"""Standard interface for forward pre hooks."""
setattr(module, self.name, self.compute_weight(module))
@staticmethod
def apply(module, name, gain=2 ** 0.5, mode='fan_in', lr_mul=1.0):
"""Apply function.
This function is to register an equalized learning rate hook in an
``nn.Module``.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
for _, hook in module._forward_pre_hooks.items():
if isinstance(hook, EqualizedLR):
raise RuntimeError(
f'Cannot register two equalized_lr hooks on the same parameter {name} in {module} module.'
)
fn = EqualizedLR(name, gain=gain, mode=mode, lr_mul=lr_mul)
weight = module._parameters[name]
delattr(module, name)
module.register_parameter(name + '_orig', weight)
setattr(module, name, weight.data)
module.register_forward_pre_hook(fn)
return fn
class EqualizedLRLinearModule(nn.Linear):
"""Equalized LR LinearModule.
In this module, we adopt equalized lr in ``nn.Linear``. The equalized
learning rate is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
Note that, the initialization of ``self.weight`` will be overwritten as
:math:`\\mathcal{N}(0, 1)`.
Args:
equalized_lr_cfg (dict | None, optional): Config for ``EqualizedLR``.
If ``None``, equalized learning rate is ignored. Defaults to
dict(mode='fan_in').
"""
def __init__(self, *args, equalized_lr_cfg=dict(mode='fan_in'), **kwargs):
super().__init__(*args, **kwargs)
self.with_equalized_lr = equalized_lr_cfg is not None
if self.with_equalized_lr:
self.lr_mul = equalized_lr_cfg.get('lr_mul', 1.0)
else:
self.lr_mul = 1.0
if self.with_equalized_lr:
equalized_lr(self, **equalized_lr_cfg)
self._init_linear_weights()
def _init_linear_weights(self):
"""Initialize linear weights as described in PGGAN."""
nn.init.normal_(self.weight, 0, 1.0 / self.lr_mul)
if self.bias is not None:
nn.init.constant_(self.bias, 0.0)
class EqualLinearActModule(nn.Module):
"""Equalized LR Linear Module with Activation Layer.
This module is modified from ``EqualizedLRLinearModule`` defined in PGGAN.
The major features updated in this module is adding support for activation
layers used in StyleGAN2.
Args:
equalized_lr_cfg (dict | None, optional): Config for equalized lr.
Defaults to dict(gain=1., lr_mul=1.).
bias (bool, optional): Whether to use bias item. Defaults to True.
bias_init (float, optional): The value for bias initialization.
Defaults to ``0.``.
act_cfg (dict | None, optional): Config for activation layer.
Defaults to None.
"""
def __init__(self, *args, equalized_lr_cfg=dict(gain=1.0, lr_mul=1.0),
bias=True, bias_init=0.0, act_cfg=None, **kwargs):
super().__init__()
self.with_activation = act_cfg is not None
self.linear = EqualizedLRLinearModule(*args, bias=False,
equalized_lr_cfg=equalized_lr_cfg, **kwargs)
if equalized_lr_cfg is not None:
self.lr_mul = equalized_lr_cfg.get('lr_mul', 1.0)
else:
self.lr_mul = 1.0
if bias:
self.bias = nn.Parameter(torch.zeros(self.linear.out_features).
fill_(bias_init))
else:
self.bias = None
if self.with_activation:
act_cfg = deepcopy(act_cfg)
if act_cfg['type'] == 'fused_bias':
self.act_type = act_cfg.pop('type')
assert self.bias is not None
self.activate = partial(fused_bias_leakyrelu, **act_cfg)
else:
self.act_type = 'normal'
self.activate = build_activation_layer(act_cfg)
else:
self.act_type = None
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input feature map with shape of (N, C, ...).
Returns:
Tensor: Output feature map.
"""
if x.ndim >= 3:
x = x.reshape(x.size(0), -1)
x = self.linear(x)
if self.with_activation and self.act_type == 'fused_bias':
x = self.activate(x, self.bias * self.lr_mul)
elif self.bias is not None and self.with_activation:
x = self.activate(x + self.bias * self.lr_mul)
elif self.bias is not None:
x = x + self.bias * self.lr_mul
elif self.with_activation:
x = self.activate(x)
return x
def get_inputs():
return [torch.rand([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
import torch.nn as nn
from copy import deepcopy
from functools import partial
from torch.nn.init import _calculate_correct_fan
import torch.utils.cpp_extension
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_sqrt_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 + x0, xmask)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_poi_fused_mul_1(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
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, 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, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_sqrt_0[grid(16)](primals_2, buf0, 16, XBLOCK=
16, num_warps=1, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
triton_poi_fused_mul_1[grid(4)](primals_3, buf1, 4, XBLOCK=4,
num_warps=1, num_stages=1)
del primals_3
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(buf1, primals_1, reinterpret_tensor(buf0, (4,
4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
return buf2, buf0, primals_1
def equalized_lr(module, name='weight', gain=2 ** 0.5, mode='fan_in',
lr_mul=1.0):
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
EqualizedLR.apply(module, name, gain=gain, mode=mode, lr_mul=lr_mul)
return module
class EqualizedLR:
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
"""
def __init__(self, name='weight', gain=2 ** 0.5, mode='fan_in', lr_mul=1.0
):
self.name = name
self.mode = mode
self.gain = gain
self.lr_mul = lr_mul
def compute_weight(self, module):
"""Compute weight with equalized learning rate.
Args:
module (nn.Module): A module that is wrapped with equalized lr.
Returns:
torch.Tensor: Updated weight.
"""
weight = getattr(module, self.name + '_orig')
if weight.ndim == 5:
fan = _calculate_correct_fan(weight[0], self.mode)
else:
assert weight.ndim <= 4
fan = _calculate_correct_fan(weight, self.mode)
weight = weight * torch.tensor(self.gain, device=weight.device
) * torch.sqrt(torch.tensor(1.0 / fan, device=weight.device)
) * self.lr_mul
return weight
def __call__(self, module, inputs):
"""Standard interface for forward pre hooks."""
setattr(module, self.name, self.compute_weight(module))
@staticmethod
def apply(module, name, gain=2 ** 0.5, mode='fan_in', lr_mul=1.0):
"""Apply function.
This function is to register an equalized learning rate hook in an
``nn.Module``.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
for _, hook in module._forward_pre_hooks.items():
if isinstance(hook, EqualizedLR):
raise RuntimeError(
f'Cannot register two equalized_lr hooks on the same parameter {name} in {module} module.'
)
fn = EqualizedLR(name, gain=gain, mode=mode, lr_mul=lr_mul)
weight = module._parameters[name]
delattr(module, name)
module.register_parameter(name + '_orig', weight)
setattr(module, name, weight.data)
module.register_forward_pre_hook(fn)
return fn
class EqualizedLRLinearModule(nn.Linear):
"""Equalized LR LinearModule.
In this module, we adopt equalized lr in ``nn.Linear``. The equalized
learning rate is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
Note that, the initialization of ``self.weight`` will be overwritten as
:math:`\\mathcal{N}(0, 1)`.
Args:
equalized_lr_cfg (dict | None, optional): Config for ``EqualizedLR``.
If ``None``, equalized learning rate is ignored. Defaults to
dict(mode='fan_in').
"""
def __init__(self, *args, equalized_lr_cfg=dict(mode='fan_in'), **kwargs):
super().__init__(*args, **kwargs)
self.with_equalized_lr = equalized_lr_cfg is not None
if self.with_equalized_lr:
self.lr_mul = equalized_lr_cfg.get('lr_mul', 1.0)
else:
self.lr_mul = 1.0
if self.with_equalized_lr:
equalized_lr(self, **equalized_lr_cfg)
self._init_linear_weights()
def _init_linear_weights(self):
"""Initialize linear weights as described in PGGAN."""
nn.init.normal_(self.weight, 0, 1.0 / self.lr_mul)
if self.bias is not None:
nn.init.constant_(self.bias, 0.0)
class EqualLinearActModuleNew(nn.Module):
"""Equalized LR Linear Module with Activation Layer.
This module is modified from ``EqualizedLRLinearModule`` defined in PGGAN.
The major features updated in this module is adding support for activation
layers used in StyleGAN2.
Args:
equalized_lr_cfg (dict | None, optional): Config for equalized lr.
Defaults to dict(gain=1., lr_mul=1.).
bias (bool, optional): Whether to use bias item. Defaults to True.
bias_init (float, optional): The value for bias initialization.
Defaults to ``0.``.
act_cfg (dict | None, optional): Config for activation layer.
Defaults to None.
"""
def __init__(self, *args, equalized_lr_cfg=dict(gain=1.0, lr_mul=1.0),
bias=True, bias_init=0.0, act_cfg=None, **kwargs):
super().__init__()
self.with_activation = act_cfg is not None
self.linear = EqualizedLRLinearModule(*args, bias=False,
equalized_lr_cfg=equalized_lr_cfg, **kwargs)
if equalized_lr_cfg is not None:
self.lr_mul = equalized_lr_cfg.get('lr_mul', 1.0)
else:
self.lr_mul = 1.0
if bias:
self.bias = nn.Parameter(torch.zeros(self.linear.out_features).
fill_(bias_init))
else:
self.bias = None
if self.with_activation:
act_cfg = deepcopy(act_cfg)
if act_cfg['type'] == 'fused_bias':
self.act_type = act_cfg.pop('type')
assert self.bias is not None
self.activate = partial(fused_bias_leakyrelu, **act_cfg)
else:
self.act_type = 'normal'
self.activate = build_activation_layer(act_cfg)
else:
self.act_type = None
def forward(self, input_0):
primals_3 = self.bias
primals_1 = self.linear.weight_orig
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| bladesaber/mmgeneration | EqualLinearActModule | false | 1,585 | [
"Apache-2.0"
] | 0 | 158b49f7efd8028f231f6e9ca758ae0e20dd72ae | https://github.com/bladesaber/mmgeneration/tree/158b49f7efd8028f231f6e9ca758ae0e20dd72ae | import torch
import torch.nn as nn
from copy import deepcopy
from functools import partial
from torch.nn.init import _calculate_correct_fan
import torch.utils.cpp_extension
def equalized_lr(module, name='weight', gain=2 ** 0.5, mode='fan_in',
lr_mul=1.0):
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
EqualizedLR.apply(module, name, gain=gain, mode=mode, lr_mul=lr_mul)
return module
class EqualizedLR:
"""Equalized Learning Rate.
This trick is proposed in:
Progressive Growing of GANs for Improved Quality, Stability, and Variation
The general idea is to dynamically rescale the weight in training instead
of in initializing so that the variance of the responses in each layer is
guaranteed with some statistical properties.
Note that this function is always combined with a convolution module which
is initialized with :math:`\\mathcal{N}(0, 1)`.
Args:
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
"""
def __init__(self, name='weight', gain=2 ** 0.5, mode='fan_in', lr_mul=1.0
):
self.name = name
self.mode = mode
self.gain = gain
self.lr_mul = lr_mul
def compute_weight(self, module):
"""Compute weight with equalized learning rate.
Args:
module (nn.Module): A module that is wrapped with equalized lr.
Returns:
torch.Tensor: Updated weight.
"""
weight = getattr(module, self.name + '_orig')
if weight.ndim == 5:
fan = _calculate_correct_fan(weight[0], self.mode)
else:
assert weight.ndim <= 4
fan = _calculate_correct_fan(weight, self.mode)
weight = weight * torch.tensor(self.gain, device=weight.device
) * torch.sqrt(torch.tensor(1.0 / fan, device=weight.device)
) * self.lr_mul
return weight
def __call__(self, module, inputs):
"""Standard interface for forward pre hooks."""
setattr(module, self.name, self.compute_weight(module))
@staticmethod
def apply(module, name, gain=2 ** 0.5, mode='fan_in', lr_mul=1.0):
"""Apply function.
This function is to register an equalized learning rate hook in an
``nn.Module``.
Args:
module (nn.Module): Module to be wrapped.
name (str | optional): The name of weights. Defaults to 'weight'.
mode (str, optional): The mode of computing ``fan`` which is the
same as ``kaiming_init`` in pytorch. You can choose one from
['fan_in', 'fan_out']. Defaults to 'fan_in'.
Returns:
nn.Module: Module that is registered with equalized lr hook.
"""
for _, hook in module._forward_pre_hooks.items():
if isinstance(hook, EqualizedLR):
raise RuntimeError(
f'Cannot register two equalized_lr
# ... truncated (>4000 chars) for memory efficiency |
NPairLoss | # 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_6/inductor_cache/4u/c4uhmnbs7kn72trgxuebp2vjyxqx5mivfbe3mxraeksihwxqhcef.py
# Topologically Sorted Source Nodes: [sub], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%slice_1, %select_5), kwargs = {})
triton_poi_fused_sub_0 = async_compile.triton('triton_poi_fused_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=[8],
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_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_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8
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 + (8 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (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_6/inductor_cache/7z/c7zej5xvyg6rsrtf2ykx2vq2ejfi63q3lyq4ntj37kjojwgfd24g.py
# Topologically Sorted Source Nodes: [exp, sum_1], Original ATen: [aten.exp, aten.sum]
# Source node to ATen node mapping:
# exp => exp
# sum_1 => sum_1
# Graph fragment:
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%mm,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%exp,), kwargs = {})
triton_per_fused_exp_sum_1 = async_compile.triton('triton_per_fused_exp_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=[1, 2],
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), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_exp_sum_1', '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_exp_sum_1(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 2
RBLOCK: tl.constexpr = 2
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.exp(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')
# kernel path: runs/run_shard_6/inductor_cache/ub/cubtidwec344qdmkunm5v7rd2usfvtr6waitafjddrd4oguz43u5.py
# Topologically Sorted Source Nodes: [sub_1], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# sub_1 => sub_1
# Graph fragment:
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%slice_2, %select_7), kwargs = {})
triton_poi_fused_sub_2 = async_compile.triton('triton_poi_fused_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=[8],
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_sub_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_sub_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8
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 + (24 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (20 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ag/cagh2yycqhzgs6poqc32f77z5mxcrxevymomw66xbzhn7dzzxhxj.py
# Topologically Sorted Source Nodes: [sub_2], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# sub_2 => sub_2
# Graph fragment:
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%slice_3, %select_9), kwargs = {})
triton_poi_fused_sub_3 = async_compile.triton('triton_poi_fused_sub_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=[8],
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_sub_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_sub_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8
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 + (40 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (36 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/wo/cwocusxjigaqowhgy5k66zdoameilv3wl2dzdznpifhe4ztviwsd.py
# Topologically Sorted Source Nodes: [sub_3], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# sub_3 => sub_3
# Graph fragment:
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%slice_4, %select_11), kwargs = {})
triton_poi_fused_sub_4 = async_compile.triton('triton_poi_fused_sub_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=[8],
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_sub_4', '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_sub_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8
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 + (56 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (52 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/yi/cyi75druk3vu5r47w5xoybfsmoi6kojnazwywqmg4tv6eygufv4d.py
# Topologically Sorted Source Nodes: [pow_1, pow_2, add_4, sum_5, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
# Source node to ATen node mapping:
# add_4 => add_4
# pow_1 => pow_1
# pow_2 => pow_2
# stack_1 => cat_1
# sum_5 => sum_5
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_16, 2), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_17, 2), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_1, %pow_2), kwargs = {})
# %sum_5 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%add_4,), kwargs = {})
# %cat_1 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze_4, %unsqueeze_5, %unsqueeze_6, %unsqueeze_7],), kwargs = {})
triton_per_fused_add_pow_stack_sum_5 = async_compile.triton('triton_per_fused_add_pow_stack_sum_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.persistent_reduction(
size_hints=[1, 4],
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), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_pow_stack_sum_5', 'mutated_arg_names': [], '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_add_pow_stack_sum_5(in_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)
tmp2 = tl.load(in_ptr0 + (4 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp7, None)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/cg/ccg544lk64vm5756ci5i5xkrfn6drkxncn3e5mgf4uxifhlu7bql.py
# Topologically Sorted Source Nodes: [pow_3, pow_4, add_5, sum_6, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
# Source node to ATen node mapping:
# add_5 => add_5
# pow_3 => pow_3
# pow_4 => pow_4
# stack_1 => cat_1
# sum_6 => sum_6
# Graph fragment:
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_18, 2), kwargs = {})
# %pow_4 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_19, 2), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_3, %pow_4), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%add_5,), kwargs = {})
# %cat_1 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze_4, %unsqueeze_5, %unsqueeze_6, %unsqueeze_7],), kwargs = {})
triton_per_fused_add_pow_stack_sum_6 = async_compile.triton('triton_per_fused_add_pow_stack_sum_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.persistent_reduction(
size_hints=[1, 4],
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,), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_pow_stack_sum_6', 'mutated_arg_names': [], '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_add_pow_stack_sum_6(in_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 + (16 + r0), None)
tmp2 = tl.load(in_ptr0 + (20 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp7, None)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ue/cueoi24wmpr3qvns2fok4rdvnsxtdq3ejvhvmo3wrne3o4ysjmdv.py
# Topologically Sorted Source Nodes: [pow_5, pow_6, add_6, sum_7, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
# Source node to ATen node mapping:
# add_6 => add_6
# pow_5 => pow_5
# pow_6 => pow_6
# stack_1 => cat_1
# sum_7 => sum_7
# Graph fragment:
# %pow_5 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_20, 2), kwargs = {})
# %pow_6 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_21, 2), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_5, %pow_6), kwargs = {})
# %sum_7 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%add_6,), kwargs = {})
# %cat_1 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze_4, %unsqueeze_5, %unsqueeze_6, %unsqueeze_7],), kwargs = {})
triton_per_fused_add_pow_stack_sum_7 = async_compile.triton('triton_per_fused_add_pow_stack_sum_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=[1, 4],
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,), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_pow_stack_sum_7', 'mutated_arg_names': [], '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_add_pow_stack_sum_7(in_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 + (32 + r0), None)
tmp2 = tl.load(in_ptr0 + (36 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp7, None)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ax/caxykxw7b5vf6nuta6muxxegjetgi26xdmxoaj23yvhwhpxwhp5u.py
# Topologically Sorted Source Nodes: [pow_7, pow_8, add_7, sum_8, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
# Source node to ATen node mapping:
# add_7 => add_7
# pow_7 => pow_7
# pow_8 => pow_8
# stack_1 => cat_1
# sum_8 => sum_8
# Graph fragment:
# %pow_7 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_22, 2), kwargs = {})
# %pow_8 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_23, 2), kwargs = {})
# %add_7 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_7, %pow_8), kwargs = {})
# %sum_8 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%add_7,), kwargs = {})
# %cat_1 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze_4, %unsqueeze_5, %unsqueeze_6, %unsqueeze_7],), kwargs = {})
triton_per_fused_add_pow_stack_sum_8 = async_compile.triton('triton_per_fused_add_pow_stack_sum_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.persistent_reduction(
size_hints=[1, 4],
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,), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_pow_stack_sum_8', 'mutated_arg_names': [], '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_add_pow_stack_sum_8(in_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 + (48 + r0), None)
tmp2 = tl.load(in_ptr0 + (52 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp7, None)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/nj/cnjfoqaewqubjnh64dw4ik6fffnr4wagnoa4qxq4bom66os7bo5g.py
# Topologically Sorted Source Nodes: [loss, mean, mul, loss_1, mean_1], Original ATen: [aten.stack, aten.mean, aten.mul, aten.add]
# Source node to ATen node mapping:
# loss => cat
# loss_1 => add_8
# mean => mean
# mean_1 => mean_1
# mul => mul
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze, %unsqueeze_1, %unsqueeze_2, %unsqueeze_3],), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%cat_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mean, 0.05), kwargs = {})
# %add_8 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%cat, %mul), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%add_8,), kwargs = {})
triton_per_fused_add_mean_mul_stack_9 = async_compile.triton('triton_per_fused_add_mean_mul_stack_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=[1, 4],
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': {6: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=(6,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_mean_mul_stack_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, '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_mean_mul_stack_9(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, 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
tmp5 = tl.load(in_ptr0 + (0))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp16 = tl.load(in_ptr1 + (0))
tmp17 = tl.broadcast_to(tmp16, [XBLOCK, RBLOCK])
tmp26 = tl.load(in_ptr2 + (0))
tmp27 = tl.broadcast_to(tmp26, [XBLOCK, RBLOCK])
tmp35 = tl.load(in_ptr3 + (0))
tmp36 = tl.broadcast_to(tmp35, [XBLOCK, RBLOCK])
tmp44 = tl.load(in_ptr4 + (r0), None)
tmp0 = r0
tmp1 = tl.full([1, 1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1, 1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp7 = 1.0
tmp8 = tmp6 + tmp7
tmp9 = tl_math.log(tmp8)
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp4, tmp9, tmp10)
tmp12 = tmp0 >= tmp3
tmp13 = tl.full([1, 1], 2, tl.int64)
tmp14 = tmp0 < tmp13
tmp15 = tmp12 & tmp14
tmp18 = tmp17 + tmp7
tmp19 = tl_math.log(tmp18)
tmp20 = tl.full(tmp19.shape, 0.0, tmp19.dtype)
tmp21 = tl.where(tmp15, tmp19, tmp20)
tmp22 = tmp0 >= tmp13
tmp23 = tl.full([1, 1], 3, tl.int64)
tmp24 = tmp0 < tmp23
tmp25 = tmp22 & tmp24
tmp28 = tmp27 + tmp7
tmp29 = tl_math.log(tmp28)
tmp30 = tl.full(tmp29.shape, 0.0, tmp29.dtype)
tmp31 = tl.where(tmp25, tmp29, tmp30)
tmp32 = tmp0 >= tmp23
tmp33 = tl.full([1, 1], 4, tl.int64)
tmp34 = tmp0 < tmp33
tmp37 = tmp36 + tmp7
tmp38 = tl_math.log(tmp37)
tmp39 = tl.full(tmp38.shape, 0.0, tmp38.dtype)
tmp40 = tl.where(tmp32, tmp38, tmp39)
tmp41 = tl.where(tmp25, tmp31, tmp40)
tmp42 = tl.where(tmp15, tmp21, tmp41)
tmp43 = tl.where(tmp4, tmp11, tmp42)
tmp45 = tl.broadcast_to(tmp44, [XBLOCK, RBLOCK])
tmp47 = tl.sum(tmp45, 1)[:, None]
tmp48 = 4.0
tmp49 = tmp47 / tmp48
tmp50 = 0.05
tmp51 = tmp49 * tmp50
tmp52 = tmp43 + tmp51
tmp53 = tl.broadcast_to(tmp52, [XBLOCK, RBLOCK])
tmp55 = tl.sum(tmp53, 1)[:, None]
tmp56 = tmp55 / tmp48
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp56, None)
''', 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((2, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sub], Original ATen: [aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_sub_0.run(arg0_1, buf0, 8, grid=grid(8), stream=stream0)
buf1 = empty_strided_cuda((1, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [mm], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 0), reinterpret_tensor(buf0, (4, 2), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [exp, sum_1], Original ATen: [aten.exp, aten.sum]
triton_per_fused_exp_sum_1.run(buf1, buf2, 1, 2, grid=grid(1), stream=stream0)
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [sub_1], Original ATen: [aten.sub]
triton_poi_fused_sub_2.run(arg0_1, buf3, 8, grid=grid(8), stream=stream0)
buf4 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [mm_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 16), reinterpret_tensor(buf3, (4, 2), (1, 4), 0), out=buf4)
buf5 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [exp_1, sum_2], Original ATen: [aten.exp, aten.sum]
triton_per_fused_exp_sum_1.run(buf4, buf5, 1, 2, grid=grid(1), stream=stream0)
buf6 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [sub_2], Original ATen: [aten.sub]
triton_poi_fused_sub_3.run(arg0_1, buf6, 8, grid=grid(8), stream=stream0)
buf7 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [mm_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 32), reinterpret_tensor(buf6, (4, 2), (1, 4), 0), out=buf7)
buf8 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [exp_2, sum_3], Original ATen: [aten.exp, aten.sum]
triton_per_fused_exp_sum_1.run(buf7, buf8, 1, 2, grid=grid(1), stream=stream0)
buf9 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [sub_3], Original ATen: [aten.sub]
triton_poi_fused_sub_4.run(arg0_1, buf9, 8, grid=grid(8), stream=stream0)
buf10 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [mm_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 48), reinterpret_tensor(buf9, (4, 2), (1, 4), 0), out=buf10)
del buf9
buf11 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [exp_3, sum_4], Original ATen: [aten.exp, aten.sum]
triton_per_fused_exp_sum_1.run(buf10, buf11, 1, 2, grid=grid(1), stream=stream0)
del buf10
buf21 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf17 = reinterpret_tensor(buf21, (1, ), (1, ), 0) # alias
# Topologically Sorted Source Nodes: [pow_1, pow_2, add_4, sum_5, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
triton_per_fused_add_pow_stack_sum_5.run(arg0_1, buf17, 1, 4, grid=grid(1), stream=stream0)
buf18 = reinterpret_tensor(buf21, (1, ), (1, ), 1) # alias
# Topologically Sorted Source Nodes: [pow_3, pow_4, add_5, sum_6, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
triton_per_fused_add_pow_stack_sum_6.run(arg0_1, buf18, 1, 4, grid=grid(1), stream=stream0)
buf19 = reinterpret_tensor(buf21, (1, ), (1, ), 2) # alias
# Topologically Sorted Source Nodes: [pow_5, pow_6, add_6, sum_7, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
triton_per_fused_add_pow_stack_sum_7.run(arg0_1, buf19, 1, 4, grid=grid(1), stream=stream0)
buf20 = reinterpret_tensor(buf21, (1, ), (1, ), 3) # alias
# Topologically Sorted Source Nodes: [pow_7, pow_8, add_7, sum_8, stack_1], Original ATen: [aten.pow, aten.add, aten.sum, aten.stack]
triton_per_fused_add_pow_stack_sum_8.run(arg0_1, buf20, 1, 4, grid=grid(1), stream=stream0)
del arg0_1
buf22 = empty_strided_cuda((), (), torch.float32)
buf23 = buf22; del buf22 # reuse
buf24 = buf23; del buf23 # reuse
# Topologically Sorted Source Nodes: [loss, mean, mul, loss_1, mean_1], Original ATen: [aten.stack, aten.mean, aten.mul, aten.add]
triton_per_fused_add_mean_mul_stack_9.run(buf24, buf2, buf5, buf8, buf11, buf21, 1, 4, grid=grid(1), stream=stream0)
del buf11
del buf17
del buf18
del buf19
del buf2
del buf20
del buf21
del buf5
del buf8
return (buf24, )
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
class NPairLoss(torch.nn.Module):
def __init__(self, l2=0.05):
"""
Basic N-Pair Loss as proposed in 'Improved Deep Metric Learning with Multi-class N-pair Loss Objective'
Args:
l2: float, weighting parameter for weight penality due to embeddings not being normalized.
Returns:
Nothing!
"""
super(NPairLoss, self).__init__()
self.l2 = l2
def npair_distance(self, anchor, positive, negatives):
"""
Compute basic N-Pair loss.
Args:
anchor, positive, negative: torch.Tensor(), resp. embeddings for anchor, positive and negative samples.
Returns:
n-pair loss (torch.Tensor())
"""
return torch.log(1 + torch.sum(torch.exp(anchor.reshape(1, -1).mm((
negatives - positive).transpose(0, 1)))))
def weightsum(self, anchor, positive):
"""
Compute weight penalty.
NOTE: Only need to penalize anchor and positive since the negatives are created based on these.
Args:
anchor, positive: torch.Tensor(), resp. embeddings for anchor and positive samples.
Returns:
torch.Tensor(), Weight penalty
"""
return torch.sum(anchor ** 2 + positive ** 2)
def forward(self, batch):
"""
Args:
batch: torch.Tensor() [(BS x embed_dim)], batch of embeddings
Returns:
n-pair loss (torch.Tensor(), batch-averaged)
"""
loss = torch.stack([self.npair_distance(npair[0], npair[1], npair[2
:]) for npair in batch])
loss = loss + self.l2 * torch.mean(torch.stack([self.weightsum(
npair[0], npair[1]) for npair in batch]))
return torch.mean(loss)
def get_inputs():
return [torch.rand([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.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
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 8
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 + (8 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_per_fused_exp_sum_1(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
RBLOCK: tl.constexpr = 2
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.exp(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)
@triton.jit
def triton_poi_fused_sub_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 8
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 + (24 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (20 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_poi_fused_sub_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 8
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 + (40 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (36 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_poi_fused_sub_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 8
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 + (56 + x2), xmask)
tmp1 = tl.load(in_ptr0 + (52 + x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
@triton.jit
def triton_per_fused_add_pow_stack_sum_5(in_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)
tmp2 = tl.load(in_ptr0 + (4 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + tl.full([XBLOCK, 1], 0, tl.int32), tmp7, None)
@triton.jit
def triton_per_fused_add_pow_stack_sum_6(in_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 + (16 + r0), None)
tmp2 = tl.load(in_ptr0 + (20 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + tl.full([XBLOCK, 1], 0, tl.int32), tmp7, None)
@triton.jit
def triton_per_fused_add_pow_stack_sum_7(in_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 + (32 + r0), None)
tmp2 = tl.load(in_ptr0 + (36 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + tl.full([XBLOCK, 1], 0, tl.int32), tmp7, None)
@triton.jit
def triton_per_fused_add_pow_stack_sum_8(in_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 + (48 + r0), None)
tmp2 = tl.load(in_ptr0 + (52 + r0), None)
tmp1 = tmp0 * tmp0
tmp3 = tmp2 * tmp2
tmp4 = tmp1 + tmp3
tmp5 = tl.broadcast_to(tmp4, [XBLOCK, RBLOCK])
tmp7 = tl.sum(tmp5, 1)[:, None]
tl.store(out_ptr1 + tl.full([XBLOCK, 1], 0, tl.int32), tmp7, None)
@triton.jit
def triton_per_fused_add_mean_mul_stack_9(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, 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
tmp5 = tl.load(in_ptr0 + 0)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp16 = tl.load(in_ptr1 + 0)
tmp17 = tl.broadcast_to(tmp16, [XBLOCK, RBLOCK])
tmp26 = tl.load(in_ptr2 + 0)
tmp27 = tl.broadcast_to(tmp26, [XBLOCK, RBLOCK])
tmp35 = tl.load(in_ptr3 + 0)
tmp36 = tl.broadcast_to(tmp35, [XBLOCK, RBLOCK])
tmp44 = tl.load(in_ptr4 + r0, None)
tmp0 = r0
tl.full([1, 1], 0, tl.int64)
tmp3 = tl.full([1, 1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp7 = 1.0
tmp8 = tmp6 + tmp7
tmp9 = tl_math.log(tmp8)
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp4, tmp9, tmp10)
tmp12 = tmp0 >= tmp3
tmp13 = tl.full([1, 1], 2, tl.int64)
tmp14 = tmp0 < tmp13
tmp15 = tmp12 & tmp14
tmp18 = tmp17 + tmp7
tmp19 = tl_math.log(tmp18)
tmp20 = tl.full(tmp19.shape, 0.0, tmp19.dtype)
tmp21 = tl.where(tmp15, tmp19, tmp20)
tmp22 = tmp0 >= tmp13
tmp23 = tl.full([1, 1], 3, tl.int64)
tmp24 = tmp0 < tmp23
tmp25 = tmp22 & tmp24
tmp28 = tmp27 + tmp7
tmp29 = tl_math.log(tmp28)
tmp30 = tl.full(tmp29.shape, 0.0, tmp29.dtype)
tmp31 = tl.where(tmp25, tmp29, tmp30)
tmp32 = tmp0 >= tmp23
tl.full([1, 1], 4, tl.int64)
tmp37 = tmp36 + tmp7
tmp38 = tl_math.log(tmp37)
tmp39 = tl.full(tmp38.shape, 0.0, tmp38.dtype)
tmp40 = tl.where(tmp32, tmp38, tmp39)
tmp41 = tl.where(tmp25, tmp31, tmp40)
tmp42 = tl.where(tmp15, tmp21, tmp41)
tmp43 = tl.where(tmp4, tmp11, tmp42)
tmp45 = tl.broadcast_to(tmp44, [XBLOCK, RBLOCK])
tmp47 = tl.sum(tmp45, 1)[:, None]
tmp48 = 4.0
tmp49 = tmp47 / tmp48
tmp50 = 0.05
tmp51 = tmp49 * tmp50
tmp52 = tmp43 + tmp51
tmp53 = tl.broadcast_to(tmp52, [XBLOCK, RBLOCK])
tmp55 = tl.sum(tmp53, 1)[:, None]
tmp56 = tmp55 / tmp48
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp56, None)
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((2, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_sub_0[grid(8)](arg0_1, buf0, 8, XBLOCK=8,
num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((1, 2), (2, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 0),
reinterpret_tensor(buf0, (4, 2), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_exp_sum_1[grid(1)](buf1, buf2, 1, 2, XBLOCK=1,
num_warps=2, num_stages=1)
buf3 = buf0
del buf0
triton_poi_fused_sub_2[grid(8)](arg0_1, buf3, 8, XBLOCK=8,
num_warps=1, num_stages=1)
buf4 = buf1
del buf1
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 16),
reinterpret_tensor(buf3, (4, 2), (1, 4), 0), out=buf4)
buf5 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_exp_sum_1[grid(1)](buf4, buf5, 1, 2, XBLOCK=1,
num_warps=2, num_stages=1)
buf6 = buf3
del buf3
triton_poi_fused_sub_3[grid(8)](arg0_1, buf6, 8, XBLOCK=8,
num_warps=1, num_stages=1)
buf7 = buf4
del buf4
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 32),
reinterpret_tensor(buf6, (4, 2), (1, 4), 0), out=buf7)
buf8 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_exp_sum_1[grid(1)](buf7, buf8, 1, 2, XBLOCK=1,
num_warps=2, num_stages=1)
buf9 = buf6
del buf6
triton_poi_fused_sub_4[grid(8)](arg0_1, buf9, 8, XBLOCK=8,
num_warps=1, num_stages=1)
buf10 = buf7
del buf7
extern_kernels.mm(reinterpret_tensor(arg0_1, (1, 4), (4, 1), 48),
reinterpret_tensor(buf9, (4, 2), (1, 4), 0), out=buf10)
del buf9
buf11 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_exp_sum_1[grid(1)](buf10, buf11, 1, 2, XBLOCK=1,
num_warps=2, num_stages=1)
del buf10
buf21 = empty_strided_cuda((4,), (1,), torch.float32)
buf17 = reinterpret_tensor(buf21, (1,), (1,), 0)
triton_per_fused_add_pow_stack_sum_5[grid(1)](arg0_1, buf17, 1, 4,
XBLOCK=1, num_warps=2, num_stages=1)
buf18 = reinterpret_tensor(buf21, (1,), (1,), 1)
triton_per_fused_add_pow_stack_sum_6[grid(1)](arg0_1, buf18, 1, 4,
XBLOCK=1, num_warps=2, num_stages=1)
buf19 = reinterpret_tensor(buf21, (1,), (1,), 2)
triton_per_fused_add_pow_stack_sum_7[grid(1)](arg0_1, buf19, 1, 4,
XBLOCK=1, num_warps=2, num_stages=1)
buf20 = reinterpret_tensor(buf21, (1,), (1,), 3)
triton_per_fused_add_pow_stack_sum_8[grid(1)](arg0_1, buf20, 1, 4,
XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
buf22 = empty_strided_cuda((), (), torch.float32)
buf23 = buf22
del buf22
buf24 = buf23
del buf23
triton_per_fused_add_mean_mul_stack_9[grid(1)](buf24, buf2, buf5,
buf8, buf11, buf21, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del buf11
del buf17
del buf18
del buf19
del buf2
del buf20
del buf21
del buf5
del buf8
return buf24,
class NPairLossNew(torch.nn.Module):
def __init__(self, l2=0.05):
"""
Basic N-Pair Loss as proposed in 'Improved Deep Metric Learning with Multi-class N-pair Loss Objective'
Args:
l2: float, weighting parameter for weight penality due to embeddings not being normalized.
Returns:
Nothing!
"""
super(NPairLossNew, self).__init__()
self.l2 = l2
def npair_distance(self, anchor, positive, negatives):
"""
Compute basic N-Pair loss.
Args:
anchor, positive, negative: torch.Tensor(), resp. embeddings for anchor, positive and negative samples.
Returns:
n-pair loss (torch.Tensor())
"""
return torch.log(1 + torch.sum(torch.exp(anchor.reshape(1, -1).mm((
negatives - positive).transpose(0, 1)))))
def weightsum(self, anchor, positive):
"""
Compute weight penalty.
NOTE: Only need to penalize anchor and positive since the negatives are created based on these.
Args:
anchor, positive: torch.Tensor(), resp. embeddings for anchor and positive samples.
Returns:
torch.Tensor(), Weight penalty
"""
return torch.sum(anchor ** 2 + positive ** 2)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bm2-lab/MDML | NPairLoss | false | 1,586 | [
"MIT"
] | 0 | 222fb22b2ee53dd3c1a6f2e99a88f71e9635e3a0 | https://github.com/bm2-lab/MDML/tree/222fb22b2ee53dd3c1a6f2e99a88f71e9635e3a0 | import torch
class Model(torch.nn.Module):
def __init__(self, l2=0.05):
"""
Basic N-Pair Loss as proposed in 'Improved Deep Metric Learning with Multi-class N-pair Loss Objective'
Args:
l2: float, weighting parameter for weight penality due to embeddings not being normalized.
Returns:
Nothing!
"""
super().__init__()
self.l2 = l2
def npair_distance(self, anchor, positive, negatives):
"""
Compute basic N-Pair loss.
Args:
anchor, positive, negative: torch.Tensor(), resp. embeddings for anchor, positive and negative samples.
Returns:
n-pair loss (torch.Tensor())
"""
return torch.log(1 + torch.sum(torch.exp(anchor.reshape(1, -1).mm((
negatives - positive).transpose(0, 1)))))
def weightsum(self, anchor, positive):
"""
Compute weight penalty.
NOTE: Only need to penalize anchor and positive since the negatives are created based on these.
Args:
anchor, positive: torch.Tensor(), resp. embeddings for anchor and positive samples.
Returns:
torch.Tensor(), Weight penalty
"""
return torch.sum(anchor ** 2 + positive ** 2)
def forward(self, batch):
"""
Args:
batch: torch.Tensor() [(BS x embed_dim)], batch of embeddings
Returns:
n-pair loss (torch.Tensor(), batch-averaged)
"""
loss = torch.stack([self.npair_distance(npair[0], npair[1], npair[2
:]) for npair in batch])
loss = loss + self.l2 * torch.mean(torch.stack([self.weightsum(
npair[0], npair[1]) for npair in batch]))
return torch.mean(loss)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return []
|
FunctionalRelu6 | # 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_6/inductor_cache/36/c36w3rtlopsrs3btjlizo5n4q3ytar7mtz2eoqnft735ksvzltlx.py
# Topologically Sorted Source Nodes: [relu6], Original ATen: [aten.hardtanh]
# Source node to ATen node mapping:
# relu6 => clamp_max, clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%arg0_1, 0), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 6), kwargs = {})
triton_poi_fused_hardtanh_0 = async_compile.triton('triton_poi_fused_hardtanh_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_hardtanh_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_hardtanh_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 = 6.0
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: [relu6], Original ATen: [aten.hardtanh]
stream0 = get_raw_stream(0)
triton_poi_fused_hardtanh_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
class FunctionalRelu6(torch.nn.Module):
def forward(self, x):
return torch.nn.functional.relu6(x)
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
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_hardtanh_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 = 6.0
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_hardtanh_0[grid(256)](arg0_1, buf0, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class FunctionalRelu6New(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | FunctionalRelu6 | false | 1,587 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x):
return torch.nn.functional.relu6(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MaxElementwise | # 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_6/inductor_cache/xn/cxncjpzgxjplm55ywcfy5vnpvvgzqmw56ruh2sgj4c3gtprfogbe.py
# Topologically Sorted Source Nodes: [max_1], Original ATen: [aten.maximum]
# Source node to ATen node mapping:
# max_1 => maximum
# Graph fragment:
# %maximum : [num_users=1] = call_function[target=torch.ops.aten.maximum.default](args = (%arg1_1, %arg0_1), kwargs = {})
triton_poi_fused_maximum_0 = async_compile.triton('triton_poi_fused_maximum_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_maximum_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_maximum_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tl.store(out_ptr0 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [max_1], Original ATen: [aten.maximum]
stream0 = get_raw_stream(0)
triton_poi_fused_maximum_0.run(arg1_1, arg0_1, buf0, 256, grid=grid(256), 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, 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
class MaxElementwise(torch.nn.Module):
def forward(self, x, y):
return torch.max(x, y)
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
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_maximum_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_maximum_0[grid(256)](arg1_1, arg0_1, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class MaxElementwiseNew(torch.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]
| bunderhi/torch2trt | MaxElementwise | false | 1,588 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x, y):
return torch.max(x, y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MultiHeadAttention | # 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_6/inductor_cache/l3/cl3mqwaki56dc4zcxfjjgkbopnejxzhksqm6egdinynmjrsrw2qw.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# score_1 => exp
# Graph fragment:
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_12, 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, 1.0), 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=[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 = tmp14 * tmp1
tmp16 = tl_math.exp(tmp15)
tl.store(out_ptr0 + (x3), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/v4/cv4nyn2kde7dd2c53ddahw4vtxyldln6pqt62jrliqindkf3sj5m.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# score_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_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 = 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, 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), (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))
assert_size_stride(primals_10, (4, 4), (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: [q], 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: [k], 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: [v], 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((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, 1), 0), reinterpret_tensor(buf1, (16, 1, 4), (4, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf3, buf4, 256, grid=grid(256), stream=stream0)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
del buf4
buf6 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [v_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0), out=buf6)
buf7 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, reinterpret_tensor(buf6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf7)
del primals_11
return (reinterpret_tensor(buf7, (4, 4, 4), (16, 4, 1), 0), 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(buf6, (16, 4), (4, 1), 0), primals_10, reinterpret_tensor(buf2, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf0, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf1, (16, 4, 1), (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, 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)
primals_10 = rand_strided((4, 4), (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)
| import math
import torch
from torch import nn
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
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, 'n_head': 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 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__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 = tmp14 * tmp1
tmp16 = tl_math.exp(tmp15)
tl.store(out_ptr0 + x3, tmp16, 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 = 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, 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), (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))
assert_size_stride(primals_10, (4, 4), (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.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((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 1),
0), reinterpret_tensor(buf1, (16, 1, 4), (4, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(256)](buf3, buf4, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused__softmax_1[grid(256)](buf4, buf5, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf4
buf6 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0), out=buf6)
buf7 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_11, reinterpret_tensor(buf6, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_10, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf7)
del primals_11
return reinterpret_tensor(buf7, (4, 4, 4), (16, 4, 1), 0
), 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(buf6, (16, 4), (4, 1), 0
), primals_10, reinterpret_tensor(buf2, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf0, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf1, (16, 4, 1), (4, 1, 4), 0)
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttentionNew(nn.Module):
def __init__(self, d_model, n_head):
super(MultiHeadAttentionNew, self).__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
def forward(self, input_0, input_1, input_2):
primals_1 = self.w_q.weight
primals_2 = self.w_q.bias
primals_4 = self.w_k.weight
primals_5 = self.w_k.bias
primals_7 = self.w_v.weight
primals_8 = self.w_v.bias
primals_10 = self.w_concat.weight
primals_11 = self.w_concat.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,
primals_10, primals_11])
return output[0]
| bsgiovanini/transformer | MultiHeadAttention | false | 1,589 | [
"Apache-2.0"
] | 0 | e128fa862f1b3d17d7b92df169a2bbee3f08366f | https://github.com/bsgiovanini/transformer/tree/e128fa862f1b3d17d7b92df169a2bbee3f08366f | import math
import torch
from torch import nn
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super().__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class Model(nn.Module):
def __init__(self, d_model, n_head):
super().__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
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]
|
WaveNetLayer | # 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_6/inductor_cache/im/cimrryk5tlijxgch3lnrftbusuhrtt63nyisb4difipuywd7wx5o.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.constant_pad_nd]
# Source node to ATen node mapping:
# x => constant_pad_nd
# Graph fragment:
# %constant_pad_nd : [num_users=2] = call_function[target=torch.ops.aten.constant_pad_nd.default](args = (%primals_1, [0, 1]), 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=[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_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 = 80
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = (xindex // 5)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 4, tl.int64)
tmp2 = tmp0 < tmp1
tmp3 = tl.load(in_ptr0 + (x0 + (4*x1)), tmp2 & xmask, other=0.0)
tl.store(out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/kx/ckxt5ip3pdehe4uvtgnb7arkqq5gkapqc7o4qwwpbxdypvnq4zxb.py
# Topologically Sorted Source Nodes: [tanh, sigmoid, x_2], Original ATen: [aten.tanh, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# sigmoid => sigmoid
# tanh => tanh
# x_2 => mul
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%getitem,), kwargs = {})
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem_1,), kwargs = {})
# %mul : [num_users=3] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh, %sigmoid), kwargs = {})
triton_poi_fused_mul_sigmoid_tanh_1 = async_compile.triton('triton_poi_fused_mul_sigmoid_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=[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_mul_sigmoid_tanh_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_mul_sigmoid_tanh_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 8)
x4 = xindex % 8
x1 = (xindex // 4) % 2
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x4 + (16*x2)), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x4 + (16*x2)), xmask)
tmp5 = tl.load(in_ptr1 + (2 + x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tmp6 = tmp4 + tmp5
tmp7 = tl.sigmoid(tmp6)
tmp8 = tmp3 * tmp7
tl.store(out_ptr0 + (x3), tmp3, xmask)
tl.store(out_ptr1 + (x3), tmp7, xmask)
tl.store(out_ptr2 + (x3), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/lf/clf7hs52i4bd5d3e73uio27ntyjfqmszkbsw6dta3r6rzgeftva3.py
# Topologically Sorted Source Nodes: [s], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# s => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%mul, %primals_4, %primals_5, [1], [0], [1], False, [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=[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_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 = 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_6/inductor_cache/jg/cjgfre7znianqklg7g5ijua6n6nby7yvw4ljytb2viu2zu3gjmst.py
# Topologically Sorted Source Nodes: [x_3, add, sqrt, x_4], Original ATen: [aten.convolution, aten.add, aten.sqrt, aten.mul]
# Source node to ATen node mapping:
# add => add
# sqrt => full_default
# x_3 => convolution_2
# x_4 => mul_1
# Graph fragment:
# %convolution_2 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%mul, %primals_6, %primals_7, [1], [0], [1], False, [0], 1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_2, %primals_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.7071067690849304), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, %full_default), kwargs = {})
triton_poi_fused_add_convolution_mul_sqrt_3 = async_compile.triton('triton_poi_fused_add_convolution_mul_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.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_convolution_mul_sqrt_3', '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_convolution_mul_sqrt_3(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
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')
tmp3 = tl.load(in_ptr1 + (x3), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = 0.7071067690849304
tmp6 = tmp4 * tmp5
tl.store(in_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, primals_6, primals_7 = 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, 2, 1), (2, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 2, 1), (2, 1, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 5), (20, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.constant_pad_nd]
stream0 = get_raw_stream(0)
triton_poi_fused_constant_pad_nd_0.run(primals_1, buf0, 80, grid=grid(80), stream=stream0)
# Topologically Sorted Source Nodes: [x], 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, 4), (16, 4, 1))
buf2 = empty_strided_cuda((4, 2, 4), (8, 4, 1), torch.float32)
buf3 = empty_strided_cuda((4, 2, 4), (8, 4, 1), torch.float32)
buf4 = empty_strided_cuda((4, 2, 4), (8, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [tanh, sigmoid, x_2], Original ATen: [aten.tanh, aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_tanh_1.run(buf1, primals_3, buf2, buf3, buf4, 32, grid=grid(32), stream=stream0)
del buf1
del primals_3
# Topologically Sorted Source Nodes: [s], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(buf4, primals_4, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf5, (4, 4, 4), (16, 4, 1))
buf6 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [s], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf6, primals_5, 64, grid=grid(64), stream=stream0)
del primals_5
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf7 = extern_kernels.convolution(buf4, primals_6, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf7, (4, 4, 4), (16, 4, 1))
buf8 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [x_3, add, sqrt, x_4], Original ATen: [aten.convolution, aten.add, aten.sqrt, aten.mul]
triton_poi_fused_add_convolution_mul_sqrt_3.run(buf8, primals_7, primals_1, 64, grid=grid(64), stream=stream0)
del primals_1
del primals_7
return (buf8, buf6, primals_2, primals_4, primals_6, buf0, buf2, 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((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, 2, 1), (2, 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, 2, 1), (2, 1, 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 typing as T
class WaveNetLayer(torch.nn.Module):
"""a single gated residual wavenet layer"""
def __init__(self, channels: 'int', kernel_size: 'int', dilation: 'int'):
super().__init__()
self._conv = torch.nn.Conv1d(in_channels=channels, out_channels=
channels, kernel_size=kernel_size, padding='same', dilation=
dilation)
self._conv_skip = torch.nn.Conv1d(in_channels=channels // 2,
out_channels=channels, kernel_size=1)
self._conv_out = torch.nn.Conv1d(in_channels=channels // 2,
out_channels=channels, kernel_size=1)
def forward(self, x: 'torch.Tensor') ->T.Tuple[torch.Tensor, torch.Tensor]:
r = x
x = self._conv(x)
x, g = x.split(x.size(1) // 2, dim=1)
x = torch.tanh(x) * torch.sigmoid(g)
s = self._conv_skip(x)
x = self._conv_out(x)
x = (x + r) * torch.tensor(0.5).sqrt()
return x, s
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'channels': 4, 'kernel_size': 4, 'dilation': 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
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 = 80
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = xindex // 5
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 4, tl.int64)
tmp2 = tmp0 < tmp1
tmp3 = tl.load(in_ptr0 + (x0 + 4 * x1), tmp2 & xmask, other=0.0)
tl.store(out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_mul_sigmoid_tanh_1(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, out_ptr2, xnumel, XBLOCK: tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 8
x4 = xindex % 8
x1 = xindex // 4 % 2
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x4 + 16 * x2), xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x4 + 16 * x2), xmask)
tmp5 = tl.load(in_ptr1 + (2 + x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tmp6 = tmp4 + tmp5
tmp7 = tl.sigmoid(tmp6)
tmp8 = tmp3 * tmp7
tl.store(out_ptr0 + x3, tmp3, xmask)
tl.store(out_ptr1 + x3, tmp7, xmask)
tl.store(out_ptr2 + x3, tmp8, xmask)
@triton.jit
def triton_poi_fused_convolution_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
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_add_convolution_mul_sqrt_3(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
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')
tmp3 = tl.load(in_ptr1 + x3, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = 0.7071067690849304
tmp6 = tmp4 * tmp5
tl.store(in_out_ptr0 + x3, tmp6, 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), (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, 2, 1), (2, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 2, 1), (2, 1, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 5), (20, 5, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_constant_pad_nd_0[grid(80)](primals_1, buf0, 80,
XBLOCK=128, num_warps=4, 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, 4), (16, 4, 1))
buf2 = empty_strided_cuda((4, 2, 4), (8, 4, 1), torch.float32)
buf3 = empty_strided_cuda((4, 2, 4), (8, 4, 1), torch.float32)
buf4 = empty_strided_cuda((4, 2, 4), (8, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_tanh_1[grid(32)](buf1, primals_3, buf2,
buf3, buf4, 32, XBLOCK=32, num_warps=1, num_stages=1)
del buf1
del primals_3
buf5 = extern_kernels.convolution(buf4, primals_4, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf5, (4, 4, 4), (16, 4, 1))
buf6 = buf5
del buf5
triton_poi_fused_convolution_2[grid(64)](buf6, primals_5, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_5
buf7 = extern_kernels.convolution(buf4, primals_6, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf7, (4, 4, 4), (16, 4, 1))
buf8 = buf7
del buf7
triton_poi_fused_add_convolution_mul_sqrt_3[grid(64)](buf8,
primals_7, primals_1, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_1
del primals_7
return buf8, buf6, primals_2, primals_4, primals_6, buf0, buf2, buf3, buf4
class WaveNetLayerNew(torch.nn.Module):
"""a single gated residual wavenet layer"""
def __init__(self, channels: 'int', kernel_size: 'int', dilation: 'int'):
super().__init__()
self._conv = torch.nn.Conv1d(in_channels=channels, out_channels=
channels, kernel_size=kernel_size, padding='same', dilation=
dilation)
self._conv_skip = torch.nn.Conv1d(in_channels=channels // 2,
out_channels=channels, kernel_size=1)
self._conv_out = torch.nn.Conv1d(in_channels=channels // 2,
out_channels=channels, kernel_size=1)
def forward(self, input_0):
primals_1 = self._conv.weight
primals_3 = self._conv.bias
primals_4 = self._conv_skip.weight
primals_5 = self._conv_skip.bias
primals_6 = self._conv_out.weight
primals_7 = self._conv_out.bias
primals_2 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0], output[1]
| brentspell/hifi-gan-bwe | WaveNetLayer | false | 1,590 | [
"MIT"
] | 0 | 63579ac8055c63fc0e5a20ae90e2a86575fc8e12 | https://github.com/brentspell/hifi-gan-bwe/tree/63579ac8055c63fc0e5a20ae90e2a86575fc8e12 | import torch
import typing as T
class Model(torch.nn.Module):
"""a single gated residual wavenet layer"""
def __init__(self, channels: 'int', kernel_size: 'int', dilation: 'int'):
super().__init__()
self._conv = torch.nn.Conv1d(in_channels=channels, out_channels=
channels, kernel_size=kernel_size, padding='same', dilation=
dilation)
self._conv_skip = torch.nn.Conv1d(in_channels=channels // 2,
out_channels=channels, kernel_size=1)
self._conv_out = torch.nn.Conv1d(in_channels=channels // 2,
out_channels=channels, kernel_size=1)
def forward(self, x: 'torch.Tensor') ->T.Tuple[torch.Tensor, torch.Tensor]:
r = x
x = self._conv(x)
x, g = x.split(x.size(1) // 2, dim=1)
x = torch.tanh(x) * torch.sigmoid(g)
s = self._conv_skip(x)
x = self._conv_out(x)
x = (x + r) * torch.tensor(0.5).sqrt()
return x, s
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4, 1]
|
IDiv | # 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_6/inductor_cache/eg/ceghw55buy2kdxurnttygvw63vke5iiofs62xigwwfgbadj5oqov.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.div]
# Source node to ATen node mapping:
# x => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %copy_ : [num_users=1] = call_function[target=torch.ops.aten.copy_.default](args = (%arg0_1, %div), kwargs = {})
triton_poi_fused_div_0 = async_compile.triton('triton_poi_fused_div_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_div_0', 'mutated_arg_names': ['in_ptr0', 'out_ptr1'], '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_0(in_ptr0, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tmp0 / tmp1
tl.store(out_ptr1 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_0.run(arg0_1, arg1_1, arg0_1, 256, grid=grid(256), stream=stream0)
del arg1_1
return (arg0_1, )
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
class IDiv(torch.nn.Module):
def __init__(self):
super(IDiv, self).__init__()
def forward(self, x, y):
x /= y
return x
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
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_div_0(in_ptr0, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 / tmp1
tl.store(out_ptr1 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_div_0[grid(256)](arg0_1, arg1_1, arg0_1, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg1_1
return arg0_1,
class IDivNew(torch.nn.Module):
def __init__(self):
super(IDivNew, 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]
| bunderhi/torch2trt | IDiv | false | 1,591 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
x /= y
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Mul | # 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_6/inductor_cache/p3/cp3qleddjiuuytozrtebx5pzf2ycpwtw4mkq2jsx7qqswymv2bm6.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, %arg1_1), 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: '*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 = 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)
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, 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: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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
class Mul(torch.nn.Module):
def __init__(self):
super(Mul, self).__init__()
def forward(self, x, y):
return x * y
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
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, 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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK
=128, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class MulNew(torch.nn.Module):
def __init__(self):
super(MulNew, 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]
| bunderhi/torch2trt | Mul | false | 1,592 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return x * y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
EQ | # 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_6/inductor_cache/al/calhisv2tizhq3s3r7egazungn7zoqv2w3alcz76uykvsknhyvtu.py
# Topologically Sorted Source Nodes: [eq], Original ATen: [aten.eq]
# Source node to ATen node mapping:
# eq => eq
# Graph fragment:
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
triton_poi_fused_eq_0 = async_compile.triton('triton_poi_fused_eq_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_eq_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_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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.bool)
# Topologically Sorted Source Nodes: [eq], Original ATen: [aten.eq]
stream0 = get_raw_stream(0)
triton_poi_fused_eq_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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
class EQ(torch.nn.Module):
def __init__(self):
super(EQ, self).__init__()
def forward(self, x, y):
return x == y
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
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_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 == tmp1
tl.store(out_ptr0 + x0, tmp2, 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.bool)
get_raw_stream(0)
triton_poi_fused_eq_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class EQNew(torch.nn.Module):
def __init__(self):
super(EQNew, 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]
| bunderhi/torch2trt | EQ | false | 1,593 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return x == y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GT | # 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_6/inductor_cache/oo/coocrqpipgemuyc3lkhsb7xfpsz2kbgvv5cqmrfukvj62n7puahy.py
# Topologically Sorted Source Nodes: [gt], Original ATen: [aten.gt]
# Source node to ATen node mapping:
# gt => gt
# Graph fragment:
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
triton_poi_fused_gt_0 = async_compile.triton('triton_poi_fused_gt_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_gt_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_gt_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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.bool)
# Topologically Sorted Source Nodes: [gt], Original ATen: [aten.gt]
stream0 = get_raw_stream(0)
triton_poi_fused_gt_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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
class GT(torch.nn.Module):
def __init__(self):
super(GT, self).__init__()
def forward(self, x, y):
return x > y
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
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_gt_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 > tmp1
tl.store(out_ptr0 + x0, tmp2, 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.bool)
get_raw_stream(0)
triton_poi_fused_gt_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class GTNew(torch.nn.Module):
def __init__(self):
super(GTNew, 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]
| bunderhi/torch2trt | GT | false | 1,594 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return x > y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RAddFloat | # 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_6/inductor_cache/vq/cvq3k2pudyocnklorpznpn2x4uwcd75kitqtmbswq44pxwjzv5lf.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 = (%arg0_1, 1.0), 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': [], '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_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
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: [add], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_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
class RAddFloat(torch.nn.Module):
def __init__(self):
super(RAddFloat, self).__init__()
def forward(self, x):
return 1.0 + x
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
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 = 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)
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_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RAddFloatNew(torch.nn.Module):
def __init__(self):
super(RAddFloatNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RAddFloat | false | 1,595 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 1.0 + x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ISub | # 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_6/inductor_cache/tk/ctkgmfmzjuitsqnampr3rgu2evmz4dngqwytnh7daxge4todv4py.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# x => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %copy_ : [num_users=1] = call_function[target=torch.ops.aten.copy_.default](args = (%arg0_1, %sub), kwargs = {})
triton_poi_fused_sub_0 = async_compile.triton('triton_poi_fused_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_sub_0', 'mutated_arg_names': ['in_ptr0', 'out_ptr1'], '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_sub_0(in_ptr0, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tmp0 - tmp1
tl.store(out_ptr1 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_sub_0.run(arg0_1, arg1_1, arg0_1, 256, grid=grid(256), stream=stream0)
del arg1_1
return (arg0_1, )
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
class ISub(torch.nn.Module):
def __init__(self):
super(ISub, self).__init__()
def forward(self, x, y):
x -= y
return x
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
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_sub_0(in_ptr0, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 - tmp1
tl.store(out_ptr1 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_sub_0[grid(256)](arg0_1, arg1_1, arg0_1, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg1_1
return arg0_1,
class ISubNew(torch.nn.Module):
def __init__(self):
super(ISubNew, 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]
| bunderhi/torch2trt | ISub | false | 1,596 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
x -= y
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
IAdd | # 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_6/inductor_cache/ex/cexeaqxslhznkuzencvh3uey3x5fj6oev5hgvga57qcwzylabzz5.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %copy_ : [num_users=1] = call_function[target=torch.ops.aten.copy_.default](args = (%arg0_1, %add), 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_ptr0', 'out_ptr1'], '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, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr1 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(arg0_1, arg1_1, arg0_1, 256, grid=grid(256), stream=stream0)
del arg1_1
return (arg0_1, )
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
class IAdd(torch.nn.Module):
def __init__(self):
super(IAdd, self).__init__()
def forward(self, x, y):
x += y
return x
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
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_add_0(in_ptr0, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr1 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](arg0_1, arg1_1, arg0_1, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg1_1
return arg0_1,
class IAddNew(torch.nn.Module):
def __init__(self):
super(IAddNew, 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]
| bunderhi/torch2trt | IAdd | false | 1,597 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
x += y
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MinElementwise | # 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_6/inductor_cache/ja/cjai25ymezyuledylgfusmsbmzcntnnlj2ymnqb47pogtjos3bke.py
# Topologically Sorted Source Nodes: [min_1], Original ATen: [aten.minimum]
# Source node to ATen node mapping:
# min_1 => minimum
# Graph fragment:
# %minimum : [num_users=1] = call_function[target=torch.ops.aten.minimum.default](args = (%arg1_1, %arg0_1), kwargs = {})
triton_poi_fused_minimum_0 = async_compile.triton('triton_poi_fused_minimum_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_minimum_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_minimum_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = triton_helpers.minimum(tmp0, tmp1)
tl.store(out_ptr0 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [min_1], Original ATen: [aten.minimum]
stream0 = get_raw_stream(0)
triton_poi_fused_minimum_0.run(arg1_1, arg0_1, buf0, 256, grid=grid(256), 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, 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
class MinElementwise(torch.nn.Module):
def forward(self, x, y):
return torch.min(x, y)
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
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_minimum_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = triton_helpers.minimum(tmp0, tmp1)
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_minimum_0[grid(256)](arg1_1, arg0_1, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class MinElementwiseNew(torch.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]
| bunderhi/torch2trt | MinElementwise | false | 1,598 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x, y):
return torch.min(x, y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
IMul | # 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_6/inductor_cache/p7/cp7q4ygkticqaby6s2m6wxiksnded4xo7kzlzdrkn2gkm4uhmfll.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# x => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %copy_ : [num_users=1] = call_function[target=torch.ops.aten.copy_.default](args = (%arg0_1, %mul), 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: '*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': ['in_ptr0', 'out_ptr1'], '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_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)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tmp0 * tmp1
tl.store(out_ptr1 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(arg0_1, arg1_1, arg0_1, 256, grid=grid(256), stream=stream0)
del arg1_1
return (arg0_1, )
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
class IMul(torch.nn.Module):
def __init__(self):
super(IMul, self).__init__()
def forward(self, x, y):
x *= y
return x
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
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, in_ptr1, 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)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 * tmp1
tl.store(out_ptr1 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(256)](arg0_1, arg1_1, arg0_1, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg1_1
return arg0_1,
class IMulNew(torch.nn.Module):
def __init__(self):
super(IMulNew, 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]
| bunderhi/torch2trt | IMul | false | 1,599 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
x *= y
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
LT | # 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_6/inductor_cache/4h/c4hafdeeg2jl3rk254ws2m7bzvz6wbkt7uvyh3ocmzpvol2ipjko.py
# Topologically Sorted Source Nodes: [lt], Original ATen: [aten.lt]
# Source node to ATen node mapping:
# lt => lt
# Graph fragment:
# %lt : [num_users=1] = call_function[target=torch.ops.aten.lt.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
triton_poi_fused_lt_0 = async_compile.triton('triton_poi_fused_lt_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_lt_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_lt_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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.bool)
# Topologically Sorted Source Nodes: [lt], Original ATen: [aten.lt]
stream0 = get_raw_stream(0)
triton_poi_fused_lt_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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
class LT(torch.nn.Module):
def __init__(self):
super(LT, self).__init__()
def forward(self, x, y):
return x < y
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
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_lt_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 < tmp1
tl.store(out_ptr0 + x0, tmp2, 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.bool)
get_raw_stream(0)
triton_poi_fused_lt_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class LTNew(torch.nn.Module):
def __init__(self):
super(LTNew, 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]
| bunderhi/torch2trt | LT | false | 1,600 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return x < y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Pow | # 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_6/inductor_cache/rw/crw3qydkgjmifxnk2lsi7cnidtaocksbhyv6htmtnrfv2zqr2lbi.py
# Topologically Sorted Source Nodes: [pow_1], Original ATen: [aten.pow]
# Source node to ATen node mapping:
# pow_1 => pow_1
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
triton_poi_fused_pow_0 = async_compile.triton('triton_poi_fused_pow_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_pow_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_pow_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = libdevice.pow(tmp0, tmp1)
tl.store(out_ptr0 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [pow_1], Original ATen: [aten.pow]
stream0 = get_raw_stream(0)
triton_poi_fused_pow_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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
class Pow(torch.nn.Module):
def __init__(self):
super(Pow, self).__init__()
def forward(self, x, y):
return x ** y
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.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_pow_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = libdevice.pow(tmp0, tmp1)
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_pow_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK
=128, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class PowNew(torch.nn.Module):
def __init__(self):
super(PowNew, 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]
| bunderhi/torch2trt | Pow | false | 1,601 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return x ** y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RDivFloat | # 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_6/inductor_cache/co/ccog5rgtzipydkyy2hkzhhuz3xrmtwbjws5ymi42pi4qrtmk6eqe.py
# Topologically Sorted Source Nodes: [truediv], Original ATen: [aten.reciprocal, aten.mul]
# Source node to ATen node mapping:
# truediv => mul, reciprocal
# Graph fragment:
# %reciprocal : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%arg0_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%reciprocal, 100.0), kwargs = {})
triton_poi_fused_mul_reciprocal_0 = async_compile.triton('triton_poi_fused_mul_reciprocal_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_reciprocal_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_reciprocal_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], 1, tl.int32)
tmp2 = tmp1 / tmp0
tmp3 = 100.0
tmp4 = 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: [truediv], Original ATen: [aten.reciprocal, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_reciprocal_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
class RDivFloat(torch.nn.Module):
def __init__(self):
super(RDivFloat, self).__init__()
def forward(self, x):
return 100.0 / x
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
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_reciprocal_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], 1, tl.int32)
tmp2 = tmp1 / tmp0
tmp3 = 100.0
tmp4 = 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_mul_reciprocal_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RDivFloatNew(torch.nn.Module):
def __init__(self):
super(RDivFloatNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RDivFloat | false | 1,602 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 100.0 / x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TensorClamp | # 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_6/inductor_cache/kk/ckkvw2eootfbcqybp52nepeyub7pn2go6gqcxynwbaothewy66ku.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, -0.1), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 0.1), 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 = -0.1
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp3 = 0.1
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
class TensorClamp(torch.nn.Module):
def forward(self, x):
return x.clamp(-0.1, 0.1)
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
from torch._inductor.runtime import triton_helpers
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 = -0.1
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp3 = 0.1
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 TensorClampNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | TensorClamp | false | 1,603 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x):
return x.clamp(-0.1, 0.1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RSubInt | # 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_6/inductor_cache/27/c27mlt7y7rt3puwrn7flijok6xmn62zm5ireirvcu7xgcaebfhih.py
# Topologically Sorted Source Nodes: [sub], Original ATen: [aten.rsub]
# Source node to ATen node mapping:
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg0_1), kwargs = {})
triton_poi_fused_rsub_0 = async_compile.triton('triton_poi_fused_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_rsub_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_rsub_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 = tmp1 - tmp0
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: [sub], Original ATen: [aten.rsub]
stream0 = get_raw_stream(0)
triton_poi_fused_rsub_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
class RSubInt(torch.nn.Module):
def __init__(self):
super(RSubInt, self).__init__()
def forward(self, x):
return 1 - x
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
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_rsub_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 = tmp1 - tmp0
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_rsub_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RSubIntNew(torch.nn.Module):
def __init__(self):
super(RSubIntNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RSubInt | false | 1,604 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 1 - x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RMulInt | # 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_6/inductor_cache/kh/ckhkvtw63aixa5pww5plsf4f5xegmzknbm6a3stzheag7xjm7n63.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, 10), 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 = 10.0
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
class RMulInt(torch.nn.Module):
def __init__(self):
super(RMulInt, self).__init__()
def forward(self, x):
return 10 * x
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
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 = 10.0
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 RMulIntNew(torch.nn.Module):
def __init__(self):
super(RMulIntNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RMulInt | false | 1,605 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 10 * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RMulFloat | # 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_6/inductor_cache/kh/ckhkvtw63aixa5pww5plsf4f5xegmzknbm6a3stzheag7xjm7n63.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, 10.0), 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 = 10.0
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
class RMulFloat(torch.nn.Module):
def __init__(self):
super(RMulFloat, self).__init__()
def forward(self, x):
return 10.0 * x
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
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 = 10.0
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=128,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RMulFloatNew(torch.nn.Module):
def __init__(self):
super(RMulFloatNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RMulFloat | false | 1,606 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 10.0 * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TensorClampOptionMax | # 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_6/inductor_cache/ob/cobbkvkoqvposl3b23hawoxlcj73gz43pznqo2momxipyniy7uqm.py
# Topologically Sorted Source Nodes: [clamp], Original ATen: [aten.clamp]
# Source node to ATen node mapping:
# clamp => clamp_max
# Graph fragment:
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%arg0_1, 0.1), 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 = 0.1
tmp2 = triton_helpers.minimum(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: [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
class TensorClampOptionMax(torch.nn.Module):
def forward(self, x):
return x.clamp(max=0.1)
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
from torch._inductor.runtime import triton_helpers
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 = 0.1
tmp2 = triton_helpers.minimum(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_clamp_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class TensorClampOptionMaxNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | TensorClampOptionMax | false | 1,607 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x):
return x.clamp(max=0.1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RDivInt | # 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_6/inductor_cache/co/ccog5rgtzipydkyy2hkzhhuz3xrmtwbjws5ymi42pi4qrtmk6eqe.py
# Topologically Sorted Source Nodes: [truediv], Original ATen: [aten.reciprocal, aten.mul]
# Source node to ATen node mapping:
# truediv => mul, reciprocal
# Graph fragment:
# %reciprocal : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%arg0_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%reciprocal, 100), kwargs = {})
triton_poi_fused_mul_reciprocal_0 = async_compile.triton('triton_poi_fused_mul_reciprocal_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_reciprocal_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_reciprocal_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], 1, tl.int32)
tmp2 = tmp1 / tmp0
tmp3 = 100.0
tmp4 = 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: [truediv], Original ATen: [aten.reciprocal, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_reciprocal_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
class RDivInt(torch.nn.Module):
def __init__(self):
super(RDivInt, self).__init__()
def forward(self, x):
return 100 / x
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
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_reciprocal_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], 1, tl.int32)
tmp2 = tmp1 / tmp0
tmp3 = 100.0
tmp4 = 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_mul_reciprocal_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RDivIntNew(torch.nn.Module):
def __init__(self):
super(RDivIntNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RDivInt | false | 1,608 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 100 / x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TorchSub | # 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_6/inductor_cache/2x/c2xnwzkmtsschsc4kip66bb4tudxm35hzl4lx5aabvhr6szhgwdb.py
# Topologically Sorted Source Nodes: [sub], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %arg0_1), kwargs = {})
triton_poi_fused_sub_0 = async_compile.triton('triton_poi_fused_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_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_sub_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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: [sub], Original ATen: [aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_sub_0.run(arg1_1, arg0_1, buf0, 256, grid=grid(256), 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, 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
class TorchSub(torch.nn.Module):
def __init__(self):
super(TorchSub, self).__init__()
def forward(self, x, y):
return torch.sub(x, y)
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
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_sub_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_sub_0[grid(256)](arg1_1, arg0_1, buf0, 256, XBLOCK
=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class TorchSubNew(torch.nn.Module):
def __init__(self):
super(TorchSubNew, 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]
| bunderhi/torch2trt | TorchSub | false | 1,609 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return torch.sub(x, y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Sub | # 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_6/inductor_cache/2x/c2xnwzkmtsschsc4kip66bb4tudxm35hzl4lx5aabvhr6szhgwdb.py
# Topologically Sorted Source Nodes: [sub], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
triton_poi_fused_sub_0 = async_compile.triton('triton_poi_fused_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_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_sub_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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: [sub], Original ATen: [aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_sub_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), 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, 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
class Sub(torch.nn.Module):
def __init__(self):
super(Sub, self).__init__()
def forward(self, x, y):
return x - y
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
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_sub_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 - tmp1
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_sub_0[grid(256)](arg0_1, arg1_1, buf0, 256, XBLOCK
=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class SubNew(torch.nn.Module):
def __init__(self):
super(SubNew, 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]
| bunderhi/torch2trt | Sub | false | 1,610 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return x - y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TorchClampMin | # 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_6/inductor_cache/gd/cgd4op35ni7syid77iv67fyz4srypibp225bds5otckh3ocqyvrf.py
# Topologically Sorted Source Nodes: [clamp_min], Original ATen: [aten.clamp_min]
# Source node to ATen node mapping:
# clamp_min => clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%arg0_1, -0.1), kwargs = {})
triton_poi_fused_clamp_min_0 = async_compile.triton('triton_poi_fused_clamp_min_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_min_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_min_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 = -0.1
tmp2 = triton_helpers.maximum(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: [clamp_min], Original ATen: [aten.clamp_min]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_min_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
class TorchClampMin(torch.nn.Module):
def forward(self, x):
return torch.clamp_min(x, -0.1)
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
from torch._inductor.runtime import triton_helpers
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_min_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 = -0.1
tmp2 = triton_helpers.maximum(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_clamp_min_0[grid(256)](arg0_1, buf0, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class TorchClampMinNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | TorchClampMin | false | 1,611 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x):
return torch.clamp_min(x, -0.1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TorchDiv | # 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_6/inductor_cache/i3/ci3wf6sq5envi5qqy36dg5xyilb5fmbazlijbvgfe3ony6kt56ir.py
# Topologically Sorted Source Nodes: [div], Original ATen: [aten.div]
# Source node to ATen node mapping:
# div => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg1_1, %arg0_1), kwargs = {})
triton_poi_fused_div_0 = async_compile.triton('triton_poi_fused_div_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_div_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_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (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, 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: [div], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_0.run(arg1_1, arg0_1, buf0, 256, grid=grid(256), 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, 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
class TorchDiv(torch.nn.Module):
def __init__(self):
super(TorchDiv, self).__init__()
def forward(self, x, y):
return torch.div(x, y)
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
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_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 / tmp1
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_div_0[grid(256)](arg1_1, arg0_1, buf0, 256, XBLOCK
=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class TorchDivNew(torch.nn.Module):
def __init__(self):
super(TorchDivNew, 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]
| bunderhi/torch2trt | TorchDiv | false | 1,612 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return torch.div(x, y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RSubFloat | # 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_6/inductor_cache/27/c27mlt7y7rt3puwrn7flijok6xmn62zm5ireirvcu7xgcaebfhih.py
# Topologically Sorted Source Nodes: [sub], Original ATen: [aten.rsub]
# Source node to ATen node mapping:
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %arg0_1), kwargs = {})
triton_poi_fused_rsub_0 = async_compile.triton('triton_poi_fused_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_rsub_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_rsub_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 = tmp1 - tmp0
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: [sub], Original ATen: [aten.rsub]
stream0 = get_raw_stream(0)
triton_poi_fused_rsub_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
class RSubFloat(torch.nn.Module):
def __init__(self):
super(RSubFloat, self).__init__()
def forward(self, x):
return 1.0 - x
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
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_rsub_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 = tmp1 - tmp0
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_rsub_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RSubFloatNew(torch.nn.Module):
def __init__(self):
super(RSubFloatNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RSubFloat | false | 1,613 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 1.0 - x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TensorClampMax | # 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_6/inductor_cache/tg/ctggfgjpksbppsxvi7yflqjzicnuldsi7ay5mnalfyyjmn4pnrxt.py
# Topologically Sorted Source Nodes: [clamp_max], Original ATen: [aten.clamp_max]
# Source node to ATen node mapping:
# clamp_max => clamp_max
# Graph fragment:
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%arg0_1, 0.1), kwargs = {})
triton_poi_fused_clamp_max_0 = async_compile.triton('triton_poi_fused_clamp_max_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_max_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_max_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 = 0.1
tmp2 = triton_helpers.minimum(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: [clamp_max], Original ATen: [aten.clamp_max]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_max_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
class TensorClampMax(torch.nn.Module):
def forward(self, x):
return x.clamp_max(0.1)
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
from torch._inductor.runtime import triton_helpers
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_max_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 = 0.1
tmp2 = triton_helpers.minimum(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_clamp_max_0[grid(256)](arg0_1, buf0, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class TensorClampMaxNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | TensorClampMax | false | 1,614 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def forward(self, x):
return x.clamp_max(0.1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
RpowFloat | # 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_6/inductor_cache/lc/clcfvdilobek33revibf6ihdojxbgxqng6nkbsrggerexhylc52x.py
# Topologically Sorted Source Nodes: [pow_1], Original ATen: [aten.pow]
# Source node to ATen node mapping:
# pow_1 => pow_1
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Scalar](args = (2.0, %arg0_1), kwargs = {})
triton_poi_fused_pow_0 = async_compile.triton('triton_poi_fused_pow_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_pow_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_pow_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 = libdevice.exp2(tmp0)
tl.store(out_ptr0 + (x0), tmp1, 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: [pow_1], Original ATen: [aten.pow]
stream0 = get_raw_stream(0)
triton_poi_fused_pow_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
class RpowFloat(torch.nn.Module):
def __init__(self):
super(RpowFloat, self).__init__()
def forward(self, x):
return 2.0 ** x
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.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_pow_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 = libdevice.exp2(tmp0)
tl.store(out_ptr0 + x0, tmp1, 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_pow_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class RpowFloatNew(torch.nn.Module):
def __init__(self):
super(RpowFloatNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| bunderhi/torch2trt | RpowFloat | false | 1,615 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return 2.0 ** x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TorchPow | # 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_6/inductor_cache/rw/crw3qydkgjmifxnk2lsi7cnidtaocksbhyv6htmtnrfv2zqr2lbi.py
# Topologically Sorted Source Nodes: [pow_1], Original ATen: [aten.pow]
# Source node to ATen node mapping:
# pow_1 => pow_1
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Tensor](args = (%arg1_1, %arg0_1), kwargs = {})
triton_poi_fused_pow_0 = async_compile.triton('triton_poi_fused_pow_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_pow_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_pow_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
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = libdevice.pow(tmp0, tmp1)
tl.store(out_ptr0 + (x0), tmp2, 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)
# Topologically Sorted Source Nodes: [pow_1], Original ATen: [aten.pow]
stream0 = get_raw_stream(0)
triton_poi_fused_pow_0.run(arg1_1, arg0_1, buf0, 256, grid=grid(256), 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, 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
class TorchPow(torch.nn.Module):
def __init__(self):
super(TorchPow, self).__init__()
def forward(self, x, y):
return torch.pow(x, y)
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.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_pow_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
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = libdevice.pow(tmp0, tmp1)
tl.store(out_ptr0 + x0, tmp2, 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)
get_raw_stream(0)
triton_poi_fused_pow_0[grid(256)](arg1_1, arg0_1, buf0, 256, XBLOCK
=128, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
return buf0,
class TorchPowNew(torch.nn.Module):
def __init__(self):
super(TorchPowNew, 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]
| bunderhi/torch2trt | TorchPow | false | 1,616 | [
"MIT"
] | 0 | fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | https://github.com/bunderhi/torch2trt/tree/fa5e31e742a0f0c9a9ee38909a6fa56bb07ba96d | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, y):
return torch.pow(x, y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TransformerEncoderLayerWithConv1d | # 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_6/inductor_cache/yr/cyrvpoicx542o7l5zk6rjoea35aunxmpz2sijrdm6ao2qoj4bvgj.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul, aten.transpose]
# Source node to ATen node mapping:
# multi_head_attention_forward => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%view_6, 1.0), kwargs = {})
# %permute_29 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%view_9, [0, 2, 1]), kwargs = {})
triton_poi_fused_mul_transpose_0 = async_compile.triton('triton_poi_fused_mul_transpose_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: '*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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_transpose_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_transpose_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, 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 + (12*y1) + (48*x2)), 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)
tl.store(out_ptr1 + (y3 + (16*x2)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/of/cofu6zktaqykiqb54lgneepflg47mm7m74y2g7tiq4m3dv42pnvl.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul, aten.transpose]
# Source node to ATen node mapping:
# multi_head_attention_forward => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Scalar](args = (%permute_5, 1.0), kwargs = {})
# %permute_30 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%view_10, [0, 2, 1]), kwargs = {})
triton_poi_fused_mul_transpose_1 = async_compile.triton('triton_poi_fused_mul_transpose_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: '*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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_transpose_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_transpose_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, 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 + (4 + y0 + (12*y1) + (48*x2)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4 + 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)
tl.store(out_ptr1 + (y3 + (16*x2)), tmp4, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/rg/crg7nqye2hlyq5u53fojts2unoebbb6m3p6j53spexhzok6ontmj.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._safe_softmax]
# Source node to ATen node mapping:
# multi_head_attention_forward => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_11, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_11, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__safe_softmax_2 = async_compile.triton('triton_poi_fused__safe_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__safe_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__safe_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 = 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_6/inductor_cache/uy/cuyqionbrqc3htd4pbnwfftu6m5ijmeoinj4crusviajhktak5nn.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._safe_softmax]
# Source node to ATen node mapping:
# multi_head_attention_forward => any_1, div, eq, full_default, logical_not, logical_not_1, sum_1, where
# 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 = {})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%view_11, -inf), kwargs = {})
# %logical_not : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%eq,), kwargs = {})
# %any_1 : [num_users=1] = call_function[target=torch.ops.aten.any.dim](args = (%logical_not, -1, True), kwargs = {})
# %logical_not_1 : [num_users=1] = call_function[target=torch.ops.aten.logical_not.default](args = (%any_1,), 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 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%logical_not_1, %full_default, %div), kwargs = {})
triton_poi_fused__safe_softmax_3 = async_compile.triton('triton_poi_fused__safe_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: '*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__safe_softmax_3', '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__safe_softmax_3(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_6/inductor_cache/ry/crymxush4rzdmsz62q7wp64whddb4k672pfna3mrkhtvuk3xdnt6.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# multi_head_attention_forward => clone
# Graph fragment:
# %clone : [num_users=3] = call_function[target=torch.ops.aten.clone.default](args = (%squeeze,), 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: '*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_4', '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_4(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 16
x2 = (xindex // 64)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (12*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/me/cmesjmsjhrj4ceejs7ktpkfmid74uo3wl5rm3dyvhuegj6bbj75o.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm, aten.transpose]
# Source node to ATen node mapping:
# multi_head_attention_forward => bmm_1
# Graph fragment:
# %bmm_1 : [num_users=1] = call_function[target=torch.ops.aten.bmm.default](args = (%view_12, %view_13), kwargs = {})
# %permute_28 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%view_13, [0, 2, 1]), kwargs = {})
triton_poi_fused_bmm_transpose_5 = async_compile.triton('triton_poi_fused_bmm_transpose_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_bmm_transpose_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_bmm_transpose_5(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 % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (128 + x0 + (4*((x0 % 4) // 4)) + (16*x1)), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
tl.store(out_ptr1 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/q3/cq3oorhy57xbcyalz2y6g76c5pd6yqgdh2qlqjwbjqxf4bqyazbf.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# multi_head_attention_forward => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_6,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_6 = async_compile.triton('triton_poi_fused_clone_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, 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_6', '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_6(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_6/inductor_cache/pw/cpw6jeu3vr4erot3bahfr3gqwa2vqailcc5xbtuiztd5l6w2zwjm.py
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# x => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_16), 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_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=[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_7', '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_7(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_6/inductor_cache/hy/chy77uudjf33jqcobrq24clniixoyn2vucadggy6c6n33mq4favb.py
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# x => add_1, add_2, mul_2, mul_3, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_16), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 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_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %primals_6), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_3, %primals_7), kwargs = {})
triton_poi_fused_add_native_layer_norm_8 = async_compile.triton('triton_poi_fused_add_native_layer_norm_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=[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_8', '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_8(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_6/inductor_cache/5s/c5sz7jq4tmcun7rcsmmfeqrom6cicyxjx2pclothphxo54ygxlxu.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_18,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_9 = async_compile.triton('triton_poi_fused_relu_threshold_backward_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: '*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_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_relu_threshold_backward_9(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')
# kernel path: runs/run_shard_6/inductor_cache/cn/ccn3ztml5qupb2u6trhxfni5siq2chr5f7il63gpx775u3hi6iz5.py
# Topologically Sorted Source Nodes: [add_1], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add_1 => add_3
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %view_20), kwargs = {})
triton_poi_fused_add_10 = async_compile.triton('triton_poi_fused_add_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=[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_10', '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_10(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_6/inductor_cache/s5/cs5qfpz5dtrtoopym4yheekk2kuetbhcvypaqjhns7stzzfhkjha.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => 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_2, 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_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=[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_11', '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_11(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_6/inductor_cache/dr/cdrqch7dkzd6wgbo46meh7vw2wmmm5avruav7pzwc4s4i4zleekn.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_4, add_5, mul_4, mul_5, 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_2, 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_3), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, %rsqrt_1), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_4, %primals_12), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_5, %primals_13), kwargs = {})
triton_poi_fused_native_layer_norm_12 = async_compile.triton('triton_poi_fused_native_layer_norm_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.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_12', '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_12(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')
# kernel path: runs/run_shard_6/inductor_cache/2w/c2w6xjm2yhfcxhyb5y5wobuyg3db26y62couv4w75feaxismungq.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_10, %primals_14, %primals_15, [1], [1], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_13 = async_compile.triton('triton_poi_fused_convolution_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.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_13', '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_13(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
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/h6/ch6c32hfwunyggngpke5ontevy2xjxnd7kqnam5djemzxd2q272i.py
# Topologically Sorted Source Nodes: [conv1d, output], Original ATen: [aten.convolution, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# conv1d => convolution
# output => relu_1
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%permute_10, %primals_14, %primals_15, [1], [1], [1], False, [0], 1), kwargs = {})
# %relu_1 : [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_1, 0), kwargs = {})
triton_poi_fused_convolution_relu_threshold_backward_14 = async_compile.triton('triton_poi_fused_convolution_relu_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.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_14', '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_threshold_backward_14(in_out_ptr0, 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
x3 = xindex
x1 = (xindex // 3) % 4
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)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x3), tmp4, xmask)
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, 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, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (12, ), (1, ))
assert_size_stride(primals_3, (12, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (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, 4), (4, 1))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, ), (1, ))
assert_size_stride(primals_13, (4, ), (1, ))
assert_size_stride(primals_14, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_15, (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_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 12), (1, 4), 0), out=buf0)
del primals_3
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
buf27 = empty_strided_cuda((16, 1, 4), (1, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul, aten.transpose]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_transpose_0.run(buf0, primals_2, buf1, buf27, 16, 4, grid=grid(16, 4), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 1, 4), (16, 4, 4, 1), torch.float32)
buf28 = empty_strided_cuda((16, 4, 1), (1, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul, aten.transpose]
triton_poi_fused_mul_transpose_1.run(buf0, primals_2, buf2, buf28, 16, 4, grid=grid(16, 4), stream=stream0)
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf1, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf2, (16, 1, 4), (4, 0, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._safe_softmax]
triton_poi_fused__safe_softmax_2.run(buf3, buf4, 256, grid=grid(256), stream=stream0)
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._safe_softmax]
triton_poi_fused__safe_softmax_3.run(buf3, buf4, buf5, 256, grid=grid(256), stream=stream0)
del buf3
del buf4
buf6 = empty_strided_cuda((3, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf0, primals_2, buf6, 192, grid=grid(192), stream=stream0)
del buf0
del primals_2
buf7 = reinterpret_tensor(buf2, (16, 4, 1), (1, 16, 64), 0); del buf2 # reuse
buf26 = reinterpret_tensor(buf1, (16, 1, 4), (1, 1, 16), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm, aten.transpose]
triton_poi_fused_bmm_transpose_5.run(buf6, buf7, buf26, 64, grid=grid(64), stream=stream0)
del buf6
buf8 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1), 0), buf7, out=buf8)
buf9 = reinterpret_tensor(buf7, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
triton_poi_fused_clone_6.run(buf8, buf9, 4, 16, grid=grid(4, 16), stream=stream0)
buf10 = reinterpret_tensor(buf8, (16, 4), (4, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf9, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf10)
del primals_5
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_7.run(primals_1, buf10, buf11, buf12, 16, grid=grid(16), stream=stream0)
buf13 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_8.run(primals_1, buf10, buf11, buf12, primals_6, primals_7, buf13, 64, grid=grid(64), stream=stream0)
del primals_7
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf14)
buf15 = reinterpret_tensor(buf14, (4, 4, 4), (16, 4, 1), 0); del buf14 # reuse
buf25 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_9.run(buf15, primals_9, buf25, 64, grid=grid(64), stream=stream0)
del primals_9
buf16 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf15, (16, 4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), out=buf16)
buf17 = reinterpret_tensor(buf16, (4, 4, 4), (16, 4, 1), 0); del buf16 # reuse
# Topologically Sorted Source Nodes: [add_1], Original ATen: [aten.add]
triton_poi_fused_add_10.run(buf17, buf13, primals_11, 64, grid=grid(64), stream=stream0)
del primals_11
buf18 = buf12; del buf12 # reuse
buf19 = buf11; del buf11 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_11.run(buf17, buf18, buf19, 16, grid=grid(16), stream=stream0)
buf20 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_12.run(buf17, buf18, buf19, primals_12, primals_13, buf20, 64, grid=grid(64), stream=stream0)
del buf18
del buf19
del primals_13
buf21 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
triton_poi_fused_convolution_13.run(buf20, buf21, 16, 4, grid=grid(16, 4), stream=stream0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf22 = extern_kernels.convolution(buf21, primals_14, stride=(1,), padding=(1,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf22, (4, 4, 3), (12, 3, 1))
del buf21
buf23 = buf22; del buf22 # reuse
buf24 = empty_strided_cuda((4, 4, 3), (12, 3, 1), torch.bool)
# Topologically Sorted Source Nodes: [conv1d, output], Original ATen: [aten.convolution, aten.relu, aten.threshold_backward]
triton_poi_fused_convolution_relu_threshold_backward_14.run(buf23, primals_15, buf24, 48, grid=grid(48), stream=stream0)
del primals_15
return (reinterpret_tensor(buf23, (3, 4, 4), (1, 12, 3), 0), primals_1, primals_6, primals_12, primals_14, buf5, reinterpret_tensor(buf9, (16, 4), (4, 1), 0), buf10, reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(buf15, (16, 4), (4, 1), 0), buf17, reinterpret_tensor(buf20, (4, 4, 4), (4, 1, 16), 0), buf24, primals_10, buf25, primals_8, primals_4, buf26, buf27, buf28, )
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((12, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((12, 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((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, 4), (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)
primals_13 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_15 = 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, 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 TransformerEncoderLayerWithConv1d(nn.Module):
"""
Input and output shape: seqlen x batch_size x dim
"""
def __init__(self, dim_model, nheads, dim_feedforward, dropout,
kernel_size, stride):
super(TransformerEncoderLayerWithConv1d, self).__init__()
self.encoder_layer = nn.TransformerEncoderLayer(dim_model, nheads,
dim_feedforward, dropout)
self.conv1d = nn.Conv1d(dim_model, dim_model, kernel_size, stride=
stride, padding=1)
def forward(self, src, src_mask=None, src_key_padding_mask=None):
output = self.encoder_layer(src, src_mask, src_key_padding_mask)
output = F.relu(self.conv1d(output.permute(1, 2, 0)))
return output.permute(2, 0, 1)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'dim_model': 4, 'nheads': 4, 'dim_feedforward': 4,
'dropout': 0.5, 'kernel_size': 4, 'stride': 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 import triton_helpers
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_mul_transpose_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
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 + 12 * y1 + 48 * x2), 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)
tl.store(out_ptr1 + (y3 + 16 * x2), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused_mul_transpose_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
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 + (4 + y0 + 12 * y1 + 48 * x2), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4 + 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)
tl.store(out_ptr1 + (y3 + 16 * x2), tmp4, xmask & ymask)
@triton.jit
def triton_poi_fused__safe_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 = 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__safe_softmax_3(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_clone_4(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 16
x2 = xindex // 64
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 12 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x3, tmp2, xmask)
@triton.jit
def triton_poi_fused_bmm_transpose_5(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 % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (128 + x0 + 4 * (x0 % 4 // 4) + 16 * x1), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
tl.store(out_ptr1 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_clone_6(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_7(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_8(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_9(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)
@triton.jit
def triton_poi_fused_add_10(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_11(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_12(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)
@triton.jit
def triton_poi_fused_convolution_13(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
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x1), xmask & ymask, eviction_policy
='evict_last')
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_convolution_relu_threshold_backward_14(in_out_ptr0,
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
x3 = xindex
x1 = xindex // 3 % 4
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)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x3, tmp4, xmask)
tl.store(out_ptr0 + x3, tmp6, 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) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (12,), (1,))
assert_size_stride(primals_3, (12, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (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, 4), (4, 1))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4,), (1,))
assert_size_stride(primals_13, (4,), (1,))
assert_size_stride(primals_14, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_15, (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_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 12), (1, 4), 0), out=buf0)
del primals_3
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
buf27 = empty_strided_cuda((16, 1, 4), (1, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_transpose_0[grid(16, 4)](buf0, primals_2, buf1,
buf27, 16, 4, XBLOCK=4, YBLOCK=8, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 1, 4), (16, 4, 4, 1), torch.float32)
buf28 = empty_strided_cuda((16, 4, 1), (1, 16, 1), torch.float32)
triton_poi_fused_mul_transpose_1[grid(16, 4)](buf0, primals_2, buf2,
buf28, 16, 4, XBLOCK=4, YBLOCK=8, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf1, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf2, (16, 1, 4), (4, 0, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__safe_softmax_2[grid(256)](buf3, buf4, 256, XBLOCK
=256, num_warps=4, num_stages=1)
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__safe_softmax_3[grid(256)](buf3, buf4, buf5, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del buf3
del buf4
buf6 = empty_strided_cuda((3, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_4[grid(192)](buf0, primals_2, buf6, 192,
XBLOCK=256, num_warps=4, num_stages=1)
del buf0
del primals_2
buf7 = reinterpret_tensor(buf2, (16, 4, 1), (1, 16, 64), 0)
del buf2
buf26 = reinterpret_tensor(buf1, (16, 1, 4), (1, 1, 16), 0)
del buf1
triton_poi_fused_bmm_transpose_5[grid(64)](buf6, buf7, buf26, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del buf6
buf8 = empty_strided_cuda((16, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1),
0), buf7, out=buf8)
buf9 = reinterpret_tensor(buf7, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf7
triton_poi_fused_clone_6[grid(4, 16)](buf8, buf9, 4, 16, XBLOCK=16,
YBLOCK=4, num_warps=1, num_stages=1)
buf10 = reinterpret_tensor(buf8, (16, 4), (4, 1), 0)
del buf8
extern_kernels.addmm(primals_5, reinterpret_tensor(buf9, (16, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf10)
del primals_5
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_7[grid(16)](primals_1, buf10,
buf11, buf12, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf13 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_8[grid(64)](primals_1, buf10,
buf11, buf12, primals_6, primals_7, buf13, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_7
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf14)
buf15 = reinterpret_tensor(buf14, (4, 4, 4), (16, 4, 1), 0)
del buf14
buf25 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_9[grid(64)](buf15,
primals_9, buf25, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_9
buf16 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf15, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), out=buf16)
buf17 = reinterpret_tensor(buf16, (4, 4, 4), (16, 4, 1), 0)
del buf16
triton_poi_fused_add_10[grid(64)](buf17, buf13, primals_11, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_11
buf18 = buf12
del buf12
buf19 = buf11
del buf11
triton_poi_fused_native_layer_norm_11[grid(16)](buf17, buf18, buf19,
16, XBLOCK=16, num_warps=1, num_stages=1)
buf20 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_12[grid(64)](buf17, buf18, buf19,
primals_12, primals_13, buf20, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf18
del buf19
del primals_13
buf21 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_convolution_13[grid(16, 4)](buf20, buf21, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf22 = extern_kernels.convolution(buf21, primals_14, stride=(1,),
padding=(1,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf22, (4, 4, 3), (12, 3, 1))
del buf21
buf23 = buf22
del buf22
buf24 = empty_strided_cuda((4, 4, 3), (12, 3, 1), torch.bool)
triton_poi_fused_convolution_relu_threshold_backward_14[grid(48)](buf23
, primals_15, buf24, 48, XBLOCK=64, num_warps=1, num_stages=1)
del primals_15
return reinterpret_tensor(buf23, (3, 4, 4), (1, 12, 3), 0
), primals_1, primals_6, primals_12, primals_14, buf5, reinterpret_tensor(
buf9, (16, 4), (4, 1), 0), buf10, reinterpret_tensor(buf13, (16, 4),
(4, 1), 0), reinterpret_tensor(buf15, (16, 4), (4, 1), 0
), buf17, reinterpret_tensor(buf20, (4, 4, 4), (4, 1, 16), 0
), buf24, primals_10, buf25, primals_8, primals_4, buf26, buf27, buf28
class TransformerEncoderLayerWithConv1dNew(nn.Module):
"""
Input and output shape: seqlen x batch_size x dim
"""
def __init__(self, dim_model, nheads, dim_feedforward, dropout,
kernel_size, stride):
super(TransformerEncoderLayerWithConv1dNew, self).__init__()
self.encoder_layer = nn.TransformerEncoderLayer(dim_model, nheads,
dim_feedforward, dropout)
self.conv1d = nn.Conv1d(dim_model, dim_model, kernel_size, stride=
stride, padding=1)
def forward(self, input_0):
primals_3 = self.encoder_layer.self_attn.in_proj_weight
primals_2 = self.encoder_layer.self_attn.in_proj_bias
primals_4 = self.encoder_layer.self_attn.out_proj.weight
primals_5 = self.encoder_layer.self_attn.out_proj.bias
primals_8 = self.encoder_layer.linear1.weight
primals_6 = self.encoder_layer.linear1.bias
primals_10 = self.encoder_layer.linear2.weight
primals_7 = self.encoder_layer.linear2.bias
primals_9 = self.encoder_layer.norm1.weight
primals_11 = self.encoder_layer.norm1.bias
primals_12 = self.encoder_layer.norm2.weight
primals_13 = self.encoder_layer.norm2.bias
primals_1 = self.conv1d.weight
primals_15 = self.conv1d.bias
primals_14 = 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]
| bliunlpr/pykaldi2 | TransformerEncoderLayerWithConv1d | false | 1,617 | [
"MIT"
] | 0 | f6020b5dd9900f97ab69c97442a91196a03dd93b | https://github.com/bliunlpr/pykaldi2/tree/f6020b5dd9900f97ab69c97442a91196a03dd93b | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
"""
Input and output shape: seqlen x batch_size x dim
"""
def __init__(self, dim_model, nheads, dim_feedforward, dropout,
kernel_size, stride):
super().__init__()
self.encoder_layer = nn.TransformerEncoderLayer(dim_model, nheads,
dim_feedforward, dropout)
self.conv1d = nn.Conv1d(dim_model, dim_model, kernel_size, stride=
stride, padding=1)
def forward(self, src, src_mask=None, src_key_padding_mask=None):
output = self.encoder_layer(src, src_mask, src_key_padding_mask)
output = F.relu(self.conv1d(output.permute(1, 2, 0)))
return output.permute(2, 0, 1)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'dim_model': 4, 'nheads': 4, 'dim_feedforward': 4,
'dropout': 0.5, 'kernel_size': 4, 'stride': 1}]
|
EncoderLayer | # 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_6/inductor_cache/fg/cfg742icmosiwp5ugziye26din5ueqx3v7ntptkkpyackudldrxs.py
# Topologically Sorted Source Nodes: [eq], Original ATen: [aten.eq]
# Source node to ATen node mapping:
# eq => eq
# Graph fragment:
# %eq : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%primals_8, 0), kwargs = {})
triton_poi_fused_eq_0 = async_compile.triton('triton_poi_fused_eq_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: '*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_eq_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_eq_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')
# kernel path: runs/run_shard_6/inductor_cache/3l/c3l3va5l626kusnmwtajfm3qn4244ouqt4fwov5x5r27uoqjzpf7.py
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
# Source node to ATen node mapping:
# score => div
# score_1 => full_default, where
# score_2 => amax, exp, sub, sum_1
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_12, 1.0), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], -9.999999960041972e-13), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %div), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where, %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, [1], True), kwargs = {})
triton_poi_fused__softmax_div_masked_fill_1 = async_compile.triton('triton_poi_fused__softmax_div_masked_fill_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: '*i1', 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_div_masked_fill_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__softmax_div_masked_fill_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
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x0 + (64*x1)), xmask)
tmp6 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask).to(tl.int1)
tmp7 = tl.load(in_ptr1 + (16 + x0 + (64*x1)), xmask)
tmp11 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask).to(tl.int1)
tmp12 = tl.load(in_ptr1 + (32 + x0 + (64*x1)), xmask)
tmp16 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask).to(tl.int1)
tmp17 = tl.load(in_ptr1 + (48 + x0 + (64*x1)), xmask)
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp8 = tmp7 * tmp2
tmp9 = tl.where(tmp6, tmp4, tmp8)
tmp10 = triton_helpers.maximum(tmp5, tmp9)
tmp13 = tmp12 * tmp2
tmp14 = tl.where(tmp11, tmp4, tmp13)
tmp15 = triton_helpers.maximum(tmp10, tmp14)
tmp18 = tmp17 * tmp2
tmp19 = tl.where(tmp16, tmp4, tmp18)
tmp20 = triton_helpers.maximum(tmp15, tmp19)
tmp21 = tmp5 - tmp20
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp9 - tmp20
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tmp14 - tmp20
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tmp19 - tmp20
tmp30 = tl_math.exp(tmp29)
tmp31 = tmp28 + tmp30
tl.store(out_ptr0 + (x2), tmp20, xmask)
tl.store(out_ptr1 + (x2), tmp31, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ug/cugbt5li6jtqpc3xeqqp52oc3obojbp6pwyxoifw5kcuiwhlus76.py
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
# Source node to ATen node mapping:
# score => div
# score_1 => full_default, where
# score_2 => amax, div_1, exp, sub
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_12, 1.0), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], -9.999999960041972e-13), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %div), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_div_masked_fill_2 = async_compile.triton('triton_poi_fused__softmax_div_masked_fill_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_div_masked_fill_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_div_masked_fill_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
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask).to(tl.int1)
tmp1 = tl.load(in_out_ptr0 + (x3), xmask)
tmp6 = tl.load(in_ptr1 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp7 = tmp5 - tmp6
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 / tmp9
tl.store(in_out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/7f/c7fwok6q7j5rvjs3ob32s2cth5xjbedhynzb5ozchylog57bhmxv.py
# Topologically Sorted Source Nodes: [add, mean, std], Original ATen: [aten.add, aten.mean, aten.std]
# Source node to ATen node mapping:
# add => add
# mean => mean
# std => var
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_18, %primals_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [-1], True), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%add, [-1]), kwargs = {correction: 1.0, keepdim: True})
triton_poi_fused_add_mean_std_3 = async_compile.triton('triton_poi_fused_add_mean_std_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: '*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_mean_std_3', 'mutated_arg_names': ['in_out_ptr0'], '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_mean_std_3(in_out_ptr0, 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 + (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 = 3.0
tmp29 = tmp27 / tmp28
tl.store(in_out_ptr0 + (x0), tmp29, xmask)
tl.store(out_ptr0 + (x0), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/rh/crh4u2dkpmnbqhpz4bseb3zzx7xjwe5wrpinkjypfpeninbp7je2.py
# Topologically Sorted Source Nodes: [add, mean, std, sub, add_1, out_1, mul, out_2], Original ATen: [aten.add, aten.mean, aten.std, aten.sub, aten.div, aten.mul]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# mean => mean
# mul => mul
# out_1 => div_2
# out_2 => add_2
# std => sqrt
# sub => sub_1
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_18, %primals_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [-1], True), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %mean), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt, 1e-12), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_1, %add_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_11, %div_2), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_12), kwargs = {})
triton_poi_fused_add_div_mean_mul_std_sub_4 = async_compile.triton('triton_poi_fused_add_div_mean_mul_std_sub_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=[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_div_mean_mul_std_sub_4', '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_mean_mul_std_sub_4(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
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_ptr2 + (x2), xmask)
tmp4 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 - tmp4
tmp7 = libdevice.sqrt(tmp6)
tmp8 = 1e-12
tmp9 = tmp7 + tmp8
tmp10 = tmp5 / tmp9
tmp11 = tmp0 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/va/cvayouropyisaprtjrhemadbdvsels72axdjsrgmbayknhu335yc.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_2 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_20,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_5 = async_compile.triton('triton_poi_fused_relu_threshold_backward_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: '*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_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_relu_threshold_backward_5(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')
# kernel path: runs/run_shard_6/inductor_cache/c3/cc3wjckmv52z5p6lagnrhsfwt53rzdfhvzlxkm5tgkwbs3kuzwax.py
# Topologically Sorted Source Nodes: [add_3], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add_3 => add_3
# Graph fragment:
# %add_3 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_22, %add_2), kwargs = {})
triton_poi_fused_add_6 = async_compile.triton('triton_poi_fused_add_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_6', '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_6(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_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_6/inductor_cache/c3/cc3ohqldegtkfvmguqkj6lczx7rbnhyzfqk4rqoloybwozl5ul4n.py
# Topologically Sorted Source Nodes: [mean_2, std_2, sub_1, add_4, out_3, mul_1, out_4], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
# Source node to ATen node mapping:
# add_4 => add_4
# mean_2 => mean_1
# mul_1 => mul_1
# out_3 => div_3
# out_4 => add_5
# std_2 => sqrt_1, var_1
# sub_1 => sub_2
# Graph fragment:
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add_3, [-1], True), kwargs = {})
# %var_1 : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%add_3, [-1]), kwargs = {correction: 1.0, keepdim: True})
# %sqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var_1,), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_3, %mean_1), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt_1, 1e-12), kwargs = {})
# %div_3 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_2, %add_4), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_17, %div_3), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_18), kwargs = {})
triton_poi_fused_add_div_mean_mul_std_sub_7 = async_compile.triton('triton_poi_fused_add_div_mean_mul_std_sub_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_std_sub_7', '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_std_sub_7(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 % 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')
tmp3 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp2 - tmp10
tmp13 = tmp12 * tmp12
tmp14 = tmp3 - tmp10
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp10
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp10
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = 3.0
tmp24 = tmp22 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = 1e-12
tmp27 = tmp25 + tmp26
tmp28 = tmp11 / tmp27
tmp29 = tmp0 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + (x2), tmp31, 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, primals_18 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (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, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4, 4, 4), (64, 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, ))
assert_size_stride(primals_13, (4, 4), (4, 1))
assert_size_stride(primals_14, (4, ), (1, ))
assert_size_stride(primals_15, (4, 4), (4, 1))
assert_size_stride(primals_16, (4, ), (1, ))
assert_size_stride(primals_17, (4, ), (1, ))
assert_size_stride(primals_18, (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: [q], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_1, (16, 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((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_1, (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: [v], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_6
del primals_7
buf3 = 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, 1), 0), reinterpret_tensor(buf1, (16, 1, 4), (4, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [eq], Original ATen: [aten.eq]
stream0 = get_raw_stream(0)
triton_poi_fused_eq_0.run(primals_8, buf4, 256, grid=grid(256), stream=stream0)
del primals_8
buf5 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
triton_poi_fused__softmax_div_masked_fill_1.run(buf4, buf3, buf5, buf6, 64, grid=grid(64), stream=stream0)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
triton_poi_fused__softmax_div_masked_fill_2.run(buf7, buf4, buf5, buf6, 256, grid=grid(256), stream=stream0)
buf8 = reinterpret_tensor(buf6, (16, 4, 1), (4, 1, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [v_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0), out=buf8)
buf9 = reinterpret_tensor(buf5, (16, 4), (4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf9)
del primals_10
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = buf10; del buf10 # reuse
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add, mean, std], Original ATen: [aten.add, aten.mean, aten.std]
triton_poi_fused_add_mean_std_3.run(buf11, buf9, primals_1, buf12, 16, grid=grid(16), stream=stream0)
buf13 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, mean, std, sub, add_1, out_1, mul, out_2], Original ATen: [aten.add, aten.mean, aten.std, aten.sub, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_std_sub_4.run(primals_11, buf9, primals_1, buf12, buf11, primals_12, buf13, 64, grid=grid(64), stream=stream0)
del buf11
del buf12
del primals_12
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf14)
buf15 = reinterpret_tensor(buf14, (4, 4, 4), (16, 4, 1), 0); del buf14 # reuse
buf19 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_5.run(buf15, primals_14, buf19, 64, grid=grid(64), stream=stream0)
del primals_14
buf16 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf15, (16, 4), (4, 1), 0), reinterpret_tensor(primals_15, (4, 4), (1, 4), 0), out=buf16)
buf17 = reinterpret_tensor(buf16, (4, 4, 4), (16, 4, 1), 0); del buf16 # reuse
# Topologically Sorted Source Nodes: [add_3], Original ATen: [aten.add]
triton_poi_fused_add_6.run(buf17, primals_16, buf13, 64, grid=grid(64), stream=stream0)
del primals_16
buf18 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean_2, std_2, sub_1, add_4, out_3, mul_1, out_4], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_std_sub_7.run(primals_17, buf17, primals_18, buf18, 64, grid=grid(64), stream=stream0)
del primals_18
return (buf18, primals_1, primals_11, primals_17, buf4, buf7, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), buf9, reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(buf15, (16, 4), (4, 1), 0), buf17, primals_15, buf19, primals_13, primals_9, reinterpret_tensor(buf2, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf0, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf1, (16, 4, 1), (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), (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)
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, 4), (64, 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)
primals_13 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_18 = 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, primals_14, primals_15, primals_16, primals_17, primals_18])
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
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super(LayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
class PositionwiseFeedForward(nn.Module):
def __init__(self, d_model, hidden, drop_prob=0.1):
super(PositionwiseFeedForward, self).__init__()
self.linear1 = nn.Linear(d_model, hidden)
self.linear2 = nn.Linear(hidden, d_model)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=drop_prob)
def forward(self, x):
x = self.linear1(x)
x = self.relu(x)
x = self.dropout(x)
x = self.linear2(x)
return x
class EncoderLayer(nn.Module):
def __init__(self, d_model, ffn_hidden, n_head, drop_prob):
super(EncoderLayer, self).__init__()
self.attention = MultiHeadAttention(d_model=d_model, n_head=n_head)
self.norm1 = LayerNorm(d_model=d_model)
self.dropout1 = nn.Dropout(p=drop_prob)
self.ffn = PositionwiseFeedForward(d_model=d_model, hidden=
ffn_hidden, drop_prob=drop_prob)
self.norm2 = LayerNorm(d_model=d_model)
self.dropout2 = nn.Dropout(p=drop_prob)
def forward(self, x, s_mask):
_x = x
x = self.attention(q=x, k=x, v=x, mask=s_mask)
x = self.norm1(x + _x)
x = self.dropout1(x)
_x = x
x = self.ffn(x)
x = self.norm2(x + _x)
x = self.dropout2(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_model': 4, 'ffn_hidden': 4, 'n_head': 4, 'drop_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_eq_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)
@triton.jit
def triton_poi_fused__softmax_div_masked_fill_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
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x0 + 64 * x1), xmask)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask).to(tl.int1)
tmp7 = tl.load(in_ptr1 + (16 + x0 + 64 * x1), xmask)
tmp11 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask).to(tl.int1)
tmp12 = tl.load(in_ptr1 + (32 + x0 + 64 * x1), xmask)
tmp16 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask).to(tl.int1)
tmp17 = tl.load(in_ptr1 + (48 + x0 + 64 * x1), xmask)
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp8 = tmp7 * tmp2
tmp9 = tl.where(tmp6, tmp4, tmp8)
tmp10 = triton_helpers.maximum(tmp5, tmp9)
tmp13 = tmp12 * tmp2
tmp14 = tl.where(tmp11, tmp4, tmp13)
tmp15 = triton_helpers.maximum(tmp10, tmp14)
tmp18 = tmp17 * tmp2
tmp19 = tl.where(tmp16, tmp4, tmp18)
tmp20 = triton_helpers.maximum(tmp15, tmp19)
tmp21 = tmp5 - tmp20
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp9 - tmp20
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tmp14 - tmp20
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tmp19 - tmp20
tmp30 = tl_math.exp(tmp29)
tmp31 = tmp28 + tmp30
tl.store(out_ptr0 + x2, tmp20, xmask)
tl.store(out_ptr1 + x2, tmp31, xmask)
@triton.jit
def triton_poi_fused__softmax_div_masked_fill_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
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask).to(tl.int1)
tmp1 = tl.load(in_out_ptr0 + x3, xmask)
tmp6 = tl.load(in_ptr1 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr2 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp7 = tmp5 - tmp6
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 / tmp9
tl.store(in_out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_mean_std_3(in_out_ptr0, 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 + 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 = 3.0
tmp29 = tmp27 / tmp28
tl.store(in_out_ptr0 + x0, tmp29, xmask)
tl.store(out_ptr0 + x0, tmp16, xmask)
@triton.jit
def triton_poi_fused_add_div_mean_mul_std_sub_4(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
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_ptr2 + x2, xmask)
tmp4 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 - tmp4
tmp7 = libdevice.sqrt(tmp6)
tmp8 = 1e-12
tmp9 = tmp7 + tmp8
tmp10 = tmp5 / tmp9
tmp11 = tmp0 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_5(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)
@triton.jit
def triton_poi_fused_add_6(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_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_add_div_mean_mul_std_sub_7(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 % 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')
tmp3 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp2 - tmp10
tmp13 = tmp12 * tmp12
tmp14 = tmp3 - tmp10
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp10
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp10
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = 3.0
tmp24 = tmp22 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = 1e-12
tmp27 = tmp25 + tmp26
tmp28 = tmp11 / tmp27
tmp29 = tmp0 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + x2, tmp31, 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, primals_18
) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (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,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4, 4, 4), (64, 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,))
assert_size_stride(primals_13, (4, 4), (4, 1))
assert_size_stride(primals_14, (4,), (1,))
assert_size_stride(primals_15, (4, 4), (4, 1))
assert_size_stride(primals_16, (4,), (1,))
assert_size_stride(primals_17, (4,), (1,))
assert_size_stride(primals_18, (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_3, reinterpret_tensor(primals_1, (16,
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((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_1, (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_7, reinterpret_tensor(primals_1, (16,
4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf2)
del primals_6
del primals_7
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 1),
0), reinterpret_tensor(buf1, (16, 1, 4), (4, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_eq_0[grid(256)](primals_8, buf4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_8
buf5 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
triton_poi_fused__softmax_div_masked_fill_1[grid(64)](buf4, buf3,
buf5, buf6, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused__softmax_div_masked_fill_2[grid(256)](buf7, buf4,
buf5, buf6, 256, XBLOCK=256, num_warps=4, num_stages=1)
buf8 = reinterpret_tensor(buf6, (16, 4, 1), (4, 1, 1), 0)
del buf6
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0), out=buf8)
buf9 = reinterpret_tensor(buf5, (16, 4), (4, 1), 0)
del buf5
extern_kernels.addmm(primals_10, reinterpret_tensor(buf8, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf9)
del primals_10
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = buf10
del buf10
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_mean_std_3[grid(16)](buf11, buf9, primals_1,
buf12, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf13 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_std_sub_4[grid(64)](primals_11,
buf9, primals_1, buf12, buf11, primals_12, buf13, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf11
del buf12
del primals_12
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf14)
buf15 = reinterpret_tensor(buf14, (4, 4, 4), (16, 4, 1), 0)
del buf14
buf19 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_5[grid(64)](buf15,
primals_14, buf19, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_14
buf16 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf15, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_15, (4, 4), (1, 4), 0), out=buf16)
buf17 = reinterpret_tensor(buf16, (4, 4, 4), (16, 4, 1), 0)
del buf16
triton_poi_fused_add_6[grid(64)](buf17, primals_16, buf13, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_16
buf18 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_std_sub_7[grid(64)](primals_17,
buf17, primals_18, buf18, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_18
return (buf18, primals_1, primals_11, primals_17, buf4, buf7,
reinterpret_tensor(buf8, (16, 4), (4, 1), 0), buf9,
reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(
buf15, (16, 4), (4, 1), 0), buf17, primals_15, buf19, primals_13,
primals_9, reinterpret_tensor(buf2, (16, 1, 4), (4, 1, 1), 0),
reinterpret_tensor(buf0, (16, 1, 4), (4, 1, 1), 0),
reinterpret_tensor(buf1, (16, 4, 1), (4, 1, 4), 0))
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super(LayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
class PositionwiseFeedForward(nn.Module):
def __init__(self, d_model, hidden, drop_prob=0.1):
super(PositionwiseFeedForward, self).__init__()
self.linear1 = nn.Linear(d_model, hidden)
self.linear2 = nn.Linear(hidden, d_model)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=drop_prob)
def forward(self, x):
x = self.linear1(x)
x = self.relu(x)
x = self.dropout(x)
x = self.linear2(x)
return x
class EncoderLayerNew(nn.Module):
def __init__(self, d_model, ffn_hidden, n_head, drop_prob):
super(EncoderLayerNew, self).__init__()
self.attention = MultiHeadAttention(d_model=d_model, n_head=n_head)
self.norm1 = LayerNorm(d_model=d_model)
self.dropout1 = nn.Dropout(p=drop_prob)
self.ffn = PositionwiseFeedForward(d_model=d_model, hidden=
ffn_hidden, drop_prob=drop_prob)
self.norm2 = LayerNorm(d_model=d_model)
self.dropout2 = nn.Dropout(p=drop_prob)
def forward(self, input_0, input_1):
primals_2 = self.attention.w_q.weight
primals_3 = self.attention.w_q.bias
primals_4 = self.attention.w_k.weight
primals_5 = self.attention.w_k.bias
primals_6 = self.attention.w_v.weight
primals_7 = self.attention.w_v.bias
primals_9 = self.attention.w_concat.weight
primals_10 = self.attention.w_concat.bias
primals_11 = self.norm1.gamma
primals_12 = self.norm1.beta
primals_13 = self.ffn.linear1.weight
primals_14 = self.ffn.linear1.bias
primals_15 = self.ffn.linear2.weight
primals_16 = self.ffn.linear2.bias
primals_17 = self.norm2.gamma
primals_18 = self.norm2.beta
primals_1 = 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, primals_13, primals_14,
primals_15, primals_16, primals_17, primals_18])
return output[0]
| bsgiovanini/transformer | EncoderLayer | false | 1,618 | [
"Apache-2.0"
] | 0 | e128fa862f1b3d17d7b92df169a2bbee3f08366f | https://github.com/bsgiovanini/transformer/tree/e128fa862f1b3d17d7b92df169a2bbee3f08366f | import math
import torch
from torch import nn
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super().__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super().__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super().__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
class PositionwiseFeedForward(nn.Module):
def __init__(self, d_model, hidden, drop_prob=0.1):
super().__init__()
self.linear1 = nn.Linear(d_model, hidden)
self.linear2 = nn.Linear(hidden, d_model)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=drop_prob)
def forward(self, x):
x = self.linear1(x)
x = self.relu(x)
x = self.dropout(x)
x = self.linear2(x)
return x
class Model(nn.Module):
def __init__(self, d_model, ffn_hidden, n_head, drop_prob):
super().__init__()
self.attention = MultiHeadAttention(d_model=d_model, n_head=n_head)
self.norm1 = LayerNorm(d_model=d_model)
self.dropout1 = nn.Dropout(p=drop_prob)
self.ffn = PositionwiseFeedForward(d_model=d_model, hidden=
ffn_hidden, drop_prob=drop_prob)
self.norm2 = LayerNorm(d_model=d_model)
self.dropout2 = nn.Dropout(p=drop_prob)
def forward(self, x, s_mask):
_x = x
x = self.attention(q=x, k=x, v=x, mask=s_mask)
x = self.norm1(x + _x)
x = self.dropout1(x)
_x = x
x = self.ffn(x)
x = self.norm2(x + _x)
x = self.dropout2(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4]), to
# ... truncated (>4000 chars) for memory efficiency |
DecoderLayer | # 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_6/inductor_cache/fg/cfg742icmosiwp5ugziye26din5ueqx3v7ntptkkpyackudldrxs.py
# Topologically Sorted Source Nodes: [eq], Original ATen: [aten.eq]
# Source node to ATen node mapping:
# eq => eq
# Graph fragment:
# %eq : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%primals_8, 0), kwargs = {})
triton_poi_fused_eq_0 = async_compile.triton('triton_poi_fused_eq_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: '*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_eq_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_eq_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')
# kernel path: runs/run_shard_6/inductor_cache/3l/c3l3va5l626kusnmwtajfm3qn4244ouqt4fwov5x5r27uoqjzpf7.py
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
# Source node to ATen node mapping:
# score => div
# score_1 => full_default, where
# score_2 => amax, exp, sub, sum_1
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_12, 1.0), kwargs = {})
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([], -9.999999960041972e-13), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %div), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where, %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, [1], True), kwargs = {})
triton_poi_fused__softmax_div_masked_fill_1 = async_compile.triton('triton_poi_fused__softmax_div_masked_fill_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: '*i1', 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_div_masked_fill_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__softmax_div_masked_fill_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
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x0 + (64*x1)), xmask)
tmp6 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask).to(tl.int1)
tmp7 = tl.load(in_ptr1 + (16 + x0 + (64*x1)), xmask)
tmp11 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask).to(tl.int1)
tmp12 = tl.load(in_ptr1 + (32 + x0 + (64*x1)), xmask)
tmp16 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask).to(tl.int1)
tmp17 = tl.load(in_ptr1 + (48 + x0 + (64*x1)), xmask)
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp8 = tmp7 * tmp2
tmp9 = tl.where(tmp6, tmp4, tmp8)
tmp10 = triton_helpers.maximum(tmp5, tmp9)
tmp13 = tmp12 * tmp2
tmp14 = tl.where(tmp11, tmp4, tmp13)
tmp15 = triton_helpers.maximum(tmp10, tmp14)
tmp18 = tmp17 * tmp2
tmp19 = tl.where(tmp16, tmp4, tmp18)
tmp20 = triton_helpers.maximum(tmp15, tmp19)
tmp21 = tmp5 - tmp20
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp9 - tmp20
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tmp14 - tmp20
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tmp19 - tmp20
tmp30 = tl_math.exp(tmp29)
tmp31 = tmp28 + tmp30
tl.store(out_ptr0 + (x2), tmp20, xmask)
tl.store(out_ptr1 + (x2), tmp31, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ug/cugbt5li6jtqpc3xeqqp52oc3obojbp6pwyxoifw5kcuiwhlus76.py
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
# Source node to ATen node mapping:
# score => div
# score_1 => full_default, where
# score_2 => amax, div_1, exp, sub
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_12, 1.0), kwargs = {})
# %full_default : [num_users=2] = call_function[target=torch.ops.aten.full.default](args = ([], -9.999999960041972e-13), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %div), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%where, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%where, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_div_masked_fill_2 = async_compile.triton('triton_poi_fused__softmax_div_masked_fill_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_div_masked_fill_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_div_masked_fill_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
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask).to(tl.int1)
tmp1 = tl.load(in_out_ptr0 + (x3), xmask)
tmp6 = tl.load(in_ptr1 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp7 = tmp5 - tmp6
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 / tmp9
tl.store(in_out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/7f/c7fwok6q7j5rvjs3ob32s2cth5xjbedhynzb5ozchylog57bhmxv.py
# Topologically Sorted Source Nodes: [add, mean, std], Original ATen: [aten.add, aten.mean, aten.std]
# Source node to ATen node mapping:
# add => add
# mean => mean
# std => var
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_18, %primals_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [-1], True), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%add, [-1]), kwargs = {correction: 1.0, keepdim: True})
triton_poi_fused_add_mean_std_3 = async_compile.triton('triton_poi_fused_add_mean_std_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: '*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_mean_std_3', 'mutated_arg_names': ['in_out_ptr0'], '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_mean_std_3(in_out_ptr0, 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 + (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 = 3.0
tmp29 = tmp27 / tmp28
tl.store(in_out_ptr0 + (x0), tmp29, xmask)
tl.store(out_ptr0 + (x0), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/rh/crh4u2dkpmnbqhpz4bseb3zzx7xjwe5wrpinkjypfpeninbp7je2.py
# Topologically Sorted Source Nodes: [add, mean, std, sub, add_1, out_1, mul, out_2], Original ATen: [aten.add, aten.mean, aten.std, aten.sub, aten.div, aten.mul]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# mean => mean
# mul => mul
# out_1 => div_2
# out_2 => add_2
# std => sqrt
# sub => sub_1
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_18, %primals_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [-1], True), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %mean), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt, 1e-12), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_1, %add_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_11, %div_2), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_12), kwargs = {})
triton_poi_fused_add_div_mean_mul_std_sub_4 = async_compile.triton('triton_poi_fused_add_div_mean_mul_std_sub_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=[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_div_mean_mul_std_sub_4', '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_mean_mul_std_sub_4(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
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_ptr2 + (x2), xmask)
tmp4 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 - tmp4
tmp7 = libdevice.sqrt(tmp6)
tmp8 = 1e-12
tmp9 = tmp7 + tmp8
tmp10 = tmp5 / tmp9
tmp11 = tmp0 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/56/c566ex7xddxc2fqpwqlmymdyd23nesbsyghxftm7cy73ebnuo3ke.py
# Topologically Sorted Source Nodes: [add_3], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add_3 => add_3
# Graph fragment:
# %add_3 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_37, %add_2), 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=[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_5', '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_5(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_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_6/inductor_cache/jj/cjjdssmzh5kz3manmdmaotcsrgjtwlvn4pxa2tdwl7lv72rrmy2z.py
# Topologically Sorted Source Nodes: [mean_2, std_2, sub_1, add_4, out_4, mul_1, out_5], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
# Source node to ATen node mapping:
# add_4 => add_4
# mean_2 => mean_1
# mul_1 => mul_1
# out_4 => div_5
# out_5 => add_5
# std_2 => sqrt_1, var_1
# sub_1 => sub_3
# Graph fragment:
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add_3, [-1], True), kwargs = {})
# %var_1 : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%add_3, [-1]), kwargs = {correction: 1.0, keepdim: True})
# %sqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var_1,), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_3, %mean_1), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt_1, 1e-12), kwargs = {})
# %div_5 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_3, %add_4), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_23, %div_5), kwargs = {})
# %add_5 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_24), kwargs = {})
triton_poi_fused_add_div_mean_mul_std_sub_6 = async_compile.triton('triton_poi_fused_add_div_mean_mul_std_sub_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_std_sub_6', '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_std_sub_6(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 % 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')
tmp3 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp2 - tmp10
tmp13 = tmp12 * tmp12
tmp14 = tmp3 - tmp10
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp10
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp10
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = 3.0
tmp24 = tmp22 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = 1e-12
tmp27 = tmp25 + tmp26
tmp28 = tmp11 / tmp27
tmp29 = tmp0 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + (x2), tmp31, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/u4/cu4mvhweewrefdurxuza5qfbqlwomkc67kmxkkaurh6luaf2e2fz.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_3 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_39,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_7 = async_compile.triton('triton_poi_fused_relu_threshold_backward_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: '*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_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_relu_threshold_backward_7(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, 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 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (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, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4, 4, 4), (64, 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, ))
assert_size_stride(primals_13, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_14, (4, 4), (4, 1))
assert_size_stride(primals_15, (4, ), (1, ))
assert_size_stride(primals_16, (4, 4), (4, 1))
assert_size_stride(primals_17, (4, ), (1, ))
assert_size_stride(primals_18, (4, 4), (4, 1))
assert_size_stride(primals_19, (4, ), (1, ))
assert_size_stride(primals_20, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_21, (4, 4), (4, 1))
assert_size_stride(primals_22, (4, ), (1, ))
assert_size_stride(primals_23, (4, ), (1, ))
assert_size_stride(primals_24, (4, ), (1, ))
assert_size_stride(primals_25, (4, 4), (4, 1))
assert_size_stride(primals_26, (4, ), (1, ))
assert_size_stride(primals_27, (4, 4), (4, 1))
assert_size_stride(primals_28, (4, ), (1, ))
assert_size_stride(primals_29, (4, ), (1, ))
assert_size_stride(primals_30, (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: [q], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_1, (16, 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((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_1, (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: [v], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_6
del primals_7
buf3 = 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, 1), 0), reinterpret_tensor(buf1, (16, 1, 4), (4, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [eq], Original ATen: [aten.eq]
stream0 = get_raw_stream(0)
triton_poi_fused_eq_0.run(primals_8, buf4, 256, grid=grid(256), stream=stream0)
del primals_8
buf5 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
triton_poi_fused__softmax_div_masked_fill_1.run(buf4, buf3, buf5, buf6, 64, grid=grid(64), stream=stream0)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [score, score_1, score_2], Original ATen: [aten.div, aten.masked_fill, aten._softmax]
triton_poi_fused__softmax_div_masked_fill_2.run(buf7, buf4, buf5, buf6, 256, grid=grid(256), stream=stream0)
buf8 = reinterpret_tensor(buf6, (16, 4, 1), (4, 1, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [v_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0), out=buf8)
buf9 = reinterpret_tensor(buf5, (16, 4), (4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf9)
del primals_10
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = buf10; del buf10 # reuse
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add, mean, std], Original ATen: [aten.add, aten.mean, aten.std]
triton_poi_fused_add_mean_std_3.run(buf11, buf9, primals_1, buf12, 16, grid=grid(16), stream=stream0)
buf13 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, mean, std, sub, add_1, out_1, mul, out_2], Original ATen: [aten.add, aten.mean, aten.std, aten.sub, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_std_sub_4.run(primals_11, buf9, primals_1, buf12, buf11, primals_12, buf13, 64, grid=grid(64), stream=stream0)
del buf11
del buf12
del primals_12
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [q_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_15, reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_14, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf14)
del primals_15
buf15 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [k_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_17, reinterpret_tensor(primals_13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_16, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf15)
del primals_16
del primals_17
buf16 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [v_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_19, reinterpret_tensor(primals_13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_18, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf16)
del primals_18
del primals_19
buf17 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul_2], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf14, (16, 4, 1), (4, 1, 1), 0), reinterpret_tensor(buf15, (16, 1, 4), (4, 4, 1), 0), out=buf17)
buf18 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [eq_1], Original ATen: [aten.eq]
triton_poi_fused_eq_0.run(primals_20, buf18, 256, grid=grid(256), stream=stream0)
del primals_20
buf19 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf20 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score_1, score_3, score_4, score_5], Original ATen: [aten.masked_fill, aten.div, aten._softmax]
triton_poi_fused__softmax_div_masked_fill_1.run(buf18, buf17, buf19, buf20, 64, grid=grid(64), stream=stream0)
buf21 = reinterpret_tensor(buf17, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf17 # reuse
# Topologically Sorted Source Nodes: [score_1, score_3, score_4, score_5], Original ATen: [aten.masked_fill, aten.div, aten._softmax]
triton_poi_fused__softmax_div_masked_fill_2.run(buf21, buf18, buf19, buf20, 256, grid=grid(256), stream=stream0)
buf22 = reinterpret_tensor(buf20, (16, 4, 1), (4, 1, 1), 0); del buf20 # reuse
# Topologically Sorted Source Nodes: [v_3], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf21, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf16, (16, 4, 1), (4, 1, 1), 0), out=buf22)
buf23 = reinterpret_tensor(buf19, (16, 4), (4, 1), 0); del buf19 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf22, (16, 4), (4, 1), 0), reinterpret_tensor(primals_21, (4, 4), (1, 4), 0), out=buf23)
buf24 = reinterpret_tensor(buf23, (4, 4, 4), (16, 4, 1), 0); del buf23 # reuse
# Topologically Sorted Source Nodes: [add_3], Original ATen: [aten.add]
triton_poi_fused_add_5.run(buf24, primals_22, buf13, 64, grid=grid(64), stream=stream0)
del primals_22
buf25 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean_2, std_2, sub_1, add_4, out_4, mul_1, out_5], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_std_sub_6.run(primals_23, buf24, primals_24, buf25, 64, grid=grid(64), stream=stream0)
del primals_24
buf26 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf25, (16, 4), (4, 1), 0), reinterpret_tensor(primals_25, (4, 4), (1, 4), 0), out=buf26)
buf27 = reinterpret_tensor(buf26, (4, 4, 4), (16, 4, 1), 0); del buf26 # reuse
buf31 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_7.run(buf27, primals_26, buf31, 64, grid=grid(64), stream=stream0)
del primals_26
buf28 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf27, (16, 4), (4, 1), 0), reinterpret_tensor(primals_27, (4, 4), (1, 4), 0), out=buf28)
buf29 = reinterpret_tensor(buf28, (4, 4, 4), (16, 4, 1), 0); del buf28 # reuse
# Topologically Sorted Source Nodes: [add_6], Original ATen: [aten.add]
triton_poi_fused_add_5.run(buf29, primals_28, buf25, 64, grid=grid(64), stream=stream0)
del primals_28
buf30 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean_4, std_4, sub_2, add_7, out_6, mul_2, out_7], Original ATen: [aten.mean, aten.std, aten.sub, aten.add, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_std_sub_6.run(primals_29, buf29, primals_30, buf30, 64, grid=grid(64), stream=stream0)
del primals_30
return (buf30, primals_1, primals_11, primals_23, primals_29, buf4, buf7, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), buf9, reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_13, (16, 4), (4, 1), 0), buf18, buf21, reinterpret_tensor(buf22, (16, 4), (4, 1), 0), buf24, reinterpret_tensor(buf25, (16, 4), (4, 1), 0), reinterpret_tensor(buf27, (16, 4), (4, 1), 0), buf29, primals_27, buf31, primals_25, primals_21, reinterpret_tensor(buf16, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf14, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf15, (16, 4, 1), (4, 1, 4), 0), primals_14, primals_9, reinterpret_tensor(buf2, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf0, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf1, (16, 4, 1), (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), (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)
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, 4), (64, 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)
primals_13 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_18 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_19 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_20 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_21 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_22 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_23 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_24 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_25 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_26 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_27 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_28 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_29 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_30 = 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, 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])
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
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super(LayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
class PositionwiseFeedForward(nn.Module):
def __init__(self, d_model, hidden, drop_prob=0.1):
super(PositionwiseFeedForward, self).__init__()
self.linear1 = nn.Linear(d_model, hidden)
self.linear2 = nn.Linear(hidden, d_model)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=drop_prob)
def forward(self, x):
x = self.linear1(x)
x = self.relu(x)
x = self.dropout(x)
x = self.linear2(x)
return x
class DecoderLayer(nn.Module):
def __init__(self, d_model, ffn_hidden, n_head, drop_prob):
super(DecoderLayer, self).__init__()
self.self_attention = MultiHeadAttention(d_model=d_model, n_head=n_head
)
self.norm1 = LayerNorm(d_model=d_model)
self.dropout1 = nn.Dropout(p=drop_prob)
self.enc_dec_attention = MultiHeadAttention(d_model=d_model, n_head
=n_head)
self.norm2 = LayerNorm(d_model=d_model)
self.dropout2 = nn.Dropout(p=drop_prob)
self.ffn = PositionwiseFeedForward(d_model=d_model, hidden=
ffn_hidden, drop_prob=drop_prob)
self.norm3 = LayerNorm(d_model=d_model)
self.dropout3 = nn.Dropout(p=drop_prob)
def forward(self, dec, enc, t_mask, s_mask):
_x = dec
x = self.self_attention(q=dec, k=dec, v=dec, mask=t_mask)
x = self.norm1(x + _x)
x = self.dropout1(x)
if enc is not None:
_x = x
x = self.enc_dec_attention(q=x, k=enc, v=enc, mask=s_mask)
x = self.norm2(x + _x)
x = self.dropout2(x)
_x = x
x = self.ffn(x)
x = self.norm3(x + _x)
x = self.dropout3(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4,
4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_model': 4, 'ffn_hidden': 4, 'n_head': 4, 'drop_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_eq_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)
@triton.jit
def triton_poi_fused__softmax_div_masked_fill_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
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x0 + 64 * x1), xmask)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask).to(tl.int1)
tmp7 = tl.load(in_ptr1 + (16 + x0 + 64 * x1), xmask)
tmp11 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask).to(tl.int1)
tmp12 = tl.load(in_ptr1 + (32 + x0 + 64 * x1), xmask)
tmp16 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask).to(tl.int1)
tmp17 = tl.load(in_ptr1 + (48 + x0 + 64 * x1), xmask)
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp8 = tmp7 * tmp2
tmp9 = tl.where(tmp6, tmp4, tmp8)
tmp10 = triton_helpers.maximum(tmp5, tmp9)
tmp13 = tmp12 * tmp2
tmp14 = tl.where(tmp11, tmp4, tmp13)
tmp15 = triton_helpers.maximum(tmp10, tmp14)
tmp18 = tmp17 * tmp2
tmp19 = tl.where(tmp16, tmp4, tmp18)
tmp20 = triton_helpers.maximum(tmp15, tmp19)
tmp21 = tmp5 - tmp20
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp9 - tmp20
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tmp14 - tmp20
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tmp19 - tmp20
tmp30 = tl_math.exp(tmp29)
tmp31 = tmp28 + tmp30
tl.store(out_ptr0 + x2, tmp20, xmask)
tl.store(out_ptr1 + x2, tmp31, xmask)
@triton.jit
def triton_poi_fused__softmax_div_masked_fill_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
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask).to(tl.int1)
tmp1 = tl.load(in_out_ptr0 + x3, xmask)
tmp6 = tl.load(in_ptr1 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr2 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = 1.0
tmp3 = tmp1 * tmp2
tmp4 = -9.999999960041972e-13
tmp5 = tl.where(tmp0, tmp4, tmp3)
tmp7 = tmp5 - tmp6
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 / tmp9
tl.store(in_out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_mean_std_3(in_out_ptr0, 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 + 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 = 3.0
tmp29 = tmp27 / tmp28
tl.store(in_out_ptr0 + x0, tmp29, xmask)
tl.store(out_ptr0 + x0, tmp16, xmask)
@triton.jit
def triton_poi_fused_add_div_mean_mul_std_sub_4(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
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_ptr2 + x2, xmask)
tmp4 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 - tmp4
tmp7 = libdevice.sqrt(tmp6)
tmp8 = 1e-12
tmp9 = tmp7 + tmp8
tmp10 = tmp5 / tmp9
tmp11 = tmp0 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_poi_fused_add_5(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_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_add_div_mean_mul_std_sub_6(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 % 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')
tmp3 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp2 - tmp10
tmp13 = tmp12 * tmp12
tmp14 = tmp3 - tmp10
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp10
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp10
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = 3.0
tmp24 = tmp22 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = 1e-12
tmp27 = tmp25 + tmp26
tmp28 = tmp11 / tmp27
tmp29 = tmp0 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + x2, tmp31, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_7(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, 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) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (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,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4, 4, 4), (64, 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,))
assert_size_stride(primals_13, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_14, (4, 4), (4, 1))
assert_size_stride(primals_15, (4,), (1,))
assert_size_stride(primals_16, (4, 4), (4, 1))
assert_size_stride(primals_17, (4,), (1,))
assert_size_stride(primals_18, (4, 4), (4, 1))
assert_size_stride(primals_19, (4,), (1,))
assert_size_stride(primals_20, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_21, (4, 4), (4, 1))
assert_size_stride(primals_22, (4,), (1,))
assert_size_stride(primals_23, (4,), (1,))
assert_size_stride(primals_24, (4,), (1,))
assert_size_stride(primals_25, (4, 4), (4, 1))
assert_size_stride(primals_26, (4,), (1,))
assert_size_stride(primals_27, (4, 4), (4, 1))
assert_size_stride(primals_28, (4,), (1,))
assert_size_stride(primals_29, (4,), (1,))
assert_size_stride(primals_30, (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_3, reinterpret_tensor(primals_1, (16,
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((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_1, (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_7, reinterpret_tensor(primals_1, (16,
4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf2)
del primals_6
del primals_7
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 1), (4, 1, 1),
0), reinterpret_tensor(buf1, (16, 1, 4), (4, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_eq_0[grid(256)](primals_8, buf4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_8
buf5 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
triton_poi_fused__softmax_div_masked_fill_1[grid(64)](buf4, buf3,
buf5, buf6, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused__softmax_div_masked_fill_2[grid(256)](buf7, buf4,
buf5, buf6, 256, XBLOCK=256, num_warps=4, num_stages=1)
buf8 = reinterpret_tensor(buf6, (16, 4, 1), (4, 1, 1), 0)
del buf6
extern_kernels.bmm(reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0), out=buf8)
buf9 = reinterpret_tensor(buf5, (16, 4), (4, 1), 0)
del buf5
extern_kernels.addmm(primals_10, reinterpret_tensor(buf8, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf9)
del primals_10
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = buf10
del buf10
buf12 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_mean_std_3[grid(16)](buf11, buf9, primals_1,
buf12, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf13 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_std_sub_4[grid(64)](primals_11,
buf9, primals_1, buf12, buf11, primals_12, buf13, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf11
del buf12
del primals_12
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_15, reinterpret_tensor(buf13, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_14, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf14)
del primals_15
buf15 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_17, reinterpret_tensor(primals_13, (16,
4), (4, 1), 0), reinterpret_tensor(primals_16, (4, 4), (1, 4),
0), alpha=1, beta=1, out=buf15)
del primals_16
del primals_17
buf16 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_19, reinterpret_tensor(primals_13, (16,
4), (4, 1), 0), reinterpret_tensor(primals_18, (4, 4), (1, 4),
0), alpha=1, beta=1, out=buf16)
del primals_18
del primals_19
buf17 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf14, (16, 4, 1), (4, 1, 1),
0), reinterpret_tensor(buf15, (16, 1, 4), (4, 4, 1), 0), out=buf17)
buf18 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_eq_0[grid(256)](primals_20, buf18, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_20
buf19 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf20 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
triton_poi_fused__softmax_div_masked_fill_1[grid(64)](buf18, buf17,
buf19, buf20, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf21 = reinterpret_tensor(buf17, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf17
triton_poi_fused__softmax_div_masked_fill_2[grid(256)](buf21, buf18,
buf19, buf20, 256, XBLOCK=256, num_warps=4, num_stages=1)
buf22 = reinterpret_tensor(buf20, (16, 4, 1), (4, 1, 1), 0)
del buf20
extern_kernels.bmm(reinterpret_tensor(buf21, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf16, (16, 4, 1), (4, 1, 1), 0), out=buf22)
buf23 = reinterpret_tensor(buf19, (16, 4), (4, 1), 0)
del buf19
extern_kernels.mm(reinterpret_tensor(buf22, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_21, (4, 4), (1, 4), 0), out=buf23)
buf24 = reinterpret_tensor(buf23, (4, 4, 4), (16, 4, 1), 0)
del buf23
triton_poi_fused_add_5[grid(64)](buf24, primals_22, buf13, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_22
buf25 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_std_sub_6[grid(64)](primals_23,
buf24, primals_24, buf25, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_24
buf26 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf25, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_25, (4, 4), (1, 4), 0), out=buf26)
buf27 = reinterpret_tensor(buf26, (4, 4, 4), (16, 4, 1), 0)
del buf26
buf31 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_7[grid(64)](buf27,
primals_26, buf31, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_26
buf28 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf27, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_27, (4, 4), (1, 4), 0), out=buf28)
buf29 = reinterpret_tensor(buf28, (4, 4, 4), (16, 4, 1), 0)
del buf28
triton_poi_fused_add_5[grid(64)](buf29, primals_28, buf25, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_28
buf30 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_std_sub_6[grid(64)](primals_29,
buf29, primals_30, buf30, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_30
return (buf30, primals_1, primals_11, primals_23, primals_29, buf4,
buf7, reinterpret_tensor(buf8, (16, 4), (4, 1), 0), buf9,
reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(
primals_13, (16, 4), (4, 1), 0), buf18, buf21, reinterpret_tensor(
buf22, (16, 4), (4, 1), 0), buf24, reinterpret_tensor(buf25, (16, 4
), (4, 1), 0), reinterpret_tensor(buf27, (16, 4), (4, 1), 0), buf29,
primals_27, buf31, primals_25, primals_21, reinterpret_tensor(buf16,
(16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf14, (16, 1, 4), (4,
1, 1), 0), reinterpret_tensor(buf15, (16, 4, 1), (4, 1, 4), 0),
primals_14, primals_9, reinterpret_tensor(buf2, (16, 1, 4), (4, 1,
1), 0), reinterpret_tensor(buf0, (16, 1, 4), (4, 1, 1), 0),
reinterpret_tensor(buf1, (16, 4, 1), (4, 1, 4), 0))
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super(LayerNorm, self).__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super(ScaleDotProductAttention, self).__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
class PositionwiseFeedForward(nn.Module):
def __init__(self, d_model, hidden, drop_prob=0.1):
super(PositionwiseFeedForward, self).__init__()
self.linear1 = nn.Linear(d_model, hidden)
self.linear2 = nn.Linear(hidden, d_model)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=drop_prob)
def forward(self, x):
x = self.linear1(x)
x = self.relu(x)
x = self.dropout(x)
x = self.linear2(x)
return x
class DecoderLayerNew(nn.Module):
def __init__(self, d_model, ffn_hidden, n_head, drop_prob):
super(DecoderLayerNew, self).__init__()
self.self_attention = MultiHeadAttention(d_model=d_model, n_head=n_head
)
self.norm1 = LayerNorm(d_model=d_model)
self.dropout1 = nn.Dropout(p=drop_prob)
self.enc_dec_attention = MultiHeadAttention(d_model=d_model, n_head
=n_head)
self.norm2 = LayerNorm(d_model=d_model)
self.dropout2 = nn.Dropout(p=drop_prob)
self.ffn = PositionwiseFeedForward(d_model=d_model, hidden=
ffn_hidden, drop_prob=drop_prob)
self.norm3 = LayerNorm(d_model=d_model)
self.dropout3 = nn.Dropout(p=drop_prob)
def forward(self, input_0, input_1, input_2, input_3):
primals_2 = self.self_attention.w_q.weight
primals_3 = self.self_attention.w_q.bias
primals_4 = self.self_attention.w_k.weight
primals_5 = self.self_attention.w_k.bias
primals_6 = self.self_attention.w_v.weight
primals_7 = self.self_attention.w_v.bias
primals_9 = self.self_attention.w_concat.weight
primals_10 = self.self_attention.w_concat.bias
primals_11 = self.norm1.gamma
primals_12 = self.norm1.beta
primals_14 = self.enc_dec_attention.w_q.weight
primals_15 = self.enc_dec_attention.w_q.bias
primals_16 = self.enc_dec_attention.w_k.weight
primals_17 = self.enc_dec_attention.w_k.bias
primals_18 = self.enc_dec_attention.w_v.weight
primals_19 = self.enc_dec_attention.w_v.bias
primals_21 = self.enc_dec_attention.w_concat.weight
primals_22 = self.enc_dec_attention.w_concat.bias
primals_23 = self.norm2.gamma
primals_24 = self.norm2.beta
primals_25 = self.ffn.linear1.weight
primals_26 = self.ffn.linear1.bias
primals_27 = self.ffn.linear2.weight
primals_28 = self.ffn.linear2.bias
primals_29 = self.norm3.gamma
primals_30 = self.norm3.beta
primals_1 = input_0
primals_13 = input_1
primals_8 = input_2
primals_20 = input_3
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])
return output[0]
| bsgiovanini/transformer | DecoderLayer | false | 1,619 | [
"Apache-2.0"
] | 0 | e128fa862f1b3d17d7b92df169a2bbee3f08366f | https://github.com/bsgiovanini/transformer/tree/e128fa862f1b3d17d7b92df169a2bbee3f08366f | import math
import torch
from torch import nn
class LayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
super().__init__()
self.gamma = nn.Parameter(torch.ones(d_model))
self.beta = nn.Parameter(torch.zeros(d_model))
self.eps = eps
def forward(self, x):
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
out = (x - mean) / (std + self.eps)
out = self.gamma * out + self.beta
return out
class ScaleDotProductAttention(nn.Module):
"""
compute scale dot product attention
Query : given sentence that we focused on (decoder)
Key : every sentence to check relationship with Qeury(encoder)
Value : every sentence same with Key (encoder)
"""
def __init__(self):
super().__init__()
self.softmax = nn.Softmax()
def forward(self, q, k, v, mask=None, e=1e-12):
batch_size, head, length, d_tensor = k.size()
k_t = k.view(batch_size, head, d_tensor, length)
score = q @ k_t / math.sqrt(d_tensor)
if mask is not None:
score = score.masked_fill(mask == 0, -e)
score = self.softmax(score)
v = score @ v
return v, score
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, n_head):
super().__init__()
self.n_head = n_head
self.attention = ScaleDotProductAttention()
self.w_q = nn.Linear(d_model, d_model)
self.w_k = nn.Linear(d_model, d_model)
self.w_v = nn.Linear(d_model, d_model)
self.w_concat = nn.Linear(d_model, d_model)
def forward(self, q, k, v, mask=None):
q, k, v = self.w_q(q), self.w_k(k), self.w_v(v)
q, k, v = self.split(q), self.split(k), self.split(v)
out, _attention = self.attention(q, k, v, mask=mask)
out = self.concat(out)
out = self.w_concat(out)
return out
def split(self, tensor):
"""
split tensor by number of head
:param tensor: [batch_size, length, d_model]
:return: [batch_size, head, length, d_tensor]
"""
batch_size, length, d_model = tensor.size()
d_tensor = d_model // self.n_head
tensor = tensor.view(batch_size, self.n_head, length, d_tensor)
return tensor
def concat(self, tensor):
"""
inverse function of self.split(tensor : torch.Tensor)
:param tensor: [batch_size, head, length, d_tensor]
:return: [batch_size, length, d_model]
"""
batch_size, head, length, d_tensor = tensor.size()
d_model = head * d_tensor
tensor = tensor.view(batch_size, length, d_model)
return tensor
class PositionwiseFeedForward(nn.Module):
def __init__(self, d_model, hidden, drop_prob=0.1):
super().__init__()
self.linear1 = nn.Linear(d_model, hidden)
self.linear2 = nn.Linear(hidden, d_model)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=drop_prob)
def forward(self, x):
x = self.linear1(x)
x = self.relu(x)
x = self.dropout(x)
x = self.linear2(x)
return x
class Model(nn.Module):
def __init__(self, d_model, ffn_hidden, n_head, drop_prob):
super().__init__()
self.self_attention = MultiHeadAttention(d_model=d_model, n_head=n_head
)
self.norm1 = LayerNorm(d_model=d_model)
self.dropout1 = nn.Dropout(p=drop_prob)
self.enc_dec_attention = MultiHeadAttention(d_model=d_model, n_head
=n_head)
self.norm2 = LayerNorm(d_model=d_model)
self.dropout2 = nn.Dropout(p=drop_prob)
self.ffn = PositionwiseFeedForward(d_model=d_model, hidden=
ffn_hidden, drop_prob=drop_prob)
self.norm3 = LayerNorm(d_model=d_model)
self.dropout3 = nn.Dropout(p=drop_prob)
def forward(self, dec, enc, t_mask, s_mask):
_x = dec
x = self.self_attention(q=dec, k=dec, v=dec, mask=t_
# ... truncated (>4000 chars) for memory efficiency |
CrossEntropyWithLogSoftmax | # 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_6/inductor_cache/cs/ccsm5hun5cjbnfbypk7rtely22ozwggyrnoadiiwljeaz4usjlv5.py
# Topologically Sorted Source Nodes: [mul, mean, neg], Original ATen: [aten.mul, aten.mean, aten.neg]
# Source node to ATen node mapping:
# mean => mean
# mul => mul
# neg => neg
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%mean,), kwargs = {})
triton_per_fused_mean_mul_neg_0 = async_compile.triton('triton_per_fused_mean_mul_neg_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_mean_mul_neg_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_mean_mul_neg_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)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = tmp0 * tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = triton_helpers.promote_to_tensor(tl.sum(tmp3, 0))
tmp6 = 256.0
tmp7 = tmp5 / tmp6
tmp8 = -tmp7
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp8, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul, mean, neg], Original ATen: [aten.mul, aten.mean, aten.neg]
stream0 = get_raw_stream(0)
triton_per_fused_mean_mul_neg_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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 CrossEntropyWithLogSoftmax(nn.Module):
def forward(self, y_hat, y):
return -(y_hat * y).mean()
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_mean_mul_neg_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)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = tmp0 * tmp1
tmp3 = tl.broadcast_to(tmp2, [RBLOCK])
tmp5 = triton_helpers.promote_to_tensor(tl.sum(tmp3, 0))
tmp6 = 256.0
tmp7 = tmp5 / tmp6
tmp8 = -tmp7
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp8, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_mul_neg_0[grid(1)](buf1, arg0_1, arg1_1, 1,
256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class CrossEntropyWithLogSoftmaxNew(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]
| cadurosar/graph_kd_dense_cifar100 | CrossEntropyWithLogSoftmax | false | 1,620 | [
"MIT"
] | 0 | 84054ab4f8f61c9db3460993661ba7bf1d951b36 | https://github.com/cadurosar/graph_kd_dense_cifar100/tree/84054ab4f8f61c9db3460993661ba7bf1d951b36 | import torch
import torch.nn as nn
class Model(nn.Module):
def forward(self, y_hat, y):
return -(y_hat * y).mean()
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
KlLoss | # 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_6/inductor_cache/jc/cjc73l2t54xbd7dueafrmv27tpu5mxfktwrqdi77aaakjgxdal3p.py
# Topologically Sorted Source Nodes: [truediv, log, x, abs_1, loss_kl], Original ATen: [aten.div, aten.log, aten.mul, aten.abs, aten.mean]
# Source node to ATen node mapping:
# abs_1 => abs_1
# log => log
# loss_kl => mean
# truediv => div
# x => mul
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%div,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %log), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%mul,), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%abs_1,), kwargs = {})
triton_per_fused_abs_div_log_mean_mul_0 = async_compile.triton('triton_per_fused_abs_div_log_mean_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.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_abs_div_log_mean_mul_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_abs_div_log_mean_mul_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)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = tmp0 / tmp1
tmp3 = tl_math.log(tmp2)
tmp4 = tmp0 * tmp3
tmp5 = tl_math.abs(tmp4)
tmp6 = tl.broadcast_to(tmp5, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = 256.0
tmp10 = tmp8 / tmp9
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp10, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [truediv, log, x, abs_1, loss_kl], Original ATen: [aten.div, aten.log, aten.mul, aten.abs, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_abs_div_log_mean_mul_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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
def kl_div(p: 'torch.Tensor', q: 'torch.Tensor') ->torch.Tensor:
x = p * torch.log(p / q)
return x.abs().mean()
class KlLoss(nn.Module):
def __init__(self) ->None:
super().__init__()
def forward(self, inputs: 'torch.Tensor', targets: 'torch.Tensor'):
loss_kl = kl_div(targets, inputs)
return loss_kl
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_per_fused_abs_div_log_mean_mul_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)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = tmp0 / tmp1
tmp3 = tl_math.log(tmp2)
tmp4 = tmp0 * tmp3
tmp5 = tl_math.abs(tmp4)
tmp6 = tl.broadcast_to(tmp5, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = 256.0
tmp10 = tmp8 / tmp9
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp10, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_abs_div_log_mean_mul_0[grid(1)](buf1, arg0_1,
arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
def kl_div(p: 'torch.Tensor', q: 'torch.Tensor') ->torch.Tensor:
x = p * torch.log(p / q)
return x.abs().mean()
class KlLossNew(nn.Module):
def __init__(self) ->None:
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]
| by-liu/RetinalApp | KlLoss | false | 1,621 | [
"MIT"
] | 0 | 53173b2b20dfcf613a3a22d6caa5178771d14225 | https://github.com/by-liu/RetinalApp/tree/53173b2b20dfcf613a3a22d6caa5178771d14225 | import torch
import torch.nn as nn
def kl_div(p: 'torch.Tensor', q: 'torch.Tensor') ->torch.Tensor:
x = p * torch.log(p / q)
return x.abs().mean()
class Model(nn.Module):
def __init__(self) ->None:
super().__init__()
def forward(self, inputs: 'torch.Tensor', targets: 'torch.Tensor'):
loss_kl = kl_div(targets, inputs)
return loss_kl
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
BatchMeanCrossEntropyWithLogSoftmax | # 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_6/inductor_cache/ij/cijsjachqiycp6ry5ykl2jjt6skuqvdpm4coostfotjab2ywqkya.py
# Topologically Sorted Source Nodes: [mul, sum_1, mean, neg], Original ATen: [aten.mul, aten.sum, aten.mean, aten.neg]
# Source node to ATen node mapping:
# mean => mean
# mul => mul
# neg => neg
# sum_1 => sum_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1]), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%sum_1, [0]), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%mean,), kwargs = {})
triton_poi_fused_mean_mul_neg_sum_0 = async_compile.triton('triton_poi_fused_mean_mul_neg_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=[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_mean_mul_neg_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 32, 'num_reduction': 0, 'backend_hash': 'A9C866B4A14FD3277824029365D703C2427B2E685E54EC9B3EF4ADC8D1EEAC1D', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_neg_sum_0(in_out_ptr0, in_ptr0, in_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 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp3 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp4 = tl.load(in_ptr1 + (16 + x0), xmask)
tmp7 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp8 = tl.load(in_ptr1 + (32 + x0), xmask)
tmp11 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp12 = tl.load(in_ptr1 + (48 + x0), xmask)
tmp15 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp16 = tl.load(in_ptr1 + (64 + x0), xmask)
tmp18 = tl.load(in_ptr0 + (80 + x0), xmask)
tmp19 = tl.load(in_ptr1 + (80 + x0), xmask)
tmp22 = tl.load(in_ptr0 + (96 + x0), xmask)
tmp23 = tl.load(in_ptr1 + (96 + x0), xmask)
tmp26 = tl.load(in_ptr0 + (112 + x0), xmask)
tmp27 = tl.load(in_ptr1 + (112 + x0), xmask)
tmp31 = tl.load(in_ptr0 + (128 + x0), xmask)
tmp32 = tl.load(in_ptr1 + (128 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (144 + x0), xmask)
tmp35 = tl.load(in_ptr1 + (144 + x0), xmask)
tmp38 = tl.load(in_ptr0 + (160 + x0), xmask)
tmp39 = tl.load(in_ptr1 + (160 + x0), xmask)
tmp42 = tl.load(in_ptr0 + (176 + x0), xmask)
tmp43 = tl.load(in_ptr1 + (176 + x0), xmask)
tmp47 = tl.load(in_ptr0 + (192 + x0), xmask)
tmp48 = tl.load(in_ptr1 + (192 + x0), xmask)
tmp50 = tl.load(in_ptr0 + (208 + x0), xmask)
tmp51 = tl.load(in_ptr1 + (208 + x0), xmask)
tmp54 = tl.load(in_ptr0 + (224 + x0), xmask)
tmp55 = tl.load(in_ptr1 + (224 + x0), xmask)
tmp58 = tl.load(in_ptr0 + (240 + x0), xmask)
tmp59 = tl.load(in_ptr1 + (240 + x0), xmask)
tmp2 = tmp0 * tmp1
tmp5 = tmp3 * tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 * tmp12
tmp14 = tmp10 + tmp13
tmp17 = tmp15 * tmp16
tmp20 = tmp18 * tmp19
tmp21 = tmp17 + tmp20
tmp24 = tmp22 * tmp23
tmp25 = tmp21 + tmp24
tmp28 = tmp26 * tmp27
tmp29 = tmp25 + tmp28
tmp30 = tmp14 + tmp29
tmp33 = tmp31 * tmp32
tmp36 = tmp34 * tmp35
tmp37 = tmp33 + tmp36
tmp40 = tmp38 * tmp39
tmp41 = tmp37 + tmp40
tmp44 = tmp42 * tmp43
tmp45 = tmp41 + tmp44
tmp46 = tmp30 + tmp45
tmp49 = tmp47 * tmp48
tmp52 = tmp50 * tmp51
tmp53 = tmp49 + tmp52
tmp56 = tmp54 * tmp55
tmp57 = tmp53 + tmp56
tmp60 = tmp58 * tmp59
tmp61 = tmp57 + tmp60
tmp62 = tmp46 + tmp61
tmp63 = 4.0
tmp64 = tmp62 / tmp63
tmp65 = -tmp64
tl.store(in_out_ptr0 + (x0), tmp65, 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, 1), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul, sum_1, mean, neg], Original ATen: [aten.mul, aten.sum, aten.mean, aten.neg]
stream0 = get_raw_stream(0)
triton_poi_fused_mean_mul_neg_sum_0.run(buf1, arg0_1, arg1_1, 16, grid=grid(16), stream=stream0)
del arg0_1
del arg1_1
return (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((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 BatchMeanCrossEntropyWithLogSoftmax(nn.Module):
def forward(self, y_hat, y):
return -(y_hat * y).sum(dim=1).mean(dim=0)
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
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_mul_neg_sum_0(in_out_ptr0, in_ptr0, in_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 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp3 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp4 = tl.load(in_ptr1 + (16 + x0), xmask)
tmp7 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp8 = tl.load(in_ptr1 + (32 + x0), xmask)
tmp11 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp12 = tl.load(in_ptr1 + (48 + x0), xmask)
tmp15 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp16 = tl.load(in_ptr1 + (64 + x0), xmask)
tmp18 = tl.load(in_ptr0 + (80 + x0), xmask)
tmp19 = tl.load(in_ptr1 + (80 + x0), xmask)
tmp22 = tl.load(in_ptr0 + (96 + x0), xmask)
tmp23 = tl.load(in_ptr1 + (96 + x0), xmask)
tmp26 = tl.load(in_ptr0 + (112 + x0), xmask)
tmp27 = tl.load(in_ptr1 + (112 + x0), xmask)
tmp31 = tl.load(in_ptr0 + (128 + x0), xmask)
tmp32 = tl.load(in_ptr1 + (128 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (144 + x0), xmask)
tmp35 = tl.load(in_ptr1 + (144 + x0), xmask)
tmp38 = tl.load(in_ptr0 + (160 + x0), xmask)
tmp39 = tl.load(in_ptr1 + (160 + x0), xmask)
tmp42 = tl.load(in_ptr0 + (176 + x0), xmask)
tmp43 = tl.load(in_ptr1 + (176 + x0), xmask)
tmp47 = tl.load(in_ptr0 + (192 + x0), xmask)
tmp48 = tl.load(in_ptr1 + (192 + x0), xmask)
tmp50 = tl.load(in_ptr0 + (208 + x0), xmask)
tmp51 = tl.load(in_ptr1 + (208 + x0), xmask)
tmp54 = tl.load(in_ptr0 + (224 + x0), xmask)
tmp55 = tl.load(in_ptr1 + (224 + x0), xmask)
tmp58 = tl.load(in_ptr0 + (240 + x0), xmask)
tmp59 = tl.load(in_ptr1 + (240 + x0), xmask)
tmp2 = tmp0 * tmp1
tmp5 = tmp3 * tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 * tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 * tmp12
tmp14 = tmp10 + tmp13
tmp17 = tmp15 * tmp16
tmp20 = tmp18 * tmp19
tmp21 = tmp17 + tmp20
tmp24 = tmp22 * tmp23
tmp25 = tmp21 + tmp24
tmp28 = tmp26 * tmp27
tmp29 = tmp25 + tmp28
tmp30 = tmp14 + tmp29
tmp33 = tmp31 * tmp32
tmp36 = tmp34 * tmp35
tmp37 = tmp33 + tmp36
tmp40 = tmp38 * tmp39
tmp41 = tmp37 + tmp40
tmp44 = tmp42 * tmp43
tmp45 = tmp41 + tmp44
tmp46 = tmp30 + tmp45
tmp49 = tmp47 * tmp48
tmp52 = tmp50 * tmp51
tmp53 = tmp49 + tmp52
tmp56 = tmp54 * tmp55
tmp57 = tmp53 + tmp56
tmp60 = tmp58 * tmp59
tmp61 = tmp57 + tmp60
tmp62 = tmp46 + tmp61
tmp63 = 4.0
tmp64 = tmp62 / tmp63
tmp65 = -tmp64
tl.store(in_out_ptr0 + x0, tmp65, 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, 1), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_mean_mul_neg_sum_0[grid(16)](buf1, arg0_1, arg1_1,
16, XBLOCK=16, num_warps=1, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class BatchMeanCrossEntropyWithLogSoftmaxNew(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]
| cadurosar/graph_kd_dense_cifar100 | BatchMeanCrossEntropyWithLogSoftmax | false | 1,622 | [
"MIT"
] | 0 | 84054ab4f8f61c9db3460993661ba7bf1d951b36 | https://github.com/cadurosar/graph_kd_dense_cifar100/tree/84054ab4f8f61c9db3460993661ba7bf1d951b36 | import torch
import torch.nn as nn
class Model(nn.Module):
def forward(self, y_hat, y):
return -(y_hat * y).sum(dim=1).mean(dim=0)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
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_6/inductor_cache/s4/cs4nfd57ofr376fmxwurzhs4gjpm5ozhhk2tbhok3earnld65s7o.py
# Topologically Sorted Source Nodes: [e, e_1], Original ATen: [aten.add, aten.leaky_relu]
# Source node to ATen node mapping:
# e => add
# e_1 => gt
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %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=[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_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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = (xindex // 4)
x0 = xindex % 4
x2 = (xindex // 16)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tl.store(out_ptr0 + (x4), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/4p/c4pjqwbjkzp7rz5o45l7yogltumroijgxjr25vimim5pknn4qw4m.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_2, 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=[64],
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 = 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 = 0.0
tmp2 = tmp0 > tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/b5/cb5nayitdupiwly7l5oyyidlurp5a3e2zgidsecoz7dssdptb2xy.py
# Topologically Sorted Source Nodes: [e, e_1, zero_vec, attention, attention_1, e_2, e_3, attention_3, attention_4, e_4, e_5, attention_6, attention_7, e_6, e_7, attention_9, attention_10], Original ATen: [aten.add, aten.leaky_relu, aten.mul, aten.where, aten._softmax]
# Source node to ATen node mapping:
# 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 => add
# e_1 => mul, where
# e_2 => add_1
# e_3 => mul_5, where_3
# e_4 => add_2
# e_5 => mul_10, where_6
# e_6 => add_3
# e_7 => mul_15, where_9
# zero_vec => full_default
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %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, 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 = (%view_12, %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 = (%view_21, %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 = (%view_30, %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=[16],
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, 17), 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 = 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 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp1 = tl.load(in_ptr1 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x2), xmask)
tmp3 = tl.load(in_ptr3 + (4*x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp11 = tl.load(in_ptr1 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp12 = tl.load(in_ptr3 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp19 = tl.load(in_ptr1 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp20 = tl.load(in_ptr3 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp27 = tl.load(in_ptr1 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp28 = tl.load(in_ptr3 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp34 = tl.load(in_ptr4 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp35 = tl.load(in_ptr5 + (x2), xmask)
tmp36 = tl.load(in_ptr6 + (4*x1), xmask, eviction_policy='evict_last')
tmp41 = tl.load(in_ptr4 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp42 = tl.load(in_ptr6 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp48 = tl.load(in_ptr4 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp49 = tl.load(in_ptr6 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp55 = tl.load(in_ptr4 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp56 = tl.load(in_ptr6 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp62 = tl.load(in_ptr7 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp63 = tl.load(in_ptr8 + (x2), xmask)
tmp64 = tl.load(in_ptr9 + (4*x1), xmask, eviction_policy='evict_last')
tmp69 = tl.load(in_ptr7 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp70 = tl.load(in_ptr9 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp76 = tl.load(in_ptr7 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp77 = tl.load(in_ptr9 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp83 = tl.load(in_ptr7 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp84 = tl.load(in_ptr9 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp90 = tl.load(in_ptr10 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp91 = tl.load(in_ptr11 + (x2), xmask)
tmp92 = tl.load(in_ptr12 + (4*x1), xmask, eviction_policy='evict_last')
tmp97 = tl.load(in_ptr10 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp98 = tl.load(in_ptr12 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp104 = tl.load(in_ptr10 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp105 = tl.load(in_ptr12 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp111 = tl.load(in_ptr10 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp112 = tl.load(in_ptr12 + (3 + (4*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)
tmp13 = tmp2 + tmp12
tmp14 = tmp13 * tmp5
tmp15 = tl.where(tmp11, tmp13, tmp14)
tmp16 = tl.where(tmp10, tmp15, tmp8)
tmp17 = triton_helpers.maximum(tmp9, tmp16)
tmp21 = tmp2 + tmp20
tmp22 = tmp21 * tmp5
tmp23 = tl.where(tmp19, tmp21, tmp22)
tmp24 = tl.where(tmp18, tmp23, tmp8)
tmp25 = triton_helpers.maximum(tmp17, tmp24)
tmp29 = tmp2 + tmp28
tmp30 = tmp29 * tmp5
tmp31 = tl.where(tmp27, tmp29, tmp30)
tmp32 = tl.where(tmp26, tmp31, tmp8)
tmp33 = triton_helpers.maximum(tmp25, tmp32)
tmp37 = tmp35 + tmp36
tmp38 = tmp37 * tmp5
tmp39 = tl.where(tmp34, tmp37, tmp38)
tmp40 = tl.where(tmp0, tmp39, tmp8)
tmp43 = tmp35 + tmp42
tmp44 = tmp43 * tmp5
tmp45 = tl.where(tmp41, tmp43, tmp44)
tmp46 = tl.where(tmp10, tmp45, tmp8)
tmp47 = triton_helpers.maximum(tmp40, tmp46)
tmp50 = tmp35 + tmp49
tmp51 = tmp50 * tmp5
tmp52 = tl.where(tmp48, tmp50, tmp51)
tmp53 = tl.where(tmp18, tmp52, tmp8)
tmp54 = triton_helpers.maximum(tmp47, tmp53)
tmp57 = tmp35 + tmp56
tmp58 = tmp57 * tmp5
tmp59 = tl.where(tmp55, tmp57, tmp58)
tmp60 = tl.where(tmp26, tmp59, tmp8)
tmp61 = triton_helpers.maximum(tmp54, tmp60)
tmp65 = tmp63 + tmp64
tmp66 = tmp65 * tmp5
tmp67 = tl.where(tmp62, tmp65, tmp66)
tmp68 = tl.where(tmp0, tmp67, tmp8)
tmp71 = tmp63 + tmp70
tmp72 = tmp71 * tmp5
tmp73 = tl.where(tmp69, tmp71, tmp72)
tmp74 = tl.where(tmp10, tmp73, tmp8)
tmp75 = triton_helpers.maximum(tmp68, tmp74)
tmp78 = tmp63 + tmp77
tmp79 = tmp78 * tmp5
tmp80 = tl.where(tmp76, tmp78, tmp79)
tmp81 = tl.where(tmp18, tmp80, tmp8)
tmp82 = triton_helpers.maximum(tmp75, tmp81)
tmp85 = tmp63 + tmp84
tmp86 = tmp85 * tmp5
tmp87 = tl.where(tmp83, tmp85, tmp86)
tmp88 = tl.where(tmp26, tmp87, tmp8)
tmp89 = triton_helpers.maximum(tmp82, tmp88)
tmp93 = tmp91 + tmp92
tmp94 = tmp93 * tmp5
tmp95 = tl.where(tmp90, tmp93, tmp94)
tmp96 = tl.where(tmp0, tmp95, tmp8)
tmp99 = tmp91 + tmp98
tmp100 = tmp99 * tmp5
tmp101 = tl.where(tmp97, tmp99, tmp100)
tmp102 = tl.where(tmp10, tmp101, tmp8)
tmp103 = triton_helpers.maximum(tmp96, tmp102)
tmp106 = tmp91 + tmp105
tmp107 = tmp106 * tmp5
tmp108 = tl.where(tmp104, tmp106, tmp107)
tmp109 = tl.where(tmp18, tmp108, tmp8)
tmp110 = triton_helpers.maximum(tmp103, tmp109)
tmp113 = tmp91 + tmp112
tmp114 = tmp113 * tmp5
tmp115 = tl.where(tmp111, tmp113, tmp114)
tmp116 = tl.where(tmp26, tmp115, tmp8)
tmp117 = triton_helpers.maximum(tmp110, tmp116)
tl.store(out_ptr0 + (x2), tmp33, xmask)
tl.store(out_ptr1 + (x2), tmp61, xmask)
tl.store(out_ptr2 + (x2), tmp89, xmask)
tl.store(out_ptr3 + (x2), tmp117, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/fb/cfbqm35w3cpnalep76jyvrmob4hwnfqcnjbdgika5kj63xczgouv.py
# Topologically Sorted Source Nodes: [e, e_1, zero_vec, attention, attention_1, e_2, e_3, attention_3, attention_4, e_4, e_5, attention_6, attention_7, e_6, e_7, attention_9, attention_10], Original ATen: [aten.add, aten.leaky_relu, aten.mul, aten.where, aten._softmax]
# Source node to ATen node mapping:
# 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 => add
# e_1 => mul, where
# e_2 => add_1
# e_3 => mul_5, where_3
# e_4 => add_2
# e_5 => mul_10, where_6
# e_6 => add_3
# e_7 => mul_15, where_9
# zero_vec => full_default
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %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, 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 = (%view_12, %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 = (%view_21, %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 = (%view_30, %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=[64],
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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x4 = (xindex // 4)
x0 = xindex % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x3), xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x4), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr5 + (x3), xmask).to(tl.int1)
tmp14 = tl.load(in_ptr6 + (x4), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr7 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr8 + (x4), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr9 + (x3), xmask).to(tl.int1)
tmp24 = tl.load(in_ptr10 + (x4), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr11 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr12 + (x4), xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr13 + (x3), xmask).to(tl.int1)
tmp34 = tl.load(in_ptr14 + (x4), xmask, eviction_policy='evict_last')
tmp35 = tl.load(in_ptr15 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp40 = tl.load(in_ptr16 + (x4), 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 + (x3), tmp12, xmask)
tl.store(out_ptr1 + (x3), tmp22, xmask)
tl.store(out_ptr2 + (x3), tmp32, xmask)
tl.store(out_ptr3 + (x3), tmp42, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/dw/cdwqsjnh2osfmjr2utzzaqdg2vrfivzkuhareq3urgidllj2bsvr.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=[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_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 = 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_6/inductor_cache/xd/cxd67j2epx2xomkthytqlsbq3x4hob3ajklr2i77yt72q4achxiv.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=1] = 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=[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_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 = 256
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_6/inductor_cache/6j/c6jpwvgkj3sqlgkyvvq2xry3p2d62ynbydcmjpmsi2uriudvkwsm.py
# Topologically Sorted Source Nodes: [zero_vec, e_8, e_9, attention_12, attention_13], Original ATen: [aten.mul, aten.add, aten.leaky_relu, aten.where, aten._softmax]
# Source node to ATen node mapping:
# attention_12 => where_13
# attention_13 => amax_4
# e_8 => add_4
# e_9 => 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, 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 = (%view_39, %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=[16],
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, 5), 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 = 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 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp1 = tl.load(in_ptr1 + (4*x2), xmask, eviction_policy='evict_last').to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x2), xmask)
tmp3 = tl.load(in_ptr3 + (4*x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp11 = tl.load(in_ptr1 + (1 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp12 = tl.load(in_ptr3 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp19 = tl.load(in_ptr1 + (2 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp20 = tl.load(in_ptr3 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp27 = tl.load(in_ptr1 + (3 + (4*x2)), xmask, eviction_policy='evict_last').to(tl.int1)
tmp28 = tl.load(in_ptr3 + (3 + (4*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)
tmp13 = tmp2 + tmp12
tmp14 = tmp13 * tmp5
tmp15 = tl.where(tmp11, tmp13, tmp14)
tmp16 = tl.where(tmp10, tmp15, tmp8)
tmp17 = triton_helpers.maximum(tmp9, tmp16)
tmp21 = tmp2 + tmp20
tmp22 = tmp21 * tmp5
tmp23 = tl.where(tmp19, tmp21, tmp22)
tmp24 = tl.where(tmp18, tmp23, tmp8)
tmp25 = triton_helpers.maximum(tmp17, tmp24)
tmp29 = tmp2 + tmp28
tmp30 = tmp29 * tmp5
tmp31 = tl.where(tmp27, tmp29, tmp30)
tmp32 = tl.where(tmp26, tmp31, tmp8)
tmp33 = triton_helpers.maximum(tmp25, tmp32)
tl.store(out_ptr0 + (x2), tmp33, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/kq/ckqsvuyjpykhrx63sgunf6ncivhca5ffsml3r6qydy4l5qz7egg5.py
# Topologically Sorted Source Nodes: [zero_vec, e_8, e_9, attention_12, attention_13], Original ATen: [aten.mul, aten.add, aten.leaky_relu, aten.where, aten._softmax]
# Source node to ATen node mapping:
# attention_12 => where_13
# attention_13 => exp_4, sub_4
# e_8 => add_4
# e_9 => 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, 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 = (%view_39, %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=[64],
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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x4 = (xindex // 4)
x0 = xindex % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + (x3), xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x4), 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 + (x3), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/2n/c2ncfhxfnnde24ns3z2dmmqvqs56zdhptmjm7a2ax5stkf247rnh.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 = (%bmm_4, 0), kwargs = {})
# %mul_22 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%bmm_4, 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=[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__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 = 64
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)
tmp8 = tl.load(in_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr0 + (12 + x0 + (16*x2)), 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 + (x3), tmp35, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/rp/crpcotgng62dxuybnefscgik6jjdcjwm6zi4ya7wklif6ha6kjym.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=[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__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 = 64
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_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (12 + x0 + (16*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, 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), (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, (8, 1), (1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (8, 1), (1, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (8, 1), (1, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (8, 1), (1, 1))
assert_size_stride(primals_11, (16, 4), (4, 1))
assert_size_stride(primals_12, (8, 1), (1, 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: [Wh], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), primals_3, out=buf0)
del primals_3
buf1 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh1], Original ATen: [aten.mm]
extern_kernels.mm(buf0, reinterpret_tensor(primals_4, (4, 1), (1, 1), 0), out=buf1)
buf2 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh2], Original ATen: [aten.mm]
extern_kernels.mm(buf0, reinterpret_tensor(primals_4, (4, 1), (1, 1), 4), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [e, e_1], Original ATen: [aten.add, aten.leaky_relu]
stream0 = get_raw_stream(0)
triton_poi_fused_add_leaky_relu_0.run(buf1, buf2, buf3, 64, grid=grid(64), stream=stream0)
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [gt], Original ATen: [aten.gt]
triton_poi_fused_gt_1.run(primals_2, buf4, 64, grid=grid(64), stream=stream0)
del primals_2
buf9 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), primals_5, out=buf9)
del primals_5
buf10 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh1_1], Original ATen: [aten.mm]
extern_kernels.mm(buf9, reinterpret_tensor(primals_6, (4, 1), (1, 1), 0), out=buf10)
buf11 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh2_1], Original ATen: [aten.mm]
extern_kernels.mm(buf9, reinterpret_tensor(primals_6, (4, 1), (1, 1), 4), out=buf11)
buf12 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [e_2, e_3], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf10, buf11, buf12, 64, grid=grid(64), stream=stream0)
buf17 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), primals_7, out=buf17)
del primals_7
buf18 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh1_2], Original ATen: [aten.mm]
extern_kernels.mm(buf17, reinterpret_tensor(primals_8, (4, 1), (1, 1), 0), out=buf18)
buf19 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh2_2], Original ATen: [aten.mm]
extern_kernels.mm(buf17, reinterpret_tensor(primals_8, (4, 1), (1, 1), 4), out=buf19)
buf20 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [e_4, e_5], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf18, buf19, buf20, 64, grid=grid(64), stream=stream0)
buf25 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), primals_9, out=buf25)
del primals_9
buf26 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh1_3], Original ATen: [aten.mm]
extern_kernels.mm(buf25, reinterpret_tensor(primals_10, (4, 1), (1, 1), 0), out=buf26)
buf27 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh2_3], Original ATen: [aten.mm]
extern_kernels.mm(buf25, reinterpret_tensor(primals_10, (4, 1), (1, 1), 4), out=buf27)
buf28 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [e_6, e_7], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf26, buf27, buf28, 64, grid=grid(64), stream=stream0)
buf5 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf21 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf29 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [e, e_1, zero_vec, attention, attention_1, e_2, e_3, attention_3, attention_4, e_4, e_5, attention_6, attention_7, e_6, e_7, 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, 16, grid=grid(16), stream=stream0)
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf22 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf30 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [e, e_1, zero_vec, attention, attention_1, e_2, e_3, attention_3, attention_4, e_4, e_5, attention_6, attention_7, e_6, e_7, 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, 64, grid=grid(64), 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), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf6, buf7, 64, grid=grid(64), stream=stream0)
buf8 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [h_prime], Original ATen: [aten.bmm]
extern_kernels.bmm(buf7, reinterpret_tensor(buf0, (4, 4, 4), (16, 4, 1), 0), out=buf8)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_4], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf14, buf15, 64, grid=grid(64), stream=stream0)
buf16 = buf14; del buf14 # reuse
# Topologically Sorted Source Nodes: [h_prime_1], Original ATen: [aten.bmm]
extern_kernels.bmm(buf15, reinterpret_tensor(buf9, (4, 4, 4), (16, 4, 1), 0), out=buf16)
buf23 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_7], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf22, buf23, 64, grid=grid(64), stream=stream0)
buf24 = buf22; del buf22 # reuse
# Topologically Sorted Source Nodes: [h_prime_2], Original ATen: [aten.bmm]
extern_kernels.bmm(buf23, reinterpret_tensor(buf17, (4, 4, 4), (16, 4, 1), 0), out=buf24)
buf31 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_10], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf30, buf31, 64, grid=grid(64), stream=stream0)
buf32 = buf30; del buf30 # reuse
# Topologically Sorted Source Nodes: [h_prime_3], Original ATen: [aten.bmm]
extern_kernels.bmm(buf31, reinterpret_tensor(buf25, (4, 4, 4), (16, 4, 1), 0), out=buf32)
buf33 = empty_strided_cuda((4, 4, 16), (64, 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, 256, grid=grid(256), stream=stream0)
buf34 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [Wh_4], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf33, (16, 16), (16, 1), 0), primals_11, out=buf34)
buf35 = reinterpret_tensor(buf5, (16, 1), (1, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [Wh1_4], Original ATen: [aten.mm]
extern_kernels.mm(buf34, reinterpret_tensor(primals_12, (4, 1), (1, 1), 0), out=buf35)
buf36 = reinterpret_tensor(buf29, (16, 1), (1, 1), 0); del buf29 # reuse
# Topologically Sorted Source Nodes: [Wh2_4], Original ATen: [aten.mm]
extern_kernels.mm(buf34, reinterpret_tensor(primals_12, (4, 1), (1, 1), 4), out=buf36)
buf37 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [e_8, e_9], Original ATen: [aten.add, aten.leaky_relu]
triton_poi_fused_add_leaky_relu_0.run(buf35, buf36, buf37, 64, grid=grid(64), stream=stream0)
buf38 = reinterpret_tensor(buf27, (4, 4, 1), (4, 1, 16), 0); del buf27 # reuse
# Topologically Sorted Source Nodes: [zero_vec, e_8, e_9, 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, 16, grid=grid(16), stream=stream0)
buf39 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [zero_vec, e_8, e_9, 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, 64, grid=grid(64), stream=stream0)
del buf35
del buf36
del buf38
buf40 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attention_13], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf39, buf40, 64, grid=grid(64), stream=stream0)
buf41 = buf39; del buf39 # reuse
# Topologically Sorted Source Nodes: [h_prime_4], Original ATen: [aten.bmm]
extern_kernels.bmm(buf40, reinterpret_tensor(buf34, (4, 4, 4), (16, 4, 1), 0), out=buf41)
buf42 = empty_strided_cuda((4, 4, 4), (16, 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, 64, grid=grid(64), stream=stream0)
buf43 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_9.run(buf42, buf43, 64, grid=grid(64), stream=stream0)
del buf42
return (buf43, buf0, buf3, buf4, buf7, buf8, buf9, buf12, buf15, buf16, buf17, buf20, buf23, buf24, buf25, buf28, buf31, buf32, buf34, buf37, buf40, buf41, buf43, reinterpret_tensor(primals_12, (1, 4), (1, 1), 4), reinterpret_tensor(primals_12, (1, 4), (1, 1), 0), reinterpret_tensor(buf33, (16, 16), (1, 16), 0), reinterpret_tensor(primals_11, (4, 16), (1, 4), 0), reinterpret_tensor(primals_10, (1, 4), (1, 1), 4), reinterpret_tensor(primals_10, (1, 4), (1, 1), 0), reinterpret_tensor(primals_1, (4, 16), (1, 4), 0), reinterpret_tensor(primals_8, (1, 4), (1, 1), 4), reinterpret_tensor(primals_8, (1, 4), (1, 1), 0), reinterpret_tensor(primals_6, (1, 4), (1, 1), 4), reinterpret_tensor(primals_6, (1, 4), (1, 1), 0), reinterpret_tensor(primals_4, (1, 4), (1, 1), 4), reinterpret_tensor(primals_4, (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), (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((8, 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((8, 1), (1, 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((8, 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((8, 1), (1, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((8, 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])
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 GraphAttentionLayer(nn.Module):
"""
https://github.com/Diego999/pyGAT/blob/master/layers.py
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, batch_num,
node_num, concat=True):
super(GraphAttentionLayer, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
self.batch_num = batch_num
self.node_num = node_num
torch.device('cuda')
self.W = nn.Parameter(torch.empty(size=(in_features, out_features)))
nn.init.xavier_uniform_(self.W.data, gain=1.414)
self.a = nn.Parameter(torch.empty(size=(2 * out_features, 1)))
nn.init.xavier_uniform_(self.a.data, gain=1.414)
self.leakyrelu = nn.LeakyReLU(self.alpha)
def forward(self, h, adj):
"""
:param h: (batch_size, number_nodes, in_features)
:param adj: (batch_size, number_nodes, number_nodes)
:return: (batch_size, number_nodes, out_features)
"""
Wh = torch.matmul(h, self.W)
e = self.prepare_batch(Wh)
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, Wh)
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def prepare_batch(self, Wh):
"""
with batch training
:param Wh: (batch_size, number_nodes, out_features)
:return:
"""
_B, _N, _E = Wh.shape
Wh1 = torch.matmul(Wh, self.a[:self.out_features, :])
Wh2 = torch.matmul(Wh, self.a[self.out_features:, :])
e = Wh1 + Wh2.permute(0, 2, 1)
return self.leakyrelu(e)
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class GAT(nn.Module):
def __init__(self, in_feat_dim, nhid, out_feat_dim, dropout, alpha,
nheads, batch_num, node_num):
"""Dense version of GAT."""
super(GAT, self).__init__()
self.dropout = dropout
self.batch_num = batch_num
self.node_num = node_num
torch.device('cuda')
self.adj = nn.Parameter(torch.ones(self.batch_num, self.node_num,
self.node_num))
self.attentions = [GraphAttentionLayer(in_feat_dim, nhid, dropout=
dropout, alpha=alpha, batch_num=batch_num, node_num=node_num,
concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
self.out_att = GraphAttentionLayer(nhid * nheads, out_feat_dim,
dropout=dropout, alpha=alpha, batch_num=batch_num, node_num=
node_num, concat=False)
def forward(self, x):
x = F.dropout(x, self.dropout, training=self.training)
x = torch.cat([att(x, self.adj) for att in self.attentions], dim=-1)
x = F.dropout(x, self.dropout, training=self.training)
x = F.elu(self.out_att(x, self.adj))
return F.log_softmax(x, dim=1)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'in_feat_dim': 4, 'nhid': 4, 'out_feat_dim': 4, 'dropout':
0.5, 'alpha': 4, 'nheads': 4, 'batch_num': 4, 'node_num': 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
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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex // 4
x0 = xindex % 4
x2 = xindex // 16
x4 = xindex
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tl.store(out_ptr0 + x4, tmp4, xmask)
@triton.jit
def triton_poi_fused_gt_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
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 = 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 + 4 * x2, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp1 = tl.load(in_ptr1 + 4 * x2, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp2 = tl.load(in_ptr2 + x2, xmask)
tmp3 = tl.load(in_ptr3 + 4 * x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp11 = tl.load(in_ptr1 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp12 = tl.load(in_ptr3 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp19 = tl.load(in_ptr1 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp20 = tl.load(in_ptr3 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp26 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp27 = tl.load(in_ptr1 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp28 = tl.load(in_ptr3 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp34 = tl.load(in_ptr4 + 4 * x2, xmask, eviction_policy='evict_last').to(
tl.int1)
tmp35 = tl.load(in_ptr5 + x2, xmask)
tmp36 = tl.load(in_ptr6 + 4 * x1, xmask, eviction_policy='evict_last')
tmp41 = tl.load(in_ptr4 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp42 = tl.load(in_ptr6 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp48 = tl.load(in_ptr4 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp49 = tl.load(in_ptr6 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp55 = tl.load(in_ptr4 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp56 = tl.load(in_ptr6 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp62 = tl.load(in_ptr7 + 4 * x2, xmask, eviction_policy='evict_last').to(
tl.int1)
tmp63 = tl.load(in_ptr8 + x2, xmask)
tmp64 = tl.load(in_ptr9 + 4 * x1, xmask, eviction_policy='evict_last')
tmp69 = tl.load(in_ptr7 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp70 = tl.load(in_ptr9 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp76 = tl.load(in_ptr7 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp77 = tl.load(in_ptr9 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp83 = tl.load(in_ptr7 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp84 = tl.load(in_ptr9 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp90 = tl.load(in_ptr10 + 4 * x2, xmask, eviction_policy='evict_last').to(
tl.int1)
tmp91 = tl.load(in_ptr11 + x2, xmask)
tmp92 = tl.load(in_ptr12 + 4 * x1, xmask, eviction_policy='evict_last')
tmp97 = tl.load(in_ptr10 + (1 + 4 * x2), xmask, eviction_policy=
'evict_last').to(tl.int1)
tmp98 = tl.load(in_ptr12 + (1 + 4 * x1), xmask, eviction_policy=
'evict_last')
tmp104 = tl.load(in_ptr10 + (2 + 4 * x2), xmask, eviction_policy=
'evict_last').to(tl.int1)
tmp105 = tl.load(in_ptr12 + (2 + 4 * x1), xmask, eviction_policy=
'evict_last')
tmp111 = tl.load(in_ptr10 + (3 + 4 * x2), xmask, eviction_policy=
'evict_last').to(tl.int1)
tmp112 = tl.load(in_ptr12 + (3 + 4 * 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)
tmp13 = tmp2 + tmp12
tmp14 = tmp13 * tmp5
tmp15 = tl.where(tmp11, tmp13, tmp14)
tmp16 = tl.where(tmp10, tmp15, tmp8)
tmp17 = triton_helpers.maximum(tmp9, tmp16)
tmp21 = tmp2 + tmp20
tmp22 = tmp21 * tmp5
tmp23 = tl.where(tmp19, tmp21, tmp22)
tmp24 = tl.where(tmp18, tmp23, tmp8)
tmp25 = triton_helpers.maximum(tmp17, tmp24)
tmp29 = tmp2 + tmp28
tmp30 = tmp29 * tmp5
tmp31 = tl.where(tmp27, tmp29, tmp30)
tmp32 = tl.where(tmp26, tmp31, tmp8)
tmp33 = triton_helpers.maximum(tmp25, tmp32)
tmp37 = tmp35 + tmp36
tmp38 = tmp37 * tmp5
tmp39 = tl.where(tmp34, tmp37, tmp38)
tmp40 = tl.where(tmp0, tmp39, tmp8)
tmp43 = tmp35 + tmp42
tmp44 = tmp43 * tmp5
tmp45 = tl.where(tmp41, tmp43, tmp44)
tmp46 = tl.where(tmp10, tmp45, tmp8)
tmp47 = triton_helpers.maximum(tmp40, tmp46)
tmp50 = tmp35 + tmp49
tmp51 = tmp50 * tmp5
tmp52 = tl.where(tmp48, tmp50, tmp51)
tmp53 = tl.where(tmp18, tmp52, tmp8)
tmp54 = triton_helpers.maximum(tmp47, tmp53)
tmp57 = tmp35 + tmp56
tmp58 = tmp57 * tmp5
tmp59 = tl.where(tmp55, tmp57, tmp58)
tmp60 = tl.where(tmp26, tmp59, tmp8)
tmp61 = triton_helpers.maximum(tmp54, tmp60)
tmp65 = tmp63 + tmp64
tmp66 = tmp65 * tmp5
tmp67 = tl.where(tmp62, tmp65, tmp66)
tmp68 = tl.where(tmp0, tmp67, tmp8)
tmp71 = tmp63 + tmp70
tmp72 = tmp71 * tmp5
tmp73 = tl.where(tmp69, tmp71, tmp72)
tmp74 = tl.where(tmp10, tmp73, tmp8)
tmp75 = triton_helpers.maximum(tmp68, tmp74)
tmp78 = tmp63 + tmp77
tmp79 = tmp78 * tmp5
tmp80 = tl.where(tmp76, tmp78, tmp79)
tmp81 = tl.where(tmp18, tmp80, tmp8)
tmp82 = triton_helpers.maximum(tmp75, tmp81)
tmp85 = tmp63 + tmp84
tmp86 = tmp85 * tmp5
tmp87 = tl.where(tmp83, tmp85, tmp86)
tmp88 = tl.where(tmp26, tmp87, tmp8)
tmp89 = triton_helpers.maximum(tmp82, tmp88)
tmp93 = tmp91 + tmp92
tmp94 = tmp93 * tmp5
tmp95 = tl.where(tmp90, tmp93, tmp94)
tmp96 = tl.where(tmp0, tmp95, tmp8)
tmp99 = tmp91 + tmp98
tmp100 = tmp99 * tmp5
tmp101 = tl.where(tmp97, tmp99, tmp100)
tmp102 = tl.where(tmp10, tmp101, tmp8)
tmp103 = triton_helpers.maximum(tmp96, tmp102)
tmp106 = tmp91 + tmp105
tmp107 = tmp106 * tmp5
tmp108 = tl.where(tmp104, tmp106, tmp107)
tmp109 = tl.where(tmp18, tmp108, tmp8)
tmp110 = triton_helpers.maximum(tmp103, tmp109)
tmp113 = tmp91 + tmp112
tmp114 = tmp113 * tmp5
tmp115 = tl.where(tmp111, tmp113, tmp114)
tmp116 = tl.where(tmp26, tmp115, tmp8)
tmp117 = triton_helpers.maximum(tmp110, tmp116)
tl.store(out_ptr0 + x2, tmp33, xmask)
tl.store(out_ptr1 + x2, tmp61, xmask)
tl.store(out_ptr2 + x2, tmp89, xmask)
tl.store(out_ptr3 + x2, tmp117, 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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x4 = xindex // 4
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + x3, xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp10 = tl.load(in_ptr4 + x4, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr5 + x3, xmask).to(tl.int1)
tmp14 = tl.load(in_ptr6 + x4, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr7 + (x0 + 4 * x2), xmask, eviction_policy=
'evict_last')
tmp20 = tl.load(in_ptr8 + x4, xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr9 + x3, xmask).to(tl.int1)
tmp24 = tl.load(in_ptr10 + x4, xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr11 + (x0 + 4 * x2), xmask, eviction_policy=
'evict_last')
tmp30 = tl.load(in_ptr12 + x4, xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr13 + x3, xmask).to(tl.int1)
tmp34 = tl.load(in_ptr14 + x4, xmask, eviction_policy='evict_last')
tmp35 = tl.load(in_ptr15 + (x0 + 4 * x2), xmask, eviction_policy=
'evict_last')
tmp40 = tl.load(in_ptr16 + x4, 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 + x3, tmp12, xmask)
tl.store(out_ptr1 + x3, tmp22, xmask)
tl.store(out_ptr2 + x3, tmp32, xmask)
tl.store(out_ptr3 + x3, tmp42, xmask)
@triton.jit
def triton_poi_fused__softmax_4(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_cat_5(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 % 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 = 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 + 4 * x2, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp1 = tl.load(in_ptr1 + 4 * x2, xmask, eviction_policy='evict_last').to(tl
.int1)
tmp2 = tl.load(in_ptr2 + x2, xmask)
tmp3 = tl.load(in_ptr3 + 4 * x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp11 = tl.load(in_ptr1 + (1 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp12 = tl.load(in_ptr3 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp19 = tl.load(in_ptr1 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp20 = tl.load(in_ptr3 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp26 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp27 = tl.load(in_ptr1 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
).to(tl.int1)
tmp28 = tl.load(in_ptr3 + (3 + 4 * 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)
tmp13 = tmp2 + tmp12
tmp14 = tmp13 * tmp5
tmp15 = tl.where(tmp11, tmp13, tmp14)
tmp16 = tl.where(tmp10, tmp15, tmp8)
tmp17 = triton_helpers.maximum(tmp9, tmp16)
tmp21 = tmp2 + tmp20
tmp22 = tmp21 * tmp5
tmp23 = tl.where(tmp19, tmp21, tmp22)
tmp24 = tl.where(tmp18, tmp23, tmp8)
tmp25 = triton_helpers.maximum(tmp17, tmp24)
tmp29 = tmp2 + tmp28
tmp30 = tmp29 * tmp5
tmp31 = tl.where(tmp27, tmp29, tmp30)
tmp32 = tl.where(tmp26, tmp31, tmp8)
tmp33 = triton_helpers.maximum(tmp25, tmp32)
tl.store(out_ptr0 + x2, tmp33, 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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x4 = xindex // 4
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask).to(tl.int1)
tmp1 = tl.load(in_ptr1 + x3, xmask).to(tl.int1)
tmp2 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr3 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp10 = tl.load(in_ptr4 + x4, 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 + x3, tmp12, xmask)
@triton.jit
def triton_poi_fused__log_softmax_elu_8(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
x3 = xindex
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp8 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp14 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp28 = tl.load(in_ptr0 + (12 + x0 + 16 * x2), 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 + x3, tmp35, xmask)
@triton.jit
def triton_poi_fused__log_softmax_9(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
x3 = xindex
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (12 + x0 + 16 * 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, 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), (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, (8, 1), (1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (8, 1), (1, 1))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (8, 1), (1, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (8, 1), (1, 1))
assert_size_stride(primals_11, (16, 4), (4, 1))
assert_size_stride(primals_12, (8, 1), (1, 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_1, (16, 4), (4, 1), 0),
primals_3, out=buf0)
del primals_3
buf1 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf0, reinterpret_tensor(primals_4, (4, 1), (1, 1
), 0), out=buf1)
buf2 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf0, reinterpret_tensor(primals_4, (4, 1), (1, 1
), 4), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_leaky_relu_0[grid(64)](buf1, buf2, buf3, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_gt_1[grid(64)](primals_2, buf4, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_2
buf9 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
primals_5, out=buf9)
del primals_5
buf10 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf9, reinterpret_tensor(primals_6, (4, 1), (1, 1
), 0), out=buf10)
buf11 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf9, reinterpret_tensor(primals_6, (4, 1), (1, 1
), 4), out=buf11)
buf12 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(64)](buf10, buf11, buf12, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf17 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
primals_7, out=buf17)
del primals_7
buf18 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf17, reinterpret_tensor(primals_8, (4, 1), (1,
1), 0), out=buf18)
buf19 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf17, reinterpret_tensor(primals_8, (4, 1), (1,
1), 4), out=buf19)
buf20 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(64)](buf18, buf19, buf20, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf25 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
primals_9, out=buf25)
del primals_9
buf26 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf25, reinterpret_tensor(primals_10, (4, 1), (1,
1), 0), out=buf26)
buf27 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf25, reinterpret_tensor(primals_10, (4, 1), (1,
1), 4), out=buf27)
buf28 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(64)](buf26, buf27, buf28, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf5 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf13 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf21 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf29 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused__softmax_add_leaky_relu_mul_where_2[grid(16)](buf4,
buf3, buf1, buf2, buf12, buf10, buf11, buf20, buf18, buf19,
buf28, buf26, buf27, buf5, buf13, buf21, buf29, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf22 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf30 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_add_leaky_relu_mul_where_3[grid(64)](buf4,
buf3, buf1, buf2, buf5, buf12, buf10, buf11, buf13, buf20,
buf18, buf19, buf21, buf28, buf26, buf27, buf29, buf6, buf14,
buf22, buf30, 64, XBLOCK=64, 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), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(64)](buf6, buf7, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf8 = buf6
del buf6
extern_kernels.bmm(buf7, reinterpret_tensor(buf0, (4, 4, 4), (16, 4,
1), 0), out=buf8)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(64)](buf14, buf15, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf16 = buf14
del buf14
extern_kernels.bmm(buf15, reinterpret_tensor(buf9, (4, 4, 4), (16,
4, 1), 0), out=buf16)
buf23 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(64)](buf22, buf23, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf24 = buf22
del buf22
extern_kernels.bmm(buf23, reinterpret_tensor(buf17, (4, 4, 4), (16,
4, 1), 0), out=buf24)
buf31 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(64)](buf30, buf31, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf32 = buf30
del buf30
extern_kernels.bmm(buf31, reinterpret_tensor(buf25, (4, 4, 4), (16,
4, 1), 0), out=buf32)
buf33 = empty_strided_cuda((4, 4, 16), (64, 16, 1), torch.float32)
triton_poi_fused_cat_5[grid(256)](buf8, buf16, buf24, buf32, buf33,
256, XBLOCK=256, num_warps=4, num_stages=1)
buf34 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf33, (16, 16), (16, 1), 0),
primals_11, out=buf34)
buf35 = reinterpret_tensor(buf5, (16, 1), (1, 1), 0)
del buf5
extern_kernels.mm(buf34, reinterpret_tensor(primals_12, (4, 1), (1,
1), 0), out=buf35)
buf36 = reinterpret_tensor(buf29, (16, 1), (1, 1), 0)
del buf29
extern_kernels.mm(buf34, reinterpret_tensor(primals_12, (4, 1), (1,
1), 4), out=buf36)
buf37 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_leaky_relu_0[grid(64)](buf35, buf36, buf37, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf38 = reinterpret_tensor(buf27, (4, 4, 1), (4, 1, 16), 0)
del buf27
triton_poi_fused__softmax_add_leaky_relu_mul_where_6[grid(16)](buf4,
buf37, buf35, buf36, buf38, 16, XBLOCK=16, num_warps=1,
num_stages=1)
buf39 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_add_leaky_relu_mul_where_7[grid(64)](buf4,
buf37, buf35, buf36, buf38, buf39, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf35
del buf36
del buf38
buf40 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_4[grid(64)](buf39, buf40, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf41 = buf39
del buf39
extern_kernels.bmm(buf40, reinterpret_tensor(buf34, (4, 4, 4), (16,
4, 1), 0), out=buf41)
buf42 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__log_softmax_elu_8[grid(64)](buf41, buf42, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf43 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__log_softmax_9[grid(64)](buf42, buf43, 64, XBLOCK=
64, num_warps=1, num_stages=1)
del buf42
return (buf43, buf0, buf3, buf4, buf7, buf8, buf9, buf12, buf15, buf16,
buf17, buf20, buf23, buf24, buf25, buf28, buf31, buf32, buf34,
buf37, buf40, buf41, buf43, reinterpret_tensor(primals_12, (1, 4),
(1, 1), 4), reinterpret_tensor(primals_12, (1, 4), (1, 1), 0),
reinterpret_tensor(buf33, (16, 16), (1, 16), 0), reinterpret_tensor
(primals_11, (4, 16), (1, 4), 0), reinterpret_tensor(primals_10, (1,
4), (1, 1), 4), reinterpret_tensor(primals_10, (1, 4), (1, 1), 0),
reinterpret_tensor(primals_1, (4, 16), (1, 4), 0),
reinterpret_tensor(primals_8, (1, 4), (1, 1), 4),
reinterpret_tensor(primals_8, (1, 4), (1, 1), 0),
reinterpret_tensor(primals_6, (1, 4), (1, 1), 4),
reinterpret_tensor(primals_6, (1, 4), (1, 1), 0),
reinterpret_tensor(primals_4, (1, 4), (1, 1), 4),
reinterpret_tensor(primals_4, (1, 4), (1, 1), 0))
class GraphAttentionLayer(nn.Module):
"""
https://github.com/Diego999/pyGAT/blob/master/layers.py
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, batch_num,
node_num, concat=True):
super(GraphAttentionLayer, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
self.batch_num = batch_num
self.node_num = node_num
torch.device('cuda')
self.W = nn.Parameter(torch.empty(size=(in_features, out_features)))
nn.init.xavier_uniform_(self.W.data, gain=1.414)
self.a = nn.Parameter(torch.empty(size=(2 * out_features, 1)))
nn.init.xavier_uniform_(self.a.data, gain=1.414)
self.leakyrelu = nn.LeakyReLU(self.alpha)
def forward(self, h, adj):
"""
:param h: (batch_size, number_nodes, in_features)
:param adj: (batch_size, number_nodes, number_nodes)
:return: (batch_size, number_nodes, out_features)
"""
Wh = torch.matmul(h, self.W)
e = self.prepare_batch(Wh)
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, Wh)
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def prepare_batch(self, Wh):
"""
with batch training
:param Wh: (batch_size, number_nodes, out_features)
:return:
"""
_B, _N, _E = Wh.shape
Wh1 = torch.matmul(Wh, self.a[:self.out_features, :])
Wh2 = torch.matmul(Wh, self.a[self.out_features:, :])
e = Wh1 + Wh2.permute(0, 2, 1)
return self.leakyrelu(e)
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class GATNew(nn.Module):
def __init__(self, in_feat_dim, nhid, out_feat_dim, dropout, alpha,
nheads, batch_num, node_num):
"""Dense version of GAT."""
super(GATNew, self).__init__()
self.dropout = dropout
self.batch_num = batch_num
self.node_num = node_num
torch.device('cuda')
self.adj = nn.Parameter(torch.ones(self.batch_num, self.node_num,
self.node_num))
self.attentions = [GraphAttentionLayer(in_feat_dim, nhid, dropout=
dropout, alpha=alpha, batch_num=batch_num, node_num=node_num,
concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
self.out_att = GraphAttentionLayer(nhid * nheads, out_feat_dim,
dropout=dropout, alpha=alpha, batch_num=batch_num, node_num=
node_num, concat=False)
def forward(self, input_0):
primals_1 = self.adj
primals_3 = self.attention_0.W
primals_4 = self.attention_0.a
primals_5 = self.attention_1.W
primals_6 = self.attention_1.a
primals_7 = self.attention_2.W
primals_8 = self.attention_2.a
primals_9 = self.attention_3.W
primals_10 = self.attention_3.a
primals_11 = self.out_att.W
primals_12 = self.out_att.a
primals_2 = 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])
return output[0]
| bpilseo/VLP_GAT | GAT | false | 1,623 | [
"MIT"
] | 0 | ca8a2594036ab8fe4a180e5ced87f59f8984e54f | https://github.com/bpilseo/VLP_GAT/tree/ca8a2594036ab8fe4a180e5ced87f59f8984e54f | import torch
from torch import nn
import torch.nn.functional as F
class GraphAttentionLayer(nn.Module):
"""
https://github.com/Diego999/pyGAT/blob/master/layers.py
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, batch_num,
node_num, concat=True):
super().__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
self.batch_num = batch_num
self.node_num = node_num
torch.device('cuda')
self.W = nn.Parameter(torch.empty(size=(in_features, out_features)))
nn.init.xavier_uniform_(self.W.data, gain=1.414)
self.a = nn.Parameter(torch.empty(size=(2 * out_features, 1)))
nn.init.xavier_uniform_(self.a.data, gain=1.414)
self.leakyrelu = nn.LeakyReLU(self.alpha)
def forward(self, h, adj):
"""
:param h: (batch_size, number_nodes, in_features)
:param adj: (batch_size, number_nodes, number_nodes)
:return: (batch_size, number_nodes, out_features)
"""
Wh = torch.matmul(h, self.W)
e = self.prepare_batch(Wh)
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, Wh)
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def prepare_batch(self, Wh):
"""
with batch training
:param Wh: (batch_size, number_nodes, out_features)
:return:
"""
_B, _N, _E = Wh.shape
Wh1 = torch.matmul(Wh, self.a[:self.out_features, :])
Wh2 = torch.matmul(Wh, self.a[self.out_features:, :])
e = Wh1 + Wh2.permute(0, 2, 1)
return self.leakyrelu(e)
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class Model(nn.Module):
def __init__(self, in_feat_dim, nhid, out_feat_dim, dropout, alpha,
nheads, batch_num, node_num):
"""Dense version of GAT."""
super().__init__()
self.dropout = dropout
self.batch_num = batch_num
self.node_num = node_num
torch.device('cuda')
self.adj = nn.Parameter(torch.ones(self.batch_num, self.node_num,
self.node_num))
self.attentions = [GraphAttentionLayer(in_feat_dim, nhid, dropout=
dropout, alpha=alpha, batch_num=batch_num, node_num=node_num,
concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
self.out_att = GraphAttentionLayer(nhid * nheads, out_feat_dim,
dropout=dropout, alpha=alpha, batch_num=batch_num, node_num=
node_num, concat=False)
def forward(self, x):
x = F.dropout(x, self.dropout, training=self.training)
x = torch.cat([att(x, self.adj) for att in self.attentions], dim=-1)
x = F.dropout(x, self.dropout, training=self.training)
x = F.elu(self.out_att(x, self.adj))
return F.log_softmax(x, dim=1)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'in_feat_dim': 4, 'nhid': 4, 'out_feat_dim': 4, 'dropout':
0.5, 'alpha': 4, 'nheads': 4, 'batch_num': 4, 'node_num': 4}]
|
RemoveChannelMeanStd | # 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_6/inductor_cache/dq/cdqspqrzle4hslc7j2642l3po3ngbns2rlbz3nf7iatgjyfmijde.py
# Topologically Sorted Source Nodes: [mean, sub, std, truediv], Original ATen: [aten.mean, aten.sub, aten.std, aten.div]
# Source node to ATen node mapping:
# mean => mean
# std => var
# sub => sub
# truediv => div
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view, [2]), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %view_1), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%view, [2]), kwargs = {correction: 1.0})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %view_2), kwargs = {})
triton_per_fused_div_mean_std_sub_0 = async_compile.triton('triton_per_fused_div_mean_std_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.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_div_mean_std_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, '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_div_mean_std_sub_0(in_ptr0, out_ptr2, 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]
tmp6 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp8 = tl.where(xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tmp10 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp11 = tmp10.to(tl.float32)
tmp12 = tmp9 / tmp11
tmp13 = tmp1 - tmp12
tmp14 = tmp13 * tmp13
tmp15 = tl.broadcast_to(tmp14, [XBLOCK, RBLOCK])
tmp17 = tl.where(xmask, tmp15, 0)
tmp18 = tl.sum(tmp17, 1)[:, None]
tmp19 = 16.0
tmp20 = tmp4 / tmp19
tmp21 = tmp0 - tmp20
tmp22 = 15.0
tmp23 = tmp18 / tmp22
tmp24 = libdevice.sqrt(tmp23)
tmp25 = tmp21 / tmp24
tl.store(out_ptr2 + (r1 + (16*x0)), tmp25, 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)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean, sub, std, truediv], Original ATen: [aten.mean, aten.sub, aten.std, aten.div]
stream0 = get_raw_stream(0)
triton_per_fused_div_mean_std_sub_0.run(arg0_1, buf4, 16, 16, grid=grid(16), stream=stream0)
del arg0_1
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)
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
class RemoveChannelMeanStd(torch.nn.Module):
def forward(self, x):
x2 = x.view(x.size(0), x.size(1), -1)
mean = x2.mean(dim=2).view(x.size(0), x.size(1), 1, 1)
std = x2.std(dim=2).view(x.size(0), x.size(1), 1, 1)
return (x - mean) / std
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.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_per_fused_div_mean_std_sub_0(in_ptr0, out_ptr2, 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]
tmp6 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp8 = tl.where(xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tmp10 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp11 = tmp10.to(tl.float32)
tmp12 = tmp9 / tmp11
tmp13 = tmp1 - tmp12
tmp14 = tmp13 * tmp13
tmp15 = tl.broadcast_to(tmp14, [XBLOCK, RBLOCK])
tmp17 = tl.where(xmask, tmp15, 0)
tmp18 = tl.sum(tmp17, 1)[:, None]
tmp19 = 16.0
tmp20 = tmp4 / tmp19
tmp21 = tmp0 - tmp20
tmp22 = 15.0
tmp23 = tmp18 / tmp22
tmp24 = libdevice.sqrt(tmp23)
tmp25 = tmp21 / tmp24
tl.store(out_ptr2 + (r1 + 16 * x0), tmp25, 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)
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_div_mean_std_sub_0[grid(16)](arg0_1, buf4, 16, 16,
XBLOCK=8, num_warps=2, num_stages=1)
del arg0_1
return buf4,
class RemoveChannelMeanStdNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| cadurosar/graph_kd_dense_cifar100 | RemoveChannelMeanStd | false | 1,624 | [
"MIT"
] | 0 | 84054ab4f8f61c9db3460993661ba7bf1d951b36 | https://github.com/cadurosar/graph_kd_dense_cifar100/tree/84054ab4f8f61c9db3460993661ba7bf1d951b36 | import torch
class Model(torch.nn.Module):
def forward(self, x):
x2 = x.view(x.size(0), x.size(1), -1)
mean = x2.mean(dim=2).view(x.size(0), x.size(1), 1, 1)
std = x2.std(dim=2).view(x.size(0), x.size(1), 1, 1)
return (x - mean) / std
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
BERTEmbedding2 | # 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_6/inductor_cache/zv/czv5weh4kag2zzt7vdzpounxpmvyi3xvjezcwjzscyisfikusgm5.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %primals_2), 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': [], '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, 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
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2), 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, (1, 4, 4), (16, 4, 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, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(primals_1, primals_2, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_2
return (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((1, 4, 4), (16, 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)
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
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class LearnedPositionalEmbedding2(nn.Module):
def __init__(self, d_model, max_len=512):
super().__init__()
pe = torch.zeros(max_len, d_model).float()
pe.require_grad = True
pe = pe.unsqueeze(0)
self.pe = nn.Parameter(pe)
torch.nn.init.normal_(self.pe, std=0.02)
def forward(self, x):
return self.pe[:, :x.size(1)]
class BERTEmbedding2(nn.Module):
"""
BERT Embedding which is consisted with under features
1. PositionalEmbedding : adding positional information using sin, cos
sum of all these features are output of BERTEmbedding
"""
def __init__(self, input_dim, max_len, dropout=0.1):
"""
:param vocab_size: total vocab size
:param embed_size: embedding size of token embedding
:param dropout: dropout rate
"""
super().__init__()
self.learnedPosition = LearnedPositionalEmbedding2(d_model=
input_dim, max_len=max_len)
self.dropout = nn.Dropout(p=dropout)
def forward(self, sequence):
x = self.learnedPosition(sequence) + sequence
return self.dropout(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'max_len': 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
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
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, 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
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (1, 4, 4), (16, 4, 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, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](primals_1, primals_2, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_1
del primals_2
return buf0,
class LearnedPositionalEmbedding2(nn.Module):
def __init__(self, d_model, max_len=512):
super().__init__()
pe = torch.zeros(max_len, d_model).float()
pe.require_grad = True
pe = pe.unsqueeze(0)
self.pe = nn.Parameter(pe)
torch.nn.init.normal_(self.pe, std=0.02)
def forward(self, x):
return self.pe[:, :x.size(1)]
class BERTEmbedding2New(nn.Module):
"""
BERT Embedding which is consisted with under features
1. PositionalEmbedding : adding positional information using sin, cos
sum of all these features are output of BERTEmbedding
"""
def __init__(self, input_dim, max_len, dropout=0.1):
"""
:param vocab_size: total vocab size
:param embed_size: embedding size of token embedding
:param dropout: dropout rate
"""
super().__init__()
self.learnedPosition = LearnedPositionalEmbedding2(d_model=
input_dim, max_len=max_len)
self.dropout = nn.Dropout(p=dropout)
def forward(self, input_0):
primals_1 = self.learnedPosition.pe
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
| byeongjokim/LateTemporalModeling3DCNN_for_sign | BERTEmbedding2 | false | 1,625 | [
"MIT"
] | 0 | e3a802fcf91dc3930aea782464ee34d9b747d3ab | https://github.com/byeongjokim/LateTemporalModeling3DCNN_for_sign/tree/e3a802fcf91dc3930aea782464ee34d9b747d3ab | import torch
import torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class LearnedPositionalEmbedding2(nn.Module):
def __init__(self, d_model, max_len=512):
super().__init__()
pe = torch.zeros(max_len, d_model).float()
pe.require_grad = True
pe = pe.unsqueeze(0)
self.pe = nn.Parameter(pe)
torch.nn.init.normal_(self.pe, std=0.02)
def forward(self, x):
return self.pe[:, :x.size(1)]
class Model(nn.Module):
"""
BERT Embedding which is consisted with under features
1. PositionalEmbedding : adding positional information using sin, cos
sum of all these features are output of BERTEmbedding
"""
def __init__(self, input_dim, max_len, dropout=0.1):
"""
:param vocab_size: total vocab size
:param embed_size: embedding size of token embedding
:param dropout: dropout rate
"""
super().__init__()
self.learnedPosition = LearnedPositionalEmbedding2(d_model=
input_dim, max_len=max_len)
self.dropout = nn.Dropout(p=dropout)
def forward(self, sequence):
x = self.learnedPosition(sequence) + sequence
return self.dropout(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
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_6/inductor_cache/v7/cv7humnywkkqhrumbeetegqlkretdwtkj5pcanrbgxrolupvobzt.py
# Topologically Sorted Source Nodes: [mul, pow_1, mul_1, add, mul_2, tanh, add_1, mul_3], Original ATen: [aten.mul, aten.pow, aten.add, aten.tanh]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# mul_3 => mul_3
# pow_1 => pow_1
# tanh => tanh
# 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=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add_1), kwargs = {})
triton_poi_fused_add_mul_pow_tanh_0 = async_compile.triton('triton_poi_fused_add_mul_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=[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_mul_pow_tanh_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_mul_pow_tanh_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 = 0.5
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
tl.store(out_ptr0 + (x0), 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, 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: [linear], 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, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, pow_1, mul_1, add, mul_2, tanh, add_1, mul_3], Original ATen: [aten.mul, aten.pow, aten.add, aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_pow_tanh_0.run(buf0, buf1, 256, grid=grid(256), 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, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 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, 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, 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 math
import torch
import torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class GELU(nn.Module):
"""
Paper Section 3.4, last paragraph notice that BERT used the GELU instead of RELU
"""
def forward(self, x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x +
0.044715 * torch.pow(x, 3))))
class PositionwiseFeedForward(nn.Module):
"""Implements FFN equation."""
def __init__(self, d_model, d_ff, dropout=0.1):
super(PositionwiseFeedForward, self).__init__()
self.w_1 = nn.Linear(d_model, d_ff)
self.w_2 = nn.Linear(d_ff, d_model)
self.dropout = nn.Dropout(dropout)
self.activation = GELU()
def forward(self, x):
return self.w_2(self.dropout(self.activation(self.w_1(x))))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_model': 4, 'd_ff': 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
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
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_pow_tanh_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 = 0.5
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
tl.store(out_ptr0 + x0, 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, 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.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, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mul_pow_tanh_0[grid(256)](buf0, buf1, 256,
XBLOCK=128, 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, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 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, 4), (4, 1), 0), primals_4
class GELU(nn.Module):
"""
Paper Section 3.4, last paragraph notice that BERT used the GELU instead of RELU
"""
def forward(self, x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x +
0.044715 * torch.pow(x, 3))))
class PositionwiseFeedForwardNew(nn.Module):
"""Implements FFN equation."""
def __init__(self, d_model, d_ff, dropout=0.1):
super(PositionwiseFeedForwardNew, self).__init__()
self.w_1 = nn.Linear(d_model, d_ff)
self.w_2 = nn.Linear(d_ff, d_model)
self.dropout = nn.Dropout(dropout)
self.activation = GELU()
def forward(self, input_0):
primals_1 = self.w_1.weight
primals_2 = self.w_1.bias
primals_4 = self.w_2.weight
primals_5 = self.w_2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| byeongjokim/LateTemporalModeling3DCNN_for_sign | PositionwiseFeedForward | false | 1,626 | [
"MIT"
] | 0 | e3a802fcf91dc3930aea782464ee34d9b747d3ab | https://github.com/byeongjokim/LateTemporalModeling3DCNN_for_sign/tree/e3a802fcf91dc3930aea782464ee34d9b747d3ab | import math
import torch
import torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class GELU(nn.Module):
"""
Paper Section 3.4, last paragraph notice that BERT used the GELU instead of RELU
"""
def forward(self, x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x +
0.044715 * torch.pow(x, 3))))
class Model(nn.Module):
"""Implements FFN equation."""
def __init__(self, d_model, d_ff, dropout=0.1):
super().__init__()
self.w_1 = nn.Linear(d_model, d_ff)
self.w_2 = nn.Linear(d_ff, d_model)
self.dropout = nn.Dropout(dropout)
self.activation = GELU()
def forward(self, x):
return self.w_2(self.dropout(self.activation(self.w_1(x))))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Evidential_layer | # 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_6/inductor_cache/pd/cpdu37l3bj63bjibgjk2ueagf7o3e26iukuvw6axiaa2bjb2e6op.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_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 = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 8
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, (8, 4), (4, 1))
assert_size_stride(primals_2, (8, ), (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, 8), (8, 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, 8), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 8), (128, 32, 8, 1), 0); del buf0 # reuse
buf2 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf2, 512, grid=grid(512), stream=stream0)
del primals_2
return (buf1, reinterpret_tensor(primals_3, (64, 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((8, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((8, ), (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 Evidential_layer(nn.Module):
def __init__(self, in_dim, num_classes):
super(Evidential_layer, self).__init__()
self.num_classes = num_classes
self.fc1 = nn.Linear(in_dim, 2 * self.num_classes)
self.relu = torch.nn.ReLU()
def forward(self, x):
x = self.fc1(x)
return self.relu(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_dim': 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 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 = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 8
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, (8, 4), (4, 1))
assert_size_stride(primals_2, (8,), (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, 8), (8, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 8), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 8), (128, 32, 8, 1), 0)
del buf0
buf2 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(512)](buf1,
primals_2, buf2, 512, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
return buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf2
class Evidential_layerNew(nn.Module):
def __init__(self, in_dim, num_classes):
super(Evidential_layerNew, self).__init__()
self.num_classes = num_classes
self.fc1 = nn.Linear(in_dim, 2 * self.num_classes)
self.relu = torch.nn.ReLU()
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| caisr-hh/DEED | Evidential_layer | false | 1,627 | [
"MIT"
] | 0 | 2a9edb1df31d99c1e8da177dec696d7c90c2e7de | https://github.com/caisr-hh/DEED/tree/2a9edb1df31d99c1e8da177dec696d7c90c2e7de | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, in_dim, num_classes):
super().__init__()
self.num_classes = num_classes
self.fc1 = nn.Linear(in_dim, 2 * self.num_classes)
self.relu = torch.nn.ReLU()
def forward(self, x):
x = self.fc1(x)
return self.relu(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
NextSentencePrediction | # 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_6/inductor_cache/yy/cyya3js6wt64vdji3sfisvrqyfvqxwkwqq5mzg5bqjl2crzjs4t3.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 = (%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_6/inductor_cache/t6/ct6t4t4f6df2rswen66qmskrag4dcnuaoxkebk4pssna63yxl3v3.py
# Topologically Sorted Source Nodes: [linear, log_softmax], Original ATen: [aten.add, aten._log_softmax]
# Source node to ATen node mapping:
# linear => add
# log_softmax => amax, sub
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_3), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add, [-1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %amax), kwargs = {})
triton_poi_fused__log_softmax_add_1 = async_compile.triton('triton_poi_fused__log_softmax_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=[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__log_softmax_add_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_add_1(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
x2 = xindex
x0 = xindex % 2
x1 = (xindex // 2)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2*x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (1 + (2*x1)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (1))
tmp9 = tl.broadcast_to(tmp8, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp6 = tmp3 + tmp5
tmp10 = tmp7 + tmp9
tmp11 = triton_helpers.maximum(tmp6, tmp10)
tmp12 = tmp2 - tmp11
tl.store(out_ptr0 + (x2), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/jt/cjtlgs566wh4nqwf6wvcsracaovzta35wbm5vllgfuavp5ebkiq4.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_2 = async_compile.triton('triton_poi_fused__log_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=[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, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_2', '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__log_softmax_2(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
x2 = xindex
x1 = (xindex // 2)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (2*x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (2*x1)), xmask, eviction_policy='evict_last')
tmp2 = tl_math.exp(tmp1)
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp2 + tmp4
tmp6 = tl_math.log(tmp5)
tmp7 = tmp0 - tmp6
tl.store(out_ptr0 + (x2), tmp7, 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, (2, 4), (4, 1))
assert_size_stride(primals_3, (2, ), (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: [linear], 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, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 2), (1, 4), 0), out=buf1)
del primals_2
buf2 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear, log_softmax], Original ATen: [aten.add, aten._log_softmax]
triton_poi_fused__log_softmax_add_1.run(buf1, primals_3, buf2, 32, grid=grid(32), stream=stream0)
del primals_3
buf3 = reinterpret_tensor(buf1, (4, 4, 2), (8, 2, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_2.run(buf2, buf3, 32, grid=grid(32), stream=stream0)
del buf2
return (buf3, 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((2, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((2, ), (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
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class NextSentencePrediction(nn.Module):
"""
2-class classification model : is_next, is_not_next
"""
def __init__(self, hidden):
"""
:param hidden: BERT model output size
"""
super().__init__()
self.linear = nn.Linear(hidden, 2)
self.softmax = nn.LogSoftmax(dim=-1)
def forward(self, x):
return self.softmax(self.linear(x[:, 0]))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'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 import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
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__log_softmax_add_1(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
x2 = xindex
x0 = xindex % 2
x1 = xindex // 2
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + 2 * x1, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (1 + 2 * x1), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + 1)
tmp9 = tl.broadcast_to(tmp8, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp6 = tmp3 + tmp5
tmp10 = tmp7 + tmp9
tmp11 = triton_helpers.maximum(tmp6, tmp10)
tmp12 = tmp2 - tmp11
tl.store(out_ptr0 + x2, tmp12, xmask)
@triton.jit
def triton_poi_fused__log_softmax_2(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
x2 = xindex
x1 = xindex // 2
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 2 * x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 2 * x1), xmask, eviction_policy='evict_last')
tmp2 = tl_math.exp(tmp1)
tmp4 = tl_math.exp(tmp3)
tmp5 = tmp2 + tmp4
tmp6 = tl_math.log(tmp5)
tmp7 = tmp0 - tmp6
tl.store(out_ptr0 + x2, tmp7, 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, (2, 4), (4, 1))
assert_size_stride(primals_3, (2,), (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, 2), (2, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 2), (1, 4), 0), out=buf1)
del primals_2
buf2 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
triton_poi_fused__log_softmax_add_1[grid(32)](buf1, primals_3, buf2,
32, XBLOCK=32, num_warps=1, num_stages=1)
del primals_3
buf3 = reinterpret_tensor(buf1, (4, 4, 2), (8, 2, 1), 0)
del buf1
triton_poi_fused__log_softmax_2[grid(32)](buf2, buf3, 32, XBLOCK=32,
num_warps=1, num_stages=1)
del buf2
return buf3, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), buf3
class NextSentencePredictionNew(nn.Module):
"""
2-class classification model : is_next, is_not_next
"""
def __init__(self, hidden):
"""
:param hidden: BERT model output size
"""
super().__init__()
self.linear = nn.Linear(hidden, 2)
self.softmax = nn.LogSoftmax(dim=-1)
def forward(self, input_0):
primals_2 = self.linear.weight
primals_3 = self.linear.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| byeongjokim/LateTemporalModeling3DCNN_for_sign | NextSentencePrediction | false | 1,628 | [
"MIT"
] | 0 | e3a802fcf91dc3930aea782464ee34d9b747d3ab | https://github.com/byeongjokim/LateTemporalModeling3DCNN_for_sign/tree/e3a802fcf91dc3930aea782464ee34d9b747d3ab | import torch
import torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class Model(nn.Module):
"""
2-class classification model : is_next, is_not_next
"""
def __init__(self, hidden):
"""
:param hidden: BERT model output size
"""
super().__init__()
self.linear = nn.Linear(hidden, 2)
self.softmax = nn.LogSoftmax(dim=-1)
def forward(self, x):
return self.softmax(self.linear(x[:, 0]))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
BatchMeanKLDivWithLogSoftmax | # 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_6/inductor_cache/r4/cr456wn6hoprbjcumpkgojrwlvg6hreubmiilb647gjqms63qtu4.py
# Topologically Sorted Source Nodes: [mul, mul_1, sub, sum_1], Original ATen: [aten.mul, aten.sub, aten.sum]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# sub => sub
# sum_1 => sum_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg2_1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %mul_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%sub, [1]), kwargs = {})
triton_poi_fused_mul_sub_sum_0 = async_compile.triton('triton_poi_fused_mul_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.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_mul_sub_sum_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_mul_sub_sum_0(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 % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (64*x1)), xmask)
tmp3 = tl.load(in_ptr2 + (x0 + (64*x1)), xmask)
tmp6 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp7 = tl.load(in_ptr1 + (16 + x0 + (64*x1)), xmask)
tmp9 = tl.load(in_ptr2 + (16 + x0 + (64*x1)), xmask)
tmp13 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp14 = tl.load(in_ptr1 + (32 + x0 + (64*x1)), xmask)
tmp16 = tl.load(in_ptr2 + (32 + x0 + (64*x1)), xmask)
tmp20 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp21 = tl.load(in_ptr1 + (48 + x0 + (64*x1)), xmask)
tmp23 = tl.load(in_ptr2 + (48 + x0 + (64*x1)), xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp0 * tmp3
tmp5 = tmp2 - tmp4
tmp8 = tmp6 * tmp7
tmp10 = tmp6 * tmp9
tmp11 = tmp8 - tmp10
tmp12 = tmp5 + tmp11
tmp15 = tmp13 * tmp14
tmp17 = tmp13 * tmp16
tmp18 = tmp15 - tmp17
tmp19 = tmp12 + tmp18
tmp22 = tmp20 * tmp21
tmp24 = tmp20 * tmp23
tmp25 = tmp22 - tmp24
tmp26 = tmp19 + tmp25
tl.store(out_ptr0 + (x2), tmp26, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/oc/cocbcl7fhmph3ec6vgzy45b4wkrfo4u7q6s5oftx57ok5tdcbqve.py
# Topologically Sorted Source Nodes: [mean], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# mean => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%sum_1, [0]), kwargs = {})
triton_poi_fused_mean_1 = async_compile.triton('triton_poi_fused_mean_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_mean_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_mean_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 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
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, arg1_1, arg2_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))
assert_size_stride(arg2_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: [mul, mul_1, sub, sum_1], Original ATen: [aten.mul, aten.sub, aten.sum]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_sub_sum_0.run(arg0_1, arg1_1, arg2_1, buf0, 64, grid=grid(64), stream=stream0)
del arg0_1
del arg1_1
del arg2_1
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean], Original ATen: [aten.mean]
triton_poi_fused_mean_1.run(buf0, buf1, 16, grid=grid(16), stream=stream0)
del buf0
return (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((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg2_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1, arg2_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 BatchMeanKLDivWithLogSoftmax(nn.Module):
def forward(self, p, log_q, log_p):
return (p * log_p - p * log_q).sum(dim=1).mean(dim=0)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), 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
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_sub_sum_0(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 % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr2 + (x0 + 64 * x1), xmask)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp7 = tl.load(in_ptr1 + (16 + x0 + 64 * x1), xmask)
tmp9 = tl.load(in_ptr2 + (16 + x0 + 64 * x1), xmask)
tmp13 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp14 = tl.load(in_ptr1 + (32 + x0 + 64 * x1), xmask)
tmp16 = tl.load(in_ptr2 + (32 + x0 + 64 * x1), xmask)
tmp20 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp21 = tl.load(in_ptr1 + (48 + x0 + 64 * x1), xmask)
tmp23 = tl.load(in_ptr2 + (48 + x0 + 64 * x1), xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp0 * tmp3
tmp5 = tmp2 - tmp4
tmp8 = tmp6 * tmp7
tmp10 = tmp6 * tmp9
tmp11 = tmp8 - tmp10
tmp12 = tmp5 + tmp11
tmp15 = tmp13 * tmp14
tmp17 = tmp13 * tmp16
tmp18 = tmp15 - tmp17
tmp19 = tmp12 + tmp18
tmp22 = tmp20 * tmp21
tmp24 = tmp20 * tmp23
tmp25 = tmp22 - tmp24
tmp26 = tmp19 + tmp25
tl.store(out_ptr0 + x2, tmp26, xmask)
@triton.jit
def triton_poi_fused_mean_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 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + x0, tmp8, xmask)
def call(args):
arg0_1, arg1_1, arg2_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))
assert_size_stride(arg2_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_mul_sub_sum_0[grid(64)](arg0_1, arg1_1, arg2_1,
buf0, 64, XBLOCK=64, num_warps=1, num_stages=1)
del arg0_1
del arg1_1
del arg2_1
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_mean_1[grid(16)](buf0, buf1, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf0
return buf1,
class BatchMeanKLDivWithLogSoftmaxNew(nn.Module):
def forward(self, input_0, input_1, input_2):
arg0_1 = input_0
arg1_1 = input_1
arg2_1 = input_2
output = call([arg0_1, arg1_1, arg2_1])
return output[0]
| cadurosar/graph_kd_dense_cifar100 | BatchMeanKLDivWithLogSoftmax | false | 1,629 | [
"MIT"
] | 0 | 84054ab4f8f61c9db3460993661ba7bf1d951b36 | https://github.com/cadurosar/graph_kd_dense_cifar100/tree/84054ab4f8f61c9db3460993661ba7bf1d951b36 | import torch
import torch.nn as nn
class Model(nn.Module):
def forward(self, p, log_q, log_p):
return (p * log_p - p * log_q).sum(dim=1).mean(dim=0)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4]), torch.rand(
[4, 4, 4, 4])]
def get_init_inputs():
return []
|
BWCEWLoss | # 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_6/inductor_cache/ls/cls2tbcj37wpwheh2vwphvrxjnksqhrpy5c3ixm3i7qzrhg4pqc6.py
# Topologically Sorted Source Nodes: [train_loss, train_mean_loss], Original ATen: [aten.binary_cross_entropy_with_logits, aten.mean]
# Source node to ATen node mapping:
# train_loss => abs_1, exp, full_default, log1p, mean, minimum, mul, neg, sub, sub_1, sub_2
# train_mean_loss => mean_1
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg0_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %arg1_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), 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 = (%full_default, %arg1_1), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%arg1_1,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %sub_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sub_2,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mean,), kwargs = {})
triton_per_fused_binary_cross_entropy_with_logits_mean_0 = async_compile.triton('triton_per_fused_binary_cross_entropy_with_logits_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=[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_binary_cross_entropy_with_logits_mean_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_binary_cross_entropy_with_logits_mean_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)
tmp3 = tl.load(in_ptr1 + (r0), None)
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp2 * tmp3
tmp5 = 0.0
tmp6 = triton_helpers.minimum(tmp5, tmp3)
tmp7 = tl_math.abs(tmp3)
tmp8 = -tmp7
tmp9 = tl_math.exp(tmp8)
tmp10 = libdevice.log1p(tmp9)
tmp11 = tmp6 - tmp10
tmp12 = tmp4 - tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 256.0
tmp17 = tmp15 / tmp16
tmp18 = tmp17 / tmp1
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp18, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [train_loss, train_mean_loss], Original ATen: [aten.binary_cross_entropy_with_logits, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_mean_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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
from torch import Tensor
from typing import Optional
from torch import nn
class BWCEWLoss(nn.Module):
""" Binary weighted cross entropy loss. """
def __init__(self, positive_class_weight: 'Optional[Tensor]'=None,
robust_lambda: 'int'=0, confidence_penalty: 'int'=0, **kwargs):
super().__init__()
self.loss_fn = nn.BCEWithLogitsLoss(pos_weight=
positive_class_weight, **kwargs)
self.robust_lambda = robust_lambda
self.confidence_penalty = confidence_penalty
def forward(self, preds: 'torch.Tensor', target: 'torch.Tensor'):
train_loss = self.loss_fn(preds, target.float())
if self.robust_lambda > 0:
train_loss = (1 - self.robust_lambda
) * train_loss + self.robust_lambda / 2
train_mean_loss = torch.mean(train_loss)
if self.confidence_penalty > 0:
probabilities = torch.sigmoid(preds)
mean_penalty = utils.mean_confidence_penalty(probabilities, 2)
train_mean_loss += self.confidence_penalty * mean_penalty
return train_mean_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 libdevice, math as tl_math
from torch import Tensor
from typing import Optional
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_per_fused_binary_cross_entropy_with_logits_mean_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)
tmp3 = tl.load(in_ptr1 + r0, None)
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp2 * tmp3
tmp5 = 0.0
tmp6 = triton_helpers.minimum(tmp5, tmp3)
tmp7 = tl_math.abs(tmp3)
tmp8 = -tmp7
tmp9 = tl_math.exp(tmp8)
tmp10 = libdevice.log1p(tmp9)
tmp11 = tmp6 - tmp10
tmp12 = tmp4 - tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 256.0
tmp17 = tmp15 / tmp16
tmp18 = tmp17 / tmp1
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp18, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_mean_0[grid(1)](buf1,
arg0_1, arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class BWCEWLossNew(nn.Module):
""" Binary weighted cross entropy loss. """
def __init__(self, positive_class_weight: 'Optional[Tensor]'=None,
robust_lambda: 'int'=0, confidence_penalty: 'int'=0, **kwargs):
super().__init__()
self.loss_fn = nn.BCEWithLogitsLoss(pos_weight=
positive_class_weight, **kwargs)
self.robust_lambda = robust_lambda
self.confidence_penalty = confidence_penalty
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| carlogrisetti/ludwig | BWCEWLoss | false | 1,630 | [
"Apache-2.0"
] | 0 | 5c0887f14867e1577e0ddc3806c5cf7a781fb665 | https://github.com/carlogrisetti/ludwig/tree/5c0887f14867e1577e0ddc3806c5cf7a781fb665 | import torch
from torch import Tensor
from typing import Optional
from torch import nn
class Model(nn.Module):
""" Binary weighted cross entropy loss. """
def __init__(self, positive_class_weight: 'Optional[Tensor]'=None,
robust_lambda: 'int'=0, confidence_penalty: 'int'=0, **kwargs):
super().__init__()
self.loss_fn = nn.BCEWithLogitsLoss(pos_weight=
positive_class_weight, **kwargs)
self.robust_lambda = robust_lambda
self.confidence_penalty = confidence_penalty
def forward(self, preds: 'torch.Tensor', target: 'torch.Tensor'):
train_loss = self.loss_fn(preds, target.float())
if self.robust_lambda > 0:
train_loss = (1 - self.robust_lambda
) * train_loss + self.robust_lambda / 2
train_mean_loss = torch.mean(train_loss)
if self.confidence_penalty > 0:
probabilities = torch.sigmoid(preds)
mean_penalty = utils.mean_confidence_penalty(probabilities, 2)
train_mean_loss += self.confidence_penalty * mean_penalty
return train_mean_loss
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ConvLayer | # 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_6/inductor_cache/r3/cr3hlg2dj2d3nmsli5wlcbgrfym3b6ux3uuxd7pl3rggj6domt5d.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.reflection_pad2d]
# Source node to ATen node mapping:
# x => _unsafe_index, _unsafe_index_1
# Graph fragment:
# %_unsafe_index : [num_users=1] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%primals_1, [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=[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_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 = 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)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (15 + ((-1)*(tl_math.abs((-3) + x0))) + ((-4)*(tl_math.abs((-3) + x1))) + (16*x2)), xmask)
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/32/c32v7egt4mupqssam3gmac2qgv3ujprjybthsgweflmot256qqw7.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 = (%_unsafe_index_1, %primals_2, %primals_3, [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 = 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, ))
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: [x], Original ATen: [aten.reflection_pad2d]
stream0 = get_raw_stream(0)
triton_poi_fused_reflection_pad2d_0.run(primals_1, buf0, 256, grid=grid(256), 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=(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 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf2, primals_3, 256, grid=grid(256), stream=stream0)
del primals_3
return (buf2, primals_2, 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, 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)
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
class ConvLayer(torch.nn.Module):
def __init__(self, in_features, out_features, kernel_size=1, stride=1):
super().__init__()
padding = kernel_size // 2
self.refpadding = torch.nn.ReflectionPad2d(padding)
self.conv = torch.nn.Conv2d(in_features, out_features, kernel_size,
stride)
def forward(self, x):
x = self.refpadding(x)
return self.conv(x)
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
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_reflection_pad2d_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
x3 = xindex
tmp0 = tl.load(in_ptr0 + (15 + -1 * tl_math.abs(-3 + x0) + -4 * tl_math
.abs(-3 + x1) + 16 * x2), xmask)
tl.store(out_ptr0 + x3, tmp0, 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 = 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,))
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_reflection_pad2d_0[grid(256)](primals_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
buf1 = extern_kernels.convolution(buf0, primals_2, 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 = buf1
del buf1
triton_poi_fused_convolution_1[grid(256)](buf2, primals_3, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_3
return buf2, primals_2, buf0
class ConvLayerNew(torch.nn.Module):
def __init__(self, in_features, out_features, kernel_size=1, stride=1):
super().__init__()
padding = kernel_size // 2
self.refpadding = torch.nn.ReflectionPad2d(padding)
self.conv = torch.nn.Conv2d(in_features, out_features, kernel_size,
stride)
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]
| bruchano/GTAVSelfDriving | ConvLayer | false | 1,631 | [
"MIT"
] | 0 | c5f929f793b48e47725dbdb8f0991e04e3c43eba | https://github.com/bruchano/GTAVSelfDriving/tree/c5f929f793b48e47725dbdb8f0991e04e3c43eba | import torch
class Model(torch.nn.Module):
def __init__(self, in_features, out_features, kernel_size=1, stride=1):
super().__init__()
padding = kernel_size // 2
self.refpadding = torch.nn.ReflectionPad2d(padding)
self.conv = torch.nn.Conv2d(in_features, out_features, kernel_size,
stride)
def forward(self, x):
x = self.refpadding(x)
return self.conv(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ReduceMax | # 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_6/inductor_cache/y7/cy7jkncvgqkch7rp5inkpciskf4ausuosjdvipi55vmji2a4yndh.py
# Topologically Sorted Source Nodes: [amax], Original ATen: [aten.amax]
# Source node to ATen node mapping:
# amax => amax
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%arg0_1, [1]), kwargs = {})
triton_poi_fused_amax_0 = async_compile.triton('triton_poi_fused_amax_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_amax_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_amax_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)
tmp1 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tl.store(out_ptr0 + (x2), tmp6, 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: [amax], Original ATen: [aten.amax]
stream0 = get_raw_stream(0)
triton_poi_fused_amax_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
class ReduceMax(torch.nn.Module):
def forward(self, inputs, mask=None):
return torch.amax(inputs, dim=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
from torch._inductor.runtime import triton_helpers
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_amax_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)
tmp1 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tl.store(out_ptr0 + x2, tmp6, 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_amax_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class ReduceMaxNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| carlogrisetti/ludwig | ReduceMax | false | 1,632 | [
"Apache-2.0"
] | 0 | 5c0887f14867e1577e0ddc3806c5cf7a781fb665 | https://github.com/carlogrisetti/ludwig/tree/5c0887f14867e1577e0ddc3806c5cf7a781fb665 | import torch
class Model(torch.nn.Module):
def forward(self, inputs, mask=None):
return torch.amax(inputs, dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SingleLayer | # 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_6/inductor_cache/3j/c3jk4fd45xsskb354bmqh5ayvalm334wxs72twddoal7gsrew3wi.py
# Topologically Sorted Source Nodes: [group_norm, relu], Original ATen: [aten.native_group_norm, aten.relu]
# Source node to ATen node mapping:
# group_norm => add, add_1, mul_1, rsqrt, var_mean
# relu => relu
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view, [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 = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, %unsqueeze_5), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %unsqueeze_2), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_1,), kwargs = {})
triton_per_fused_native_group_norm_relu_0 = async_compile.triton('triton_per_fused_native_group_norm_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.persistent_reduction(
size_hints=[16, 16],
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_group_norm_relu_0', 'mutated_arg_names': [], 'no_x_dim': False, '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_group_norm_relu_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr2, out_ptr3, 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
x2 = xindex % 4
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp24 = tl.load(in_ptr1 + (x2), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = tmp0 - tmp10
tmp18 = 16.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp25 = tmp23 * tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.full([1, 1], 0, tl.int32)
tmp29 = triton_helpers.maximum(tmp28, tmp27)
tl.store(out_ptr2 + (r1 + (16*x0)), tmp29, xmask)
tl.store(out_ptr3 + (x0), tmp22, xmask)
tl.store(out_ptr0 + (x0), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/2f/c2fom4sdctqko5bxlqb4lq3ngyr5wpcuqvrqjkl76hv6o4dyucyu.py
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# out_1 => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_3, %convolution], 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=[512],
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_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_cat_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 16) % 8
x0 = xindex % 16
x2 = (xindex // 128)
x3 = xindex
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) + (64*x2)), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x0 + (16*((-4) + x1)) + (64*x2)), tmp6 & xmask, other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x3), tmp10, 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, ), (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, 3, 3), (36, 9, 3, 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)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
# Topologically Sorted Source Nodes: [group_norm, relu], Original ATen: [aten.native_group_norm, aten.relu]
stream0 = get_raw_stream(0)
triton_per_fused_native_group_norm_relu_0.run(primals_3, primals_1, primals_2, buf0, buf3, buf6, 16, 16, grid=grid(16), stream=stream0)
del primals_1
del primals_2
# Topologically Sorted Source Nodes: [out], 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, 4, 4), (64, 16, 4, 1))
buf5 = empty_strided_cuda((4, 8, 4, 4), (128, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.cat]
triton_poi_fused_cat_1.run(primals_3, buf4, buf5, 512, grid=grid(512), stream=stream0)
del buf4
return (buf5, primals_3, primals_4, buf3, reinterpret_tensor(buf0, (4, 4, 1), (4, 1, 1), 0), reinterpret_tensor(buf6, (4, 4, 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, ), (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, 3, 3), (36, 9, 3, 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 torch.nn as nn
class SingleLayer(nn.Module):
def __init__(self, nChannels, growthRate):
super(SingleLayer, self).__init__()
self.bn1 = nn.GroupNorm(nChannels, nChannels, affine=True)
self.conv1 = nn.Conv2d(nChannels, growthRate, kernel_size=3,
padding=1, bias=False)
self.relu = nn.ReLU(inplace=True)
def forward(self, x):
out = self.conv1(self.relu(self.bn1(x)))
out = torch.cat((x, out), 1)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nChannels': 4, 'growthRate': 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
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_native_group_norm_relu_0(in_ptr0, in_ptr1, in_ptr2,
out_ptr0, out_ptr2, out_ptr3, 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
x2 = xindex % 4
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp24 = tl.load(in_ptr1 + x2, xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tl.where(xmask, tmp1, 0)
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = tmp0 - tmp10
tmp18 = 16.0
tmp19 = tmp16 / tmp18
tmp20 = 1e-05
tmp21 = tmp19 + tmp20
tmp22 = libdevice.rsqrt(tmp21)
tmp23 = tmp17 * tmp22
tmp25 = tmp23 * tmp24
tmp27 = tmp25 + tmp26
tmp28 = tl.full([1, 1], 0, tl.int32)
tmp29 = triton_helpers.maximum(tmp28, tmp27)
tl.store(out_ptr2 + (r1 + 16 * x0), tmp29, xmask)
tl.store(out_ptr3 + x0, tmp22, xmask)
tl.store(out_ptr0 + x0, tmp10, xmask)
@triton.jit
def triton_poi_fused_cat_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 16 % 8
x0 = xindex % 16
x2 = xindex // 128
x3 = xindex
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 + 64 * x2), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (x0 + 16 * (-4 + x1) + 64 * x2), tmp6 & xmask,
other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x3, tmp10, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = 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, 3, 3), (36, 9, 3, 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)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
get_raw_stream(0)
triton_per_fused_native_group_norm_relu_0[grid(16)](primals_3,
primals_1, primals_2, buf0, buf3, buf6, 16, 16, XBLOCK=1,
num_warps=2, num_stages=1)
del primals_1
del primals_2
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, 4, 4), (64, 16, 4, 1))
buf5 = empty_strided_cuda((4, 8, 4, 4), (128, 16, 4, 1), torch.float32)
triton_poi_fused_cat_1[grid(512)](primals_3, buf4, buf5, 512,
XBLOCK=128, num_warps=4, num_stages=1)
del buf4
return buf5, primals_3, primals_4, buf3, reinterpret_tensor(buf0, (4, 4,
1), (4, 1, 1), 0), reinterpret_tensor(buf6, (4, 4, 1), (4, 1, 1), 0)
class SingleLayerNew(nn.Module):
def __init__(self, nChannels, growthRate):
super(SingleLayerNew, self).__init__()
self.bn1 = nn.GroupNorm(nChannels, nChannels, affine=True)
self.conv1 = nn.Conv2d(nChannels, growthRate, kernel_size=3,
padding=1, bias=False)
self.relu = nn.ReLU(inplace=True)
def forward(self, input_0):
primals_1 = self.bn1.weight
primals_2 = self.bn1.bias
primals_4 = self.conv1.weight
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| cadurosar/graph_kd_dense_cifar100 | SingleLayer | false | 1,633 | [
"MIT"
] | 0 | 84054ab4f8f61c9db3460993661ba7bf1d951b36 | https://github.com/cadurosar/graph_kd_dense_cifar100/tree/84054ab4f8f61c9db3460993661ba7bf1d951b36 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, nChannels, growthRate):
super().__init__()
self.bn1 = nn.GroupNorm(nChannels, nChannels, affine=True)
self.conv1 = nn.Conv2d(nChannels, growthRate, kernel_size=3,
padding=1, bias=False)
self.relu = nn.ReLU(inplace=True)
def forward(self, x):
out = self.conv1(self.relu(self.bn1(x)))
out = torch.cat((x, out), 1)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
TransformerLayer | # 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_6/inductor_cache/52/c525by6g4qqvzpkamxv56vfyu6zlvbqfepthe4npppzrrv2boata.py
# Topologically Sorted Source Nodes: [means, x_zeromean], Original ATen: [aten.mean, aten.sub]
# Source node to ATen node mapping:
# means => mean
# x_zeromean => sub
# 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=[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_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 = 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 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = tmp0 - tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/jb/cjbxtrhaay7rotsyysuftg23isdvjimy5xsuoa3afjgeuvhmcet3.py
# Topologically Sorted Source Nodes: [pow_1, variances, add, sqrt, x, mul, x_1], Original ATen: [aten.pow, aten.mean, aten.add, aten.sqrt, aten.div, aten.mul]
# Source node to ATen node mapping:
# add => add
# mul => mul
# pow_1 => pow_1
# sqrt => sqrt
# variances => mean_1
# x => div
# x_1 => add_1
# 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=[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_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 = 64
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')
# kernel path: runs/run_shard_6/inductor_cache/uc/cucgrga44gtlwzw6ehoy4jrwzm5fghn3ljf7iugmdjhe6m7mjcas.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 = ([%primals_4, %primals_5, %primals_6],), kwargs = {})
triton_poi_fused_cat_2 = async_compile.triton('triton_poi_fused_cat_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_2', '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_cat_2(in_ptr0, in_ptr1, in_ptr2, 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
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 + (x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr1 + ((-4) + x0), tmp9 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1], 12, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tl.load(in_ptr2 + ((-8) + x0), tmp11 & xmask, eviction_policy='evict_last', other=0.0)
tmp15 = tl.where(tmp9, tmp10, tmp14)
tmp16 = tl.where(tmp4, tmp5, tmp15)
tl.store(out_ptr0 + (x0), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/7v/c7vmyyo5wxjwhczncwe26c5vvxpnuc4d63pyhv44adhxrqqhbkd6.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# multi_head_attention_forward => cat_2
# Graph fragment:
# %cat_2 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%view_5, %repeat_1],), kwargs = {})
triton_poi_fused_cat_3 = async_compile.triton('triton_poi_fused_cat_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: '*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_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_cat_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 80
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 16)
x3 = xindex % 16
x0 = xindex % 4
x4 = xindex
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 = tmp0 >= tmp3
tmp7 = tl.full([1], 5, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x0), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x4), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/hr/chreiskn3cjyo3eff2gsxuz7ampglelwlsrzs7kmnwjm4l5lzi5w.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# multi_head_attention_forward => mul_1
# Graph fragment:
# %mul_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_3, 1.0), kwargs = {})
triton_poi_fused_mul_4 = async_compile.triton('triton_poi_fused_mul_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=[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_mul_4', '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_mul_4(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
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = x2 % 4
tmp2 = tl.full([1], 0, tl.int64)
tmp3 = tmp1 >= tmp2
tmp4 = tl.full([1], 4, tl.int64)
tmp5 = tmp1 < tmp4
tmp6 = tl.load(in_ptr0 + (x0 % 4), tmp5 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = tmp1 >= tmp4
tmp8 = tl.full([1], 8, tl.int64)
tmp9 = tmp1 < tmp8
tmp10 = tmp7 & tmp9
tmp11 = tl.load(in_ptr1 + ((-4) + (x0 % 4)), tmp10 & xmask, eviction_policy='evict_last', other=0.0)
tmp12 = tmp1 >= tmp8
tmp13 = tl.full([1], 12, tl.int64)
tmp14 = tmp1 < tmp13
tmp15 = tl.load(in_ptr2 + ((-8) + (x0 % 4)), tmp12 & xmask, eviction_policy='evict_last', other=0.0)
tmp16 = tl.where(tmp10, tmp11, tmp15)
tmp17 = tl.where(tmp5, tmp6, tmp16)
tmp18 = tmp0 + tmp17
tmp19 = 1.0
tmp20 = tmp18 * tmp19
tl.store(in_out_ptr0 + (x2), tmp20, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/qq/cqqiwi6cphov3y4cw45pihso7kygfbxrztamsxi6ywv3sllhzvz5.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# multi_head_attention_forward => amax, exp, sub_1, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub_1 : [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_1,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
triton_poi_fused__softmax_5 = async_compile.triton('triton_poi_fused__softmax_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_5', '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_5(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 + (5*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (5*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (5*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (5*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (4 + (5*x0)), xmask, eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tmp8 = triton_helpers.maximum(tmp6, tmp7)
tmp9 = tmp0 - tmp8
tmp10 = tl_math.exp(tmp9)
tmp11 = tmp1 - tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tl_math.exp(tmp14)
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp8
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tl.store(out_ptr0 + (x0), tmp8, xmask)
tl.store(out_ptr1 + (x0), tmp22, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ti/ctik3dxqudjbuxk6lecrd6cxqp5ncnz6kele7crpn6ziyqip447n.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# multi_head_attention_forward => amax, div_1, exp, sub_1, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub_1 : [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_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=3] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_6 = async_compile.triton('triton_poi_fused__softmax_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=[512],
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_6', '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__softmax_6(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 320
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')
tmp4 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tl.store(in_out_ptr0 + (x2), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/na/cnaxt66xj2zhaned4fgx5elukrpsiq623ib46puxb6rcdh4iismy.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# multi_head_attention_forward => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), 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=[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_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 = 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_6/inductor_cache/yx/cyx6uhp26itou7morodlakbwephiwz7r7jx56snt2ukwrojhlmfp.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# multi_head_attention_forward => mean_2
# Graph fragment:
# %mean_2 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view_11, [1]), kwargs = {})
triton_poi_fused_mean_8 = async_compile.triton('triton_poi_fused_mean_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=[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_mean_8', '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_8(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 80
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 20
x1 = (xindex // 20)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (80*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (20 + x0 + (80*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (40 + x0 + (80*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (60 + x0 + (80*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_6/inductor_cache/hu/chuhaqnv3eocytas2o5zkt2laav3cpr7gqx5s2m2tsctpi2kooe7.py
# Topologically Sorted Source Nodes: [x_3, means_1, x_zeromean_1, pow_2, variances_1], Original ATen: [aten.add, aten.mean, aten.sub, aten.pow]
# Source node to ATen node mapping:
# means_1 => mean_3
# pow_2 => pow_2
# variances_1 => mean_4
# x_3 => add_2
# x_zeromean_1 => sub_2
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_10), kwargs = {})
# %mean_3 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add_2, [-1], True), kwargs = {})
# %sub_2 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %mean_3), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_2, 2), kwargs = {})
# %mean_4 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%pow_2, [-1], True), kwargs = {})
triton_poi_fused_add_mean_pow_sub_9 = async_compile.triton('triton_poi_fused_add_mean_pow_sub_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: '*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_mean_pow_sub_9', '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_mean_pow_sub_9(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_6/inductor_cache/ps/cpsrkyasbkreyck4njrazduiuuaugpsanaxlzzqhnuxs7rd6xtwj.py
# Topologically Sorted Source Nodes: [x_3, means_1, x_zeromean_1, add_3, sqrt_1, x_4, mul_1, x_5], Original ATen: [aten.add, aten.mean, aten.sub, aten.sqrt, aten.div, aten.mul]
# Source node to ATen node mapping:
# add_3 => add_3
# means_1 => mean_3
# mul_1 => mul_2
# sqrt_1 => sqrt_1
# x_3 => add_2
# x_4 => div_2
# x_5 => add_4
# x_zeromean_1 => sub_2
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_10), kwargs = {})
# %mean_3 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add_2, [-1], True), kwargs = {})
# %sub_2 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %mean_3), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean_4, 1e-12), kwargs = {})
# %sqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add_3,), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_2, %sqrt_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_14, %div_2), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_15), kwargs = {})
triton_poi_fused_add_div_mean_mul_sqrt_sub_10 = async_compile.triton('triton_poi_fused_add_div_mean_mul_sqrt_sub_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=[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_div_mean_mul_sqrt_sub_10', '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_mean_mul_sqrt_sub_10(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
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_ptr2 + (x2), xmask)
tmp4 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 - tmp4
tmp7 = 1e-12
tmp8 = tmp6 + tmp7
tmp9 = libdevice.sqrt(tmp8)
tmp10 = tmp5 / tmp9
tmp11 = tmp0 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/in/cingjrjcsnplp4mnf7j3qaenyi3g6h7mejtd6hllewoqj33nanxc.py
# Topologically Sorted Source Nodes: [mul_2, truediv_2, erf, add_5, x_6], Original ATen: [aten.mul, aten.div, aten.erf, aten.add]
# Source node to ATen node mapping:
# add_5 => add_5
# erf => erf
# mul_2 => mul_3
# truediv_2 => div_3
# x_6 => mul_4
# Graph fragment:
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_13, 0.5), kwargs = {})
# %div_3 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_13, 1.4142135623730951), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%div_3,), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1.0), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %add_5), kwargs = {})
triton_poi_fused_add_div_erf_mul_11 = async_compile.triton('triton_poi_fused_add_div_erf_mul_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_erf_mul_11', '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_erf_mul_11(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 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tl.store(out_ptr0 + (x0), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/55/c55kf6uasfvlyi4fs3esg22nysynhdxbi47sxnyfzfqq64pxh7xj.py
# Topologically Sorted Source Nodes: [x_3, x_8], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x_3 => add_2
# x_8 => add_6
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_10), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %view_15), kwargs = {})
triton_poi_fused_add_12 = async_compile.triton('triton_poi_fused_add_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.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_add_12', '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_12(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
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_out_ptr0 + (x2), xmask)
tmp4 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tl.store(in_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, 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 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (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, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (1, 1, 4), (4, 4, 1))
assert_size_stride(primals_8, (1, 1, 4), (4, 4, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4, 4), (4, 1))
assert_size_stride(primals_13, (4, 4), (4, 1))
assert_size_stride(primals_14, (4, ), (1, ))
assert_size_stride(primals_15, (4, ), (1, ))
assert_size_stride(primals_16, (4, 4), (4, 1))
assert_size_stride(primals_17, (4, ), (1, ))
assert_size_stride(primals_18, (4, 4), (4, 1))
assert_size_stride(primals_19, (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: [means, x_zeromean], Original ATen: [aten.mean, aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_mean_sub_0.run(primals_1, buf0, 64, grid=grid(64), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pow_1, variances, add, sqrt, x, mul, x_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, 64, grid=grid(64), stream=stream0)
del primals_2
del primals_3
buf2 = reinterpret_tensor(buf0, (16, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((12, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
triton_poi_fused_cat_2.run(primals_4, primals_5, primals_6, buf3, 12, grid=grid(12), stream=stream0)
buf4 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(buf3, (4, ), (1, ), 4), reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_12, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf4)
buf5 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(buf3, (4, ), (1, ), 8), reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf5)
del buf3
buf6 = empty_strided_cuda((5, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.cat]
triton_poi_fused_cat_3.run(buf5, primals_8, buf6, 80, grid=grid(80), stream=stream0)
del primals_8
buf7 = reinterpret_tensor(buf2, (16, 4, 1), (1, 16, 64), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
triton_poi_fused_mul_4.run(buf7, primals_4, primals_5, primals_6, 64, grid=grid(64), stream=stream0)
del primals_4
del primals_5
del primals_6
buf8 = empty_strided_cuda((5, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.cat]
triton_poi_fused_cat_3.run(buf4, primals_7, buf8, 80, grid=grid(80), stream=stream0)
del primals_7
buf9 = empty_strided_cuda((16, 4, 5), (20, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 1, 5), (1, 0, 16), 0), out=buf9)
buf10 = reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 64), 0); del buf4 # reuse
buf11 = reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 64), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_5.run(buf9, buf10, buf11, 64, grid=grid(64), stream=stream0)
buf12 = buf9; del buf9 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_6.run(buf12, buf10, buf11, 320, grid=grid(320), stream=stream0)
buf13 = reinterpret_tensor(buf11, (16, 4, 1), (4, 1, 1), 0); del buf11 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(buf12, reinterpret_tensor(buf6, (16, 5, 1), (1, 16, 0), 0), out=buf13)
buf14 = reinterpret_tensor(buf10, (4, 16, 1), (16, 1, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
triton_poi_fused_clone_7.run(buf13, buf14, 4, 16, grid=grid(4, 16), stream=stream0)
buf15 = reinterpret_tensor(buf13, (16, 4), (4, 1), 0); del buf13 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(buf14, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf15)
del primals_10
buf16 = empty_strided_cuda((4, 4, 5), (20, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mean]
triton_poi_fused_mean_8.run(buf12, buf16, 80, grid=grid(80), stream=stream0)
buf17 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf18 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [x_3, means_1, x_zeromean_1, pow_2, variances_1], Original ATen: [aten.add, aten.mean, aten.sub, aten.pow]
triton_poi_fused_add_mean_pow_sub_9.run(primals_1, buf15, buf17, buf18, 16, grid=grid(16), stream=stream0)
buf19 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3, means_1, x_zeromean_1, add_3, sqrt_1, x_4, mul_1, x_5], Original ATen: [aten.add, aten.mean, aten.sub, aten.sqrt, aten.div, aten.mul]
triton_poi_fused_add_div_mean_mul_sqrt_sub_10.run(primals_14, primals_1, buf15, buf17, buf18, primals_15, buf19, 64, grid=grid(64), stream=stream0)
del buf17
del buf18
del primals_15
buf20 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_17, reinterpret_tensor(buf19, (16, 4), (4, 1), 0), reinterpret_tensor(primals_16, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf20)
del primals_17
buf21 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_2, truediv_2, erf, add_5, x_6], Original ATen: [aten.mul, aten.div, aten.erf, aten.add]
triton_poi_fused_add_div_erf_mul_11.run(buf20, buf21, 64, grid=grid(64), stream=stream0)
buf22 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf21, (16, 4), (4, 1), 0), reinterpret_tensor(primals_18, (4, 4), (1, 4), 0), out=buf22)
buf23 = reinterpret_tensor(buf22, (4, 4, 4), (16, 4, 1), 0); del buf22 # reuse
# Topologically Sorted Source Nodes: [x_3, x_8], Original ATen: [aten.add]
triton_poi_fused_add_12.run(buf23, primals_1, buf15, primals_19, 64, grid=grid(64), stream=stream0)
del primals_19
return (buf23, buf16, primals_1, primals_14, reinterpret_tensor(buf1, (16, 4), (4, 1), 0), buf12, reinterpret_tensor(buf14, (16, 4), (4, 1), 0), buf15, reinterpret_tensor(buf19, (16, 4), (4, 1), 0), buf20, reinterpret_tensor(buf21, (16, 4), (4, 1), 0), primals_18, primals_16, primals_9, reinterpret_tensor(buf6, (16, 1, 5), (1, 1, 16), 0), reinterpret_tensor(buf7, (16, 1, 4), (1, 1, 16), 0), reinterpret_tensor(buf8, (16, 5, 1), (1, 16, 1), 0), primals_13, primals_12, primals_11, )
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, ), (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, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, 1, 4), (4, 4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((1, 1, 4), (4, 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, 4), (4, 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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_18 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_19 = 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, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19])
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 uuid
from torch import Tensor
import torch.nn as nn
from typing import Tuple
import torch.nn.functional as F
from typing import Optional
from typing import Dict
from torch.nn import Parameter
def gelu(x):
"""Implementation of the gelu activation function.
For information: OpenAI GPT's gelu is slightly different
(and gives slightly different results):
0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
def utils_softmax(x, dim: 'int', onnx_trace: 'bool'=False):
if onnx_trace:
return F.softmax(x.float(), dim=dim)
else:
return F.softmax(x, dim=dim, dtype=torch.float32)
def with_incremental_state(cls):
cls.__bases__ = (FairseqIncrementalState,) + tuple(b for b in cls.
__bases__ if b != FairseqIncrementalState)
return cls
class ESM1LayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12, affine=True):
"""Construct a layernorm layer in the TF style (eps inside the sqrt)."""
super().__init__()
self.hidden_size = (hidden_size,) if isinstance(hidden_size, int
) else tuple(hidden_size)
self.eps = eps
self.affine = bool(affine)
if self.affine:
self.weight = nn.Parameter(torch.ones(hidden_size))
self.bias = nn.Parameter(torch.zeros(hidden_size))
else:
self.weight, self.bias = None, None
def forward(self, x):
dims = tuple(-(i + 1) for i in range(len(self.hidden_size)))
means = x.mean(dims, keepdim=True)
x_zeromean = x - means
variances = x_zeromean.pow(2).mean(dims, keepdim=True)
x = x_zeromean / torch.sqrt(variances + self.eps)
if self.affine:
x = self.weight * x + self.bias
return x
class FairseqIncrementalState(object):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.init_incremental_state()
def init_incremental_state(self):
self._incremental_state_id = str(uuid.uuid4())
def _get_full_incremental_state_key(self, key: 'str') ->str:
return '{}.{}'.format(self._incremental_state_id, key)
def get_incremental_state(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]', key: 'str'
) ->Optional[Dict[str, Optional[Tensor]]]:
"""Helper for getting incremental state for an nn.Module."""
full_key = self._get_full_incremental_state_key(key)
if incremental_state is None or full_key not in incremental_state:
return None
return incremental_state[full_key]
def set_incremental_state(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]', key: 'str',
value: 'Dict[str, Optional[Tensor]]') ->Optional[Dict[str, Dict[str,
Optional[Tensor]]]]:
"""Helper for setting incremental state for an nn.Module."""
if incremental_state is not None:
full_key = self._get_full_incremental_state_key(key)
incremental_state[full_key] = value
return incremental_state
@with_incremental_state
class MultiheadAttention(nn.Module):
"""Multi-headed attention.
See "Attention Is All You Need" for more details.
"""
def __init__(self, embed_dim, num_heads, kdim=None, vdim=None, dropout=
0.0, bias=True, add_bias_kv=False, add_zero_attn=False,
self_attention=False, encoder_decoder_attention=False):
super().__init__()
self.embed_dim = embed_dim
self.kdim = kdim if kdim is not None else embed_dim
self.vdim = vdim if vdim is not None else embed_dim
self.qkv_same_dim = self.kdim == embed_dim and self.vdim == embed_dim
self.num_heads = num_heads
self.dropout = dropout
self.head_dim = embed_dim // num_heads
assert self.head_dim * num_heads == self.embed_dim, 'embed_dim must be divisible by num_heads'
self.scaling = self.head_dim ** -0.5
self.self_attention = self_attention
self.encoder_decoder_attention = encoder_decoder_attention
assert not self.self_attention or self.qkv_same_dim, 'Self-attention requires query, key and value to be of the same size'
self.k_proj = nn.Linear(self.kdim, embed_dim, bias=bias)
self.v_proj = nn.Linear(self.vdim, embed_dim, bias=bias)
self.q_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
self.out_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
if add_bias_kv:
self.bias_k = Parameter(torch.Tensor(1, 1, embed_dim))
self.bias_v = Parameter(torch.Tensor(1, 1, embed_dim))
else:
self.bias_k = self.bias_v = None
self.add_zero_attn = add_zero_attn
self.reset_parameters()
self.onnx_trace = False
self.enable_torch_version = False
if hasattr(F, 'multi_head_attention_forward'):
self.enable_torch_version = True
else:
self.enable_torch_version = False
def prepare_for_onnx_export_(self):
self.onnx_trace = True
def reset_parameters(self):
if self.qkv_same_dim:
nn.init.xavier_uniform_(self.k_proj.weight, gain=1 / math.sqrt(2))
nn.init.xavier_uniform_(self.v_proj.weight, gain=1 / math.sqrt(2))
nn.init.xavier_uniform_(self.q_proj.weight, gain=1 / math.sqrt(2))
else:
nn.init.xavier_uniform_(self.k_proj.weight)
nn.init.xavier_uniform_(self.v_proj.weight)
nn.init.xavier_uniform_(self.q_proj.weight)
nn.init.xavier_uniform_(self.out_proj.weight)
if self.out_proj.bias is not None:
nn.init.constant_(self.out_proj.bias, 0.0)
if self.bias_k is not None:
nn.init.xavier_normal_(self.bias_k)
if self.bias_v is not None:
nn.init.xavier_normal_(self.bias_v)
def forward(self, query, key: 'Optional[Tensor]', value:
'Optional[Tensor]', key_padding_mask: 'Optional[Tensor]'=None,
incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]'=None,
need_weights: 'bool'=True, static_kv: 'bool'=False, attn_mask:
'Optional[Tensor]'=None, before_softmax: 'bool'=False,
need_head_weights: 'bool'=False) ->Tuple[Tensor, Optional[Tensor]]:
"""Input shape: Time x Batch x Channel
Args:
key_padding_mask (ByteTensor, optional): mask to exclude
keys that are pads, of shape `(batch, src_len)`, where
padding elements are indicated by 1s.
need_weights (bool, optional): return the attention weights,
averaged over heads (default: False).
attn_mask (ByteTensor, optional): typically used to
implement causal attention, where the mask prevents the
attention from looking forward in time (default: None).
before_softmax (bool, optional): return the raw attention
weights and values before the attention softmax.
need_head_weights (bool, optional): return the attention
weights for each head. Implies *need_weights*. Default:
return the average attention weights over all heads.
"""
if need_head_weights:
need_weights = True
tgt_len, bsz, embed_dim = query.size()
assert embed_dim == self.embed_dim
assert list(query.size()) == [tgt_len, bsz, embed_dim]
if (self.enable_torch_version and not self.onnx_trace and
incremental_state is None and not static_kv and not torch.jit.
is_scripting() and not need_head_weights):
assert key is not None and value is not None
return F.multi_head_attention_forward(query, key, value, self.
embed_dim, self.num_heads, torch.empty([0]), torch.cat((
self.q_proj.bias, self.k_proj.bias, self.v_proj.bias)),
self.bias_k, self.bias_v, self.add_zero_attn, self.dropout,
self.out_proj.weight, self.out_proj.bias, self.training,
key_padding_mask, need_weights, 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)
if incremental_state is not None:
saved_state = self._get_input_buffer(incremental_state)
if saved_state is not None and 'prev_key' in saved_state:
if static_kv:
assert self.encoder_decoder_attention and not self.self_attention
key = value = None
else:
saved_state = None
if self.self_attention:
q = self.q_proj(query)
k = self.k_proj(query)
v = self.v_proj(query)
elif self.encoder_decoder_attention:
q = self.q_proj(query)
if key is None:
assert value is None
k = v = None
else:
k = self.k_proj(key)
v = self.v_proj(key)
else:
assert key is not None and value is not None
q = self.q_proj(query)
k = self.k_proj(key)
v = self.v_proj(value)
q *= self.scaling
if self.bias_k is not None:
assert self.bias_v is not None
k = torch.cat([k, self.bias_k.repeat(1, bsz, 1)])
v = torch.cat([v, self.bias_v.repeat(1, bsz, 1)])
if attn_mask is not None:
attn_mask = torch.cat([attn_mask, attn_mask.new_zeros(
attn_mask.size(0), 1)], dim=1)
if key_padding_mask is not None:
key_padding_mask = torch.cat([key_padding_mask,
key_padding_mask.new_zeros(key_padding_mask.size(0), 1)
], dim=1)
q = q.contiguous().view(tgt_len, bsz * self.num_heads, self.head_dim
).transpose(0, 1)
if k is not None:
k = k.contiguous().view(-1, bsz * self.num_heads, self.head_dim
).transpose(0, 1)
if v is not None:
v = v.contiguous().view(-1, bsz * self.num_heads, self.head_dim
).transpose(0, 1)
if saved_state is not None:
if 'prev_key' in saved_state:
_prev_key = saved_state['prev_key']
assert _prev_key is not None
prev_key = _prev_key.view(bsz * self.num_heads, -1, self.
head_dim)
if static_kv:
k = prev_key
else:
assert k is not None
k = torch.cat([prev_key, k], dim=1)
if 'prev_value' in saved_state:
_prev_value = saved_state['prev_value']
assert _prev_value is not None
prev_value = _prev_value.view(bsz * self.num_heads, -1,
self.head_dim)
if static_kv:
v = prev_value
else:
assert v is not None
v = torch.cat([prev_value, v], dim=1)
prev_key_padding_mask: 'Optional[Tensor]' = None
if 'prev_key_padding_mask' in saved_state:
prev_key_padding_mask = saved_state['prev_key_padding_mask']
assert k is not None and v is not None
key_padding_mask = (MultiheadAttention.
_append_prev_key_padding_mask(key_padding_mask=
key_padding_mask, prev_key_padding_mask=
prev_key_padding_mask, batch_size=bsz, src_len=k.size(1),
static_kv=static_kv))
saved_state['prev_key'] = k.view(bsz, self.num_heads, -1, self.
head_dim)
saved_state['prev_value'] = v.view(bsz, self.num_heads, -1,
self.head_dim)
saved_state['prev_key_padding_mask'] = key_padding_mask
assert incremental_state is not None
incremental_state = self._set_input_buffer(incremental_state,
saved_state)
assert k is not None
src_len = k.size(1)
if key_padding_mask is not None and key_padding_mask.dim() == 0:
key_padding_mask = None
if key_padding_mask is not None:
assert key_padding_mask.size(0) == bsz
assert key_padding_mask.size(1) == src_len
if self.add_zero_attn:
assert v is not None
src_len += 1
k = torch.cat([k, k.new_zeros((k.size(0), 1) + k.size()[2:])],
dim=1)
v = torch.cat([v, v.new_zeros((v.size(0), 1) + v.size()[2:])],
dim=1)
if attn_mask is not None:
attn_mask = torch.cat([attn_mask, attn_mask.new_zeros(
attn_mask.size(0), 1)], dim=1)
if key_padding_mask is not None:
key_padding_mask = torch.cat([key_padding_mask, torch.zeros
(key_padding_mask.size(0), 1).type_as(key_padding_mask)
], dim=1)
attn_weights = torch.bmm(q, k.transpose(1, 2))
attn_weights = MultiheadAttention.apply_sparse_mask(attn_weights,
tgt_len, src_len, bsz)
assert list(attn_weights.size()) == [bsz * self.num_heads, tgt_len,
src_len]
if attn_mask is not None:
attn_mask = attn_mask.unsqueeze(0)
if self.onnx_trace:
attn_mask = attn_mask.repeat(attn_weights.size(0), 1, 1)
attn_weights += attn_mask
if key_padding_mask is not None:
attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len,
src_len)
attn_weights = attn_weights.masked_fill(key_padding_mask.
unsqueeze(1).unsqueeze(2), float('-inf'))
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len,
src_len)
if before_softmax:
return attn_weights, v
attn_weights_float = utils_softmax(attn_weights, dim=-1, onnx_trace
=self.onnx_trace)
attn_weights = attn_weights_float.type_as(attn_weights)
attn_probs = F.dropout(attn_weights_float.type_as(attn_weights), p=
self.dropout, training=self.training)
assert v is not None
attn = torch.bmm(attn_probs, v)
assert list(attn.size()) == [bsz * self.num_heads, tgt_len, self.
head_dim]
if self.onnx_trace and attn.size(1) == 1:
attn = attn.contiguous().view(tgt_len, bsz, embed_dim)
else:
attn = attn.transpose(0, 1).contiguous().view(tgt_len, bsz,
embed_dim)
attn = self.out_proj(attn)
attn_weights: 'Optional[Tensor]' = None
if need_weights:
attn_weights = attn_weights_float.view(bsz, self.num_heads,
tgt_len, src_len).transpose(1, 0)
if not need_head_weights:
attn_weights = attn_weights.mean(dim=0)
return attn, attn_weights
@staticmethod
def _append_prev_key_padding_mask(key_padding_mask: 'Optional[Tensor]',
prev_key_padding_mask: 'Optional[Tensor]', batch_size: 'int',
src_len: 'int', static_kv: 'bool') ->Optional[Tensor]:
if prev_key_padding_mask is not None and static_kv:
new_key_padding_mask = prev_key_padding_mask
elif prev_key_padding_mask is not None and key_padding_mask is not None:
new_key_padding_mask = torch.cat([prev_key_padding_mask.float(),
key_padding_mask.float()], dim=1)
elif prev_key_padding_mask is not None:
filler = torch.zeros((batch_size, src_len -
prev_key_padding_mask.size(1)), device=
prev_key_padding_mask.device)
new_key_padding_mask = torch.cat([prev_key_padding_mask.float(),
filler.float()], dim=1)
elif key_padding_mask is not None:
filler = torch.zeros((batch_size, src_len - key_padding_mask.
size(1)), device=key_padding_mask.device)
new_key_padding_mask = torch.cat([filler.float(),
key_padding_mask.float()], dim=1)
else:
new_key_padding_mask = prev_key_padding_mask
return new_key_padding_mask
@torch.jit.export
def reorder_incremental_state(self, incremental_state:
'Dict[str, Dict[str, Optional[Tensor]]]', new_order: 'Tensor'):
"""Reorder buffered internal state (for incremental generation)."""
input_buffer = self._get_input_buffer(incremental_state)
if input_buffer is not None:
for k in input_buffer.keys():
input_buffer_k = input_buffer[k]
if input_buffer_k is not None:
if self.encoder_decoder_attention and input_buffer_k.size(0
) == new_order.size(0):
break
input_buffer[k] = input_buffer_k.index_select(0, new_order)
incremental_state = self._set_input_buffer(incremental_state,
input_buffer)
return incremental_state
def _get_input_buffer(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]') ->Dict[str,
Optional[Tensor]]:
result = self.get_incremental_state(incremental_state, 'attn_state')
if result is not None:
return result
else:
empty_result: 'Dict[str, Optional[Tensor]]' = {}
return empty_result
def _set_input_buffer(self, incremental_state:
'Dict[str, Dict[str, Optional[Tensor]]]', buffer:
'Dict[str, Optional[Tensor]]'):
return self.set_incremental_state(incremental_state, 'attn_state',
buffer)
def apply_sparse_mask(attn_weights, tgt_len: 'int', src_len: 'int', bsz:
'int'):
return attn_weights
def upgrade_state_dict_named(self, state_dict, name):
prefix = name + '.' if name != '' else ''
items_to_add = {}
keys_to_remove = []
for k in state_dict.keys():
if k.endswith(prefix + 'in_proj_weight'):
dim = int(state_dict[k].shape[0] / 3)
items_to_add[prefix + 'q_proj.weight'] = state_dict[k][:dim]
items_to_add[prefix + 'k_proj.weight'] = state_dict[k][dim:
2 * dim]
items_to_add[prefix + 'v_proj.weight'] = state_dict[k][2 * dim:
]
keys_to_remove.append(k)
k_bias = prefix + 'in_proj_bias'
if k_bias in state_dict.keys():
dim = int(state_dict[k].shape[0] / 3)
items_to_add[prefix + 'q_proj.bias'] = state_dict[k_bias][:
dim]
items_to_add[prefix + 'k_proj.bias'] = state_dict[k_bias][
dim:2 * dim]
items_to_add[prefix + 'v_proj.bias'] = state_dict[k_bias][
2 * dim:]
keys_to_remove.append(prefix + 'in_proj_bias')
for k in keys_to_remove:
del state_dict[k]
for key, value in items_to_add.items():
state_dict[key] = value
class TransformerLayer(nn.Module):
"""Transformer layer block."""
def __init__(self, embed_dim, ffn_embed_dim, attention_heads,
add_bias_kv=True, use_esm1b_layer_norm=False):
super().__init__()
self.embed_dim = embed_dim
self.ffn_embed_dim = ffn_embed_dim
self.attention_heads = attention_heads
self._init_submodules(add_bias_kv, use_esm1b_layer_norm)
def _init_submodules(self, add_bias_kv, use_esm1b_layer_norm):
BertLayerNorm = (ESM1bLayerNorm if use_esm1b_layer_norm else
ESM1LayerNorm)
self.self_attn = MultiheadAttention(self.embed_dim, self.
attention_heads, add_bias_kv=add_bias_kv, add_zero_attn=False)
self.self_attn_layer_norm = BertLayerNorm(self.embed_dim)
self.fc1 = nn.Linear(self.embed_dim, self.ffn_embed_dim)
self.fc2 = nn.Linear(self.ffn_embed_dim, self.embed_dim)
self.final_layer_norm = BertLayerNorm(self.embed_dim)
def forward(self, x, self_attn_mask=None, self_attn_padding_mask=None,
need_head_weights=False):
residual = x
x = self.self_attn_layer_norm(x)
x, attn = self.self_attn(query=x, key=x, value=x, key_padding_mask=
self_attn_padding_mask, need_weights=True, need_head_weights=
need_head_weights, attn_mask=self_attn_mask)
x = residual + x
residual = x
x = self.final_layer_norm(x)
x = gelu(self.fc1(x))
x = self.fc2(x)
x = residual + x
return x, attn
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'embed_dim': 4, 'ffn_embed_dim': 4, 'attention_heads': 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 uuid
from torch import Tensor
import torch.nn as nn
from typing import Tuple
import torch.nn.functional as F
from typing import Optional
from typing import Dict
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_poi_fused_mean_sub_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)
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 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = tmp0 - tmp9
tl.store(out_ptr0 + x2, 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 = 64
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)
@triton.jit
def triton_poi_fused_cat_2(in_ptr0, in_ptr1, in_ptr2, 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
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + x0, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr1 + (-4 + x0), tmp9 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp11 = tmp0 >= tmp7
tl.full([1], 12, tl.int64)
tmp14 = tl.load(in_ptr2 + (-8 + x0), tmp11 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp15 = tl.where(tmp9, tmp10, tmp14)
tmp16 = tl.where(tmp4, tmp5, tmp15)
tl.store(out_ptr0 + x0, tmp16, xmask)
@triton.jit
def triton_poi_fused_cat_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 80
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 16
x3 = xindex % 16
x0 = xindex % 4
x4 = xindex
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 = tmp0 >= tmp3
tl.full([1], 5, tl.int64)
tmp9 = tl.load(in_ptr1 + x0, tmp6 & xmask, eviction_policy='evict_last',
other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x4, tmp10, xmask)
@triton.jit
def triton_poi_fused_mul_4(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
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = x2 % 4
tl.full([1], 0, tl.int64)
tmp4 = tl.full([1], 4, tl.int64)
tmp5 = tmp1 < tmp4
tmp6 = tl.load(in_ptr0 + x0 % 4, tmp5 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp7 = tmp1 >= tmp4
tmp8 = tl.full([1], 8, tl.int64)
tmp9 = tmp1 < tmp8
tmp10 = tmp7 & tmp9
tmp11 = tl.load(in_ptr1 + (-4 + x0 % 4), tmp10 & xmask, eviction_policy
='evict_last', other=0.0)
tmp12 = tmp1 >= tmp8
tl.full([1], 12, tl.int64)
tmp15 = tl.load(in_ptr2 + (-8 + x0 % 4), tmp12 & xmask, eviction_policy
='evict_last', other=0.0)
tmp16 = tl.where(tmp10, tmp11, tmp15)
tmp17 = tl.where(tmp5, tmp6, tmp16)
tmp18 = tmp0 + tmp17
tmp19 = 1.0
tmp20 = tmp18 * tmp19
tl.store(in_out_ptr0 + x2, tmp20, xmask)
@triton.jit
def triton_poi_fused__softmax_5(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 + 5 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 5 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 5 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 5 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (4 + 5 * x0), xmask, eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tmp8 = triton_helpers.maximum(tmp6, tmp7)
tmp9 = tmp0 - tmp8
tmp10 = tl_math.exp(tmp9)
tmp11 = tmp1 - tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tl_math.exp(tmp14)
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp8
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp22, xmask)
@triton.jit
def triton_poi_fused__softmax_6(in_out_ptr0, in_ptr0, in_ptr1, xnumel,
XBLOCK: tl.constexpr):
xnumel = 320
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')
tmp4 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tl.store(in_out_ptr0 + x2, tmp5, xmask)
@triton.jit
def triton_poi_fused_clone_7(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_mean_8(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 80
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 20
x1 = xindex // 20
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 80 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (20 + x0 + 80 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (40 + x0 + 80 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (60 + x0 + 80 * 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_add_mean_pow_sub_9(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_div_mean_mul_sqrt_sub_10(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
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_ptr2 + x2, xmask)
tmp4 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 - tmp4
tmp7 = 1e-12
tmp8 = tmp6 + tmp7
tmp9 = libdevice.sqrt(tmp8)
tmp10 = tmp5 / tmp9
tmp11 = tmp0 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_poi_fused_add_div_erf_mul_11(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 = 0.5
tmp2 = tmp0 * tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp0 * tmp3
tmp5 = libdevice.erf(tmp4)
tmp6 = 1.0
tmp7 = tmp5 + tmp6
tmp8 = tmp2 * tmp7
tl.store(out_ptr0 + x0, tmp8, xmask)
@triton.jit
def triton_poi_fused_add_12(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
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_out_ptr0 + x2, xmask)
tmp4 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tl.store(in_out_ptr0 + x2, tmp6, 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,
primals_18, primals_19) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (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,), (1,))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (1, 1, 4), (4, 4, 1))
assert_size_stride(primals_8, (1, 1, 4), (4, 4, 1))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4, 4), (4, 1))
assert_size_stride(primals_13, (4, 4), (4, 1))
assert_size_stride(primals_14, (4,), (1,))
assert_size_stride(primals_15, (4,), (1,))
assert_size_stride(primals_16, (4, 4), (4, 1))
assert_size_stride(primals_17, (4,), (1,))
assert_size_stride(primals_18, (4, 4), (4, 1))
assert_size_stride(primals_19, (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_sub_0[grid(64)](primals_1, buf0, 64, XBLOCK=
64, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_mean_mul_pow_sqrt_1[grid(64)](primals_2,
buf0, primals_3, buf1, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_2
del primals_3
buf2 = reinterpret_tensor(buf0, (16, 4), (4, 1), 0)
del buf0
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((12,), (1,), torch.float32)
triton_poi_fused_cat_2[grid(12)](primals_4, primals_5, primals_6,
buf3, 12, XBLOCK=16, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(buf3, (4,), (1,), 4),
reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_12, (4, 4), (1, 4), 0), alpha=1,
beta=1, out=buf4)
buf5 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(buf3, (4,), (1,), 8),
reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), alpha=1,
beta=1, out=buf5)
del buf3
buf6 = empty_strided_cuda((5, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_cat_3[grid(80)](buf5, primals_8, buf6, 80, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_8
buf7 = reinterpret_tensor(buf2, (16, 4, 1), (1, 16, 64), 0)
del buf2
triton_poi_fused_mul_4[grid(64)](buf7, primals_4, primals_5,
primals_6, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_4
del primals_5
del primals_6
buf8 = empty_strided_cuda((5, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_cat_3[grid(80)](buf4, primals_7, buf8, 80, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_7
buf9 = empty_strided_cuda((16, 4, 5), (20, 5, 1), torch.float32)
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 1, 5), (1, 0,
16), 0), out=buf9)
buf10 = reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 64), 0)
del buf4
buf11 = reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 64), 0)
del buf5
triton_poi_fused__softmax_5[grid(64)](buf9, buf10, buf11, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf12 = buf9
del buf9
triton_poi_fused__softmax_6[grid(320)](buf12, buf10, buf11, 320,
XBLOCK=256, num_warps=4, num_stages=1)
buf13 = reinterpret_tensor(buf11, (16, 4, 1), (4, 1, 1), 0)
del buf11
extern_kernels.bmm(buf12, reinterpret_tensor(buf6, (16, 5, 1), (1,
16, 0), 0), out=buf13)
buf14 = reinterpret_tensor(buf10, (4, 16, 1), (16, 1, 1), 0)
del buf10
triton_poi_fused_clone_7[grid(4, 16)](buf13, buf14, 4, 16, XBLOCK=
16, YBLOCK=4, num_warps=1, num_stages=1)
buf15 = reinterpret_tensor(buf13, (16, 4), (4, 1), 0)
del buf13
extern_kernels.addmm(primals_10, reinterpret_tensor(buf14, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf15)
del primals_10
buf16 = empty_strided_cuda((4, 4, 5), (20, 5, 1), torch.float32)
triton_poi_fused_mean_8[grid(80)](buf12, buf16, 80, XBLOCK=128,
num_warps=4, num_stages=1)
buf17 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf18 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_mean_pow_sub_9[grid(16)](primals_1, buf15,
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_add_div_mean_mul_sqrt_sub_10[grid(64)](primals_14,
primals_1, buf15, buf17, buf18, primals_15, buf19, 64, XBLOCK=
64, num_warps=1, num_stages=1)
del buf17
del buf18
del primals_15
buf20 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_17, reinterpret_tensor(buf19, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_16, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf20)
del primals_17
buf21 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_erf_mul_11[grid(64)](buf20, buf21, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf22 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf21, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_18, (4, 4), (1, 4), 0), out=buf22)
buf23 = reinterpret_tensor(buf22, (4, 4, 4), (16, 4, 1), 0)
del buf22
triton_poi_fused_add_12[grid(64)](buf23, primals_1, buf15,
primals_19, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_19
return buf23, buf16, primals_1, primals_14, reinterpret_tensor(buf1, (
16, 4), (4, 1), 0), buf12, reinterpret_tensor(buf14, (16, 4), (4, 1), 0
), buf15, reinterpret_tensor(buf19, (16, 4), (4, 1), 0
), buf20, reinterpret_tensor(buf21, (16, 4), (4, 1), 0
), primals_18, primals_16, primals_9, reinterpret_tensor(buf6, (16,
1, 5), (1, 1, 16), 0), reinterpret_tensor(buf7, (16, 1, 4), (1, 1,
16), 0), reinterpret_tensor(buf8, (16, 5, 1), (1, 16, 1), 0
), primals_13, primals_12, primals_11
def gelu(x):
"""Implementation of the gelu activation function.
For information: OpenAI GPT's gelu is slightly different
(and gives slightly different results):
0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
def utils_softmax(x, dim: 'int', onnx_trace: 'bool'=False):
if onnx_trace:
return F.softmax(x.float(), dim=dim)
else:
return F.softmax(x, dim=dim, dtype=torch.float32)
def with_incremental_state(cls):
cls.__bases__ = (FairseqIncrementalState,) + tuple(b for b in cls.
__bases__ if b != FairseqIncrementalState)
return cls
class ESM1LayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12, affine=True):
"""Construct a layernorm layer in the TF style (eps inside the sqrt)."""
super().__init__()
self.hidden_size = (hidden_size,) if isinstance(hidden_size, int
) else tuple(hidden_size)
self.eps = eps
self.affine = bool(affine)
if self.affine:
self.weight = nn.Parameter(torch.ones(hidden_size))
self.bias = nn.Parameter(torch.zeros(hidden_size))
else:
self.weight, self.bias = None, None
def forward(self, x):
dims = tuple(-(i + 1) for i in range(len(self.hidden_size)))
means = x.mean(dims, keepdim=True)
x_zeromean = x - means
variances = x_zeromean.pow(2).mean(dims, keepdim=True)
x = x_zeromean / torch.sqrt(variances + self.eps)
if self.affine:
x = self.weight * x + self.bias
return x
class FairseqIncrementalState(object):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.init_incremental_state()
def init_incremental_state(self):
self._incremental_state_id = str(uuid.uuid4())
def _get_full_incremental_state_key(self, key: 'str') ->str:
return '{}.{}'.format(self._incremental_state_id, key)
def get_incremental_state(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]', key: 'str'
) ->Optional[Dict[str, Optional[Tensor]]]:
"""Helper for getting incremental state for an nn.Module."""
full_key = self._get_full_incremental_state_key(key)
if incremental_state is None or full_key not in incremental_state:
return None
return incremental_state[full_key]
def set_incremental_state(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]', key: 'str',
value: 'Dict[str, Optional[Tensor]]') ->Optional[Dict[str, Dict[str,
Optional[Tensor]]]]:
"""Helper for setting incremental state for an nn.Module."""
if incremental_state is not None:
full_key = self._get_full_incremental_state_key(key)
incremental_state[full_key] = value
return incremental_state
@with_incremental_state
class MultiheadAttention(nn.Module):
"""Multi-headed attention.
See "Attention Is All You Need" for more details.
"""
def __init__(self, embed_dim, num_heads, kdim=None, vdim=None, dropout=
0.0, bias=True, add_bias_kv=False, add_zero_attn=False,
self_attention=False, encoder_decoder_attention=False):
super().__init__()
self.embed_dim = embed_dim
self.kdim = kdim if kdim is not None else embed_dim
self.vdim = vdim if vdim is not None else embed_dim
self.qkv_same_dim = self.kdim == embed_dim and self.vdim == embed_dim
self.num_heads = num_heads
self.dropout = dropout
self.head_dim = embed_dim // num_heads
assert self.head_dim * num_heads == self.embed_dim, 'embed_dim must be divisible by num_heads'
self.scaling = self.head_dim ** -0.5
self.self_attention = self_attention
self.encoder_decoder_attention = encoder_decoder_attention
assert not self.self_attention or self.qkv_same_dim, 'Self-attention requires query, key and value to be of the same size'
self.k_proj = nn.Linear(self.kdim, embed_dim, bias=bias)
self.v_proj = nn.Linear(self.vdim, embed_dim, bias=bias)
self.q_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
self.out_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
if add_bias_kv:
self.bias_k = Parameter(torch.Tensor(1, 1, embed_dim))
self.bias_v = Parameter(torch.Tensor(1, 1, embed_dim))
else:
self.bias_k = self.bias_v = None
self.add_zero_attn = add_zero_attn
self.reset_parameters()
self.onnx_trace = False
self.enable_torch_version = False
if hasattr(F, 'multi_head_attention_forward'):
self.enable_torch_version = True
else:
self.enable_torch_version = False
def prepare_for_onnx_export_(self):
self.onnx_trace = True
def reset_parameters(self):
if self.qkv_same_dim:
nn.init.xavier_uniform_(self.k_proj.weight, gain=1 / math.sqrt(2))
nn.init.xavier_uniform_(self.v_proj.weight, gain=1 / math.sqrt(2))
nn.init.xavier_uniform_(self.q_proj.weight, gain=1 / math.sqrt(2))
else:
nn.init.xavier_uniform_(self.k_proj.weight)
nn.init.xavier_uniform_(self.v_proj.weight)
nn.init.xavier_uniform_(self.q_proj.weight)
nn.init.xavier_uniform_(self.out_proj.weight)
if self.out_proj.bias is not None:
nn.init.constant_(self.out_proj.bias, 0.0)
if self.bias_k is not None:
nn.init.xavier_normal_(self.bias_k)
if self.bias_v is not None:
nn.init.xavier_normal_(self.bias_v)
def forward(self, query, key: 'Optional[Tensor]', value:
'Optional[Tensor]', key_padding_mask: 'Optional[Tensor]'=None,
incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]'=None,
need_weights: 'bool'=True, static_kv: 'bool'=False, attn_mask:
'Optional[Tensor]'=None, before_softmax: 'bool'=False,
need_head_weights: 'bool'=False) ->Tuple[Tensor, Optional[Tensor]]:
"""Input shape: Time x Batch x Channel
Args:
key_padding_mask (ByteTensor, optional): mask to exclude
keys that are pads, of shape `(batch, src_len)`, where
padding elements are indicated by 1s.
need_weights (bool, optional): return the attention weights,
averaged over heads (default: False).
attn_mask (ByteTensor, optional): typically used to
implement causal attention, where the mask prevents the
attention from looking forward in time (default: None).
before_softmax (bool, optional): return the raw attention
weights and values before the attention softmax.
need_head_weights (bool, optional): return the attention
weights for each head. Implies *need_weights*. Default:
return the average attention weights over all heads.
"""
if need_head_weights:
need_weights = True
tgt_len, bsz, embed_dim = query.size()
assert embed_dim == self.embed_dim
assert list(query.size()) == [tgt_len, bsz, embed_dim]
if (self.enable_torch_version and not self.onnx_trace and
incremental_state is None and not static_kv and not torch.jit.
is_scripting() and not need_head_weights):
assert key is not None and value is not None
return F.multi_head_attention_forward(query, key, value, self.
embed_dim, self.num_heads, torch.empty([0]), torch.cat((
self.q_proj.bias, self.k_proj.bias, self.v_proj.bias)),
self.bias_k, self.bias_v, self.add_zero_attn, self.dropout,
self.out_proj.weight, self.out_proj.bias, self.training,
key_padding_mask, need_weights, 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)
if incremental_state is not None:
saved_state = self._get_input_buffer(incremental_state)
if saved_state is not None and 'prev_key' in saved_state:
if static_kv:
assert self.encoder_decoder_attention and not self.self_attention
key = value = None
else:
saved_state = None
if self.self_attention:
q = self.q_proj(query)
k = self.k_proj(query)
v = self.v_proj(query)
elif self.encoder_decoder_attention:
q = self.q_proj(query)
if key is None:
assert value is None
k = v = None
else:
k = self.k_proj(key)
v = self.v_proj(key)
else:
assert key is not None and value is not None
q = self.q_proj(query)
k = self.k_proj(key)
v = self.v_proj(value)
q *= self.scaling
if self.bias_k is not None:
assert self.bias_v is not None
k = torch.cat([k, self.bias_k.repeat(1, bsz, 1)])
v = torch.cat([v, self.bias_v.repeat(1, bsz, 1)])
if attn_mask is not None:
attn_mask = torch.cat([attn_mask, attn_mask.new_zeros(
attn_mask.size(0), 1)], dim=1)
if key_padding_mask is not None:
key_padding_mask = torch.cat([key_padding_mask,
key_padding_mask.new_zeros(key_padding_mask.size(0), 1)
], dim=1)
q = q.contiguous().view(tgt_len, bsz * self.num_heads, self.head_dim
).transpose(0, 1)
if k is not None:
k = k.contiguous().view(-1, bsz * self.num_heads, self.head_dim
).transpose(0, 1)
if v is not None:
v = v.contiguous().view(-1, bsz * self.num_heads, self.head_dim
).transpose(0, 1)
if saved_state is not None:
if 'prev_key' in saved_state:
_prev_key = saved_state['prev_key']
assert _prev_key is not None
prev_key = _prev_key.view(bsz * self.num_heads, -1, self.
head_dim)
if static_kv:
k = prev_key
else:
assert k is not None
k = torch.cat([prev_key, k], dim=1)
if 'prev_value' in saved_state:
_prev_value = saved_state['prev_value']
assert _prev_value is not None
prev_value = _prev_value.view(bsz * self.num_heads, -1,
self.head_dim)
if static_kv:
v = prev_value
else:
assert v is not None
v = torch.cat([prev_value, v], dim=1)
prev_key_padding_mask: 'Optional[Tensor]' = None
if 'prev_key_padding_mask' in saved_state:
prev_key_padding_mask = saved_state['prev_key_padding_mask']
assert k is not None and v is not None
key_padding_mask = (MultiheadAttention.
_append_prev_key_padding_mask(key_padding_mask=
key_padding_mask, prev_key_padding_mask=
prev_key_padding_mask, batch_size=bsz, src_len=k.size(1),
static_kv=static_kv))
saved_state['prev_key'] = k.view(bsz, self.num_heads, -1, self.
head_dim)
saved_state['prev_value'] = v.view(bsz, self.num_heads, -1,
self.head_dim)
saved_state['prev_key_padding_mask'] = key_padding_mask
assert incremental_state is not None
incremental_state = self._set_input_buffer(incremental_state,
saved_state)
assert k is not None
src_len = k.size(1)
if key_padding_mask is not None and key_padding_mask.dim() == 0:
key_padding_mask = None
if key_padding_mask is not None:
assert key_padding_mask.size(0) == bsz
assert key_padding_mask.size(1) == src_len
if self.add_zero_attn:
assert v is not None
src_len += 1
k = torch.cat([k, k.new_zeros((k.size(0), 1) + k.size()[2:])],
dim=1)
v = torch.cat([v, v.new_zeros((v.size(0), 1) + v.size()[2:])],
dim=1)
if attn_mask is not None:
attn_mask = torch.cat([attn_mask, attn_mask.new_zeros(
attn_mask.size(0), 1)], dim=1)
if key_padding_mask is not None:
key_padding_mask = torch.cat([key_padding_mask, torch.zeros
(key_padding_mask.size(0), 1).type_as(key_padding_mask)
], dim=1)
attn_weights = torch.bmm(q, k.transpose(1, 2))
attn_weights = MultiheadAttention.apply_sparse_mask(attn_weights,
tgt_len, src_len, bsz)
assert list(attn_weights.size()) == [bsz * self.num_heads, tgt_len,
src_len]
if attn_mask is not None:
attn_mask = attn_mask.unsqueeze(0)
if self.onnx_trace:
attn_mask = attn_mask.repeat(attn_weights.size(0), 1, 1)
attn_weights += attn_mask
if key_padding_mask is not None:
attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len,
src_len)
attn_weights = attn_weights.masked_fill(key_padding_mask.
unsqueeze(1).unsqueeze(2), float('-inf'))
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len,
src_len)
if before_softmax:
return attn_weights, v
attn_weights_float = utils_softmax(attn_weights, dim=-1, onnx_trace
=self.onnx_trace)
attn_weights = attn_weights_float.type_as(attn_weights)
attn_probs = F.dropout(attn_weights_float.type_as(attn_weights), p=
self.dropout, training=self.training)
assert v is not None
attn = torch.bmm(attn_probs, v)
assert list(attn.size()) == [bsz * self.num_heads, tgt_len, self.
head_dim]
if self.onnx_trace and attn.size(1) == 1:
attn = attn.contiguous().view(tgt_len, bsz, embed_dim)
else:
attn = attn.transpose(0, 1).contiguous().view(tgt_len, bsz,
embed_dim)
attn = self.out_proj(attn)
attn_weights: 'Optional[Tensor]' = None
if need_weights:
attn_weights = attn_weights_float.view(bsz, self.num_heads,
tgt_len, src_len).transpose(1, 0)
if not need_head_weights:
attn_weights = attn_weights.mean(dim=0)
return attn, attn_weights
@staticmethod
def _append_prev_key_padding_mask(key_padding_mask: 'Optional[Tensor]',
prev_key_padding_mask: 'Optional[Tensor]', batch_size: 'int',
src_len: 'int', static_kv: 'bool') ->Optional[Tensor]:
if prev_key_padding_mask is not None and static_kv:
new_key_padding_mask = prev_key_padding_mask
elif prev_key_padding_mask is not None and key_padding_mask is not None:
new_key_padding_mask = torch.cat([prev_key_padding_mask.float(),
key_padding_mask.float()], dim=1)
elif prev_key_padding_mask is not None:
filler = torch.zeros((batch_size, src_len -
prev_key_padding_mask.size(1)), device=
prev_key_padding_mask.device)
new_key_padding_mask = torch.cat([prev_key_padding_mask.float(),
filler.float()], dim=1)
elif key_padding_mask is not None:
filler = torch.zeros((batch_size, src_len - key_padding_mask.
size(1)), device=key_padding_mask.device)
new_key_padding_mask = torch.cat([filler.float(),
key_padding_mask.float()], dim=1)
else:
new_key_padding_mask = prev_key_padding_mask
return new_key_padding_mask
@torch.jit.export
def reorder_incremental_state(self, incremental_state:
'Dict[str, Dict[str, Optional[Tensor]]]', new_order: 'Tensor'):
"""Reorder buffered internal state (for incremental generation)."""
input_buffer = self._get_input_buffer(incremental_state)
if input_buffer is not None:
for k in input_buffer.keys():
input_buffer_k = input_buffer[k]
if input_buffer_k is not None:
if self.encoder_decoder_attention and input_buffer_k.size(0
) == new_order.size(0):
break
input_buffer[k] = input_buffer_k.index_select(0, new_order)
incremental_state = self._set_input_buffer(incremental_state,
input_buffer)
return incremental_state
def _get_input_buffer(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]') ->Dict[str,
Optional[Tensor]]:
result = self.get_incremental_state(incremental_state, 'attn_state')
if result is not None:
return result
else:
empty_result: 'Dict[str, Optional[Tensor]]' = {}
return empty_result
def _set_input_buffer(self, incremental_state:
'Dict[str, Dict[str, Optional[Tensor]]]', buffer:
'Dict[str, Optional[Tensor]]'):
return self.set_incremental_state(incremental_state, 'attn_state',
buffer)
def apply_sparse_mask(attn_weights, tgt_len: 'int', src_len: 'int', bsz:
'int'):
return attn_weights
def upgrade_state_dict_named(self, state_dict, name):
prefix = name + '.' if name != '' else ''
items_to_add = {}
keys_to_remove = []
for k in state_dict.keys():
if k.endswith(prefix + 'in_proj_weight'):
dim = int(state_dict[k].shape[0] / 3)
items_to_add[prefix + 'q_proj.weight'] = state_dict[k][:dim]
items_to_add[prefix + 'k_proj.weight'] = state_dict[k][dim:
2 * dim]
items_to_add[prefix + 'v_proj.weight'] = state_dict[k][2 * dim:
]
keys_to_remove.append(k)
k_bias = prefix + 'in_proj_bias'
if k_bias in state_dict.keys():
dim = int(state_dict[k].shape[0] / 3)
items_to_add[prefix + 'q_proj.bias'] = state_dict[k_bias][:
dim]
items_to_add[prefix + 'k_proj.bias'] = state_dict[k_bias][
dim:2 * dim]
items_to_add[prefix + 'v_proj.bias'] = state_dict[k_bias][
2 * dim:]
keys_to_remove.append(prefix + 'in_proj_bias')
for k in keys_to_remove:
del state_dict[k]
for key, value in items_to_add.items():
state_dict[key] = value
class TransformerLayerNew(nn.Module):
"""Transformer layer block."""
def __init__(self, embed_dim, ffn_embed_dim, attention_heads,
add_bias_kv=True, use_esm1b_layer_norm=False):
super().__init__()
self.embed_dim = embed_dim
self.ffn_embed_dim = ffn_embed_dim
self.attention_heads = attention_heads
self._init_submodules(add_bias_kv, use_esm1b_layer_norm)
def _init_submodules(self, add_bias_kv, use_esm1b_layer_norm):
BertLayerNorm = (ESM1bLayerNorm if use_esm1b_layer_norm else
ESM1LayerNorm)
self.self_attn = MultiheadAttention(self.embed_dim, self.
attention_heads, add_bias_kv=add_bias_kv, add_zero_attn=False)
self.self_attn_layer_norm = BertLayerNorm(self.embed_dim)
self.fc1 = nn.Linear(self.embed_dim, self.ffn_embed_dim)
self.fc2 = nn.Linear(self.ffn_embed_dim, self.embed_dim)
self.final_layer_norm = BertLayerNorm(self.embed_dim)
def forward(self, input_0):
primals_7 = self.self_attn.bias_k
primals_8 = self.self_attn.bias_v
primals_9 = self.self_attn.k_proj.weight
primals_2 = self.self_attn.k_proj.bias
primals_11 = self.self_attn.v_proj.weight
primals_3 = self.self_attn.v_proj.bias
primals_12 = self.self_attn.q_proj.weight
primals_4 = self.self_attn.q_proj.bias
primals_13 = self.self_attn.out_proj.weight
primals_5 = self.self_attn.out_proj.bias
primals_6 = self.self_attn_layer_norm.weight
primals_10 = self.self_attn_layer_norm.bias
primals_16 = self.fc1.weight
primals_14 = self.fc1.bias
primals_18 = self.fc2.weight
primals_15 = self.fc2.bias
primals_17 = self.final_layer_norm.weight
primals_19 = self.final_layer_norm.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, primals_16, primals_17, primals_18, primals_19])
return output[0], output[1]
| boxiangliu/esm | TransformerLayer | false | 1,634 | [
"MIT"
] | 0 | 3c143d99103e0ea38a9455f30a73cd9c87376606 | https://github.com/boxiangliu/esm/tree/3c143d99103e0ea38a9455f30a73cd9c87376606 | import math
import torch
import uuid
from torch import Tensor
import torch.nn as nn
from typing import Tuple
import torch.nn.functional as F
from typing import Optional
from typing import Dict
from torch.nn import Parameter
def gelu(x):
"""Implementation of the gelu activation function.
For information: OpenAI GPT's gelu is slightly different
(and gives slightly different results):
0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
"""
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
def utils_softmax(x, dim: 'int', onnx_trace: 'bool'=False):
if onnx_trace:
return F.softmax(x.float(), dim=dim)
else:
return F.softmax(x, dim=dim, dtype=torch.float32)
def with_incremental_state(cls):
cls.__bases__ = (FairseqIncrementalState,) + tuple(b for b in cls.
__bases__ if b != FairseqIncrementalState)
return cls
class ESM1LayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12, affine=True):
"""Construct a layernorm layer in the TF style (eps inside the sqrt)."""
super().__init__()
self.hidden_size = (hidden_size,) if isinstance(hidden_size, int
) else tuple(hidden_size)
self.eps = eps
self.affine = bool(affine)
if self.affine:
self.weight = nn.Parameter(torch.ones(hidden_size))
self.bias = nn.Parameter(torch.zeros(hidden_size))
else:
self.weight, self.bias = None, None
def forward(self, x):
dims = tuple(-(i + 1) for i in range(len(self.hidden_size)))
means = x.mean(dims, keepdim=True)
x_zeromean = x - means
variances = x_zeromean.pow(2).mean(dims, keepdim=True)
x = x_zeromean / torch.sqrt(variances + self.eps)
if self.affine:
x = self.weight * x + self.bias
return x
class FairseqIncrementalState(object):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.init_incremental_state()
def init_incremental_state(self):
self._incremental_state_id = str(uuid.uuid4())
def _get_full_incremental_state_key(self, key: 'str') ->str:
return '{}.{}'.format(self._incremental_state_id, key)
def get_incremental_state(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]', key: 'str'
) ->Optional[Dict[str, Optional[Tensor]]]:
"""Helper for getting incremental state for an nn.Module."""
full_key = self._get_full_incremental_state_key(key)
if incremental_state is None or full_key not in incremental_state:
return None
return incremental_state[full_key]
def set_incremental_state(self, incremental_state:
'Optional[Dict[str, Dict[str, Optional[Tensor]]]]', key: 'str',
value: 'Dict[str, Optional[Tensor]]') ->Optional[Dict[str, Dict[str,
Optional[Tensor]]]]:
"""Helper for setting incremental state for an nn.Module."""
if incremental_state is not None:
full_key = self._get_full_incremental_state_key(key)
incremental_state[full_key] = value
return incremental_state
@with_incremental_state
class MultiheadAttention(nn.Module):
"""Multi-headed attention.
See "Attention Is All You Need" for more details.
"""
def __init__(self, embed_dim, num_heads, kdim=None, vdim=None, dropout=
0.0, bias=True, add_bias_kv=False, add_zero_attn=False,
self_attention=False, encoder_decoder_attention=False):
super().__init__()
self.embed_dim = embed_dim
self.kdim = kdim if kdim is not None else embed_dim
self.vdim = vdim if vdim is not None else embed_dim
self.qkv_same_dim = self.kdim == embed_dim and self.vdim == embed_dim
self.num_heads = num_heads
self.dropout = dropout
self.head_dim = embed_dim // num_heads
assert self.head_dim * num_heads == self.embe
# ... truncated (>4000 chars) for memory efficiency |
my_Hingeloss | # 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_6/inductor_cache/xo/cxo7navcvfdngvouei3426orwczlourihu7qpxajq6oy5kvzifnp.py
# Topologically Sorted Source Nodes: [sub, mul_1, max_1, mul, pos, sub_1, loss, setitem, loss_1], Original ATen: [aten.rsub, aten.mul, aten.max, aten.sum, aten.sub, aten.add, aten.lift_fresh, aten.index_put, aten.mean]
# Source node to ATen node mapping:
# loss => add
# loss_1 => mean
# max_1 => max_1
# mul => mul
# mul_1 => mul_1
# pos => sum_1
# setitem => full_default, index_put
# sub => sub
# sub_1 => sub_1
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %arg0_1), kwargs = {})
# %max_1 : [num_users=1] = call_function[target=torch.ops.aten.max.dim](args = (%mul_1, 2), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [2]), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%getitem, %sum_1), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_1, 1), 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: cpu, pin_memory: False})
# %index_put : [num_users=1] = call_function[target=torch.ops.aten.index_put_.default](args = (%add, [%lt], %full_default), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%index_put,), kwargs = {})
triton_per_fused_add_index_put_lift_fresh_max_mean_mul_rsub_sub_sum_0 = async_compile.triton('triton_per_fused_add_index_put_lift_fresh_max_mean_mul_rsub_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=[1, 64],
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_index_put_lift_fresh_max_mean_mul_rsub_sub_sum_0', 'mutated_arg_names': ['in_out_ptr1'], '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_add_index_put_lift_fresh_max_mean_mul_rsub_sub_sum_0(in_out_ptr1, in_ptr0, in_ptr1, 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)
r0 = rindex % 4
r1 = (rindex // 4)
r2 = rindex
tmp0 = tl.load(in_ptr0 + (r0 + (16*r1)), None)
tmp3 = tl.load(in_ptr1 + (r0 + (16*r1)), None)
tmp5 = tl.load(in_ptr0 + (4 + r0 + (16*r1)), None)
tmp7 = tl.load(in_ptr1 + (4 + r0 + (16*r1)), None)
tmp10 = tl.load(in_ptr0 + (8 + r0 + (16*r1)), None)
tmp12 = tl.load(in_ptr1 + (8 + r0 + (16*r1)), None)
tmp15 = tl.load(in_ptr0 + (12 + r0 + (16*r1)), None)
tmp17 = tl.load(in_ptr1 + (12 + r0 + (16*r1)), None)
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp2 * tmp3
tmp6 = tmp1 - tmp5
tmp8 = tmp6 * tmp7
tmp9 = triton_helpers.maximum(tmp4, tmp8)
tmp11 = tmp1 - tmp10
tmp13 = tmp11 * tmp12
tmp14 = triton_helpers.maximum(tmp9, tmp13)
tmp16 = tmp1 - tmp15
tmp18 = tmp16 * tmp17
tmp19 = triton_helpers.maximum(tmp14, tmp18)
tmp20 = tmp3 * tmp0
tmp21 = tmp7 * tmp5
tmp22 = tmp20 + tmp21
tmp23 = tmp12 * tmp10
tmp24 = tmp22 + tmp23
tmp25 = tmp17 * tmp15
tmp26 = tmp24 + tmp25
tmp27 = tmp19 - tmp26
tmp28 = tmp27 + tmp1
tmp29 = 0.0
tmp30 = tmp28 < tmp29
tmp31 = tl.where(tmp30, tmp29, tmp28)
tmp32 = tl.broadcast_to(tmp31, [XBLOCK, RBLOCK])
tmp34 = tl.sum(tmp32, 1)[:, None]
tmp35 = 64.0
tmp36 = tmp34 / tmp35
tl.debug_barrier()
tl.store(in_out_ptr1 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp36, 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)
buf2 = empty_strided_cuda((), (), torch.float32)
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [sub, mul_1, max_1, mul, pos, sub_1, loss, setitem, loss_1], Original ATen: [aten.rsub, aten.mul, aten.max, aten.sum, aten.sub, aten.add, aten.lift_fresh, aten.index_put, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_add_index_put_lift_fresh_max_mean_mul_rsub_sub_sum_0.run(buf3, arg1_1, arg0_1, 1, 64, 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 my_Hingeloss(nn.Module):
def __init__(self):
super(my_Hingeloss, self).__init__()
def forward(self, output, target):
pos = torch.sum(output * target, 2)
neg = torch.max((1 - target) * output, 2)
loss = neg[0] - pos + 1
loss[loss < 0] = 0
loss = torch.mean(loss)
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
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_index_put_lift_fresh_max_mean_mul_rsub_sub_sum_0(
in_out_ptr1, in_ptr0, in_ptr1, 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)
r0 = rindex % 4
r1 = rindex // 4
tmp0 = tl.load(in_ptr0 + (r0 + 16 * r1), None)
tmp3 = tl.load(in_ptr1 + (r0 + 16 * r1), None)
tmp5 = tl.load(in_ptr0 + (4 + r0 + 16 * r1), None)
tmp7 = tl.load(in_ptr1 + (4 + r0 + 16 * r1), None)
tmp10 = tl.load(in_ptr0 + (8 + r0 + 16 * r1), None)
tmp12 = tl.load(in_ptr1 + (8 + r0 + 16 * r1), None)
tmp15 = tl.load(in_ptr0 + (12 + r0 + 16 * r1), None)
tmp17 = tl.load(in_ptr1 + (12 + r0 + 16 * r1), None)
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp2 * tmp3
tmp6 = tmp1 - tmp5
tmp8 = tmp6 * tmp7
tmp9 = triton_helpers.maximum(tmp4, tmp8)
tmp11 = tmp1 - tmp10
tmp13 = tmp11 * tmp12
tmp14 = triton_helpers.maximum(tmp9, tmp13)
tmp16 = tmp1 - tmp15
tmp18 = tmp16 * tmp17
tmp19 = triton_helpers.maximum(tmp14, tmp18)
tmp20 = tmp3 * tmp0
tmp21 = tmp7 * tmp5
tmp22 = tmp20 + tmp21
tmp23 = tmp12 * tmp10
tmp24 = tmp22 + tmp23
tmp25 = tmp17 * tmp15
tmp26 = tmp24 + tmp25
tmp27 = tmp19 - tmp26
tmp28 = tmp27 + tmp1
tmp29 = 0.0
tmp30 = tmp28 < tmp29
tmp31 = tl.where(tmp30, tmp29, tmp28)
tmp32 = tl.broadcast_to(tmp31, [XBLOCK, RBLOCK])
tmp34 = tl.sum(tmp32, 1)[:, None]
tmp35 = 64.0
tmp36 = tmp34 / tmp35
tl.debug_barrier()
tl.store(in_out_ptr1 + tl.full([XBLOCK, 1], 0, tl.int32), tmp36, 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)
buf2 = empty_strided_cuda((), (), torch.float32)
buf3 = buf2
del buf2
get_raw_stream(0)
triton_per_fused_add_index_put_lift_fresh_max_mean_mul_rsub_sub_sum_0[
grid(1)](buf3, arg1_1, arg0_1, 1, 64, XBLOCK=1, num_warps=2,
num_stages=1)
del arg0_1
del arg1_1
return buf3,
class my_HingelossNew(nn.Module):
def __init__(self):
super(my_HingelossNew, 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]
| carsault/chord_sequence_prediction | my_Hingeloss | false | 1,635 | [
"MIT"
] | 0 | 6eb539a963ca6350bcf0c88b8d8756775ad7c488 | https://github.com/carsault/chord_sequence_prediction/tree/6eb539a963ca6350bcf0c88b8d8756775ad7c488 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, output, target):
pos = torch.sum(output * target, 2)
neg = torch.max((1 - target) * output, 2)
loss = neg[0] - pos + 1
loss[loss < 0] = 0
loss = torch.mean(loss)
return loss
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SigmoidCrossEntropyLoss | # 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_6/inductor_cache/bv/cbvp5u5kif72bdcqavxdzrj2w7nrgeddwxfpnbuaa5qdr5ycayqv.py
# Topologically Sorted Source Nodes: [element_loss, loss, loss_1], Original ATen: [aten.binary_cross_entropy_with_logits, aten.sum, aten.mean]
# Source node to ATen node mapping:
# element_loss => abs_1, exp, full_default, log1p, minimum, mul, neg, sub, sub_1, sub_2
# loss => sum_1
# loss_1 => mean
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %arg0_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), 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 = (%full_default, %arg0_1), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%arg0_1,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %sub_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%sub_2, [1]), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sum_1,), kwargs = {})
triton_per_fused_binary_cross_entropy_with_logits_mean_sum_0 = async_compile.triton('triton_per_fused_binary_cross_entropy_with_logits_mean_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, 4],
reduction_hint=ReductionHint.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': {3: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=(3,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_binary_cross_entropy_with_logits_mean_sum_0', 'mutated_arg_names': ['in_out_ptr0'], '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_binary_cross_entropy_with_logits_mean_sum_0(in_out_ptr0, in_ptr0, in_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 + (4*r0), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (4*r0), None, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp37 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp39 = tl.load(in_ptr1 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp2 * tmp3
tmp5 = 0.0
tmp6 = triton_helpers.minimum(tmp5, tmp3)
tmp7 = tl_math.abs(tmp3)
tmp8 = -tmp7
tmp9 = tl_math.exp(tmp8)
tmp10 = libdevice.log1p(tmp9)
tmp11 = tmp6 - tmp10
tmp12 = tmp4 - tmp11
tmp14 = tmp1 - tmp13
tmp16 = tmp14 * tmp15
tmp17 = triton_helpers.minimum(tmp5, tmp15)
tmp18 = tl_math.abs(tmp15)
tmp19 = -tmp18
tmp20 = tl_math.exp(tmp19)
tmp21 = libdevice.log1p(tmp20)
tmp22 = tmp17 - tmp21
tmp23 = tmp16 - tmp22
tmp24 = tmp12 + tmp23
tmp26 = tmp1 - tmp25
tmp28 = tmp26 * tmp27
tmp29 = triton_helpers.minimum(tmp5, tmp27)
tmp30 = tl_math.abs(tmp27)
tmp31 = -tmp30
tmp32 = tl_math.exp(tmp31)
tmp33 = libdevice.log1p(tmp32)
tmp34 = tmp29 - tmp33
tmp35 = tmp28 - tmp34
tmp36 = tmp24 + tmp35
tmp38 = tmp1 - tmp37
tmp40 = tmp38 * tmp39
tmp41 = triton_helpers.minimum(tmp5, tmp39)
tmp42 = tl_math.abs(tmp39)
tmp43 = -tmp42
tmp44 = tl_math.exp(tmp43)
tmp45 = libdevice.log1p(tmp44)
tmp46 = tmp41 - tmp45
tmp47 = tmp40 - tmp46
tmp48 = tmp36 + tmp47
tmp49 = tl.broadcast_to(tmp48, [XBLOCK, RBLOCK])
tmp51 = tl.sum(tmp49, 1)[:, None]
tmp52 = 4.0
tmp53 = tmp51 / tmp52
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp53, 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, 1))
assert_size_stride(arg1_1, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [element_loss, loss, loss_1], Original ATen: [aten.binary_cross_entropy_with_logits, aten.sum, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_mean_sum_0.run(buf2, arg1_1, arg0_1, 1, 4, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (buf2, )
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
from torch import Tensor
from typing import List
from typing import Optional
from typing import Union
from torch import nn
class SigmoidCrossEntropyLoss(nn.Module):
def __init__(self, class_weights: 'Optional[Union[Tensor, List]]'=None,
**kwargs):
"""
Params:
class_weights: List or 1D tensor of length equal to number of classes.
"""
super().__init__()
if class_weights:
self.loss_fn = nn.BCEWithLogitsLoss(reduction='none',
pos_weight=torch.Tensor(class_weights))
else:
self.loss_fn = nn.BCEWithLogitsLoss(reduction='none')
def forward(self, preds: 'Tensor', target: 'Tensor') ->Tensor:
if preds.ndim != 2:
raise RuntimeError(
'SigmoidCrossEntropyLoss currently supported for 2D tensors.')
element_loss = self.loss_fn(preds.type(torch.float32), target.type(
torch.float32))
loss = torch.sum(element_loss, dim=1)
loss = torch.mean(loss)
return loss
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
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch import Tensor
from typing import List
from typing import Optional
from typing import Union
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_per_fused_binary_cross_entropy_with_logits_mean_sum_0(in_out_ptr0,
in_ptr0, in_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 + 4 * r0, None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + 4 * r0, None, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp37 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp39 = tl.load(in_ptr1 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp1 - tmp0
tmp4 = tmp2 * tmp3
tmp5 = 0.0
tmp6 = triton_helpers.minimum(tmp5, tmp3)
tmp7 = tl_math.abs(tmp3)
tmp8 = -tmp7
tmp9 = tl_math.exp(tmp8)
tmp10 = libdevice.log1p(tmp9)
tmp11 = tmp6 - tmp10
tmp12 = tmp4 - tmp11
tmp14 = tmp1 - tmp13
tmp16 = tmp14 * tmp15
tmp17 = triton_helpers.minimum(tmp5, tmp15)
tmp18 = tl_math.abs(tmp15)
tmp19 = -tmp18
tmp20 = tl_math.exp(tmp19)
tmp21 = libdevice.log1p(tmp20)
tmp22 = tmp17 - tmp21
tmp23 = tmp16 - tmp22
tmp24 = tmp12 + tmp23
tmp26 = tmp1 - tmp25
tmp28 = tmp26 * tmp27
tmp29 = triton_helpers.minimum(tmp5, tmp27)
tmp30 = tl_math.abs(tmp27)
tmp31 = -tmp30
tmp32 = tl_math.exp(tmp31)
tmp33 = libdevice.log1p(tmp32)
tmp34 = tmp29 - tmp33
tmp35 = tmp28 - tmp34
tmp36 = tmp24 + tmp35
tmp38 = tmp1 - tmp37
tmp40 = tmp38 * tmp39
tmp41 = triton_helpers.minimum(tmp5, tmp39)
tmp42 = tl_math.abs(tmp39)
tmp43 = -tmp42
tmp44 = tl_math.exp(tmp43)
tmp45 = libdevice.log1p(tmp44)
tmp46 = tmp41 - tmp45
tmp47 = tmp40 - tmp46
tmp48 = tmp36 + tmp47
tmp49 = tl.broadcast_to(tmp48, [XBLOCK, RBLOCK])
tmp51 = tl.sum(tmp49, 1)[:, None]
tmp52 = 4.0
tmp53 = tmp51 / tmp52
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp53, None)
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)
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_mean_sum_0[grid(1)](
buf2, arg1_1, arg0_1, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf2,
class SigmoidCrossEntropyLossNew(nn.Module):
def __init__(self, class_weights: 'Optional[Union[Tensor, List]]'=None,
**kwargs):
"""
Params:
class_weights: List or 1D tensor of length equal to number of classes.
"""
super().__init__()
if class_weights:
self.loss_fn = nn.BCEWithLogitsLoss(reduction='none',
pos_weight=torch.Tensor(class_weights))
else:
self.loss_fn = nn.BCEWithLogitsLoss(reduction='none')
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| carlogrisetti/ludwig | SigmoidCrossEntropyLoss | false | 1,636 | [
"Apache-2.0"
] | 0 | 5c0887f14867e1577e0ddc3806c5cf7a781fb665 | https://github.com/carlogrisetti/ludwig/tree/5c0887f14867e1577e0ddc3806c5cf7a781fb665 | import torch
from torch import Tensor
from typing import List
from typing import Optional
from typing import Union
from torch import nn
class Model(nn.Module):
def __init__(self, class_weights: 'Optional[Union[Tensor, List]]'=None,
**kwargs):
"""
Params:
class_weights: List or 1D tensor of length equal to number of classes.
"""
super().__init__()
if class_weights:
self.loss_fn = nn.BCEWithLogitsLoss(reduction='none',
pos_weight=torch.Tensor(class_weights))
else:
self.loss_fn = nn.BCEWithLogitsLoss(reduction='none')
def forward(self, preds: 'Tensor', target: 'Tensor') ->Tensor:
if preds.ndim != 2:
raise RuntimeError(
'SigmoidCrossEntropyLoss currently supported for 2D tensors.')
element_loss = self.loss_fn(preds.type(torch.float32), target.type(
torch.float32))
loss = torch.sum(element_loss, dim=1)
loss = torch.mean(loss)
return loss
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return []
|
Explorer | # 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_6/inductor_cache/a2/ca2wr2cvkya5clovpxidv7ia56pdcyp7uq4omtpg5m2nr7ya3ryn.py
# Topologically Sorted Source Nodes: [a], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# a => 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_6/inductor_cache/uy/cuyajkf6kbza7jdm3zwubrxncylawdnwlrjm47ro7zzjxnyusskg.py
# Topologically Sorted Source Nodes: [tanh_2, mul, mul_1], Original ATen: [aten.tanh, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# tanh_2 => tanh_2
# Graph fragment:
# %tanh_2 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%view_5,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%tanh_2, 4), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, 16), kwargs = {})
triton_poi_fused_mul_tanh_1 = async_compile.triton('triton_poi_fused_mul_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_tanh_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_mul_tanh_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 = libdevice.tanh(tmp0)
tmp2 = 4.0
tmp3 = tmp1 * tmp2
tmp4 = 16.0
tmp5 = tmp3 * tmp4
tl.store(out_ptr0 + (x0), tmp5, 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: [a], 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: [a_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: [tanh_2, mul, mul_1], Original ATen: [aten.tanh, aten.mul]
triton_poi_fused_mul_tanh_1.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
return (buf5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, buf3, buf4, 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 numpy as np
import torch.nn as nn
def init(module, weight_init, bias_init, gain=1):
weight_init(module.weight.data, gain=gain)
bias_init(module.bias.data)
return module
class Explorer(nn.Module):
def __init__(self, state_dim, max_action, exp_regularization):
super(Explorer, self).__init__()
def init_(m):
return init(m, nn.init.orthogonal_, lambda x: nn.init.constant_
(x, 0), np.sqrt(2))
self.l1 = init_(nn.Linear(state_dim, 64))
self.l2 = init_(nn.Linear(64, 64))
self.l3 = init_(nn.Linear(64, state_dim))
self.max_action = max_action
self.exp_regularization = exp_regularization
def forward(self, state):
a = torch.tanh(self.l1(state))
a = torch.tanh(self.l2(a))
return self.max_action * torch.tanh(self.l3(a)
) * self.exp_regularization ** 2
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4, 'max_action': 4, 'exp_regularization': 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 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
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_mul_tanh_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 = libdevice.tanh(tmp0)
tmp2 = 4.0
tmp3 = tmp1 * tmp2
tmp4 = 16.0
tmp5 = tmp3 * tmp4
tl.store(out_ptr0 + x0, tmp5, 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=
128, 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=
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, 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_mul_tanh_1[grid(256)](buf4, buf5, 256, XBLOCK=128,
num_warps=4, num_stages=1)
return buf5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, buf3, buf4, primals_6, primals_4
def init(module, weight_init, bias_init, gain=1):
weight_init(module.weight.data, gain=gain)
bias_init(module.bias.data)
return module
class ExplorerNew(nn.Module):
def __init__(self, state_dim, max_action, exp_regularization):
super(ExplorerNew, self).__init__()
def init_(m):
return init(m, nn.init.orthogonal_, lambda x: nn.init.constant_
(x, 0), np.sqrt(2))
self.l1 = init_(nn.Linear(state_dim, 64))
self.l2 = init_(nn.Linear(64, 64))
self.l3 = init_(nn.Linear(64, state_dim))
self.max_action = max_action
self.exp_regularization = exp_regularization
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]
| baturaysaglam/DISCOVER | Explorer | false | 1,637 | [
"MIT"
] | 0 | 423158c84a5935ca5755ccad06ea5fe20fb57d76 | https://github.com/baturaysaglam/DISCOVER/tree/423158c84a5935ca5755ccad06ea5fe20fb57d76 | import torch
import numpy as np
import torch.nn as nn
def init(module, weight_init, bias_init, gain=1):
weight_init(module.weight.data, gain=gain)
bias_init(module.bias.data)
return module
class Model(nn.Module):
def __init__(self, state_dim, max_action, exp_regularization):
super().__init__()
def init_(m):
return init(m, nn.init.orthogonal_, lambda x: nn.init.constant_
(x, 0), np.sqrt(2))
self.l1 = init_(nn.Linear(state_dim, 64))
self.l2 = init_(nn.Linear(64, 64))
self.l3 = init_(nn.Linear(64, state_dim))
self.max_action = max_action
self.exp_regularization = exp_regularization
def forward(self, state):
a = torch.tanh(self.l1(state))
a = torch.tanh(self.l2(a))
return self.max_action * torch.tanh(self.l3(a)
) * self.exp_regularization ** 2
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
ReduceLast | # 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_6/inductor_cache/sz/cszdqebg5pns6veqfzmjstoqvgro5kzpfkqix2nafzrmf7gzh3he.py
# Topologically Sorted Source Nodes: [abs_1, amax, used, length, length_1, sequence_length, setitem], Original ATen: [aten.abs, aten.amax, aten.sign, aten.sum, aten._to_copy, aten.sub, aten.lift_fresh, aten.index_put]
# Source node to ATen node mapping:
# abs_1 => abs_1
# amax => amax
# length => sum_1
# length_1 => convert_element_type
# sequence_length => sub
# setitem => full_default, index_put
# used => sign
# Graph fragment:
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%arg0_1,), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%abs_1, [2]), kwargs = {})
# %sign : [num_users=1] = call_function[target=torch.ops.aten.sign.default](args = (%amax,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%sign, [1]), kwargs = {})
# %convert_element_type : [num_users=1] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%sum_1, torch.int32), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%convert_element_type, 1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), kwargs = {dtype: torch.int32, layout: torch.strided, device: cpu, pin_memory: False})
# %index_put : [num_users=1] = call_function[target=torch.ops.aten.index_put_.default](args = (%sub, [%lt], %full_default), kwargs = {})
triton_poi_fused__to_copy_abs_amax_index_put_lift_fresh_sign_sub_sum_0 = async_compile.triton('triton_poi_fused__to_copy_abs_amax_index_put_lift_fresh_sign_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.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i32', 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__to_copy_abs_amax_index_put_lift_fresh_sign_sub_sum_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__to_copy_abs_amax_index_put_lift_fresh_sign_sub_sum_0(in_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 + (64*x1)), xmask)
tmp2 = tl.load(in_ptr0 + (4 + x0 + (64*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (8 + x0 + (64*x1)), xmask)
tmp8 = tl.load(in_ptr0 + (12 + x0 + (64*x1)), xmask)
tmp18 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp20 = tl.load(in_ptr0 + (20 + x0 + (64*x1)), xmask)
tmp23 = tl.load(in_ptr0 + (24 + x0 + (64*x1)), xmask)
tmp26 = tl.load(in_ptr0 + (28 + x0 + (64*x1)), xmask)
tmp36 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp38 = tl.load(in_ptr0 + (36 + x0 + (64*x1)), xmask)
tmp41 = tl.load(in_ptr0 + (40 + x0 + (64*x1)), xmask)
tmp44 = tl.load(in_ptr0 + (44 + x0 + (64*x1)), xmask)
tmp54 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp56 = tl.load(in_ptr0 + (52 + x0 + (64*x1)), xmask)
tmp59 = tl.load(in_ptr0 + (56 + x0 + (64*x1)), xmask)
tmp62 = tl.load(in_ptr0 + (60 + x0 + (64*x1)), xmask)
tmp1 = tl_math.abs(tmp0)
tmp3 = tl_math.abs(tmp2)
tmp4 = triton_helpers.maximum(tmp1, tmp3)
tmp6 = tl_math.abs(tmp5)
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tl_math.abs(tmp8)
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp11 = tl.full([1], 0, tl.int32)
tmp12 = tmp11 < tmp10
tmp13 = tmp12.to(tl.int8)
tmp14 = tmp10 < tmp11
tmp15 = tmp14.to(tl.int8)
tmp16 = tmp13 - tmp15
tmp17 = tmp16.to(tmp10.dtype)
tmp19 = tl_math.abs(tmp18)
tmp21 = tl_math.abs(tmp20)
tmp22 = triton_helpers.maximum(tmp19, tmp21)
tmp24 = tl_math.abs(tmp23)
tmp25 = triton_helpers.maximum(tmp22, tmp24)
tmp27 = tl_math.abs(tmp26)
tmp28 = triton_helpers.maximum(tmp25, tmp27)
tmp29 = tmp11 < tmp28
tmp30 = tmp29.to(tl.int8)
tmp31 = tmp28 < tmp11
tmp32 = tmp31.to(tl.int8)
tmp33 = tmp30 - tmp32
tmp34 = tmp33.to(tmp28.dtype)
tmp35 = tmp17 + tmp34
tmp37 = tl_math.abs(tmp36)
tmp39 = tl_math.abs(tmp38)
tmp40 = triton_helpers.maximum(tmp37, tmp39)
tmp42 = tl_math.abs(tmp41)
tmp43 = triton_helpers.maximum(tmp40, tmp42)
tmp45 = tl_math.abs(tmp44)
tmp46 = triton_helpers.maximum(tmp43, tmp45)
tmp47 = tmp11 < tmp46
tmp48 = tmp47.to(tl.int8)
tmp49 = tmp46 < tmp11
tmp50 = tmp49.to(tl.int8)
tmp51 = tmp48 - tmp50
tmp52 = tmp51.to(tmp46.dtype)
tmp53 = tmp35 + tmp52
tmp55 = tl_math.abs(tmp54)
tmp57 = tl_math.abs(tmp56)
tmp58 = triton_helpers.maximum(tmp55, tmp57)
tmp60 = tl_math.abs(tmp59)
tmp61 = triton_helpers.maximum(tmp58, tmp60)
tmp63 = tl_math.abs(tmp62)
tmp64 = triton_helpers.maximum(tmp61, tmp63)
tmp65 = tmp11 < tmp64
tmp66 = tmp65.to(tl.int8)
tmp67 = tmp64 < tmp11
tmp68 = tmp67.to(tl.int8)
tmp69 = tmp66 - tmp68
tmp70 = tmp69.to(tmp64.dtype)
tmp71 = tmp53 + tmp70
tmp72 = tmp71.to(tl.int32)
tmp73 = tl.full([1], 1, tl.int32)
tmp74 = tmp72 - tmp73
tmp75 = tmp74 < tmp11
tmp76 = tl.where(tmp75, tmp11, tmp74)
tl.store(out_ptr1 + (x2), tmp76, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/rf/crff2v5g62gmwittje4cxykzyvx5ipq7hs2mh2ykeyfvspvt4gxy.py
# Topologically Sorted Source Nodes: [gathered], Original ATen: [aten.index]
# Source node to ATen node mapping:
# gathered => index
# Graph fragment:
# %index : [num_users=1] = call_function[target=torch.ops.aten.index.Tensor](args = (%arg0_1, [%iota_default, %convert_element_type_1]), kwargs = {})
triton_poi_fused_index_1 = async_compile.triton('triton_poi_fused_index_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: '*i32', 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_index_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_index_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
x3 = (xindex // 16)
x0 = xindex % 16
x1 = (xindex // 16) % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp1 = tmp0.to(tl.int64)
tmp2 = tl.full([XBLOCK], 4, tl.int32)
tmp3 = tmp1 + tmp2
tmp4 = tmp1 < 0
tmp5 = tl.where(tmp4, tmp3, tmp1)
tl.device_assert(((0 <= tmp5) & (tmp5 < 4)) | ~(xmask), "index out of bounds: 0 <= tmp5 < 4")
tmp7 = tl.load(in_ptr1 + (x0 + (16*tmp5) + (64*x1)), xmask)
tl.store(out_ptr0 + (x4), 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)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.int32)
# Topologically Sorted Source Nodes: [abs_1, amax, used, length, length_1, sequence_length, setitem], Original ATen: [aten.abs, aten.amax, aten.sign, aten.sum, aten._to_copy, aten.sub, aten.lift_fresh, aten.index_put]
stream0 = get_raw_stream(0)
triton_poi_fused__to_copy_abs_amax_index_put_lift_fresh_sign_sub_sum_0.run(arg0_1, buf1, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gathered], Original ATen: [aten.index]
triton_poi_fused_index_1.run(buf1, arg0_1, buf2, 256, grid=grid(256), stream=stream0)
del arg0_1
del buf1
return (buf2, )
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
def sequence_length_3D(sequence: 'torch.Tensor') ->torch.Tensor:
used = torch.sign(torch.amax(torch.abs(sequence), dim=2))
length = torch.sum(used, 1)
length = length.int()
return length
class ReduceLast(torch.nn.Module):
def forward(self, inputs, mask=None):
batch_size = inputs.shape[0]
sequence_length = sequence_length_3D(inputs) - 1
sequence_length[sequence_length < 0] = 0
gathered = inputs[torch.arange(batch_size), sequence_length.type(
torch.int64)]
return gathered
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._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__to_copy_abs_amax_index_put_lift_fresh_sign_sub_sum_0(
in_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 + 64 * x1), xmask)
tmp2 = tl.load(in_ptr0 + (4 + x0 + 64 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (8 + x0 + 64 * x1), xmask)
tmp8 = tl.load(in_ptr0 + (12 + x0 + 64 * x1), xmask)
tmp18 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp20 = tl.load(in_ptr0 + (20 + x0 + 64 * x1), xmask)
tmp23 = tl.load(in_ptr0 + (24 + x0 + 64 * x1), xmask)
tmp26 = tl.load(in_ptr0 + (28 + x0 + 64 * x1), xmask)
tmp36 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp38 = tl.load(in_ptr0 + (36 + x0 + 64 * x1), xmask)
tmp41 = tl.load(in_ptr0 + (40 + x0 + 64 * x1), xmask)
tmp44 = tl.load(in_ptr0 + (44 + x0 + 64 * x1), xmask)
tmp54 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp56 = tl.load(in_ptr0 + (52 + x0 + 64 * x1), xmask)
tmp59 = tl.load(in_ptr0 + (56 + x0 + 64 * x1), xmask)
tmp62 = tl.load(in_ptr0 + (60 + x0 + 64 * x1), xmask)
tmp1 = tl_math.abs(tmp0)
tmp3 = tl_math.abs(tmp2)
tmp4 = triton_helpers.maximum(tmp1, tmp3)
tmp6 = tl_math.abs(tmp5)
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tl_math.abs(tmp8)
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp11 = tl.full([1], 0, tl.int32)
tmp12 = tmp11 < tmp10
tmp13 = tmp12.to(tl.int8)
tmp14 = tmp10 < tmp11
tmp15 = tmp14.to(tl.int8)
tmp16 = tmp13 - tmp15
tmp17 = tmp16.to(tmp10.dtype)
tmp19 = tl_math.abs(tmp18)
tmp21 = tl_math.abs(tmp20)
tmp22 = triton_helpers.maximum(tmp19, tmp21)
tmp24 = tl_math.abs(tmp23)
tmp25 = triton_helpers.maximum(tmp22, tmp24)
tmp27 = tl_math.abs(tmp26)
tmp28 = triton_helpers.maximum(tmp25, tmp27)
tmp29 = tmp11 < tmp28
tmp30 = tmp29.to(tl.int8)
tmp31 = tmp28 < tmp11
tmp32 = tmp31.to(tl.int8)
tmp33 = tmp30 - tmp32
tmp34 = tmp33.to(tmp28.dtype)
tmp35 = tmp17 + tmp34
tmp37 = tl_math.abs(tmp36)
tmp39 = tl_math.abs(tmp38)
tmp40 = triton_helpers.maximum(tmp37, tmp39)
tmp42 = tl_math.abs(tmp41)
tmp43 = triton_helpers.maximum(tmp40, tmp42)
tmp45 = tl_math.abs(tmp44)
tmp46 = triton_helpers.maximum(tmp43, tmp45)
tmp47 = tmp11 < tmp46
tmp48 = tmp47.to(tl.int8)
tmp49 = tmp46 < tmp11
tmp50 = tmp49.to(tl.int8)
tmp51 = tmp48 - tmp50
tmp52 = tmp51.to(tmp46.dtype)
tmp53 = tmp35 + tmp52
tmp55 = tl_math.abs(tmp54)
tmp57 = tl_math.abs(tmp56)
tmp58 = triton_helpers.maximum(tmp55, tmp57)
tmp60 = tl_math.abs(tmp59)
tmp61 = triton_helpers.maximum(tmp58, tmp60)
tmp63 = tl_math.abs(tmp62)
tmp64 = triton_helpers.maximum(tmp61, tmp63)
tmp65 = tmp11 < tmp64
tmp66 = tmp65.to(tl.int8)
tmp67 = tmp64 < tmp11
tmp68 = tmp67.to(tl.int8)
tmp69 = tmp66 - tmp68
tmp70 = tmp69.to(tmp64.dtype)
tmp71 = tmp53 + tmp70
tmp72 = tmp71.to(tl.int32)
tmp73 = tl.full([1], 1, tl.int32)
tmp74 = tmp72 - tmp73
tmp75 = tmp74 < tmp11
tmp76 = tl.where(tmp75, tmp11, tmp74)
tl.store(out_ptr1 + x2, tmp76, xmask)
@triton.jit
def triton_poi_fused_index_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
x3 = xindex // 16
x0 = xindex % 16
x1 = xindex // 16 % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp1 = tmp0.to(tl.int64)
tmp2 = tl.full([XBLOCK], 4, tl.int32)
tmp3 = tmp1 + tmp2
tmp4 = tmp1 < 0
tmp5 = tl.where(tmp4, tmp3, tmp1)
tl.device_assert((0 <= tmp5) & (tmp5 < 4) | ~xmask,
'index out of bounds: 0 <= tmp5 < 4')
tmp7 = tl.load(in_ptr1 + (x0 + 16 * tmp5 + 64 * x1), xmask)
tl.store(out_ptr0 + x4, 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)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.int32)
get_raw_stream(0)
triton_poi_fused__to_copy_abs_amax_index_put_lift_fresh_sign_sub_sum_0[
grid(16)](arg0_1, buf1, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_index_1[grid(256)](buf1, arg0_1, buf2, 256, XBLOCK
=256, num_warps=4, num_stages=1)
del arg0_1
del buf1
return buf2,
def sequence_length_3D(sequence: 'torch.Tensor') ->torch.Tensor:
used = torch.sign(torch.amax(torch.abs(sequence), dim=2))
length = torch.sum(used, 1)
length = length.int()
return length
class ReduceLastNew(torch.nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| carlogrisetti/ludwig | ReduceLast | false | 1,638 | [
"Apache-2.0"
] | 0 | 5c0887f14867e1577e0ddc3806c5cf7a781fb665 | https://github.com/carlogrisetti/ludwig/tree/5c0887f14867e1577e0ddc3806c5cf7a781fb665 | import torch
def sequence_length_3D(sequence: 'torch.Tensor') ->torch.Tensor:
used = torch.sign(torch.amax(torch.abs(sequence), dim=2))
length = torch.sum(used, 1)
length = length.int()
return length
class Model(torch.nn.Module):
def forward(self, inputs, mask=None):
batch_size = inputs.shape[0]
sequence_length = sequence_length_3D(inputs) - 1
sequence_length[sequence_length < 0] = 0
gathered = inputs[torch.arange(batch_size), sequence_length.type(
torch.int64)]
return gathered
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
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_6/inductor_cache/5l/c5lgdvchkbmivjnvcs2kuewrdqraps4bwq34ys346n3lh32qesvr.py
# Topologically Sorted Source Nodes: [out, mean, std, sub, out_1, relu1], Original ATen: [aten.convolution, aten.mean, aten.std, aten.sub, aten.div, aten.relu]
# Source node to ATen node mapping:
# mean => mean
# out => convolution
# out_1 => div
# relu1 => relu
# std => sqrt, var
# sub => sub
# Graph fragment:
# %convolution : [num_users=3] = 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 = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view, [2]), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%view, [2]), kwargs = {correction: 1.0})
# %sqrt : [num_users=2] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%convolution, %view_1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %view_2), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%div,), kwargs = {})
triton_per_fused_convolution_div_mean_relu_std_sub_0 = async_compile.triton('triton_per_fused_convolution_div_mean_relu_std_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.persistent_reduction(
size_hints=[16, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32', 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_per_fused_convolution_div_mean_relu_std_sub_0', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1', 'in_out_ptr2'], 'no_x_dim': False, '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_convolution_div_mean_relu_std_sub_0(in_out_ptr0, in_out_ptr1, in_out_ptr2, in_ptr0, out_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)
r2 = rindex
x3 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (r2 + (16*x3)), xmask, other=0.0)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.where(xmask, tmp3, 0)
tmp6 = tl.sum(tmp5, 1)[:, None]
tmp8 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp13 = tmp12.to(tl.float32)
tmp14 = tmp11 / tmp13
tmp15 = tmp3 - tmp14
tmp16 = tmp15 * tmp15
tmp17 = tl.broadcast_to(tmp16, [XBLOCK, RBLOCK])
tmp19 = tl.where(xmask, tmp17, 0)
tmp20 = tl.sum(tmp19, 1)[:, None]
tmp21 = 16.0
tmp22 = tmp6 / tmp21
tmp23 = 15.0
tmp24 = tmp20 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = tmp2 - tmp22
tmp27 = tmp26 / tmp25
tmp28 = tl.full([1, 1], 0, tl.int32)
tmp29 = triton_helpers.maximum(tmp28, tmp27)
tl.store(in_out_ptr0 + (r2 + (16*x3)), tmp2, xmask)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x3), tmp22, xmask)
tl.debug_barrier()
tl.store(in_out_ptr2 + (x3), tmp25, xmask)
tl.store(out_ptr0 + (r2 + (16*x3)), tmp29, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/7d/c7dol2m4pzrwhlm3ie26bkh5wyzfwzx2tgk5pk3xbho7pf72e76y.py
# Topologically Sorted Source Nodes: [out_2, mean_2, std_2, sub_1, out_3, out_4, out_5], Original ATen: [aten.convolution, aten.mean, aten.std, aten.sub, aten.div, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# mean_2 => mean_1
# out_2 => convolution_1
# out_3 => div_1
# out_4 => add
# out_5 => relu_1
# std_2 => sqrt_1, var_1
# sub_1 => sub_1
# Graph fragment:
# %convolution_1 : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view_3, [2]), kwargs = {})
# %var_1 : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%view_3, [2]), kwargs = {correction: 1.0})
# %sqrt_1 : [num_users=2] = call_function[target=torch.ops.aten.sqrt.default](args = (%var_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%convolution_1, %view_4), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_1, %view_5), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_1, %primals_3), kwargs = {})
# %relu_1 : [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_1, 0), kwargs = {})
triton_per_fused_add_convolution_div_mean_relu_std_sub_threshold_backward_1 = async_compile.triton('triton_per_fused_add_convolution_div_mean_relu_std_sub_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.persistent_reduction(
size_hints=[16, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*i1', 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_convolution_div_mean_relu_std_sub_threshold_backward_1', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1', 'in_out_ptr2'], 'no_x_dim': False, '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_add_convolution_div_mean_relu_std_sub_threshold_backward_1(in_out_ptr0, in_out_ptr1, in_out_ptr2, in_ptr0, in_ptr1, out_ptr0, out_ptr1, 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)
r2 = rindex
x3 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (r2 + (16*x3)), xmask, other=0.0)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (r2 + (16*x3)), xmask, other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.where(xmask, tmp3, 0)
tmp6 = tl.sum(tmp5, 1)[:, None]
tmp8 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp13 = tmp12.to(tl.float32)
tmp14 = tmp11 / tmp13
tmp15 = tmp3 - tmp14
tmp16 = tmp15 * tmp15
tmp17 = tl.broadcast_to(tmp16, [XBLOCK, RBLOCK])
tmp19 = tl.where(xmask, tmp17, 0)
tmp20 = tl.sum(tmp19, 1)[:, None]
tmp21 = 16.0
tmp22 = tmp6 / tmp21
tmp23 = 15.0
tmp24 = tmp20 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = tmp2 - tmp22
tmp27 = tmp26 / tmp25
tmp29 = tmp27 + tmp28
tmp30 = tl.full([1, 1], 0, tl.int32)
tmp31 = triton_helpers.maximum(tmp30, tmp29)
tmp32 = 0.0
tmp33 = tmp31 <= tmp32
tl.store(in_out_ptr0 + (r2 + (16*x3)), tmp2, xmask)
tl.debug_barrier()
tl.store(in_out_ptr1 + (x3), tmp22, xmask)
tl.debug_barrier()
tl.store(in_out_ptr2 + (x3), tmp25, xmask)
tl.store(out_ptr0 + (r2 + (16*x3)), tmp31, xmask)
tl.store(out_ptr1 + (r2 + (16*x3)), tmp33, 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, 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))
assert_size_stride(primals_4, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [out], 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, 4, 4, 4), (64, 16, 4, 1))
buf1 = buf0; del buf0 # reuse
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf3 = buf2; del buf2 # reuse
buf7 = buf5; del buf5 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out, mean, std, sub, out_1, relu1], Original ATen: [aten.convolution, aten.mean, aten.std, aten.sub, aten.div, aten.relu]
stream0 = get_raw_stream(0)
triton_per_fused_convolution_div_mean_relu_std_sub_0.run(buf1, buf3, buf7, primals_2, buf8, 16, 16, grid=grid(16), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.convolution]
buf9 = extern_kernels.convolution(buf8, primals_4, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf9, (4, 4, 4, 4), (64, 16, 4, 1))
buf10 = buf9; del buf9 # reuse
buf11 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf12 = buf11; del buf11 # reuse
buf16 = buf14; del buf14 # reuse
buf17 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf18 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [out_2, mean_2, std_2, sub_1, out_3, out_4, out_5], Original ATen: [aten.convolution, aten.mean, aten.std, aten.sub, aten.div, aten.add, aten.relu, aten.threshold_backward]
triton_per_fused_add_convolution_div_mean_relu_std_sub_threshold_backward_1.run(buf10, buf12, buf16, primals_5, primals_3, buf17, buf18, 16, 16, grid=grid(16), stream=stream0)
del primals_5
return (buf8, buf17, primals_1, primals_3, primals_4, buf1, reinterpret_tensor(buf3, (4, 4, 1, 1), (4, 1, 1, 1), 0), buf7, buf8, buf10, reinterpret_tensor(buf12, (4, 4, 1, 1), (4, 1, 1, 1), 0), buf16, buf18, )
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)
primals_4 = rand_strided((4, 4, 3, 3), (36, 9, 3, 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.functional as F
class RemoveChannelMeanStd(torch.nn.Module):
def forward(self, x):
x2 = x.view(x.size(0), x.size(1), -1)
mean = x2.mean(dim=2).view(x.size(0), x.size(1), 1, 1)
std = x2.std(dim=2).view(x.size(0), x.size(1), 1, 1)
return (x - mean) / std
class Block(nn.Module):
def __init__(self, in_planes, planes, stride=1, groups=False):
super(Block, self).__init__()
self.bn1 = RemoveChannelMeanStd()
if not groups:
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride
=stride, padding=1, bias=True)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1,
padding=1, bias=True)
else:
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride
=stride, padding=1, bias=True, groups=min(in_planes, planes))
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1,
padding=1, bias=True, groups=planes)
self.bn2 = RemoveChannelMeanStd()
if stride != 1 or in_planes != planes:
self.shortcut = nn.Sequential(nn.Conv2d(in_planes, planes,
kernel_size=1, stride=stride, bias=True),
RemoveChannelMeanStd())
def forward(self, x):
out = self.conv1(x)
out = self.bn1(out)
relu1 = F.relu(out)
out = self.conv2(relu1)
out = self.bn2(out)
shortcut = self.shortcut(x) if hasattr(self, 'shortcut') else x
out += shortcut
out = F.relu(out)
return relu1, out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_planes': 4, 'planes': 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
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_convolution_div_mean_relu_std_sub_0(in_out_ptr0,
in_out_ptr1, in_out_ptr2, in_ptr0, out_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)
r2 = rindex
x3 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (r2 + 16 * x3), xmask, other=0.0)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.where(xmask, tmp3, 0)
tmp6 = tl.sum(tmp5, 1)[:, None]
tmp8 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp13 = tmp12.to(tl.float32)
tmp14 = tmp11 / tmp13
tmp15 = tmp3 - tmp14
tmp16 = tmp15 * tmp15
tmp17 = tl.broadcast_to(tmp16, [XBLOCK, RBLOCK])
tmp19 = tl.where(xmask, tmp17, 0)
tmp20 = tl.sum(tmp19, 1)[:, None]
tmp21 = 16.0
tmp22 = tmp6 / tmp21
tmp23 = 15.0
tmp24 = tmp20 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = tmp2 - tmp22
tmp27 = tmp26 / tmp25
tmp28 = tl.full([1, 1], 0, tl.int32)
tmp29 = triton_helpers.maximum(tmp28, tmp27)
tl.store(in_out_ptr0 + (r2 + 16 * x3), tmp2, xmask)
tl.debug_barrier()
tl.store(in_out_ptr1 + x3, tmp22, xmask)
tl.debug_barrier()
tl.store(in_out_ptr2 + x3, tmp25, xmask)
tl.store(out_ptr0 + (r2 + 16 * x3), tmp29, xmask)
@triton.jit
def triton_per_fused_add_convolution_div_mean_relu_std_sub_threshold_backward_1(
in_out_ptr0, in_out_ptr1, in_out_ptr2, in_ptr0, in_ptr1, out_ptr0,
out_ptr1, 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)
r2 = rindex
x3 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (r2 + 16 * x3), xmask, other=0.0)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (r2 + 16 * x3), xmask, other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.where(xmask, tmp3, 0)
tmp6 = tl.sum(tmp5, 1)[:, None]
tmp8 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp13 = tmp12.to(tl.float32)
tmp14 = tmp11 / tmp13
tmp15 = tmp3 - tmp14
tmp16 = tmp15 * tmp15
tmp17 = tl.broadcast_to(tmp16, [XBLOCK, RBLOCK])
tmp19 = tl.where(xmask, tmp17, 0)
tmp20 = tl.sum(tmp19, 1)[:, None]
tmp21 = 16.0
tmp22 = tmp6 / tmp21
tmp23 = 15.0
tmp24 = tmp20 / tmp23
tmp25 = libdevice.sqrt(tmp24)
tmp26 = tmp2 - tmp22
tmp27 = tmp26 / tmp25
tmp29 = tmp27 + tmp28
tmp30 = tl.full([1, 1], 0, tl.int32)
tmp31 = triton_helpers.maximum(tmp30, tmp29)
tmp32 = 0.0
tmp33 = tmp31 <= tmp32
tl.store(in_out_ptr0 + (r2 + 16 * x3), tmp2, xmask)
tl.debug_barrier()
tl.store(in_out_ptr1 + x3, tmp22, xmask)
tl.debug_barrier()
tl.store(in_out_ptr2 + x3, tmp25, xmask)
tl.store(out_ptr0 + (r2 + 16 * x3), tmp31, xmask)
tl.store(out_ptr1 + (r2 + 16 * x3), tmp33, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = 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))
assert_size_stride(primals_4, (4, 4, 3, 3), (36, 9, 3, 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=(1, 1), 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 = buf0
del buf0
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf3 = buf2
del buf2
buf7 = buf5
del buf5
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_convolution_div_mean_relu_std_sub_0[grid(16)](buf1,
buf3, buf7, primals_2, buf8, 16, 16, XBLOCK=8, num_warps=2,
num_stages=1)
del primals_2
buf9 = extern_kernels.convolution(buf8, primals_4, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf9, (4, 4, 4, 4), (64, 16, 4, 1))
buf10 = buf9
del buf9
buf11 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf12 = buf11
del buf11
buf16 = buf14
del buf14
buf17 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf18 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_per_fused_add_convolution_div_mean_relu_std_sub_threshold_backward_1[
grid(16)](buf10, buf12, buf16, primals_5, primals_3, buf17,
buf18, 16, 16, XBLOCK=1, num_warps=2, num_stages=1)
del primals_5
return (buf8, buf17, primals_1, primals_3, primals_4, buf1,
reinterpret_tensor(buf3, (4, 4, 1, 1), (4, 1, 1, 1), 0), buf7, buf8,
buf10, reinterpret_tensor(buf12, (4, 4, 1, 1), (4, 1, 1, 1), 0),
buf16, buf18)
class RemoveChannelMeanStd(torch.nn.Module):
def forward(self, x):
x2 = x.view(x.size(0), x.size(1), -1)
mean = x2.mean(dim=2).view(x.size(0), x.size(1), 1, 1)
std = x2.std(dim=2).view(x.size(0), x.size(1), 1, 1)
return (x - mean) / std
class BlockNew(nn.Module):
def __init__(self, in_planes, planes, stride=1, groups=False):
super(BlockNew, self).__init__()
self.bn1 = RemoveChannelMeanStd()
if not groups:
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride
=stride, padding=1, bias=True)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1,
padding=1, bias=True)
else:
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride
=stride, padding=1, bias=True, groups=min(in_planes, planes))
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1,
padding=1, bias=True, groups=planes)
self.bn2 = RemoveChannelMeanStd()
if stride != 1 or in_planes != planes:
self.shortcut = nn.Sequential(nn.Conv2d(in_planes, planes,
kernel_size=1, stride=stride, bias=True),
RemoveChannelMeanStd())
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], output[1]
| cadurosar/graph_kd_dense_cifar100 | Block | false | 1,639 | [
"MIT"
] | 0 | 84054ab4f8f61c9db3460993661ba7bf1d951b36 | https://github.com/cadurosar/graph_kd_dense_cifar100/tree/84054ab4f8f61c9db3460993661ba7bf1d951b36 | import torch
import torch.nn as nn
import torch.nn.functional as F
class RemoveChannelMeanStd(torch.nn.Module):
def forward(self, x):
x2 = x.view(x.size(0), x.size(1), -1)
mean = x2.mean(dim=2).view(x.size(0), x.size(1), 1, 1)
std = x2.std(dim=2).view(x.size(0), x.size(1), 1, 1)
return (x - mean) / std
class Model(nn.Module):
def __init__(self, in_planes, planes, stride=1, groups=False):
super().__init__()
self.bn1 = RemoveChannelMeanStd()
if not groups:
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride
=stride, padding=1, bias=True)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1,
padding=1, bias=True)
else:
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride
=stride, padding=1, bias=True, groups=min(in_planes, planes))
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1,
padding=1, bias=True, groups=planes)
self.bn2 = RemoveChannelMeanStd()
if stride != 1 or in_planes != planes:
self.shortcut = nn.Sequential(nn.Conv2d(in_planes, planes,
kernel_size=1, stride=stride, bias=True),
RemoveChannelMeanStd())
def forward(self, x):
out = self.conv1(x)
out = self.bn1(out)
relu1 = F.relu(out)
out = self.conv2(relu1)
out = self.bn2(out)
shortcut = self.shortcut(x) if hasattr(self, 'shortcut') else x
out += shortcut
out = F.relu(out)
return relu1, out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
NPairsLoss | # 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_6/inductor_cache/5u/c5uuve7l53u3gk5jcyv7dpxtt73wwahvpgg4puf25vz73bgkynpq.py
# Topologically Sorted Source Nodes: [diag, diagonal_mean, neg, logsumexp, mean_log_row_sum_exp, add], Original ATen: [aten.diagonal_copy, aten.mean, aten.neg, aten.logsumexp, aten.add]
# Source node to ATen node mapping:
# add => add_1
# diag => diagonal_copy
# diagonal_mean => mean
# logsumexp => abs_1, add, amax, eq, exp, full_default, log, sub, sum_1, where
# mean_log_row_sum_exp => mean_1
# neg => neg
# Graph fragment:
# %diagonal_copy : [num_users=1] = call_function[target=torch.ops.aten.diagonal_copy.default](args = (%mm,), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%diagonal_copy,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%mean,), kwargs = {})
# %amax : [num_users=2] = call_function[target=torch.ops.aten.amax.default](args = (%mm, [1], True), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%amax,), kwargs = {})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%abs_1, inf), 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})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %amax), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mm, %where), 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]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%log, %squeeze), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%add,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg, %mean_1), kwargs = {})
triton_per_fused_add_diagonal_copy_logsumexp_mean_neg_0 = async_compile.triton('triton_per_fused_add_diagonal_copy_logsumexp_mean_neg_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: '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), equal_to_1=(2,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_diagonal_copy_logsumexp_mean_neg_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, '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_diagonal_copy_logsumexp_mean_neg_0(in_out_ptr0, in_ptr0, 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 + (5*r0), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (4*r0), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.sum(tmp1, 1)[:, None]
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tmp8 = triton_helpers.maximum(tmp6, tmp7)
tmp10 = triton_helpers.maximum(tmp8, tmp9)
tmp11 = tl_math.abs(tmp10)
tmp12 = float("inf")
tmp13 = tmp11 == tmp12
tmp14 = 0.0
tmp15 = tl.where(tmp13, tmp14, tmp10)
tmp16 = tmp4 - tmp15
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp5 - tmp15
tmp19 = tl_math.exp(tmp18)
tmp20 = tmp17 + tmp19
tmp21 = tmp7 - tmp15
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp20 + tmp22
tmp24 = tmp9 - tmp15
tmp25 = tl_math.exp(tmp24)
tmp26 = tmp23 + tmp25
tmp27 = tl_math.log(tmp26)
tmp28 = tmp27 + tmp15
tmp29 = tl.broadcast_to(tmp28, [XBLOCK, RBLOCK])
tmp31 = tl.sum(tmp29, 1)[:, None]
tmp32 = 4.0
tmp33 = tmp3 / tmp32
tmp34 = -tmp33
tmp35 = tmp31 / tmp32
tmp36 = tmp34 + tmp35
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp36, 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, 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, 1), torch.float32)
# Topologically Sorted Source Nodes: [scores], Original ATen: [aten.mm]
extern_kernels.mm(arg1_1, reinterpret_tensor(arg0_1, (4, 4), (1, 4), 0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf3 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [diag, diagonal_mean, neg, logsumexp, mean_log_row_sum_exp, add], Original ATen: [aten.diagonal_copy, aten.mean, aten.neg, aten.logsumexp, aten.add]
stream0 = get_raw_stream(0)
triton_per_fused_add_diagonal_copy_logsumexp_mean_neg_0.run(buf3, buf0, 1, 4, grid=grid(1), stream=stream0)
del buf0
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, 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
from torch import nn
class NPairsLoss(nn.Module):
def __init__(self, name):
super(NPairsLoss, self).__init__()
self.name = name
def forward(self, r1, r2):
"""
Computes the N-Pairs Loss between the r1 and r2 representations.
:param r1: Tensor of shape (batch_size, representation_size)
:param r2: Tensor of shape (batch_size, representation_size)
:return: he scalar loss
"""
scores = torch.matmul(r1, r2.t())
diagonal_mean = torch.mean(torch.diag(scores))
mean_log_row_sum_exp = torch.mean(torch.logsumexp(scores, dim=1))
return -diagonal_mean + mean_log_row_sum_exp
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'name': 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_per_fused_add_diagonal_copy_logsumexp_mean_neg_0(in_out_ptr0,
in_ptr0, 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 + 5 * r0, None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + 4 * r0, None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.sum(tmp1, 1)[:, None]
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tmp8 = triton_helpers.maximum(tmp6, tmp7)
tmp10 = triton_helpers.maximum(tmp8, tmp9)
tmp11 = tl_math.abs(tmp10)
tmp12 = float('inf')
tmp13 = tmp11 == tmp12
tmp14 = 0.0
tmp15 = tl.where(tmp13, tmp14, tmp10)
tmp16 = tmp4 - tmp15
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp5 - tmp15
tmp19 = tl_math.exp(tmp18)
tmp20 = tmp17 + tmp19
tmp21 = tmp7 - tmp15
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp20 + tmp22
tmp24 = tmp9 - tmp15
tmp25 = tl_math.exp(tmp24)
tmp26 = tmp23 + tmp25
tmp27 = tl_math.log(tmp26)
tmp28 = tmp27 + tmp15
tmp29 = tl.broadcast_to(tmp28, [XBLOCK, RBLOCK])
tmp31 = tl.sum(tmp29, 1)[:, None]
tmp32 = 4.0
tmp33 = tmp3 / tmp32
tmp34 = -tmp33
tmp35 = tmp31 / tmp32
tmp36 = tmp34 + tmp35
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp36, None)
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, 1), torch.float32)
extern_kernels.mm(arg1_1, reinterpret_tensor(arg0_1, (4, 4), (1, 4),
0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf3 = buf1
del buf1
get_raw_stream(0)
triton_per_fused_add_diagonal_copy_logsumexp_mean_neg_0[grid(1)](buf3,
buf0, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del buf0
return buf3,
class NPairsLossNew(nn.Module):
def __init__(self, name):
super(NPairsLossNew, self).__init__()
self.name = name
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| celsofranssa/PAWS | NPairsLoss | false | 1,640 | [
"MIT"
] | 0 | 3171c8a22990059f5d4c0e7e81cc0299a716efb2 | https://github.com/celsofranssa/PAWS/tree/3171c8a22990059f5d4c0e7e81cc0299a716efb2 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, name):
super().__init__()
self.name = name
def forward(self, r1, r2):
"""
Computes the N-Pairs Loss between the r1 and r2 representations.
:param r1: Tensor of shape (batch_size, representation_size)
:param r2: Tensor of shape (batch_size, representation_size)
:return: he scalar loss
"""
scores = torch.matmul(r1, r2.t())
diagonal_mean = torch.mean(torch.diag(scores))
mean_log_row_sum_exp = torch.mean(torch.logsumexp(scores, dim=1))
return -diagonal_mean + mean_log_row_sum_exp
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [4]
|
CNNCifar | # 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_6/inductor_cache/zv/czvfpj3ah2lefbwpcuw4esv23bxs5a3ab63ply3ntgbsdktepd5v.py
# Topologically Sorted Source Nodes: [conv2d, relu], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d => convolution
# relu => 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=[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_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 = 18816
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 784) % 6
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_6/inductor_cache/v7/cv7qi7gg3bpfwb3hj7zgy5jlgh7x7wdgqsfsodkjsoverxdjlf6z.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x => 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=[8192],
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 = 4704
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 14
x3 = (xindex // 14)
x2 = (xindex // 1176)
x4 = xindex % 1176
tmp0 = tl.load(in_ptr0 + ((2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (28 + (2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (29 + (2*x0) + (56*x3)), xmask, 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 + (x4 + (1184*x2)), tmp6, xmask)
tl.store(out_ptr1 + (x4 + (1280*x2)), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/xe/cxelxvpw3asckozc53rh36773aohp5hqpbp2nos5ymcdqhxvo4bl.py
# Topologically Sorted Source Nodes: [conv2d_1, relu_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d_1 => convolution_1
# relu_1 => 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], [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_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=[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_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 = 6400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 100) % 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_6/inductor_cache/tn/ctnw4tbgfy47ppke77vu7rtiz7dl5o3ahickx4p64n7c5rmrrix6.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x_1 => _low_memory_max_pool2d_with_offsets_1, getitem_3
# Graph fragment:
# %_low_memory_max_pool2d_with_offsets_1 : [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_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=[2048],
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_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 = 1600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = (xindex // 5)
x2 = xindex
tmp0 = tl.load(in_ptr0 + ((2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (10 + (2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (11 + (2*x0) + (20*x1)), 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 + (x2), tmp15, xmask)
tl.store(out_ptr1 + (x2), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/jn/cjnqv3sgcv5x2iz7ij5zdad6ofabcnonrlksgsxu2ob7n274gz6b.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_3 => relu_2
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_7), kwargs = {})
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_1,), kwargs = {})
triton_poi_fused_relu_4 = async_compile.triton('triton_poi_fused_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=[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_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_relu_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 480
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 120
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_6/inductor_cache/6m/c6m6u2ctjb4r4ra3sizrwezzkzegfp2ombflmfg3dwjfci2pen7h.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_5 = async_compile.triton('triton_poi_fused_relu_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=[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_relu_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_relu_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 336
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 84
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_6/inductor_cache/gs/cgsvpzwol2pyh6klnjmwgbogbcrewrnkr3diy2tntyhkzjwywqsz.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 = (%addmm_2, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%addmm_2, %amax), kwargs = {})
triton_poi_fused__log_softmax_6 = async_compile.triton('triton_poi_fused__log_softmax_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=80, major=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_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__log_softmax_6(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 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/o6/co6cntxjesemjsuiiy4lyctyi4xtwek53cbirwghtvdb5hcnjiws.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_7 = async_compile.triton('triton_poi_fused__log_softmax_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: '*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_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__log_softmax_7(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 = args
args.clear()
assert_size_stride(primals_1, (6, 3, 5, 5), (75, 25, 5, 1))
assert_size_stride(primals_2, (6, ), (1, ))
assert_size_stride(primals_3, (4, 3, 32, 32), (3072, 1024, 32, 1))
assert_size_stride(primals_4, (16, 6, 5, 5), (150, 25, 5, 1))
assert_size_stride(primals_5, (16, ), (1, ))
assert_size_stride(primals_6, (120, 400), (400, 1))
assert_size_stride(primals_7, (120, ), (1, ))
assert_size_stride(primals_8, (84, 120), (120, 1))
assert_size_stride(primals_9, (84, ), (1, ))
assert_size_stride(primals_10, (4, 84), (84, 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(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, 6, 28, 28), (4704, 784, 28, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv2d, relu], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 18816, grid=grid(18816), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((4, 6, 14, 14), (1184, 196, 14, 1), torch.float32)
buf3 = empty_strided_cuda((4, 6, 14, 14), (1280, 196, 14, 1), torch.int8)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_1.run(buf1, buf2, buf3, 4704, grid=grid(4704), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf4 = 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(buf4, (4, 16, 10, 10), (1600, 100, 10, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [conv2d_1, relu_1], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf5, primals_5, 6400, grid=grid(6400), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.int8)
buf7 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_3.run(buf5, buf6, buf7, 1600, grid=grid(1600), stream=stream0)
buf8 = empty_strided_cuda((4, 120), (120, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf7, (4, 400), (400, 1), 0), reinterpret_tensor(primals_6, (400, 120), (1, 400), 0), out=buf8)
buf9 = buf8; del buf8 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu]
triton_poi_fused_relu_4.run(buf9, primals_7, 480, grid=grid(480), stream=stream0)
del primals_7
buf10 = empty_strided_cuda((4, 84), (84, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf9, reinterpret_tensor(primals_8, (120, 84), (1, 120), 0), out=buf10)
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
triton_poi_fused_relu_5.run(buf11, primals_9, 336, grid=grid(336), stream=stream0)
del primals_9
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, buf11, reinterpret_tensor(primals_10, (84, 4), (1, 84), 0), alpha=1, beta=1, out=buf12)
del primals_11
buf13 = 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_6.run(buf12, buf13, 16, grid=grid(16), stream=stream0)
buf14 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_7.run(buf13, buf14, 16, grid=grid(16), stream=stream0)
del buf13
return (buf14, primals_1, primals_3, primals_4, buf1, buf2, buf3, buf5, buf6, reinterpret_tensor(buf7, (4, 400), (400, 1), 0), buf9, buf11, buf14, primals_10, primals_8, 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((6, 3, 5, 5), (75, 25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((6, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 32, 32), (3072, 1024, 32, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, 6, 5, 5), (150, 25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((120, 400), (400, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((120, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((84, 120), (120, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((84, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 84), (84, 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 torch
from torch import nn
import torch.nn.functional as F
class CNNCifar(nn.Module):
def __init__(self, args):
super(CNNCifar, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, args.num_classes)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = x.view(-1, 16 * 5 * 5)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return F.log_softmax(x, dim=1)
def get_inputs():
return [torch.rand([4, 3, 32, 32])]
def get_init_inputs():
return [[], {'args': _mock_config(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 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_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 18816
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 784 % 6
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_max_pool2d_with_indices_1(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 4704
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 14
x3 = xindex // 14
x2 = xindex // 1176
x4 = xindex % 1176
tmp0 = tl.load(in_ptr0 + (2 * x0 + 56 * x3), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 56 * x3), xmask, eviction_policy
='evict_last')
tmp3 = tl.load(in_ptr0 + (28 + 2 * x0 + 56 * x3), xmask,
eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (29 + 2 * x0 + 56 * x3), xmask,
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 + (x4 + 1184 * x2), tmp6, xmask)
tl.store(out_ptr1 + (x4 + 1280 * x2), tmp16, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_2(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 6400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 100 % 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_max_pool2d_with_indices_3(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 1600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = xindex // 5
x2 = xindex
tmp0 = tl.load(in_ptr0 + (2 * x0 + 20 * x1), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 20 * x1), xmask, eviction_policy
='evict_last')
tmp7 = tl.load(in_ptr0 + (10 + 2 * x0 + 20 * x1), xmask,
eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (11 + 2 * x0 + 20 * x1), 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 + x2, tmp15, xmask)
tl.store(out_ptr1 + x2, tmp16, xmask)
@triton.jit
def triton_poi_fused_relu_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 480
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 120
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_relu_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 336
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 84
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__log_softmax_6(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 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused__log_softmax_7(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) = args
args.clear()
assert_size_stride(primals_1, (6, 3, 5, 5), (75, 25, 5, 1))
assert_size_stride(primals_2, (6,), (1,))
assert_size_stride(primals_3, (4, 3, 32, 32), (3072, 1024, 32, 1))
assert_size_stride(primals_4, (16, 6, 5, 5), (150, 25, 5, 1))
assert_size_stride(primals_5, (16,), (1,))
assert_size_stride(primals_6, (120, 400), (400, 1))
assert_size_stride(primals_7, (120,), (1,))
assert_size_stride(primals_8, (84, 120), (120, 1))
assert_size_stride(primals_9, (84,), (1,))
assert_size_stride(primals_10, (4, 84), (84, 1))
assert_size_stride(primals_11, (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, 6, 28, 28), (4704, 784, 28, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(18816)](buf1, primals_2,
18816, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((4, 6, 14, 14), (1184, 196, 14, 1), torch
.float32)
buf3 = empty_strided_cuda((4, 6, 14, 14), (1280, 196, 14, 1), torch
.int8)
triton_poi_fused_max_pool2d_with_indices_1[grid(4704)](buf1, buf2,
buf3, 4704, XBLOCK=128, num_warps=4, num_stages=1)
buf4 = 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(buf4, (4, 16, 10, 10), (1600, 100, 10, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_relu_2[grid(6400)](buf5, primals_5,
6400, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf6 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.int8)
buf7 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.float32
)
triton_poi_fused_max_pool2d_with_indices_3[grid(1600)](buf5, buf6,
buf7, 1600, XBLOCK=256, num_warps=4, num_stages=1)
buf8 = empty_strided_cuda((4, 120), (120, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf7, (4, 400), (400, 1), 0),
reinterpret_tensor(primals_6, (400, 120), (1, 400), 0), out=buf8)
buf9 = buf8
del buf8
triton_poi_fused_relu_4[grid(480)](buf9, primals_7, 480, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_7
buf10 = empty_strided_cuda((4, 84), (84, 1), torch.float32)
extern_kernels.mm(buf9, reinterpret_tensor(primals_8, (120, 84), (1,
120), 0), out=buf10)
buf11 = buf10
del buf10
triton_poi_fused_relu_5[grid(336)](buf11, primals_9, 336, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_9
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_11, buf11, reinterpret_tensor(
primals_10, (84, 4), (1, 84), 0), alpha=1, beta=1, out=buf12)
del primals_11
buf13 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__log_softmax_6[grid(16)](buf12, buf13, 16, XBLOCK=
16, num_warps=1, num_stages=1)
buf14 = buf12
del buf12
triton_poi_fused__log_softmax_7[grid(16)](buf13, buf14, 16, XBLOCK=
16, num_warps=1, num_stages=1)
del buf13
return (buf14, primals_1, primals_3, primals_4, buf1, buf2, buf3, buf5,
buf6, reinterpret_tensor(buf7, (4, 400), (400, 1), 0), buf9, buf11,
buf14, primals_10, primals_8, primals_6)
class CNNCifarNew(nn.Module):
def __init__(self, args):
super(CNNCifarNew, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, args.num_classes)
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.fc1.weight
primals_7 = self.fc1.bias
primals_8 = self.fc2.weight
primals_9 = self.fc2.bias
primals_10 = self.fc3.weight
primals_11 = self.fc3.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]
| Joey61Liuyi/Federated-Learning-PyTorch | CNNCifar | false | 1,641 | [
"MIT"
] | 0 | e95f096b18c5a1bf30fc0485acd5a15c84327f2e | https://github.com/Joey61Liuyi/Federated-Learning-PyTorch/tree/e95f096b18c5a1bf30fc0485acd5a15c84327f2e | from _paritybench_helpers import _mock_config
import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, args):
super().__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, args.num_classes)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = x.view(-1, 16 * 5 * 5)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return F.log_softmax(x, dim=1)
def get_inputs():
return [torch.rand([4, 3, 32, 32])]
def get_init_inputs():
return []
|
RMSELoss | # 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_6/inductor_cache/d5/cd5s53yhudkvpmmaicopy6xu2boxzspih5czefns3qdpdfmxci7z.py
# Topologically Sorted Source Nodes: [mse_loss, add, sqrt], Original ATen: [aten.mse_loss, aten.add, aten.sqrt]
# Source node to ATen node mapping:
# add => add
# mse_loss => mean, pow_1, sub
# sqrt => sqrt
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %arg0_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%pow_1,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean, 1e-08), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add,), kwargs = {})
triton_per_fused_add_mse_loss_sqrt_0 = async_compile.triton('triton_per_fused_add_mse_loss_sqrt_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_mse_loss_sqrt_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_add_mse_loss_sqrt_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)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = triton_helpers.promote_to_tensor(tl.sum(tmp4, 0))
tmp7 = 256.0
tmp8 = tmp6 / tmp7
tmp9 = 1e-08
tmp10 = tmp8 + tmp9
tmp11 = libdevice.sqrt(tmp10)
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp11, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mse_loss, add, sqrt], Original ATen: [aten.mse_loss, aten.add, aten.sqrt]
stream0 = get_raw_stream(0)
triton_per_fused_add_mse_loss_sqrt_0.run(buf1, arg1_1, arg0_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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 RMSELoss(nn.Module):
def __init__(self, eps=1e-08):
super(RMSELoss, self).__init__()
self.mse = nn.MSELoss()
self.eps = eps
def forward(self, y_hat, y):
return torch.sqrt(self.mse(y_hat, y) + self.eps)
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 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_per_fused_add_mse_loss_sqrt_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)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp4 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = triton_helpers.promote_to_tensor(tl.sum(tmp4, 0))
tmp7 = 256.0
tmp8 = tmp6 / tmp7
tmp9 = 1e-08
tmp10 = tmp8 + tmp9
tmp11 = libdevice.sqrt(tmp10)
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp11, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_mse_loss_sqrt_0[grid(1)](buf1, arg1_1, arg0_1,
1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class RMSELossNew(nn.Module):
def __init__(self, eps=1e-08):
super(RMSELossNew, self).__init__()
self.mse = nn.MSELoss()
self.eps = eps
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| cerisara/weibull-knowledge-informed-ml | RMSELoss | false | 1,642 | [
"MIT"
] | 0 | 19017817f5324fb1fffd8322d2d3567a6271948c | https://github.com/cerisara/weibull-knowledge-informed-ml/tree/19017817f5324fb1fffd8322d2d3567a6271948c | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, eps=1e-08):
super().__init__()
self.mse = nn.MSELoss()
self.eps = eps
def forward(self, y_hat, y):
return torch.sqrt(self.mse(y_hat, y) + self.eps)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MAPELoss | # 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_6/inductor_cache/n2/cn23lclcjjov3evcgbnbohwdk7oavkwcoywfbrr3el4gtu3acmwc.py
# Topologically Sorted Source Nodes: [sub, abs_1, add, abs_2, truediv, mean, mul], Original ATen: [aten.sub, aten.abs, aten.add, aten.div, aten.mean, aten.mul]
# Source node to ATen node mapping:
# abs_1 => abs_1
# abs_2 => abs_2
# add => add
# mean => mean
# mul => mul
# sub => sub
# truediv => div
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%sub,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-08), kwargs = {})
# %abs_2 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%add,), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%abs_1, %abs_2), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%div,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mean, 100), kwargs = {})
triton_per_fused_abs_add_div_mean_mul_sub_0 = async_compile.triton('triton_per_fused_abs_add_div_mean_mul_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.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_abs_add_div_mean_mul_sub_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_abs_add_div_mean_mul_sub_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)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.abs(tmp2)
tmp4 = 1e-08
tmp5 = tmp0 + tmp4
tmp6 = tl_math.abs(tmp5)
tmp7 = tmp3 / tmp6
tmp8 = tl.broadcast_to(tmp7, [RBLOCK])
tmp10 = triton_helpers.promote_to_tensor(tl.sum(tmp8, 0))
tmp11 = 256.0
tmp12 = tmp10 / tmp11
tmp13 = 100.0
tmp14 = tmp12 * tmp13
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp14, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [sub, abs_1, add, abs_2, truediv, mean, mul], Original ATen: [aten.sub, aten.abs, aten.add, aten.div, aten.mean, aten.mul]
stream0 = get_raw_stream(0)
triton_per_fused_abs_add_div_mean_mul_sub_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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 MAPELoss(nn.Module):
def __init__(self, eps=1e-08):
super(MAPELoss, self).__init__()
self.eps = eps
def forward(self, y_hat, y):
return torch.mean(torch.abs(y - y_hat) / torch.abs(y + self.eps)) * 100
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_per_fused_abs_add_div_mean_mul_sub_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)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.abs(tmp2)
tmp4 = 1e-08
tmp5 = tmp0 + tmp4
tmp6 = tl_math.abs(tmp5)
tmp7 = tmp3 / tmp6
tmp8 = tl.broadcast_to(tmp7, [RBLOCK])
tmp10 = triton_helpers.promote_to_tensor(tl.sum(tmp8, 0))
tmp11 = 256.0
tmp12 = tmp10 / tmp11
tmp13 = 100.0
tmp14 = tmp12 * tmp13
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp14, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_abs_add_div_mean_mul_sub_0[grid(1)](buf1, arg0_1,
arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class MAPELossNew(nn.Module):
def __init__(self, eps=1e-08):
super(MAPELossNew, self).__init__()
self.eps = eps
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| cerisara/weibull-knowledge-informed-ml | MAPELoss | false | 1,643 | [
"MIT"
] | 0 | 19017817f5324fb1fffd8322d2d3567a6271948c | https://github.com/cerisara/weibull-knowledge-informed-ml/tree/19017817f5324fb1fffd8322d2d3567a6271948c | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, eps=1e-08):
super().__init__()
self.eps = eps
def forward(self, y_hat, y):
return torch.mean(torch.abs(y - y_hat) / torch.abs(y + self.eps)) * 100
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MY_NGM_FFNN | # 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_6/inductor_cache/r3/cr3febcwm3t44fuoitsx3ou2p6xg4sk4f7unagmmrvffasxf47te.py
# Topologically Sorted Source Nodes: [hidden1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# hidden1 => 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=[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_6/inductor_cache/g5/cg5f2rptqnpi2mrqpqc4tujqpbrrrjrse6plhgftx425znsffpfv.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_5, [-1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), 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
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
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/yh/cyhogxneodczl7mcnuf7mkhxldvr2nc5wj5e42agntthff4e45p7.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_2 = async_compile.triton('triton_poi_fused__log_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__log_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__log_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')
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 = 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, ))
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
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [hidden1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf8, 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
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [hidden2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf7, 256, grid=grid(256), 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, 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: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_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: [log_softmax], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_2.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, 4), (4, 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((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)
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 random
import torch
import torch.nn as nn
from collections import defaultdict
import torch.nn.functional as F
import torch.optim as optim
class MY_NGM_FFNN(nn.Module):
def __init__(self, alpha, input_dim, hidden1_dim, hidden2_dim,
output_dim, device=torch.device('cpu')):
super(MY_NGM_FFNN, self).__init__()
self.hidden1 = nn.Linear(input_dim, hidden1_dim)
self.hidden2 = nn.Linear(hidden1_dim, hidden2_dim)
self.output = nn.Linear(hidden2_dim, output_dim)
self.alpha = alpha
self.device = device
self
def save(self, output_dir, model_name):
None
torch.save(self.state_dict(), output_dir + model_name + '.pt')
None
def load(self, output_dir, model_name):
None
self.load_state_dict(torch.load(output_dir + model_name + '.pt'))
None
def forward(self, tf_idf_vec):
hidden1 = F.relu(self.hidden1(tf_idf_vec))
hidden2 = F.relu(self.hidden2(hidden1))
return F.log_softmax(self.output(hidden2), -1)
def reset_parameters(self):
self.hidden1.reset_parameters()
self.hidden2.reset_parameters()
self.output.reset_parameters()
def get_last_hidden(self, tf_idf_vec):
hidden1 = F.relu(self.hidden1(tf_idf_vec))
return F.relu(self.hidden2(hidden1))
def train_(self, seed_nodes, train_node_pairs, node2vec, node2label,
num_epoch, batch_size, learning_rate):
None
self.train()
loss_function = nn.NLLLoss()
optimizer = optim.SGD(self.parameters(), lr=learning_rate)
node2neighbors = defaultdict(list)
for src, dest in train_node_pairs:
node2neighbors[src].append(dest)
node2neighbors[dest].append(src)
labeled_nodes = dict()
for node in seed_nodes:
labeled_nodes[node] = node2label[node]
iteration = 1
while iteration < 2:
None
None
iteration += 1
for e in range(NUM_EPOCH):
train_node_pairs_cpy = train_node_pairs[:]
total_loss = 0
count = 0
while train_node_pairs_cpy:
optimizer.zero_grad()
loss = torch.tensor(0, dtype=torch.float32, device=self
.device)
try:
batch = random.sample(train_node_pairs_cpy, batch_size)
except ValueError:
break
for src, dest in batch:
count += 1
train_node_pairs_cpy.remove((src, dest))
src_vec = torch.tensor(node2vec[src])
dest_vec = torch.tensor(node2vec[dest])
if src in labeled_nodes:
src_target = torch.tensor([labeled_nodes[src]])
src_softmax = self.forward(torch.tensor(src_vec))
src_incident_edges = len(node2neighbors[src])
loss += loss_function(src_softmax.view(1, -1),
src_target) * (1 / src_incident_edges)
if dest in labeled_nodes:
dest_target = torch.tensor([labeled_nodes[dest]])
dest_softmax = self.forward(torch.tensor(dest_vec))
dest_incident_edges = len(node2neighbors[dest])
loss += loss_function(dest_softmax.view(1, -1),
dest_target) * (1 / dest_incident_edges)
loss += self.alpha * torch.dist(self.
get_last_hidden(src_vec), self.get_last_hidden(
dest_vec))
if loss.item() != 0:
assert not torch.isnan(loss)
loss.backward()
optimizer.step()
total_loss += loss.item()
del loss
total_loss / len(labeled_nodes)
None
for node in list(labeled_nodes.keys()):
label = labeled_nodes[node]
for neighbor in node2neighbors[node]:
if neighbor not in labeled_nodes:
labeled_nodes[neighbor] = label
def predict(self, tf_idf_vec):
return torch.argmax(self.forward(tf_idf_vec)).item()
def evaluate(self, test_nodes, node2vec, node2label):
self.eval()
None
correct_count = 0
for node in test_nodes:
predicted = self.predict(torch.tensor(node2vec[node], device=
self.device))
None
if predicted == node2label[node]:
correct_count += 1
return float(correct_count) / len(test_nodes)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'alpha': 4, 'input_dim': 4, 'hidden1_dim': 4,
'hidden2_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 random
import torch.nn as nn
from collections import defaultdict
import torch.nn.functional as F
import torch.optim as optim
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__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
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
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused__log_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')
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) = 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,))
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
buf8 = 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, buf8, 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
buf7 = 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, buf7, 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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__log_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__log_softmax_2[grid(256)](buf5, buf6, 256, 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, 4), (4, 1), 0), reinterpret_tensor(
buf3, (64, 4), (4, 1), 0), buf6, primals_6, buf7, primals_4, buf8
class MY_NGM_FFNNNew(nn.Module):
def __init__(self, alpha, input_dim, hidden1_dim, hidden2_dim,
output_dim, device=torch.device('cpu')):
super(MY_NGM_FFNNNew, self).__init__()
self.hidden1 = nn.Linear(input_dim, hidden1_dim)
self.hidden2 = nn.Linear(hidden1_dim, hidden2_dim)
self.output = nn.Linear(hidden2_dim, output_dim)
self.alpha = alpha
self.device = device
self
def save(self, output_dir, model_name):
None
torch.save(self.state_dict(), output_dir + model_name + '.pt')
None
def load(self, output_dir, model_name):
None
self.load_state_dict(torch.load(output_dir + model_name + '.pt'))
None
def reset_parameters(self):
self.hidden1.reset_parameters()
self.hidden2.reset_parameters()
self.output.reset_parameters()
def get_last_hidden(self, tf_idf_vec):
hidden1 = F.relu(self.hidden1(tf_idf_vec))
return F.relu(self.hidden2(hidden1))
def train_(self, seed_nodes, train_node_pairs, node2vec, node2label,
num_epoch, batch_size, learning_rate):
None
self.train()
loss_function = nn.NLLLoss()
optimizer = optim.SGD(self.parameters(), lr=learning_rate)
node2neighbors = defaultdict(list)
for src, dest in train_node_pairs:
node2neighbors[src].append(dest)
node2neighbors[dest].append(src)
labeled_nodes = dict()
for node in seed_nodes:
labeled_nodes[node] = node2label[node]
iteration = 1
while iteration < 2:
None
None
iteration += 1
for e in range(NUM_EPOCH):
train_node_pairs_cpy = train_node_pairs[:]
total_loss = 0
count = 0
while train_node_pairs_cpy:
optimizer.zero_grad()
loss = torch.tensor(0, dtype=torch.float32, device=self
.device)
try:
batch = random.sample(train_node_pairs_cpy, batch_size)
except ValueError:
break
for src, dest in batch:
count += 1
train_node_pairs_cpy.remove((src, dest))
src_vec = torch.tensor(node2vec[src])
dest_vec = torch.tensor(node2vec[dest])
if src in labeled_nodes:
src_target = torch.tensor([labeled_nodes[src]])
src_softmax = self.forward(torch.tensor(src_vec))
src_incident_edges = len(node2neighbors[src])
loss += loss_function(src_softmax.view(1, -1),
src_target) * (1 / src_incident_edges)
if dest in labeled_nodes:
dest_target = torch.tensor([labeled_nodes[dest]])
dest_softmax = self.forward(torch.tensor(dest_vec))
dest_incident_edges = len(node2neighbors[dest])
loss += loss_function(dest_softmax.view(1, -1),
dest_target) * (1 / dest_incident_edges)
loss += self.alpha * torch.dist(self.
get_last_hidden(src_vec), self.get_last_hidden(
dest_vec))
if loss.item() != 0:
assert not torch.isnan(loss)
loss.backward()
optimizer.step()
total_loss += loss.item()
del loss
total_loss / len(labeled_nodes)
None
for node in list(labeled_nodes.keys()):
label = labeled_nodes[node]
for neighbor in node2neighbors[node]:
if neighbor not in labeled_nodes:
labeled_nodes[neighbor] = label
def predict(self, tf_idf_vec):
return torch.argmax(self.forward(tf_idf_vec)).item()
def evaluate(self, test_nodes, node2vec, node2label):
self.eval()
None
correct_count = 0
for node in test_nodes:
predicted = self.predict(torch.tensor(node2vec[node], device=
self.device))
None
if predicted == node2label[node]:
correct_count += 1
return float(correct_count) / len(test_nodes)
def forward(self, input_0):
primals_1 = self.hidden1.weight
primals_2 = self.hidden1.bias
primals_4 = self.hidden2.weight
primals_5 = self.hidden2.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]
| calpoly-bioinf/knowledge_driven_modeling | MY_NGM_FFNN | false | 1,644 | [
"MIT"
] | 0 | dbe55d5bb07f7c5a1834a21fde8833f295e3ac96 | https://github.com/calpoly-bioinf/knowledge_driven_modeling/tree/dbe55d5bb07f7c5a1834a21fde8833f295e3ac96 | import random
import torch
import torch.nn as nn
from collections import defaultdict
import torch.nn.functional as F
import torch.optim as optim
class Model(nn.Module):
def __init__(self, alpha, input_dim, hidden1_dim, hidden2_dim,
output_dim, device=torch.device('cpu')):
super().__init__()
self.hidden1 = nn.Linear(input_dim, hidden1_dim)
self.hidden2 = nn.Linear(hidden1_dim, hidden2_dim)
self.output = nn.Linear(hidden2_dim, output_dim)
self.alpha = alpha
self.device = device
self
def save(self, output_dir, model_name):
None
torch.save(self.state_dict(), output_dir + model_name + '.pt')
None
def load(self, output_dir, model_name):
None
self.load_state_dict(torch.load(output_dir + model_name + '.pt'))
None
def forward(self, tf_idf_vec):
hidden1 = F.relu(self.hidden1(tf_idf_vec))
hidden2 = F.relu(self.hidden2(hidden1))
return F.log_softmax(self.output(hidden2), -1)
def reset_parameters(self):
self.hidden1.reset_parameters()
self.hidden2.reset_parameters()
self.output.reset_parameters()
def get_last_hidden(self, tf_idf_vec):
hidden1 = F.relu(self.hidden1(tf_idf_vec))
return F.relu(self.hidden2(hidden1))
def train_(self, seed_nodes, train_node_pairs, node2vec, node2label,
num_epoch, batch_size, learning_rate):
None
self.train()
loss_function = nn.NLLLoss()
optimizer = optim.SGD(self.parameters(), lr=learning_rate)
node2neighbors = defaultdict(list)
for src, dest in train_node_pairs:
node2neighbors[src].append(dest)
node2neighbors[dest].append(src)
labeled_nodes = dict()
for node in seed_nodes:
labeled_nodes[node] = node2label[node]
iteration = 1
while iteration < 2:
None
None
iteration += 1
for e in range(NUM_EPOCH):
train_node_pairs_cpy = train_node_pairs[:]
total_loss = 0
count = 0
while train_node_pairs_cpy:
optimizer.zero_grad()
loss = torch.tensor(0, dtype=torch.float32, device=self
.device)
try:
batch = random.sample(train_node_pairs_cpy, batch_size)
except ValueError:
break
for src, dest in batch:
count += 1
train_node_pairs_cpy.remove((src, dest))
src_vec = torch.tensor(node2vec[src])
dest_vec = torch.tensor(node2vec[dest])
if src in labeled_nodes:
src_target = torch.tensor([labeled_nodes[src]])
src_softmax = self.forward(torch.tensor(src_vec))
src_incident_edges = len(node2neighbors[src])
loss += loss_function(src_softmax.view(1, -1),
src_target) * (1 / src_incident_edges)
if dest in labeled_nodes:
dest_target = torch.tensor([labeled_nodes[dest]])
dest_softmax = self.forward(torch.tensor(dest_vec))
dest_incident_edges = len(node2neighbors[dest])
loss += loss_function(dest_softmax.view(1, -1),
dest_target) * (1 / dest_incident_edges)
loss += self.alpha * torch.dist(self.
get_last_hidden(src_vec), self.get_last_hidden(
dest_vec))
if loss.item() != 0:
assert not torch.isnan(loss)
loss.backward()
optimizer.step()
# ... truncated (>4000 chars) for memory efficiency |
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_6/inductor_cache/au/cau6qypw2vz4drppp6yr6chutchyhnniousxhhlq2y5r3yu3gep5.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_1 => relu
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_3), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_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': ['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_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 = 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_6/inductor_cache/vh/cvhowampoosezwy5zm5vfkdmhzrvsn2u2gxpn4cchngk4b74ympu.py
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => amax, div, exp, sub, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%addmm_2, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%addmm_2, %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, [1], True), kwargs = {})
# %div : [num_users=1] = 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=[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_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__softmax_1(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
x2 = xindex
x1 = (xindex // 2)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (2*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (2*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp4 = tmp0 - tmp3
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp1 - tmp3
tmp7 = tl_math.exp(tmp6)
tmp8 = tmp2 - tmp3
tmp9 = tl_math.exp(tmp8)
tmp10 = tmp7 + tmp9
tmp11 = tmp5 / tmp10
tl.store(out_ptr0 + (x2), tmp11, 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, 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, ))
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, 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 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_0.run(buf1, primals_3, 256, grid=grid(256), stream=stream0)
del primals_3
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf1, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu]
triton_poi_fused_relu_0.run(buf3, primals_5, 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, buf3, reinterpret_tensor(primals_6, (4, 2), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf4, buf5, 128, grid=grid(128), stream=stream0)
del buf4
return (buf5, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), buf1, buf3, buf5, 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, 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)
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 as nn
import torch.nn.functional as F
class Discriminator(nn.Module):
def __init__(self, n_layersDecod, hidden_size, output_size=2):
super(Discriminator, self).__init__()
self.map1 = nn.Linear(n_layersDecod * hidden_size, hidden_size)
self.map2 = nn.Linear(hidden_size, hidden_size)
self.map3 = nn.Linear(hidden_size, output_size)
self.n_layersDecod = n_layersDecod
self.hidden_size = hidden_size
def forward(self, x):
x = x.view(-1, self.n_layersDecod * self.hidden_size)
x = F.relu(self.map1(x))
x = F.relu(self.map2(x))
return nn.Softmax()(self.map3(x))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_layersDecod': 1, '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
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_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.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__softmax_1(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
x2 = xindex
x1 = xindex // 2
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 2 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 2 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp4 = tmp0 - tmp3
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp1 - tmp3
tmp7 = tl_math.exp(tmp6)
tmp8 = tmp2 - tmp3
tmp9 = tl_math.exp(tmp8)
tmp10 = tmp7 + tmp9
tmp11 = tmp5 / tmp10
tl.store(out_ptr0 + x2, tmp11, 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, 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,))
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, 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 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_relu_0[grid(256)](buf1, primals_3, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_3
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(buf1, reinterpret_tensor(primals_4, (4, 4), (1, 4
), 0), out=buf2)
buf3 = buf2
del buf2
triton_poi_fused_relu_0[grid(256)](buf3, primals_5, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
extern_kernels.addmm(primals_7, buf3, reinterpret_tensor(primals_6,
(4, 2), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
triton_poi_fused__softmax_1[grid(128)](buf4, buf5, 128, XBLOCK=128,
num_warps=4, num_stages=1)
del buf4
return buf5, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0
), buf1, buf3, buf5, primals_6, primals_4
class DiscriminatorNew(nn.Module):
def __init__(self, n_layersDecod, hidden_size, output_size=2):
super(DiscriminatorNew, self).__init__()
self.map1 = nn.Linear(n_layersDecod * hidden_size, hidden_size)
self.map2 = nn.Linear(hidden_size, hidden_size)
self.map3 = nn.Linear(hidden_size, output_size)
self.n_layersDecod = n_layersDecod
self.hidden_size = hidden_size
def forward(self, input_0):
primals_2 = self.map1.weight
primals_3 = self.map1.bias
primals_4 = self.map2.weight
primals_5 = self.map2.bias
primals_6 = self.map3.weight
primals_7 = self.map3.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| carsault/chord_sequence_prediction | Discriminator | false | 1,645 | [
"MIT"
] | 0 | 6eb539a963ca6350bcf0c88b8d8756775ad7c488 | https://github.com/carsault/chord_sequence_prediction/tree/6eb539a963ca6350bcf0c88b8d8756775ad7c488 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, n_layersDecod, hidden_size, output_size=2):
super().__init__()
self.map1 = nn.Linear(n_layersDecod * hidden_size, hidden_size)
self.map2 = nn.Linear(hidden_size, hidden_size)
self.map3 = nn.Linear(hidden_size, output_size)
self.n_layersDecod = n_layersDecod
self.hidden_size = hidden_size
def forward(self, x):
x = x.view(-1, self.n_layersDecod * self.hidden_size)
x = F.relu(self.map1(x))
x = F.relu(self.map2(x))
return nn.Softmax()(self.map3(x))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [1, 4]
|
NeuralNet | # 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_6/inductor_cache/r6/cr6zjqblm6bzp3ra5bxxxousvsozkpcmx4vt2ow2xooiqo45tdke.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten._to_copy]
# Source node to ATen node mapping:
# x => convert_element_type
# Graph fragment:
# %convert_element_type : [num_users=1] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%primals_1, torch.float64), kwargs = {})
triton_poi_fused__to_copy_0 = async_compile.triton('triton_poi_fused__to_copy_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: '*fp64', 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__to_copy_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__to_copy_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 = tmp0.to(tl.float64)
tl.store(out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/ff/cffiuioxgfnd7gb3ukcfrnvq6qtjvejobnrgl3kgqc2qcmiswonl.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_3 => 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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp64', 1: '*fp64', 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 = tl.full([1], 0.0, tl.float64)
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_6/inductor_cache/tx/ctxcn7g5u6cl3pjvoozpr3ibzoccygyhuddalmafgdrhgwsazwae.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_5 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_3, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_3, %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: '*fp64', 1: '*fp64', 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 = libdevice.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_6/inductor_cache/wc/cwceujsgrc3jjty6om26j2viqcp75hv5c3fbl3mlyr7yjhoifkpm.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_5 => 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: '*fp64', 1: '*fp64', 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 = 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((4, 4, 4, 4), (64, 16, 4, 1), torch.float64)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten._to_copy]
stream0 = get_raw_stream(0)
triton_poi_fused__to_copy_0.run(primals_1, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float64)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf0, (64, 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, 4), (64, 16, 4, 1), 0); del buf1 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf2, primals_3, buf6, 256, grid=grid(256), stream=stream0)
del primals_3
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float64)
# Topologically Sorted Source Nodes: [x_4], 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
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float64)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf3, buf4, 256, grid=grid(256), stream=stream0)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
del buf4
return (buf5, reinterpret_tensor(buf0, (64, 4), (4, 1), 0), reinterpret_tensor(buf2, (64, 4), (4, 1), 0), 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((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.float64)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float64)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float64)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float64)
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
class NeuralNet(torch.nn.Module):
def __init__(self, input_size, output_size, hidden_size, lr=0.0001):
super(NeuralNet, self).__init__()
self.input_size = input_size
self.hidden_size = hidden_size
self.output_size = output_size
self.HiddenLayer = torch.nn.Linear(self.input_size, self.
hidden_size, bias=True).double()
self.DropoutLayer = torch.nn.Dropout(p=0.0)
self.OutputLayer = torch.nn.Linear(self.hidden_size, self.
output_size, bias=True).double()
self.neural_tangent_kernel = None
self.learning_rate = lr
self.optimizer = torch.optim.SGD(self.parameters(), lr=self.
learning_rate)
self.loss_arr = []
self.__policy_loss = []
def forward(self, x):
x = x.double()
x = self.HiddenLayer(x)
x = self.DropoutLayer(x)
x = torch.relu(x)
x = self.OutputLayer(x)
x = torch.softmax(x, dim=1)
return x
def __partial_derivatives(self, x):
self.zero_grad()
w1 = torch.empty(self.output_size, self.hidden_size * self.
input_size + self.hidden_size, dtype=torch.float64)
w2 = torch.empty(self.output_size, self.hidden_size * self.
output_size + self.output_size, dtype=torch.float64)
for i in range(self.output_size):
y = self.forward(x)
y = y[0][i]
y.backward()
wi1 = self.HiddenLayer.weight.grad
wi1 = torch.reshape(wi1, [wi1.shape[0] * wi1.shape[1], 1])
wi1 = torch.cat([wi1, self.HiddenLayer.bias.grad.unsqueeze(1)])
wi2 = self.OutputLayer.weight.grad
wi2 = torch.reshape(wi2, [wi2.shape[0] * wi2.shape[1], 1])
wi2 = torch.cat([wi2, self.OutputLayer.bias.grad.unsqueeze(1)])
wi1g = wi1.clone().detach()
wi2g = wi2.clone().detach()
w1[i] = wi1g.squeeze()
w2[i] = wi2g.squeeze()
self.zero_grad()
return w1, w2
def compute_neural_tangent_kernel(self, x):
kernel = torch.zeros([x.shape[0] * self.output_size, x.shape[0] *
self.output_size], dtype=torch.float64, requires_grad=False)
i = 0
for x1 in x.data:
w1x1, w2x1 = self.__partial_derivatives(x1.unsqueeze(dim=0))
j = 0
for x2 in x.data:
w1x2, w2x2 = self.__partial_derivatives(x2.unsqueeze(dim=0))
kernel[self.output_size * i:self.output_size * i + self.
output_size, self.output_size * j:self.output_size * j +
self.output_size] = torch.matmul(w1x1, w1x2.transpose(0, 1)
) + torch.matmul(w2x1, w2x2.transpose(0, 1))
j += 1
i += 1
self.neural_tangent_kernel = kernel
return kernel
def train_network(self, log_probs, gains, gains_normed=0):
self.__policy_loss = []
eps = 1e-08
batch_len = len(gains)
gains.clone().detach().requires_grad_(True)
if gains_normed:
gains_norm = gains
else:
gains_norm = (gains - gains.mean()) / (gains.std() + eps)
for k in range(batch_len):
self.__policy_loss.append(-log_probs[k] * gains_norm[k])
self.__policy_loss = torch.cat(self.__policy_loss).sum()
self.__policy_loss.backward()
self.optimizer.step()
self.optimizer.zero_grad()
self.loss_arr.append(self.__policy_loss.item())
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4, 'output_size': 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 libdevice
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__to_copy_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 = tmp0.to(tl.float64)
tl.store(out_ptr0 + x0, tmp1, 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 = tl.full([1], 0.0, tl.float64)
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 = libdevice.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 = 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((4, 4, 4, 4), (64, 16, 4, 1), torch.float64)
get_raw_stream(0)
triton_poi_fused__to_copy_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.float64)
extern_kernels.mm(reinterpret_tensor(buf0, (64, 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, 4), (64, 16, 4, 1), 0)
del buf1
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(256)](buf2,
primals_3, buf6, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_3
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float64)
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
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float64)
triton_poi_fused__softmax_2[grid(256)](buf3, buf4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused__softmax_3[grid(256)](buf4, buf5, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf4
return buf5, reinterpret_tensor(buf0, (64, 4), (4, 1), 0
), reinterpret_tensor(buf2, (64, 4), (4, 1), 0), buf5, primals_4, buf6
class NeuralNetNew(torch.nn.Module):
def __init__(self, input_size, output_size, hidden_size, lr=0.0001):
super(NeuralNetNew, self).__init__()
self.input_size = input_size
self.hidden_size = hidden_size
self.output_size = output_size
self.HiddenLayer = torch.nn.Linear(self.input_size, self.
hidden_size, bias=True).double()
self.DropoutLayer = torch.nn.Dropout(p=0.0)
self.OutputLayer = torch.nn.Linear(self.hidden_size, self.
output_size, bias=True).double()
self.neural_tangent_kernel = None
self.learning_rate = lr
self.optimizer = torch.optim.SGD(self.parameters(), lr=self.
learning_rate)
self.loss_arr = []
self.__policy_loss = []
def __partial_derivatives(self, x):
self.zero_grad()
w1 = torch.empty(self.output_size, self.hidden_size * self.
input_size + self.hidden_size, dtype=torch.float64)
w2 = torch.empty(self.output_size, self.hidden_size * self.
output_size + self.output_size, dtype=torch.float64)
for i in range(self.output_size):
y = self.forward(x)
y = y[0][i]
y.backward()
wi1 = self.HiddenLayer.weight.grad
wi1 = torch.reshape(wi1, [wi1.shape[0] * wi1.shape[1], 1])
wi1 = torch.cat([wi1, self.HiddenLayer.bias.grad.unsqueeze(1)])
wi2 = self.OutputLayer.weight.grad
wi2 = torch.reshape(wi2, [wi2.shape[0] * wi2.shape[1], 1])
wi2 = torch.cat([wi2, self.OutputLayer.bias.grad.unsqueeze(1)])
wi1g = wi1.clone().detach()
wi2g = wi2.clone().detach()
w1[i] = wi1g.squeeze()
w2[i] = wi2g.squeeze()
self.zero_grad()
return w1, w2
def compute_neural_tangent_kernel(self, x):
kernel = torch.zeros([x.shape[0] * self.output_size, x.shape[0] *
self.output_size], dtype=torch.float64, requires_grad=False)
i = 0
for x1 in x.data:
w1x1, w2x1 = self.__partial_derivatives(x1.unsqueeze(dim=0))
j = 0
for x2 in x.data:
w1x2, w2x2 = self.__partial_derivatives(x2.unsqueeze(dim=0))
kernel[self.output_size * i:self.output_size * i + self.
output_size, self.output_size * j:self.output_size * j +
self.output_size] = torch.matmul(w1x1, w1x2.transpose(0, 1)
) + torch.matmul(w2x1, w2x2.transpose(0, 1))
j += 1
i += 1
self.neural_tangent_kernel = kernel
return kernel
def train_network(self, log_probs, gains, gains_normed=0):
self.__policy_loss = []
eps = 1e-08
batch_len = len(gains)
gains.clone().detach().requires_grad_(True)
if gains_normed:
gains_norm = gains
else:
gains_norm = (gains - gains.mean()) / (gains.std() + eps)
for k in range(batch_len):
self.__policy_loss.append(-log_probs[k] * gains_norm[k])
self.__policy_loss = torch.cat(self.__policy_loss).sum()
self.__policy_loss.backward()
self.optimizer.step()
self.optimizer.zero_grad()
self.loss_arr.append(self.__policy_loss.item())
def forward(self, input_0):
primals_2 = self.HiddenLayer.weight
primals_3 = self.HiddenLayer.bias
primals_4 = self.OutputLayer.weight
primals_5 = self.OutputLayer.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| bva-bme/Constrained_Policy_Gradient | NeuralNet | false | 1,646 | [
"MIT"
] | 0 | 2331f55ff3bf06e2276662517c34cc45d5a51da8 | https://github.com/bva-bme/Constrained_Policy_Gradient/tree/2331f55ff3bf06e2276662517c34cc45d5a51da8 | import torch
class Model(torch.nn.Module):
def __init__(self, input_size, output_size, hidden_size, lr=0.0001):
super().__init__()
self.input_size = input_size
self.hidden_size = hidden_size
self.output_size = output_size
self.HiddenLayer = torch.nn.Linear(self.input_size, self.
hidden_size, bias=True).double()
self.DropoutLayer = torch.nn.Dropout(p=0.0)
self.OutputLayer = torch.nn.Linear(self.hidden_size, self.
output_size, bias=True).double()
self.neural_tangent_kernel = None
self.learning_rate = lr
self.optimizer = torch.optim.SGD(self.parameters(), lr=self.
learning_rate)
self.loss_arr = []
self.__policy_loss = []
def forward(self, x):
x = x.double()
x = self.HiddenLayer(x)
x = self.DropoutLayer(x)
x = torch.relu(x)
x = self.OutputLayer(x)
x = torch.softmax(x, dim=1)
return x
def __partial_derivatives(self, x):
self.zero_grad()
w1 = torch.empty(self.output_size, self.hidden_size * self.
input_size + self.hidden_size, dtype=torch.float64)
w2 = torch.empty(self.output_size, self.hidden_size * self.
output_size + self.output_size, dtype=torch.float64)
for i in range(self.output_size):
y = self.forward(x)
y = y[0][i]
y.backward()
wi1 = self.HiddenLayer.weight.grad
wi1 = torch.reshape(wi1, [wi1.shape[0] * wi1.shape[1], 1])
wi1 = torch.cat([wi1, self.HiddenLayer.bias.grad.unsqueeze(1)])
wi2 = self.OutputLayer.weight.grad
wi2 = torch.reshape(wi2, [wi2.shape[0] * wi2.shape[1], 1])
wi2 = torch.cat([wi2, self.OutputLayer.bias.grad.unsqueeze(1)])
wi1g = wi1.clone().detach()
wi2g = wi2.clone().detach()
w1[i] = wi1g.squeeze()
w2[i] = wi2g.squeeze()
self.zero_grad()
return w1, w2
def compute_neural_tangent_kernel(self, x):
kernel = torch.zeros([x.shape[0] * self.output_size, x.shape[0] *
self.output_size], dtype=torch.float64, requires_grad=False)
i = 0
for x1 in x.data:
w1x1, w2x1 = self.__partial_derivatives(x1.unsqueeze(dim=0))
j = 0
for x2 in x.data:
w1x2, w2x2 = self.__partial_derivatives(x2.unsqueeze(dim=0))
kernel[self.output_size * i:self.output_size * i + self.
output_size, self.output_size * j:self.output_size * j +
self.output_size] = torch.matmul(w1x1, w1x2.transpose(0, 1)
) + torch.matmul(w2x1, w2x2.transpose(0, 1))
j += 1
i += 1
self.neural_tangent_kernel = kernel
return kernel
def train_network(self, log_probs, gains, gains_normed=0):
self.__policy_loss = []
eps = 1e-08
batch_len = len(gains)
gains.clone().detach().requires_grad_(True)
if gains_normed:
gains_norm = gains
else:
gains_norm = (gains - gains.mean()) / (gains.std() + eps)
for k in range(batch_len):
self.__policy_loss.append(-log_probs[k] * gains_norm[k])
self.__policy_loss = torch.cat(self.__policy_loss).sum()
self.__policy_loss.backward()
self.optimizer.step()
self.optimizer.zero_grad()
self.loss_arr.append(self.__policy_loss.item())
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
RMSLELoss | # 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_6/inductor_cache/qm/cqmset2orwdkxsryywb7cbljhejyscv2fzgw3cfi4bew2zmihtkv.py
# Topologically Sorted Source Nodes: [add, log, add_1, log_1, mse_loss, add_2, sqrt], Original ATen: [aten.add, aten.log, aten.mse_loss, aten.sqrt]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# log => log
# log_1 => log_1
# mse_loss => mean, pow_1, sub
# sqrt => sqrt
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg1_1, 1), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add_1,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %log_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%pow_1,), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean, 1e-08), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add_2,), kwargs = {})
triton_per_fused_add_log_mse_loss_sqrt_0 = async_compile.triton('triton_per_fused_add_log_mse_loss_sqrt_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_log_mse_loss_sqrt_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, '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_add_log_mse_loss_sqrt_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)
tmp4 = tl.load(in_ptr1 + (r0), None)
tmp1 = 1.0
tmp2 = tmp0 + tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp4 + tmp1
tmp6 = tl_math.log(tmp5)
tmp7 = tmp3 - tmp6
tmp8 = tmp7 * tmp7
tmp9 = tl.broadcast_to(tmp8, [RBLOCK])
tmp11 = triton_helpers.promote_to_tensor(tl.sum(tmp9, 0))
tmp12 = 256.0
tmp13 = tmp11 / tmp12
tmp14 = 1e-08
tmp15 = tmp13 + tmp14
tmp16 = libdevice.sqrt(tmp15)
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp16, 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)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [add, log, add_1, log_1, mse_loss, add_2, sqrt], Original ATen: [aten.add, aten.log, aten.mse_loss, aten.sqrt]
stream0 = get_raw_stream(0)
triton_per_fused_add_log_mse_loss_sqrt_0.run(buf1, arg0_1, arg1_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (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((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 RMSLELoss(nn.Module):
def __init__(self, eps=1e-08):
super(RMSLELoss, self).__init__()
self.mse = nn.MSELoss()
self.eps = eps
def forward(self, y_hat, y):
return torch.sqrt(self.mse(torch.log(y_hat + 1), torch.log(y + 1)) +
self.eps)
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 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_per_fused_add_log_mse_loss_sqrt_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)
tmp4 = tl.load(in_ptr1 + r0, None)
tmp1 = 1.0
tmp2 = tmp0 + tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp4 + tmp1
tmp6 = tl_math.log(tmp5)
tmp7 = tmp3 - tmp6
tmp8 = tmp7 * tmp7
tmp9 = tl.broadcast_to(tmp8, [RBLOCK])
tmp11 = triton_helpers.promote_to_tensor(tl.sum(tmp9, 0))
tmp12 = 256.0
tmp13 = tmp11 / tmp12
tmp14 = 1e-08
tmp15 = tmp13 + tmp14
tmp16 = libdevice.sqrt(tmp15)
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp16, 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)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_log_mse_loss_sqrt_0[grid(1)](buf1, arg0_1,
arg1_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf1,
class RMSLELossNew(nn.Module):
def __init__(self, eps=1e-08):
super(RMSLELossNew, self).__init__()
self.mse = nn.MSELoss()
self.eps = eps
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| cerisara/weibull-knowledge-informed-ml | RMSLELoss | false | 1,647 | [
"MIT"
] | 0 | 19017817f5324fb1fffd8322d2d3567a6271948c | https://github.com/cerisara/weibull-knowledge-informed-ml/tree/19017817f5324fb1fffd8322d2d3567a6271948c | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, eps=1e-08):
super().__init__()
self.mse = nn.MSELoss()
self.eps = eps
def forward(self, y_hat, y):
return torch.sqrt(self.mse(torch.log(y_hat + 1), torch.log(y + 1)) +
self.eps)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
BERTEmbedding4 | # 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_6/inductor_cache/uq/cuq6dnw6x5lmjovxm3lh3rsy464zindfpkgwt3vkq3t2hvc6cdlj.py
# Topologically Sorted Source Nodes: [mul, add, x], Original ATen: [aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# x => add_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %primals_2), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_4), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %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: '*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_mul_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_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 % 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x2), xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
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 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (1, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (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, x], Original ATen: [aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_0.run(primals_1, primals_2, primals_4, primals_3, buf0, 256, grid=grid(256), stream=stream0)
del primals_3
del primals_4
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((4, 4), (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, 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)
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 torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class LearnedPositionalEmbedding3(nn.Module):
def __init__(self, d_model, max_len=512):
super().__init__()
pe = torch.zeros(max_len, d_model).float()
self.a_2 = nn.Parameter(torch.ones_like(pe))
self.b_2 = nn.Parameter(torch.zeros_like(pe))
pe.require_grad = True
pe = pe.unsqueeze(0)
self.pe = nn.Parameter(pe)
torch.nn.init.normal_(self.pe, std=d_model ** -0.5)
def forward(self, x):
return self.a_2 * self.pe[:, :x.size(1)] + self.b_2
class BERTEmbedding4(nn.Module):
"""
BERT Embedding which is consisted with under features
1. PositionalEmbedding : adding positional information using sin, cos
sum of all these features are output of BERTEmbedding
"""
def __init__(self, input_dim, max_len, dropout=0.1):
"""
:param vocab_size: total vocab size
:param embed_size: embedding size of token embedding
:param dropout: dropout rate
"""
super().__init__()
self.learnedPosition = LearnedPositionalEmbedding3(d_model=
input_dim, max_len=max_len)
self.dropout = nn.Dropout(p=dropout)
def forward(self, sequence):
x = self.learnedPosition(sequence) + sequence
return self.dropout(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'max_len': 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
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
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, 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 % 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x2, xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tl.store(out_ptr0 + x2, tmp6, 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, (1, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (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_1, primals_2,
primals_4, primals_3, buf0, 256, XBLOCK=128, num_warps=4,
num_stages=1)
del primals_3
del primals_4
return buf0, primals_1, primals_2
class LearnedPositionalEmbedding3(nn.Module):
def __init__(self, d_model, max_len=512):
super().__init__()
pe = torch.zeros(max_len, d_model).float()
self.a_2 = nn.Parameter(torch.ones_like(pe))
self.b_2 = nn.Parameter(torch.zeros_like(pe))
pe.require_grad = True
pe = pe.unsqueeze(0)
self.pe = nn.Parameter(pe)
torch.nn.init.normal_(self.pe, std=d_model ** -0.5)
def forward(self, x):
return self.a_2 * self.pe[:, :x.size(1)] + self.b_2
class BERTEmbedding4New(nn.Module):
"""
BERT Embedding which is consisted with under features
1. PositionalEmbedding : adding positional information using sin, cos
sum of all these features are output of BERTEmbedding
"""
def __init__(self, input_dim, max_len, dropout=0.1):
"""
:param vocab_size: total vocab size
:param embed_size: embedding size of token embedding
:param dropout: dropout rate
"""
super().__init__()
self.learnedPosition = LearnedPositionalEmbedding3(d_model=
input_dim, max_len=max_len)
self.dropout = nn.Dropout(p=dropout)
def forward(self, input_0):
primals_1 = self.learnedPosition.a_2
primals_4 = self.learnedPosition.b_2
primals_2 = self.learnedPosition.pe
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| byeongjokim/LateTemporalModeling3DCNN_for_sign | BERTEmbedding4 | false | 1,648 | [
"MIT"
] | 0 | e3a802fcf91dc3930aea782464ee34d9b747d3ab | https://github.com/byeongjokim/LateTemporalModeling3DCNN_for_sign/tree/e3a802fcf91dc3930aea782464ee34d9b747d3ab | import torch
import torch.nn as nn
from itertools import chain as chain
import torch.utils.data
import torch.hub
import torch.nn.parallel
import torch.optim
class LearnedPositionalEmbedding3(nn.Module):
def __init__(self, d_model, max_len=512):
super().__init__()
pe = torch.zeros(max_len, d_model).float()
self.a_2 = nn.Parameter(torch.ones_like(pe))
self.b_2 = nn.Parameter(torch.zeros_like(pe))
pe.require_grad = True
pe = pe.unsqueeze(0)
self.pe = nn.Parameter(pe)
torch.nn.init.normal_(self.pe, std=d_model ** -0.5)
def forward(self, x):
return self.a_2 * self.pe[:, :x.size(1)] + self.b_2
class Model(nn.Module):
"""
BERT Embedding which is consisted with under features
1. PositionalEmbedding : adding positional information using sin, cos
sum of all these features are output of BERTEmbedding
"""
def __init__(self, input_dim, max_len, dropout=0.1):
"""
:param vocab_size: total vocab size
:param embed_size: embedding size of token embedding
:param dropout: dropout rate
"""
super().__init__()
self.learnedPosition = LearnedPositionalEmbedding3(d_model=
input_dim, max_len=max_len)
self.dropout = nn.Dropout(p=dropout)
def forward(self, sequence):
x = self.learnedPosition(sequence) + sequence
return self.dropout(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
SpaceToBatch | # 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_6/inductor_cache/3u/c3ub52l73zdv4klgqzgxmtzrzxvztuyczv2jksnvrjr7erq7guxd.py
# Topologically Sorted Source Nodes: [reshape_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# reshape_1 => 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_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, 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 = 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')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', 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, 1, 1), (64, 16, 4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [reshape_1], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(arg0_1, buf0, 64, 4, grid=grid(64, 4), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf0, (64, 4, 1, 1), (4, 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
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.utils.data
class SpaceToDim(nn.Module):
def __init__(self, scale_factor, dims=(-2, -1), dim=0):
super(SpaceToDim, self).__init__()
self.scale_factor = scale_factor
self.dims = dims
self.dim = dim
def forward(self, x):
_shape = list(x.shape)
shape = _shape.copy()
dims = [x.dim() + self.dims[0] if self.dims[0] < 0 else self.dims[0
], x.dim() + self.dims[1] if self.dims[1] < 0 else self.dims[1]]
dims = [max(abs(dims[0]), abs(dims[1])), min(abs(dims[0]), abs(dims
[1]))]
if self.dim in dims:
raise RuntimeError("Integrate dimension can't be space dimension!")
shape[dims[0]] //= self.scale_factor
shape[dims[1]] //= self.scale_factor
shape.insert(dims[0] + 1, self.scale_factor)
shape.insert(dims[1] + 1, self.scale_factor)
dim = self.dim if self.dim < dims[1] else self.dim + 1
dim = dim if dim <= dims[0] else dim + 1
x = x.reshape(*shape)
perm = [dim, dims[1] + 1, dims[0] + 2]
perm = [i for i in range(min(perm))] + perm
perm.extend(i for i in range(x.dim()) if i not in perm)
x = x.permute(*perm)
shape = _shape
shape[self.dim] *= self.scale_factor ** 2
shape[self.dims[0]] //= self.scale_factor
shape[self.dims[1]] //= self.scale_factor
return x.reshape(*shape)
def extra_repr(self):
return f'scale_factor={self.scale_factor}'
class SpaceToBatch(nn.Module):
def __init__(self, block_size):
super(SpaceToBatch, self).__init__()
self.body = SpaceToDim(block_size, dim=0)
def forward(self, x):
return self.body(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'block_size': 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
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_clone_0(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
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')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
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, 1, 1), (64, 16, 4, 1, 1, 1),
torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64, 4)](arg0_1, buf0, 64, 4, XBLOCK=4,
YBLOCK=32, num_warps=4, num_stages=1)
del arg0_1
return reinterpret_tensor(buf0, (64, 4, 1, 1), (4, 1, 1, 1), 0),
class SpaceToDim(nn.Module):
def __init__(self, scale_factor, dims=(-2, -1), dim=0):
super(SpaceToDim, self).__init__()
self.scale_factor = scale_factor
self.dims = dims
self.dim = dim
def forward(self, x):
_shape = list(x.shape)
shape = _shape.copy()
dims = [x.dim() + self.dims[0] if self.dims[0] < 0 else self.dims[0
], x.dim() + self.dims[1] if self.dims[1] < 0 else self.dims[1]]
dims = [max(abs(dims[0]), abs(dims[1])), min(abs(dims[0]), abs(dims
[1]))]
if self.dim in dims:
raise RuntimeError("Integrate dimension can't be space dimension!")
shape[dims[0]] //= self.scale_factor
shape[dims[1]] //= self.scale_factor
shape.insert(dims[0] + 1, self.scale_factor)
shape.insert(dims[1] + 1, self.scale_factor)
dim = self.dim if self.dim < dims[1] else self.dim + 1
dim = dim if dim <= dims[0] else dim + 1
x = x.reshape(*shape)
perm = [dim, dims[1] + 1, dims[0] + 2]
perm = [i for i in range(min(perm))] + perm
perm.extend(i for i in range(x.dim()) if i not in perm)
x = x.permute(*perm)
shape = _shape
shape[self.dim] *= self.scale_factor ** 2
shape[self.dims[0]] //= self.scale_factor
shape[self.dims[1]] //= self.scale_factor
return x.reshape(*shape)
def extra_repr(self):
return f'scale_factor={self.scale_factor}'
class SpaceToBatchNew(nn.Module):
def __init__(self, block_size):
super(SpaceToBatchNew, self).__init__()
self.body = SpaceToDim(block_size, dim=0)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| cestcedric/TSSR-GAN | SpaceToBatch | false | 1,649 | [
"BSD-2-Clause",
"MIT"
] | 0 | d6e1b50409e0f0591660552993e6d5b70d41e766 | https://github.com/cestcedric/TSSR-GAN/tree/d6e1b50409e0f0591660552993e6d5b70d41e766 | import torch
import torch.nn as nn
import torch.utils.data
class SpaceToDim(nn.Module):
def __init__(self, scale_factor, dims=(-2, -1), dim=0):
super().__init__()
self.scale_factor = scale_factor
self.dims = dims
self.dim = dim
def forward(self, x):
_shape = list(x.shape)
shape = _shape.copy()
dims = [x.dim() + self.dims[0] if self.dims[0] < 0 else self.dims[0
], x.dim() + self.dims[1] if self.dims[1] < 0 else self.dims[1]]
dims = [max(abs(dims[0]), abs(dims[1])), min(abs(dims[0]), abs(dims
[1]))]
if self.dim in dims:
raise RuntimeError("Integrate dimension can't be space dimension!")
shape[dims[0]] //= self.scale_factor
shape[dims[1]] //= self.scale_factor
shape.insert(dims[0] + 1, self.scale_factor)
shape.insert(dims[1] + 1, self.scale_factor)
dim = self.dim if self.dim < dims[1] else self.dim + 1
dim = dim if dim <= dims[0] else dim + 1
x = x.reshape(*shape)
perm = [dim, dims[1] + 1, dims[0] + 2]
perm = [i for i in range(min(perm))] + perm
perm.extend(i for i in range(x.dim()) if i not in perm)
x = x.permute(*perm)
shape = _shape
shape[self.dim] *= self.scale_factor ** 2
shape[self.dims[0]] //= self.scale_factor
shape[self.dims[1]] //= self.scale_factor
return x.reshape(*shape)
def extra_repr(self):
return f'scale_factor={self.scale_factor}'
class Model(nn.Module):
def __init__(self, block_size):
super().__init__()
self.body = SpaceToDim(block_size, dim=0)
def forward(self, x):
return self.body(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.