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
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
Conv | # 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_1/inductor_cache/3r/c3runacu4pkgvdlmsngxxodg4pf6xmzvzxpf7xzbkjc3ay27rdj3.py
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4, 4), (64, 16, 4, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_2, 256, grid=grid(256), stream=stream0)
del primals_2
return (buf1, primals_1, primals_3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
def spectral_norm(module, mode=True):
if mode:
return nn.utils.spectral_norm(module)
return module
class Conv(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1,
padding=1, transpose=False, use_spectral_norm=False):
super(Conv, self).__init__()
self.out_channels = out_channels
if transpose:
self.conv = nn.ConvTranspose2d(in_channels, out_channels,
kernel_size=kernel_size, stride=stride, padding=padding,
bias=not use_spectral_norm)
else:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=
kernel_size, stride=stride, padding=padding, bias=not
use_spectral_norm)
if use_spectral_norm:
self.conv = spectral_norm(self.conv)
def forward(self, x):
return self.conv(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 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, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(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
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(256)](buf1, primals_2, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
return buf1, primals_1, primals_3
def spectral_norm(module, mode=True):
if mode:
return nn.utils.spectral_norm(module)
return module
class ConvNew(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1,
padding=1, transpose=False, use_spectral_norm=False):
super(ConvNew, self).__init__()
self.out_channels = out_channels
if transpose:
self.conv = nn.ConvTranspose2d(in_channels, out_channels,
kernel_size=kernel_size, stride=stride, padding=padding,
bias=not use_spectral_norm)
else:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=
kernel_size, stride=stride, padding=padding, bias=not
use_spectral_norm)
if use_spectral_norm:
self.conv = spectral_norm(self.conv)
def forward(self, input_0):
primals_1 = self.conv.weight
primals_2 = self.conv.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| DQiaole/ZITS | Conv | false | 7,946 | [
"Apache-2.0"
] | 40 | 5f7a060167790789d5e29a3d14d3c2ef8a34e765 | https://github.com/DQiaole/ZITS/tree/5f7a060167790789d5e29a3d14d3c2ef8a34e765 | import torch
import torch.nn as nn
def spectral_norm(module, mode=True):
if mode:
return nn.utils.spectral_norm(module)
return module
class Model(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1,
padding=1, transpose=False, use_spectral_norm=False):
super().__init__()
self.out_channels = out_channels
if transpose:
self.conv = nn.ConvTranspose2d(in_channels, out_channels,
kernel_size=kernel_size, stride=stride, padding=padding,
bias=not use_spectral_norm)
else:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=
kernel_size, stride=stride, padding=padding, bias=not
use_spectral_norm)
if use_spectral_norm:
self.conv = spectral_norm(self.conv)
def forward(self, x):
return self.conv(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
MaxLayer | # 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_1/inductor_cache/7u/c7ui4psgjsumcyq4ae6yfg3chazsbheap2vpg7rr7wd4jzw6akrc.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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
import torch.nn as nn
class MaxLayer(nn.Module):
def __init__(self):
super(MaxLayer, self).__init__()
def forward(self, a, b):
return torch.max(a, b)
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_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 MaxLayerNew(nn.Module):
def __init__(self):
super(MaxLayerNew, 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]
| DingXiaoH/Centripetal-SGD | MaxLayer | false | 7,947 | [
"Apache-2.0"
] | 35 | 992dd0fb31ee47a79cb0891f4f231707abd0c5c6 | https://github.com/DingXiaoH/Centripetal-SGD/tree/992dd0fb31ee47a79cb0891f4f231707abd0c5c6 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, a, b):
return torch.max(a, b)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
BinaryCrossEntropyLoss | # 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_1/inductor_cache/kw/ckwjemcedomvwlqylptnreenbwkgo4vjpkyxs7e2cj4t4ms2iohd.py
# Topologically Sorted Source Nodes: [binary_cross_entropy_with_logits], Original ATen: [aten.binary_cross_entropy_with_logits]
# Source node to ATen node mapping:
# binary_cross_entropy_with_logits => abs_1, exp, full_default, log1p, mean, minimum, mul, neg, sub, sub_1, sub_2
# 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 = {})
triton_per_fused_binary_cross_entropy_with_logits_0 = async_compile.triton('triton_per_fused_binary_cross_entropy_with_logits_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_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
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp17, 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_with_logits], Original ATen: [aten.binary_cross_entropy_with_logits]
stream0 = get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_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 BinaryCrossEntropyLoss(nn.Module):
def __init__(self, pos_weight=None, reduction='mean'):
super(BinaryCrossEntropyLoss, self).__init__()
self.BCE_loss = nn.BCEWithLogitsLoss(pos_weight=pos_weight,
reduction=reduction)
def forward(self, inputs, targets):
return self.BCE_loss(inputs, targets)
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_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
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp17, 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_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 BinaryCrossEntropyLossNew(nn.Module):
def __init__(self, pos_weight=None, reduction='mean'):
super(BinaryCrossEntropyLossNew, self).__init__()
self.BCE_loss = nn.BCEWithLogitsLoss(pos_weight=pos_weight,
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]
| DerekRay/2020-instanceSeg | BinaryCrossEntropyLoss | false | 7,948 | [
"MIT"
] | 25 | a08ad95e64726db53cc32a5f90aaa13ae3cdb6a3 | https://github.com/DerekRay/2020-instanceSeg/tree/a08ad95e64726db53cc32a5f90aaa13ae3cdb6a3 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, pos_weight=None, reduction='mean'):
super().__init__()
self.BCE_loss = nn.BCEWithLogitsLoss(pos_weight=pos_weight,
reduction=reduction)
def forward(self, inputs, targets):
return self.BCE_loss(inputs, targets)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GateConv | # 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_1/inductor_cache/bp/cbpjopd6zppixbmwr3t6xfwm3lgdzrlygvpd2itzq6lgthdc2ykj.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, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 8
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_1/inductor_cache/az/cazawsdhbkjp23y4nbl3u7u7ow7isqbyfetsclcygjvaf3ggiw3n.py
# Topologically Sorted Source Nodes: [sigmoid, mul], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# sigmoid => sigmoid
# Graph fragment:
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%getitem, %sigmoid), kwargs = {})
triton_poi_fused_mul_sigmoid_1 = async_compile.triton('triton_poi_fused_mul_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_sigmoid_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_sigmoid_1(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x1 = (xindex // 64)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (64 + x0 + (128*x1)), xmask)
tmp2 = tl.load(in_ptr0 + (x0 + (128*x1)), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp2 * tmp1
tl.store(out_ptr0 + (x2), tmp1, xmask)
tl.store(out_ptr1 + (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, (8, 4, 3, 3), (36, 9, 3, 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)
# Topologically Sorted Source Nodes: [x], 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, 8, 4, 4), (128, 16, 4, 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, 512, grid=grid(512), stream=stream0)
del primals_2
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)
# Topologically Sorted Source Nodes: [sigmoid, mul], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_1.run(buf1, buf2, buf3, 256, grid=grid(256), stream=stream0)
return (buf3, primals_1, primals_3, reinterpret_tensor(buf1, (4, 4, 4, 4), (128, 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
primals_1 = rand_strided((8, 4, 3, 3), (36, 9, 3, 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 GateConv(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1,
padding=1, transpose=False):
super(GateConv, self).__init__()
self.out_channels = out_channels
if transpose:
self.gate_conv = nn.ConvTranspose2d(in_channels, out_channels *
2, kernel_size=kernel_size, stride=stride, padding=padding)
else:
self.gate_conv = nn.Conv2d(in_channels, out_channels * 2,
kernel_size=kernel_size, stride=stride, padding=padding)
def forward(self, x):
x = self.gate_conv(x)
x, g = torch.split(x, self.out_channels, dim=1)
return x * torch.sigmoid(g)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 8
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_mul_sigmoid_1(in_ptr0, out_ptr0, out_ptr1, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x1 = xindex // 64
x2 = xindex
tmp0 = tl.load(in_ptr0 + (64 + x0 + 128 * x1), xmask)
tmp2 = tl.load(in_ptr0 + (x0 + 128 * x1), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp2 * tmp1
tl.store(out_ptr0 + x2, tmp1, xmask)
tl.store(out_ptr1 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (8, 4, 3, 3), (36, 9, 3, 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 = 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, 8, 4, 4), (128, 16, 4, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(512)](buf1, primals_2, 512,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
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)
triton_poi_fused_mul_sigmoid_1[grid(256)](buf1, buf2, buf3, 256,
XBLOCK=256, num_warps=4, num_stages=1)
return buf3, primals_1, primals_3, reinterpret_tensor(buf1, (4, 4, 4, 4
), (128, 16, 4, 1), 0), buf2
class GateConvNew(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1,
padding=1, transpose=False):
super(GateConvNew, self).__init__()
self.out_channels = out_channels
if transpose:
self.gate_conv = nn.ConvTranspose2d(in_channels, out_channels *
2, kernel_size=kernel_size, stride=stride, padding=padding)
else:
self.gate_conv = nn.Conv2d(in_channels, out_channels * 2,
kernel_size=kernel_size, stride=stride, padding=padding)
def forward(self, input_0):
primals_1 = self.gate_conv.weight
primals_2 = self.gate_conv.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| DQiaole/ZITS | GateConv | false | 7,949 | [
"Apache-2.0"
] | 40 | 5f7a060167790789d5e29a3d14d3c2ef8a34e765 | https://github.com/DQiaole/ZITS/tree/5f7a060167790789d5e29a3d14d3c2ef8a34e765 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1,
padding=1, transpose=False):
super().__init__()
self.out_channels = out_channels
if transpose:
self.gate_conv = nn.ConvTranspose2d(in_channels, out_channels *
2, kernel_size=kernel_size, stride=stride, padding=padding)
else:
self.gate_conv = nn.Conv2d(in_channels, out_channels * 2,
kernel_size=kernel_size, stride=stride, padding=padding)
def forward(self, x):
x = self.gate_conv(x)
x, g = torch.split(x, self.out_channels, dim=1)
return x * torch.sigmoid(g)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
GELU | # 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_1/inductor_cache/aj/caj5tg5luyk7zbg5qjfjybkw7uy53iqt5pkour3hfbw7ahus5poa.py
# Topologically Sorted Source Nodes: [mul, truediv, erf, add, mul_1], Original ATen: [aten.mul, aten.div, aten.erf, aten.add]
# Source node to ATen node mapping:
# add => add
# erf => erf
# mul => mul
# mul_1 => mul_1
# truediv => div
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 0.5), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, 1.4142135623730951), kwargs = {})
# %erf : [num_users=1] = call_function[target=torch.ops.aten.erf.default](args = (%div,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%erf, 1.0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add), kwargs = {})
triton_poi_fused_add_div_erf_mul_0 = async_compile.triton('triton_poi_fused_add_div_erf_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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 = 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')
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, truediv, erf, add, mul_1], Original ATen: [aten.mul, aten.div, aten.erf, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_erf_mul_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
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)))
class GELU(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return gelu(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
import math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_div_erf_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.5
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)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_div_erf_mul_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
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)))
class GELUNew(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DQiaole/ZITS | GELU | false | 7,950 | [
"Apache-2.0"
] | 40 | 5f7a060167790789d5e29a3d14d3c2ef8a34e765 | https://github.com/DQiaole/ZITS/tree/5f7a060167790789d5e29a3d14d3c2ef8a34e765 | import math
import torch
import torch.nn as nn
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)))
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return gelu(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ScaledDotProductAttentionMemory | # 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_1/inductor_cache/f2/cf2u4d3unsat27x3zknrgw2nin4lsudw2qnsxjvfhacio5k5uugt.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_2,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1) + (64*x3)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x4), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ls/cls2rxeezdp5ao3stkexqn3nvyql3kgy7logqhpln7azlk2d2urq.py
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_3,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_1 = async_compile.triton('triton_poi_fused_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8) % 16
x2 = (xindex // 128)
x3 = 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 + (x1 + (16*x0) + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x1 + (16*((-4) + x0))), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = 2.0
tmp11 = tmp10 * tmp9
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp6, tmp11, tmp12)
tmp14 = tl.where(tmp4, tmp5, tmp13)
tl.store(out_ptr0 + (x3), tmp14, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/6s/c6sskpghchozk4czzjlhfungxmdggnn4gokovushs67vb2vqu5ky.py
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
# Source node to ATen node mapping:
# att_1 => div_1, exp, sum_1
# wrapped_sqrt => full_default
# Graph fragment:
# %full_default : [num_users=4] = call_function[target=torch.ops.aten.full.default](args = ([], 2.0), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %scalar_tensor_default : [num_users=2] = call_function[target=torch.ops.aten.scalar_tensor.default](args = (1,), kwargs = {dtype: torch.float32, device: cuda:0, pin_memory: False})
# %ge_scalar : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%full_default, 0), kwargs = {})
# %neg_default : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%scalar_tensor_default,), kwargs = {})
# %where_self : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ge_scalar, %scalar_tensor_default, %neg_default), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_11, %where_self), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [-1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where_self, %full_default), kwargs = {})
# %div_tensor : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_tensor, %mul_tensor_1), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_tensor,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_per_fused__softmax_sqrt_2 = async_compile.triton('triton_per_fused__softmax_sqrt_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[64, 8],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_sqrt_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 2, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_sqrt_2(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 64
rnumel = 8
RBLOCK: tl.constexpr = 8
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 + (8*x0)), xmask, other=0.0)
tmp1 = tl.full([1, 1], 2.0, tl.float64)
tmp2 = tl.full([1, 1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, float("-inf"))
tmp11 = triton_helpers.max2(tmp10, 1)[:, None]
tmp12 = tmp7 - tmp11
tmp13 = tmp6.to(tl.float64)
tmp14 = tmp13 * tmp1
tmp15 = tmp14.to(tl.float32)
tmp16 = tmp12 / tmp15
tmp17 = tl_math.exp(tmp16)
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp20 = tl.where(xmask, tmp18, 0)
tmp21 = tl.sum(tmp20, 1)[:, None]
tmp22 = tmp17 / tmp21
tl.store(out_ptr2 + (r1 + (8*x0)), tmp22, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/2p/c2pdjkvlplag67u7k6g3qkmaexh45o6ukwcwi4zycxldvypfsvjv.py
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# matmul_1 => clone_2
# Graph fragment:
# %clone_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_5,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_3 = async_compile.triton('triton_poi_fused_clone_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4) % 8
x0 = xindex % 4
x2 = (xindex // 32) % 4
x3 = (xindex // 128)
x4 = 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 + (4*x2) + (16*x1) + (64*x3)), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x0 + (4*x2) + (16*((-4) + x1))), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = 2.0
tmp11 = tmp10 * tmp9
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp6, tmp11, tmp12)
tmp14 = tl.where(tmp4, tmp5, tmp13)
tl.store(out_ptr0 + (x4), tmp14, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/dj/cdjffwvn2idikuml44jvdav3vzvow2i4vqtqbymwksc7y55376j6.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_6,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1) + (64*x3)), xmask)
tl.store(out_ptr0 + (x4), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13 = 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, (1, 4, 16), (64, 16, 1))
assert_size_stride(primals_4, (1, 4, 16), (64, 16, 1))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16, ), (1, ))
assert_size_stride(primals_7, (16, 4), (4, 1))
assert_size_stride(primals_8, (16, ), (1, ))
assert_size_stride(primals_9, (16, 4), (4, 1))
assert_size_stride(primals_10, (16, ), (1, ))
assert_size_stride(primals_11, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_12, (4, 16), (16, 1))
assert_size_stride(primals_13, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf0)
del primals_5
buf1 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 16), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_7
del primals_8
buf2 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(primals_11, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 16), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_10
del primals_9
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, primals_6, buf3, 256, grid=grid(256), stream=stream0)
del buf0
del primals_6
buf4 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf1, primals_3, buf4, 512, grid=grid(512), stream=stream0)
del primals_3
buf5 = empty_strided_cuda((16, 4, 8), (32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf4, (16, 4, 8), (32, 8, 1), 0), out=buf5)
buf8 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [wrapped_sqrt, att_1], Original ATen: [aten.sqrt, aten._softmax]
triton_per_fused__softmax_sqrt_2.run(buf5, buf8, 64, 8, grid=grid(64), stream=stream0)
buf9 = reinterpret_tensor(buf5, (4, 4, 8, 4), (128, 32, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf2, primals_4, buf9, 512, grid=grid(512), stream=stream0)
del primals_4
buf10 = reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [matmul_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf8, (16, 4, 8), (32, 8, 1), 0), reinterpret_tensor(buf9, (16, 8, 4), (32, 4, 1), 0), out=buf10)
buf11 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf10, buf11, 256, grid=grid(256), stream=stream0)
del buf10
buf12 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_13, reinterpret_tensor(buf11, (16, 16), (16, 1), 0), reinterpret_tensor(primals_12, (16, 4), (1, 16), 0), alpha=1, beta=1, out=buf12)
del primals_13
return (reinterpret_tensor(buf12, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (16, 4), (4, 1), 0), buf8, reinterpret_tensor(buf11, (16, 16), (16, 1), 0), primals_12, reinterpret_tensor(buf9, (16, 4, 8), (32, 1, 4), 0), reinterpret_tensor(buf3, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf4, (16, 8, 4), (32, 1, 8), 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((1, 4, 16), (64, 16, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 4, 16), (64, 16, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
import torch.nn as nn
class ScaledDotProductAttentionMemory(nn.Module):
"""
Scaled dot-product attention with memory
"""
def __init__(self, d_model, d_k, d_v, h, m):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
:param m: Number of memory slots
"""
super(ScaledDotProductAttentionMemory, self).__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.m_k = nn.Parameter(torch.FloatTensor(1, m, h * d_k))
self.m_v = nn.Parameter(torch.FloatTensor(1, m, h * d_v))
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.m = m
self.init_weights()
def init_weights(self):
nn.init.xavier_uniform_(self.fc_q.weight)
nn.init.xavier_uniform_(self.fc_k.weight)
nn.init.xavier_uniform_(self.fc_v.weight)
nn.init.xavier_uniform_(self.fc_o.weight)
nn.init.normal_(self.m_k, 0, 1 / self.d_k)
nn.init.normal_(self.m_v, 0, 1 / self.m)
nn.init.constant_(self.fc_q.bias, 0)
nn.init.constant_(self.fc_k.bias, 0)
nn.init.constant_(self.fc_v.bias, 0)
nn.init.constant_(self.fc_o.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
m_k = np.sqrt(self.d_k) * self.m_k.expand(b_s, self.m, self.h *
self.d_k)
m_v = np.sqrt(self.m) * self.m_v.expand(b_s, self.m, self.h * self.d_v)
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = torch.cat([self.fc_k(keys), m_k], 1).view(b_s, nk + self.m,
self.h, self.d_k).permute(0, 2, 3, 1)
v = torch.cat([self.fc_v(values), m_v], 1).view(b_s, nk + self.m,
self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = torch.cat([att[:, :, :, :nk] * attention_weights, att[:,
:, :, nk:]], -1)
if attention_mask is not None:
att[:, :, :, :nk] = att[:, :, :, :nk].masked_fill(attention_mask,
-np.inf)
att = torch.softmax(att, -1)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [[], {'d_model': 4, 'd_k': 4, 'd_v': 4, 'h': 4, 'm': 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_clone_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1 + 64 * x3), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x4, tmp2, xmask)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = xindex // 8 % 16
x2 = xindex // 128
x3 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x1 + 16 * x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (x1 + 16 * (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = 2.0
tmp11 = tmp10 * tmp9
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp6, tmp11, tmp12)
tmp14 = tl.where(tmp4, tmp5, tmp13)
tl.store(out_ptr0 + x3, tmp14, xmask)
@triton.jit
def triton_per_fused__softmax_sqrt_2(in_ptr0, out_ptr2, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 64
RBLOCK: tl.constexpr = 8
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 + 8 * x0), xmask, other=0.0)
tmp1 = tl.full([1, 1], 2.0, tl.float64)
tmp2 = tl.full([1, 1], 0.0, tl.float64)
tmp3 = tmp1 >= tmp2
tmp4 = 1.0
tmp5 = -1.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = tl.broadcast_to(tmp7, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, float('-inf'))
tmp11 = triton_helpers.max2(tmp10, 1)[:, None]
tmp12 = tmp7 - tmp11
tmp13 = tmp6.to(tl.float64)
tmp14 = tmp13 * tmp1
tmp15 = tmp14.to(tl.float32)
tmp16 = tmp12 / tmp15
tmp17 = tl_math.exp(tmp16)
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp20 = tl.where(xmask, tmp18, 0)
tmp21 = tl.sum(tmp20, 1)[:, None]
tmp22 = tmp17 / tmp21
tl.store(out_ptr2 + (r1 + 8 * x0), tmp22, xmask)
@triton.jit
def triton_poi_fused_clone_3(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 // 4 % 8
x0 = xindex % 4
x2 = xindex // 32 % 4
x3 = xindex // 128
x4 = 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 + 4 * x2 + 16 * x1 + 64 * x3), tmp4 &
xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (x0 + 4 * x2 + 16 * (-4 + x1)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = 2.0
tmp11 = tmp10 * tmp9
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp6, tmp11, tmp12)
tmp14 = tl.where(tmp4, tmp5, tmp13)
tl.store(out_ptr0 + x4, tmp14, xmask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1 + 64 * x3), xmask)
tl.store(out_ptr0 + x4, tmp0, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13) = 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, (1, 4, 16), (64, 16, 1))
assert_size_stride(primals_4, (1, 4, 16), (64, 16, 1))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16,), (1,))
assert_size_stride(primals_7, (16, 4), (4, 1))
assert_size_stride(primals_8, (16,), (1,))
assert_size_stride(primals_9, (16, 4), (4, 1))
assert_size_stride(primals_10, (16,), (1,))
assert_size_stride(primals_11, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_12, (4, 16), (16, 1))
assert_size_stride(primals_13, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf0)
del primals_5
buf1 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_2, (16,
4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 16), (1, 4),
0), alpha=1, beta=1, out=buf1)
del primals_7
del primals_8
buf2 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.addmm(primals_10, reinterpret_tensor(primals_11, (16,
4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 16), (1, 4),
0), alpha=1, beta=1, out=buf2)
del primals_10
del primals_9
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(256)](buf0, primals_6, buf3, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del buf0
del primals_6
buf4 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.float32)
triton_poi_fused_clone_1[grid(512)](buf1, primals_3, buf4, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_3
buf5 = empty_strided_cuda((16, 4, 8), (32, 8, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf4, (16, 4, 8), (32, 8, 1), 0), out=buf5)
buf8 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.float32)
triton_per_fused__softmax_sqrt_2[grid(64)](buf5, buf8, 64, 8,
XBLOCK=32, num_warps=2, num_stages=1)
buf9 = reinterpret_tensor(buf5, (4, 4, 8, 4), (128, 32, 4, 1), 0)
del buf5
triton_poi_fused_clone_3[grid(512)](buf2, primals_4, buf9, 512,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_4
buf10 = reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0)
del buf2
extern_kernels.bmm(reinterpret_tensor(buf8, (16, 4, 8), (32, 8, 1),
0), reinterpret_tensor(buf9, (16, 8, 4), (32, 4, 1), 0), out=buf10)
buf11 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf1
triton_poi_fused_clone_4[grid(256)](buf10, buf11, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf10
buf12 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_13, reinterpret_tensor(buf11, (16, 16),
(16, 1), 0), reinterpret_tensor(primals_12, (16, 4), (1, 16), 0
), alpha=1, beta=1, out=buf12)
del primals_13
return reinterpret_tensor(buf12, (4, 4, 4), (16, 4, 1), 0
), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_11, (16, 4), (4, 1), 0
), buf8, reinterpret_tensor(buf11, (16, 16), (16, 1), 0
), primals_12, reinterpret_tensor(buf9, (16, 4, 8), (32, 1, 4), 0
), reinterpret_tensor(buf3, (16, 4, 4), (16, 1, 4), 0
), reinterpret_tensor(buf4, (16, 8, 4), (32, 1, 8), 0)
class ScaledDotProductAttentionMemoryNew(nn.Module):
"""
Scaled dot-product attention with memory
"""
def __init__(self, d_model, d_k, d_v, h, m):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
:param m: Number of memory slots
"""
super(ScaledDotProductAttentionMemoryNew, self).__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.m_k = nn.Parameter(torch.FloatTensor(1, m, h * d_k))
self.m_v = nn.Parameter(torch.FloatTensor(1, m, h * d_v))
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.m = m
self.init_weights()
def init_weights(self):
nn.init.xavier_uniform_(self.fc_q.weight)
nn.init.xavier_uniform_(self.fc_k.weight)
nn.init.xavier_uniform_(self.fc_v.weight)
nn.init.xavier_uniform_(self.fc_o.weight)
nn.init.normal_(self.m_k, 0, 1 / self.d_k)
nn.init.normal_(self.m_v, 0, 1 / self.m)
nn.init.constant_(self.fc_q.bias, 0)
nn.init.constant_(self.fc_k.bias, 0)
nn.init.constant_(self.fc_v.bias, 0)
nn.init.constant_(self.fc_o.bias, 0)
def forward(self, input_0, input_1, input_2):
primals_3 = self.m_k
primals_4 = self.m_v
primals_5 = self.fc_q.weight
primals_6 = self.fc_q.bias
primals_7 = self.fc_k.weight
primals_8 = self.fc_k.bias
primals_9 = self.fc_v.weight
primals_10 = self.fc_v.bias
primals_12 = self.fc_o.weight
primals_13 = self.fc_o.bias
primals_1 = input_0
primals_2 = input_1
primals_11 = 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, primals_12, primals_13])
return output[0]
| CurryYuan/X-Trans2Cap | ScaledDotProductAttentionMemory | false | 7,951 | [
"Apache-2.0"
] | 11 | c78a27209f14fcbbec74fe8b5edc06faea2e7d44 | https://github.com/CurryYuan/X-Trans2Cap/tree/c78a27209f14fcbbec74fe8b5edc06faea2e7d44 | import torch
import numpy as np
import torch.nn as nn
class Model(nn.Module):
"""
Scaled dot-product attention with memory
"""
def __init__(self, d_model, d_k, d_v, h, m):
"""
:param d_model: Output dimensionality of the model
:param d_k: Dimensionality of queries and keys
:param d_v: Dimensionality of values
:param h: Number of heads
:param m: Number of memory slots
"""
super().__init__()
self.fc_q = nn.Linear(d_model, h * d_k)
self.fc_k = nn.Linear(d_model, h * d_k)
self.fc_v = nn.Linear(d_model, h * d_v)
self.fc_o = nn.Linear(h * d_v, d_model)
self.m_k = nn.Parameter(torch.FloatTensor(1, m, h * d_k))
self.m_v = nn.Parameter(torch.FloatTensor(1, m, h * d_v))
self.d_model = d_model
self.d_k = d_k
self.d_v = d_v
self.h = h
self.m = m
self.init_weights()
def init_weights(self):
nn.init.xavier_uniform_(self.fc_q.weight)
nn.init.xavier_uniform_(self.fc_k.weight)
nn.init.xavier_uniform_(self.fc_v.weight)
nn.init.xavier_uniform_(self.fc_o.weight)
nn.init.normal_(self.m_k, 0, 1 / self.d_k)
nn.init.normal_(self.m_v, 0, 1 / self.m)
nn.init.constant_(self.fc_q.bias, 0)
nn.init.constant_(self.fc_k.bias, 0)
nn.init.constant_(self.fc_v.bias, 0)
nn.init.constant_(self.fc_o.bias, 0)
def forward(self, queries, keys, values, attention_mask=None,
attention_weights=None):
"""
Computes
:param queries: Queries (b_s, nq, d_model)
:param keys: Keys (b_s, nk, d_model)
:param values: Values (b_s, nk, d_model)
:param attention_mask: Mask over attention values (b_s, h, nq, nk). True indicates masking.
:param attention_weights: Multiplicative weights for attention values (b_s, h, nq, nk).
:return:
"""
b_s, nq = queries.shape[:2]
nk = keys.shape[1]
m_k = np.sqrt(self.d_k) * self.m_k.expand(b_s, self.m, self.h *
self.d_k)
m_v = np.sqrt(self.m) * self.m_v.expand(b_s, self.m, self.h * self.d_v)
q = self.fc_q(queries).view(b_s, nq, self.h, self.d_k).permute(0, 2,
1, 3)
k = torch.cat([self.fc_k(keys), m_k], 1).view(b_s, nk + self.m,
self.h, self.d_k).permute(0, 2, 3, 1)
v = torch.cat([self.fc_v(values), m_v], 1).view(b_s, nk + self.m,
self.h, self.d_v).permute(0, 2, 1, 3)
att = torch.matmul(q, k) / np.sqrt(self.d_k)
if attention_weights is not None:
att = torch.cat([att[:, :, :, :nk] * attention_weights, att[:,
:, :, nk:]], -1)
if attention_mask is not None:
att[:, :, :, :nk] = att[:, :, :, :nk].masked_fill(attention_mask,
-np.inf)
att = torch.softmax(att, -1)
out = torch.matmul(att, v).permute(0, 2, 1, 3).contiguous().view(b_s,
nq, self.h * self.d_v)
out = self.fc_o(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [4, 4, 4, 4, 4]
|
BinaryLoss | # 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_1/inductor_cache/jk/cjksfpx3iwyilcfx4dpxbs6ltmwazirnk4kq6ec5mqttaqmxvhog.py
# Topologically Sorted Source Nodes: [sub, abs_1, lt, less_than_theta, pow_1, mul, mul_1, sub_1, sub_2, add, mul_2, loss, loss_1, loss_2], Original ATen: [aten.sub, aten.abs, aten.lt, aten._to_copy, aten.pow, aten.mul, aten.rsub, aten.add, aten.sum, aten.mean]
# Source node to ATen node mapping:
# abs_1 => abs_1
# add => add
# less_than_theta => convert_element_type
# loss => add_1
# loss_1 => sum_1
# loss_2 => mean
# lt => lt
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# pow_1 => pow_1
# sub => sub
# sub_1 => sub_1
# sub_2 => sub_2
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%slice_6, %slice_2), kwargs = {})
# %abs_1 : [num_users=3] = call_function[target=torch.ops.aten.abs.default](args = (%sub,), kwargs = {})
# %lt : [num_users=1] = call_function[target=torch.ops.aten.lt.Scalar](args = (%abs_1, 0.1), kwargs = {})
# %convert_element_type : [num_users=2] = call_function[target=torch.ops.prims.convert_element_type.default](args = (%lt, torch.float32), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%abs_1, 2), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convert_element_type, %pow_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, 5.0), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %convert_element_type), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%abs_1, 0.1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_2, 0.05000000000000001), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %add), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %mul_2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%add_1, [1], True), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sum_1,), kwargs = {})
triton_per_fused__to_copy_abs_add_lt_mean_mul_pow_rsub_sub_sum_0 = async_compile.triton('triton_per_fused__to_copy_abs_add_lt_mean_mul_pow_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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__to_copy_abs_add_lt_mean_mul_pow_rsub_sub_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_abs_add_lt_mean_mul_pow_rsub_sub_sum_0(in_out_ptr0, 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 % 16
r1 = (rindex // 16)
tmp0 = tl.load(in_ptr0 + (r0 + (64*r1)), None)
tmp1 = tl.load(in_ptr1 + (r0 + (64*r1)), None)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.abs(tmp2)
tmp4 = 0.1
tmp5 = tmp3 < tmp4
tmp6 = tmp5.to(tl.float32)
tmp7 = tmp3 * tmp3
tmp8 = tmp6 * tmp7
tmp9 = 5.0
tmp10 = tmp8 * tmp9
tmp11 = 1.0
tmp12 = tmp11 - tmp6
tmp13 = tmp3 - tmp4
tmp14 = 0.05000000000000001
tmp15 = tmp13 + tmp14
tmp16 = tmp12 * tmp15
tmp17 = tmp10 + tmp16
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp20 = tl.sum(tmp18, 1)[:, None]
tmp21 = 64.0
tmp22 = tmp20 / tmp21
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp22, 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, lt, less_than_theta, pow_1, mul, mul_1, sub_1, sub_2, add, mul_2, loss, loss_1, loss_2], Original ATen: [aten.sub, aten.abs, aten.lt, aten._to_copy, aten.pow, aten.mul, aten.rsub, aten.add, aten.sum, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused__to_copy_abs_add_lt_mean_mul_pow_rsub_sub_sum_0.run(buf1, arg1_1, arg0_1, 1, 64, 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
import torch.nn.functional as F
def adjust_smooth_l1_loss(y_pred, theta=0.1):
less_grad_factor = 1.0 / (2 * theta)
less_loss_bias = less_grad_factor * theta ** 2
less_than_theta = (y_pred < theta).float()
loss = less_than_theta * y_pred ** 2 * less_grad_factor + (1 -
less_than_theta) * (y_pred - theta + less_loss_bias)
return loss
class CrossEntropyLoss(nn.Module):
def __init__(self, weight=None, ignore_index=255, reduction='mean'):
""" NLLLoss: negative log likelihood loss.
# nll_loss: weights: None | a tensor of size C
pred in [N, C, d1, d2, ..., dk]
target in [N, d1, d2, ..., dk]
output in [N, d1, d2, ..., dk]
"""
super(CrossEntropyLoss, self).__init__()
self.nll_loss = nn.NLLLoss(weight, ignore_index=ignore_index,
reduction=reduction)
def forward(self, inputs, targets):
return self.nll_loss(F.log_softmax(inputs, dim=1), targets)
class BinaryLoss(nn.Module):
"""This class computes the Binary loss to force BG pixels close to 0 and FG pixels far away.
"""
def __init__(self, margin=2.0, FG_stCH=1, loss_type='l1', weights=None):
"""
margin: minimum distance between FG/BG if prediction with 1 channel
FG_stCH: start channel of FG objects on prediction with multiple channels
loss_type: 'l1' | 'CE', works for prediction with multiple channel.
if 'l1', prediction is expected to be softmax2d output.
if 'CE', prediction is expected to be net logits
if prediction has channel=1,
weights: if not None, a tensor of size C
"""
super(BinaryLoss, self).__init__()
self.margin = margin
self.FG_stCH = FG_stCH
self.loss_type = loss_type if FG_stCH > 1 else 'l1'
if self.loss_type == 'CE':
self.CE_loss = CrossEntropyLoss(weight=weights, reduction='none')
def forward(self, preds, targets, weights=None):
"""
Params:
preds/targets: [bs, ch, ht, wd]
weights:[bs, 1, ht, wd]
"""
_bs, ch, _ht, _wd = preds.size()
if ch > 1:
if self.loss_type == 'l1':
preds_0 = preds[:, :self.FG_stCH, :, :]
targets_0 = targets[:, :self.FG_stCH, :, :].float()
loss = adjust_smooth_l1_loss(torch.abs(targets_0 - preds_0))
loss = loss.sum(axis=1, keepdim=True)
else:
preds_0 = preds[:, :self.FG_stCH, :, :].float()
targets_0 = targets[:, :self.FG_stCH, :, :]
preds_1, _ = preds[:, self.FG_stCH:, :, :].float().max(axis
=1, keepdim=True)
targets_1 = targets[:, self.FG_stCH:, :, :].sum(axis=1,
keepdim=True).int()
_, target_id = torch.cat([targets_0, targets_1], axis=1).max(
axis=1)
loss = self.CE_loss(torch.cat((preds_0, preds_1), axis=1),
target_id)
loss = loss[:, None, :, :]
else:
isFG = (targets > 0.5).float()
loss_0 = adjust_smooth_l1_loss(F.relu(preds))
loss_1 = adjust_smooth_l1_loss(F.relu(self.margin - preds))
loss = loss_0 * (1.0 - isFG) + loss_1 * isFG
if weights is not None:
loss = torch.mul(loss, weights).sum() / (weights.sum() + 0.0001)
else:
loss = loss.mean()
return loss.float()
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 math as tl_math
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_per_fused__to_copy_abs_add_lt_mean_mul_pow_rsub_sub_sum_0(
in_out_ptr0, 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 % 16
r1 = rindex // 16
tmp0 = tl.load(in_ptr0 + (r0 + 64 * r1), None)
tmp1 = tl.load(in_ptr1 + (r0 + 64 * r1), None)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.abs(tmp2)
tmp4 = 0.1
tmp5 = tmp3 < tmp4
tmp6 = tmp5.to(tl.float32)
tmp7 = tmp3 * tmp3
tmp8 = tmp6 * tmp7
tmp9 = 5.0
tmp10 = tmp8 * tmp9
tmp11 = 1.0
tmp12 = tmp11 - tmp6
tmp13 = tmp3 - tmp4
tmp14 = 0.05000000000000001
tmp15 = tmp13 + tmp14
tmp16 = tmp12 * tmp15
tmp17 = tmp10 + tmp16
tmp18 = tl.broadcast_to(tmp17, [XBLOCK, RBLOCK])
tmp20 = tl.sum(tmp18, 1)[:, None]
tmp21 = 64.0
tmp22 = tmp20 / tmp21
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp22, 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__to_copy_abs_add_lt_mean_mul_pow_rsub_sub_sum_0[grid
(1)](buf1, arg1_1, arg0_1, 1, 64, XBLOCK=1, num_warps=2,
num_stages=1)
del arg0_1
del arg1_1
return buf1,
def adjust_smooth_l1_loss(y_pred, theta=0.1):
less_grad_factor = 1.0 / (2 * theta)
less_loss_bias = less_grad_factor * theta ** 2
less_than_theta = (y_pred < theta).float()
loss = less_than_theta * y_pred ** 2 * less_grad_factor + (1 -
less_than_theta) * (y_pred - theta + less_loss_bias)
return loss
class CrossEntropyLoss(nn.Module):
def __init__(self, weight=None, ignore_index=255, reduction='mean'):
""" NLLLoss: negative log likelihood loss.
# nll_loss: weights: None | a tensor of size C
pred in [N, C, d1, d2, ..., dk]
target in [N, d1, d2, ..., dk]
output in [N, d1, d2, ..., dk]
"""
super(CrossEntropyLoss, self).__init__()
self.nll_loss = nn.NLLLoss(weight, ignore_index=ignore_index,
reduction=reduction)
def forward(self, inputs, targets):
return self.nll_loss(F.log_softmax(inputs, dim=1), targets)
class BinaryLossNew(nn.Module):
"""This class computes the Binary loss to force BG pixels close to 0 and FG pixels far away.
"""
def __init__(self, margin=2.0, FG_stCH=1, loss_type='l1', weights=None):
"""
margin: minimum distance between FG/BG if prediction with 1 channel
FG_stCH: start channel of FG objects on prediction with multiple channels
loss_type: 'l1' | 'CE', works for prediction with multiple channel.
if 'l1', prediction is expected to be softmax2d output.
if 'CE', prediction is expected to be net logits
if prediction has channel=1,
weights: if not None, a tensor of size C
"""
super(BinaryLossNew, self).__init__()
self.margin = margin
self.FG_stCH = FG_stCH
self.loss_type = loss_type if FG_stCH > 1 else 'l1'
if self.loss_type == 'CE':
self.CE_loss = CrossEntropyLoss(weight=weights, 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]
| DerekRay/2020-instanceSeg | BinaryLoss | false | 7,952 | [
"MIT"
] | 25 | a08ad95e64726db53cc32a5f90aaa13ae3cdb6a3 | https://github.com/DerekRay/2020-instanceSeg/tree/a08ad95e64726db53cc32a5f90aaa13ae3cdb6a3 | import torch
import torch.nn as nn
import torch.nn.functional as F
def adjust_smooth_l1_loss(y_pred, theta=0.1):
less_grad_factor = 1.0 / (2 * theta)
less_loss_bias = less_grad_factor * theta ** 2
less_than_theta = (y_pred < theta).float()
loss = less_than_theta * y_pred ** 2 * less_grad_factor + (1 -
less_than_theta) * (y_pred - theta + less_loss_bias)
return loss
class CrossEntropyLoss(nn.Module):
def __init__(self, weight=None, ignore_index=255, reduction='mean'):
""" NLLLoss: negative log likelihood loss.
# nll_loss: weights: None | a tensor of size C
pred in [N, C, d1, d2, ..., dk]
target in [N, d1, d2, ..., dk]
output in [N, d1, d2, ..., dk]
"""
super().__init__()
self.nll_loss = nn.NLLLoss(weight, ignore_index=ignore_index,
reduction=reduction)
def forward(self, inputs, targets):
return self.nll_loss(F.log_softmax(inputs, dim=1), targets)
class Model(nn.Module):
"""This class computes the Binary loss to force BG pixels close to 0 and FG pixels far away.
"""
def __init__(self, margin=2.0, FG_stCH=1, loss_type='l1', weights=None):
"""
margin: minimum distance between FG/BG if prediction with 1 channel
FG_stCH: start channel of FG objects on prediction with multiple channels
loss_type: 'l1' | 'CE', works for prediction with multiple channel.
if 'l1', prediction is expected to be softmax2d output.
if 'CE', prediction is expected to be net logits
if prediction has channel=1,
weights: if not None, a tensor of size C
"""
super().__init__()
self.margin = margin
self.FG_stCH = FG_stCH
self.loss_type = loss_type if FG_stCH > 1 else 'l1'
if self.loss_type == 'CE':
self.CE_loss = CrossEntropyLoss(weight=weights, reduction='none')
def forward(self, preds, targets, weights=None):
"""
Params:
preds/targets: [bs, ch, ht, wd]
weights:[bs, 1, ht, wd]
"""
_bs, ch, _ht, _wd = preds.size()
if ch > 1:
if self.loss_type == 'l1':
preds_0 = preds[:, :self.FG_stCH, :, :]
targets_0 = targets[:, :self.FG_stCH, :, :].float()
loss = adjust_smooth_l1_loss(torch.abs(targets_0 - preds_0))
loss = loss.sum(axis=1, keepdim=True)
else:
preds_0 = preds[:, :self.FG_stCH, :, :].float()
targets_0 = targets[:, :self.FG_stCH, :, :]
preds_1, _ = preds[:, self.FG_stCH:, :, :].float().max(axis
=1, keepdim=True)
targets_1 = targets[:, self.FG_stCH:, :, :].sum(axis=1,
keepdim=True).int()
_, target_id = torch.cat([targets_0, targets_1], axis=1).max(
axis=1)
loss = self.CE_loss(torch.cat((preds_0, preds_1), axis=1),
target_id)
loss = loss[:, None, :, :]
else:
isFG = (targets > 0.5).float()
loss_0 = adjust_smooth_l1_loss(F.relu(preds))
loss_1 = adjust_smooth_l1_loss(F.relu(self.margin - preds))
loss = loss_0 * (1.0 - isFG) + loss_1 * isFG
if weights is not None:
loss = torch.mul(loss, weights).sum() / (weights.sum() + 0.0001)
else:
loss = loss.mean()
return loss.float()
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
BCEDiceLoss | # 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_1/inductor_cache/j7/cj7zq6t3fadbf5cj73xvwpauitv3wlnszfdqyykqgzgicr7x6zb2.py
# Topologically Sorted Source Nodes: [bce], Original ATen: [aten.binary_cross_entropy_with_logits]
# Source node to ATen node mapping:
# bce => abs_1, exp, full_default, log1p, mean, minimum, mul, neg, sub, sub_1, sub_2
# 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 = {})
triton_per_fused_binary_cross_entropy_with_logits_0 = async_compile.triton('triton_per_fused_binary_cross_entropy_with_logits_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_0', 'mutated_arg_names': [], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_0(in_ptr0, in_ptr1, out_ptr0, 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))
tl.store(out_ptr0 + (tl.full([1], 0, tl.int32)), tmp15, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qn/cqnkpvlcoqle63qrpkimappm6tpho6wyebadqrt5ogzeqkjkdoxq.py
# Topologically Sorted Source Nodes: [intersection, sum_1, sum_2, sum_3], Original ATen: [aten.mul, aten.sum]
# Source node to ATen node mapping:
# intersection => mul_1
# sum_1 => sum_1
# sum_2 => sum_2
# sum_3 => sum_3
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %view_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [1]), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%view, [1]), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%view_1, [1]), kwargs = {})
triton_per_fused_mul_sum_1 = async_compile.triton('triton_per_fused_mul_sum_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 64],
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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mul_sum_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 3, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mul_sum_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, 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)
tmp2 = tl.load(in_ptr1 + (r1 + (64*x0)), xmask, other=0.0)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp14 = tl.where(xmask, tmp12, 0)
tmp15 = tl.sum(tmp14, 1)[:, None]
tl.store(out_ptr0 + (x0), tmp7, xmask)
tl.store(out_ptr1 + (x0), tmp11, xmask)
tl.store(out_ptr2 + (x0), tmp15, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cx/ccxl4ot76narch337s3yzdr2kc5f7mx4prnbdlldsn45cvcled6g.py
# Topologically Sorted Source Nodes: [bce, mul_1, add, add_1, add_2, dice, sum_4, truediv_1, dice_1, add_3], Original ATen: [aten.binary_cross_entropy_with_logits, aten.mul, aten.add, aten.div, aten.sum, aten.rsub]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# add_3 => add_3
# bce => abs_1, exp, full_default, log1p, mean, minimum, mul, neg, sub, sub_1, sub_2
# dice => div
# dice_1 => sub_3
# mul_1 => mul_2
# sum_4 => sum_4
# truediv_1 => div_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 = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 2.0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, 1e-05), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, %sum_3), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, 1e-05), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add, %add_2), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%div,), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_4, 4), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %div_1), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean, %sub_3), kwargs = {})
triton_per_fused_add_binary_cross_entropy_with_logits_div_mul_rsub_sum_2 = async_compile.triton('triton_per_fused_add_binary_cross_entropy_with_logits_div_mul_rsub_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 4],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {4: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=(4,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_binary_cross_entropy_with_logits_div_mul_rsub_sum_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_div_mul_rsub_sum_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp5 = tl.load(in_ptr1 + (r0), None)
tmp6 = tl.load(in_ptr2 + (r0), None)
tmp13 = tl.load(in_out_ptr0 + (0))
tmp14 = tl.broadcast_to(tmp13, [XBLOCK, 1])
tmp1 = 2.0
tmp2 = tmp0 * tmp1
tmp3 = 1e-05
tmp4 = tmp2 + tmp3
tmp7 = tmp5 + tmp6
tmp8 = tmp7 + tmp3
tmp9 = tmp4 / tmp8
tmp10 = tl.broadcast_to(tmp9, [XBLOCK, RBLOCK])
tmp12 = tl.sum(tmp10, 1)[:, None]
tmp15 = 256.0
tmp16 = tmp14 / tmp15
tmp17 = 0.25
tmp18 = tmp12 * tmp17
tmp19 = 1.0
tmp20 = tmp19 - tmp18
tmp21 = tmp16 + tmp20
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp21, 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)
# Topologically Sorted Source Nodes: [bce], Original ATen: [aten.binary_cross_entropy_with_logits]
stream0 = get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_0.run(arg0_1, arg1_1, buf0, 1, 256, grid=grid(1), stream=stream0)
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf2 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf3 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [intersection, sum_1, sum_2, sum_3], Original ATen: [aten.mul, aten.sum]
triton_per_fused_mul_sum_1.run(arg1_1, arg0_1, buf1, buf2, buf3, 4, 64, grid=grid(4), stream=stream0)
del arg0_1
del arg1_1
buf5 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [bce, mul_1, add, add_1, add_2, dice, sum_4, truediv_1, dice_1, add_3], Original ATen: [aten.binary_cross_entropy_with_logits, aten.mul, aten.add, aten.div, aten.sum, aten.rsub]
triton_per_fused_add_binary_cross_entropy_with_logits_div_mul_rsub_sum_2.run(buf5, buf1, buf2, buf3, 1, 4, grid=grid(1), stream=stream0)
del buf1
del buf2
del buf3
return (buf5, )
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 BCEDiceLoss(nn.Module):
def __init__(self):
super(BCEDiceLoss, self).__init__()
def forward(self, input, target):
bce = F.binary_cross_entropy_with_logits(input, target)
input = torch.sigmoid(input)
num = target.size(0)
input = input.view(num, -1)
target = target.view(num, -1)
smooth = 1e-05
intersection = input * target
dice = (2.0 * intersection.sum(1) + smooth) / (input.sum(1) +
target.sum(1) + smooth)
dice = 1 - dice.sum() / num
return bce + dice
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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_0(in_ptr0, in_ptr1,
out_ptr0, 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))
tl.store(out_ptr0 + tl.full([1], 0, tl.int32), tmp15, None)
@triton.jit
def triton_per_fused_mul_sum_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
out_ptr2, 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)
tmp2 = tl.load(in_ptr1 + (r1 + 64 * x0), xmask, other=0.0)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp14 = tl.where(xmask, tmp12, 0)
tmp15 = tl.sum(tmp14, 1)[:, None]
tl.store(out_ptr0 + x0, tmp7, xmask)
tl.store(out_ptr1 + x0, tmp11, xmask)
tl.store(out_ptr2 + x0, tmp15, xmask)
@triton.jit
def triton_per_fused_add_binary_cross_entropy_with_logits_div_mul_rsub_sum_2(
in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, rnumel, XBLOCK: tl.
constexpr):
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp5 = tl.load(in_ptr1 + r0, None)
tmp6 = tl.load(in_ptr2 + r0, None)
tmp13 = tl.load(in_out_ptr0 + 0)
tmp14 = tl.broadcast_to(tmp13, [XBLOCK, 1])
tmp1 = 2.0
tmp2 = tmp0 * tmp1
tmp3 = 1e-05
tmp4 = tmp2 + tmp3
tmp7 = tmp5 + tmp6
tmp8 = tmp7 + tmp3
tmp9 = tmp4 / tmp8
tmp10 = tl.broadcast_to(tmp9, [XBLOCK, RBLOCK])
tmp12 = tl.sum(tmp10, 1)[:, None]
tmp15 = 256.0
tmp16 = tmp14 / tmp15
tmp17 = 0.25
tmp18 = tmp12 * tmp17
tmp19 = 1.0
tmp20 = tmp19 - tmp18
tmp21 = tmp16 + tmp20
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp21, 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)
get_raw_stream(0)
triton_per_fused_binary_cross_entropy_with_logits_0[grid(1)](arg0_1,
arg1_1, buf0, 1, 256, num_warps=2, num_stages=1)
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
buf2 = empty_strided_cuda((4,), (1,), torch.float32)
buf3 = empty_strided_cuda((4,), (1,), torch.float32)
triton_per_fused_mul_sum_1[grid(4)](arg1_1, arg0_1, buf1, buf2,
buf3, 4, 64, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
buf5 = buf0
del buf0
triton_per_fused_add_binary_cross_entropy_with_logits_div_mul_rsub_sum_2[
grid(1)](buf5, buf1, buf2, buf3, 1, 4, XBLOCK=1, num_warps=2,
num_stages=1)
del buf1
del buf2
del buf3
return buf5,
class BCEDiceLossNew(nn.Module):
def __init__(self):
super(BCEDiceLossNew, 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]
| DLWK/EANet | BCEDiceLoss | false | 7,953 | [
"MIT"
] | 14 | 3680e099dd815117d4a54f928fb8247aa2f0b71a | https://github.com/DLWK/EANet/tree/3680e099dd815117d4a54f928fb8247aa2f0b71a | import torch
import torch.nn as nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input, target):
bce = F.binary_cross_entropy_with_logits(input, target)
input = torch.sigmoid(input)
num = target.size(0)
input = input.view(num, -1)
target = target.view(num, -1)
smooth = 1e-05
intersection = input * target
dice = (2.0 * intersection.sum(1) + smooth) / (input.sum(1) +
target.sum(1) + smooth)
dice = 1 - dice.sum() / num
return bce + dice
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
PointerAttention | # 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_1/inductor_cache/go/cgobu5sa2tyxtk6mrlov2ofgra2mvlz7hlazhico2xqgi24yq2dl.py
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# align_vectors => amax, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view, [-1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view, %amax), kwargs = {})
triton_poi_fused__log_softmax_0 = async_compile.triton('triton_poi_fused__log_softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__log_softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ra/crarp4npflln6yt45bhalbbozscbsjsfwkxoelsyyluujbbwkkhj.py
# Topologically Sorted Source Nodes: [align_vectors_2], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# align_vectors_2 => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_1 = async_compile.triton('triton_poi_fused_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, 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
x1 = (xindex // 4) % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (4*x4), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x4)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + (4*x4)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + (4*x4)), 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 + (x0 + (4*x2) + (16*x1)), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4), (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: [align], Original ATen: [aten.bmm]
extern_kernels.bmm(arg0_1, reinterpret_tensor(arg1_1, (4, 4, 4), (16, 1, 4), 0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._log_softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__log_softmax_0.run(buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [align_vectors_2], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf1, buf2, 64, grid=grid(64), stream=stream0)
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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class PointerAttention(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax'):
super(PointerAttention, self).__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
if self.attn_type == 'general':
self.linear_in = nn.Linear(dim, dim, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view(tgt_batch, tgt_len, 1, dim)
wq = wq.expand(tgt_batch, tgt_len, src_len, dim)
uh = self.linear_context(h_s.contiguous().view(-1, dim))
uh = uh.view(src_batch, 1, src_len, dim)
uh = uh.expand(src_batch, tgt_len, src_len, dim)
wquh = torch.tanh(wq + uh)
return self.v(wquh.view(-1, dim)).view(tgt_batch, tgt_len, src_len)
def forward(self, source, memory_bank, memory_lengths=None, coverage=None):
"""
Args:
source (FloatTensor): query vectors ``(batch, tgt_len, dim)``
memory_bank (FloatTensor): source vectors ``(batch, src_len, dim)``
memory_lengths (LongTensor): the source context lengths ``(batch,)``
coverage (FloatTensor): None (not supported yet)
Returns:
(FloatTensor, FloatTensor):
* Computed vector ``(tgt_len, batch, dim)``
* Attention distribtutions for each query
``(tgt_len, batch, src_len)``
"""
if source.dim() == 2:
one_step = True
source = source.unsqueeze(1)
else:
one_step = False
batch, source_l, dim = memory_bank.size()
batch_, target_l, dim_ = source.size()
aeq(batch, batch_)
aeq(dim, dim_)
aeq(self.dim, dim)
if coverage is not None:
batch_, source_l_ = coverage.size()
aeq(batch, batch_)
aeq(source_l, source_l_)
if coverage is not None:
cover = coverage.view(-1).unsqueeze(1)
memory_bank += self.linear_cover(cover).view_as(memory_bank)
memory_bank = torch.tanh(memory_bank)
align = self.score(source, memory_bank)
if memory_lengths is not None:
mask = sequence_mask(memory_lengths, max_len=align.size(-1))
mask = mask.unsqueeze(1)
align.masked_fill_(~mask, -float('inf'))
if self.attn_func == 'softmax':
align_vectors = torch.log_softmax(align.view(batch * target_l,
source_l), -1)
else:
align_vectors = sparsemax(align.view(batch * target_l, source_l
), -1)
align_vectors = align_vectors.view(batch, target_l, source_l)
if one_step:
align_vectors = align_vectors.squeeze(1)
batch_, source_l_ = align_vectors.size()
aeq(batch, batch_)
aeq(source_l, source_l_)
else:
align_vectors = align_vectors.transpose(0, 1).contiguous()
target_l_, batch_, source_l_ = align_vectors.size()
aeq(target_l, target_l_)
aeq(batch, batch_)
aeq(source_l, source_l_)
return align_vectors
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.cuda
import torch.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__log_softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
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_clone_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
x3 = xindex
x4 = xindex // 4
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x4, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x4), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + 4 * x4), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + 4 * x4), 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 + (x0 + 4 * x2 + 16 * x1), tmp13, xmask)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4), (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)
extern_kernels.bmm(arg0_1, reinterpret_tensor(arg1_1, (4, 4, 4), (
16, 1, 4), 0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__log_softmax_0[grid(64)](buf0, buf1, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf2 = buf0
del buf0
triton_poi_fused_clone_1[grid(64)](buf1, buf2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf1
return buf2,
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class PointerAttentionNew(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax'):
super(PointerAttentionNew, self).__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
if self.attn_type == 'general':
self.linear_in = nn.Linear(dim, dim, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view(tgt_batch, tgt_len, 1, dim)
wq = wq.expand(tgt_batch, tgt_len, src_len, dim)
uh = self.linear_context(h_s.contiguous().view(-1, dim))
uh = uh.view(src_batch, 1, src_len, dim)
uh = uh.expand(src_batch, tgt_len, src_len, dim)
wquh = torch.tanh(wq + uh)
return self.v(wquh.view(-1, dim)).view(tgt_batch, tgt_len, src_len)
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DenDen047/data2text-macro-plan-py | PointerAttention | false | 7,954 | [
"MIT"
] | 20 | bb01ec6e23dab28c1e969f23bd55776b597fb995 | https://github.com/DenDen047/data2text-macro-plan-py/tree/bb01ec6e23dab28c1e969f23bd55776b597fb995 | import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class Model(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax'):
super().__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
if self.attn_type == 'general':
self.linear_in = nn.Linear(dim, dim, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view
# ... truncated (>4000 chars) for memory efficiency |
PolicyNetwork | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/cn/ccnvkf7kfnskbbfy2kwx55oghjftngamwdttghryrfs4g3fay72l.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 256
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gr/cgrekmp3vmfaskaxgm7wq3mhpcpisrqek37fdagqcpd5rqlnxvep.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_2 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_5, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_5, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 32)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vv/cvvhapfmnf3saoxluehcmlbqma5latd76fq263yvdkloxowagavl.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_2 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 32)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (256, 4), (4, 1))
assert_size_stride(primals_2, (256, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (256, 256), (256, 1))
assert_size_stride(primals_5, (256, ), (1, ))
assert_size_stride(primals_6, (2, 256), (256, 1))
assert_size_stride(primals_7, (2, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 256), (256, 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, 256), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 256), (4096, 1024, 256, 1), 0); del buf0 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf8, 16384, grid=grid(16384), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 256), (256, 1), 0), reinterpret_tensor(primals_4, (256, 256), (1, 256), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 256), (4096, 1024, 256, 1), 0); del buf2 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf7, 16384, grid=grid(16384), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 256), (256, 1), 0), reinterpret_tensor(primals_6, (256, 2), (1, 256), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 2), (32, 8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf4, buf5, 128, grid=grid(128), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 2), (32, 8, 2, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf5, buf6, 128, grid=grid(128), stream=stream0)
del buf5
return (buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 256), (256, 1), 0), reinterpret_tensor(buf3, (64, 256), (256, 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((256, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((256, ), (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((256, 256), (256, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((2, 256), (256, 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 PolicyNetwork(nn.Module):
def __init__(self):
super(PolicyNetwork, self).__init__()
self.fc1 = nn.Linear(4, 256)
self.fc2 = nn.Linear(256, 256)
self.fc3 = nn.Linear(256, 2)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.softmax(self.fc3(x))
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 256
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
@triton.jit
def triton_poi_fused__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
x3 = xindex
x0 = xindex % 8
x2 = xindex // 32
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = xindex // 32
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (256, 4), (4, 1))
assert_size_stride(primals_2, (256,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (256, 256), (256, 1))
assert_size_stride(primals_5, (256,), (1,))
assert_size_stride(primals_6, (2, 256), (256, 1))
assert_size_stride(primals_7, (2,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 256), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 256), (4096, 1024, 256, 1), 0
)
del buf0
buf8 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(16384)](buf1,
primals_2, buf8, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 256), (256, 1), 0),
reinterpret_tensor(primals_4, (256, 256), (1, 256), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 256), (4096, 1024, 256, 1), 0
)
del buf2
buf7 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(16384)](buf3,
primals_5, buf7, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 256),
(256, 1), 0), reinterpret_tensor(primals_6, (256, 2), (1, 256),
0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((4, 4, 4, 2), (32, 8, 2, 1), torch.float32)
triton_poi_fused__softmax_1[grid(128)](buf4, buf5, 128, XBLOCK=128,
num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 2), (32, 8, 2, 1), 0)
del buf4
triton_poi_fused__softmax_2[grid(128)](buf5, buf6, 128, XBLOCK=128,
num_warps=4, num_stages=1)
del buf5
return buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 256), (256, 1), 0
), reinterpret_tensor(buf3, (64, 256), (256, 1), 0
), buf6, primals_6, buf7, primals_4, buf8
class PolicyNetworkNew(nn.Module):
def __init__(self):
super(PolicyNetworkNew, self).__init__()
self.fc1 = nn.Linear(4, 256)
self.fc2 = nn.Linear(256, 256)
self.fc3 = nn.Linear(256, 2)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| DensoITLab/spinningup_in_pytorch | PolicyNetwork | false | 7,955 | [
"MIT"
] | 11 | 612d8c4c6593c8c5ecb5a939bf43085daac9e552 | https://github.com/DensoITLab/spinningup_in_pytorch/tree/612d8c4c6593c8c5ecb5a939bf43085daac9e552 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(4, 256)
self.fc2 = nn.Linear(256, 256)
self.fc3 = nn.Linear(256, 2)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.softmax(self.fc3(x))
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
InformedSender | # 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_1/inductor_cache/lj/cljoh642mrhblbqz4ipwrm2lotsbebgxx7pbgrugnexl2ewfvl4y.py
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# h_4 => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze_1, %unsqueeze_3, %unsqueeze_5, %unsqueeze_7], 2), 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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4) % 4
x0 = xindex % 4
x2 = (xindex // 16)
x3 = xindex
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (4*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 2, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr1 + (x0 + (4*x2)), tmp9 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1], 3, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tmp11 & tmp13
tmp15 = tl.load(in_ptr2 + (x0 + (4*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp16 = tmp0 >= tmp12
tmp17 = tl.full([1], 4, tl.int64)
tmp18 = tmp0 < tmp17
tmp19 = tl.load(in_ptr3 + (x0 + (4*x2)), 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)
tl.store(out_ptr0 + (x3), tmp22, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/65/c65iltr2pquzye7migngokhhtu5qf7ejrjl4tw7zz7m5pjbserj7.py
# Topologically Sorted Source Nodes: [h_6], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# h_6 => sigmoid
# Graph fragment:
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution,), kwargs = {})
triton_poi_fused_sigmoid_1 = async_compile.triton('triton_poi_fused_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_1(in_out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tl.store(in_out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kf/ckfojkadxu3pzjhmlbtcn6s4ufg52wics6gt2nnjnfq7kpsh4knh.py
# Topologically Sorted Source Nodes: [h_9], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# h_9 => sigmoid_1
# Graph fragment:
# %sigmoid_1 : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_sigmoid_2 = async_compile.triton('triton_poi_fused_sigmoid_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_2(in_out_ptr0, 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.sigmoid(tmp0)
tl.store(in_out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vd/cvdcxgajpj4qce4mul4ippu7ofcaej5wcuy7os5qdu26qgoxn6mi.py
# Topologically Sorted Source Nodes: [logits], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# logits => exp, log, sub_1, sum_1
# Graph fragment:
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mm_4, 1), kwargs = {})
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_tensor, 1.0), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%mul_tensor_1,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %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 = (%mul_tensor_1, %log), kwargs = {})
triton_per_fused__log_softmax_3 = async_compile.triton('triton_per_fused__log_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.persistent_reduction(
size_hints=[4, 128],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__log_softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 2, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__log_softmax_3(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 100
RBLOCK: tl.constexpr = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (100*x0)), rmask & xmask, other=0.0)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.where(rmask & xmask, tmp3, float("-inf"))
tmp6 = triton_helpers.max2(tmp5, 1)[:, None]
tmp7 = tmp2 - tmp6
tmp8 = tmp7 * tmp1
tmp9 = tl_math.exp(tmp8)
tmp10 = tl.broadcast_to(tmp9, [XBLOCK, RBLOCK])
tmp12 = tl.where(rmask & xmask, tmp10, 0)
tmp13 = tl.sum(tmp12, 1)[:, None]
tmp14 = tl_math.log(tmp13)
tmp15 = tmp8 - tmp14
tl.store(out_ptr2 + (r1 + (100*x0)), tmp15, rmask & xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 1, 4, 1), (4, 4, 1, 1))
assert_size_stride(primals_4, (1, 1, 4, 1), (4, 4, 1, 1))
assert_size_stride(primals_5, (100, 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: [h_i], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_i_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 16), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_i_6], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 32), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_i_9], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 48), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf3)
del primals_2
buf4 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(buf0, buf1, buf2, buf3, buf4, 64, grid=grid(64), stream=stream0)
del buf0
del buf1
del buf2
del buf3
# Topologically Sorted Source Nodes: [h_5], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(buf4, primals_3, stride=(4, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 4, 1, 4), (16, 4, 4, 1))
buf6 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [h_6], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_1.run(buf6, 64, grid=grid(64), stream=stream0)
# Topologically Sorted Source Nodes: [h_8], Original ATen: [aten.convolution]
buf7 = extern_kernels.convolution(reinterpret_tensor(buf6, (4, 1, 4, 4), (16, 4, 4, 1), 0), primals_4, stride=(4, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf7, (4, 1, 1, 4), (4, 4, 4, 1))
buf8 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [h_9], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_2.run(buf8, 16, grid=grid(16), stream=stream0)
buf9 = empty_strided_cuda((4, 100), (100, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_12], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf8, (4, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 100), (1, 4), 0), out=buf9)
buf12 = empty_strided_cuda((4, 100), (100, 1), torch.float32)
# Topologically Sorted Source Nodes: [logits], Original ATen: [aten._log_softmax]
triton_per_fused__log_softmax_3.run(buf9, buf12, 4, 100, grid=grid(4), stream=stream0)
del buf9
return (buf12, primals_3, primals_4, reinterpret_tensor(primals_1, (4, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (4, 1), 16), reinterpret_tensor(primals_1, (4, 4), (4, 1), 32), reinterpret_tensor(primals_1, (4, 4), (4, 1), 48), buf4, buf6, buf8, buf12, primals_5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (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, 4, 1), (4, 4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 1, 4, 1), (4, 4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((100, 4), (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
import torch.nn.parallel
import torch.utils.data
import torch.distributions
class InformedSender(nn.Module):
def __init__(self, game_size, feat_size, embedding_size, hidden_size,
vocab_size=100, temp=1.0):
super(InformedSender, self).__init__()
self.game_size = game_size
self.embedding_size = embedding_size
self.hidden_size = hidden_size
self.vocab_size = vocab_size
self.temp = temp
self.lin1 = nn.Linear(feat_size, embedding_size, bias=False)
self.conv2 = nn.Conv2d(1, hidden_size, kernel_size=(game_size, 1),
stride=(game_size, 1), bias=False)
self.conv3 = nn.Conv2d(1, 1, kernel_size=(hidden_size, 1), stride=(
hidden_size, 1), bias=False)
self.lin4 = nn.Linear(embedding_size, vocab_size, bias=False)
def forward(self, x, return_embeddings=False):
emb = self.return_embeddings(x)
h = self.conv2(emb)
h = torch.sigmoid(h)
h = h.transpose(1, 2)
h = self.conv3(h)
h = torch.sigmoid(h)
h = h.squeeze(dim=1)
h = h.squeeze(dim=1)
h = self.lin4(h)
h = h.mul(1.0 / self.temp)
logits = F.log_softmax(h, dim=1)
return logits
def return_embeddings(self, x):
embs = []
for i in range(self.game_size):
h = x[i]
if len(h.size()) == 3:
h = h.squeeze(dim=-1)
h_i = self.lin1(h)
h_i = h_i.unsqueeze(dim=1)
h_i = h_i.unsqueeze(dim=1)
embs.append(h_i)
h = torch.cat(embs, dim=2)
return h
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'game_size': 4, 'feat_size': 4, 'embedding_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 math as tl_math
import torch.nn as nn
import torch.nn.parallel
import torch.utils.data
import torch.distributions
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, in_ptr3, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4 % 4
x0 = xindex % 4
x2 = xindex // 16
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 4 * x2), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 2, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr1 + (x0 + 4 * x2), tmp9 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1], 3, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tmp11 & tmp13
tmp15 = tl.load(in_ptr2 + (x0 + 4 * x2), tmp14 & xmask, eviction_policy
='evict_last', other=0.0)
tmp16 = tmp0 >= tmp12
tl.full([1], 4, tl.int64)
tmp19 = tl.load(in_ptr3 + (x0 + 4 * x2), 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)
tl.store(out_ptr0 + x3, tmp22, xmask)
@triton.jit
def triton_poi_fused_sigmoid_1(in_out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tl.store(in_out_ptr0 + x0, tmp1, xmask)
@triton.jit
def triton_poi_fused_sigmoid_2(in_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.sigmoid(tmp0)
tl.store(in_out_ptr0 + x0, tmp1, xmask)
@triton.jit
def triton_per_fused__log_softmax_3(in_ptr0, out_ptr2, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 4
rnumel = 100
RBLOCK: tl.constexpr = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 100 * x0), rmask & xmask, other=0.0)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.where(rmask & xmask, tmp3, float('-inf'))
tmp6 = triton_helpers.max2(tmp5, 1)[:, None]
tmp7 = tmp2 - tmp6
tmp8 = tmp7 * tmp1
tmp9 = tl_math.exp(tmp8)
tmp10 = tl.broadcast_to(tmp9, [XBLOCK, RBLOCK])
tmp12 = tl.where(rmask & xmask, tmp10, 0)
tmp13 = tl.sum(tmp12, 1)[:, None]
tmp14 = tl_math.log(tmp13)
tmp15 = tmp8 - tmp14
tl.store(out_ptr2 + (r1 + 100 * x0), tmp15, rmask & xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 1, 4, 1), (4, 4, 1, 1))
assert_size_stride(primals_4, (1, 1, 4, 1), (4, 4, 1, 1))
assert_size_stride(primals_5, (100, 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(reinterpret_tensor(primals_1, (4, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 16),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 32),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (4, 4), (4, 1), 48),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf3)
del primals_2
buf4 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(64)](buf0, buf1, buf2, buf3, buf4, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del buf0
del buf1
del buf2
del buf3
buf5 = extern_kernels.convolution(buf4, primals_3, stride=(4, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 4, 1, 4), (16, 4, 4, 1))
buf6 = buf5
del buf5
triton_poi_fused_sigmoid_1[grid(64)](buf6, 64, XBLOCK=64, num_warps
=1, num_stages=1)
buf7 = extern_kernels.convolution(reinterpret_tensor(buf6, (4, 1, 4,
4), (16, 4, 4, 1), 0), primals_4, stride=(4, 1), padding=(0, 0),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=1, bias=None)
assert_size_stride(buf7, (4, 1, 1, 4), (4, 4, 4, 1))
buf8 = buf7
del buf7
triton_poi_fused_sigmoid_2[grid(16)](buf8, 16, XBLOCK=16, num_warps
=1, num_stages=1)
buf9 = empty_strided_cuda((4, 100), (100, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf8, (4, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 100), (1, 4), 0), out=buf9)
buf12 = empty_strided_cuda((4, 100), (100, 1), torch.float32)
triton_per_fused__log_softmax_3[grid(4)](buf9, buf12, 4, 100,
XBLOCK=1, num_warps=2, num_stages=1)
del buf9
return buf12, primals_3, primals_4, reinterpret_tensor(primals_1, (4, 4
), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (4, 1), 16
), reinterpret_tensor(primals_1, (4, 4), (4, 1), 32
), reinterpret_tensor(primals_1, (4, 4), (4, 1), 48
), buf4, buf6, buf8, buf12, primals_5
class InformedSenderNew(nn.Module):
def __init__(self, game_size, feat_size, embedding_size, hidden_size,
vocab_size=100, temp=1.0):
super(InformedSenderNew, self).__init__()
self.game_size = game_size
self.embedding_size = embedding_size
self.hidden_size = hidden_size
self.vocab_size = vocab_size
self.temp = temp
self.lin1 = nn.Linear(feat_size, embedding_size, bias=False)
self.conv2 = nn.Conv2d(1, hidden_size, kernel_size=(game_size, 1),
stride=(game_size, 1), bias=False)
self.conv3 = nn.Conv2d(1, 1, kernel_size=(hidden_size, 1), stride=(
hidden_size, 1), bias=False)
self.lin4 = nn.Linear(embedding_size, vocab_size, bias=False)
def return_embeddings(self, x):
embs = []
for i in range(self.game_size):
h = x[i]
if len(h.size()) == 3:
h = h.squeeze(dim=-1)
h_i = self.lin1(h)
h_i = h_i.unsqueeze(dim=1)
h_i = h_i.unsqueeze(dim=1)
embs.append(h_i)
h = torch.cat(embs, dim=2)
return h
def forward(self, input_0):
primals_2 = self.lin1.weight
primals_3 = self.conv2.weight
primals_4 = self.conv3.weight
primals_5 = self.lin4.weight
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| Daetheys/Lazimpa | InformedSender | false | 7,956 | [
"MIT"
] | 15 | 21f4f4ebcdfa8b6a775b64673dd3001763c91cf1 | https://github.com/Daetheys/Lazimpa/tree/21f4f4ebcdfa8b6a775b64673dd3001763c91cf1 | import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.utils.data
import torch.distributions
class Model(nn.Module):
def __init__(self, game_size, feat_size, embedding_size, hidden_size,
vocab_size=100, temp=1.0):
super().__init__()
self.game_size = game_size
self.embedding_size = embedding_size
self.hidden_size = hidden_size
self.vocab_size = vocab_size
self.temp = temp
self.lin1 = nn.Linear(feat_size, embedding_size, bias=False)
self.conv2 = nn.Conv2d(1, hidden_size, kernel_size=(game_size, 1),
stride=(game_size, 1), bias=False)
self.conv3 = nn.Conv2d(1, 1, kernel_size=(hidden_size, 1), stride=(
hidden_size, 1), bias=False)
self.lin4 = nn.Linear(embedding_size, vocab_size, bias=False)
def forward(self, x, return_embeddings=False):
emb = self.return_embeddings(x)
h = self.conv2(emb)
h = torch.sigmoid(h)
h = h.transpose(1, 2)
h = self.conv3(h)
h = torch.sigmoid(h)
h = h.squeeze(dim=1)
h = h.squeeze(dim=1)
h = self.lin4(h)
h = h.mul(1.0 / self.temp)
logits = F.log_softmax(h, dim=1)
return logits
def return_embeddings(self, x):
embs = []
for i in range(self.game_size):
h = x[i]
if len(h.size()) == 3:
h = h.squeeze(dim=-1)
h_i = self.lin1(h)
h_i = h_i.unsqueeze(dim=1)
h_i = h_i.unsqueeze(dim=1)
embs.append(h_i)
h = torch.cat(embs, dim=2)
return h
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'game_size': 4, 'feat_size': 4, 'embedding_size': 4,
'hidden_size': 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_1/inductor_cache/4u/c4u3liwgy6ah3xlxfhvbczvvfare3yvkd7osc2eyowkxndto3p24.py
# Topologically Sorted Source Nodes: [layer_norm], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# layer_norm => add, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-06), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_native_layer_norm_0 = async_compile.triton('triton_poi_fused_native_layer_norm_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-06
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_1/inductor_cache/gn/cgn3tpasui6fv3xxba47jzqip7bgipyrz4akedry64e2fx5k4rvd.py
# Topologically Sorted Source Nodes: [layer_norm], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# layer_norm => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-06), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_3, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_2), kwargs = {})
triton_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/3a/c3asrsyuybhvyyokqadgkn4dnhzrdoetfuhpfcg2pkqy2j4beaia.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_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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_2(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/h3/ch3dkn75z5nv2s6poei22lobtkafusftzt2ks6goill4cq3nfbmj.py
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add => add_2
# Graph fragment:
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_3), kwargs = {})
triton_poi_fused_add_3 = async_compile.triton('triton_poi_fused_add_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_3', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_3(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
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, ), (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((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [layer_norm], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_0.run(primals_3, buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [layer_norm], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(primals_3, buf0, buf1, primals_1, primals_2, buf2, 256, grid=grid(256), stream=stream0)
del buf0
del buf1
del primals_1
del primals_2
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
buf4 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_2.run(buf4, primals_5, buf7, 256, grid=grid(256), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf4, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf5)
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
triton_poi_fused_add_3.run(buf6, primals_7, primals_3, 256, grid=grid(256), stream=stream0)
del primals_7
return (buf6, primals_3, reinterpret_tensor(buf2, (64, 4), (4, 1), 0), reinterpret_tensor(buf4, (64, 4), (4, 1), 0), 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((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 torch
import torch.nn as nn
import torch.cuda
import torch.distributed
class PositionwiseFeedForward(nn.Module):
""" A two-layer Feed-Forward-Network with residual layer norm.
Args:
d_model (int): the size of input for the first-layer of the FFN.
d_ff (int): the hidden layer size of the second-layer
of the FNN.
dropout (float): dropout probability in :math:`[0, 1)`.
"""
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.layer_norm = nn.LayerNorm(d_model, eps=1e-06)
self.dropout_1 = nn.Dropout(dropout)
self.relu = nn.ReLU()
self.dropout_2 = nn.Dropout(dropout)
def forward(self, x):
"""Layer definition.
Args:
x: ``(batch_size, input_len, model_dim)``
Returns:
(FloatTensor): Output ``(batch_size, input_len, model_dim)``.
"""
inter = self.dropout_1(self.relu(self.w_1(self.layer_norm(x))))
output = self.dropout_2(self.w_2(inter))
return output + x
def update_dropout(self, dropout):
self.dropout_1.p = dropout
self.dropout_2.p = dropout
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 import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
import torch.cuda
import torch.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp9 = tmp0 - tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp1 - tmp8
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp3 - tmp8
tmp15 = tmp14 * tmp14
tmp16 = tmp13 + tmp15
tmp17 = tmp5 - tmp8
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp19 / tmp7
tmp21 = 1e-06
tmp22 = tmp20 + tmp21
tmp23 = libdevice.rsqrt(tmp22)
tl.store(out_ptr0 + x0, tmp8, xmask)
tl.store(out_ptr1 + x0, tmp23, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_2(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_add_3(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
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,), (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((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
get_raw_stream(0)
triton_poi_fused_native_layer_norm_0[grid(64)](primals_3, buf0,
buf1, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(256)](primals_3, buf0,
buf1, primals_1, primals_2, buf2, 256, XBLOCK=128, num_warps=4,
num_stages=1)
del buf0
del buf1
del primals_1
del primals_2
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
buf4 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_2[grid(256)](buf4,
primals_5, buf7, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf4, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf5)
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused_add_3[grid(256)](buf6, primals_7, primals_3, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
return buf6, primals_3, reinterpret_tensor(buf2, (64, 4), (4, 1), 0
), reinterpret_tensor(buf4, (64, 4), (4, 1), 0
), primals_6, buf7, primals_4
class PositionwiseFeedForwardNew(nn.Module):
""" A two-layer Feed-Forward-Network with residual layer norm.
Args:
d_model (int): the size of input for the first-layer of the FFN.
d_ff (int): the hidden layer size of the second-layer
of the FNN.
dropout (float): dropout probability in :math:`[0, 1)`.
"""
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.layer_norm = nn.LayerNorm(d_model, eps=1e-06)
self.dropout_1 = nn.Dropout(dropout)
self.relu = nn.ReLU()
self.dropout_2 = nn.Dropout(dropout)
def update_dropout(self, dropout):
self.dropout_1.p = dropout
self.dropout_2.p = dropout
def forward(self, input_0):
primals_4 = self.w_1.weight
primals_1 = self.w_1.bias
primals_6 = self.w_2.weight
primals_2 = self.w_2.bias
primals_5 = self.layer_norm.weight
primals_7 = self.layer_norm.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| DenDen047/data2text-macro-plan-py | PositionwiseFeedForward | false | 7,957 | [
"MIT"
] | 20 | bb01ec6e23dab28c1e969f23bd55776b597fb995 | https://github.com/DenDen047/data2text-macro-plan-py/tree/bb01ec6e23dab28c1e969f23bd55776b597fb995 | import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
class Model(nn.Module):
""" A two-layer Feed-Forward-Network with residual layer norm.
Args:
d_model (int): the size of input for the first-layer of the FFN.
d_ff (int): the hidden layer size of the second-layer
of the FNN.
dropout (float): dropout probability in :math:`[0, 1)`.
"""
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.layer_norm = nn.LayerNorm(d_model, eps=1e-06)
self.dropout_1 = nn.Dropout(dropout)
self.relu = nn.ReLU()
self.dropout_2 = nn.Dropout(dropout)
def forward(self, x):
"""Layer definition.
Args:
x: ``(batch_size, input_len, model_dim)``
Returns:
(FloatTensor): Output ``(batch_size, input_len, model_dim)``.
"""
inter = self.dropout_1(self.relu(self.w_1(self.layer_norm(x))))
output = self.dropout_2(self.w_2(inter))
return output + x
def update_dropout(self, dropout):
self.dropout_1.p = dropout
self.dropout_2.p = dropout
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
GELU2 | # 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_1/inductor_cache/mu/cmul3o6p3oz2yowa3zfyaufvv47formp6gyzzdxmgfkua4nkp6b5.py
# Topologically Sorted Source Nodes: [mul, sigmoid, mul_1], Original ATen: [aten.mul, aten.sigmoid]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# sigmoid => sigmoid
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 1.702), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%mul,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %sigmoid), 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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), 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': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, 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 = 1.702
tmp2 = tmp0 * tmp1
tmp3 = tl.sigmoid(tmp2)
tmp4 = tmp0 * 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: [mul, sigmoid, mul_1], Original ATen: [aten.mul, aten.sigmoid]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_sigmoid_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 GELU2(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return x * torch.sigmoid(1.702 * 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
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_sigmoid_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 = 1.702
tmp2 = tmp0 * tmp1
tmp3 = tl.sigmoid(tmp2)
tmp4 = tmp0 * 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_sigmoid_0[grid(256)](arg0_1, buf0, 256, XBLOCK
=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class GELU2New(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DQiaole/ZITS | GELU2 | false | 7,958 | [
"Apache-2.0"
] | 40 | 5f7a060167790789d5e29a3d14d3c2ef8a34e765 | https://github.com/DQiaole/ZITS/tree/5f7a060167790789d5e29a3d14d3c2ef8a34e765 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
return x * torch.sigmoid(1.702 * x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GlobalAttentionContext | # 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_1/inductor_cache/cv/ccv6wuuk2f3t36jk6z6siizklj3kweutyi3yvvr2vh7sazv7nrd4.py
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# align_vectors => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/yt/cytqs2smvhzqhhhv5nhgfsoz7g7pop2pi3eoeztc4dtuktnwv56m.py
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# align_vectors => 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_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/es/cesgynyfcvg3brsnh76xkd5a6rbhk6wlzv735etzvza7fyynryvy.py
# Topologically Sorted Source Nodes: [c_1], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# c_1 => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%bmm_1, [1]), kwargs = {})
triton_poi_fused_mean_2 = async_compile.triton('triton_poi_fused_mean_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mean_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_2(in_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 = 1.0
tmp2 = tmp0 / tmp1
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 4), (4, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [align], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(primals_2, (4, 1, 4), (0, 0, 1), 0), reinterpret_tensor(primals_1, (4, 4, 4), (16, 1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf0, buf1, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 16, grid=grid(16), stream=stream0)
buf3 = reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [c], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf2, (4, 1, 4), (4, 0, 1), 0), primals_1, out=buf3)
del buf2
buf4 = reinterpret_tensor(buf3, (4, 4), (4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [c_1], Original ATen: [aten.mean]
triton_poi_fused_mean_2.run(buf4, 16, grid=grid(16), stream=stream0)
return (buf4, reinterpret_tensor(primals_1, (4, 4, 4), (16, 1, 4), 0), 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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 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
import torch.cuda
import torch.distributed
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class GlobalAttentionContext(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax'):
super(GlobalAttentionContext, self).__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
self.source = nn.Parameter(torch.Tensor(1, dim))
if self.attn_type == 'general':
self.linear_in = nn.Linear(dim, dim, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view(tgt_batch, tgt_len, 1, dim)
wq = wq.expand(tgt_batch, tgt_len, src_len, dim)
uh = self.linear_context(h_s.contiguous().view(-1, dim))
uh = uh.view(src_batch, 1, src_len, dim)
uh = uh.expand(src_batch, tgt_len, src_len, dim)
wquh = torch.tanh(wq + uh)
return self.v(wquh.view(-1, dim)).view(tgt_batch, tgt_len, src_len)
def forward(self, memory_bank, memory_lengths=None, coverage=None):
"""
Args:
source (FloatTensor): query vectors ``(batch, tgt_len, dim)``
memory_bank (FloatTensor): source vectors ``(batch, src_len, dim)``
memory_lengths (LongTensor): the source context lengths ``(batch,)``
coverage (FloatTensor): None (not supported yet)
Returns:
(FloatTensor, FloatTensor):
* Computed vector ``(tgt_len, batch, dim)``
* Attention distribtutions for each query
``(tgt_len, batch, src_len)``
"""
batch, source_l, dim = memory_bank.size()
source = self.source
source = source.expand(batch, -1)
source = source.unsqueeze(1)
batch_, target_l, dim_ = source.size()
aeq(batch, batch_)
aeq(dim, dim_)
aeq(self.dim, dim)
if coverage is not None:
batch_, source_l_ = coverage.size()
aeq(batch, batch_)
aeq(source_l, source_l_)
if coverage is not None:
cover = coverage.view(-1).unsqueeze(1)
memory_bank += self.linear_cover(cover).view_as(memory_bank)
memory_bank = torch.tanh(memory_bank)
align = self.score(source, memory_bank)
if memory_lengths is not None:
mask = sequence_mask(memory_lengths, max_len=align.size(-1))
mask = mask.unsqueeze(1)
align.masked_fill_(~mask, -float('inf'))
if self.attn_func == 'softmax':
align_vectors = F.softmax(align.view(batch * target_l, source_l
), -1)
else:
align_vectors = sparsemax(align.view(batch * target_l, source_l
), -1)
align_vectors = align_vectors.view(batch, target_l, source_l)
c = torch.bmm(align_vectors, memory_bank)
c = c.mean(dim=1)
batch_, dim_ = c.size()
aeq(batch, batch_)
aeq(dim, dim_)
return c
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.cuda
import torch.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__softmax_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
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_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
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_mean_2(in_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 = 1.0
tmp2 = tmp0 / tmp1
tl.store(in_out_ptr0 + x0, tmp2, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 4), (4, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(primals_2, (4, 1, 4), (0, 0,
1), 0), reinterpret_tensor(primals_1, (4, 4, 4), (16, 1, 4), 0),
out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(16)](buf0, buf1, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(16)](buf1, buf2, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf3 = reinterpret_tensor(buf1, (4, 1, 4), (4, 4, 1), 0)
del buf1
extern_kernels.bmm(reinterpret_tensor(buf2, (4, 1, 4), (4, 0, 1), 0
), primals_1, out=buf3)
del buf2
buf4 = reinterpret_tensor(buf3, (4, 4), (4, 1), 0)
del buf3
triton_poi_fused_mean_2[grid(16)](buf4, 16, XBLOCK=16, num_warps=1,
num_stages=1)
return buf4, reinterpret_tensor(primals_1, (4, 4, 4), (16, 1, 4), 0), buf0
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class GlobalAttentionContextNew(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax'):
super(GlobalAttentionContextNew, self).__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
self.source = nn.Parameter(torch.Tensor(1, dim))
if self.attn_type == 'general':
self.linear_in = nn.Linear(dim, dim, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view(tgt_batch, tgt_len, 1, dim)
wq = wq.expand(tgt_batch, tgt_len, src_len, dim)
uh = self.linear_context(h_s.contiguous().view(-1, dim))
uh = uh.view(src_batch, 1, src_len, dim)
uh = uh.expand(src_batch, tgt_len, src_len, dim)
wquh = torch.tanh(wq + uh)
return self.v(wquh.view(-1, dim)).view(tgt_batch, tgt_len, src_len)
def forward(self, input_0):
primals_2 = self.source
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
| DenDen047/data2text-macro-plan-py | GlobalAttentionContext | false | 7,959 | [
"MIT"
] | 20 | bb01ec6e23dab28c1e969f23bd55776b597fb995 | https://github.com/DenDen047/data2text-macro-plan-py/tree/bb01ec6e23dab28c1e969f23bd55776b597fb995 | import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.cuda
import torch.distributed
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class Model(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax'):
super().__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
self.source = nn.Parameter(torch.Tensor(1, dim))
if self.attn_type == 'general':
self.linear_in = nn.Linear(dim, dim, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim
# ... truncated (>4000 chars) for memory efficiency |
SEBlock | # 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_1/inductor_cache/e7/ce73hyb6fl47lsvuo6oc4nyc7nbjn2cooo36plrte4gsotp7fcxm.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.avg_pool2d]
# Source node to ATen node mapping:
# x => avg_pool2d
# Graph fragment:
# %avg_pool2d : [num_users=2] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%primals_1, [4, 4]), kwargs = {})
triton_poi_fused_avg_pool2d_0 = async_compile.triton('triton_poi_fused_avg_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_avg_pool2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 16, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (16*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (16*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (16*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (16*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (4 + (16*x0)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (5 + (16*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (6 + (16*x0)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (7 + (16*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (8 + (16*x0)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (9 + (16*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (10 + (16*x0)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (11 + (16*x0)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (12 + (16*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (13 + (16*x0)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (14 + (16*x0)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr0 + (15 + (16*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp8 = tmp7 + tmp6
tmp10 = tmp9 + tmp8
tmp12 = tmp11 + tmp10
tmp14 = tmp13 + tmp12
tmp16 = tmp15 + tmp14
tmp18 = tmp17 + tmp16
tmp20 = tmp19 + tmp18
tmp22 = tmp21 + tmp20
tmp24 = tmp23 + tmp22
tmp26 = tmp25 + tmp24
tmp28 = tmp27 + tmp26
tmp30 = tmp29 + tmp28
tmp31 = 0.0625
tmp32 = tmp30 * tmp31
tl.store(out_ptr0 + (x0), tmp32, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bq/cbqs25ilz2wluqikclvslctdlvul4rcbcdk3m2pcvrxffj6hdrw5.py
# Topologically Sorted Source Nodes: [x_1, x_2], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x_1 => convolution
# x_2 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%avg_pool2d, %primals_2, %primals_3, [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_1 = async_compile.triton('triton_poi_fused_convolution_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_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_1/inductor_cache/hq/chqoxrrsmivwbzcaxvjamxpovu4dlvpia7zse2rga3c6sey6mx7z.py
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.convolution, aten.sigmoid]
# Source node to ATen node mapping:
# x_3 => convolution_1
# x_4 => sigmoid
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_sigmoid_2 = async_compile.triton('triton_poi_fused_convolution_sigmoid_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_sigmoid_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_sigmoid_2(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.sigmoid(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ai/caitvadvtu3nfxa2adqfgvuevqbtp2q64rafga5jav4tlnlsdts4.py
# Topologically Sorted Source Nodes: [x_5, mul], Original ATen: [aten.repeat, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# x_5 => repeat
# Graph fragment:
# %repeat : [num_users=1] = call_function[target=torch.ops.aten.repeat.default](args = (%sigmoid, [1, 1, 4, 4]), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %repeat), kwargs = {})
triton_poi_fused_mul_repeat_3 = async_compile.triton('triton_poi_fused_mul_repeat_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_repeat_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_repeat_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 // 16)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
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, 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, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.avg_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_avg_pool2d_0.run(primals_1, buf0, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [x_1], 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, 1, 1), (4, 1, 1, 1))
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [x_1, x_2], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_1.run(buf2, primals_3, 16, grid=grid(16), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf3 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 1, 1), (4, 1, 1, 1))
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.convolution, aten.sigmoid]
triton_poi_fused_convolution_sigmoid_2.run(buf4, primals_5, 16, grid=grid(16), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_5, mul], Original ATen: [aten.repeat, aten.mul]
triton_poi_fused_mul_repeat_3.run(primals_1, buf4, buf5, 256, grid=grid(256), stream=stream0)
return (buf5, primals_1, primals_2, primals_4, buf0, buf2, 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, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
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 SEBlock(nn.Module):
def __init__(self, input_channels, internal_neurons):
super(SEBlock, self).__init__()
self.down = nn.Conv2d(in_channels=input_channels, out_channels=
internal_neurons, kernel_size=1, stride=1, bias=True)
self.up = nn.Conv2d(in_channels=internal_neurons, out_channels=
input_channels, kernel_size=1, stride=1, bias=True)
def forward(self, inputs):
x = F.avg_pool2d(inputs, kernel_size=inputs.size(3))
x = self.down(x)
x = F.relu(x)
x = self.up(x)
x = F.sigmoid(x)
x = x.repeat(1, 1, inputs.size(2), inputs.size(3))
return inputs * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_channels': 4, 'internal_neurons': 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
@triton.jit
def triton_poi_fused_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 16 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr0 + (2 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr0 + (3 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp7 = tl.load(in_ptr0 + (4 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp9 = tl.load(in_ptr0 + (5 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (6 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp13 = tl.load(in_ptr0 + (7 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp15 = tl.load(in_ptr0 + (8 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp17 = tl.load(in_ptr0 + (9 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr0 + (10 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (11 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr0 + (12 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr0 + (13 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp27 = tl.load(in_ptr0 + (14 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr0 + (15 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp8 = tmp7 + tmp6
tmp10 = tmp9 + tmp8
tmp12 = tmp11 + tmp10
tmp14 = tmp13 + tmp12
tmp16 = tmp15 + tmp14
tmp18 = tmp17 + tmp16
tmp20 = tmp19 + tmp18
tmp22 = tmp21 + tmp20
tmp24 = tmp23 + tmp22
tmp26 = tmp25 + tmp24
tmp28 = tmp27 + tmp26
tmp30 = tmp29 + tmp28
tmp31 = 0.0625
tmp32 = tmp30 * tmp31
tl.store(out_ptr0 + x0, tmp32, xmask)
@triton.jit
def triton_poi_fused_convolution_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_convolution_sigmoid_2(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.sigmoid(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_mul_repeat_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 // 16
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_avg_pool2d_0[grid(16)](primals_1, buf0, 16, XBLOCK
=16, num_warps=1, num_stages=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, 1, 1), (4, 1, 1, 1))
buf2 = buf1
del buf1
triton_poi_fused_convolution_relu_1[grid(16)](buf2, primals_3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf3 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 1, 1), (4, 1, 1, 1))
buf4 = buf3
del buf3
triton_poi_fused_convolution_sigmoid_2[grid(16)](buf4, primals_5,
16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_repeat_3[grid(256)](primals_1, buf4, buf5, 256,
XBLOCK=256, num_warps=4, num_stages=1)
return buf5, primals_1, primals_2, primals_4, buf0, buf2, buf4
class SEBlockNew(nn.Module):
def __init__(self, input_channels, internal_neurons):
super(SEBlockNew, self).__init__()
self.down = nn.Conv2d(in_channels=input_channels, out_channels=
internal_neurons, kernel_size=1, stride=1, bias=True)
self.up = nn.Conv2d(in_channels=internal_neurons, out_channels=
input_channels, kernel_size=1, stride=1, bias=True)
def forward(self, input_0):
primals_2 = self.down.weight
primals_3 = self.down.bias
primals_4 = self.up.weight
primals_5 = self.up.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DingXiaoH/GSM-SGD | SEBlock | false | 7,960 | [
"MIT"
] | 40 | ffc605651c4c5115dfb8659ebe48ccf71d3955a0 | https://github.com/DingXiaoH/GSM-SGD/tree/ffc605651c4c5115dfb8659ebe48ccf71d3955a0 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, input_channels, internal_neurons):
super().__init__()
self.down = nn.Conv2d(in_channels=input_channels, out_channels=
internal_neurons, kernel_size=1, stride=1, bias=True)
self.up = nn.Conv2d(in_channels=internal_neurons, out_channels=
input_channels, kernel_size=1, stride=1, bias=True)
def forward(self, inputs):
x = F.avg_pool2d(inputs, kernel_size=inputs.size(3))
x = self.down(x)
x = F.relu(x)
x = self.up(x)
x = F.sigmoid(x)
x = x.repeat(1, 1, inputs.size(2), inputs.size(3))
return inputs * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Contract | # 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_1/inductor_cache/62/c6226xjgxsv2qrqtyizdd4zsuyuwempbuxvcommmakgcjleoyel2.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# x_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=[16, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 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
x3 = xindex % 2
x4 = (xindex // 2)
y0 = yindex % 2
y1 = (yindex // 2) % 2
y2 = (yindex // 4)
x6 = xindex
y5 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (2*x3) + (4*y1) + (8*x4) + (64*y2)), xmask & ymask)
tl.store(out_ptr0 + (x6 + (16*y5)), 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, 2, 2, 4, 2, 2), (64, 32, 16, 4, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(arg0_1, buf0, 16, 16, grid=grid(16, 16), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf0, (4, 16, 2, 2), (64, 4, 2, 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.onnx
class Contract(nn.Module):
def __init__(self, gain=2):
super().__init__()
self.gain = gain
def forward(self, x):
b, c, h, w = x.size()
s = self.gain
x = x.view(b, c, h // s, s, w // s, s)
x = x.permute(0, 3, 5, 1, 2, 4).contiguous()
return x.view(b, c * s * s, h // s, w // s)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.onnx
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 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
x3 = xindex % 2
x4 = xindex // 2
y0 = yindex % 2
y1 = yindex // 2 % 2
y2 = yindex // 4
x6 = xindex
y5 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 2 * x3 + 4 * y1 + 8 * x4 + 64 * y2),
xmask & ymask)
tl.store(out_ptr0 + (x6 + 16 * y5), 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, 2, 2, 4, 2, 2), (64, 32, 16, 4, 2, 1),
torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(16, 16)](arg0_1, buf0, 16, 16, XBLOCK
=16, YBLOCK=16, num_warps=4, num_stages=1)
del arg0_1
return reinterpret_tensor(buf0, (4, 16, 2, 2), (64, 4, 2, 1), 0),
class ContractNew(nn.Module):
def __init__(self, gain=2):
super().__init__()
self.gain = gain
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DDGRCF/YOLOX_OBB | Contract | false | 7,961 | [
"Apache-2.0"
] | 39 | 27b80953306492b8bc83b86b1353d8cee01ef9b6 | https://github.com/DDGRCF/YOLOX_OBB/tree/27b80953306492b8bc83b86b1353d8cee01ef9b6 | import torch
import torch.nn as nn
import torch.onnx
class Model(nn.Module):
def __init__(self, gain=2):
super().__init__()
self.gain = gain
def forward(self, x):
b, c, h, w = x.size()
s = self.gain
x = x.view(b, c, h // s, s, w // s, s)
x = x.permute(0, 3, 5, 1, 2, 4).contiguous()
return x.view(b, c * s * s, h // s, w // s)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ValueNetwork | # 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_1/inductor_cache/cn/ccnvkf7kfnskbbfy2kwx55oghjftngamwdttghryrfs4g3fay72l.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 256
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (256, 4), (4, 1))
assert_size_stride(primals_2, (256, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (256, 256), (256, 1))
assert_size_stride(primals_5, (256, ), (1, ))
assert_size_stride(primals_6, (1, 256), (256, 1))
assert_size_stride(primals_7, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 256), (256, 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, 256), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 256), (4096, 1024, 256, 1), 0); del buf0 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf7, 16384, grid=grid(16384), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 256), (256, 1), 0), reinterpret_tensor(primals_4, (256, 256), (1, 256), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 256), (4096, 1024, 256, 1), 0); del buf2 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf6, 16384, grid=grid(16384), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 256), (256, 1), 0), reinterpret_tensor(primals_6, (256, 1), (1, 256), 0), alpha=1, beta=1, out=buf5)
del primals_7
return (reinterpret_tensor(buf5, (4, 4, 4, 1), (16, 4, 1, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 256), (256, 1), 0), reinterpret_tensor(buf3, (64, 256), (256, 1), 0), primals_6, buf6, primals_4, buf7, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((256, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((256, ), (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((256, 256), (256, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((1, 256), (256, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
class ValueNetwork(nn.Module):
def __init__(self):
super(ValueNetwork, self).__init__()
self.fc1 = nn.Linear(4, 256)
self.fc2 = nn.Linear(256, 256)
self.fc3 = nn.Linear(256, 1)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 256
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (256, 4), (4, 1))
assert_size_stride(primals_2, (256,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (256, 256), (256, 1))
assert_size_stride(primals_5, (256,), (1,))
assert_size_stride(primals_6, (1, 256), (256, 1))
assert_size_stride(primals_7, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 256), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 256), (4096, 1024, 256, 1), 0
)
del buf0
buf7 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(16384)](buf1,
primals_2, buf7, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 256), (256, 1), 0),
reinterpret_tensor(primals_4, (256, 256), (1, 256), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 256), (4096, 1024, 256, 1), 0
)
del buf2
buf6 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(16384)](buf3,
primals_5, buf6, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 256),
(256, 1), 0), reinterpret_tensor(primals_6, (256, 1), (1, 256),
0), alpha=1, beta=1, out=buf5)
del primals_7
return reinterpret_tensor(buf5, (4, 4, 4, 1), (16, 4, 1, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 256), (256, 1), 0
), reinterpret_tensor(buf3, (64, 256), (256, 1), 0
), primals_6, buf6, primals_4, buf7
class ValueNetworkNew(nn.Module):
def __init__(self):
super(ValueNetworkNew, self).__init__()
self.fc1 = nn.Linear(4, 256)
self.fc2 = nn.Linear(256, 256)
self.fc3 = nn.Linear(256, 1)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
| DensoITLab/spinningup_in_pytorch | ValueNetwork | false | 7,962 | [
"MIT"
] | 11 | 612d8c4c6593c8c5ecb5a939bf43085daac9e552 | https://github.com/DensoITLab/spinningup_in_pytorch/tree/612d8c4c6593c8c5ecb5a939bf43085daac9e552 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(4, 256)
self.fc2 = nn.Linear(256, 256)
self.fc3 = nn.Linear(256, 1)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Sum | # 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_1/inductor_cache/kc/ckcqdc4e3rohzicuch2u6npffd7qsqteolddbvlph7otuvom45tp.py
# Topologically Sorted Source Nodes: [y_1, y_2, y_3], Original ATen: [aten.add]
# Source node to ATen node mapping:
# y_1 => add
# y_2 => add_1
# y_3 => add_2
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select, %select_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %select_2), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %select_3), kwargs = {})
triton_poi_fused_add_0 = async_compile.triton('triton_poi_fused_add_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (128 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (192 + x0), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tl.store(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), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [y_1, y_2, y_3], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(arg0_1, buf0, 64, grid=grid(64), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.onnx
class Sum(nn.Module):
def __init__(self, n, weight=False):
super().__init__()
self.weight = weight
self.iter = range(n - 1)
if weight:
self.w = nn.Parameter(-torch.arange(1.0, n) / 2, requires_grad=True
)
def forward(self, x):
y = x[0]
if self.weight:
w = torch.sigmoid(self.w) * 2
for i in self.iter:
y = y + x[i + 1] * w[i]
else:
for i in self.iter:
y = y + x[i + 1]
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n': 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.onnx
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + (64 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (128 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (192 + x0), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tl.store(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), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class SumNew(nn.Module):
def __init__(self, n, weight=False):
super().__init__()
self.weight = weight
self.iter = range(n - 1)
if weight:
self.w = nn.Parameter(-torch.arange(1.0, n) / 2, requires_grad=True
)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DDGRCF/YOLOX_OBB | Sum | false | 7,963 | [
"Apache-2.0"
] | 39 | 27b80953306492b8bc83b86b1353d8cee01ef9b6 | https://github.com/DDGRCF/YOLOX_OBB/tree/27b80953306492b8bc83b86b1353d8cee01ef9b6 | import torch
import torch.nn as nn
import torch.onnx
class Model(nn.Module):
def __init__(self, n, weight=False):
super().__init__()
self.weight = weight
self.iter = range(n - 1)
if weight:
self.w = nn.Parameter(-torch.arange(1.0, n) / 2, requires_grad=True
)
def forward(self, x):
y = x[0]
if self.weight:
w = torch.sigmoid(self.w) * 2
for i in self.iter:
y = y + x[i + 1] * w[i]
else:
for i in self.iter:
y = y + x[i + 1]
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
DenseSAGEConv | # 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_1/inductor_cache/ks/cksmzjpji5wopzx7ygvaueq5yanyhzrhgd53oqwifrxqardohxup.py
# Topologically Sorted Source Nodes: [adj], Original ATen: [aten.add]
# Source node to ATen node mapping:
# adj => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_2, %expand), 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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
x3 = xindex
x1 = (xindex // 4) % 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = x1
tmp2 = x0
tmp3 = tmp1 == tmp2
tmp4 = 1.0
tmp5 = 0.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 + tmp6
tl.store(out_ptr0 + (x3), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/tj/ctjellkfxbwog6naw6ljwd5qmjza64ibofheztgtji4ltt4sqwsy.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# out => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_2,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_1 = async_compile.triton('triton_poi_fused_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 64
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/s6/cs6py6vk32wnwfs7rofgbld4vhlb2ap67kti5z5vq5uxlm27cno6.py
# Topologically Sorted Source Nodes: [sum_1, out_1], Original ATen: [aten.sum, aten.div]
# Source node to ATen node mapping:
# out_1 => div
# sum_1 => sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%add, [-1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_2, %sum_1), kwargs = {})
triton_poi_fused_div_sum_2 = async_compile.triton('triton_poi_fused_div_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_sum_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_div_sum_2(in_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
x1 = (xindex // 4)
tmp0 = tl.load(in_out_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(in_out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zb/czbosywypt6ykzsg3qgk4xqpum6bbus7ewvx3a4isas4grv4ruv5.py
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.add, aten.linalg_vector_norm, aten.clamp_min]
# Source node to ATen node mapping:
# out_3 => add_1
# out_4 => clamp_min, pow_1, pow_2, sum_2
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_4, %primals_4), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_1, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [-1], True), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 0.5), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_2, 1e-12), kwargs = {})
triton_poi_fused_add_clamp_min_linalg_vector_norm_3 = async_compile.triton('triton_poi_fused_add_clamp_min_linalg_vector_norm_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_clamp_min_linalg_vector_norm_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_min_linalg_vector_norm_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp5 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr1 + (1))
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp11 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (2))
tmp13 = tl.broadcast_to(tmp12, [XBLOCK])
tmp17 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr1 + (3))
tmp19 = tl.broadcast_to(tmp18, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tmp3 * tmp3
tmp8 = tmp5 + tmp7
tmp9 = tmp8 * tmp8
tmp10 = tmp4 + tmp9
tmp14 = tmp11 + tmp13
tmp15 = tmp14 * tmp14
tmp16 = tmp10 + tmp15
tmp20 = tmp17 + tmp19
tmp21 = tmp20 * tmp20
tmp22 = tmp16 + tmp21
tmp23 = libdevice.sqrt(tmp22)
tmp24 = 1e-12
tmp25 = triton_helpers.maximum(tmp23, tmp24)
tl.store(out_ptr0 + (x0), tmp25, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ui/cuiinzjyexhjxr6gshfqf4tzx4bxt3ta5qu63fe7nrwn55g62ck7.py
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.add, aten.div]
# Source node to ATen node mapping:
# out_3 => add_1
# out_4 => div_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_4, %primals_4), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_1, %expand_3), kwargs = {})
triton_poi_fused_add_div_4 = async_compile.triton('triton_poi_fused_add_div_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: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_4(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 / tmp3
tl.store(out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_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: [adj], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(primals_2, buf0, 256, grid=grid(256), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(primals_1, buf1, 256, grid=grid(256), stream=stream0)
del primals_1
buf2 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf1, (16, 4, 4), (16, 4, 1), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [sum_1, out_1], Original ATen: [aten.sum, aten.div]
triton_poi_fused_div_sum_2.run(buf3, buf0, 256, grid=grid(256), stream=stream0)
buf4 = reinterpret_tensor(buf0, (64, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), primals_3, out=buf4)
del primals_3
buf5 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.add, aten.linalg_vector_norm, aten.clamp_min]
triton_poi_fused_add_clamp_min_linalg_vector_norm_3.run(buf4, primals_4, buf5, 64, grid=grid(64), stream=stream0)
buf6 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.add, aten.div]
triton_poi_fused_add_div_4.run(buf4, primals_4, buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
return (buf6, primals_4, buf4, reinterpret_tensor(buf3, (4, 64), (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, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (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 math
import torch
import torch.nn.functional as F
import torch.utils.data
from torch.nn import Parameter
def uniform(size, tensor):
stdv = 1.0 / math.sqrt(size)
if tensor is not None:
tensor.data.uniform_(-stdv, stdv)
class DenseSAGEConv(torch.nn.Module):
"""See :class:`torch_geometric.nn.conv.sage_conv.SAGEConv`.
:rtype: :class:`Tensor`
"""
def __init__(self, in_channels, out_channels, normalize=True, bias=True):
super(DenseSAGEConv, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.normalize = normalize
self.weight = Parameter(torch.Tensor(self.in_channels, out_channels))
if bias:
self.bias = Parameter(torch.Tensor(out_channels))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
uniform(self.in_channels, self.weight)
uniform(self.in_channels, self.bias)
def forward(self, x, adj, mask=None, add_loop=True):
"""
Args:
x (Tensor): Node feature tensor :math:`\\mathbf{X} \\in \\mathbb{R}^{B
\\times N \\times F}`, with batch-size :math:`B`, (maximum)
number of nodes :math:`N` for each graph, and feature
dimension :math:`F`.
adj (Tensor): Adjacency tensor :math:`\\mathbf{A} \\in \\mathbb{R}^{B
\\times N \\times N}`.
mask (ByteTensor, optional): Mask matrix
:math:`\\mathbf{M} \\in {\\{ 0, 1 \\}}^{B \\times N}` indicating
the valid nodes for each graph. (default: :obj:`None`)
add_loop (bool, optional): If set to :obj:`False`, the layer will
not automatically add self-loops to the adjacency matrices.
(default: :obj:`True`)
"""
x = x.unsqueeze(0) if x.dim() == 2 else x
adj = adj.unsqueeze(0) if adj.dim() == 2 else adj
B, N, _ = x.size()
if add_loop:
eye = torch.eye(N, dtype=adj.dtype, device=adj.device)
adj = adj + eye.unsqueeze(0).expand_as(adj)
out = torch.matmul(adj, x)
out = out / adj.sum(dim=-1, keepdim=True)
out = torch.matmul(out, self.weight)
if self.bias is not None:
out = out + self.bias
if self.normalize:
out = F.normalize(out, p=2, dim=-1)
if mask is not None:
mask = mask.view(B, N, 1)
out = out * mask
return out
def __repr__(self):
return '{}({}, {})'.format(self.__class__.__name__, self.
in_channels, self.out_channels)
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import math
import torch.utils.data
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_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
x3 = xindex
x1 = xindex // 4 % 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = x1
tmp2 = x0
tmp3 = tmp1 == tmp2
tmp4 = 1.0
tmp5 = 0.0
tmp6 = tl.where(tmp3, tmp4, tmp5)
tmp7 = tmp0 + tmp6
tl.store(out_ptr0 + x3, tmp7, xmask)
@triton.jit
def triton_poi_fused_clone_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 % 64
x2 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_div_sum_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_out_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(in_out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_add_clamp_min_linalg_vector_norm_3(in_ptr0, in_ptr1,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp5 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr1 + 1)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp11 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + 2)
tmp13 = tl.broadcast_to(tmp12, [XBLOCK])
tmp17 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr1 + 3)
tmp19 = tl.broadcast_to(tmp18, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tmp3 * tmp3
tmp8 = tmp5 + tmp7
tmp9 = tmp8 * tmp8
tmp10 = tmp4 + tmp9
tmp14 = tmp11 + tmp13
tmp15 = tmp14 * tmp14
tmp16 = tmp10 + tmp15
tmp20 = tmp17 + tmp19
tmp21 = tmp20 * tmp20
tmp22 = tmp16 + tmp21
tmp23 = libdevice.sqrt(tmp22)
tmp24 = 1e-12
tmp25 = triton_helpers.maximum(tmp23, tmp24)
tl.store(out_ptr0 + x0, tmp25, xmask)
@triton.jit
def triton_poi_fused_add_div_4(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 / tmp3
tl.store(out_ptr0 + x2, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_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_0[grid(256)](primals_2, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_1[grid(256)](primals_1, buf1, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_1
buf2 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf1, (16, 4, 4), (16, 4, 1), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
triton_poi_fused_div_sum_2[grid(256)](buf3, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf4 = reinterpret_tensor(buf0, (64, 4), (4, 1), 0)
del buf0
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
primals_3, out=buf4)
del primals_3
buf5 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused_add_clamp_min_linalg_vector_norm_3[grid(64)](buf4,
primals_4, buf5, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf6 = buf1
del buf1
triton_poi_fused_add_div_4[grid(256)](buf4, primals_4, buf5, buf6,
256, XBLOCK=256, num_warps=4, num_stages=1)
del buf5
return buf6, primals_4, buf4, reinterpret_tensor(buf3, (4, 64), (1, 4), 0)
def uniform(size, tensor):
stdv = 1.0 / math.sqrt(size)
if tensor is not None:
tensor.data.uniform_(-stdv, stdv)
class DenseSAGEConvNew(torch.nn.Module):
"""See :class:`torch_geometric.nn.conv.sage_conv.SAGEConv`.
:rtype: :class:`Tensor`
"""
def __init__(self, in_channels, out_channels, normalize=True, bias=True):
super(DenseSAGEConvNew, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.normalize = normalize
self.weight = Parameter(torch.Tensor(self.in_channels, out_channels))
if bias:
self.bias = Parameter(torch.Tensor(out_channels))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
uniform(self.in_channels, self.weight)
uniform(self.in_channels, self.bias)
def __repr__(self):
return '{}({}, {})'.format(self.__class__.__name__, self.
in_channels, self.out_channels)
def forward(self, input_0, input_1):
primals_3 = self.weight
primals_4 = self.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| Cyanogenoid/fspool | DenseSAGEConv | false | 7,964 | [
"MIT"
] | 41 | 7525cb17992ec7a1bb7f92996c2b31a65aa8eba2 | https://github.com/Cyanogenoid/fspool/tree/7525cb17992ec7a1bb7f92996c2b31a65aa8eba2 | import math
import torch
import torch.nn.functional as F
import torch.utils.data
from torch.nn import Parameter
def uniform(size, tensor):
stdv = 1.0 / math.sqrt(size)
if tensor is not None:
tensor.data.uniform_(-stdv, stdv)
class Model(torch.nn.Module):
"""See :class:`torch_geometric.nn.conv.sage_conv.SAGEConv`.
:rtype: :class:`Tensor`
"""
def __init__(self, in_channels, out_channels, normalize=True, bias=True):
super().__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.normalize = normalize
self.weight = Parameter(torch.Tensor(self.in_channels, out_channels))
if bias:
self.bias = Parameter(torch.Tensor(out_channels))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
uniform(self.in_channels, self.weight)
uniform(self.in_channels, self.bias)
def forward(self, x, adj, mask=None, add_loop=True):
"""
Args:
x (Tensor): Node feature tensor :math:`\\mathbf{X} \\in \\mathbb{R}^{B
\\times N \\times F}`, with batch-size :math:`B`, (maximum)
number of nodes :math:`N` for each graph, and feature
dimension :math:`F`.
adj (Tensor): Adjacency tensor :math:`\\mathbf{A} \\in \\mathbb{R}^{B
\\times N \\times N}`.
mask (ByteTensor, optional): Mask matrix
:math:`\\mathbf{M} \\in {\\{ 0, 1 \\}}^{B \\times N}` indicating
the valid nodes for each graph. (default: :obj:`None`)
add_loop (bool, optional): If set to :obj:`False`, the layer will
not automatically add self-loops to the adjacency matrices.
(default: :obj:`True`)
"""
x = x.unsqueeze(0) if x.dim() == 2 else x
adj = adj.unsqueeze(0) if adj.dim() == 2 else adj
B, N, _ = x.size()
if add_loop:
eye = torch.eye(N, dtype=adj.dtype, device=adj.device)
adj = adj + eye.unsqueeze(0).expand_as(adj)
out = torch.matmul(adj, x)
out = out / adj.sum(dim=-1, keepdim=True)
out = torch.matmul(out, self.weight)
if self.bias is not None:
out = out + self.bias
if self.normalize:
out = F.normalize(out, p=2, dim=-1)
if mask is not None:
mask = mask.view(B, N, 1)
out = out * mask
return out
def __repr__(self):
return '{}({}, {})'.format(self.__class__.__name__, self.
in_channels, self.out_channels)
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
MultiplicationComposition | # 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_1/inductor_cache/dv/cdvmujdx4wzp3xf3q2rdnt72xso7clbkadprbhlirztwjsalq3rx.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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
from torch import nn
from abc import abstractmethod
import torch.utils.data
class Composition(nn.Module):
"""A base class for compositions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compose two batches vectors.
.. note ::
The two batches have to be of broadcastable shape.
:param x: shape: s_x
The first batch of vectors.
:param y: shape: s_y
The second batch of vectors.
:return: shape: s
The compositionm, where `s` is the broadcasted shape.
"""
raise NotImplementedError
class MultiplicationComposition(Composition):
"""Element-wise multiplication, a.k.a. Hadamard product."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
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 import nn
from abc import abstractmethod
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, 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 Composition(nn.Module):
"""A base class for compositions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compose two batches vectors.
.. note ::
The two batches have to be of broadcastable shape.
:param x: shape: s_x
The first batch of vectors.
:param y: shape: s_y
The second batch of vectors.
:return: shape: s
The compositionm, where `s` is the broadcasted shape.
"""
raise NotImplementedError
class MultiplicationCompositionNew(Composition):
"""Element-wise multiplication, a.k.a. Hadamard product."""
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DimitrisAlivas/StarQE | MultiplicationComposition | false | 7,965 | [
"MIT"
] | 11 | c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | https://github.com/DimitrisAlivas/StarQE/tree/c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | import torch
from torch import nn
from abc import abstractmethod
import torch.utils.data
class Composition(nn.Module):
"""A base class for compositions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compose two batches vectors.
.. note ::
The two batches have to be of broadcastable shape.
:param x: shape: s_x
The first batch of vectors.
:param y: shape: s_y
The second batch of vectors.
:return: shape: s
The compositionm, where `s` is the broadcasted shape.
"""
raise NotImplementedError
class Model(Composition):
"""Element-wise multiplication, a.k.a. Hadamard product."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return x * y
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Cutout | # 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_1/inductor_cache/d4/cd4j5qccwplmyca4cjdyf6n335mujnfm6mo5qx4t4j34mw3oyfr6.py
# Topologically Sorted Source Nodes: [mask, setitem, mul], Original ATen: [aten.ones_like, aten.lift_fresh, aten.fill, aten.mul]
# Source node to ATen node mapping:
# mask => full_default
# mul => mul
# setitem => copy, full_default_1
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%full_default, %full_default_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %copy), kwargs = {})
triton_poi_fused_fill_lift_fresh_mul_ones_like_0 = async_compile.triton('triton_poi_fused_fill_lift_fresh_mul_ones_like_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_fill_lift_fresh_mul_ones_like_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_fill_lift_fresh_mul_ones_like_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mask, setitem, mul], Original ATen: [aten.ones_like, aten.lift_fresh, aten.fill, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_fill_lift_fresh_mul_ones_like_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 random
import torch
def _gen_cutout_coord(height, width, size):
height_loc = random.randint(0, height - 1)
width_loc = random.randint(0, width - 1)
upper_coord = max(0, height_loc - size // 2), max(0, width_loc - size // 2)
lower_coord = min(height, height_loc + size // 2), min(width, width_loc +
size // 2)
return upper_coord, lower_coord
class Cutout(torch.nn.Module):
def __init__(self, size=16):
super().__init__()
self.size = size
def forward(self, img):
h, w = img.shape[-2:]
upper_coord, lower_coord = _gen_cutout_coord(h, w, self.size)
mask_height = lower_coord[0] - upper_coord[0]
mask_width = lower_coord[1] - upper_coord[1]
assert mask_height > 0
assert mask_width > 0
mask = torch.ones_like(img)
mask[..., upper_coord[0]:lower_coord[0], upper_coord[1]:lower_coord[1]
] = 0
return img * mask
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 random
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_fill_lift_fresh_mul_ones_like_0(in_ptr0, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_fill_lift_fresh_mul_ones_like_0[grid(256)](arg0_1,
buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
def _gen_cutout_coord(height, width, size):
height_loc = random.randint(0, height - 1)
width_loc = random.randint(0, width - 1)
upper_coord = max(0, height_loc - size // 2), max(0, width_loc - size // 2)
lower_coord = min(height, height_loc + size // 2), min(width, width_loc +
size // 2)
return upper_coord, lower_coord
class CutoutNew(torch.nn.Module):
def __init__(self, size=16):
super().__init__()
self.size = size
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DensoITLab/TeachAugment | Cutout | false | 7,966 | [
"BSD-2-Clause"
] | 20 | 66ec099a0afab99e18531c5437182cfe17dc30c8 | https://github.com/DensoITLab/TeachAugment/tree/66ec099a0afab99e18531c5437182cfe17dc30c8 | import random
import torch
def _gen_cutout_coord(height, width, size):
height_loc = random.randint(0, height - 1)
width_loc = random.randint(0, width - 1)
upper_coord = max(0, height_loc - size // 2), max(0, width_loc - size // 2)
lower_coord = min(height, height_loc + size // 2), min(width, width_loc +
size // 2)
return upper_coord, lower_coord
class Model(torch.nn.Module):
def __init__(self, size=16):
super().__init__()
self.size = size
def forward(self, img):
h, w = img.shape[-2:]
upper_coord, lower_coord = _gen_cutout_coord(h, w, self.size)
mask_height = lower_coord[0] - upper_coord[0]
mask_width = lower_coord[1] - upper_coord[1]
assert mask_height > 0
assert mask_width > 0
mask = torch.ones_like(img)
mask[..., upper_coord[0]:lower_coord[0], upper_coord[1]:lower_coord[1]
] = 0
return img * mask
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GlobalSelfAttention | # 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_1/inductor_cache/wz/cwzlgmghy6nxuchbiog4puo46i4tq7yhd3qu6ftkgjf3gwib6hxn.py
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# align_vectors => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
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_1/inductor_cache/yh/cyhf6bhaqimi2pucos5fnrpvhrt4vuaetbxnooyr5pvgjt7s6fgo.py
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# align_vectors => 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_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ph/cph7lssg6ok5pn3rbmbprurourgd72q6n57v7b4qu2gqs3wcxbt7.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 = ([%bmm_1, %primals_1], 2), 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=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + ((4*x1) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/25/c25xc4wf5xgns46eqiqtzdlbw736adueljtcx7ngbvo6jrcznpv2.py
# Topologically Sorted Source Nodes: [attn_h_2], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# attn_h_2 => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_2,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_3 = async_compile.triton('triton_poi_fused_clone_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp2 = tl.load(in_ptr1 + (x3), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x0 + (4*x2) + (16*x1)), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/sx/csxt2z4nihjsbxcckhpeiwgke5usifqk6ao5saydggkozmd2prcm.py
# Topologically Sorted Source Nodes: [align_vectors_2], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# align_vectors_2 => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_3,), 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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1)), xmask)
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 8), (8, 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: [align], Original ATen: [aten.bmm]
extern_kernels.bmm(primals_1, reinterpret_tensor(primals_2, (4, 4, 4), (16, 1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = reinterpret_tensor(buf0, (16, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [align_vectors], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [c], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf2, (4, 4, 4), (16, 4, 1), 0), primals_2, out=buf3)
del primals_2
buf4 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
triton_poi_fused_cat_2.run(buf3, primals_1, buf4, 128, grid=grid(128), stream=stream0)
buf5 = reinterpret_tensor(buf3, (16, 4), (4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf4, (16, 8), (8, 1), 0), reinterpret_tensor(primals_3, (8, 4), (1, 8), 0), out=buf5)
del primals_3
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_h_2], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf5, primals_1, buf6, 64, grid=grid(64), stream=stream0)
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [align_vectors_2], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf2, buf7, 64, grid=grid(64), stream=stream0)
del buf2
return (buf6, buf7, primals_1, reinterpret_tensor(buf4, (16, 8), (8, 1), 0), buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 8), (8, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.cuda
import torch.distributed
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class GlobalSelfAttention(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax', attn_hidden=0):
super(GlobalSelfAttention, self).__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
self.attn_hidden = attn_hidden
if attn_hidden > 0:
self.transform_in = nn.Sequential(nn.Linear(dim, attn_hidden),
nn.ELU(0.1))
if self.attn_type == 'general':
d = attn_hidden if attn_hidden > 0 else dim
self.linear_in = nn.Linear(d, d, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
out_bias = self.attn_type == 'mlp'
self.linear_out = nn.Linear(dim * 2, dim, bias=out_bias)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_hidden > 0:
h_t = self.transform_in(h_t)
h_s = self.transform_in(h_s)
tgt_dim = self.attn_hidden
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view(tgt_batch, tgt_len, 1, dim)
wq = wq.expand(tgt_batch, tgt_len, src_len, dim)
uh = self.linear_context(h_s.contiguous().view(-1, dim))
uh = uh.view(src_batch, 1, src_len, dim)
uh = uh.expand(src_batch, tgt_len, src_len, dim)
wquh = torch.tanh(wq + uh)
return self.v(wquh.view(-1, dim)).view(tgt_batch, tgt_len, src_len)
def forward(self, source, memory_bank, memory_lengths=None, coverage=None):
"""
Args:
source (FloatTensor): query vectors ``(batch, tgt_len, dim)``
memory_bank (FloatTensor): source vectors ``(batch, src_len, dim)``
memory_lengths (LongTensor): the source context lengths ``(batch,)``
coverage (FloatTensor): None (not supported yet)
Returns:
(FloatTensor, FloatTensor):
* Computed vector ``(tgt_len, batch, dim)``
* Attention distribtutions for each query
``(tgt_len, batch, src_len)``
"""
if source.dim() == 2:
one_step = True
source = source.unsqueeze(1)
else:
one_step = False
batch, source_l, dim = memory_bank.size()
batch_, target_l, dim_ = source.size()
aeq(batch, batch_)
aeq(dim, dim_)
aeq(self.dim, dim)
if coverage is not None:
batch_, source_l_ = coverage.size()
aeq(batch, batch_)
aeq(source_l, source_l_)
if coverage is not None:
cover = coverage.view(-1).unsqueeze(1)
memory_bank += self.linear_cover(cover).view_as(memory_bank)
memory_bank = torch.tanh(memory_bank)
align = self.score(source, memory_bank)
if memory_lengths is not None:
mask = sequence_mask(memory_lengths, max_len=align.size(-1))
mask = mask.unsqueeze(1)
mask = mask.repeat(1, source_l, 1)
mask_self_index = list(range(source_l))
mask[:, mask_self_index, mask_self_index] = 0
align.masked_fill_(~mask, -float('inf'))
if self.attn_func == 'softmax':
align_vectors = F.softmax(align.view(batch * target_l, source_l
), -1)
else:
align_vectors = sparsemax(align.view(batch * target_l, source_l
), -1)
align_vectors = align_vectors.view(batch, target_l, source_l)
c = torch.bmm(align_vectors, memory_bank)
concat_c = torch.cat([c, source], 2).view(batch * target_l, dim * 2)
attn_h = self.linear_out(concat_c).view(batch, target_l, dim)
if self.attn_type in ['general', 'dot']:
attn_h = F.sigmoid(attn_h).mul(source)
if one_step:
attn_h = attn_h.squeeze(1)
align_vectors = align_vectors.squeeze(1)
batch_, dim_ = attn_h.size()
aeq(batch, batch_)
aeq(dim, dim_)
batch_, source_l_ = align_vectors.size()
aeq(batch, batch_)
aeq(source_l, source_l_)
else:
attn_h = attn_h.transpose(0, 1).contiguous()
align_vectors = align_vectors.transpose(0, 1).contiguous()
target_l_, batch_, dim_ = attn_h.size()
aeq(target_l, target_l_)
aeq(batch, batch_)
aeq(dim, dim_)
target_l_, batch_, source_l_ = align_vectors.size()
aeq(target_l, target_l_)
aeq(batch, batch_)
aeq(source_l, source_l_)
return attn_h, align_vectors
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.cuda
import torch.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_cat_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = xindex // 8
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (4 * x1 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_clone_3(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
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp2 = tl.load(in_ptr1 + x3, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x0 + 4 * x2 + 16 * x1), tmp3, xmask)
@triton.jit
def triton_poi_fused_clone_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
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1), xmask)
tl.store(out_ptr0 + x3, tmp0, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 8), (8, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(primals_1, reinterpret_tensor(primals_2, (4, 4,
4), (16, 1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(64)](buf0, buf1, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf2 = reinterpret_tensor(buf0, (16, 4), (4, 1), 0)
del buf0
triton_poi_fused__softmax_1[grid(64)](buf1, buf2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf3 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0)
del buf1
extern_kernels.bmm(reinterpret_tensor(buf2, (4, 4, 4), (16, 4, 1),
0), primals_2, out=buf3)
del primals_2
buf4 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
triton_poi_fused_cat_2[grid(128)](buf3, primals_1, buf4, 128,
XBLOCK=128, num_warps=4, num_stages=1)
buf5 = reinterpret_tensor(buf3, (16, 4), (4, 1), 0)
del buf3
extern_kernels.mm(reinterpret_tensor(buf4, (16, 8), (8, 1), 0),
reinterpret_tensor(primals_3, (8, 4), (1, 8), 0), out=buf5)
del primals_3
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clone_3[grid(64)](buf5, primals_1, buf6, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clone_4[grid(64)](buf2, buf7, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf2
return buf6, buf7, primals_1, reinterpret_tensor(buf4, (16, 8), (8, 1), 0
), buf5
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class GlobalSelfAttentionNew(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax', attn_hidden=0):
super(GlobalSelfAttentionNew, self).__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
self.attn_hidden = attn_hidden
if attn_hidden > 0:
self.transform_in = nn.Sequential(nn.Linear(dim, attn_hidden),
nn.ELU(0.1))
if self.attn_type == 'general':
d = attn_hidden if attn_hidden > 0 else dim
self.linear_in = nn.Linear(d, d, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
out_bias = self.attn_type == 'mlp'
self.linear_out = nn.Linear(dim * 2, dim, bias=out_bias)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_hidden > 0:
h_t = self.transform_in(h_t)
h_s = self.transform_in(h_s)
tgt_dim = self.attn_hidden
if self.attn_type == 'general':
h_t_ = h_t.view(tgt_batch * tgt_len, tgt_dim)
h_t_ = self.linear_in(h_t_)
h_t = h_t_.view(tgt_batch, tgt_len, tgt_dim)
h_s_ = h_s.transpose(1, 2)
return torch.bmm(h_t, h_s_)
else:
dim = self.dim
wq = self.linear_query(h_t.view(-1, dim))
wq = wq.view(tgt_batch, tgt_len, 1, dim)
wq = wq.expand(tgt_batch, tgt_len, src_len, dim)
uh = self.linear_context(h_s.contiguous().view(-1, dim))
uh = uh.view(src_batch, 1, src_len, dim)
uh = uh.expand(src_batch, tgt_len, src_len, dim)
wquh = torch.tanh(wq + uh)
return self.v(wquh.view(-1, dim)).view(tgt_batch, tgt_len, src_len)
def forward(self, input_0, input_1):
primals_3 = self.linear_out.weight
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3])
return output[0], output[1]
| DenDen047/data2text-macro-plan-py | GlobalSelfAttention | false | 7,967 | [
"MIT"
] | 20 | bb01ec6e23dab28c1e969f23bd55776b597fb995 | https://github.com/DenDen047/data2text-macro-plan-py/tree/bb01ec6e23dab28c1e969f23bd55776b597fb995 | import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.cuda
import torch.distributed
def aeq(*args):
"""
Assert all arguments have the same value
"""
arguments = (arg for arg in args)
first = next(arguments)
assert all(arg == first for arg in arguments
), 'Not all arguments have the same value: ' + str(args)
def sequence_mask(lengths, max_len=None):
"""
Creates a boolean mask from sequence lengths.
"""
batch_size = lengths.numel()
max_len = max_len or lengths.max()
return torch.arange(0, max_len, device=lengths.device).type_as(lengths
).repeat(batch_size, 1).lt(lengths.unsqueeze(1))
class Model(nn.Module):
"""
Global attention takes a matrix and a query vector. It
then computes a parameterized convex combination of the matrix
based on the input query.
Constructs a unit mapping a query `q` of size `dim`
and a source matrix `H` of size `n x dim`, to an output
of size `dim`.
.. mermaid::
graph BT
A[Query]
subgraph RNN
C[H 1]
D[H 2]
E[H N]
end
F[Attn]
G[Output]
A --> F
C --> F
D --> F
E --> F
C -.-> G
D -.-> G
E -.-> G
F --> G
All models compute the output as
:math:`c = \\sum_{j=1}^{\\text{SeqLength}} a_j H_j` where
:math:`a_j` is the softmax of a score function.
Then then apply a projection layer to [q, c].
However they
differ on how they compute the attention score.
* Luong Attention (dot, general):
* dot: :math:`\\text{score}(H_j,q) = H_j^T q`
* general: :math:`\\text{score}(H_j, q) = H_j^T W_a q`
* Bahdanau Attention (mlp):
* :math:`\\text{score}(H_j, q) = v_a^T \\text{tanh}(W_a q + U_a h_j)`
Args:
dim (int): dimensionality of query and key
coverage (bool): use coverage term
attn_type (str): type of attention to use, options [dot,general,mlp]
attn_func (str): attention function to use, options [softmax,sparsemax]
"""
def __init__(self, dim, coverage=False, attn_type='dot', attn_func=
'softmax', attn_hidden=0):
super().__init__()
self.dim = dim
assert attn_type in ['dot', 'general', 'mlp'
], 'Please select a valid attention type (got {:s}).'.format(
attn_type)
self.attn_type = attn_type
assert attn_func in ['softmax', 'sparsemax'
], 'Please select a valid attention function.'
self.attn_func = attn_func
self.attn_hidden = attn_hidden
if attn_hidden > 0:
self.transform_in = nn.Sequential(nn.Linear(dim, attn_hidden),
nn.ELU(0.1))
if self.attn_type == 'general':
d = attn_hidden if attn_hidden > 0 else dim
self.linear_in = nn.Linear(d, d, bias=False)
elif self.attn_type == 'mlp':
self.linear_context = nn.Linear(dim, dim, bias=False)
self.linear_query = nn.Linear(dim, dim, bias=True)
self.v = nn.Linear(dim, 1, bias=False)
out_bias = self.attn_type == 'mlp'
self.linear_out = nn.Linear(dim * 2, dim, bias=out_bias)
if coverage:
self.linear_cover = nn.Linear(1, dim, bias=False)
def score(self, h_t, h_s):
"""
Args:
h_t (FloatTensor): sequence of queries ``(batch, tgt_len, dim)``
h_s (FloatTensor): sequence of sources ``(batch, src_len, dim``
Returns:
FloatTensor: raw attention scores (unnormalized) for each src index
``(batch, tgt_len, src_len)``
"""
src_batch, src_len, src_dim = h_s.size()
tgt_batch, tgt_len, tgt_dim = h_t.size()
aeq(src_batch, tgt_batch)
aeq(src_dim, tgt_dim)
aeq(self.dim, src_dim)
if self.attn_type in ['general', 'dot']:
if self.attn_hidden
# ... truncated (>4000 chars) for memory efficiency |
AverageAttention | # 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_1/inductor_cache/q2/cq22izcnzuual6ccs7wsaoh7bd5alf6ddh6n7insrqrakgdkqdid.py
# Topologically Sorted Source Nodes: [ones, triangle, mask], Original ATen: [aten.ones, aten.tril, aten.mul]
# Source node to ATen node mapping:
# mask => mul_1
# ones => full_default
# triangle => full_default_1, le, sub, where
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%unsqueeze, %unsqueeze_1), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%sub, 0), kwargs = {})
# %full_default_1 : [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 = (%le, %full_default, %full_default_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where, %permute), kwargs = {})
triton_poi_fused_mul_ones_tril_0 = async_compile.triton('triton_poi_fused_mul_ones_tril_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'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_ones_tril_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_ones_tril_0(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
x1 = (xindex // 4)
x2 = xindex
tmp0 = x0 + ((-1)*x1)
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 <= tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = 1 + x1
tmp7 = tmp6.to(tl.float32)
tmp8 = tmp3 / tmp7
tmp9 = tmp5 * tmp8
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/y7/cy7tcpwbeg52c2ad3m5ifhfcq4knlp4gckg4theqnizar4dm7g75.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_1, %bmm], -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=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + ((4*x1) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/v6/cv6akht44uwlvnqzu4nztxj4gv7hktsqrt77wlzb42fe4llua6lm.py
# Topologically Sorted Source Nodes: [sigmoid, mul_1, sigmoid_1, mul_2, gating_outputs_1], Original ATen: [aten.sigmoid, aten.mul, aten.add, aten.sigmoid_backward]
# Source node to ATen node mapping:
# gating_outputs_1 => add_1
# mul_1 => mul_2
# mul_2 => mul_3
# sigmoid => sigmoid
# sigmoid_1 => sigmoid_1
# Graph fragment:
# %sigmoid : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %primals_1), kwargs = {})
# %sigmoid_1 : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem_1,), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid_1, %bmm), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %mul_3), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid_1), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid_1, %sub_1), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %mul_8 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %sub_2), kwargs = {})
triton_poi_fused_add_mul_sigmoid_sigmoid_backward_2 = async_compile.triton('triton_poi_fused_add_mul_sigmoid_sigmoid_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=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mul_sigmoid_sigmoid_backward_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_sigmoid_backward_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (8*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + (x2), xmask)
tmp6 = tl.load(in_ptr0 + (4 + x0 + (8*x1)), xmask)
tmp7 = tl.load(in_ptr1 + (4 + x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr3 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tmp5 = tmp3 * tmp4
tmp8 = tmp6 + tmp7
tmp9 = tl.sigmoid(tmp8)
tmp11 = tmp9 * tmp10
tmp12 = tmp5 + tmp11
tmp13 = 1.0
tmp14 = tmp13 - tmp9
tmp15 = tmp9 * tmp14
tmp16 = tmp13 - tmp3
tmp17 = tmp3 * tmp16
tl.store(out_ptr0 + (x2), tmp12, xmask)
tl.store(out_ptr1 + (x2), tmp15, xmask)
tl.store(out_ptr2 + (x2), tmp17, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (8, 8), (8, 1))
assert_size_stride(primals_3, (8, ), (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: [ones, triangle, mask], Original ATen: [aten.ones, aten.tril, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_ones_tril_0.run(buf0, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [average_outputs], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (4, 4, 4), (0, 4, 1), 0), primals_1, out=buf1)
del buf0
buf2 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
triton_poi_fused_cat_1.run(primals_1, buf1, buf2, 128, grid=grid(128), stream=stream0)
buf3 = empty_strided_cuda((16, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf2, (16, 8), (8, 1), 0), reinterpret_tensor(primals_2, (8, 8), (1, 8), 0), out=buf3)
del primals_2
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sigmoid, mul_1, sigmoid_1, mul_2, gating_outputs_1], Original ATen: [aten.sigmoid, aten.mul, aten.add, aten.sigmoid_backward]
triton_poi_fused_add_mul_sigmoid_sigmoid_backward_2.run(buf3, primals_3, primals_1, buf1, buf4, buf5, buf6, 64, grid=grid(64), stream=stream0)
del buf3
del primals_3
return (buf4, buf1, primals_1, buf1, reinterpret_tensor(buf2, (16, 8), (8, 1), 0), buf5, 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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((8, 8), (8, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((8, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
class PositionwiseFeedForward(nn.Module):
""" A two-layer Feed-Forward-Network with residual layer norm.
Args:
d_model (int): the size of input for the first-layer of the FFN.
d_ff (int): the hidden layer size of the second-layer
of the FNN.
dropout (float): dropout probability in :math:`[0, 1)`.
"""
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.layer_norm = nn.LayerNorm(d_model, eps=1e-06)
self.dropout_1 = nn.Dropout(dropout)
self.relu = nn.ReLU()
self.dropout_2 = nn.Dropout(dropout)
def forward(self, x):
"""Layer definition.
Args:
x: ``(batch_size, input_len, model_dim)``
Returns:
(FloatTensor): Output ``(batch_size, input_len, model_dim)``.
"""
inter = self.dropout_1(self.relu(self.w_1(self.layer_norm(x))))
output = self.dropout_2(self.w_2(inter))
return output + x
def update_dropout(self, dropout):
self.dropout_1.p = dropout
self.dropout_2.p = dropout
class AverageAttention(nn.Module):
"""
Average Attention module from
"Accelerating Neural Transformer via an Average Attention Network"
:cite:`DBLP:journals/corr/abs-1805-00631`.
Args:
model_dim (int): the dimension of keys/values/queries,
must be divisible by head_count
dropout (float): dropout parameter
"""
def __init__(self, model_dim, dropout=0.1, aan_useffn=False):
self.model_dim = model_dim
self.aan_useffn = aan_useffn
super(AverageAttention, self).__init__()
if aan_useffn:
self.average_layer = PositionwiseFeedForward(model_dim,
model_dim, dropout)
self.gating_layer = nn.Linear(model_dim * 2, model_dim * 2)
def cumulative_average_mask(self, batch_size, inputs_len, device):
"""
Builds the mask to compute the cumulative average as described in
:cite:`DBLP:journals/corr/abs-1805-00631` -- Figure 3
Args:
batch_size (int): batch size
inputs_len (int): length of the inputs
Returns:
(FloatTensor):
* A Tensor of shape ``(batch_size, input_len, input_len)``
"""
triangle = torch.tril(torch.ones(inputs_len, inputs_len, dtype=
torch.float, device=device))
weights = torch.ones(1, inputs_len, dtype=torch.float, device=device
) / torch.arange(1, inputs_len + 1, dtype=torch.float, device=
device)
mask = triangle * weights.transpose(0, 1)
return mask.unsqueeze(0).expand(batch_size, inputs_len, inputs_len)
def cumulative_average(self, inputs, mask_or_step, layer_cache=None,
step=None):
"""
Computes the cumulative average as described in
:cite:`DBLP:journals/corr/abs-1805-00631` -- Equations (1) (5) (6)
Args:
inputs (FloatTensor): sequence to average
``(batch_size, input_len, dimension)``
mask_or_step: if cache is set, this is assumed
to be the current step of the
dynamic decoding. Otherwise, it is the mask matrix
used to compute the cumulative average.
layer_cache: a dictionary containing the cumulative average
of the previous step.
Returns:
a tensor of the same shape and type as ``inputs``.
"""
if layer_cache is not None:
step = mask_or_step
average_attention = (inputs + step * layer_cache['prev_g']) / (step
+ 1)
layer_cache['prev_g'] = average_attention
return average_attention
else:
mask = mask_or_step
return torch.matmul(mask, inputs)
def forward(self, inputs, mask=None, layer_cache=None, step=None):
"""
Args:
inputs (FloatTensor): ``(batch_size, input_len, model_dim)``
Returns:
(FloatTensor, FloatTensor):
* gating_outputs ``(batch_size, input_len, model_dim)``
* average_outputs average attention
``(batch_size, input_len, model_dim)``
"""
batch_size = inputs.size(0)
inputs_len = inputs.size(1)
average_outputs = self.cumulative_average(inputs, self.
cumulative_average_mask(batch_size, inputs_len, inputs.device) if
layer_cache is None else step, layer_cache=layer_cache)
if self.aan_useffn:
average_outputs = self.average_layer(average_outputs)
gating_outputs = self.gating_layer(torch.cat((inputs,
average_outputs), -1))
input_gate, forget_gate = torch.chunk(gating_outputs, 2, dim=2)
gating_outputs = torch.sigmoid(input_gate) * inputs + torch.sigmoid(
forget_gate) * average_outputs
return gating_outputs, average_outputs
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'model_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.cuda
import torch.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_mul_ones_tril_0(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
x1 = xindex // 4
x2 = xindex
tmp0 = x0 + -1 * x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 <= tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = 1 + x1
tmp7 = tmp6.to(tl.float32)
tmp8 = tmp3 / tmp7
tmp9 = tmp5 * tmp8
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused_cat_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = xindex // 8
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (4 * x1 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_mul_sigmoid_sigmoid_backward_2(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 8 * x1), xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr2 + x2, xmask)
tmp6 = tl.load(in_ptr0 + (4 + x0 + 8 * x1), xmask)
tmp7 = tl.load(in_ptr1 + (4 + x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr3 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tmp5 = tmp3 * tmp4
tmp8 = tmp6 + tmp7
tmp9 = tl.sigmoid(tmp8)
tmp11 = tmp9 * tmp10
tmp12 = tmp5 + tmp11
tmp13 = 1.0
tmp14 = tmp13 - tmp9
tmp15 = tmp9 * tmp14
tmp16 = tmp13 - tmp3
tmp17 = tmp3 * tmp16
tl.store(out_ptr0 + x2, tmp12, xmask)
tl.store(out_ptr1 + x2, tmp15, xmask)
tl.store(out_ptr2 + x2, tmp17, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (8, 8), (8, 1))
assert_size_stride(primals_3, (8,), (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_ones_tril_0[grid(16)](buf0, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (4, 4, 4), (0, 4, 1), 0
), primals_1, out=buf1)
del buf0
buf2 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
triton_poi_fused_cat_1[grid(128)](primals_1, buf1, buf2, 128,
XBLOCK=128, num_warps=4, num_stages=1)
buf3 = empty_strided_cuda((16, 8), (8, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 8), (8, 1), 0),
reinterpret_tensor(primals_2, (8, 8), (1, 8), 0), out=buf3)
del primals_2
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_mul_sigmoid_sigmoid_backward_2[grid(64)](buf3,
primals_3, primals_1, buf1, buf4, buf5, buf6, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf3
del primals_3
return buf4, buf1, primals_1, buf1, reinterpret_tensor(buf2, (16, 8), (
8, 1), 0), buf5, buf6
class PositionwiseFeedForward(nn.Module):
""" A two-layer Feed-Forward-Network with residual layer norm.
Args:
d_model (int): the size of input for the first-layer of the FFN.
d_ff (int): the hidden layer size of the second-layer
of the FNN.
dropout (float): dropout probability in :math:`[0, 1)`.
"""
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.layer_norm = nn.LayerNorm(d_model, eps=1e-06)
self.dropout_1 = nn.Dropout(dropout)
self.relu = nn.ReLU()
self.dropout_2 = nn.Dropout(dropout)
def forward(self, x):
"""Layer definition.
Args:
x: ``(batch_size, input_len, model_dim)``
Returns:
(FloatTensor): Output ``(batch_size, input_len, model_dim)``.
"""
inter = self.dropout_1(self.relu(self.w_1(self.layer_norm(x))))
output = self.dropout_2(self.w_2(inter))
return output + x
def update_dropout(self, dropout):
self.dropout_1.p = dropout
self.dropout_2.p = dropout
class AverageAttentionNew(nn.Module):
"""
Average Attention module from
"Accelerating Neural Transformer via an Average Attention Network"
:cite:`DBLP:journals/corr/abs-1805-00631`.
Args:
model_dim (int): the dimension of keys/values/queries,
must be divisible by head_count
dropout (float): dropout parameter
"""
def __init__(self, model_dim, dropout=0.1, aan_useffn=False):
self.model_dim = model_dim
self.aan_useffn = aan_useffn
super(AverageAttentionNew, self).__init__()
if aan_useffn:
self.average_layer = PositionwiseFeedForward(model_dim,
model_dim, dropout)
self.gating_layer = nn.Linear(model_dim * 2, model_dim * 2)
def cumulative_average_mask(self, batch_size, inputs_len, device):
"""
Builds the mask to compute the cumulative average as described in
:cite:`DBLP:journals/corr/abs-1805-00631` -- Figure 3
Args:
batch_size (int): batch size
inputs_len (int): length of the inputs
Returns:
(FloatTensor):
* A Tensor of shape ``(batch_size, input_len, input_len)``
"""
triangle = torch.tril(torch.ones(inputs_len, inputs_len, dtype=
torch.float, device=device))
weights = torch.ones(1, inputs_len, dtype=torch.float, device=device
) / torch.arange(1, inputs_len + 1, dtype=torch.float, device=
device)
mask = triangle * weights.transpose(0, 1)
return mask.unsqueeze(0).expand(batch_size, inputs_len, inputs_len)
def cumulative_average(self, inputs, mask_or_step, layer_cache=None,
step=None):
"""
Computes the cumulative average as described in
:cite:`DBLP:journals/corr/abs-1805-00631` -- Equations (1) (5) (6)
Args:
inputs (FloatTensor): sequence to average
``(batch_size, input_len, dimension)``
mask_or_step: if cache is set, this is assumed
to be the current step of the
dynamic decoding. Otherwise, it is the mask matrix
used to compute the cumulative average.
layer_cache: a dictionary containing the cumulative average
of the previous step.
Returns:
a tensor of the same shape and type as ``inputs``.
"""
if layer_cache is not None:
step = mask_or_step
average_attention = (inputs + step * layer_cache['prev_g']) / (step
+ 1)
layer_cache['prev_g'] = average_attention
return average_attention
else:
mask = mask_or_step
return torch.matmul(mask, inputs)
def forward(self, input_0):
primals_2 = self.gating_layer.weight
primals_3 = self.gating_layer.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0], output[1]
| DenDen047/data2text-macro-plan-py | AverageAttention | false | 7,968 | [
"MIT"
] | 20 | bb01ec6e23dab28c1e969f23bd55776b597fb995 | https://github.com/DenDen047/data2text-macro-plan-py/tree/bb01ec6e23dab28c1e969f23bd55776b597fb995 | import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
class PositionwiseFeedForward(nn.Module):
""" A two-layer Feed-Forward-Network with residual layer norm.
Args:
d_model (int): the size of input for the first-layer of the FFN.
d_ff (int): the hidden layer size of the second-layer
of the FNN.
dropout (float): dropout probability in :math:`[0, 1)`.
"""
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.layer_norm = nn.LayerNorm(d_model, eps=1e-06)
self.dropout_1 = nn.Dropout(dropout)
self.relu = nn.ReLU()
self.dropout_2 = nn.Dropout(dropout)
def forward(self, x):
"""Layer definition.
Args:
x: ``(batch_size, input_len, model_dim)``
Returns:
(FloatTensor): Output ``(batch_size, input_len, model_dim)``.
"""
inter = self.dropout_1(self.relu(self.w_1(self.layer_norm(x))))
output = self.dropout_2(self.w_2(inter))
return output + x
def update_dropout(self, dropout):
self.dropout_1.p = dropout
self.dropout_2.p = dropout
class Model(nn.Module):
"""
Average Attention module from
"Accelerating Neural Transformer via an Average Attention Network"
:cite:`DBLP:journals/corr/abs-1805-00631`.
Args:
model_dim (int): the dimension of keys/values/queries,
must be divisible by head_count
dropout (float): dropout parameter
"""
def __init__(self, model_dim, dropout=0.1, aan_useffn=False):
self.model_dim = model_dim
self.aan_useffn = aan_useffn
super().__init__()
if aan_useffn:
self.average_layer = PositionwiseFeedForward(model_dim,
model_dim, dropout)
self.gating_layer = nn.Linear(model_dim * 2, model_dim * 2)
def cumulative_average_mask(self, batch_size, inputs_len, device):
"""
Builds the mask to compute the cumulative average as described in
:cite:`DBLP:journals/corr/abs-1805-00631` -- Figure 3
Args:
batch_size (int): batch size
inputs_len (int): length of the inputs
Returns:
(FloatTensor):
* A Tensor of shape ``(batch_size, input_len, input_len)``
"""
triangle = torch.tril(torch.ones(inputs_len, inputs_len, dtype=
torch.float, device=device))
weights = torch.ones(1, inputs_len, dtype=torch.float, device=device
) / torch.arange(1, inputs_len + 1, dtype=torch.float, device=
device)
mask = triangle * weights.transpose(0, 1)
return mask.unsqueeze(0).expand(batch_size, inputs_len, inputs_len)
def cumulative_average(self, inputs, mask_or_step, layer_cache=None,
step=None):
"""
Computes the cumulative average as described in
:cite:`DBLP:journals/corr/abs-1805-00631` -- Equations (1) (5) (6)
Args:
inputs (FloatTensor): sequence to average
``(batch_size, input_len, dimension)``
mask_or_step: if cache is set, this is assumed
to be the current step of the
dynamic decoding. Otherwise, it is the mask matrix
used to compute the cumulative average.
layer_cache: a dictionary containing the cumulative average
of the previous step.
Returns:
a tensor of the same shape and type as ``inputs``.
"""
if layer_cache is not None:
step = mask_or_step
average_attention = (inputs + step * layer_cache['prev_g']) / (step
+ 1)
layer_cache['prev_g'] = average_attention
return average_attention
else:
mask = mask_or_step
return torch.matmul(mask, inputs)
def fo
# ... truncated (>4000 chars) for memory efficiency |
ComplexRotationComposition | # 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_1/inductor_cache/is/cisyzgbpkg3wuzhtj635xzxgcwj4tbkcss54skf2m3d4muwronkh.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 = (%abs_1, 1e-08), 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=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_min_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = 1e-08
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tl.store(in_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)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.view_as_complex]
buf0 = torch.ops.aten.view_as_complex.default(reinterpret_tensor(arg0_1, (4, 4, 4, 2, 2), (64, 16, 4, 2, 1), 0))
buf1 = buf0
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.view_as_complex]
buf2 = torch.ops.aten.view_as_complex.default(reinterpret_tensor(arg1_1, (4, 4, 4, 2, 2), (64, 16, 4, 2, 1), 0))
buf3 = buf2
# Topologically Sorted Source Nodes: [abs_1], Original ATen: [aten.abs]
buf4 = torch.ops.aten.abs.default(buf3)
buf5 = buf4
del buf4
buf6 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [clamp_min], Original ATen: [aten.clamp_min]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_min_0.run(buf6, 128, grid=grid(128), stream=stream0)
# Topologically Sorted Source Nodes: [clamp_min, y_1], Original ATen: [aten.clamp_min, aten.div]
buf7 = torch.ops.aten.div.Tensor(buf3, buf6)
del arg1_1
del buf2
del buf3
del buf6
buf8 = buf7
del buf7
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.mul]
buf9 = torch.ops.aten.mul.Tensor(buf1, buf8)
del arg0_1
del buf0
del buf1
del buf8
buf10 = buf9
del buf9
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.view_as_real]
buf11 = torch.ops.aten.view_as_real.default(buf10)
buf12 = buf11
return (reinterpret_tensor(buf12, (4, 4, 4, 4), (64, 16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
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 nn
from abc import abstractmethod
import torch.utils.data
def _to_complex(x: 'torch.Tensor') ->torch.Tensor:
"""View real tensor as complex."""
return torch.view_as_complex(x.view(*x.shape[:-1], -1, 2))
def _to_real(x: 'torch.Tensor') ->torch.Tensor:
"""View complex tensor as real."""
x = torch.view_as_real(x)
return x.view(*x.shape[:-2], -1)
def _complex_multiplication(x: 'torch.Tensor', y: 'torch.Tensor', y_norm:
'bool'=False) ->torch.Tensor:
"""Element-wise multiplication as complex numbers."""
x = _to_complex(x)
y = _to_complex(y)
if y_norm:
y = y / y.abs().clamp_min(1e-08)
x = x * y
return _to_real(x)
class Composition(nn.Module):
"""A base class for compositions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compose two batches vectors.
.. note ::
The two batches have to be of broadcastable shape.
:param x: shape: s_x
The first batch of vectors.
:param y: shape: s_y
The second batch of vectors.
:return: shape: s
The compositionm, where `s` is the broadcasted shape.
"""
raise NotImplementedError
class ComplexRotationComposition(Composition):
"""Composition by rotation in complex plane."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return _complex_multiplication(x, y, y_norm=True)
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 import nn
from abc import abstractmethod
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clamp_min_0(in_out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = 1e-08
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tl.store(in_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 = torch.ops.aten.view_as_complex.default(reinterpret_tensor(
arg0_1, (4, 4, 4, 2, 2), (64, 16, 4, 2, 1), 0))
buf1 = buf0
buf2 = torch.ops.aten.view_as_complex.default(reinterpret_tensor(
arg1_1, (4, 4, 4, 2, 2), (64, 16, 4, 2, 1), 0))
buf3 = buf2
buf4 = torch.ops.aten.abs.default(buf3)
buf5 = buf4
del buf4
buf6 = buf5
del buf5
get_raw_stream(0)
triton_poi_fused_clamp_min_0[grid(128)](buf6, 128, XBLOCK=128,
num_warps=4, num_stages=1)
buf7 = torch.ops.aten.div.Tensor(buf3, buf6)
del arg1_1
del buf2
del buf3
del buf6
buf8 = buf7
del buf7
buf9 = torch.ops.aten.mul.Tensor(buf1, buf8)
del arg0_1
del buf0
del buf1
del buf8
buf10 = buf9
del buf9
buf11 = torch.ops.aten.view_as_real.default(buf10)
buf12 = buf11
return reinterpret_tensor(buf12, (4, 4, 4, 4), (64, 16, 4, 1), 0),
def _to_complex(x: 'torch.Tensor') ->torch.Tensor:
"""View real tensor as complex."""
return torch.view_as_complex(x.view(*x.shape[:-1], -1, 2))
def _to_real(x: 'torch.Tensor') ->torch.Tensor:
"""View complex tensor as real."""
x = torch.view_as_real(x)
return x.view(*x.shape[:-2], -1)
def _complex_multiplication(x: 'torch.Tensor', y: 'torch.Tensor', y_norm:
'bool'=False) ->torch.Tensor:
"""Element-wise multiplication as complex numbers."""
x = _to_complex(x)
y = _to_complex(y)
if y_norm:
y = y / y.abs().clamp_min(1e-08)
x = x * y
return _to_real(x)
class Composition(nn.Module):
"""A base class for compositions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compose two batches vectors.
.. note ::
The two batches have to be of broadcastable shape.
:param x: shape: s_x
The first batch of vectors.
:param y: shape: s_y
The second batch of vectors.
:return: shape: s
The compositionm, where `s` is the broadcasted shape.
"""
raise NotImplementedError
class ComplexRotationCompositionNew(Composition):
"""Composition by rotation in complex plane."""
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DimitrisAlivas/StarQE | ComplexRotationComposition | false | 7,969 | [
"MIT"
] | 11 | c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | https://github.com/DimitrisAlivas/StarQE/tree/c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | import torch
from torch import nn
from abc import abstractmethod
import torch.utils.data
def _to_complex(x: 'torch.Tensor') ->torch.Tensor:
"""View real tensor as complex."""
return torch.view_as_complex(x.view(*x.shape[:-1], -1, 2))
def _to_real(x: 'torch.Tensor') ->torch.Tensor:
"""View complex tensor as real."""
x = torch.view_as_real(x)
return x.view(*x.shape[:-2], -1)
def _complex_multiplication(x: 'torch.Tensor', y: 'torch.Tensor', y_norm:
'bool'=False) ->torch.Tensor:
"""Element-wise multiplication as complex numbers."""
x = _to_complex(x)
y = _to_complex(y)
if y_norm:
y = y / y.abs().clamp_min(1e-08)
x = x * y
return _to_real(x)
class Composition(nn.Module):
"""A base class for compositions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compose two batches vectors.
.. note ::
The two batches have to be of broadcastable shape.
:param x: shape: s_x
The first batch of vectors.
:param y: shape: s_y
The second batch of vectors.
:return: shape: s
The compositionm, where `s` is the broadcasted shape.
"""
raise NotImplementedError
class Model(Composition):
"""Composition by rotation in complex plane."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return _complex_multiplication(x, y, y_norm=True)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
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_1/inductor_cache/7c/c7cf3rnrkj5lotnmd4dh5tr4ha7znfx67bot2ljihp6wwwjt3kkk.py
# Topologically Sorted Source Nodes: [mul, sub, sub_1, mul_1, p_t, sub_2, pow_1, neg, add_1, log, loss, loss_1], Original ATen: [aten.mul, aten.rsub, aten.add, aten.pow, aten.neg, aten.log, aten.mean]
# Source node to ATen node mapping:
# add_1 => add_1
# log => log
# loss => mul_2
# loss_1 => mean
# mul => mul
# mul_1 => mul_1
# neg => neg
# p_t => add
# pow_1 => pow_1
# sub => sub
# sub_1 => sub_1
# sub_2 => sub_2
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg0_1), kwargs = {})
# %sub_1 : [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, %sub_1), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %add), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_2, 2), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%pow_1,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, 1e-07), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add_1,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%neg, %log), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul_2,), kwargs = {})
triton_per_fused_add_log_mean_mul_neg_pow_rsub_0 = async_compile.triton('triton_per_fused_add_log_mean_mul_neg_pow_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mean_mul_neg_pow_rsub_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mean_mul_neg_pow_rsub_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 = 1.0
tmp4 = tmp3 - tmp0
tmp5 = tmp3 - tmp1
tmp6 = tmp4 * tmp5
tmp7 = tmp2 + tmp6
tmp8 = tmp3 - tmp7
tmp9 = tmp8 * tmp8
tmp10 = -tmp9
tmp11 = 1e-07
tmp12 = tmp7 + tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = tmp10 * tmp13
tmp15 = tl.broadcast_to(tmp14, [RBLOCK])
tmp17 = triton_helpers.promote_to_tensor(tl.sum(tmp15, 0))
tmp18 = 256.0
tmp19 = tmp17 / tmp18
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp19, 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, sub, sub_1, mul_1, p_t, sub_2, pow_1, neg, add_1, log, loss, loss_1], Original ATen: [aten.mul, aten.rsub, aten.add, aten.pow, aten.neg, aten.log, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_add_log_mean_mul_neg_pow_rsub_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, gamma=2, eps=1e-07, size_average=True):
super(FocalLoss, self).__init__()
self.gamma = gamma
self.eps = eps
self.size_average = size_average
def forward(self, prob, labels):
p_t = prob * labels + (1 - prob) * (1 - labels)
loss = -(1.0 - p_t) ** self.gamma * torch.log(p_t + self.eps)
if self.size_average:
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
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_add_log_mean_mul_neg_pow_rsub_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 = 1.0
tmp4 = tmp3 - tmp0
tmp5 = tmp3 - tmp1
tmp6 = tmp4 * tmp5
tmp7 = tmp2 + tmp6
tmp8 = tmp3 - tmp7
tmp9 = tmp8 * tmp8
tmp10 = -tmp9
tmp11 = 1e-07
tmp12 = tmp7 + tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = tmp10 * tmp13
tmp15 = tl.broadcast_to(tmp14, [RBLOCK])
tmp17 = triton_helpers.promote_to_tensor(tl.sum(tmp15, 0))
tmp18 = 256.0
tmp19 = tmp17 / tmp18
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp19, 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_mean_mul_neg_pow_rsub_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, gamma=2, eps=1e-07, size_average=True):
super(FocalLossNew, self).__init__()
self.gamma = gamma
self.eps = eps
self.size_average = size_average
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| Dong-JinKim/ActionCooccurrencePriors | FocalLoss | false | 7,970 | [
"MIT"
] | 27 | 110dbeecf4c25955b5b0160bd7d31d25c759ef21 | https://github.com/Dong-JinKim/ActionCooccurrencePriors/tree/110dbeecf4c25955b5b0160bd7d31d25c759ef21 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, gamma=2, eps=1e-07, size_average=True):
super().__init__()
self.gamma = gamma
self.eps = eps
self.size_average = size_average
def forward(self, prob, labels):
p_t = prob * labels + (1 - prob) * (1 - labels)
loss = -(1.0 - p_t) ** self.gamma * torch.log(p_t + self.eps)
if self.size_average:
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 []
|
ScalarMix | # 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_1/inductor_cache/li/clil4ahhftmdme7pj5ln5bz44vfsgvmapktoaxnsmxvtulfxdac5.py
# Topologically Sorted Source Nodes: [mul, weighted_sum, mul_1], Original ATen: [aten.mul, aten.add]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# weighted_sum => add
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select, %select_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, 0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, %add), 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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (0))
tmp1 = tl.broadcast_to(tmp0, [XBLOCK])
tmp2 = tl.load(in_ptr1 + (0))
tmp3 = tl.broadcast_to(tmp2, [XBLOCK])
tmp7 = tl.load(in_ptr2 + (x0), xmask)
tmp4 = tmp3 - tmp3
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp5 / tmp5
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tmp8 + tmp9
tmp11 = tmp1 * tmp10
tl.store(out_ptr0 + (x0), tmp11, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, ), (1, ))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (1, ), (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, weighted_sum, mul_1], Original ATen: [aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_0.run(primals_3, primals_1, primals_2, buf0, 64, grid=grid(64), stream=stream0)
return (buf0, primals_1, primals_3, reinterpret_tensor(primals_2, (4, 4, 4), (16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((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)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class ScalarMix(nn.Module):
def __init__(self, n_layers, dropout=0):
super(ScalarMix, self).__init__()
self.n_layers = n_layers
self.dropout = dropout
self.weights = nn.Parameter(torch.zeros(n_layers))
self.gamma = nn.Parameter(torch.tensor([1.0]))
self.dropout = nn.Dropout(dropout)
def extra_repr(self):
s = f'n_layers={self.n_layers}'
if self.dropout.p > 0:
s += f', dropout={self.dropout.p}'
return s
def forward(self, tensors):
normed_weights = self.dropout(self.weights.softmax(-1))
weighted_sum = sum(w * h for w, h in zip(normed_weights, tensors))
return self.gamma * weighted_sum
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_layers': 1}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_mul_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
tmp0 = tl.load(in_ptr0 + 0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK])
tmp2 = tl.load(in_ptr1 + 0)
tmp3 = tl.broadcast_to(tmp2, [XBLOCK])
tmp7 = tl.load(in_ptr2 + x0, xmask)
tmp4 = tmp3 - tmp3
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp5 / tmp5
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tmp8 + tmp9
tmp11 = tmp1 * tmp10
tl.store(out_ptr0 + x0, tmp11, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1,), (1,))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mul_0[grid(64)](primals_3, primals_1,
primals_2, buf0, 64, XBLOCK=64, num_warps=1, num_stages=1)
return buf0, primals_1, primals_3, reinterpret_tensor(primals_2, (4, 4,
4), (16, 4, 1), 0)
class ScalarMixNew(nn.Module):
def __init__(self, n_layers, dropout=0):
super(ScalarMixNew, self).__init__()
self.n_layers = n_layers
self.dropout = dropout
self.weights = nn.Parameter(torch.zeros(n_layers))
self.gamma = nn.Parameter(torch.tensor([1.0]))
self.dropout = nn.Dropout(dropout)
def extra_repr(self):
s = f'n_layers={self.n_layers}'
if self.dropout.p > 0:
s += f', dropout={self.dropout.p}'
return s
def forward(self, input_0):
primals_1 = self.weights
primals_3 = self.gamma
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| Dadmatech/DadmaTools | ScalarMix | false | 7,971 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class Model(nn.Module):
def __init__(self, n_layers, dropout=0):
super().__init__()
self.n_layers = n_layers
self.dropout = dropout
self.weights = nn.Parameter(torch.zeros(n_layers))
self.gamma = nn.Parameter(torch.tensor([1.0]))
self.dropout = nn.Dropout(dropout)
def extra_repr(self):
s = f'n_layers={self.n_layers}'
if self.dropout.p > 0:
s += f', dropout={self.dropout.p}'
return s
def forward(self, tensors):
normed_weights = self.dropout(self.weights.softmax(-1))
weighted_sum = sum(w * h for w, h in zip(normed_weights, tensors))
return self.gamma * weighted_sum
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [1]
|
CosineSimilarity | # 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_1/inductor_cache/kz/ckz4kdygd3gfajopqwzvtecsmzyghdhlg3mc3xzmysjall7bayjn.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, %expand), 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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
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 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tl.store(out_ptr0 + (x2), tmp15, 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: [x], 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
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.div]
triton_poi_fused_div_0.run(arg1_1, buf1, 256, grid=grid(256), stream=stream0)
del arg1_1
buf2 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf1, (16, 4, 4), (16, 1, 4), 0), out=buf2)
del buf0
del buf1
return (reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
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 nn
from abc import abstractmethod
import torch.utils.data
from torch.nn import functional
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class DotProductSimilarity(Similarity):
"""Dot product similarity."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return x @ y.transpose(-2, -1)
class CosineSimilarity(DotProductSimilarity):
"""Cosine similarity."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
x = functional.normalize(x, p=2, dim=-1)
y = functional.normalize(y, p=2, dim=-1)
return super().forward(x=x, y=y)
def get_inputs():
return [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 libdevice
from torch import nn
from abc import abstractmethod
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_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
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 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tl.store(out_ptr0 + x2, tmp15, 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, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_div_0[grid(256)](arg1_1, buf1, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg1_1
buf2 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf1, (16, 4, 4), (16, 1, 4), 0), out=buf2)
del buf0
del buf1
return reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0),
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class DotProductSimilarity(Similarity):
"""Dot product similarity."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return x @ y.transpose(-2, -1)
class CosineSimilarityNew(DotProductSimilarity):
"""Cosine similarity."""
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DimitrisAlivas/StarQE | CosineSimilarity | false | 7,972 | [
"MIT"
] | 11 | c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | https://github.com/DimitrisAlivas/StarQE/tree/c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | import torch
from torch import nn
from abc import abstractmethod
import torch.utils.data
from torch.nn import functional
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class DotProductSimilarity(Similarity):
"""Dot product similarity."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return x @ y.transpose(-2, -1)
class Model(DotProductSimilarity):
"""Cosine similarity."""
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
x = functional.normalize(x, p=2, dim=-1)
y = functional.normalize(y, p=2, dim=-1)
return super().forward(x=x, y=y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Biaffine | # 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_1/inductor_cache/2d/c2dfcb3rmvwx6dpm7qnv4bificlcp6hquadzm45twrcgdyefq3b5.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# x => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_1, %full_default], -1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 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], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 5, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = 1.0
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp6, tmp9, tmp10)
tmp12 = tl.where(tmp4, tmp5, tmp11)
tl.store(out_ptr0 + (x2), tmp12, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (1, 5, 5), (25, 5, 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.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, buf0, 80, grid=grid(80), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((1, 16, 5), (80, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [s], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (1, 16, 5), (0, 5, 1), 0), primals_3, out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4, 5), (20, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [y], Original ATen: [aten.cat]
triton_poi_fused_cat_0.run(primals_2, buf2, 80, grid=grid(80), stream=stream0)
del primals_2
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [s], Original ATen: [aten.bmm]
extern_kernels.bmm(buf2, reinterpret_tensor(buf1, (4, 5, 4), (20, 1, 5), 0), out=buf3)
del buf1
return (reinterpret_tensor(buf3, (4, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf2, (4, 5, 4), (20, 1, 5), 0), reinterpret_tensor(buf0, (1, 5, 16), (80, 1, 5), 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((1, 5, 5), (25, 5, 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.dataloader
import torch.nn as nn
import torch.nn
class Biaffine(nn.Module):
def __init__(self, n_in, n_out=1, bias_x=True, bias_y=True, diagonal=False
):
super(Biaffine, self).__init__()
self.n_in = n_in
self.n_out = n_out
self.bias_x = bias_x
self.bias_y = bias_y
self.diagonal = diagonal
if self.diagonal:
self.weight = nn.Parameter(torch.Tensor(n_out, n_in + bias_x))
else:
self.weight = nn.Parameter(torch.Tensor(n_out, n_in + bias_x,
n_in + bias_y))
self.reset_parameters()
def extra_repr(self):
s = f'n_in={self.n_in}, n_out={self.n_out}'
if self.bias_x:
s += f', bias_x={self.bias_x}'
if self.bias_y:
s += f', bias_y={self.bias_y}'
return s
def reset_parameters(self):
nn.init.zeros_(self.weight)
def forward(self, x, y):
if self.bias_x:
x = torch.cat((x, torch.ones_like(x[..., :1])), -1)
if self.bias_y:
y = torch.cat((y, torch.ones_like(y[..., :1])), -1)
if self.diagonal:
s = torch.einsum('bxi,byi,oi->boxy', x, y, self.weight)
else:
s = torch.einsum('bxi,oij,byj->boxy', x, self.weight, y)
s = s.squeeze(1)
return s
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'n_in': 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.dataloader
import torch.nn as nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_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
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 5, tl.int64)
tmp9 = 1.0
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp6, tmp9, tmp10)
tmp12 = tl.where(tmp4, tmp5, tmp11)
tl.store(out_ptr0 + x2, tmp12, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (1, 5, 5), (25, 5, 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_cat_0[grid(80)](primals_1, buf0, 80, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((1, 16, 5), (80, 5, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (1, 16, 5), (0, 5, 1),
0), primals_3, out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4, 5), (20, 5, 1), torch.float32)
triton_poi_fused_cat_0[grid(80)](primals_2, buf2, 80, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_2
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(buf2, reinterpret_tensor(buf1, (4, 5, 4), (20, 1,
5), 0), out=buf3)
del buf1
return reinterpret_tensor(buf3, (4, 4, 4), (16, 1, 4), 0
), reinterpret_tensor(buf2, (4, 5, 4), (20, 1, 5), 0
), reinterpret_tensor(buf0, (1, 5, 16), (80, 1, 5), 0)
class BiaffineNew(nn.Module):
def __init__(self, n_in, n_out=1, bias_x=True, bias_y=True, diagonal=False
):
super(BiaffineNew, self).__init__()
self.n_in = n_in
self.n_out = n_out
self.bias_x = bias_x
self.bias_y = bias_y
self.diagonal = diagonal
if self.diagonal:
self.weight = nn.Parameter(torch.Tensor(n_out, n_in + bias_x))
else:
self.weight = nn.Parameter(torch.Tensor(n_out, n_in + bias_x,
n_in + bias_y))
self.reset_parameters()
def extra_repr(self):
s = f'n_in={self.n_in}, n_out={self.n_out}'
if self.bias_x:
s += f', bias_x={self.bias_x}'
if self.bias_y:
s += f', bias_y={self.bias_y}'
return s
def reset_parameters(self):
nn.init.zeros_(self.weight)
def forward(self, input_0, input_1):
primals_3 = self.weight
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3])
return output[0]
| Dadmatech/DadmaTools | Biaffine | false | 7,973 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class Model(nn.Module):
def __init__(self, n_in, n_out=1, bias_x=True, bias_y=True, diagonal=False
):
super().__init__()
self.n_in = n_in
self.n_out = n_out
self.bias_x = bias_x
self.bias_y = bias_y
self.diagonal = diagonal
if self.diagonal:
self.weight = nn.Parameter(torch.Tensor(n_out, n_in + bias_x))
else:
self.weight = nn.Parameter(torch.Tensor(n_out, n_in + bias_x,
n_in + bias_y))
self.reset_parameters()
def extra_repr(self):
s = f'n_in={self.n_in}, n_out={self.n_out}'
if self.bias_x:
s += f', bias_x={self.bias_x}'
if self.bias_y:
s += f', bias_y={self.bias_y}'
return s
def reset_parameters(self):
nn.init.zeros_(self.weight)
def forward(self, x, y):
if self.bias_x:
x = torch.cat((x, torch.ones_like(x[..., :1])), -1)
if self.bias_y:
y = torch.cat((y, torch.ones_like(y[..., :1])), -1)
if self.diagonal:
s = torch.einsum('bxi,byi,oi->boxy', x, y, self.weight)
else:
s = torch.einsum('bxi,oij,byj->boxy', x, self.weight, y)
s = s.squeeze(1)
return s
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [4]
|
BiaffineScorer | # 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_1/inductor_cache/x7/cx7fj3qnorx5ukbeh73veybsg37egnwowogmnvtlpel64yduxdoi.py
# Topologically Sorted Source Nodes: [input1], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# input1 => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_1, %full_default], 3), 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=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 320
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], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 5, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = 1.0
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp6, tmp9, tmp10)
tmp12 = tl.where(tmp4, tmp5, tmp11)
tl.store(out_ptr0 + (x2), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/nd/cndmfvh7lrojuqo776y2t4r3iwqonobmuxwcqkx2pam6myjioeyl.py
# Topologically Sorted Source Nodes: [bilinear], Original ATen: [aten.add]
# Source node to ATen node mapping:
# bilinear => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_2, %primals_4), kwargs = {})
triton_poi_fused_add_1 = async_compile.triton('triton_poi_fused_add_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 5, 5), (25, 5, 1))
assert_size_stride(primals_4, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 5), (80, 20, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [input1], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, buf0, 320, grid=grid(320), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 5), (80, 20, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [input2], Original ATen: [aten.cat]
triton_poi_fused_cat_0.run(primals_2, buf1, 320, grid=grid(320), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [bilinear], Original ATen: [aten._trilinear]
buf2 = torch.ops.aten._trilinear.default(reinterpret_tensor(buf0, (64, 5), (5, 1), 0), primals_3, reinterpret_tensor(buf1, (64, 5), (5, 1), 0), [1, 3], [0], [1, 2], [2, 3])
del primals_3
buf3 = buf2
del buf2
buf4 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [bilinear], Original ATen: [aten.add]
triton_poi_fused_add_1.run(buf4, primals_4, 256, grid=grid(256), stream=stream0)
del primals_4
return (buf4, reinterpret_tensor(buf0, (64, 5), (5, 1), 0), reinterpret_tensor(buf1, (64, 5), (5, 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, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 5, 5), (25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((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.utils.data.dataloader
import torch.nn as nn
import torch.nn
class BiaffineScorer(nn.Module):
def __init__(self, input1_size, input2_size, output_size):
super().__init__()
self.W_bilin = nn.Bilinear(input1_size + 1, input2_size + 1,
output_size)
self.W_bilin.weight.data.zero_()
self.W_bilin.bias.data.zero_()
def forward(self, input1, input2):
input1 = torch.cat([input1, input1.new_ones(*input1.size()[:-1], 1)
], len(input1.size()) - 1)
input2 = torch.cat([input2, input2.new_ones(*input2.size()[:-1], 1)
], len(input2.size()) - 1)
return self.W_bilin(input1, input2)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input1_size': 4, 'input2_size': 4, 'output_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.utils.data.dataloader
import torch.nn as nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 320
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = xindex // 5
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 5, tl.int64)
tmp9 = 1.0
tmp10 = tl.full(tmp9.shape, 0.0, tmp9.dtype)
tmp11 = tl.where(tmp6, tmp9, tmp10)
tmp12 = tl.where(tmp4, tmp5, tmp11)
tl.store(out_ptr0 + x2, tmp12, xmask)
@triton.jit
def triton_poi_fused_add_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 5, 5), (25, 5, 1))
assert_size_stride(primals_4, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 5), (80, 20, 5, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(320)](primals_1, buf0, 320, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 5), (80, 20, 5, 1), torch.float32)
triton_poi_fused_cat_0[grid(320)](primals_2, buf1, 320, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_2
buf2 = torch.ops.aten._trilinear.default(reinterpret_tensor(buf0, (
64, 5), (5, 1), 0), primals_3, reinterpret_tensor(buf1, (64, 5),
(5, 1), 0), [1, 3], [0], [1, 2], [2, 3])
del primals_3
buf3 = buf2
del buf2
buf4 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused_add_1[grid(256)](buf4, primals_4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_4
return buf4, reinterpret_tensor(buf0, (64, 5), (5, 1), 0
), reinterpret_tensor(buf1, (64, 5), (5, 1), 0)
class BiaffineScorerNew(nn.Module):
def __init__(self, input1_size, input2_size, output_size):
super().__init__()
self.W_bilin = nn.Bilinear(input1_size + 1, input2_size + 1,
output_size)
self.W_bilin.weight.data.zero_()
self.W_bilin.bias.data.zero_()
def forward(self, input_0, input_1):
primals_3 = self.W_bilin.weight
primals_4 = self.W_bilin.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| Dadmatech/DadmaTools | BiaffineScorer | false | 7,974 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class Model(nn.Module):
def __init__(self, input1_size, input2_size, output_size):
super().__init__()
self.W_bilin = nn.Bilinear(input1_size + 1, input2_size + 1,
output_size)
self.W_bilin.weight.data.zero_()
self.W_bilin.bias.data.zero_()
def forward(self, input1, input2):
input1 = torch.cat([input1, input1.new_ones(*input1.size()[:-1], 1)
], len(input1.size()) - 1)
input2 = torch.cat([input2, input2.new_ones(*input2.size()[:-1], 1)
], len(input2.size()) - 1)
return self.W_bilin(input1, input2)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
MsgNorm | # 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_1/inductor_cache/xi/cxijcjpd6bjpfmwuerbhwwmbvtvfg6m4yoik7o4dfucfi6czb7jk.py
# Topologically Sorted Source Nodes: [msg, x_norm, mul, msg_1], Original ATen: [aten.div, aten.linalg_vector_norm, aten.mul]
# Source node to ATen node mapping:
# msg => div
# msg_1 => mul_1
# mul => mul
# x_norm => pow_3, pow_4, sum_2
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, %expand), kwargs = {})
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%arg1_1, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_3, [1], True), kwargs = {})
# %pow_4 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 0.5), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %pow_4), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %arg2_1), kwargs = {})
triton_poi_fused_div_linalg_vector_norm_mul_0 = async_compile.triton('triton_poi_fused_div_linalg_vector_norm_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_linalg_vector_norm_mul_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 10, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_linalg_vector_norm_mul_0(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)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr1 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr1 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr1 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr2 + (0))
tmp30 = tl.broadcast_to(tmp29, [XBLOCK])
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tmp17 = tmp16 * tmp16
tmp19 = tmp18 * tmp18
tmp20 = tmp17 + tmp19
tmp22 = tmp21 * tmp21
tmp23 = tmp20 + tmp22
tmp25 = tmp24 * tmp24
tmp26 = tmp23 + tmp25
tmp27 = libdevice.sqrt(tmp26)
tmp28 = tmp15 * tmp27
tmp31 = tmp28 * tmp30
tl.store(in_out_ptr0 + (x3), tmp31, 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, (1, ), (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: [msg, x_norm, mul, msg_1], Original ATen: [aten.div, aten.linalg_vector_norm, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_div_linalg_vector_norm_mul_0.run(buf1, arg0_1, arg1_1, arg2_1, 256, grid=grid(256), stream=stream0)
del arg0_1
del arg1_1
del arg2_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)
arg2_1 = rand_strided((1, ), (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
from torch.nn import functional as F
class MsgNorm(torch.nn.Module):
def __init__(self, learn_msg_scale=False):
super(MsgNorm, self).__init__()
self.msg_scale = torch.nn.Parameter(torch.Tensor([1.0]),
requires_grad=learn_msg_scale)
def forward(self, x, msg, p=2):
msg = F.normalize(msg, p=p, dim=1)
x_norm = x.norm(p=p, dim=1, keepdim=True)
msg = msg * x_norm * self.msg_scale
return msg
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
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_linalg_vector_norm_mul_0(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)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp18 = tl.load(in_ptr1 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr1 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp24 = tl.load(in_ptr1 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr2 + 0)
tmp30 = tl.broadcast_to(tmp29, [XBLOCK])
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tmp17 = tmp16 * tmp16
tmp19 = tmp18 * tmp18
tmp20 = tmp17 + tmp19
tmp22 = tmp21 * tmp21
tmp23 = tmp20 + tmp22
tmp25 = tmp24 * tmp24
tmp26 = tmp23 + tmp25
tmp27 = libdevice.sqrt(tmp26)
tmp28 = tmp15 * tmp27
tmp31 = tmp28 * tmp30
tl.store(in_out_ptr0 + x3, tmp31, 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, (1,), (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_div_linalg_vector_norm_mul_0[grid(256)](buf1,
arg0_1, arg1_1, arg2_1, 256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
del arg2_1
return buf1,
class MsgNormNew(torch.nn.Module):
def __init__(self, learn_msg_scale=False):
super(MsgNormNew, self).__init__()
self.msg_scale = torch.nn.Parameter(torch.Tensor([1.0]),
requires_grad=learn_msg_scale)
def forward(self, input_0, input_1):
arg2_1 = self.msg_scale
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1, arg2_1])
return output[0]
| Dianezzy/YOLaT-VectorGraphicsRecognition | MsgNorm | false | 7,975 | [
"MIT"
] | 44 | ae21ad5850a49048f639d9b283ded927c3b367f7 | https://github.com/Dianezzy/YOLaT-VectorGraphicsRecognition/tree/ae21ad5850a49048f639d9b283ded927c3b367f7 | import torch
from torch.nn import functional as F
class Model(torch.nn.Module):
def __init__(self, learn_msg_scale=False):
super().__init__()
self.msg_scale = torch.nn.Parameter(torch.Tensor([1.0]),
requires_grad=learn_msg_scale)
def forward(self, x, msg, p=2):
msg = F.normalize(msg, p=p, dim=1)
x_norm = x.norm(p=p, dim=1, keepdim=True)
msg = msg * x_norm * self.msg_scale
return msg
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
NegativeLpSimilarity | # 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_1/inductor_cache/o7/co7hdtq5qp4nbdmvy6pj4kybd5f5zozkazu54jkgbw6vpqlturix.py
# Topologically Sorted Source Nodes: [neg], Original ATen: [aten.neg]
# Source node to ATen node mapping:
# neg => neg
# Graph fragment:
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%_cdist_forward,), kwargs = {})
triton_poi_fused_neg_0 = async_compile.triton('triton_poi_fused_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.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_neg_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_neg_0(in_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_out_ptr0 + (x0), xmask)
tmp1 = -tmp0
tl.store(in_out_ptr0 + (x0), tmp1, 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: [cdist], Original ATen: [aten._cdist_forward]
buf0 = torch.ops.aten._cdist_forward.default(arg1_1, arg0_1, 2.0, None)
del arg0_1
del arg1_1
buf1 = buf0
del buf0
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [neg], Original ATen: [aten.neg]
stream0 = get_raw_stream(0)
triton_poi_fused_neg_0.run(buf2, 256, grid=grid(256), stream=stream0)
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)
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 nn
from abc import abstractmethod
import torch.utils.data
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class NegativeLpSimilarity(Similarity):
"""Negative l_p distance similarity."""
def __init__(self, p: 'float'=2.0):
"""
Initialize the similarity.
:param p:
The parameter p for the l_p distance. See also: torch.cdist
"""
super().__init__()
self.p = p
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return -torch.cdist(x, y, p=self.p)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
from abc import abstractmethod
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_neg_0(in_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_out_ptr0 + x0, xmask)
tmp1 = -tmp0
tl.store(in_out_ptr0 + x0, tmp1, 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 = torch.ops.aten._cdist_forward.default(arg1_1, arg0_1, 2.0, None)
del arg0_1
del arg1_1
buf1 = buf0
del buf0
buf2 = buf1
del buf1
get_raw_stream(0)
triton_poi_fused_neg_0[grid(256)](buf2, 256, XBLOCK=128, num_warps=
4, num_stages=1)
return buf2,
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class NegativeLpSimilarityNew(Similarity):
"""Negative l_p distance similarity."""
def __init__(self, p: 'float'=2.0):
"""
Initialize the similarity.
:param p:
The parameter p for the l_p distance. See also: torch.cdist
"""
super().__init__()
self.p = p
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DimitrisAlivas/StarQE | NegativeLpSimilarity | false | 7,976 | [
"MIT"
] | 11 | c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | https://github.com/DimitrisAlivas/StarQE/tree/c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | import torch
from torch import nn
from abc import abstractmethod
import torch.utils.data
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class Model(Similarity):
"""Negative l_p distance similarity."""
def __init__(self, p: 'float'=2.0):
"""
Initialize the similarity.
:param p:
The parameter p for the l_p distance. See also: torch.cdist
"""
super().__init__()
self.p = p
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return -torch.cdist(x, y, p=self.p)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Classifier | # 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_1/inductor_cache/zy/czyl2ydsxrvv7k5pxfpvw4j4cfw37fquv2imk5g3njtvfdfnxire.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_3 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 6400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 100
x2 = xindex % 1600
x3 = (xindex // 1600)
tmp0 = tl.load(in_out_ptr0 + (x4), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x4), tmp4, xmask)
tl.store(out_ptr0 + (x2 + (1664*x3)), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mv/cmvfajbub4yzkkxuxfxdp62hkaezpil47tilswbhhstayb6xpakz.py
# Topologically Sorted Source Nodes: [relu_2], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# relu_2 => relu_2
# Graph fragment:
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_5,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_2, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 3200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 50
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_1/inductor_cache/5n/c5nmaiwcf2lcyj3c2wsrpchfvkhvfhn5dgcdd4z6rjarqb7wivli.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# out => amax, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_9, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_9, %amax), 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=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 32)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/rh/crhslhf33gkqkoiejny5xwqnutjtjja3bjazxvowm42ck7b2e4ch.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# out => 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_3 = async_compile.triton('triton_poi_fused__log_softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 32)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (8 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (16 + x0 + (32*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (24 + x0 + (32*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 = args
args.clear()
assert_size_stride(primals_1, (100, 4), (4, 1))
assert_size_stride(primals_2, (100, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (100, 100), (100, 1))
assert_size_stride(primals_5, (100, ), (1, ))
assert_size_stride(primals_6, (50, 100), (100, 1))
assert_size_stride(primals_7, (50, ), (1, ))
assert_size_stride(primals_8, (50, 50), (50, 1))
assert_size_stride(primals_9, (50, ), (1, ))
assert_size_stride(primals_10, (2, 50), (50, 1))
assert_size_stride(primals_11, (2, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 100), (100, 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, 100), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 100), (1600, 400, 100, 1), 0); del buf0 # reuse
buf14 = empty_strided_cuda((4, 4, 4, 100), (1664, 400, 100, 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, buf14, 6400, grid=grid(6400), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 100), (100, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 100), (100, 1), 0), reinterpret_tensor(primals_4, (100, 100), (1, 100), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 100), (1600, 400, 100, 1), 0); del buf2 # reuse
buf13 = empty_strided_cuda((4, 4, 4, 100), (1664, 400, 100, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf13, 6400, grid=grid(6400), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 50), (50, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf3, (64, 100), (100, 1), 0), reinterpret_tensor(primals_6, (100, 50), (1, 100), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 50), (800, 200, 50, 1), 0); del buf4 # reuse
buf12 = empty_strided_cuda((4, 4, 4, 50), (800, 200, 50, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu_2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf5, primals_7, buf12, 3200, grid=grid(3200), stream=stream0)
del primals_7
buf6 = empty_strided_cuda((64, 50), (50, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf5, (64, 50), (50, 1), 0), reinterpret_tensor(primals_8, (50, 50), (1, 50), 0), out=buf6)
buf7 = reinterpret_tensor(buf6, (4, 4, 4, 50), (800, 200, 50, 1), 0); del buf6 # reuse
buf11 = empty_strided_cuda((4, 4, 4, 50), (800, 200, 50, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu_3], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf7, primals_9, buf11, 3200, grid=grid(3200), stream=stream0)
del primals_9
buf8 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_4], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, reinterpret_tensor(buf7, (64, 50), (50, 1), 0), reinterpret_tensor(primals_10, (50, 2), (1, 50), 0), alpha=1, beta=1, out=buf8)
del primals_11
buf9 = empty_strided_cuda((4, 4, 4, 2), (32, 8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_2.run(buf8, buf9, 128, grid=grid(128), stream=stream0)
buf10 = reinterpret_tensor(buf8, (4, 4, 4, 2), (32, 8, 2, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten._log_softmax]
triton_poi_fused__log_softmax_3.run(buf9, buf10, 128, grid=grid(128), stream=stream0)
del buf9
return (buf10, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 100), (100, 1), 0), reinterpret_tensor(buf3, (64, 100), (100, 1), 0), reinterpret_tensor(buf5, (64, 50), (50, 1), 0), reinterpret_tensor(buf7, (64, 50), (50, 1), 0), buf10, primals_10, buf11, primals_8, buf12, primals_6, buf13, primals_4, buf14, )
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((100, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((100, ), (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((100, 100), (100, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((100, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((50, 100), (100, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((50, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((50, 50), (50, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((50, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((2, 50), (50, 1), device='cuda:0', dtype=torch.float32)
primals_11 = 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, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.nn.functional as F
class Classifier(nn.Module):
def __init__(self, input_size):
super().__init__()
self.hidden_1 = nn.Linear(input_size, 100)
self.hidden_2 = nn.Linear(100, 100)
self.hidden_3 = nn.Linear(100, 50)
self.hidden_4 = nn.Linear(50, 50)
self.output = nn.Linear(50, 2)
self.dropout = nn.Dropout(p=0.1)
def forward(self, x):
z = self.dropout(F.relu(self.hidden_1(x)))
z = self.dropout(F.relu(self.hidden_2(z)))
z = self.dropout(F.relu(self.hidden_3(z)))
z = self.dropout(F.relu(self.hidden_4(z)))
out = F.log_softmax(self.output(z), dim=1)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 6400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x0 = xindex % 100
x2 = xindex % 1600
x3 = xindex // 1600
tmp0 = tl.load(in_out_ptr0 + x4, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x4, tmp4, xmask)
tl.store(out_ptr0 + (x2 + 1664 * x3), tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 3200
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 50
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_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = xindex // 32
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (8 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (16 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (24 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
@triton.jit
def triton_poi_fused__log_softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 8
x2 = xindex // 32
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (8 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (16 + x0 + 32 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (24 + x0 + 32 * 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) = args
args.clear()
assert_size_stride(primals_1, (100, 4), (4, 1))
assert_size_stride(primals_2, (100,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (100, 100), (100, 1))
assert_size_stride(primals_5, (100,), (1,))
assert_size_stride(primals_6, (50, 100), (100, 1))
assert_size_stride(primals_7, (50,), (1,))
assert_size_stride(primals_8, (50, 50), (50, 1))
assert_size_stride(primals_9, (50,), (1,))
assert_size_stride(primals_10, (2, 50), (50, 1))
assert_size_stride(primals_11, (2,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 100), (100, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 100), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 100), (1600, 400, 100, 1), 0)
del buf0
buf14 = empty_strided_cuda((4, 4, 4, 100), (1664, 400, 100, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(6400)](buf1,
primals_2, buf14, 6400, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 100), (100, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 100), (100, 1), 0),
reinterpret_tensor(primals_4, (100, 100), (1, 100), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 100), (1600, 400, 100, 1), 0)
del buf2
buf13 = empty_strided_cuda((4, 4, 4, 100), (1664, 400, 100, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(6400)](buf3,
primals_5, buf13, 6400, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 50), (50, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 100), (100, 1), 0),
reinterpret_tensor(primals_6, (100, 50), (1, 100), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 50), (800, 200, 50, 1), 0)
del buf4
buf12 = empty_strided_cuda((4, 4, 4, 50), (800, 200, 50, 1), torch.bool
)
triton_poi_fused_relu_threshold_backward_1[grid(3200)](buf5,
primals_7, buf12, 3200, XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
buf6 = empty_strided_cuda((64, 50), (50, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf5, (64, 50), (50, 1), 0),
reinterpret_tensor(primals_8, (50, 50), (1, 50), 0), out=buf6)
buf7 = reinterpret_tensor(buf6, (4, 4, 4, 50), (800, 200, 50, 1), 0)
del buf6
buf11 = empty_strided_cuda((4, 4, 4, 50), (800, 200, 50, 1), torch.bool
)
triton_poi_fused_relu_threshold_backward_1[grid(3200)](buf7,
primals_9, buf11, 3200, XBLOCK=256, num_warps=4, num_stages=1)
del primals_9
buf8 = empty_strided_cuda((64, 2), (2, 1), torch.float32)
extern_kernels.addmm(primals_11, reinterpret_tensor(buf7, (64, 50),
(50, 1), 0), reinterpret_tensor(primals_10, (50, 2), (1, 50), 0
), alpha=1, beta=1, out=buf8)
del primals_11
buf9 = empty_strided_cuda((4, 4, 4, 2), (32, 8, 2, 1), torch.float32)
triton_poi_fused__log_softmax_2[grid(128)](buf8, buf9, 128, XBLOCK=
128, num_warps=4, num_stages=1)
buf10 = reinterpret_tensor(buf8, (4, 4, 4, 2), (32, 8, 2, 1), 0)
del buf8
triton_poi_fused__log_softmax_3[grid(128)](buf9, buf10, 128, XBLOCK
=128, num_warps=4, num_stages=1)
del buf9
return (buf10, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(buf1, (64, 100), (100, 1), 0),
reinterpret_tensor(buf3, (64, 100), (100, 1), 0),
reinterpret_tensor(buf5, (64, 50), (50, 1), 0), reinterpret_tensor(
buf7, (64, 50), (50, 1), 0), buf10, primals_10, buf11, primals_8,
buf12, primals_6, buf13, primals_4, buf14)
class ClassifierNew(nn.Module):
def __init__(self, input_size):
super().__init__()
self.hidden_1 = nn.Linear(input_size, 100)
self.hidden_2 = nn.Linear(100, 100)
self.hidden_3 = nn.Linear(100, 50)
self.hidden_4 = nn.Linear(50, 50)
self.output = nn.Linear(50, 2)
self.dropout = nn.Dropout(p=0.1)
def forward(self, input_0):
primals_1 = self.hidden_1.weight
primals_2 = self.hidden_1.bias
primals_4 = self.hidden_2.weight
primals_5 = self.hidden_2.bias
primals_6 = self.hidden_3.weight
primals_7 = self.hidden_3.bias
primals_8 = self.hidden_4.weight
primals_9 = self.hidden_4.bias
primals_10 = self.output.weight
primals_11 = self.output.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]
| Develop-Packt/Solving-a-Classification-Problem-with-DNNs-Using-PyTorch | Classifier | false | 7,977 | [
"MIT"
] | 16 | d0fe33c71242da256e3727bb49417a08de39c85c | https://github.com/Develop-Packt/Solving-a-Classification-Problem-with-DNNs-Using-PyTorch/tree/d0fe33c71242da256e3727bb49417a08de39c85c | import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, input_size):
super().__init__()
self.hidden_1 = nn.Linear(input_size, 100)
self.hidden_2 = nn.Linear(100, 100)
self.hidden_3 = nn.Linear(100, 50)
self.hidden_4 = nn.Linear(50, 50)
self.output = nn.Linear(50, 2)
self.dropout = nn.Dropout(p=0.1)
def forward(self, x):
z = self.dropout(F.relu(self.hidden_1(x)))
z = self.dropout(F.relu(self.hidden_2(z)))
z = self.dropout(F.relu(self.hidden_3(z)))
z = self.dropout(F.relu(self.hidden_4(z)))
out = F.log_softmax(self.output(z), dim=1)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
NegativePowerNormSimilarity | # 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_1/inductor_cache/ap/capaaxd63aalh4fqdcmjippyj3avhxm5vrcqzwydwiwppycshonw.py
# Topologically Sorted Source Nodes: [sub, pow_1, sum_1, neg], Original ATen: [aten.sub, aten.pow, aten.sum, aten.neg]
# Source node to ATen node mapping:
# neg => neg
# pow_1 => pow_1
# sub => sub
# sum_1 => sum_1
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%unsqueeze, %unsqueeze_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [-1]), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%sum_1,), kwargs = {})
triton_poi_fused_neg_pow_sub_sum_0 = async_compile.triton('triton_poi_fused_neg_pow_sub_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_neg_pow_sub_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_neg_pow_sub_sum_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
x3 = (xindex // 4)
x0 = xindex % 4
x2 = (xindex // 16)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (4*x3), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + ((4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (1 + (4*x3)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (1 + (4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (2 + (4*x3)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (2 + (4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (3 + (4*x3)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (3 + (4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp6
tmp8 = tmp3 + tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = -tmp18
tl.store(out_ptr0 + (x4), tmp19, 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, pow_1, sum_1, neg], Original ATen: [aten.sub, aten.pow, aten.sum, aten.neg]
stream0 = get_raw_stream(0)
triton_poi_fused_neg_pow_sub_sum_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
from torch import nn
from abc import abstractmethod
from typing import Union
import torch.utils.data
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class NegativePowerNormSimilarity(Similarity):
"""Negative power norm: -\\|x - y\\|_p^p."""
def __init__(self, p: 'Union[int, float]'=2):
"""
Initialize the similarity.
:param p:
The parameter p for the p-norm.
"""
super().__init__()
self.p = p
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return -(x.unsqueeze(dim=-2) - y.unsqueeze(dim=-3)).pow(self.p).sum(dim
=-1)
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 import nn
from abc import abstractmethod
from typing import Union
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_neg_pow_sub_sum_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
x3 = xindex // 4
x0 = xindex % 4
x2 = xindex // 16
x4 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x3, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4 * x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (1 + 4 * x3), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (1 + 4 * x0 + 16 * x2), xmask, eviction_policy
='evict_last')
tmp9 = tl.load(in_ptr0 + (2 + 4 * x3), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (2 + 4 * x0 + 16 * x2), xmask,
eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (3 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr1 + (3 + 4 * x0 + 16 * x2), xmask,
eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp6
tmp8 = tmp3 + tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = -tmp18
tl.store(out_ptr0 + x4, tmp19, 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_neg_pow_sub_sum_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 Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class NegativePowerNormSimilarityNew(Similarity):
"""Negative power norm: -\\|x - y\\|_p^p."""
def __init__(self, p: 'Union[int, float]'=2):
"""
Initialize the similarity.
:param p:
The parameter p for the p-norm.
"""
super().__init__()
self.p = p
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DimitrisAlivas/StarQE | NegativePowerNormSimilarity | false | 7,978 | [
"MIT"
] | 11 | c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | https://github.com/DimitrisAlivas/StarQE/tree/c17676e5f1e3f19c0c4c117a50abe2ce22ffef28 | import torch
from torch import nn
from abc import abstractmethod
from typing import Union
import torch.utils.data
class Similarity(nn.Module):
"""Base class for similarity functions."""
@abstractmethod
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute pair-wise similarities.
:param x: shape: (*, n, d)
The first vectors.
:param y: shape: (*, m, d)
The second vectors.
:return: shape: (*, n, m)
The similarity values.
"""
raise NotImplementedError
def one_to_one(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
"""
Compute batched one-to-one similarities.
:param x: shape: (*, d)
The first vectors.
:param y: shape: (*, d)
The second vectors.
:return: shape: (*)
The similarity values.
"""
return self(x.unsqueeze(dim=-2), y.unsqueeze(dim=-2)).squeeze(dim=-1
).squeeze(dim=-1)
class Model(Similarity):
"""Negative power norm: -\\|x - y\\|_p^p."""
def __init__(self, p: 'Union[int, float]'=2):
"""
Initialize the similarity.
:param p:
The parameter p for the p-norm.
"""
super().__init__()
self.p = p
def forward(self, x: 'torch.Tensor', y: 'torch.Tensor') ->torch.Tensor:
return -(x.unsqueeze(dim=-2) - y.unsqueeze(dim=-3)).pow(self.p).sum(dim
=-1)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
StableBCELoss | # 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_1/inductor_cache/72/c72ouahxhcvd752zff376qx4gjtkrglch2b3q5qnruxyzp3qss2e.py
# Topologically Sorted Source Nodes: [clamp, mul, sub, abs_1, neg_abs, exp, add, log, loss, mean], Original ATen: [aten.clamp, aten.mul, aten.sub, aten.abs, aten.neg, aten.exp, aten.add, aten.log, aten.mean]
# Source node to ATen node mapping:
# abs_1 => abs_1
# add => add
# clamp => clamp_min
# exp => exp
# log => log
# loss => add_1
# mean => mean
# mul => mul
# neg_abs => neg
# sub => sub
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%arg0_1, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%clamp_min, %mul), 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 = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%exp, 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 = (%sub, %log), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%add_1,), kwargs = {})
triton_per_fused_abs_add_clamp_exp_log_mean_mul_neg_sub_0 = async_compile.triton('triton_per_fused_abs_add_clamp_exp_log_mean_mul_neg_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_clamp_exp_log_mean_mul_neg_sub_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_clamp_exp_log_mean_mul_neg_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)
tmp3 = tl.load(in_ptr1 + (r0), None)
tmp1 = 0.0
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = tmp0 * tmp3
tmp5 = tmp2 - tmp4
tmp6 = tl_math.abs(tmp0)
tmp7 = -tmp6
tmp8 = tl_math.exp(tmp7)
tmp9 = 1.0
tmp10 = tmp8 + tmp9
tmp11 = tl_math.log(tmp10)
tmp12 = tmp5 + tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 256.0
tmp17 = tmp15 / tmp16
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp17, 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: [clamp, mul, sub, abs_1, neg_abs, exp, add, log, loss, mean], Original ATen: [aten.clamp, aten.mul, aten.sub, aten.abs, aten.neg, aten.exp, aten.add, aten.log, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_abs_add_clamp_exp_log_mean_mul_neg_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
class StableBCELoss(torch.nn.modules.Module):
def __init__(self):
super(StableBCELoss, self).__init__()
def forward(self, input, target):
neg_abs = -input.abs()
loss = input.clamp(min=0) - input * target + (1 + neg_abs.exp()).log()
return loss.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
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_per_fused_abs_add_clamp_exp_log_mean_mul_neg_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)
tmp3 = tl.load(in_ptr1 + r0, None)
tmp1 = 0.0
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = tmp0 * tmp3
tmp5 = tmp2 - tmp4
tmp6 = tl_math.abs(tmp0)
tmp7 = -tmp6
tmp8 = tl_math.exp(tmp7)
tmp9 = 1.0
tmp10 = tmp8 + tmp9
tmp11 = tl_math.log(tmp10)
tmp12 = tmp5 + tmp11
tmp13 = tl.broadcast_to(tmp12, [RBLOCK])
tmp15 = triton_helpers.promote_to_tensor(tl.sum(tmp13, 0))
tmp16 = 256.0
tmp17 = tmp15 / tmp16
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp17, 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_clamp_exp_log_mean_mul_neg_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 StableBCELossNew(torch.nn.modules.Module):
def __init__(self):
super(StableBCELossNew, 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]
| Diyago/Automatic-salt-deposits-segmentation | StableBCELoss | false | 7,979 | [
"MIT"
] | 20 | fedfc7f1d9878674382eeb16a820b5f16791f4ab | https://github.com/Diyago/Automatic-salt-deposits-segmentation/tree/fedfc7f1d9878674382eeb16a820b5f16791f4ab | import torch
class Model(torch.nn.modules.Module):
def __init__(self):
super().__init__()
def forward(self, input, target):
neg_abs = -input.abs()
loss = input.clamp(min=0) - input * target + (1 + neg_abs.exp()).log()
return loss.mean()
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
NormalizeLinear | # 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_1/inductor_cache/54/c54anezm5tp7djzvtp7bcl42ylzihy5446xuvla3ymvra5m7c4ma.py
# Topologically Sorted Source Nodes: [w], Original ATen: [aten.div]
# Source node to ATen node mapping:
# w => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_1, %expand), 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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tl.store(out_ptr0 + (x2), tmp15, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (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, 1), torch.float32)
# Topologically Sorted Source Nodes: [w], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_0.run(primals_1, buf0, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_2, (64, 4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0), out=buf1)
del buf0
return (reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_1, reinterpret_tensor(primals_2, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 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 math
import torch
from torch.nn import functional as F
import torch.nn as nn
import torch.nn.init as init
class NormalizeLinear(nn.Module):
def __init__(self, in_features, num_class):
super(NormalizeLinear, self).__init__()
self.weight = nn.Parameter(torch.Tensor(num_class, in_features))
self.reset_parameters()
def reset_parameters(self):
init.kaiming_uniform_(self.weight, a=math.sqrt(5))
def forward(self, x):
w = F.normalize(self.weight.float(), p=2, dim=1)
return F.linear(x.float(), w)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_features': 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._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import math
import torch.nn as nn
import torch.nn.init as 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_div_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
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 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tl.store(out_ptr0 + x2, tmp15, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (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, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_div_0[grid(16)](primals_1, buf0, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_2, (64, 4), (4, 1), 0),
reinterpret_tensor(buf0, (4, 4), (1, 4), 0), out=buf1)
del buf0
return reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0
), primals_1, reinterpret_tensor(primals_2, (64, 4), (4, 1), 0)
class NormalizeLinearNew(nn.Module):
def __init__(self, in_features, num_class):
super(NormalizeLinearNew, self).__init__()
self.weight = nn.Parameter(torch.Tensor(num_class, in_features))
self.reset_parameters()
def reset_parameters(self):
init.kaiming_uniform_(self.weight, a=math.sqrt(5))
def forward(self, input_0):
primals_1 = self.weight
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
| DoubtedSteam/MPANet | NormalizeLinear | false | 7,980 | [
"MIT"
] | 25 | fe4f3f1d83c45485b1498786f89ace96c634f187 | https://github.com/DoubtedSteam/MPANet/tree/fe4f3f1d83c45485b1498786f89ace96c634f187 | import math
import torch
from torch.nn import functional as F
import torch.nn as nn
import torch.nn.init as init
class Model(nn.Module):
def __init__(self, in_features, num_class):
super().__init__()
self.weight = nn.Parameter(torch.Tensor(num_class, in_features))
self.reset_parameters()
def reset_parameters(self):
init.kaiming_uniform_(self.weight, a=math.sqrt(5))
def forward(self, x):
w = F.normalize(self.weight.float(), p=2, dim=1)
return F.linear(x.float(), w)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
MatrixTree | # 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_1/inductor_cache/vp/cvpuatto4d65brcast5qwxjbpm7oqa4pzv6ettrz3hj3hvdmoopb.py
# Topologically Sorted Source Nodes: [eye, ne, lap, sum_1], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
# Source node to ATen node mapping:
# eye => eq, full_default, full_default_1, iota_1, where
# lap => full_default_2, where_1
# ne => ne
# sum_1 => sum_1
# Graph fragment:
# %iota_1 : [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 = (%unsqueeze, %iota_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 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, %full_default, %full_default_1), kwargs = {})
# %ne : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where, 0), kwargs = {})
# %full_default_2 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne, %full_default_2, %select), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%where_1, [0]), kwargs = {})
triton_poi_fused_eye_masked_fill_ne_sum_0 = async_compile.triton('triton_poi_fused_eye_masked_fill_ne_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_eye_masked_fill_ne_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp7 = tl.load(in_ptr0 + (x0), xmask)
tmp16 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + (x0), tmp38, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/e6/ce6wm6ousb32bcuykza77m6mtkc2z5cngagyvrqesohqbv3mai2n.py
# Topologically Sorted Source Nodes: [eye, ne, lap, neg, diag, lap_1, diag_1, exp_1], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
# Source node to ATen node mapping:
# diag => eq_1, full_default_3, iota_2, where_2
# diag_1 => diagonal_copy
# exp_1 => exp_1
# eye => eq, full_default, full_default_1, iota_1, where
# lap => full_default_2, where_1
# lap_1 => add_1
# ne => ne
# neg => neg
# Graph fragment:
# %iota_1 : [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 = (%unsqueeze, %iota_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 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, %full_default, %full_default_1), kwargs = {})
# %ne : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where, 0), kwargs = {})
# %full_default_2 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne, %full_default_2, %select), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%where_1,), kwargs = {})
# %iota_2 : [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_1 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_2, %unsqueeze_2), kwargs = {})
# %full_default_3 : [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_2 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_1, %permute, %full_default_3), kwargs = {})
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg, %where_2), kwargs = {})
# %diagonal_copy : [num_users=1] = call_function[target=torch.ops.aten.diagonal_copy.default](args = (%select_1,), kwargs = {})
# %exp_1 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%diagonal_copy,), kwargs = {})
# %select_scatter_default : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%add_1, %exp_1, 0, 0), kwargs = {})
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_1 = async_compile.triton('triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (5*x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (x2), xmask)
tmp18 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + (x2), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/du/cdu7ybzc3j3r3wh4temrbxtitshfr6exh5habsjqpjnbub2gzazc.py
# Topologically Sorted Source Nodes: [eye_1, ne_1, lap_2, sum_2], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
# Source node to ATen node mapping:
# eye_1 => eq_3, full_default_7, full_default_8, iota_7, where_4
# lap_2 => full_default_9, where_5
# ne_1 => ne_1
# sum_2 => sum_2
# Graph fragment:
# %iota_7 : [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_3 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_6, %iota_7), kwargs = {})
# %full_default_7 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_8 : [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_4 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_3, %full_default_7, %full_default_8), kwargs = {})
# %ne_1 : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where_4, 0), kwargs = {})
# %full_default_9 : [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_5 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne_1, %full_default_9, %select_20), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%where_5, [0]), kwargs = {})
triton_poi_fused_eye_masked_fill_ne_sum_2 = async_compile.triton('triton_poi_fused_eye_masked_fill_ne_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_eye_masked_fill_ne_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_2(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
tmp7 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp16 = tl.load(in_ptr0 + (20 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (24 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (28 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + (x0), tmp38, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/2z/c2zdtqflxv2exb2bqmsudm5ctzjvnpxehknamdqaosucojhwc2gm.py
# Topologically Sorted Source Nodes: [eye_1, ne_1, lap_2, neg_1, diag_5, lap_3, diag_6, exp_5], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
# Source node to ATen node mapping:
# diag_5 => eq_4, full_default_10, iota_8, where_6
# diag_6 => diagonal_copy_3
# exp_5 => exp_5
# eye_1 => eq_3, full_default_7, full_default_8, iota_7, where_4
# lap_2 => full_default_9, where_5
# lap_3 => add_3
# ne_1 => ne_1
# neg_1 => neg_1
# Graph fragment:
# %iota_7 : [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_3 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_6, %iota_7), kwargs = {})
# %full_default_7 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_8 : [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_4 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_3, %full_default_7, %full_default_8), kwargs = {})
# %ne_1 : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where_4, 0), kwargs = {})
# %full_default_9 : [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_5 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne_1, %full_default_9, %select_20), kwargs = {})
# %neg_1 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%where_5,), kwargs = {})
# %iota_8 : [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_4 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_8, %unsqueeze_8), kwargs = {})
# %full_default_10 : [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_6 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_4, %permute_5, %full_default_10), kwargs = {})
# %add_3 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg_1, %where_6), kwargs = {})
# %diagonal_copy_3 : [num_users=1] = call_function[target=torch.ops.aten.diagonal_copy.default](args = (%select_21,), kwargs = {})
# %exp_5 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%diagonal_copy_3,), kwargs = {})
# %select_scatter_default_4 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%add_3, %exp_5, 0, 0), kwargs = {})
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_3 = async_compile.triton('triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (16 + (5*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (16 + x2), xmask)
tmp18 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + (x2), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/uj/cuj4vy6frw232alf4fbgm6krjokwhs7z3gq4vye7xmetrcmifmks.py
# Topologically Sorted Source Nodes: [diag_4, add_2], Original ATen: [aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_2 => add_2
# diag_4 => eq_2, full_default_6, iota_4, where_3
# Graph fragment:
# %iota_4 : [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_2 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_4, %unsqueeze_5), kwargs = {})
# %full_default_6 : [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_3 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_2, %permute_4, %full_default_6), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_16, %where_3), kwargs = {})
triton_poi_fused_add_diag_embed_4 = async_compile.triton('triton_poi_fused_add_diag_embed_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_diag_embed_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_4(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 % 4
x2 = xindex
x1 = (xindex // 4)
tmp4 = tl.load(in_ptr0 + (x2), xmask)
tmp6 = tl.load(in_ptr1 + (5*x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (x2), xmask)
tmp18 = tl.load(in_ptr0 + (5*x0), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 0, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tmp2 == tmp0
tmp5 = tl_math.exp(tmp4)
tmp7 = tmp5 * tmp6
tmp8 = 0.0
tmp9 = tl.where(tmp3, tmp8, tmp7)
tmp10 = x1
tmp11 = tmp10 == tmp0
tmp13 = tmp5 * tmp12
tmp14 = tl.where(tmp11, tmp8, tmp13)
tmp15 = tmp9 - tmp14
tmp16 = tl.where(tmp1, tmp15, tmp4)
tmp17 = tmp2 == tmp10
tmp19 = tl_math.exp(tmp18)
tmp21 = tmp19 * tmp20
tmp22 = tl.where(tmp17, tmp21, tmp8)
tmp23 = tmp16 + tmp22
tl.store(out_ptr0 + (x2), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pn/cpns45qzeow55p653fci3bv3aw3di4uxbd5eqhbtsahfi4sjyrcv.py
# Topologically Sorted Source Nodes: [exp_2, mul, setitem_1, exp_3, mul_1, setitem_2, sub, diag_4, add_2], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_2 => add_2
# diag_4 => eq_2, full_default_6, iota_4, where_3
# exp_2 => exp_2
# exp_3 => exp_3
# mul => mul
# mul_1 => mul_1
# setitem_1 => copy_1, full_default_4
# setitem_2 => copy_2, full_default_5
# sub => sub
# Graph fragment:
# %exp_2 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_5,), kwargs = {})
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_2, %permute_1), kwargs = {})
# %full_default_4 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_1 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_7, %full_default_4), kwargs = {})
# %select_scatter_default_1 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul, %copy_1, 1, 0), kwargs = {})
# %exp_3 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_6,), kwargs = {})
# %mul_1 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_3, %permute_2), kwargs = {})
# %full_default_5 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_2 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_9, %full_default_5), kwargs = {})
# %select_scatter_default_2 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_1, %copy_2, 0, 0), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%select_scatter_default_1, %select_scatter_default_2), kwargs = {})
# %select_scatter_default_3 : [num_users=3] = call_function[target=torch.ops.aten.select_scatter.default](args = (%arg0_1, %sub, 0, 0), kwargs = {})
# %iota_4 : [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_2 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_4, %unsqueeze_5), kwargs = {})
# %full_default_6 : [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_3 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_2, %permute_4, %full_default_6), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_16, %where_3), kwargs = {})
# %select_scatter_default_5 : [num_users=2] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_3, %add_2, 0, 0), kwargs = {})
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_5 = async_compile.triton('triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_exp_fill_lift_fresh_mul_sub_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_exp_fill_lift_fresh_mul_sub_5(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
x2 = (xindex // 16)
x3 = xindex % 16
x0 = xindex % 4
x1 = (xindex // 4) % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr1 + (x3), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + (5*x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr1 + (x5), xmask)
tmp0 = x2
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tmp4 == tmp1
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 * tmp8
tmp10 = 0.0
tmp11 = tl.where(tmp5, tmp10, tmp9)
tmp12 = x1
tmp13 = tmp12 == tmp1
tmp15 = tmp7 * tmp14
tmp16 = tl.where(tmp13, tmp10, tmp15)
tmp17 = tmp11 - tmp16
tmp19 = tl.where(tmp2, tmp17, tmp18)
tmp20 = tl.where(tmp2, tmp3, tmp19)
tl.store(out_ptr0 + (x5), tmp20, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/xu/cxuu67kd2ih7hlfxvzxdodliqss5vd2dnlmciv7rtywdzwxj2owo.py
# Topologically Sorted Source Nodes: [eye_2, ne_2, lap_4, sum_3], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
# Source node to ATen node mapping:
# eye_2 => eq_6, full_default_14, full_default_15, iota_13, where_8
# lap_4 => full_default_16, where_9
# ne_2 => ne_2
# sum_3 => sum_3
# Graph fragment:
# %iota_13 : [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_6 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_12, %iota_13), kwargs = {})
# %full_default_14 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_15 : [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_8 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_6, %full_default_14, %full_default_15), kwargs = {})
# %ne_2 : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where_8, 0), kwargs = {})
# %full_default_16 : [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_9 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne_2, %full_default_16, %select_41), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%where_9, [0]), kwargs = {})
triton_poi_fused_eye_masked_fill_ne_sum_6 = async_compile.triton('triton_poi_fused_eye_masked_fill_ne_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.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_eye_masked_fill_ne_sum_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_6(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
tmp7 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp16 = tl.load(in_ptr0 + (36 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (40 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (44 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + (x0), tmp38, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/uv/cuv5b6xmsdhxebbwbekrugevp7tta3qd3pbwz7r5f25lrlr6kvyp.py
# Topologically Sorted Source Nodes: [eye_2, ne_2, lap_4, neg_2, diag_10, lap_5, diag_11, exp_9], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
# Source node to ATen node mapping:
# diag_10 => eq_7, full_default_17, iota_14, where_10
# diag_11 => diagonal_copy_6
# exp_9 => exp_9
# eye_2 => eq_6, full_default_14, full_default_15, iota_13, where_8
# lap_4 => full_default_16, where_9
# lap_5 => add_5
# ne_2 => ne_2
# neg_2 => neg_2
# Graph fragment:
# %iota_13 : [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_6 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_12, %iota_13), kwargs = {})
# %full_default_14 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_15 : [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_8 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_6, %full_default_14, %full_default_15), kwargs = {})
# %ne_2 : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where_8, 0), kwargs = {})
# %full_default_16 : [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_9 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne_2, %full_default_16, %select_41), kwargs = {})
# %neg_2 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%where_9,), kwargs = {})
# %iota_14 : [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_7 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_14, %unsqueeze_14), kwargs = {})
# %full_default_17 : [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_10 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_7, %permute_10, %full_default_17), kwargs = {})
# %add_5 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg_2, %where_10), kwargs = {})
# %diagonal_copy_6 : [num_users=1] = call_function[target=torch.ops.aten.diagonal_copy.default](args = (%select_42,), kwargs = {})
# %exp_9 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%diagonal_copy_6,), kwargs = {})
# %select_scatter_default_9 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%add_5, %exp_9, 0, 0), kwargs = {})
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_7 = async_compile.triton('triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_7(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (32 + (5*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (32 + x2), xmask)
tmp18 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + (x2), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zx/czxsczymlsvzz4xfytxkrcne4jghw4l6hqt2r7yfmmnomn5tjbcp.py
# Topologically Sorted Source Nodes: [diag_9, add_4], Original ATen: [aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_4 => add_4
# diag_9 => eq_5, full_default_13, iota_10, where_7
# Graph fragment:
# %iota_10 : [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_5 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_10, %unsqueeze_11), kwargs = {})
# %full_default_13 : [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_7 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_5, %permute_9, %full_default_13), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_37, %where_7), kwargs = {})
triton_poi_fused_add_diag_embed_8 = async_compile.triton('triton_poi_fused_add_diag_embed_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_8', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_8(in_ptr0, in_ptr1, in_ptr2, 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)
tmp5 = tl.load(in_ptr0 + (16 + x2), xmask)
tmp7 = tl.load(in_ptr1 + (5*x0), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + (x2), xmask)
tmp17 = tl.load(in_ptr2 + (16 + x2), xmask)
tmp20 = tl.load(in_ptr0 + (16 + (5*x0)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 1, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp6 = tl_math.exp(tmp5)
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = x1
tmp12 = tmp11 == tmp3
tmp14 = tmp6 * tmp13
tmp15 = tl.where(tmp12, tmp9, tmp14)
tmp16 = tmp10 - tmp15
tmp18 = tl.where(tmp1, tmp16, tmp17)
tmp19 = tmp2 == tmp11
tmp21 = tl_math.exp(tmp20)
tmp23 = tmp21 * tmp22
tmp24 = tl.where(tmp19, tmp23, tmp9)
tmp25 = tmp18 + tmp24
tl.store(out_ptr0 + (x2), tmp25, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ql/cqlqqv6rrej7gm4ylduo4gscfvd3erfpuwkmg2d6mwja4zzb745y.py
# Topologically Sorted Source Nodes: [exp_6, mul_3, setitem_6, exp_7, mul_4, setitem_7, sub_1, diag_9, add_4], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_4 => add_4
# diag_9 => eq_5, full_default_13, iota_10, where_7
# exp_6 => exp_6
# exp_7 => exp_7
# mul_3 => mul_3
# mul_4 => mul_4
# setitem_6 => copy_6, full_default_11
# setitem_7 => copy_7, full_default_12
# sub_1 => sub_1
# Graph fragment:
# %exp_6 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_25,), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_6, %permute_6), kwargs = {})
# %full_default_11 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_6 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_27, %full_default_11), kwargs = {})
# %select_scatter_default_6 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_3, %copy_6, 1, 0), kwargs = {})
# %exp_7 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_26,), kwargs = {})
# %mul_4 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_7, %permute_7), kwargs = {})
# %full_default_12 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_7 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_29, %full_default_12), kwargs = {})
# %select_scatter_default_7 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_4, %copy_7, 0, 0), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%select_scatter_default_6, %select_scatter_default_7), kwargs = {})
# %select_scatter_default_8 : [num_users=3] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_5, %sub_1, 0, 1), kwargs = {})
# %iota_10 : [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_5 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_10, %unsqueeze_11), kwargs = {})
# %full_default_13 : [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_7 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_5, %permute_9, %full_default_13), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_37, %where_7), kwargs = {})
# %select_scatter_default_10 : [num_users=2] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_8, %add_4, 0, 1), kwargs = {})
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_9 = async_compile.triton('triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_exp_fill_lift_fresh_mul_sub_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_exp_fill_lift_fresh_mul_sub_9(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 // 16)
x3 = xindex % 16
x0 = xindex % 4
x1 = (xindex // 4) % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (16 + x3), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (5*x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_out_ptr0 + (x5), xmask)
tmp0 = x2
tmp1 = tl.full([1], 1, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = tmp4 == tmp5
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = 0.0
tmp12 = tl.where(tmp6, tmp11, tmp10)
tmp13 = x1
tmp14 = tmp13 == tmp5
tmp16 = tmp8 * tmp15
tmp17 = tl.where(tmp14, tmp11, tmp16)
tmp18 = tmp12 - tmp17
tmp20 = tl.where(tmp2, tmp18, tmp19)
tmp21 = tl.where(tmp2, tmp3, tmp20)
tl.store(in_out_ptr0 + (x5), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zn/cznct6seufqtrnxq7k5wa23k5tsr4a6s4kwoqkt6vdlglpootd3q.py
# Topologically Sorted Source Nodes: [eye_3, ne_3, lap_6, sum_4], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
# Source node to ATen node mapping:
# eye_3 => eq_9, full_default_21, full_default_22, iota_19, where_12
# lap_6 => full_default_23, where_13
# ne_3 => ne_3
# sum_4 => sum_4
# Graph fragment:
# %iota_19 : [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_9 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_18, %iota_19), kwargs = {})
# %full_default_21 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_22 : [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_12 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_9, %full_default_21, %full_default_22), kwargs = {})
# %ne_3 : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where_12, 0), kwargs = {})
# %full_default_23 : [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_13 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne_3, %full_default_23, %select_62), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%where_13, [0]), kwargs = {})
triton_poi_fused_eye_masked_fill_ne_sum_10 = async_compile.triton('triton_poi_fused_eye_masked_fill_ne_sum_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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_eye_masked_fill_ne_sum_10', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_10(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
tmp7 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp16 = tl.load(in_ptr0 + (52 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (56 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (60 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + (x0), tmp38, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cj/ccje3ga6gm7gxrs6lb3vzzspmz3npzhdx4eufnbnn4bza7ikwddb.py
# Topologically Sorted Source Nodes: [eye_3, ne_3, lap_6, neg_3, diag_15, lap_7, diag_16, exp_13], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
# Source node to ATen node mapping:
# diag_15 => eq_10, full_default_24, iota_20, where_14
# diag_16 => diagonal_copy_9
# exp_13 => exp_13
# eye_3 => eq_9, full_default_21, full_default_22, iota_19, where_12
# lap_6 => full_default_23, where_13
# lap_7 => add_7
# ne_3 => ne_3
# neg_3 => neg_3
# Graph fragment:
# %iota_19 : [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_9 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_18, %iota_19), kwargs = {})
# %full_default_21 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %full_default_22 : [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_12 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_9, %full_default_21, %full_default_22), kwargs = {})
# %ne_3 : [num_users=1] = call_function[target=torch.ops.aten.ne.Scalar](args = (%where_12, 0), kwargs = {})
# %full_default_23 : [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_13 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%ne_3, %full_default_23, %select_62), kwargs = {})
# %neg_3 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%where_13,), kwargs = {})
# %iota_20 : [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_10 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_20, %unsqueeze_20), kwargs = {})
# %full_default_24 : [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_14 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_10, %permute_15, %full_default_24), kwargs = {})
# %add_7 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg_3, %where_14), kwargs = {})
# %diagonal_copy_9 : [num_users=1] = call_function[target=torch.ops.aten.diagonal_copy.default](args = (%select_63,), kwargs = {})
# %exp_13 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%diagonal_copy_9,), kwargs = {})
# %select_scatter_default_14 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%add_7, %exp_13, 0, 0), kwargs = {})
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_11 = async_compile.triton('triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_11', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_11(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (48 + (5*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (48 + x2), xmask)
tmp18 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + (x2), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pn/cpn2v3a4lgavycp7pw6hgu5hqzgq46epn65bvsewcsgil723mihm.py
# Topologically Sorted Source Nodes: [diag_14, add_6], Original ATen: [aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_6 => add_6
# diag_14 => eq_8, full_default_20, iota_16, where_11
# Graph fragment:
# %iota_16 : [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_8 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_16, %unsqueeze_17), kwargs = {})
# %full_default_20 : [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_11 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_8, %permute_14, %full_default_20), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_58, %where_11), kwargs = {})
triton_poi_fused_add_diag_embed_12 = async_compile.triton('triton_poi_fused_add_diag_embed_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_12', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_12(in_ptr0, in_ptr1, in_ptr2, 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)
tmp5 = tl.load(in_ptr0 + (32 + x2), xmask)
tmp7 = tl.load(in_ptr1 + (5*x0), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + (x2), xmask)
tmp17 = tl.load(in_ptr2 + (32 + x2), xmask)
tmp20 = tl.load(in_ptr0 + (32 + (5*x0)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 2, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp6 = tl_math.exp(tmp5)
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = x1
tmp12 = tmp11 == tmp3
tmp14 = tmp6 * tmp13
tmp15 = tl.where(tmp12, tmp9, tmp14)
tmp16 = tmp10 - tmp15
tmp18 = tl.where(tmp1, tmp16, tmp17)
tmp19 = tmp2 == tmp11
tmp21 = tl_math.exp(tmp20)
tmp23 = tmp21 * tmp22
tmp24 = tl.where(tmp19, tmp23, tmp9)
tmp25 = tmp18 + tmp24
tl.store(out_ptr0 + (x2), tmp25, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pf/cpfq63tk4m2243aundthr33vn54ek4yy5rpvd4qx3nysiiqhk7an.py
# Topologically Sorted Source Nodes: [exp_10, mul_6, setitem_11, exp_11, mul_7, setitem_12, sub_2, diag_14, add_6], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_6 => add_6
# diag_14 => eq_8, full_default_20, iota_16, where_11
# exp_10 => exp_10
# exp_11 => exp_11
# mul_6 => mul_6
# mul_7 => mul_7
# setitem_11 => copy_11, full_default_18
# setitem_12 => copy_12, full_default_19
# sub_2 => sub_2
# Graph fragment:
# %exp_10 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_46,), kwargs = {})
# %mul_6 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_10, %permute_11), kwargs = {})
# %full_default_18 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_11 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_48, %full_default_18), kwargs = {})
# %select_scatter_default_11 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_6, %copy_11, 1, 0), kwargs = {})
# %exp_11 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_47,), kwargs = {})
# %mul_7 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_11, %permute_12), kwargs = {})
# %full_default_19 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_12 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_50, %full_default_19), kwargs = {})
# %select_scatter_default_12 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_7, %copy_12, 0, 0), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%select_scatter_default_11, %select_scatter_default_12), kwargs = {})
# %select_scatter_default_13 : [num_users=3] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_10, %sub_2, 0, 2), kwargs = {})
# %iota_16 : [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_8 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_16, %unsqueeze_17), kwargs = {})
# %full_default_20 : [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_11 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_8, %permute_14, %full_default_20), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_58, %where_11), kwargs = {})
# %select_scatter_default_15 : [num_users=2] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_13, %add_6, 0, 2), kwargs = {})
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_13 = async_compile.triton('triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_exp_fill_lift_fresh_mul_sub_13', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_exp_fill_lift_fresh_mul_sub_13(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 // 16)
x3 = xindex % 16
x0 = xindex % 4
x1 = (xindex // 4) % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (32 + x3), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (5*x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_out_ptr0 + (x5), xmask)
tmp0 = x2
tmp1 = tl.full([1], 2, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = tmp4 == tmp5
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = 0.0
tmp12 = tl.where(tmp6, tmp11, tmp10)
tmp13 = x1
tmp14 = tmp13 == tmp5
tmp16 = tmp8 * tmp15
tmp17 = tl.where(tmp14, tmp11, tmp16)
tmp18 = tmp12 - tmp17
tmp20 = tl.where(tmp2, tmp18, tmp19)
tmp21 = tl.where(tmp2, tmp3, tmp20)
tl.store(in_out_ptr0 + (x5), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mn/cmnzq7y6l7bbvbguv4zgtgukuir43qo6xnjghi3eqjocn7zcxcwm.py
# Topologically Sorted Source Nodes: [diag_19, add_8], Original ATen: [aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_8 => add_8
# diag_19 => eq_11, full_default_27, iota_22, where_15
# Graph fragment:
# %iota_22 : [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_11 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_22, %unsqueeze_23), kwargs = {})
# %full_default_27 : [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_15 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_11, %permute_19, %full_default_27), kwargs = {})
# %add_8 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_79, %where_15), kwargs = {})
triton_poi_fused_add_diag_embed_14 = async_compile.triton('triton_poi_fused_add_diag_embed_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_14', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_14(in_ptr0, in_ptr1, in_ptr2, 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)
tmp5 = tl.load(in_ptr0 + (48 + x2), xmask)
tmp7 = tl.load(in_ptr1 + (5*x0), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + (x2), xmask)
tmp17 = tl.load(in_ptr2 + (48 + x2), xmask)
tmp20 = tl.load(in_ptr0 + (48 + (5*x0)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 3, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp6 = tl_math.exp(tmp5)
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = x1
tmp12 = tmp11 == tmp3
tmp14 = tmp6 * tmp13
tmp15 = tl.where(tmp12, tmp9, tmp14)
tmp16 = tmp10 - tmp15
tmp18 = tl.where(tmp1, tmp16, tmp17)
tmp19 = tmp2 == tmp11
tmp21 = tl_math.exp(tmp20)
tmp23 = tmp21 * tmp22
tmp24 = tl.where(tmp19, tmp23, tmp9)
tmp25 = tmp18 + tmp24
tl.store(out_ptr0 + (x2), tmp25, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/3p/c3p4mgzau45jj52ckoani65yfa6a567t65g72xqsdrdzfv7abi5b.py
# Topologically Sorted Source Nodes: [exp_14, mul_9, setitem_16, exp_15, mul_10, setitem_17, sub_3, diag_19, add_8], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
# Source node to ATen node mapping:
# add_8 => add_8
# diag_19 => eq_11, full_default_27, iota_22, where_15
# exp_14 => exp_14
# exp_15 => exp_15
# mul_10 => mul_10
# mul_9 => mul_9
# setitem_16 => copy_16, full_default_25
# setitem_17 => copy_17, full_default_26
# sub_3 => sub_3
# Graph fragment:
# %exp_14 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_67,), kwargs = {})
# %mul_9 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_14, %permute_16), kwargs = {})
# %full_default_25 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_16 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_69, %full_default_25), kwargs = {})
# %select_scatter_default_16 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_9, %copy_16, 1, 0), kwargs = {})
# %exp_15 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%select_68,), kwargs = {})
# %mul_10 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_15, %permute_17), kwargs = {})
# %full_default_26 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %copy_17 : [num_users=1] = call_function[target=torch.ops.aten.copy.default](args = (%select_71, %full_default_26), kwargs = {})
# %select_scatter_default_17 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%mul_10, %copy_17, 0, 0), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%select_scatter_default_16, %select_scatter_default_17), kwargs = {})
# %select_scatter_default_18 : [num_users=3] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_15, %sub_3, 0, 3), kwargs = {})
# %iota_22 : [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_11 : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%iota_22, %unsqueeze_23), kwargs = {})
# %full_default_27 : [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_15 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq_11, %permute_19, %full_default_27), kwargs = {})
# %add_8 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%select_79, %where_15), kwargs = {})
# %select_scatter_default_19 : [num_users=1] = call_function[target=torch.ops.aten.select_scatter.default](args = (%select_scatter_default_18, %add_8, 0, 3), kwargs = {})
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_15 = async_compile.triton('triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_15', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_diag_embed_exp_fill_lift_fresh_mul_sub_15', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_diag_embed_exp_fill_lift_fresh_mul_sub_15(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 // 16)
x3 = xindex % 16
x0 = xindex % 4
x1 = (xindex // 4) % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (48 + x3), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (5*x0), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_out_ptr0 + (x5), xmask)
tmp0 = x2
tmp1 = tl.full([1], 3, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = tmp4 == tmp5
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = 0.0
tmp12 = tl.where(tmp6, tmp11, tmp10)
tmp13 = x1
tmp14 = tmp13 == tmp5
tmp16 = tmp8 * tmp15
tmp17 = tl.where(tmp14, tmp11, tmp16)
tmp18 = tmp12 - tmp17
tmp20 = tl.where(tmp2, tmp18, tmp19)
tmp21 = tl.where(tmp2, tmp3, tmp20)
tl.store(in_out_ptr0 + (x5), tmp21, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [eye, ne, lap, sum_1], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
stream0 = get_raw_stream(0)
triton_poi_fused_eye_masked_fill_ne_sum_0.run(arg0_1, buf0, 4, grid=grid(4), stream=stream0)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [eye, ne, lap, neg, diag, lap_1, diag_1, exp_1], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_1.run(arg0_1, buf0, buf1, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [eye, ne, lap, neg, diag, lap_1, diag_1, exp_1, inv_laplacian], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp, aten.linalg_inv_ex]
buf2 = torch.ops.aten.linalg_inv_ex.default(buf1)
buf3 = buf2[0]
del buf2
buf5 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [eye_1, ne_1, lap_2, sum_2], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
triton_poi_fused_eye_masked_fill_ne_sum_2.run(arg0_1, buf5, 4, grid=grid(4), stream=stream0)
buf6 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [eye_1, ne_1, lap_2, neg_1, diag_5, lap_3, diag_6, exp_5], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_3.run(arg0_1, buf5, buf6, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [eye_1, ne_1, lap_2, neg_1, diag_5, lap_3, diag_6, exp_5, inv_laplacian_1], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp, aten.linalg_inv_ex]
buf7 = torch.ops.aten.linalg_inv_ex.default(buf6)
buf8 = buf7[0]
del buf7
buf10 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [diag_4, add_2], Original ATen: [aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_4.run(arg0_1, buf3, buf10, 16, grid=grid(16), stream=stream0)
buf11 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [exp_2, mul, setitem_1, exp_3, mul_1, setitem_2, sub, diag_4, add_2], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_5.run(buf10, arg0_1, buf3, buf11, 64, grid=grid(64), stream=stream0)
del buf10
buf12 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [eye_2, ne_2, lap_4, sum_3], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
triton_poi_fused_eye_masked_fill_ne_sum_6.run(arg0_1, buf12, 4, grid=grid(4), stream=stream0)
buf13 = reinterpret_tensor(buf3, (4, 4), (4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [eye_2, ne_2, lap_4, neg_2, diag_10, lap_5, diag_11, exp_9], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_7.run(arg0_1, buf12, buf13, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [eye_2, ne_2, lap_4, neg_2, diag_10, lap_5, diag_11, exp_9, inv_laplacian_2], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp, aten.linalg_inv_ex]
buf14 = torch.ops.aten.linalg_inv_ex.default(buf13)
buf15 = buf14[0]
del buf14
buf17 = buf13; del buf13 # reuse
# Topologically Sorted Source Nodes: [diag_9, add_4], Original ATen: [aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_8.run(arg0_1, buf8, buf11, buf17, 16, grid=grid(16), stream=stream0)
buf18 = buf11; del buf11 # reuse
# Topologically Sorted Source Nodes: [exp_6, mul_3, setitem_6, exp_7, mul_4, setitem_7, sub_1, diag_9, add_4], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_9.run(buf18, buf17, arg0_1, buf8, 64, grid=grid(64), stream=stream0)
del buf17
buf19 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [eye_3, ne_3, lap_6, sum_4], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.sum]
triton_poi_fused_eye_masked_fill_ne_sum_10.run(arg0_1, buf19, 4, grid=grid(4), stream=stream0)
buf20 = reinterpret_tensor(buf8, (4, 4), (4, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [eye_3, ne_3, lap_6, neg_3, diag_15, lap_7, diag_16, exp_13], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp]
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_11.run(arg0_1, buf19, buf20, 16, grid=grid(16), stream=stream0)
del buf19
# Topologically Sorted Source Nodes: [eye_3, ne_3, lap_6, neg_3, diag_15, lap_7, diag_16, exp_13, inv_laplacian_3], Original ATen: [aten.eye, aten.ne, aten.masked_fill, aten.neg, aten.diag_embed, aten.add, aten.diagonal_copy, aten.exp, aten.linalg_inv_ex]
buf21 = torch.ops.aten.linalg_inv_ex.default(buf20)
buf22 = buf21[0]
del buf21
buf24 = buf20; del buf20 # reuse
# Topologically Sorted Source Nodes: [diag_14, add_6], Original ATen: [aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_12.run(arg0_1, buf15, buf18, buf24, 16, grid=grid(16), stream=stream0)
buf25 = buf18; del buf18 # reuse
# Topologically Sorted Source Nodes: [exp_10, mul_6, setitem_11, exp_11, mul_7, setitem_12, sub_2, diag_14, add_6], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_13.run(buf25, buf24, arg0_1, buf15, 64, grid=grid(64), stream=stream0)
del buf15
buf26 = buf24; del buf24 # reuse
# Topologically Sorted Source Nodes: [diag_19, add_8], Original ATen: [aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_14.run(arg0_1, buf22, buf25, buf26, 16, grid=grid(16), stream=stream0)
buf27 = buf25; del buf25 # reuse
# Topologically Sorted Source Nodes: [exp_14, mul_9, setitem_16, exp_15, mul_10, setitem_17, sub_3, diag_19, add_8], Original ATen: [aten.exp, aten.mul, aten.lift_fresh, aten.fill, aten.sub, aten.diag_embed, aten.add]
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_15.run(buf27, buf26, arg0_1, buf22, 64, grid=grid(64), stream=stream0)
del arg0_1
del buf22
del buf26
return (buf27, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
class MatrixTree(nn.Module):
"""Implementation of the matrix-tree theorem for computing marginals
of non-projective dependency parsing. This attention layer is used
in the paper "Learning Structured Text Representations"
:cite:`DBLP:journals/corr/LiuL17d`.
"""
def __init__(self, eps=1e-05):
self.eps = eps
super(MatrixTree, self).__init__()
def forward(self, input):
laplacian = input.exp() + self.eps
output = input.clone()
for b in range(input.size(0)):
lap = laplacian[b].masked_fill(torch.eye(input.size(1), device=
input.device).ne(0), 0)
lap = -lap + torch.diag(lap.sum(0))
lap[0] = input[b].diag().exp()
inv_laplacian = lap.inverse()
factor = inv_laplacian.diag().unsqueeze(1).expand_as(input[b]
).transpose(0, 1)
term1 = input[b].exp().mul(factor).clone()
term2 = input[b].exp().mul(inv_laplacian.transpose(0, 1)).clone()
term1[:, 0] = 0
term2[0] = 0
output[b] = term1 - term2
roots_output = input[b].diag().exp().mul(inv_laplacian.
transpose(0, 1)[0])
output[b] = output[b] + torch.diag(roots_output)
return output
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.cuda
import torch.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_0(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp7 = tl.load(in_ptr0 + x0, xmask)
tmp16 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + x0, tmp38, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_1(
in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + 5 * x0, xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + x2, xmask)
tmp18 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + x2, tmp21, xmask)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_2(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
tmp7 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp16 = tl.load(in_ptr0 + (20 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (24 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (28 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + x0, tmp38, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_3(
in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (16 + 5 * x0), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (16 + x2), xmask)
tmp18 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + x2, tmp21, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_4(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 % 4
x2 = xindex
x1 = xindex // 4
tmp4 = tl.load(in_ptr0 + x2, xmask)
tmp6 = tl.load(in_ptr1 + 5 * x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + x2, xmask)
tmp18 = tl.load(in_ptr0 + 5 * x0, xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 0, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tmp2 == tmp0
tmp5 = tl_math.exp(tmp4)
tmp7 = tmp5 * tmp6
tmp8 = 0.0
tmp9 = tl.where(tmp3, tmp8, tmp7)
tmp10 = x1
tmp11 = tmp10 == tmp0
tmp13 = tmp5 * tmp12
tmp14 = tl.where(tmp11, tmp8, tmp13)
tmp15 = tmp9 - tmp14
tmp16 = tl.where(tmp1, tmp15, tmp4)
tmp17 = tmp2 == tmp10
tmp19 = tl_math.exp(tmp18)
tmp21 = tmp19 * tmp20
tmp22 = tl.where(tmp17, tmp21, tmp8)
tmp23 = tmp16 + tmp22
tl.store(out_ptr0 + x2, tmp23, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_5(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
x2 = xindex // 16
x3 = xindex % 16
x0 = xindex % 4
x1 = xindex // 4 % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr1 + x3, xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + 5 * x0, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr1 + x5, xmask)
tmp0 = x2
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tmp4 == tmp1
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 * tmp8
tmp10 = 0.0
tmp11 = tl.where(tmp5, tmp10, tmp9)
tmp12 = x1
tmp13 = tmp12 == tmp1
tmp15 = tmp7 * tmp14
tmp16 = tl.where(tmp13, tmp10, tmp15)
tmp17 = tmp11 - tmp16
tmp19 = tl.where(tmp2, tmp17, tmp18)
tmp20 = tl.where(tmp2, tmp3, tmp19)
tl.store(out_ptr0 + x5, tmp20, xmask)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_6(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
tmp7 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp16 = tl.load(in_ptr0 + (36 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (40 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (44 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + x0, tmp38, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_7(
in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (32 + 5 * x0), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (32 + x2), xmask)
tmp18 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + x2, tmp21, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_8(in_ptr0, in_ptr1, in_ptr2, 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
tmp5 = tl.load(in_ptr0 + (16 + x2), xmask)
tmp7 = tl.load(in_ptr1 + 5 * x0, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + x2, xmask)
tmp17 = tl.load(in_ptr2 + (16 + x2), xmask)
tmp20 = tl.load(in_ptr0 + (16 + 5 * x0), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 1, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp6 = tl_math.exp(tmp5)
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = x1
tmp12 = tmp11 == tmp3
tmp14 = tmp6 * tmp13
tmp15 = tl.where(tmp12, tmp9, tmp14)
tmp16 = tmp10 - tmp15
tmp18 = tl.where(tmp1, tmp16, tmp17)
tmp19 = tmp2 == tmp11
tmp21 = tl_math.exp(tmp20)
tmp23 = tmp21 * tmp22
tmp24 = tl.where(tmp19, tmp23, tmp9)
tmp25 = tmp18 + tmp24
tl.store(out_ptr0 + x2, tmp25, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_9(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 // 16
x3 = xindex % 16
x0 = xindex % 4
x1 = xindex // 4 % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (16 + x3), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + 5 * x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_out_ptr0 + x5, xmask)
tmp0 = x2
tmp1 = tl.full([1], 1, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = tmp4 == tmp5
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = 0.0
tmp12 = tl.where(tmp6, tmp11, tmp10)
tmp13 = x1
tmp14 = tmp13 == tmp5
tmp16 = tmp8 * tmp15
tmp17 = tl.where(tmp14, tmp11, tmp16)
tmp18 = tmp12 - tmp17
tmp20 = tl.where(tmp2, tmp18, tmp19)
tmp21 = tl.where(tmp2, tmp3, tmp20)
tl.store(in_out_ptr0 + x5, tmp21, xmask)
@triton.jit
def triton_poi_fused_eye_masked_fill_ne_sum_10(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
tmp7 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp16 = tl.load(in_ptr0 + (52 + x0), xmask)
tmp25 = tl.load(in_ptr0 + (56 + x0), xmask)
tmp34 = tl.load(in_ptr0 + (60 + x0), xmask)
tmp0 = tl.full([1], 0, tl.int64)
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = tmp5 != tmp4
tmp8 = tl_math.exp(tmp7)
tmp9 = 1e-05
tmp10 = tmp8 + tmp9
tmp11 = tl.where(tmp6, tmp4, tmp10)
tmp12 = tl.full([1], 1, tl.int64)
tmp13 = tmp12 == tmp1
tmp14 = tl.where(tmp13, tmp3, tmp4)
tmp15 = tmp14 != tmp4
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 + tmp9
tmp19 = tl.where(tmp15, tmp4, tmp18)
tmp20 = tmp11 + tmp19
tmp21 = tl.full([1], 2, tl.int64)
tmp22 = tmp21 == tmp1
tmp23 = tl.where(tmp22, tmp3, tmp4)
tmp24 = tmp23 != tmp4
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp26 + tmp9
tmp28 = tl.where(tmp24, tmp4, tmp27)
tmp29 = tmp20 + tmp28
tmp30 = tl.full([1], 3, tl.int64)
tmp31 = tmp30 == tmp1
tmp32 = tl.where(tmp31, tmp3, tmp4)
tmp33 = tmp32 != tmp4
tmp35 = tl_math.exp(tmp34)
tmp36 = tmp35 + tmp9
tmp37 = tl.where(tmp33, tmp4, tmp36)
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + x0, tmp38, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_11(
in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp3 = tl.load(in_ptr0 + (48 + 5 * x0), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (48 + x2), xmask)
tmp18 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = tl_math.exp(tmp3)
tmp5 = x0
tmp6 = tmp0 == tmp5
tmp7 = 1.0
tmp8 = 0.0
tmp9 = tl.where(tmp6, tmp7, tmp8)
tmp10 = tmp9 != tmp8
tmp12 = tl_math.exp(tmp11)
tmp13 = 1e-05
tmp14 = tmp12 + tmp13
tmp15 = tl.where(tmp10, tmp8, tmp14)
tmp16 = -tmp15
tmp17 = tmp5 == tmp0
tmp19 = tl.where(tmp17, tmp18, tmp8)
tmp20 = tmp16 + tmp19
tmp21 = tl.where(tmp2, tmp4, tmp20)
tl.store(out_ptr0 + x2, tmp21, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_12(in_ptr0, in_ptr1, in_ptr2, 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
tmp5 = tl.load(in_ptr0 + (32 + x2), xmask)
tmp7 = tl.load(in_ptr1 + 5 * x0, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + x2, xmask)
tmp17 = tl.load(in_ptr2 + (32 + x2), xmask)
tmp20 = tl.load(in_ptr0 + (32 + 5 * x0), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 2, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp6 = tl_math.exp(tmp5)
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = x1
tmp12 = tmp11 == tmp3
tmp14 = tmp6 * tmp13
tmp15 = tl.where(tmp12, tmp9, tmp14)
tmp16 = tmp10 - tmp15
tmp18 = tl.where(tmp1, tmp16, tmp17)
tmp19 = tmp2 == tmp11
tmp21 = tl_math.exp(tmp20)
tmp23 = tmp21 * tmp22
tmp24 = tl.where(tmp19, tmp23, tmp9)
tmp25 = tmp18 + tmp24
tl.store(out_ptr0 + x2, tmp25, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_13(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 // 16
x3 = xindex % 16
x0 = xindex % 4
x1 = xindex // 4 % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (32 + x3), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + 5 * x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_out_ptr0 + x5, xmask)
tmp0 = x2
tmp1 = tl.full([1], 2, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = tmp4 == tmp5
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = 0.0
tmp12 = tl.where(tmp6, tmp11, tmp10)
tmp13 = x1
tmp14 = tmp13 == tmp5
tmp16 = tmp8 * tmp15
tmp17 = tl.where(tmp14, tmp11, tmp16)
tmp18 = tmp12 - tmp17
tmp20 = tl.where(tmp2, tmp18, tmp19)
tmp21 = tl.where(tmp2, tmp3, tmp20)
tl.store(in_out_ptr0 + x5, tmp21, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_14(in_ptr0, in_ptr1, in_ptr2, 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
tmp5 = tl.load(in_ptr0 + (48 + x2), xmask)
tmp7 = tl.load(in_ptr1 + 5 * x0, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + x2, xmask)
tmp17 = tl.load(in_ptr2 + (48 + x2), xmask)
tmp20 = tl.load(in_ptr0 + (48 + 5 * x0), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp0 = tl.full([1], 3, tl.int32)
tmp1 = tmp0 == tmp0
tmp2 = x0
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = tmp2 == tmp3
tmp6 = tl_math.exp(tmp5)
tmp8 = tmp6 * tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = x1
tmp12 = tmp11 == tmp3
tmp14 = tmp6 * tmp13
tmp15 = tl.where(tmp12, tmp9, tmp14)
tmp16 = tmp10 - tmp15
tmp18 = tl.where(tmp1, tmp16, tmp17)
tmp19 = tmp2 == tmp11
tmp21 = tl_math.exp(tmp20)
tmp23 = tmp21 * tmp22
tmp24 = tl.where(tmp19, tmp23, tmp9)
tmp25 = tmp18 + tmp24
tl.store(out_ptr0 + x2, tmp25, xmask)
@triton.jit
def triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_15(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 // 16
x3 = xindex % 16
x0 = xindex % 4
x1 = xindex // 4 % 4
x5 = xindex
tmp3 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (48 + x3), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + 5 * x0, xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_out_ptr0 + x5, xmask)
tmp0 = x2
tmp1 = tl.full([1], 3, tl.int32)
tmp2 = tmp0 == tmp1
tmp4 = x0
tmp5 = tl.full([1], 0, tl.int32)
tmp6 = tmp4 == tmp5
tmp8 = tl_math.exp(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = 0.0
tmp12 = tl.where(tmp6, tmp11, tmp10)
tmp13 = x1
tmp14 = tmp13 == tmp5
tmp16 = tmp8 * tmp15
tmp17 = tl.where(tmp14, tmp11, tmp16)
tmp18 = tmp12 - tmp17
tmp20 = tl.where(tmp2, tmp18, tmp19)
tmp21 = tl.where(tmp2, tmp3, tmp20)
tl.store(in_out_ptr0 + x5, tmp21, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4,), (1,), torch.float32)
get_raw_stream(0)
triton_poi_fused_eye_masked_fill_ne_sum_0[grid(4)](arg0_1, buf0, 4,
XBLOCK=4, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_1[
grid(16)](arg0_1, buf0, buf1, 16, XBLOCK=16, num_warps=1,
num_stages=1)
buf2 = torch.ops.aten.linalg_inv_ex.default(buf1)
buf3 = buf2[0]
del buf2
buf5 = buf0
del buf0
triton_poi_fused_eye_masked_fill_ne_sum_2[grid(4)](arg0_1, buf5, 4,
XBLOCK=4, num_warps=1, num_stages=1)
buf6 = buf1
del buf1
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_3[
grid(16)](arg0_1, buf5, buf6, 16, XBLOCK=16, num_warps=1,
num_stages=1)
buf7 = torch.ops.aten.linalg_inv_ex.default(buf6)
buf8 = buf7[0]
del buf7
buf10 = buf6
del buf6
triton_poi_fused_add_diag_embed_4[grid(16)](arg0_1, buf3, buf10, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf11 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_5[grid(64)
](buf10, arg0_1, buf3, buf11, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf10
buf12 = buf5
del buf5
triton_poi_fused_eye_masked_fill_ne_sum_6[grid(4)](arg0_1, buf12, 4,
XBLOCK=4, num_warps=1, num_stages=1)
buf13 = reinterpret_tensor(buf3, (4, 4), (4, 1), 0)
del buf3
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_7[
grid(16)](arg0_1, buf12, buf13, 16, XBLOCK=16, num_warps=1,
num_stages=1)
buf14 = torch.ops.aten.linalg_inv_ex.default(buf13)
buf15 = buf14[0]
del buf14
buf17 = buf13
del buf13
triton_poi_fused_add_diag_embed_8[grid(16)](arg0_1, buf8, buf11,
buf17, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf18 = buf11
del buf11
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_9[grid(64)
](buf18, buf17, arg0_1, buf8, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf17
buf19 = buf12
del buf12
triton_poi_fused_eye_masked_fill_ne_sum_10[grid(4)](arg0_1, buf19,
4, XBLOCK=4, num_warps=1, num_stages=1)
buf20 = reinterpret_tensor(buf8, (4, 4), (4, 1), 0)
del buf8
triton_poi_fused_add_diag_embed_diagonal_copy_exp_eye_masked_fill_ne_neg_11[
grid(16)](arg0_1, buf19, buf20, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del buf19
buf21 = torch.ops.aten.linalg_inv_ex.default(buf20)
buf22 = buf21[0]
del buf21
buf24 = buf20
del buf20
triton_poi_fused_add_diag_embed_12[grid(16)](arg0_1, buf15, buf18,
buf24, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf25 = buf18
del buf18
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_13[grid(64)
](buf25, buf24, arg0_1, buf15, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf15
buf26 = buf24
del buf24
triton_poi_fused_add_diag_embed_14[grid(16)](arg0_1, buf22, buf25,
buf26, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf27 = buf25
del buf25
triton_poi_fused_add_diag_embed_exp_fill_lift_fresh_mul_sub_15[grid(64)
](buf27, buf26, arg0_1, buf22, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del arg0_1
del buf22
del buf26
return buf27,
class MatrixTreeNew(nn.Module):
"""Implementation of the matrix-tree theorem for computing marginals
of non-projective dependency parsing. This attention layer is used
in the paper "Learning Structured Text Representations"
:cite:`DBLP:journals/corr/LiuL17d`.
"""
def __init__(self, eps=1e-05):
self.eps = eps
super(MatrixTreeNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DenDen047/data2text-macro-plan-py | MatrixTree | false | 7,981 | [
"MIT"
] | 20 | bb01ec6e23dab28c1e969f23bd55776b597fb995 | https://github.com/DenDen047/data2text-macro-plan-py/tree/bb01ec6e23dab28c1e969f23bd55776b597fb995 | import torch
import torch.nn as nn
import torch.cuda
import torch.distributed
class Model(nn.Module):
"""Implementation of the matrix-tree theorem for computing marginals
of non-projective dependency parsing. This attention layer is used
in the paper "Learning Structured Text Representations"
:cite:`DBLP:journals/corr/LiuL17d`.
"""
def __init__(self, eps=1e-05):
self.eps = eps
super().__init__()
def forward(self, input):
laplacian = input.exp() + self.eps
output = input.clone()
for b in range(input.size(0)):
lap = laplacian[b].masked_fill(torch.eye(input.size(1), device=
input.device).ne(0), 0)
lap = -lap + torch.diag(lap.sum(0))
lap[0] = input[b].diag().exp()
inv_laplacian = lap.inverse()
factor = inv_laplacian.diag().unsqueeze(1).expand_as(input[b]
).transpose(0, 1)
term1 = input[b].exp().mul(factor).clone()
term2 = input[b].exp().mul(inv_laplacian.transpose(0, 1)).clone()
term1[:, 0] = 0
term2[0] = 0
output[b] = term1 - term2
roots_output = input[b].diag().exp().mul(inv_laplacian.
transpose(0, 1)[0])
output[b] = output[b] + torch.diag(roots_output)
return output
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return []
|
SpatialAttention | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/g4/cg4ktrykrbz7pvagnx6ibqvgol6i2m6cad224574tjik5bwhlbin.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# x => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%mean, %getitem], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 16) % 2
x0 = xindex % 16
x2 = (xindex // 32)
x3 = xindex
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp9 = tmp7 + tmp8
tmp10 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp9 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp4, tmp13, tmp14)
tmp16 = tmp0 >= tmp3
tmp17 = tl.full([1], 2, tl.int64)
tmp18 = tmp0 < tmp17
tmp19 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp20 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp21 = triton_helpers.maximum(tmp19, tmp20)
tmp22 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp23 = triton_helpers.maximum(tmp21, tmp22)
tmp24 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp25 = triton_helpers.maximum(tmp23, tmp24)
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp16, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp15, tmp27)
tl.store(out_ptr0 + (x3), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/65/c65iltr2pquzye7migngokhhtu5qf7ejrjl4tw7zz7m5pjbserj7.py
# Topologically Sorted Source Nodes: [sigmoid], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# sigmoid => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution,), kwargs = {})
triton_poi_fused_sigmoid_1 = async_compile.triton('triton_poi_fused_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sigmoid_1(in_out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tl.store(in_out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 7, 7), (98, 49, 7, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, buf0, 128, grid=grid(128), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1), padding=(3, 3), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [sigmoid], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_1.run(buf2, 64, grid=grid(64), stream=stream0)
return (buf2, primals_2, buf0, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 2, 7, 7), (98, 49, 7, 1), device='cuda:0', dtype=torch.float32)
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 SpatialAttention(nn.Module):
def __init__(self, kernel_size=7):
super(SpatialAttention, self).__init__()
assert kernel_size in (3, 7), 'kernel size must be 3 or 7'
padding = 3 if kernel_size == 7 else 1
self.conv = nn.Conv2d(2, 1, kernel_size, padding=padding, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
avgout = torch.mean(x, dim=1, keepdim=True)
maxout, _ = torch.max(x, dim=1, keepdim=True)
x = torch.cat([avgout, maxout], dim=1)
x = self.conv(x)
return self.sigmoid(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 16 % 2
x0 = xindex % 16
x2 = xindex // 32
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp7 = tmp5 + tmp6
tmp8 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp9 = tmp7 + tmp8
tmp10 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp11 = tmp9 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp4, tmp13, tmp14)
tmp16 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp19 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp20 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp21 = triton_helpers.maximum(tmp19, tmp20)
tmp22 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp23 = triton_helpers.maximum(tmp21, tmp22)
tmp24 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp16 & xmask,
eviction_policy='evict_last', other=0.0)
tmp25 = triton_helpers.maximum(tmp23, tmp24)
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp16, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp15, tmp27)
tl.store(out_ptr0 + x3, tmp28, xmask)
@triton.jit
def triton_poi_fused_sigmoid_1(in_out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tl.store(in_out_ptr0 + x0, tmp1, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 7, 7), (98, 49, 7, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(128)](primals_1, buf0, 128, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1),
padding=(3, 3), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = buf1
del buf1
triton_poi_fused_sigmoid_1[grid(64)](buf2, 64, XBLOCK=64, num_warps
=1, num_stages=1)
return buf2, primals_2, buf0, buf2
class SpatialAttentionNew(nn.Module):
def __init__(self, kernel_size=7):
super(SpatialAttentionNew, self).__init__()
assert kernel_size in (3, 7), 'kernel size must be 3 or 7'
padding = 3 if kernel_size == 7 else 1
self.conv = nn.Conv2d(2, 1, kernel_size, padding=padding, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.conv.weight
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
| DoubtedSteam/MPANet | SpatialAttention | false | 7,982 | [
"MIT"
] | 25 | fe4f3f1d83c45485b1498786f89ace96c634f187 | https://github.com/DoubtedSteam/MPANet/tree/fe4f3f1d83c45485b1498786f89ace96c634f187 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, kernel_size=7):
super().__init__()
assert kernel_size in (3, 7), 'kernel size must be 3 or 7'
padding = 3 if kernel_size == 7 else 1
self.conv = nn.Conv2d(2, 1, kernel_size, padding=padding, bias=False)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
avgout = torch.mean(x, dim=1, keepdim=True)
maxout, _ = torch.max(x, dim=1, keepdim=True)
x = torch.cat([avgout, maxout], dim=1)
x = self.conv(x)
return self.sigmoid(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
NonSaturatingLoss | # 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_1/inductor_cache/cv/ccv6wuuk2f3t36jk6z6siizklj3kweutyi3yvvr2vh7sazv7nrd4.py
# Topologically Sorted Source Nodes: [probs], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# probs => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%arg0_1, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vm/cvm6me542idst2rxx56hwna5utkvygubsm35ar54r2uw4lutswa4.py
# Topologically Sorted Source Nodes: [probs, sub, add, log_prob, mul, sum_1, mean, neg], Original ATen: [aten._softmax, aten.rsub, aten.add, aten.log, aten.mul, aten.sum, aten.mean, aten.neg]
# Source node to ATen node mapping:
# add => add
# log_prob => log
# mean => mean
# mul => mul
# neg => neg
# probs => div, sum_1
# sub => sub_1
# sum_1 => sum_2
# 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 = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %div), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_1, 1e-12), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, %log), kwargs = {})
# %sum_2 : [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.default](args = (%sum_2,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%mean,), kwargs = {})
triton_per_fused__softmax_add_log_mean_mul_neg_rsub_sum_1 = async_compile.triton('triton_per_fused__softmax_add_log_mean_mul_neg_rsub_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, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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__softmax_add_log_mean_mul_neg_rsub_sum_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__softmax_add_log_mean_mul_neg_rsub_sum_1(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')
tmp1 = tl.load(in_ptr1 + (4*r0), None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr1 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp1 / tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = 1e-12
tmp12 = tmp10 + tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = tmp0 * tmp13
tmp16 = tmp2 / tmp7
tmp17 = tmp9 - tmp16
tmp18 = tmp17 + tmp11
tmp19 = tl_math.log(tmp18)
tmp20 = tmp15 * tmp19
tmp21 = tmp14 + tmp20
tmp23 = tmp4 / tmp7
tmp24 = tmp9 - tmp23
tmp25 = tmp24 + tmp11
tmp26 = tl_math.log(tmp25)
tmp27 = tmp22 * tmp26
tmp28 = tmp21 + tmp27
tmp30 = tmp6 / tmp7
tmp31 = tmp9 - tmp30
tmp32 = tmp31 + tmp11
tmp33 = tl_math.log(tmp32)
tmp34 = tmp29 * tmp33
tmp35 = tmp28 + tmp34
tmp36 = tl.broadcast_to(tmp35, [XBLOCK, RBLOCK])
tmp38 = tl.sum(tmp36, 1)[:, None]
tmp39 = 4.0
tmp40 = tmp38 / tmp39
tmp41 = -tmp40
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp41, 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: [probs], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(arg0_1, buf0, 16, grid=grid(16), stream=stream0)
del arg0_1
buf2 = empty_strided_cuda((), (), torch.float32)
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [probs, sub, add, log_prob, mul, sum_1, mean, neg], Original ATen: [aten._softmax, aten.rsub, aten.add, aten.log, aten.mul, aten.sum, aten.mean, aten.neg]
triton_per_fused__softmax_add_log_mean_mul_neg_rsub_sum_1.run(buf3, arg1_1, buf0, 1, 4, grid=grid(1), stream=stream0)
del arg1_1
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
import torch.nn.functional as F
def non_saturating_loss(logits, targets):
probs = logits.softmax(1)
log_prob = torch.log(1 - probs + 1e-12)
if targets.ndim == 2:
return -(targets * log_prob).sum(1).mean()
else:
return F.nll_loss(log_prob, targets)
class NonSaturatingLoss(torch.nn.Module):
def __init__(self, epsilon=0):
super().__init__()
self.epsilon = epsilon
def forward(self, logits, targets):
if self.epsilon > 0:
n_classes = logits.shape[1]
onehot_targets = F.one_hot(targets, n_classes).float()
targets = (1 - self.epsilon
) * onehot_targets + self.epsilon / n_classes
return non_saturating_loss(logits, targets)
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 math as tl_math
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused__softmax_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
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_per_fused__softmax_add_log_mean_mul_neg_rsub_sum_1(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')
tmp1 = tl.load(in_ptr1 + 4 * r0, None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr1 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp1 / tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = 1e-12
tmp12 = tmp10 + tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = tmp0 * tmp13
tmp16 = tmp2 / tmp7
tmp17 = tmp9 - tmp16
tmp18 = tmp17 + tmp11
tmp19 = tl_math.log(tmp18)
tmp20 = tmp15 * tmp19
tmp21 = tmp14 + tmp20
tmp23 = tmp4 / tmp7
tmp24 = tmp9 - tmp23
tmp25 = tmp24 + tmp11
tmp26 = tl_math.log(tmp25)
tmp27 = tmp22 * tmp26
tmp28 = tmp21 + tmp27
tmp30 = tmp6 / tmp7
tmp31 = tmp9 - tmp30
tmp32 = tmp31 + tmp11
tmp33 = tl_math.log(tmp32)
tmp34 = tmp29 * tmp33
tmp35 = tmp28 + tmp34
tmp36 = tl.broadcast_to(tmp35, [XBLOCK, RBLOCK])
tmp38 = tl.sum(tmp36, 1)[:, None]
tmp39 = 4.0
tmp40 = tmp38 / tmp39
tmp41 = -tmp40
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp41, 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)
get_raw_stream(0)
triton_poi_fused__softmax_0[grid(16)](arg0_1, buf0, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del arg0_1
buf2 = empty_strided_cuda((), (), torch.float32)
buf3 = buf2
del buf2
triton_per_fused__softmax_add_log_mean_mul_neg_rsub_sum_1[grid(1)](buf3
, arg1_1, buf0, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del arg1_1
del buf0
return buf3,
def non_saturating_loss(logits, targets):
probs = logits.softmax(1)
log_prob = torch.log(1 - probs + 1e-12)
if targets.ndim == 2:
return -(targets * log_prob).sum(1).mean()
else:
return F.nll_loss(log_prob, targets)
class NonSaturatingLossNew(torch.nn.Module):
def __init__(self, epsilon=0):
super().__init__()
self.epsilon = epsilon
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| DensoITLab/TeachAugment | NonSaturatingLoss | false | 7,983 | [
"BSD-2-Clause"
] | 20 | 66ec099a0afab99e18531c5437182cfe17dc30c8 | https://github.com/DensoITLab/TeachAugment/tree/66ec099a0afab99e18531c5437182cfe17dc30c8 | import torch
import torch.nn.functional as F
def non_saturating_loss(logits, targets):
probs = logits.softmax(1)
log_prob = torch.log(1 - probs + 1e-12)
if targets.ndim == 2:
return -(targets * log_prob).sum(1).mean()
else:
return F.nll_loss(log_prob, targets)
class Model(torch.nn.Module):
def __init__(self, epsilon=0):
super().__init__()
self.epsilon = epsilon
def forward(self, logits, targets):
if self.epsilon > 0:
n_classes = logits.shape[1]
onehot_targets = F.one_hot(targets, n_classes).float()
targets = (1 - self.epsilon
) * onehot_targets + self.epsilon / n_classes
return non_saturating_loss(logits, targets)
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return []
|
ShakeResNeXt | # 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_1/inductor_cache/nd/cndrb2by2g3ednlxsjtytxjhjdm7bdlrgvzzodhsxvxnpml2di7p.py
# Topologically Sorted Source Nodes: [h, h_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# h => convolution
# h_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1048576],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1048576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 4096) % 64
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, None)
''', device_str='cuda')
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, (64, 3, 3, 3), (27, 9, 3, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (4, 1024), (1024, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [h, h_1], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 1048576, grid=grid(1048576), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [h_2], Original ATen: [aten.avg_pool2d]
buf2 = torch.ops.aten.avg_pool2d.default(buf1, [8, 8], [8, 8], [0, 0], False, True, None)
buf3 = buf2
del buf2
buf4 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf3, (16, 1024), (1024, 1), 0), reinterpret_tensor(primals_4, (1024, 4), (1, 1024), 0), alpha=1, beta=1, out=buf4)
del primals_5
return (buf4, primals_1, primals_3, buf1, reinterpret_tensor(buf3, (16, 1024), (1024, 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((64, 3, 3, 3), (27, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 64, 64), (12288, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 1024), (1024, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
class ShakeShake(torch.autograd.Function):
@staticmethod
def forward(ctx, x1, x2, training=True):
if training:
alpha = torch.FloatTensor(x1.size(0)).uniform_()
alpha = alpha.view(alpha.size(0), 1, 1, 1).expand_as(x1)
else:
alpha = 0.5
return alpha * x1 + (1 - alpha) * x2
@staticmethod
def backward(ctx, grad_output):
beta = torch.FloatTensor(grad_output.size(0)).uniform_()
beta = beta.view(beta.size(0), 1, 1, 1).expand_as(grad_output)
beta = Variable(beta)
return beta * grad_output, (1 - beta) * grad_output, None
class Shortcut(nn.Module):
def __init__(self, in_ch, out_ch, stride):
super(Shortcut, self).__init__()
self.stride = stride
self.conv1 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.conv2 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.bn = nn.BatchNorm2d(out_ch)
def forward(self, x):
h = F.relu(x)
h1 = F.avg_pool2d(h, 1, self.stride)
h1 = self.conv1(h1)
h2 = F.avg_pool2d(F.pad(h, (-1, 1, -1, 1)), 1, self.stride)
h2 = self.conv2(h2)
h = torch.cat((h1, h2), 1)
return self.bn(h)
class ShakeBottleNeck(nn.Module):
def __init__(self, in_ch, mid_ch, out_ch, cardinary, stride=1):
super(ShakeBottleNeck, self).__init__()
self.equal_io = in_ch == out_ch
self.shortcut = None if self.equal_io else Shortcut(in_ch, out_ch,
stride=stride)
self.branch1 = self._make_branch(in_ch, mid_ch, out_ch, cardinary,
stride)
self.branch2 = self._make_branch(in_ch, mid_ch, out_ch, cardinary,
stride)
def forward(self, x):
h1 = self.branch1(x)
h2 = self.branch2(x)
if self.training:
h = ShakeShake.apply(h1, h2, self.training)
else:
h = 0.5 * (h1 + h2)
h0 = x if self.equal_io else self.shortcut(x)
return h + h0
def _make_branch(self, in_ch, mid_ch, out_ch, cardinary, stride=1):
return nn.Sequential(nn.Conv2d(in_ch, mid_ch, 1, padding=0, bias=
False), nn.BatchNorm2d(mid_ch), nn.ReLU(inplace=False), nn.
Conv2d(mid_ch, mid_ch, 3, padding=1, stride=stride, groups=
cardinary, bias=False), nn.BatchNorm2d(mid_ch), nn.ReLU(inplace
=False), nn.Conv2d(mid_ch, out_ch, 1, padding=0, bias=False),
nn.BatchNorm2d(out_ch))
class ShakeResNeXt(nn.Module):
def __init__(self, depth, w_base, cardinary, label):
super(ShakeResNeXt, self).__init__()
n_units = (depth - 2) // 9
n_chs = [64, 128, 256, 1024]
self.n_chs = n_chs
self.in_ch = n_chs[0]
self.c_in = nn.Conv2d(3, n_chs[0], 3, padding=1)
self.layer1 = self._make_layer(n_units, n_chs[0], w_base, cardinary)
self.layer2 = self._make_layer(n_units, n_chs[1], w_base, cardinary, 2)
self.layer3 = self._make_layer(n_units, n_chs[2], w_base, cardinary, 2)
self.fc_out = nn.Linear(n_chs[3], label)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def forward(self, x):
h = self.c_in(x)
h = self.layer1(h)
h = self.layer2(h)
h = self.layer3(h)
h = F.relu(h)
h = F.avg_pool2d(h, 8)
h = h.view(-1, self.n_chs[3])
h = self.fc_out(h)
return h
def _make_layer(self, n_units, n_ch, w_base, cardinary, stride=1):
layers = []
mid_ch, out_ch = n_ch * (w_base // 64) * cardinary, n_ch * 4
for i in range(n_units):
layers.append(ShakeBottleNeck(self.in_ch, mid_ch, out_ch,
cardinary, stride=stride))
self.in_ch, stride = out_ch, 1
return nn.Sequential(*layers)
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return [[], {'depth': 1, 'w_base': 4, 'cardinary': 4, 'label': 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 math
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
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):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 4096 % 64
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, None)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (64, 3, 3, 3), (27, 9, 3, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (4, 1024), (1024, 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, 64, 64, 64), (262144, 4096, 64, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(1048576)](buf1, primals_2,
1048576, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_2
buf2 = torch.ops.aten.avg_pool2d.default(buf1, [8, 8], [8, 8], [0,
0], False, True, None)
buf3 = buf2
del buf2
buf4 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf3, (16, 1024),
(1024, 1), 0), reinterpret_tensor(primals_4, (1024, 4), (1,
1024), 0), alpha=1, beta=1, out=buf4)
del primals_5
return buf4, primals_1, primals_3, buf1, reinterpret_tensor(buf3, (16,
1024), (1024, 1), 0), primals_4
class ShakeShake(torch.autograd.Function):
@staticmethod
def forward(ctx, x1, x2, training=True):
if training:
alpha = torch.FloatTensor(x1.size(0)).uniform_()
alpha = alpha.view(alpha.size(0), 1, 1, 1).expand_as(x1)
else:
alpha = 0.5
return alpha * x1 + (1 - alpha) * x2
@staticmethod
def backward(ctx, grad_output):
beta = torch.FloatTensor(grad_output.size(0)).uniform_()
beta = beta.view(beta.size(0), 1, 1, 1).expand_as(grad_output)
beta = Variable(beta)
return beta * grad_output, (1 - beta) * grad_output, None
class Shortcut(nn.Module):
def __init__(self, in_ch, out_ch, stride):
super(Shortcut, self).__init__()
self.stride = stride
self.conv1 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.conv2 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.bn = nn.BatchNorm2d(out_ch)
def forward(self, x):
h = F.relu(x)
h1 = F.avg_pool2d(h, 1, self.stride)
h1 = self.conv1(h1)
h2 = F.avg_pool2d(F.pad(h, (-1, 1, -1, 1)), 1, self.stride)
h2 = self.conv2(h2)
h = torch.cat((h1, h2), 1)
return self.bn(h)
class ShakeBottleNeck(nn.Module):
def __init__(self, in_ch, mid_ch, out_ch, cardinary, stride=1):
super(ShakeBottleNeck, self).__init__()
self.equal_io = in_ch == out_ch
self.shortcut = None if self.equal_io else Shortcut(in_ch, out_ch,
stride=stride)
self.branch1 = self._make_branch(in_ch, mid_ch, out_ch, cardinary,
stride)
self.branch2 = self._make_branch(in_ch, mid_ch, out_ch, cardinary,
stride)
def forward(self, x):
h1 = self.branch1(x)
h2 = self.branch2(x)
if self.training:
h = ShakeShake.apply(h1, h2, self.training)
else:
h = 0.5 * (h1 + h2)
h0 = x if self.equal_io else self.shortcut(x)
return h + h0
def _make_branch(self, in_ch, mid_ch, out_ch, cardinary, stride=1):
return nn.Sequential(nn.Conv2d(in_ch, mid_ch, 1, padding=0, bias=
False), nn.BatchNorm2d(mid_ch), nn.ReLU(inplace=False), nn.
Conv2d(mid_ch, mid_ch, 3, padding=1, stride=stride, groups=
cardinary, bias=False), nn.BatchNorm2d(mid_ch), nn.ReLU(inplace
=False), nn.Conv2d(mid_ch, out_ch, 1, padding=0, bias=False),
nn.BatchNorm2d(out_ch))
class ShakeResNeXtNew(nn.Module):
def __init__(self, depth, w_base, cardinary, label):
super(ShakeResNeXtNew, self).__init__()
n_units = (depth - 2) // 9
n_chs = [64, 128, 256, 1024]
self.n_chs = n_chs
self.in_ch = n_chs[0]
self.c_in = nn.Conv2d(3, n_chs[0], 3, padding=1)
self.layer1 = self._make_layer(n_units, n_chs[0], w_base, cardinary)
self.layer2 = self._make_layer(n_units, n_chs[1], w_base, cardinary, 2)
self.layer3 = self._make_layer(n_units, n_chs[2], w_base, cardinary, 2)
self.fc_out = nn.Linear(n_chs[3], label)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def _make_layer(self, n_units, n_ch, w_base, cardinary, stride=1):
layers = []
mid_ch, out_ch = n_ch * (w_base // 64) * cardinary, n_ch * 4
for i in range(n_units):
layers.append(ShakeBottleNeck(self.in_ch, mid_ch, out_ch,
cardinary, stride=stride))
self.in_ch, stride = out_ch, 1
return nn.Sequential(*layers)
def forward(self, input_0):
primals_1 = self.c_in.weight
primals_2 = self.c_in.bias
primals_4 = self.fc_out.weight
primals_5 = self.fc_out.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DensoITLab/TeachAugment | ShakeResNeXt | false | 7,984 | [
"BSD-2-Clause"
] | 20 | 66ec099a0afab99e18531c5437182cfe17dc30c8 | https://github.com/DensoITLab/TeachAugment/tree/66ec099a0afab99e18531c5437182cfe17dc30c8 | import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
class ShakeShake(torch.autograd.Function):
@staticmethod
def forward(ctx, x1, x2, training=True):
if training:
alpha = torch.FloatTensor(x1.size(0)).uniform_()
alpha = alpha.view(alpha.size(0), 1, 1, 1).expand_as(x1)
else:
alpha = 0.5
return alpha * x1 + (1 - alpha) * x2
@staticmethod
def backward(ctx, grad_output):
beta = torch.FloatTensor(grad_output.size(0)).uniform_()
beta = beta.view(beta.size(0), 1, 1, 1).expand_as(grad_output)
beta = Variable(beta)
return beta * grad_output, (1 - beta) * grad_output, None
class Shortcut(nn.Module):
def __init__(self, in_ch, out_ch, stride):
super().__init__()
self.stride = stride
self.conv1 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.conv2 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.bn = nn.BatchNorm2d(out_ch)
def forward(self, x):
h = F.relu(x)
h1 = F.avg_pool2d(h, 1, self.stride)
h1 = self.conv1(h1)
h2 = F.avg_pool2d(F.pad(h, (-1, 1, -1, 1)), 1, self.stride)
h2 = self.conv2(h2)
h = torch.cat((h1, h2), 1)
return self.bn(h)
class ShakeBottleNeck(nn.Module):
def __init__(self, in_ch, mid_ch, out_ch, cardinary, stride=1):
super().__init__()
self.equal_io = in_ch == out_ch
self.shortcut = None if self.equal_io else Shortcut(in_ch, out_ch,
stride=stride)
self.branch1 = self._make_branch(in_ch, mid_ch, out_ch, cardinary,
stride)
self.branch2 = self._make_branch(in_ch, mid_ch, out_ch, cardinary,
stride)
def forward(self, x):
h1 = self.branch1(x)
h2 = self.branch2(x)
if self.training:
h = ShakeShake.apply(h1, h2, self.training)
else:
h = 0.5 * (h1 + h2)
h0 = x if self.equal_io else self.shortcut(x)
return h + h0
def _make_branch(self, in_ch, mid_ch, out_ch, cardinary, stride=1):
return nn.Sequential(nn.Conv2d(in_ch, mid_ch, 1, padding=0, bias=
False), nn.BatchNorm2d(mid_ch), nn.ReLU(inplace=False), nn.
Conv2d(mid_ch, mid_ch, 3, padding=1, stride=stride, groups=
cardinary, bias=False), nn.BatchNorm2d(mid_ch), nn.ReLU(inplace
=False), nn.Conv2d(mid_ch, out_ch, 1, padding=0, bias=False),
nn.BatchNorm2d(out_ch))
class Model(nn.Module):
def __init__(self, depth, w_base, cardinary, label):
super().__init__()
n_units = (depth - 2) // 9
n_chs = [64, 128, 256, 1024]
self.n_chs = n_chs
self.in_ch = n_chs[0]
self.c_in = nn.Conv2d(3, n_chs[0], 3, padding=1)
self.layer1 = self._make_layer(n_units, n_chs[0], w_base, cardinary)
self.layer2 = self._make_layer(n_units, n_chs[1], w_base, cardinary, 2)
self.layer3 = self._make_layer(n_units, n_chs[2], w_base, cardinary, 2)
self.fc_out = nn.Linear(n_chs[3], label)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def forward(self, x):
h = self.c_in(x)
h = self.layer1(h)
h = self.layer2(h)
h = self.layer3(h)
h = F.relu(h)
h = F.avg_pool2d(h, 8)
h = h.view(-1, self.n_chs[3])
h = self.fc_out(h)
return h
def _make_layer(self, n_units, n_ch, w_base, cardinary, stride=1):
layers = []
mid_ch, out_c
# ... truncated (>4000 chars) for memory efficiency |
FiLM | # 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_1/inductor_cache/zu/czul7scqkgxy4zj4xoaqxd5pj7fc4yfabomtbhizo5vc6h7oxvha.py
# Topologically Sorted Source Nodes: [mul, add], Original ATen: [aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %arg2_1), kwargs = {})
triton_poi_fused_add_mul_0 = async_compile.triton('triton_poi_fused_add_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp3 = tl.load(in_ptr2 + (x0), xmask)
tmp2 = tmp0 * tmp1
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, 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, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, add], Original ATen: [aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_0.run(arg0_1, arg1_1, arg2_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
del arg1_1
del arg2_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)
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
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class FiLM(nn.Module):
"""
A Feature-wise Linear Modulation Layer from
'FiLM: Visual Reasoning with a General Conditioning Layer'
"""
def forward(self, x, gammas, betas):
return gammas * x + betas
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
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
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp3 = tl.load(in_ptr2 + x0, xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tl.store(out_ptr0 + x0, tmp4, 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, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mul_0[grid(256)](arg0_1, arg1_1, arg2_1, buf0,
256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
del arg2_1
return buf0,
class FiLMNew(nn.Module):
"""
A Feature-wise Linear Modulation Layer from
'FiLM: Visual Reasoning with a General Conditioning Layer'
"""
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]
| Dou-Yiming/YouRefIt_ERU | FiLM | false | 7,985 | [
"MIT"
] | 13 | 2a8e849380ed2d253c467b1af744a514bc171372 | https://github.com/Dou-Yiming/YouRefIt_ERU/tree/2a8e849380ed2d253c467b1af744a514bc171372 | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class Model(nn.Module):
"""
A Feature-wise Linear Modulation Layer from
'FiLM: Visual Reasoning with a General Conditioning Layer'
"""
def forward(self, x, gammas, betas):
return gammas * x + betas
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 []
|
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_1/inductor_cache/no/cnon5ajue75qf7yiwbkruc77mekmrelvorvagtbhdc4kdbzwzdin.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# multi_head_attention_forward => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_6, 1.0), kwargs = {})
triton_poi_fused_mul_0 = async_compile.triton('triton_poi_fused_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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 = 1.0
tmp4 = tmp2 * tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/a5/ca56rgpdjbilaspdaau44lnrilsxekhmcnbwzhtrcgfbilkik27p.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
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = 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_1/inductor_cache/xl/cxls6dl5dz3ua4ilno7rjcfd6m7p4ydnd3mzfaq2cepnph6e2y7h.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# multi_head_attention_forward => 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_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = 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_1/inductor_cache/mf/cmfwwg65sbylkho5zzcgwfl3cw2tru7uwmgab3xaed7bzoxeyhoe.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_10,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_3 = async_compile.triton('triton_poi_fused_clone_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
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 + (4*x1)), xmask & ymask)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/tc/ctci7shbeip5tioyeh6fiaku6tkwoufw2eypxedyp7mdhyexbptb.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x => add
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%squeeze, %primals_2), kwargs = {})
triton_poi_fused_add_4 = async_compile.triton('triton_poi_fused_add_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_4', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_4(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
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')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (12, 4), (4, 1))
assert_size_stride(primals_6, (12, ), (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, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(primals_2, reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.mm]
extern_kernels.mm(primals_2, reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.mm]
extern_kernels.mm(primals_2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
del primals_4
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf0, reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_6, (4, ), (1, ), 4), buf1, reinterpret_tensor(primals_5, (4, 4), (1, 4), 16), alpha=1, beta=1, out=buf4)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_6, (4, ), (1, ), 8), buf2, reinterpret_tensor(primals_5, (4, 4), (1, 4), 32), alpha=1, beta=1, out=buf5)
buf6 = reinterpret_tensor(buf3, (4, 4, 1), (1, 4, 16), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(buf6, primals_6, 16, grid=grid(16), stream=stream0)
del primals_6
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(buf6, reinterpret_tensor(buf4, (4, 1, 4), (1, 1, 4), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf7, buf8, 64, grid=grid(64), stream=stream0)
buf9 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf8, buf9, 64, grid=grid(64), stream=stream0)
del buf8
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(buf9, reinterpret_tensor(buf5, (4, 4, 1), (1, 4, 1), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf10, buf11, 4, 4, grid=grid(4, 4), stream=stream0)
buf12 = reinterpret_tensor(buf10, (4, 4), (4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf11, (4, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf12)
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
triton_poi_fused_add_4.run(buf13, primals_8, primals_2, 16, grid=grid(16), stream=stream0)
del primals_8
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_3], Original ATen: [aten.mm]
extern_kernels.mm(buf13, reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf14)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.addmm(buf13, buf14, reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf15)
return (buf15, primals_2, buf0, buf1, buf2, buf9, reinterpret_tensor(buf11, (4, 4), (4, 1), 0), buf13, buf14, primals_10, primals_9, primals_7, reinterpret_tensor(buf5, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf6, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf4, (4, 4, 1), (1, 4, 1), 0), reinterpret_tensor(primals_5, (4, 4), (4, 1), 32), reinterpret_tensor(primals_5, (4, 4), (4, 1), 16), reinterpret_tensor(primals_5, (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, 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, 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((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((12, ), (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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.onnx
class TransformerLayer(nn.Module):
def __init__(self, channels, num_heads):
super().__init__()
self.q = nn.Linear(channels, channels, bias=False)
self.k = nn.Linear(channels, channels, bias=False)
self.v = nn.Linear(channels, channels, bias=False)
self.ma = nn.MultiheadAttention(embed_dim=channels, num_heads=num_heads
)
self.fc1 = nn.Linear(channels, channels, bias=False)
self.fc2 = nn.Linear(channels, channels, bias=False)
def forward(self, x):
x = self.ma(self.q(x), self.k(x), self.v(x))[0] + x
x = self.fc2(self.fc1(x)) + x
return x
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'channels': 4, 'num_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 math as tl_math
import torch.nn as nn
import torch.onnx
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_mul_0(in_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 = 1.0
tmp4 = tmp2 * tmp3
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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 4
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 + 4 * x1), xmask & ymask)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_4(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
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)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (12, 4), (4, 1))
assert_size_stride(primals_6, (12,), (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, 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(primals_2, reinterpret_tensor(primals_1, (4, 4),
(1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_2, reinterpret_tensor(primals_3, (4, 4),
(1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_2, reinterpret_tensor(primals_4, (4, 4),
(1, 4), 0), out=buf2)
del primals_4
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf0, reinterpret_tensor(primals_5, (4, 4), (1, 4
), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_6, (4,), (1,), 4),
buf1, reinterpret_tensor(primals_5, (4, 4), (1, 4), 16), alpha=
1, beta=1, out=buf4)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_6, (4,), (1,), 8),
buf2, reinterpret_tensor(primals_5, (4, 4), (1, 4), 32), alpha=
1, beta=1, out=buf5)
buf6 = reinterpret_tensor(buf3, (4, 4, 1), (1, 4, 16), 0)
del buf3
get_raw_stream(0)
triton_poi_fused_mul_0[grid(16)](buf6, primals_6, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_6
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(buf6, reinterpret_tensor(buf4, (4, 1, 4), (1, 1,
4), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(64)](buf7, buf8, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf9 = buf7
del buf7
triton_poi_fused__softmax_2[grid(64)](buf8, buf9, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf8
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(buf9, reinterpret_tensor(buf5, (4, 4, 1), (1, 4,
1), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
triton_poi_fused_clone_3[grid(4, 4)](buf10, buf11, 4, 4, XBLOCK=4,
YBLOCK=4, num_warps=1, num_stages=1)
buf12 = reinterpret_tensor(buf10, (4, 4), (4, 1), 0)
del buf10
extern_kernels.mm(reinterpret_tensor(buf11, (4, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf12)
buf13 = buf12
del buf12
triton_poi_fused_add_4[grid(16)](buf13, primals_8, primals_2, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_8
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf13, reinterpret_tensor(primals_9, (4, 4), (1,
4), 0), out=buf14)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(buf13, buf14, reinterpret_tensor(primals_10, (
4, 4), (1, 4), 0), alpha=1, beta=1, out=buf15)
return buf15, primals_2, buf0, buf1, buf2, buf9, reinterpret_tensor(buf11,
(4, 4), (4, 1), 0
), buf13, buf14, primals_10, primals_9, primals_7, reinterpret_tensor(
buf5, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf6, (4, 1, 4),
(1, 1, 4), 0), reinterpret_tensor(buf4, (4, 4, 1), (1, 4, 1), 0
), reinterpret_tensor(primals_5, (4, 4), (4, 1), 32
), reinterpret_tensor(primals_5, (4, 4), (4, 1), 16
), reinterpret_tensor(primals_5, (4, 4), (4, 1), 0)
class TransformerLayerNew(nn.Module):
def __init__(self, channels, num_heads):
super().__init__()
self.q = nn.Linear(channels, channels, bias=False)
self.k = nn.Linear(channels, channels, bias=False)
self.v = nn.Linear(channels, channels, bias=False)
self.ma = nn.MultiheadAttention(embed_dim=channels, num_heads=num_heads
)
self.fc1 = nn.Linear(channels, channels, bias=False)
self.fc2 = nn.Linear(channels, channels, bias=False)
def forward(self, input_0):
primals_1 = self.q.weight
primals_2 = self.k.weight
primals_3 = self.v.weight
primals_5 = self.ma.in_proj_weight
primals_6 = self.ma.in_proj_bias
primals_4 = self.ma.out_proj.weight
primals_8 = self.ma.out_proj.bias
primals_7 = self.fc1.weight
primals_9 = self.fc2.weight
primals_10 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return output[0]
| DDGRCF/YOLOX_OBB | TransformerLayer | false | 7,986 | [
"Apache-2.0"
] | 39 | 27b80953306492b8bc83b86b1353d8cee01ef9b6 | https://github.com/DDGRCF/YOLOX_OBB/tree/27b80953306492b8bc83b86b1353d8cee01ef9b6 | import torch
import torch.nn as nn
import torch.onnx
class Model(nn.Module):
def __init__(self, channels, num_heads):
super().__init__()
self.q = nn.Linear(channels, channels, bias=False)
self.k = nn.Linear(channels, channels, bias=False)
self.v = nn.Linear(channels, channels, bias=False)
self.ma = nn.MultiheadAttention(embed_dim=channels, num_heads=num_heads
)
self.fc1 = nn.Linear(channels, channels, bias=False)
self.fc2 = nn.Linear(channels, channels, bias=False)
def forward(self, x):
x = self.ma(self.q(x), self.k(x), self.v(x))[0] + x
x = self.fc2(self.fc1(x)) + x
return x
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [4, 4]
|
QuadriLinearScore | # 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_1/inductor_cache/mz/cmzab224ctbcchtlimpgvqkspaz3nkja2roqpom4mydflmrj3hj7.py
# Topologically Sorted Source Nodes: [g0], Original ATen: [aten.clone, aten._unsafe_view]
# Source node to ATen node mapping:
# g0 => clone, view
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%slice_2,), kwargs = {memory_format: torch.contiguous_format})
# %view : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone, [12, 4]), kwargs = {})
triton_poi_fused__unsafe_view_clone_0 = async_compile.triton('triton_poi_fused__unsafe_view_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_view_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_view_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*(x1 % 3)) + (16*(x1 // 3))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qm/cqm3wuwynxkh6pgn4sqcx6amwqhmvhbfyakbcepgfx6jyjtbtlvd.py
# Topologically Sorted Source Nodes: [g1], Original ATen: [aten.clone, aten._unsafe_view]
# Source node to ATen node mapping:
# g1 => clone_1, view_2
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%slice_4,), kwargs = {memory_format: torch.contiguous_format})
# %view_2 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone_1, [12, 4]), kwargs = {})
triton_poi_fused__unsafe_view_clone_1 = async_compile.triton('triton_poi_fused__unsafe_view_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_view_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_view_clone_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4 + x0 + (4*(x1 % 3)) + (16*(x1 // 3))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/sc/cscrjp3hgaurq2sdlsdzgevaj23dewosj2dxqxdka6sucoj6gc7l.py
# Topologically Sorted Source Nodes: [temp012], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# temp012 => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute, %permute_1), kwargs = {})
triton_poi_fused_mul_2 = async_compile.triton('triton_poi_fused_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 19008
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 396
x2 = (xindex // 1584)
x3 = xindex % 1584
tmp0 = tl.load(in_ptr0 + (x0 + (396*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + (396*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 + (x3 + (1600*x2)), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pw/cpwvi2tv5jwixczxm6tdxkennjv3x5vpd3vgaeiu5ubjwv4kax3a.py
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.bmm, aten.transpose]
# Source node to ATen node mapping:
# score => bmm
# Graph fragment:
# %bmm : [num_users=1] = call_function[target=torch.ops.aten.bmm.default](args = (%view_4, %view_5), kwargs = {})
# %permute_8 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%view_4, [0, 2, 1]), kwargs = {})
triton_poi_fused_bmm_transpose_3 = async_compile.triton('triton_poi_fused_bmm_transpose_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_3(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 19008
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 396
x1 = (xindex // 396)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (396*(x1 % 4)) + (1600*(x1 // 4))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
tl.store(out_ptr1 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/xs/cxsqvqteve6uvby5qw37bjyxf7g4tkwjggk3xvdttbk55tby2v2w.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten.div]
# Source node to ATen node mapping:
# score_1 => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_7, 19.8997487421324), kwargs = {})
triton_poi_fused_div_4 = async_compile.triton('triton_poi_fused_div_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_4', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_4(in_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
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = 0.050251890762960605
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 396), (396, 1))
assert_size_stride(primals_3, (4, 396), (396, 1))
assert_size_stride(primals_4, (4, 20), (20, 1))
assert_size_stride(primals_5, (20, 396), (396, 1))
assert_size_stride(primals_6, (20, 396), (396, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((12, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [g0], Original ATen: [aten.clone, aten._unsafe_view]
stream0 = get_raw_stream(0)
triton_poi_fused__unsafe_view_clone_0.run(primals_1, buf0, 48, grid=grid(48), stream=stream0)
buf1 = empty_strided_cuda((12, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g0], Original ATen: [aten.mm]
extern_kernels.mm(buf0, primals_2, out=buf1)
del primals_2
buf2 = empty_strided_cuda((12, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [g1], Original ATen: [aten.clone, aten._unsafe_view]
triton_poi_fused__unsafe_view_clone_1.run(primals_1, buf2, 48, grid=grid(48), stream=stream0)
del primals_1
buf3 = empty_strided_cuda((12, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g1], Original ATen: [aten.mm]
extern_kernels.mm(buf2, primals_3, out=buf3)
del primals_3
buf4 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g2], Original ATen: [aten.mm]
extern_kernels.mm(primals_4, primals_5, out=buf4)
buf5 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g3], Original ATen: [aten.mm]
extern_kernels.mm(primals_4, primals_6, out=buf5)
buf6 = empty_strided_cuda((4, 3, 4, 396), (4800, 1600, 396, 1), torch.float32)
# Topologically Sorted Source Nodes: [temp012], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(buf1, buf3, buf4, buf6, 19008, grid=grid(19008), stream=stream0)
buf7 = empty_strided_cuda((1, 48, 396), (19008, 396, 1), torch.float32)
buf10 = empty_strided_cuda((1, 396, 48), (19008, 1, 396), torch.float32)
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.bmm, aten.transpose]
triton_poi_fused_bmm_transpose_3.run(buf6, buf7, buf10, 19008, grid=grid(19008), stream=stream0)
del buf6
buf8 = empty_strided_cuda((1, 48, 4), (192, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.bmm]
extern_kernels.bmm(buf7, reinterpret_tensor(buf5, (1, 396, 4), (0, 1, 396), 0), out=buf8)
del buf7
buf9 = reinterpret_tensor(buf8, (4, 3, 4, 4), (48, 16, 4, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten.div]
triton_poi_fused_div_4.run(buf9, 192, grid=grid(192), stream=stream0)
return (buf9, buf1, buf3, buf4, buf10, reinterpret_tensor(buf5, (1, 4, 396), (396, 396, 1), 0), reinterpret_tensor(primals_4, (20, 4), (1, 20), 0), reinterpret_tensor(primals_6, (396, 20), (1, 396), 0), reinterpret_tensor(primals_5, (396, 20), (1, 396), 0), reinterpret_tensor(buf2, (4, 12), (1, 4), 0), reinterpret_tensor(buf0, (4, 12), (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, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 20), (20, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((20, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((20, 396), (396, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class QuadriLinearScore(nn.Module):
"""
Outer product version of quadrilinear function for sequence labeling.
"""
def __init__(self, wemb_size, tagset_size, temb_size=20, rank=396, std=
0.1545, window_size=1, normalization=True, **kwargs):
"""
Args:
wemb_size: word embedding hidden size
tagset_size: tag set size
temb_size: tag embedding size
rank: rank of the weight tensor
std: standard deviation of the tensor
"""
super(QuadriLinearScore, self).__init__()
self.wemb_size = wemb_size
self.tagset_size = tagset_size
self.temb_size = temb_size
self.rank = rank
self.std = std
self.window_size = window_size
self.normalization = normalization
self.tag_emd = nn.Parameter(torch.Tensor(self.tagset_size, self.
temb_size))
self.T = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.U = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.V = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.W = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.rand_init()
self
def rand_init(self):
"""random initialization
"""
nn.init.uniform_(self.tag_emd, a=math.sqrt(6 / self.temb_size), b=
math.sqrt(6 / self.temb_size))
nn.init.normal_(self.T, std=self.std)
nn.init.normal_(self.U, std=self.std)
nn.init.normal_(self.V, std=self.std)
nn.init.normal_(self.W, std=self.std)
def forward(self, word_emb):
"""
Args:
word_emb: [batch, sent_length, wemb_size]
Returns: Tensor
[batch, sent_length-window_size, tagset_size, tagset_size]
"""
assert word_emb.size(2
) == self.wemb_size, 'batch sizes of encoder and decoder are requires to be equal.'
g0 = torch.matmul(word_emb[:, :-self.window_size], self.U)
g1 = torch.matmul(word_emb[:, self.window_size:], self.T)
g2 = torch.matmul(self.tag_emd, self.V)
g3 = torch.matmul(self.tag_emd, self.W)
temp01 = g0 * g1
temp012 = torch.einsum('nak,bk->nabk', [temp01, g2])
score = torch.einsum('nabk,ck->nabc', [temp012, g3])
if self.normalization:
score = score / math.sqrt(self.rank)
return score
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'wemb_size': 4, 'tagset_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import math
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__unsafe_view_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * (x1 % 3) + 16 * (x1 // 3)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused__unsafe_view_clone_1(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 48
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4 + x0 + 4 * (x1 % 3) + 16 * (x1 // 3)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 19008
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 396
x2 = xindex // 1584
x3 = xindex % 1584
tmp0 = tl.load(in_ptr0 + (x0 + 396 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + 396 * 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 + (x3 + 1600 * x2), tmp4, xmask)
@triton.jit
def triton_poi_fused_bmm_transpose_3(in_ptr0, out_ptr0, out_ptr1, xnumel,
XBLOCK: tl.constexpr):
xnumel = 19008
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 396
x1 = xindex // 396
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 396 * (x1 % 4) + 1600 * (x1 // 4)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
tl.store(out_ptr1 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_div_4(in_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
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = 0.050251890762960605
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + x0, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 396), (396, 1))
assert_size_stride(primals_3, (4, 396), (396, 1))
assert_size_stride(primals_4, (4, 20), (20, 1))
assert_size_stride(primals_5, (20, 396), (396, 1))
assert_size_stride(primals_6, (20, 396), (396, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((12, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__unsafe_view_clone_0[grid(48)](primals_1, buf0, 48,
XBLOCK=64, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((12, 396), (396, 1), torch.float32)
extern_kernels.mm(buf0, primals_2, out=buf1)
del primals_2
buf2 = empty_strided_cuda((12, 4), (4, 1), torch.float32)
triton_poi_fused__unsafe_view_clone_1[grid(48)](primals_1, buf2, 48,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_1
buf3 = empty_strided_cuda((12, 396), (396, 1), torch.float32)
extern_kernels.mm(buf2, primals_3, out=buf3)
del primals_3
buf4 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
extern_kernels.mm(primals_4, primals_5, out=buf4)
buf5 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
extern_kernels.mm(primals_4, primals_6, out=buf5)
buf6 = empty_strided_cuda((4, 3, 4, 396), (4800, 1600, 396, 1),
torch.float32)
triton_poi_fused_mul_2[grid(19008)](buf1, buf3, buf4, buf6, 19008,
XBLOCK=256, num_warps=4, num_stages=1)
buf7 = empty_strided_cuda((1, 48, 396), (19008, 396, 1), torch.float32)
buf10 = empty_strided_cuda((1, 396, 48), (19008, 1, 396), torch.float32
)
triton_poi_fused_bmm_transpose_3[grid(19008)](buf6, buf7, buf10,
19008, XBLOCK=256, num_warps=4, num_stages=1)
del buf6
buf8 = empty_strided_cuda((1, 48, 4), (192, 4, 1), torch.float32)
extern_kernels.bmm(buf7, reinterpret_tensor(buf5, (1, 396, 4), (0,
1, 396), 0), out=buf8)
del buf7
buf9 = reinterpret_tensor(buf8, (4, 3, 4, 4), (48, 16, 4, 1), 0)
del buf8
triton_poi_fused_div_4[grid(192)](buf9, 192, XBLOCK=256, num_warps=
4, num_stages=1)
return buf9, buf1, buf3, buf4, buf10, reinterpret_tensor(buf5, (1, 4,
396), (396, 396, 1), 0), reinterpret_tensor(primals_4, (20, 4), (1,
20), 0), reinterpret_tensor(primals_6, (396, 20), (1, 396), 0
), reinterpret_tensor(primals_5, (396, 20), (1, 396), 0
), reinterpret_tensor(buf2, (4, 12), (1, 4), 0), reinterpret_tensor(
buf0, (4, 12), (1, 4), 0)
class QuadriLinearScoreNew(nn.Module):
"""
Outer product version of quadrilinear function for sequence labeling.
"""
def __init__(self, wemb_size, tagset_size, temb_size=20, rank=396, std=
0.1545, window_size=1, normalization=True, **kwargs):
"""
Args:
wemb_size: word embedding hidden size
tagset_size: tag set size
temb_size: tag embedding size
rank: rank of the weight tensor
std: standard deviation of the tensor
"""
super(QuadriLinearScoreNew, self).__init__()
self.wemb_size = wemb_size
self.tagset_size = tagset_size
self.temb_size = temb_size
self.rank = rank
self.std = std
self.window_size = window_size
self.normalization = normalization
self.tag_emd = nn.Parameter(torch.Tensor(self.tagset_size, self.
temb_size))
self.T = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.U = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.V = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.W = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.rand_init()
self
def rand_init(self):
"""random initialization
"""
nn.init.uniform_(self.tag_emd, a=math.sqrt(6 / self.temb_size), b=
math.sqrt(6 / self.temb_size))
nn.init.normal_(self.T, std=self.std)
nn.init.normal_(self.U, std=self.std)
nn.init.normal_(self.V, std=self.std)
nn.init.normal_(self.W, std=self.std)
def forward(self, input_0):
primals_4 = self.tag_emd
primals_2 = self.T
primals_3 = self.U
primals_5 = self.V
primals_6 = self.W
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| Dadmatech/DadmaTools | QuadriLinearScore | false | 7,987 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import math
import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class Model(nn.Module):
"""
Outer product version of quadrilinear function for sequence labeling.
"""
def __init__(self, wemb_size, tagset_size, temb_size=20, rank=396, std=
0.1545, window_size=1, normalization=True, **kwargs):
"""
Args:
wemb_size: word embedding hidden size
tagset_size: tag set size
temb_size: tag embedding size
rank: rank of the weight tensor
std: standard deviation of the tensor
"""
super().__init__()
self.wemb_size = wemb_size
self.tagset_size = tagset_size
self.temb_size = temb_size
self.rank = rank
self.std = std
self.window_size = window_size
self.normalization = normalization
self.tag_emd = nn.Parameter(torch.Tensor(self.tagset_size, self.
temb_size))
self.T = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.U = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.V = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.W = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.rand_init()
self
def rand_init(self):
"""random initialization
"""
nn.init.uniform_(self.tag_emd, a=math.sqrt(6 / self.temb_size), b=
math.sqrt(6 / self.temb_size))
nn.init.normal_(self.T, std=self.std)
nn.init.normal_(self.U, std=self.std)
nn.init.normal_(self.V, std=self.std)
nn.init.normal_(self.W, std=self.std)
def forward(self, word_emb):
"""
Args:
word_emb: [batch, sent_length, wemb_size]
Returns: Tensor
[batch, sent_length-window_size, tagset_size, tagset_size]
"""
assert word_emb.size(2
) == self.wemb_size, 'batch sizes of encoder and decoder are requires to be equal.'
g0 = torch.matmul(word_emb[:, :-self.window_size], self.U)
g1 = torch.matmul(word_emb[:, self.window_size:], self.T)
g2 = torch.matmul(self.tag_emd, self.V)
g3 = torch.matmul(self.tag_emd, self.W)
temp01 = g0 * g1
temp012 = torch.einsum('nak,bk->nabk', [temp01, g2])
score = torch.einsum('nabk,ck->nabc', [temp012, g3])
if self.normalization:
score = score / math.sqrt(self.rank)
return score
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
GlobalAvgPool2dResNext | # 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_1/inductor_cache/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.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 = (%view, [2]), 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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):
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), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mean], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, arg0_1, 16, 16, grid=grid(16), 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
from torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class GlobalAvgPool2dResNext(nn.Module):
def __init__(self):
"""Global average pooling over the input's spatial dimensions"""
super(GlobalAvgPool2dResNext, self).__init__()
def forward(self, inputs):
in_size = inputs.size()
return inputs.view((in_size[0], in_size[1], -1)).mean(dim=2)
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 torchvision import datasets as datasets
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
@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):
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), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(16)](buf1, arg0_1, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
del arg0_1
return buf1,
class GlobalAvgPool2dResNextNew(nn.Module):
def __init__(self):
"""Global average pooling over the input's spatial dimensions"""
super(GlobalAvgPool2dResNextNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Alibaba-MIIL/ZS_SDL | GlobalAvgPool2dResNext | false | 7,988 | [
"MIT"
] | 20 | 769fe4f57d2d458a7c4b5468a6395c9b296b1dad | https://github.com/Alibaba-MIIL/ZS_SDL/tree/769fe4f57d2d458a7c4b5468a6395c9b296b1dad | import torch
from torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class Model(nn.Module):
def __init__(self):
"""Global average pooling over the input's spatial dimensions"""
super().__init__()
def forward(self, inputs):
in_size = inputs.size()
return inputs.view((in_size[0], in_size[1], -1)).mean(dim=2)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MyUpsample2 | # 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_1/inductor_cache/za/czambx4debjna7weirsx3k25o6zgirc3niva7su5rel266wjplok.py
# Topologically Sorted Source Nodes: [reshape], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# reshape => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_0 = async_compile.triton('triton_poi_fused_clone_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 2) % 4
x3 = (xindex // 16)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x1 + (4*x3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x4), 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)
buf0 = empty_strided_cuda((4, 4, 4, 2, 4, 2), (256, 64, 16, 8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [reshape], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(arg0_1, buf0, 1024, grid=grid(1024), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf0, (4, 4, 8, 8), (256, 64, 8, 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.nn.parallel
import torch.optim
import torch.utils.data.distributed
class MyUpsample2(nn.Module):
def forward(self, x):
return x[:, :, :, None, :, None].expand(-1, -1, -1, 2, -1, 2).reshape(x
.size(0), x.size(1), x.size(2) * 2, x.size(3) * 2)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 2 % 4
x3 = xindex // 16
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x1 + 4 * x3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + x4, 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)
buf0 = empty_strided_cuda((4, 4, 4, 2, 4, 2), (256, 64, 16, 8, 2, 1
), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(1024)](arg0_1, buf0, 1024, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return reinterpret_tensor(buf0, (4, 4, 8, 8), (256, 64, 8, 1), 0),
class MyUpsample2New(nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Dou-Yiming/YouRefIt_ERU | MyUpsample2 | false | 7,989 | [
"MIT"
] | 13 | 2a8e849380ed2d253c467b1af744a514bc171372 | https://github.com/Dou-Yiming/YouRefIt_ERU/tree/2a8e849380ed2d253c467b1af744a514bc171372 | import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class Model(nn.Module):
def forward(self, x):
return x[:, :, :, None, :, None].expand(-1, -1, -1, 2, -1, 2).reshape(x
.size(0), x.size(1), x.size(2) * 2, x.size(3) * 2)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
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_1/inductor_cache/5m/c5mqjkgr5v5727r267mbxh7dsgbrwetftfoe6333a3eri2wcmk42.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.leaky_relu]
# Source node to ATen node mapping:
# x_1 => gt, mul, where
# Graph fragment:
# %gt : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%view_1, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.1), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %view_1, %mul), kwargs = {})
triton_poi_fused_leaky_relu_0 = async_compile.triton('triton_poi_fused_leaky_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_leaky_relu_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_leaky_relu_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.1
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.leaky_relu]
stream0 = get_raw_stream(0)
triton_poi_fused_leaky_relu_0.run(buf0, primals_2, buf1, buf2, 256, grid=grid(256), stream=stream0)
del buf0
del primals_2
return (buf2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
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.dataloader
import torch.nn as nn
import torch.nn
class SharedDropout(nn.Module):
def __init__(self, p=0.5, batch_first=True):
super(SharedDropout, self).__init__()
self.p = p
self.batch_first = batch_first
def extra_repr(self):
s = f'p={self.p}'
if self.batch_first:
s += f', batch_first={self.batch_first}'
return s
def forward(self, x):
if self.training:
if self.batch_first:
mask = self.get_mask(x[:, 0], self.p)
else:
mask = self.get_mask(x[0], self.p)
x *= mask.unsqueeze(1) if self.batch_first else mask
return x
@staticmethod
def get_mask(x, p):
mask = x.new_empty(x.shape).bernoulli_(1 - p)
mask = mask / (1 - p)
return mask
class MLP(nn.Module):
def __init__(self, n_in, n_hidden, dropout=0, identity=False):
super(MLP, self).__init__()
self.linear = nn.Linear(n_in, n_hidden)
self.identity = identity
if not self.identity:
self.activation = nn.LeakyReLU(negative_slope=0.1)
self.dropout = SharedDropout(p=dropout)
self.reset_parameters()
def reset_parameters(self):
nn.init.orthogonal_(self.linear.weight)
nn.init.zeros_(self.linear.bias)
def forward(self, x):
x = self.linear(x)
if not self.identity:
x = self.activation(x)
x = self.dropout(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_in': 4, 'n_hidden': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_leaky_relu_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.1
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_leaky_relu_0[grid(256)](buf0, primals_2, buf1,
buf2, 256, XBLOCK=128, num_warps=4, num_stages=1)
del buf0
del primals_2
return buf2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1
class SharedDropout(nn.Module):
def __init__(self, p=0.5, batch_first=True):
super(SharedDropout, self).__init__()
self.p = p
self.batch_first = batch_first
def extra_repr(self):
s = f'p={self.p}'
if self.batch_first:
s += f', batch_first={self.batch_first}'
return s
def forward(self, x):
if self.training:
if self.batch_first:
mask = self.get_mask(x[:, 0], self.p)
else:
mask = self.get_mask(x[0], self.p)
x *= mask.unsqueeze(1) if self.batch_first else mask
return x
@staticmethod
def get_mask(x, p):
mask = x.new_empty(x.shape).bernoulli_(1 - p)
mask = mask / (1 - p)
return mask
class MLPNew(nn.Module):
def __init__(self, n_in, n_hidden, dropout=0, identity=False):
super(MLPNew, self).__init__()
self.linear = nn.Linear(n_in, n_hidden)
self.identity = identity
if not self.identity:
self.activation = nn.LeakyReLU(negative_slope=0.1)
self.dropout = SharedDropout(p=dropout)
self.reset_parameters()
def reset_parameters(self):
nn.init.orthogonal_(self.linear.weight)
nn.init.zeros_(self.linear.bias)
def forward(self, input_0):
primals_1 = self.linear.weight
primals_2 = self.linear.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| Dadmatech/DadmaTools | MLP | false | 7,990 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class SharedDropout(nn.Module):
def __init__(self, p=0.5, batch_first=True):
super().__init__()
self.p = p
self.batch_first = batch_first
def extra_repr(self):
s = f'p={self.p}'
if self.batch_first:
s += f', batch_first={self.batch_first}'
return s
def forward(self, x):
if self.training:
if self.batch_first:
mask = self.get_mask(x[:, 0], self.p)
else:
mask = self.get_mask(x[0], self.p)
x *= mask.unsqueeze(1) if self.batch_first else mask
return x
@staticmethod
def get_mask(x, p):
mask = x.new_empty(x.shape).bernoulli_(1 - p)
mask = mask / (1 - p)
return mask
class Model(nn.Module):
def __init__(self, n_in, n_hidden, dropout=0, identity=False):
super().__init__()
self.linear = nn.Linear(n_in, n_hidden)
self.identity = identity
if not self.identity:
self.activation = nn.LeakyReLU(negative_slope=0.1)
self.dropout = SharedDropout(p=dropout)
self.reset_parameters()
def reset_parameters(self):
nn.init.orthogonal_(self.linear.weight)
nn.init.zeros_(self.linear.bias)
def forward(self, x):
x = self.linear(x)
if not self.identity:
x = self.activation(x)
x = self.dropout(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
BiaffineAttention | # 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_1/inductor_cache/bp/cbpkro4y3tdg4ug7qmmydmvue6tznnxkwjb7b4r7ju2y3aalctcv.py
# Topologically Sorted Source Nodes: [add, add_1, output_2, output_3], Original ATen: [aten.add, aten._softmax]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# output_2 => add_2
# output_3 => amax
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_13, %view_5), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %permute_3), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %primals_10), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_2, [1], True), kwargs = {})
triton_poi_fused__softmax_add_0 = async_compile.triton('triton_poi_fused__softmax_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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 10, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x0), xmask)
tmp5 = tl.load(in_ptr3 + (0))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp8 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp7 = tmp4 + tmp6
tmp10 = tmp8 + tmp9
tmp11 = tmp10 + tmp3
tmp12 = tmp11 + tmp6
tmp13 = triton_helpers.maximum(tmp7, tmp12)
tmp16 = tmp14 + tmp15
tmp17 = tmp16 + tmp3
tmp18 = tmp17 + tmp6
tmp19 = triton_helpers.maximum(tmp13, tmp18)
tmp22 = tmp20 + tmp21
tmp23 = tmp22 + tmp3
tmp24 = tmp23 + tmp6
tmp25 = triton_helpers.maximum(tmp19, tmp24)
tl.store(out_ptr0 + (x0), tmp25, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/o7/co7d4yyfde7favjhlve466f3vkmmrgfappddg5regmbx6egs544c.py
# Topologically Sorted Source Nodes: [add, add_1, output_2, output_3], Original ATen: [aten.add, aten._softmax]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# output_2 => add_2
# output_3 => exp, sub
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_13, %view_5), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %permute_3), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %primals_10), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_add_1 = async_compile.triton('triton_poi_fused__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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_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 + (x2), xmask)
tmp1 = tl.load(in_out_ptr0 + (x2), xmask)
tmp3 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (0))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp8 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp7 = tmp4 + tmp6
tmp9 = tmp7 - tmp8
tmp10 = tl_math.exp(tmp9)
tl.store(in_out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wp/cwpr2jp7bpqwkbaxgttcl2ekjdtkf6lezuspoiunutzd3j2qgeyk.py
# Topologically Sorted Source Nodes: [output_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# output_3 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (150, 4), (4, 1))
assert_size_stride(primals_4, (150, ), (1, ))
assert_size_stride(primals_5, (150, 4), (4, 1))
assert_size_stride(primals_6, (150, ), (1, ))
assert_size_stride(primals_7, (1, 150), (150, 1))
assert_size_stride(primals_8, (1, 150), (150, 1))
assert_size_stride(primals_9, (1, 150, 150), (22500, 150, 1))
assert_size_stride(primals_10, (1, 1), (1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 150), (150, 1), torch.float32)
# Topologically Sorted Source Nodes: [input_s], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_4, primals_1, reinterpret_tensor(primals_3, (4, 150), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_3
del primals_4
buf1 = empty_strided_cuda((16, 150), (150, 1), torch.float32)
# Topologically Sorted Source Nodes: [input_t], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_6, reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 150), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_5
del primals_6
buf2 = empty_strided_cuda((1, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.mm]
extern_kernels.mm(primals_7, reinterpret_tensor(buf0, (150, 4), (1, 150), 0), out=buf2)
buf3 = empty_strided_cuda((1, 1, 16), (16, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_d], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(primals_8, (1, 1, 150), (150, 150, 1), 0), reinterpret_tensor(buf1, (1, 150, 16), (0, 1, 150), 0), out=buf3)
buf4 = empty_strided_cuda((1, 4, 150), (600, 150, 1), torch.float32)
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (1, 4, 150), (600, 150, 1), 0), primals_9, out=buf4)
buf5 = empty_strided_cuda((4, 1, 4), (4, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf4, (4, 1, 150), (150, 150, 1), 0), reinterpret_tensor(buf1, (4, 150, 4), (600, 1, 150), 0), out=buf5)
buf6 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [add, add_1, output_2, output_3], Original ATen: [aten.add, aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_add_0.run(buf5, buf3, buf2, primals_10, buf6, 4, grid=grid(4), stream=stream0)
buf7 = reinterpret_tensor(buf3, (4, 4), (4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [add, add_1, output_2, output_3], Original ATen: [aten.add, aten._softmax]
triton_poi_fused__softmax_add_1.run(buf7, buf5, buf2, primals_10, buf6, 16, grid=grid(16), stream=stream0)
del buf2
del buf6
del primals_10
buf8 = reinterpret_tensor(buf5, (4, 4), (4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [output_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf7, buf8, 16, grid=grid(16), stream=stream0)
del buf7
return (buf8, primals_1, primals_7, buf0, reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), buf1, buf8, reinterpret_tensor(buf4, (4, 150, 1), (150, 1, 150), 0), reinterpret_tensor(primals_9, (1, 150, 150), (22500, 1, 150), 0), reinterpret_tensor(primals_8, (1, 150, 1), (150, 1, 150), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((150, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((150, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((150, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((150, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, 150), (150, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((1, 150), (150, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((1, 150, 150), (22500, 150, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((1, 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])
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.dataloader
from torch.nn import Parameter
import torch.nn as nn
import torch.nn
from torch.nn.parameter import Parameter
class BiaffineAttention(nn.Module):
"""
Adopted from NeuroNLP2:
https://github.com/XuezheMax/NeuroNLP2/blob/master/neuronlp2/nn/modules/attention.py
Bi-Affine attention layer.
"""
def __init__(self, input_size_encoder, input_size_decoder, hidden_size=
150, num_labels=1, biaffine=True, **kwargs):
"""
Args:
input_size_encoder: int
the dimension of the encoder input.
input_size_decoder: int
the dimension of the decoder input.
num_labels: int
the number of labels of the crf layer
biaffine: bool
if apply bi-affine parameter.
**kwargs:
"""
super(BiaffineAttention, self).__init__()
self.input_size_encoder = input_size_encoder
self.input_size_decoder = input_size_decoder
self.hidden_size = hidden_size
self.linear_encoder = torch.nn.Linear(self.input_size_encoder, self
.hidden_size)
self.linear_decoder = torch.nn.Linear(self.input_size_decoder, self
.hidden_size)
self.num_labels = num_labels
self.biaffine = biaffine
self.W_d = Parameter(torch.Tensor(self.num_labels, self.hidden_size))
self.W_e = Parameter(torch.Tensor(self.num_labels, self.hidden_size))
self.b = Parameter(torch.Tensor(1, self.num_labels))
if self.biaffine:
self.U = Parameter(torch.Tensor(self.num_labels, self.
hidden_size, self.hidden_size))
else:
self.register_parameter('U', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.xavier_normal_(self.W_d)
nn.init.xavier_normal_(self.W_e)
nn.init.constant_(self.b, 0.0)
if self.biaffine:
nn.init.xavier_normal_(self.U)
def forward(self, input_s, input_t, mask_d=None, mask_e=None):
"""
Args:
input_s: Tensor
the student input tensor with shape = [batch, input_size]
input_t: Tensor
the teacher input tensor with shape = [batch, num_teachers, input_size]
mask_d: None
mask_e: None
Returns: Tensor
the energy tensor with shape = [batch, length]
"""
assert input_s.size(0) == input_t.size(0
), 'batch sizes of encoder and decoder are requires to be equal.'
input_s.size()
_, _num_teachers, _ = input_t.size()
input_s = self.linear_encoder(input_s)
input_t = self.linear_decoder(input_t)
out_e = torch.matmul(self.W_e, input_s.transpose(1, 0)).transpose(1, 0)
out_d = torch.einsum('nd,bnd->bn', self.W_d, input_t)
if self.biaffine:
output = torch.einsum('bd,nde->bne', input_s, self.U)
output = torch.einsum('bne,bne->bn', output, input_t)
output = output + out_d + out_e + self.b
else:
output = out_d + out_d + self.b
if mask_d is not None and mask_e is not None:
output = output * mask_d.unsqueeze(1).unsqueeze(3
) * mask_e.unsqueeze(1).unsqueeze(2)
output = torch.nn.functional.softmax(output, 1)
return output
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_size_encoder': 4, 'input_size_decoder': 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.utils.data.dataloader
from torch.nn import Parameter
import torch.nn as nn
import torch.nn
from torch.nn.parameter 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__softmax_add_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x0, xmask)
tmp5 = tl.load(in_ptr3 + 0)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp8 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp20 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp21 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp7 = tmp4 + tmp6
tmp10 = tmp8 + tmp9
tmp11 = tmp10 + tmp3
tmp12 = tmp11 + tmp6
tmp13 = triton_helpers.maximum(tmp7, tmp12)
tmp16 = tmp14 + tmp15
tmp17 = tmp16 + tmp3
tmp18 = tmp17 + tmp6
tmp19 = triton_helpers.maximum(tmp13, tmp18)
tmp22 = tmp20 + tmp21
tmp23 = tmp22 + tmp3
tmp24 = tmp23 + tmp6
tmp25 = triton_helpers.maximum(tmp19, tmp24)
tl.store(out_ptr0 + x0, tmp25, xmask)
@triton.jit
def triton_poi_fused__softmax_add_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2,
in_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 + x2, xmask)
tmp1 = tl.load(in_out_ptr0 + x2, xmask)
tmp3 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + 0)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp8 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp7 = tmp4 + tmp6
tmp9 = tmp7 - tmp8
tmp10 = tl_math.exp(tmp9)
tl.store(in_out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (150, 4), (4, 1))
assert_size_stride(primals_4, (150,), (1,))
assert_size_stride(primals_5, (150, 4), (4, 1))
assert_size_stride(primals_6, (150,), (1,))
assert_size_stride(primals_7, (1, 150), (150, 1))
assert_size_stride(primals_8, (1, 150), (150, 1))
assert_size_stride(primals_9, (1, 150, 150), (22500, 150, 1))
assert_size_stride(primals_10, (1, 1), (1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 150), (150, 1), torch.float32)
extern_kernels.addmm(primals_4, primals_1, reinterpret_tensor(
primals_3, (4, 150), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_3
del primals_4
buf1 = empty_strided_cuda((16, 150), (150, 1), torch.float32)
extern_kernels.addmm(primals_6, reinterpret_tensor(primals_2, (16,
4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 150), (1, 4),
0), alpha=1, beta=1, out=buf1)
del primals_5
del primals_6
buf2 = empty_strided_cuda((1, 4), (4, 1), torch.float32)
extern_kernels.mm(primals_7, reinterpret_tensor(buf0, (150, 4), (1,
150), 0), out=buf2)
buf3 = empty_strided_cuda((1, 1, 16), (16, 16, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(primals_8, (1, 1, 150), (150,
150, 1), 0), reinterpret_tensor(buf1, (1, 150, 16), (0, 1, 150),
0), out=buf3)
buf4 = empty_strided_cuda((1, 4, 150), (600, 150, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (1, 4, 150), (600, 150,
1), 0), primals_9, out=buf4)
buf5 = empty_strided_cuda((4, 1, 4), (4, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf4, (4, 1, 150), (150, 150,
1), 0), reinterpret_tensor(buf1, (4, 150, 4), (600, 1, 150), 0),
out=buf5)
buf6 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_add_0[grid(4)](buf5, buf3, buf2,
primals_10, buf6, 4, XBLOCK=4, num_warps=1, num_stages=1)
buf7 = reinterpret_tensor(buf3, (4, 4), (4, 1), 0)
del buf3
triton_poi_fused__softmax_add_1[grid(16)](buf7, buf5, buf2,
primals_10, buf6, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf2
del buf6
del primals_10
buf8 = reinterpret_tensor(buf5, (4, 4), (4, 1), 0)
del buf5
triton_poi_fused__softmax_2[grid(16)](buf7, buf8, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf7
return buf8, primals_1, primals_7, buf0, reinterpret_tensor(primals_2,
(16, 4), (4, 1), 0), buf1, buf8, reinterpret_tensor(buf4, (4, 150,
1), (150, 1, 150), 0), reinterpret_tensor(primals_9, (1, 150, 150),
(22500, 1, 150), 0), reinterpret_tensor(primals_8, (1, 150, 1), (
150, 1, 150), 0)
class BiaffineAttentionNew(nn.Module):
"""
Adopted from NeuroNLP2:
https://github.com/XuezheMax/NeuroNLP2/blob/master/neuronlp2/nn/modules/attention.py
Bi-Affine attention layer.
"""
def __init__(self, input_size_encoder, input_size_decoder, hidden_size=
150, num_labels=1, biaffine=True, **kwargs):
"""
Args:
input_size_encoder: int
the dimension of the encoder input.
input_size_decoder: int
the dimension of the decoder input.
num_labels: int
the number of labels of the crf layer
biaffine: bool
if apply bi-affine parameter.
**kwargs:
"""
super(BiaffineAttentionNew, self).__init__()
self.input_size_encoder = input_size_encoder
self.input_size_decoder = input_size_decoder
self.hidden_size = hidden_size
self.linear_encoder = torch.nn.Linear(self.input_size_encoder, self
.hidden_size)
self.linear_decoder = torch.nn.Linear(self.input_size_decoder, self
.hidden_size)
self.num_labels = num_labels
self.biaffine = biaffine
self.W_d = Parameter(torch.Tensor(self.num_labels, self.hidden_size))
self.W_e = Parameter(torch.Tensor(self.num_labels, self.hidden_size))
self.b = Parameter(torch.Tensor(1, self.num_labels))
if self.biaffine:
self.U = Parameter(torch.Tensor(self.num_labels, self.
hidden_size, self.hidden_size))
else:
self.register_parameter('U', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.xavier_normal_(self.W_d)
nn.init.xavier_normal_(self.W_e)
nn.init.constant_(self.b, 0.0)
if self.biaffine:
nn.init.xavier_normal_(self.U)
def forward(self, input_0, input_1):
primals_7 = self.W_d
primals_8 = self.W_e
primals_10 = self.b
primals_9 = self.U
primals_3 = self.linear_encoder.weight
primals_4 = self.linear_encoder.bias
primals_5 = self.linear_decoder.weight
primals_6 = self.linear_decoder.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return output[0]
| Dadmatech/DadmaTools | BiaffineAttention | false | 7,991 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import torch
import torch.utils.data.dataloader
from torch.nn import Parameter
import torch.nn as nn
import torch.nn
from torch.nn.parameter import Parameter
class Model(nn.Module):
"""
Adopted from NeuroNLP2:
https://github.com/XuezheMax/NeuroNLP2/blob/master/neuronlp2/nn/modules/attention.py
Bi-Affine attention layer.
"""
def __init__(self, input_size_encoder, input_size_decoder, hidden_size=
150, num_labels=1, biaffine=True, **kwargs):
"""
Args:
input_size_encoder: int
the dimension of the encoder input.
input_size_decoder: int
the dimension of the decoder input.
num_labels: int
the number of labels of the crf layer
biaffine: bool
if apply bi-affine parameter.
**kwargs:
"""
super().__init__()
self.input_size_encoder = input_size_encoder
self.input_size_decoder = input_size_decoder
self.hidden_size = hidden_size
self.linear_encoder = torch.nn.Linear(self.input_size_encoder, self
.hidden_size)
self.linear_decoder = torch.nn.Linear(self.input_size_decoder, self
.hidden_size)
self.num_labels = num_labels
self.biaffine = biaffine
self.W_d = Parameter(torch.Tensor(self.num_labels, self.hidden_size))
self.W_e = Parameter(torch.Tensor(self.num_labels, self.hidden_size))
self.b = Parameter(torch.Tensor(1, self.num_labels))
if self.biaffine:
self.U = Parameter(torch.Tensor(self.num_labels, self.
hidden_size, self.hidden_size))
else:
self.register_parameter('U', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.xavier_normal_(self.W_d)
nn.init.xavier_normal_(self.W_e)
nn.init.constant_(self.b, 0.0)
if self.biaffine:
nn.init.xavier_normal_(self.U)
def forward(self, input_s, input_t, mask_d=None, mask_e=None):
"""
Args:
input_s: Tensor
the student input tensor with shape = [batch, input_size]
input_t: Tensor
the teacher input tensor with shape = [batch, num_teachers, input_size]
mask_d: None
mask_e: None
Returns: Tensor
the energy tensor with shape = [batch, length]
"""
assert input_s.size(0) == input_t.size(0
), 'batch sizes of encoder and decoder are requires to be equal.'
input_s.size()
_, _num_teachers, _ = input_t.size()
input_s = self.linear_encoder(input_s)
input_t = self.linear_decoder(input_t)
out_e = torch.matmul(self.W_e, input_s.transpose(1, 0)).transpose(1, 0)
out_d = torch.einsum('nd,bnd->bn', self.W_d, input_t)
if self.biaffine:
output = torch.einsum('bd,nde->bne', input_s, self.U)
output = torch.einsum('bne,bne->bn', output, input_t)
output = output + out_d + out_e + self.b
else:
output = out_d + out_d + self.b
if mask_d is not None and mask_e is not None:
output = output * mask_d.unsqueeze(1).unsqueeze(3
) * mask_e.unsqueeze(1).unsqueeze(2)
output = torch.nn.functional.softmax(output, 1)
return output
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ShakeResNet | # 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_1/inductor_cache/ar/car7yauiypuyt2aup7rqtsj4glkttpfh7y4zxutghlnunswymzjp.py
# Topologically Sorted Source Nodes: [h, h_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# h => convolution
# h_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 262144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 4096) % 16
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 3, 3, 3), (27, 9, 3, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (4, 16), (16, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [h], 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, 16, 64, 64), (65536, 4096, 64, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [h, h_1], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 262144, grid=grid(262144), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [h_2], Original ATen: [aten.avg_pool2d]
buf2 = torch.ops.aten.avg_pool2d.default(buf1, [8, 8], [8, 8], [0, 0], False, True, None)
buf3 = buf2
del buf2
buf4 = empty_strided_cuda((256, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf3, (256, 16), (16, 1), 0), reinterpret_tensor(primals_4, (16, 4), (1, 16), 0), alpha=1, beta=1, out=buf4)
del primals_5
return (buf4, primals_1, primals_3, buf1, reinterpret_tensor(buf3, (256, 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((16, 3, 3, 3), (27, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 64, 64), (12288, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
class ShakeShake(torch.autograd.Function):
@staticmethod
def forward(ctx, x1, x2, training=True):
if training:
alpha = torch.FloatTensor(x1.size(0)).uniform_()
alpha = alpha.view(alpha.size(0), 1, 1, 1).expand_as(x1)
else:
alpha = 0.5
return alpha * x1 + (1 - alpha) * x2
@staticmethod
def backward(ctx, grad_output):
beta = torch.FloatTensor(grad_output.size(0)).uniform_()
beta = beta.view(beta.size(0), 1, 1, 1).expand_as(grad_output)
beta = Variable(beta)
return beta * grad_output, (1 - beta) * grad_output, None
class Shortcut(nn.Module):
def __init__(self, in_ch, out_ch, stride):
super(Shortcut, self).__init__()
self.stride = stride
self.conv1 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.conv2 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.bn = nn.BatchNorm2d(out_ch)
def forward(self, x):
h = F.relu(x)
h1 = F.avg_pool2d(h, 1, self.stride)
h1 = self.conv1(h1)
h2 = F.avg_pool2d(F.pad(h, (-1, 1, -1, 1)), 1, self.stride)
h2 = self.conv2(h2)
h = torch.cat((h1, h2), 1)
return self.bn(h)
class ShakeBlock(nn.Module):
def __init__(self, in_ch, out_ch, stride=1):
super(ShakeBlock, self).__init__()
self.equal_io = in_ch == out_ch
self.shortcut = self.equal_io and None or Shortcut(in_ch, out_ch,
stride=stride)
self.branch1 = self._make_branch(in_ch, out_ch, stride)
self.branch2 = self._make_branch(in_ch, out_ch, stride)
def forward(self, x):
h1 = self.branch1(x)
h2 = self.branch2(x)
h = ShakeShake.apply(h1, h2, self.training)
h0 = x if self.equal_io else self.shortcut(x)
return h + h0
def _make_branch(self, in_ch, out_ch, stride=1):
return nn.Sequential(nn.ReLU(inplace=False), nn.Conv2d(in_ch,
out_ch, 3, padding=1, stride=stride, bias=False), nn.
BatchNorm2d(out_ch), nn.ReLU(inplace=False), nn.Conv2d(out_ch,
out_ch, 3, padding=1, stride=1, bias=False), nn.BatchNorm2d(out_ch)
)
class ShakeResNet(nn.Module):
def __init__(self, depth, w_base, label):
super(ShakeResNet, self).__init__()
n_units = (depth - 2) / 6
in_chs = [16, w_base, w_base * 2, w_base * 4]
self.in_chs = in_chs
self.c_in = nn.Conv2d(3, in_chs[0], 3, padding=1)
self.layer1 = self._make_layer(n_units, in_chs[0], in_chs[1])
self.layer2 = self._make_layer(n_units, in_chs[1], in_chs[2], 2)
self.layer3 = self._make_layer(n_units, in_chs[2], in_chs[3], 2)
self.fc = nn.Linear(in_chs[3], label)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def forward(self, x):
h = self.c_in(x)
h = self.layer1(h)
h = self.layer2(h)
h = self.layer3(h)
h = F.relu(h)
h = F.avg_pool2d(h, 8)
h = h.view(-1, self.in_chs[3])
h = self.fc(h)
return h
def _make_layer(self, n_units, in_ch, out_ch, stride=1):
layers = []
for i in range(int(n_units)):
layers.append(ShakeBlock(in_ch, out_ch, stride=stride))
in_ch, stride = out_ch, 1
return nn.Sequential(*layers)
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return [[], {'depth': 1, 'w_base': 4, 'label': 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 math
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
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):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 4096 % 16
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, None)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 3, 3, 3), (27, 9, 3, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (4, 3, 64, 64), (12288, 4096, 64, 1))
assert_size_stride(primals_4, (4, 16), (16, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = 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, 16, 64, 64), (65536, 4096, 64, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(262144)](buf1, primals_2,
262144, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_2
buf2 = torch.ops.aten.avg_pool2d.default(buf1, [8, 8], [8, 8], [0,
0], False, True, None)
buf3 = buf2
del buf2
buf4 = empty_strided_cuda((256, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf3, (256, 16),
(16, 1), 0), reinterpret_tensor(primals_4, (16, 4), (1, 16), 0),
alpha=1, beta=1, out=buf4)
del primals_5
return buf4, primals_1, primals_3, buf1, reinterpret_tensor(buf3, (256,
16), (16, 1), 0), primals_4
class ShakeShake(torch.autograd.Function):
@staticmethod
def forward(ctx, x1, x2, training=True):
if training:
alpha = torch.FloatTensor(x1.size(0)).uniform_()
alpha = alpha.view(alpha.size(0), 1, 1, 1).expand_as(x1)
else:
alpha = 0.5
return alpha * x1 + (1 - alpha) * x2
@staticmethod
def backward(ctx, grad_output):
beta = torch.FloatTensor(grad_output.size(0)).uniform_()
beta = beta.view(beta.size(0), 1, 1, 1).expand_as(grad_output)
beta = Variable(beta)
return beta * grad_output, (1 - beta) * grad_output, None
class Shortcut(nn.Module):
def __init__(self, in_ch, out_ch, stride):
super(Shortcut, self).__init__()
self.stride = stride
self.conv1 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.conv2 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.bn = nn.BatchNorm2d(out_ch)
def forward(self, x):
h = F.relu(x)
h1 = F.avg_pool2d(h, 1, self.stride)
h1 = self.conv1(h1)
h2 = F.avg_pool2d(F.pad(h, (-1, 1, -1, 1)), 1, self.stride)
h2 = self.conv2(h2)
h = torch.cat((h1, h2), 1)
return self.bn(h)
class ShakeBlock(nn.Module):
def __init__(self, in_ch, out_ch, stride=1):
super(ShakeBlock, self).__init__()
self.equal_io = in_ch == out_ch
self.shortcut = self.equal_io and None or Shortcut(in_ch, out_ch,
stride=stride)
self.branch1 = self._make_branch(in_ch, out_ch, stride)
self.branch2 = self._make_branch(in_ch, out_ch, stride)
def forward(self, x):
h1 = self.branch1(x)
h2 = self.branch2(x)
h = ShakeShake.apply(h1, h2, self.training)
h0 = x if self.equal_io else self.shortcut(x)
return h + h0
def _make_branch(self, in_ch, out_ch, stride=1):
return nn.Sequential(nn.ReLU(inplace=False), nn.Conv2d(in_ch,
out_ch, 3, padding=1, stride=stride, bias=False), nn.
BatchNorm2d(out_ch), nn.ReLU(inplace=False), nn.Conv2d(out_ch,
out_ch, 3, padding=1, stride=1, bias=False), nn.BatchNorm2d(out_ch)
)
class ShakeResNetNew(nn.Module):
def __init__(self, depth, w_base, label):
super(ShakeResNetNew, self).__init__()
n_units = (depth - 2) / 6
in_chs = [16, w_base, w_base * 2, w_base * 4]
self.in_chs = in_chs
self.c_in = nn.Conv2d(3, in_chs[0], 3, padding=1)
self.layer1 = self._make_layer(n_units, in_chs[0], in_chs[1])
self.layer2 = self._make_layer(n_units, in_chs[1], in_chs[2], 2)
self.layer3 = self._make_layer(n_units, in_chs[2], in_chs[3], 2)
self.fc = nn.Linear(in_chs[3], label)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def _make_layer(self, n_units, in_ch, out_ch, stride=1):
layers = []
for i in range(int(n_units)):
layers.append(ShakeBlock(in_ch, out_ch, stride=stride))
in_ch, stride = out_ch, 1
return nn.Sequential(*layers)
def forward(self, input_0):
primals_1 = self.c_in.weight
primals_2 = self.c_in.bias
primals_4 = self.fc.weight
primals_5 = self.fc.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DensoITLab/TeachAugment | ShakeResNet | false | 7,992 | [
"BSD-2-Clause"
] | 20 | 66ec099a0afab99e18531c5437182cfe17dc30c8 | https://github.com/DensoITLab/TeachAugment/tree/66ec099a0afab99e18531c5437182cfe17dc30c8 | import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
class ShakeShake(torch.autograd.Function):
@staticmethod
def forward(ctx, x1, x2, training=True):
if training:
alpha = torch.FloatTensor(x1.size(0)).uniform_()
alpha = alpha.view(alpha.size(0), 1, 1, 1).expand_as(x1)
else:
alpha = 0.5
return alpha * x1 + (1 - alpha) * x2
@staticmethod
def backward(ctx, grad_output):
beta = torch.FloatTensor(grad_output.size(0)).uniform_()
beta = beta.view(beta.size(0), 1, 1, 1).expand_as(grad_output)
beta = Variable(beta)
return beta * grad_output, (1 - beta) * grad_output, None
class Shortcut(nn.Module):
def __init__(self, in_ch, out_ch, stride):
super().__init__()
self.stride = stride
self.conv1 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.conv2 = nn.Conv2d(in_ch, out_ch // 2, 1, stride=1, padding=0,
bias=False)
self.bn = nn.BatchNorm2d(out_ch)
def forward(self, x):
h = F.relu(x)
h1 = F.avg_pool2d(h, 1, self.stride)
h1 = self.conv1(h1)
h2 = F.avg_pool2d(F.pad(h, (-1, 1, -1, 1)), 1, self.stride)
h2 = self.conv2(h2)
h = torch.cat((h1, h2), 1)
return self.bn(h)
class ShakeBlock(nn.Module):
def __init__(self, in_ch, out_ch, stride=1):
super().__init__()
self.equal_io = in_ch == out_ch
self.shortcut = self.equal_io and None or Shortcut(in_ch, out_ch,
stride=stride)
self.branch1 = self._make_branch(in_ch, out_ch, stride)
self.branch2 = self._make_branch(in_ch, out_ch, stride)
def forward(self, x):
h1 = self.branch1(x)
h2 = self.branch2(x)
h = ShakeShake.apply(h1, h2, self.training)
h0 = x if self.equal_io else self.shortcut(x)
return h + h0
def _make_branch(self, in_ch, out_ch, stride=1):
return nn.Sequential(nn.ReLU(inplace=False), nn.Conv2d(in_ch,
out_ch, 3, padding=1, stride=stride, bias=False), nn.
BatchNorm2d(out_ch), nn.ReLU(inplace=False), nn.Conv2d(out_ch,
out_ch, 3, padding=1, stride=1, bias=False), nn.BatchNorm2d(out_ch)
)
class Model(nn.Module):
def __init__(self, depth, w_base, label):
super().__init__()
n_units = (depth - 2) / 6
in_chs = [16, w_base, w_base * 2, w_base * 4]
self.in_chs = in_chs
self.c_in = nn.Conv2d(3, in_chs[0], 3, padding=1)
self.layer1 = self._make_layer(n_units, in_chs[0], in_chs[1])
self.layer2 = self._make_layer(n_units, in_chs[1], in_chs[2], 2)
self.layer3 = self._make_layer(n_units, in_chs[2], in_chs[3], 2)
self.fc = nn.Linear(in_chs[3], label)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
def forward(self, x):
h = self.c_in(x)
h = self.layer1(h)
h = self.layer2(h)
h = self.layer3(h)
h = F.relu(h)
h = F.avg_pool2d(h, 8)
h = h.view(-1, self.in_chs[3])
h = self.fc(h)
return h
def _make_layer(self, n_units, in_ch, out_ch, stride=1):
layers = []
for i in range(int(n_units)):
layers.append(ShakeBlock(in_ch, out_ch, stride=stride))
in_ch, stride = out_ch, 1
return nn.Sequential(*layers)
def get_inputs():
return [torch.rand([4, 3, 64, 64])]
def get_init_inputs():
return [1, 4, 4]
|
PT | # 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_1/inductor_cache/d2/cd2onkfnnqqav3af2iyomd6l7n3rqdwwvxsyw3ac7zfntydugnbi.py
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add => add
# x => relu
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_3), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_add_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_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_1/inductor_cache/nd/cndmfvh7lrojuqo776y2t4r3iwqonobmuxwcqkx2pam6myjioeyl.py
# Topologically Sorted Source Nodes: [add_1], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add_1 => add_1
# Graph fragment:
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_5), kwargs = {})
triton_poi_fused_add_1 = async_compile.triton('triton_poi_fused_add_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), primals_2, out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0.run(buf1, primals_3, buf4, 256, grid=grid(256), stream=stream0)
del primals_3
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_4, out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [add_1], Original ATen: [aten.add]
triton_poi_fused_add_1.run(buf3, primals_5, 256, grid=grid(256), stream=stream0)
del primals_5
return (buf3, reinterpret_tensor(buf1, (4, 64), (1, 4), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), buf4, reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((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 numpy as np
import torch.nn as nn
import torch.nn.functional as F
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super(Linear, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class PT(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, epsilon, mode, K, alpha):
super(PT, self).__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
self.epsilon = epsilon
self.mode = mode
self.K = K
self.alpha = alpha
self.number_class = nclass
def forward(self, x):
x = torch.relu(self.Linear1(x))
return self.Linear2(x)
def loss_function(self, y_hat, y_soft, epoch=0):
if self.training:
y_hat_con = torch.detach(torch.softmax(y_hat, dim=-1))
exp = np.log(epoch / self.epsilon + 1)
if self.mode == 2:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), torch.mul(y_soft, y_hat_con ** exp))
) / self.number_class
elif self.mode == 1:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), torch.mul(y_soft, y_hat_con))) / self.number_class
else:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), y_soft)) / self.number_class
else:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1),
y_soft)) / self.number_class
return loss
def inference(self, h, adj):
y0 = torch.softmax(h, dim=-1)
y = y0
for _ in range(self.K):
y = (1 - self.alpha) * torch.matmul(adj, y) + self.alpha * y0
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nfeat': 4, 'nhid': 4, 'nclass': 4, 'dropout': 0.5,
'epsilon': 4, 'mode': 4, 'K': 4, 'alpha': 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 math
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_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_add_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0),
primals_2, out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0[grid(256)](buf1,
primals_3, buf4, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_3
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
primals_4, out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
triton_poi_fused_add_1[grid(256)](buf3, primals_5, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_5
return buf3, reinterpret_tensor(buf1, (4, 64), (1, 4), 0
), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0
), buf4, reinterpret_tensor(primals_1, (4, 64), (1, 4), 0)
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super(Linear, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class PTNew(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, epsilon, mode, K, alpha):
super(PTNew, self).__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
self.epsilon = epsilon
self.mode = mode
self.K = K
self.alpha = alpha
self.number_class = nclass
def loss_function(self, y_hat, y_soft, epoch=0):
if self.training:
y_hat_con = torch.detach(torch.softmax(y_hat, dim=-1))
exp = np.log(epoch / self.epsilon + 1)
if self.mode == 2:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), torch.mul(y_soft, y_hat_con ** exp))
) / self.number_class
elif self.mode == 1:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), torch.mul(y_soft, y_hat_con))) / self.number_class
else:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), y_soft)) / self.number_class
else:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1),
y_soft)) / self.number_class
return loss
def inference(self, h, adj):
y0 = torch.softmax(h, dim=-1)
y = y0
for _ in range(self.K):
y = (1 - self.alpha) * torch.matmul(adj, y) + self.alpha * y0
return y
def forward(self, input_0):
primals_2 = self.Linear1.weight
primals_3 = self.Linear1.bias
primals_4 = self.Linear2.weight
primals_5 = self.Linear2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DongHande/PT_propagation_then_training | PT | false | 7,993 | [
"MIT"
] | 21 | 3f346ff161d2a0b807e3c0269ad26a7266305cc3 | https://github.com/DongHande/PT_propagation_then_training/tree/3f346ff161d2a0b807e3c0269ad26a7266305cc3 | import math
import torch
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super().__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class Model(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, epsilon, mode, K, alpha):
super().__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
self.epsilon = epsilon
self.mode = mode
self.K = K
self.alpha = alpha
self.number_class = nclass
def forward(self, x):
x = torch.relu(self.Linear1(x))
return self.Linear2(x)
def loss_function(self, y_hat, y_soft, epoch=0):
if self.training:
y_hat_con = torch.detach(torch.softmax(y_hat, dim=-1))
exp = np.log(epoch / self.epsilon + 1)
if self.mode == 2:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), torch.mul(y_soft, y_hat_con ** exp))
) / self.number_class
elif self.mode == 1:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), torch.mul(y_soft, y_hat_con))) / self.number_class
else:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1
), y_soft)) / self.number_class
else:
loss = -torch.sum(torch.mul(torch.log_softmax(y_hat, dim=-1),
y_soft)) / self.number_class
return loss
def inference(self, h, adj):
y0 = torch.softmax(h, dim=-1)
y = y0
for _ in range(self.K):
y = (1 - self.alpha) * torch.matmul(adj, y) + self.alpha * y0
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nfeat': 4, 'nhid': 4, 'nclass': 4, 'dropout': 0.5,
'epsilon': 4, 'mode': 4, 'K': 4, 'alpha': 4}]
|
PoolReducer | # 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_1/inductor_cache/v5/cv5s5nofrt3xilwwaczjjgmovzeqkvmyyldkduxjqjzazpbu4gsa.py
# Topologically Sorted Source Nodes: [max_1], Original ATen: [aten.max]
# Source node to ATen node mapping:
# max_1 => getitem
# Graph fragment:
# %getitem : [num_users=1] = call_function[target=operator.getitem](args = (%max_1, 0), kwargs = {})
triton_poi_fused_max_0 = async_compile.triton('triton_poi_fused_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_max_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + (16*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + (16*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + (16*x1)), xmask)
tmp2 = 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: [max_1], Original ATen: [aten.max]
stream0 = get_raw_stream(0)
triton_poi_fused_max_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 torch import nn
class PoolReducer(nn.Module):
def __init__(self, groups, pool_operation='max'):
super(PoolReducer, self).__init__()
if pool_operation == 'max':
self.pool = lambda x: x.max(2)[0]
elif pool_operation == 'average':
self.pool = lambda x: x.mean(2)
else:
raise ValueError('pool_operation must be in ["max","average"]')
def forward(self, x):
return self.pool(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'groups': 1}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_max_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + 16 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + 16 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + 16 * x1), xmask)
tmp2 = 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_max_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class PoolReducerNew(nn.Module):
def __init__(self, groups, pool_operation='max'):
super(PoolReducerNew, self).__init__()
if pool_operation == 'max':
self.pool = lambda x: x.max(2)[0]
elif pool_operation == 'average':
self.pool = lambda x: x.mean(2)
else:
raise ValueError('pool_operation must be in ["max","average"]')
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Dreem-Organization/RobustSleepNet | PoolReducer | false | 7,994 | [
"MIT"
] | 16 | c8ff3f6f857299eb2bf2e9400483084d5ecd4106 | https://github.com/Dreem-Organization/RobustSleepNet/tree/c8ff3f6f857299eb2bf2e9400483084d5ecd4106 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, groups, pool_operation='max'):
super().__init__()
if pool_operation == 'max':
self.pool = lambda x: x.max(2)[0]
elif pool_operation == 'average':
self.pool = lambda x: x.mean(2)
else:
raise ValueError('pool_operation must be in ["max","average"]')
def forward(self, x):
return self.pool(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [1]
|
HyperSphereLoss | # 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_1/inductor_cache/wm/cwmtdji7wew23pxblgrnor37cmtzgq5ehdup65sglxkhagz5loak.py
# Topologically Sorted Source Nodes: [q], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# q => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%div, %div_1], 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=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 320
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 16) % 5
x0 = xindex % 16
x2 = (xindex // 80)
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 = 2.0
tmp7 = tmp5 * tmp6
tmp8 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp9 = tmp8 * tmp8
tmp10 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp10 * tmp10
tmp12 = tmp9 + tmp11
tmp13 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp14 = tmp13 * tmp13
tmp15 = tmp12 + tmp14
tmp16 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp17 = tmp16 * tmp16
tmp18 = tmp15 + tmp17
tmp19 = libdevice.sqrt(tmp18)
tmp20 = tmp19 * tmp19
tmp21 = 1.0
tmp22 = tmp20 + tmp21
tmp23 = tmp7 / tmp22
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp4, tmp23, tmp24)
tmp26 = tmp0 >= tmp3
tmp27 = tl.full([1], 5, tl.int64)
tmp28 = tmp0 < tmp27
tmp29 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp26 & xmask, eviction_policy='evict_last', other=0.0)
tmp30 = tmp29 * tmp29
tmp31 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp26 & xmask, eviction_policy='evict_last', other=0.0)
tmp32 = tmp31 * tmp31
tmp33 = tmp30 + tmp32
tmp34 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp26 & xmask, eviction_policy='evict_last', other=0.0)
tmp35 = tmp34 * tmp34
tmp36 = tmp33 + tmp35
tmp37 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp26 & xmask, eviction_policy='evict_last', other=0.0)
tmp38 = tmp37 * tmp37
tmp39 = tmp36 + tmp38
tmp40 = libdevice.sqrt(tmp39)
tmp41 = tmp40 * tmp40
tmp42 = tmp41 - tmp21
tmp43 = tmp41 + tmp21
tmp44 = tmp42 / tmp43
tmp45 = tl.full(tmp44.shape, 0.0, tmp44.dtype)
tmp46 = tl.where(tmp26, tmp44, tmp45)
tmp47 = tl.where(tmp4, tmp25, tmp46)
tl.store(out_ptr0 + (x3), tmp47, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4s/c4sf274gpf37u37symxy5dtiezp6lkdsditc3vzvonvuszsmn3yr.py
# Topologically Sorted Source Nodes: [mul_1, norm_1, q_norm, add_2, loss, acos, mean], Original ATen: [aten.mul, aten.linalg_vector_norm, aten.pow, aten.add, aten.div, aten.acos, aten.mean]
# Source node to ATen node mapping:
# acos => acos
# add_2 => add_2
# loss => div_2
# mean => mean
# mul_1 => mul_1
# norm_1 => pow_4, pow_5, sum_2
# q_norm => pow_6
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select, 2), kwargs = {})
# %pow_4 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%cat, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_4, [1]), kwargs = {})
# %pow_5 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 0.5), kwargs = {})
# %pow_6 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%pow_5, 2), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_6, 1), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul_1, %add_2), kwargs = {})
# %acos : [num_users=1] = call_function[target=torch.ops.aten.acos.default](args = (%div_2,), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%acos,), kwargs = {})
triton_per_fused_acos_add_div_linalg_vector_norm_mean_mul_pow_1 = async_compile.triton('triton_per_fused_acos_add_div_linalg_vector_norm_mean_mul_pow_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, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_acos_add_div_linalg_vector_norm_mean_mul_pow_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_acos_add_div_linalg_vector_norm_mean_mul_pow_1(in_out_ptr0, in_ptr0, 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 % 16
r1 = (rindex // 16)
tmp0 = tl.load(in_ptr0 + (64 + r0 + (80*r1)), None)
tmp3 = tl.load(in_ptr0 + (r0 + (80*r1)), None)
tmp5 = tl.load(in_ptr0 + (16 + r0 + (80*r1)), None)
tmp8 = tl.load(in_ptr0 + (32 + r0 + (80*r1)), None)
tmp11 = tl.load(in_ptr0 + (48 + r0 + (80*r1)), None)
tmp1 = 2.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp3
tmp6 = tmp5 * tmp5
tmp7 = tmp4 + tmp6
tmp9 = tmp8 * tmp8
tmp10 = tmp7 + tmp9
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp0 * tmp0
tmp15 = tmp13 + tmp14
tmp16 = libdevice.sqrt(tmp15)
tmp17 = tmp16 * tmp16
tmp18 = 1.0
tmp19 = tmp17 + tmp18
tmp20 = tmp2 / tmp19
tmp21 = libdevice.acos(tmp20)
tmp22 = tl.broadcast_to(tmp21, [XBLOCK, RBLOCK])
tmp24 = tl.sum(tmp22, 1)[:, None]
tmp25 = 64.0
tmp26 = tmp24 / tmp25
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp26, 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, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 5, 4, 4), (80, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [q], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(arg0_1, buf0, 320, grid=grid(320), stream=stream0)
del arg0_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [mul_1, norm_1, q_norm, add_2, loss, acos, mean], Original ATen: [aten.mul, aten.linalg_vector_norm, aten.pow, aten.add, aten.div, aten.acos, aten.mean]
triton_per_fused_acos_add_div_linalg_vector_norm_mean_mul_pow_1.run(buf2, buf0, 1, 64, 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, 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 HyperSphereLoss(nn.Module):
def forward(self, input):
"""
Calcuate distance between input and N(North Pole) using hypersphere metrics.
Woo Park, Sung, and Junseok Kwon.
"Sphere Generative Adversarial Network Based on Geometric Moment Matching."
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019.
"""
q = self.project_to_hypersphere(input)
q_norm = torch.norm(q, dim=1) ** 2
loss = 2 * q[:, -1] / (1 + q_norm)
return torch.mean(torch.acos(loss))
def project_to_hypersphere(self, v):
v_norm = torch.norm(v, dim=1, keepdim=True) ** 2
a = 2 * v / (v_norm + 1)
b = (v_norm - 1) / (v_norm + 1)
return torch.cat([a, b], 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
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 320
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 16 % 5
x0 = xindex % 16
x2 = xindex // 80
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 = 2.0
tmp7 = tmp5 * tmp6
tmp8 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp9 = tmp8 * tmp8
tmp10 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp11 = tmp10 * tmp10
tmp12 = tmp9 + tmp11
tmp13 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp14 = tmp13 * tmp13
tmp15 = tmp12 + tmp14
tmp16 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp17 = tmp16 * tmp16
tmp18 = tmp15 + tmp17
tmp19 = libdevice.sqrt(tmp18)
tmp20 = tmp19 * tmp19
tmp21 = 1.0
tmp22 = tmp20 + tmp21
tmp23 = tmp7 / tmp22
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp4, tmp23, tmp24)
tmp26 = tmp0 >= tmp3
tl.full([1], 5, tl.int64)
tmp29 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp26 & xmask,
eviction_policy='evict_last', other=0.0)
tmp30 = tmp29 * tmp29
tmp31 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp26 & xmask,
eviction_policy='evict_last', other=0.0)
tmp32 = tmp31 * tmp31
tmp33 = tmp30 + tmp32
tmp34 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp26 & xmask,
eviction_policy='evict_last', other=0.0)
tmp35 = tmp34 * tmp34
tmp36 = tmp33 + tmp35
tmp37 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp26 & xmask,
eviction_policy='evict_last', other=0.0)
tmp38 = tmp37 * tmp37
tmp39 = tmp36 + tmp38
tmp40 = libdevice.sqrt(tmp39)
tmp41 = tmp40 * tmp40
tmp42 = tmp41 - tmp21
tmp43 = tmp41 + tmp21
tmp44 = tmp42 / tmp43
tmp45 = tl.full(tmp44.shape, 0.0, tmp44.dtype)
tmp46 = tl.where(tmp26, tmp44, tmp45)
tmp47 = tl.where(tmp4, tmp25, tmp46)
tl.store(out_ptr0 + x3, tmp47, xmask)
@triton.jit
def triton_per_fused_acos_add_div_linalg_vector_norm_mean_mul_pow_1(in_out_ptr0
, in_ptr0, 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 % 16
r1 = rindex // 16
tmp0 = tl.load(in_ptr0 + (64 + r0 + 80 * r1), None)
tmp3 = tl.load(in_ptr0 + (r0 + 80 * r1), None)
tmp5 = tl.load(in_ptr0 + (16 + r0 + 80 * r1), None)
tmp8 = tl.load(in_ptr0 + (32 + r0 + 80 * r1), None)
tmp11 = tl.load(in_ptr0 + (48 + r0 + 80 * r1), None)
tmp1 = 2.0
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp3
tmp6 = tmp5 * tmp5
tmp7 = tmp4 + tmp6
tmp9 = tmp8 * tmp8
tmp10 = tmp7 + tmp9
tmp12 = tmp11 * tmp11
tmp13 = tmp10 + tmp12
tmp14 = tmp0 * tmp0
tmp15 = tmp13 + tmp14
tmp16 = libdevice.sqrt(tmp15)
tmp17 = tmp16 * tmp16
tmp18 = 1.0
tmp19 = tmp17 + tmp18
tmp20 = tmp2 / tmp19
tmp21 = libdevice.acos(tmp20)
tmp22 = tl.broadcast_to(tmp21, [XBLOCK, RBLOCK])
tmp24 = tl.sum(tmp22, 1)[:, None]
tmp25 = 64.0
tmp26 = tmp24 / tmp25
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp26, None)
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, 5, 4, 4), (80, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(320)](arg0_1, buf0, 320, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
triton_per_fused_acos_add_div_linalg_vector_norm_mean_mul_pow_1[grid(1)
](buf2, buf0, 1, 64, XBLOCK=1, num_warps=2, num_stages=1)
del buf0
return buf2,
class HyperSphereLossNew(nn.Module):
def project_to_hypersphere(self, v):
v_norm = torch.norm(v, dim=1, keepdim=True) ** 2
a = 2 * v / (v_norm + 1)
b = (v_norm - 1) / (v_norm + 1)
return torch.cat([a, b], dim=1)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Dotori-HJ/SphereGAN-Pytorch-implementation | HyperSphereLoss | false | 7,995 | [
"MIT"
] | 11 | fe7843545388ecdae34f374e7f1c42300ab12689 | https://github.com/Dotori-HJ/SphereGAN-Pytorch-implementation/tree/fe7843545388ecdae34f374e7f1c42300ab12689 | import torch
import torch.nn as nn
class Model(nn.Module):
def forward(self, input):
"""
Calcuate distance between input and N(North Pole) using hypersphere metrics.
Woo Park, Sung, and Junseok Kwon.
"Sphere Generative Adversarial Network Based on Geometric Moment Matching."
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019.
"""
q = self.project_to_hypersphere(input)
q_norm = torch.norm(q, dim=1) ** 2
loss = 2 * q[:, -1] / (1 + q_norm)
return torch.mean(torch.acos(loss))
def project_to_hypersphere(self, v):
v_norm = torch.norm(v, dim=1, keepdim=True) ** 2
a = 2 * v / (v_norm + 1)
b = (v_norm - 1) / (v_norm + 1)
return torch.cat([a, b], dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GCN | # 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_1/inductor_cache/d2/cd2onkfnnqqav3af2iyomd6l7n3rqdwwvxsyw3ac7zfntydugnbi.py
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add => add
# x => relu
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_4, %primals_4), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_add_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_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_1/inductor_cache/6e/c6ebscol22x55zycphnwr3uy4ba7rbar46xom3qgmqtrqan24cdq.py
# Topologically Sorted Source Nodes: [h, log_softmax], Original ATen: [aten.add, aten._log_softmax]
# Source node to ATen node mapping:
# h => add_1
# log_softmax => amax, exp, sub, sum_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_9, %primals_6), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_1, [-1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_1, %amax), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
triton_poi_fused__log_softmax_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_add_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_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 + (1))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp9 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (2))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp14 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (3))
tmp16 = tl.broadcast_to(tmp15, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp7 = tmp4 + tmp6
tmp8 = triton_helpers.maximum(tmp3, tmp7)
tmp12 = tmp9 + tmp11
tmp13 = triton_helpers.maximum(tmp8, tmp12)
tmp17 = tmp14 + tmp16
tmp18 = triton_helpers.maximum(tmp13, tmp17)
tmp19 = tmp3 - tmp18
tmp20 = tl_math.exp(tmp19)
tmp21 = tmp7 - tmp18
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp20 + tmp22
tmp24 = tmp12 - tmp18
tmp25 = tl_math.exp(tmp24)
tmp26 = tmp23 + tmp25
tmp27 = tmp17 - tmp18
tmp28 = tl_math.exp(tmp27)
tmp29 = tmp26 + tmp28
tl.store(out_ptr0 + (x0), tmp18, xmask)
tl.store(out_ptr1 + (x0), tmp29, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/nv/cnv34oqgp2piwfu4kb4d3grqyqas6ixhkszs2ofup3p3l3i7ccvo.py
# Topologically Sorted Source Nodes: [h, log_softmax], Original ATen: [aten.add, aten._log_softmax]
# Source node to ATen node mapping:
# h => add_1
# log_softmax => amax, log, sub, sub_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_9, %primals_6), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_1, [-1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_1, %amax), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub, %log), kwargs = {})
triton_poi_fused__log_softmax_add_2 = async_compile.triton('triton_poi_fused__log_softmax_add_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_add_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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
x2 = xindex
x0 = xindex % 4
x1 = (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 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = tl_math.log(tmp5)
tmp7 = tmp4 - tmp6
tl.store(in_out_ptr0 + (x2), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (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(primals_2, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(primals_1, (16, 4, 4), (16, 4, 1), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (64, 4), (4, 1), 0), primals_3, out=buf1)
del primals_3
buf2 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf1 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0.run(buf2, primals_4, buf8, 256, grid=grid(256), stream=stream0)
del primals_4
buf3 = 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(primals_2, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0), out=buf3)
buf4 = reinterpret_tensor(buf2, (64, 4), (4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), primals_5, out=buf4)
buf5 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf6 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [h, log_softmax], Original ATen: [aten.add, aten._log_softmax]
triton_poi_fused__log_softmax_add_1.run(buf4, primals_6, buf5, buf6, 64, grid=grid(64), stream=stream0)
buf7 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [h, log_softmax], Original ATen: [aten.add, aten._log_softmax]
triton_poi_fused__log_softmax_add_2.run(buf7, primals_6, buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
del buf6
del primals_6
return (buf7, buf7, reinterpret_tensor(buf3, (4, 64), (1, 4), 0), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), reinterpret_tensor(primals_2, (16, 4, 4), (16, 1, 4), 0), buf8, reinterpret_tensor(buf0, (4, 64), (1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
import torch.nn.functional as F
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super(Linear, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class GCN(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super(GCN, self).__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
def forward(self, x, adj):
x = torch.relu(self.Linear1(torch.matmul(adj, x)))
h = self.Linear2(torch.matmul(adj, x))
return torch.log_softmax(h, dim=-1)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nfeat': 4, 'nhid': 4, 'nclass': 4, 'dropout': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import math
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_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_add_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
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 + 1)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK])
tmp9 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + 2)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp14 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr1 + 3)
tmp16 = tl.broadcast_to(tmp15, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp7 = tmp4 + tmp6
tmp8 = triton_helpers.maximum(tmp3, tmp7)
tmp12 = tmp9 + tmp11
tmp13 = triton_helpers.maximum(tmp8, tmp12)
tmp17 = tmp14 + tmp16
tmp18 = triton_helpers.maximum(tmp13, tmp17)
tmp19 = tmp3 - tmp18
tmp20 = tl_math.exp(tmp19)
tmp21 = tmp7 - tmp18
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp20 + tmp22
tmp24 = tmp12 - tmp18
tmp25 = tl_math.exp(tmp24)
tmp26 = tmp23 + tmp25
tmp27 = tmp17 - tmp18
tmp28 = tl_math.exp(tmp27)
tmp29 = tmp26 + tmp28
tl.store(out_ptr0 + x0, tmp18, xmask)
tl.store(out_ptr1 + x0, tmp29, xmask)
@triton.jit
def triton_poi_fused__log_softmax_add_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
x2 = xindex
x0 = xindex % 4
x1 = 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 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = tl_math.log(tmp5)
tmp7 = tmp4 - tmp6
tl.store(in_out_ptr0 + x2, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (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(primals_2, (16, 4, 4), (16, 4,
1), 0), reinterpret_tensor(primals_1, (16, 4, 4), (16, 4, 1), 0
), out=buf0)
del primals_1
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (64, 4), (4, 1), 0),
primals_3, out=buf1)
del primals_3
buf2 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf1
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0[grid(256)](buf2,
primals_4, buf8, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(primals_2, (16, 4, 4), (16, 4,
1), 0), reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0),
out=buf3)
buf4 = reinterpret_tensor(buf2, (64, 4), (4, 1), 0)
del buf2
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
primals_5, out=buf4)
buf5 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf6 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused__log_softmax_add_1[grid(64)](buf4, primals_6, buf5,
buf6, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf7 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf4
triton_poi_fused__log_softmax_add_2[grid(256)](buf7, primals_6,
buf5, buf6, 256, XBLOCK=256, num_warps=4, num_stages=1)
del buf5
del buf6
del primals_6
return buf7, buf7, reinterpret_tensor(buf3, (4, 64), (1, 4), 0
), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0
), reinterpret_tensor(primals_2, (16, 4, 4), (16, 1, 4), 0
), buf8, reinterpret_tensor(buf0, (4, 64), (1, 4), 0)
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super(Linear, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class GCNNew(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super(GCNNew, self).__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
def forward(self, input_0, input_1):
primals_3 = self.Linear1.weight
primals_4 = self.Linear1.bias
primals_5 = self.Linear2.weight
primals_6 = self.Linear2.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| DongHande/PT_propagation_then_training | GCN | false | 7,996 | [
"MIT"
] | 21 | 3f346ff161d2a0b807e3c0269ad26a7266305cc3 | https://github.com/DongHande/PT_propagation_then_training/tree/3f346ff161d2a0b807e3c0269ad26a7266305cc3 | import math
import torch
import torch.nn as nn
import torch.nn.functional as F
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super().__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class Model(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super().__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
def forward(self, x, adj):
x = torch.relu(self.Linear1(torch.matmul(adj, x)))
h = self.Linear2(torch.matmul(adj, x))
return torch.log_softmax(h, dim=-1)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4, 0.5]
|
TransformerEncoderLayer | # 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_1/inductor_cache/25/c25xezvaufcxosn2zaablzq3d4dz4ocigm5s7w52r36lqwucoptj.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x => add, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_1, [1]), 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=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_native_layer_norm_0 = async_compile.triton('triton_poi_fused_native_layer_norm_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_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_1/inductor_cache/uv/cuvyzdrclvsiubqwnkqyzqogybjelcuaoywggvz52bz5abo4hmua.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_1, [1]), 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=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_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_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
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_1/inductor_cache/5o/c5oauw3z3ep2ftanta4riae32t6z5u67plt43eekrznregbmlpjr.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# multi_head_attention_forward => mul_2
# Graph fragment:
# %mul_2 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_3, 1.0), kwargs = {})
triton_poi_fused_mul_2 = async_compile.triton('triton_poi_fused_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_2(in_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 = 1.0
tmp4 = tmp2 * tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kj/ckjie4jfhwucdjwrkoxyorl5ltzj7tre6m3jengzvslyxc6fnmbt.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
# 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 = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = 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_1/inductor_cache/bt/cbt34fhodpavx2oq24hpku3ag5gk3mibbr2ozkxz4z7bs2m23gfy.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# multi_head_attention_forward => 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_1/inductor_cache/fe/cfech2rxtnpxedtqlj5kjrdsdntbttrruuwxu7hmdxbzkefn6e6q.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_5 = async_compile.triton('triton_poi_fused_clone_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_5(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 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 + (4*x1)), xmask & ymask)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4s/c4sghfflzlvhg76ckzrkj2yv7ethqif4fe437f2dxqtwuh4ahinp.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_2
# x_3 => var_mean_1
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %squeeze), kwargs = {})
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_2, [1]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 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 + (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_1/inductor_cache/cm/ccma2nw3h6tzmwtufo45byghevpzamho2ay363inl5jh2l67anpn.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_2
# x_3 => add_3, add_4, mul_3, mul_4, rsqrt_1, sub_2
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %squeeze), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_8, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_3,), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %getitem_9), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, %rsqrt_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %primals_8), kwargs = {})
# %add_4 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %primals_9), kwargs = {})
triton_poi_fused_add_native_layer_norm_7 = async_compile.triton('triton_poi_fused_add_native_layer_norm_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
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_1/inductor_cache/xy/cxyww2txopwpx25tsbwly3yxypvgzdkmo5kybevltblykxeczke5.py
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_4 => relu
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_11), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_1,), kwargs = {})
triton_poi_fused_relu_8 = async_compile.triton('triton_poi_fused_relu_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_8', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_8(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_1/inductor_cache/cb/ccbk5xotcsqtl2ksko2mwsbwmw3wuum5asogvfe7nzhucp3hqc2c.py
# Topologically Sorted Source Nodes: [x_2, x_8], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x_2 => add_2
# x_8 => add_5
# Graph fragment:
# %add_2 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %squeeze), kwargs = {})
# %add_tensor : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default, %primals_13), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %add_tensor), kwargs = {})
triton_poi_fused_add_9 = async_compile.triton('triton_poi_fused_add_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_9(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
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 = 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, ), (1, ))
assert_size_stride(primals_4, (12, 4), (4, 1))
assert_size_stride(primals_5, (12, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (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, 4), (4, 1))
assert_size_stride(primals_13, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf1 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_0.run(primals_1, buf0, buf1, 4, grid=grid(4), stream=stream0)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(primals_1, buf0, buf1, primals_2, primals_3, buf2, 16, grid=grid(16), stream=stream0)
del primals_2
del primals_3
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_5, (4, ), (1, ), 4), buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 16), alpha=1, beta=1, out=buf4)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_5, (4, ), (1, ), 8), buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 32), alpha=1, beta=1, out=buf5)
buf6 = reinterpret_tensor(buf3, (4, 4, 1), (1, 4, 16), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(buf6, primals_5, 16, grid=grid(16), stream=stream0)
del primals_5
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(buf6, reinterpret_tensor(buf4, (4, 1, 4), (1, 1, 4), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf7, buf8, 64, grid=grid(64), stream=stream0)
buf9 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_4.run(buf8, buf9, 64, grid=grid(64), stream=stream0)
del buf8
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.bmm]
extern_kernels.bmm(buf9, reinterpret_tensor(buf5, (4, 4, 1), (1, 4, 1), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.clone]
triton_poi_fused_clone_5.run(buf10, buf11, 4, 4, grid=grid(4, 4), stream=stream0)
buf12 = reinterpret_tensor(buf10, (4, 4), (4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf11, (4, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf12)
del primals_7
buf13 = buf1; del buf1 # reuse
buf14 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(primals_1, buf12, buf13, buf14, 4, grid=grid(4), stream=stream0)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_7.run(primals_1, buf12, buf13, buf14, primals_8, primals_9, buf15, 16, grid=grid(16), stream=stream0)
del buf13
del buf14
del primals_9
buf16 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf15, reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), out=buf16)
buf17 = buf16; del buf16 # reuse
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
triton_poi_fused_relu_8.run(buf17, primals_11, 16, grid=grid(16), stream=stream0)
del primals_11
buf18 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf17, reinterpret_tensor(primals_12, (4, 4), (1, 4), 0), out=buf18)
buf19 = buf18; del buf18 # reuse
# Topologically Sorted Source Nodes: [x_2, x_8], Original ATen: [aten.add]
triton_poi_fused_add_9.run(buf19, primals_1, buf12, primals_13, 16, grid=grid(16), stream=stream0)
del primals_13
return (buf19, primals_1, primals_8, buf2, buf9, reinterpret_tensor(buf11, (4, 4), (4, 1), 0), buf12, buf15, buf17, primals_12, primals_10, primals_6, reinterpret_tensor(buf5, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf6, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf4, (4, 4, 1), (1, 4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (4, 1), 32), reinterpret_tensor(primals_4, (4, 4), (4, 1), 16), reinterpret_tensor(primals_4, (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, 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((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((12, ), (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, ), (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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.utils.data
import torch.distributions
class TransformerEncoderLayer(nn.Module):
def __init__(self, embed_dim, num_heads, hidden_size, dropout=0.0,
attention_dropout=0.0, activation_dropout=0.0):
super().__init__()
self.embed_dim = embed_dim
self.self_attn = torch.nn.MultiheadAttention(embed_dim=self.
embed_dim, num_heads=num_heads, dropout=attention_dropout)
self.self_attn_layer_norm = torch.nn.LayerNorm(self.embed_dim)
self.dropout = dropout
self.activation_dropout = activation_dropout
self.normalize_before = True
self.fc1 = torch.nn.Linear(self.embed_dim, hidden_size)
self.fc2 = torch.nn.Linear(hidden_size, self.embed_dim)
self.layer_norm = torch.nn.LayerNorm(self.embed_dim)
self.init_parameters()
def forward(self, x, key_padding_mask=None, attn_mask=None):
residual = x
x = self.self_attn_layer_norm(x)
x, _att = self.self_attn(query=x, key=x, value=x, key_padding_mask=
key_padding_mask, attn_mask=attn_mask)
x = F.dropout(x, p=self.dropout, training=self.training)
x = residual + x
residual = x
x = self.layer_norm(x)
x = F.relu(self.fc1(x))
x = F.dropout(x, p=self.activation_dropout, training=self.training)
x = self.fc2(x)
x = F.dropout(x, p=self.dropout, training=self.training)
x = residual + x
return x
def init_parameters(self):
nn.init.xavier_uniform_(self.fc1.weight)
nn.init.constant_(self.fc1.bias, 0.0)
nn.init.xavier_uniform_(self.fc2.weight)
nn.init.constant_(self.fc2.bias, 0.0)
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'embed_dim': 4, 'num_heads': 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, math as tl_math
import torch.nn as nn
import torch.nn.parallel
import torch.utils.data
import torch.distributions
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 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_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_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
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_mul_2(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 = 1.0
tmp4 = tmp2 * tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_4(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 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_clone_5(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 4
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 + 4 * x1), xmask & ymask)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 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 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
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_8(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_add_9(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel,
XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
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) = 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,), (1,))
assert_size_stride(primals_4, (12, 4), (4, 1))
assert_size_stride(primals_5, (12,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (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, 4), (4, 1))
assert_size_stride(primals_13, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf1 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused_native_layer_norm_0[grid(4)](primals_1, buf0, buf1,
4, XBLOCK=4, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(16)](primals_1, buf0,
buf1, primals_2, primals_3, buf2, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del primals_2
del primals_3
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4
), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_5, (4,), (1,), 4),
buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 16), alpha=
1, beta=1, out=buf4)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_5, (4,), (1,), 8),
buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 32), alpha=
1, beta=1, out=buf5)
buf6 = reinterpret_tensor(buf3, (4, 4, 1), (1, 4, 16), 0)
del buf3
triton_poi_fused_mul_2[grid(16)](buf6, primals_5, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_5
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(buf6, reinterpret_tensor(buf4, (4, 1, 4), (1, 1,
4), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_3[grid(64)](buf7, buf8, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf9 = buf7
del buf7
triton_poi_fused__softmax_4[grid(64)](buf8, buf9, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf8
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(buf9, reinterpret_tensor(buf5, (4, 4, 1), (1, 4,
1), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
triton_poi_fused_clone_5[grid(4, 4)](buf10, buf11, 4, 4, XBLOCK=4,
YBLOCK=4, num_warps=1, num_stages=1)
buf12 = reinterpret_tensor(buf10, (4, 4), (4, 1), 0)
del buf10
extern_kernels.addmm(primals_7, reinterpret_tensor(buf11, (4, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf12)
del primals_7
buf13 = buf1
del buf1
buf14 = buf0
del buf0
triton_poi_fused_add_native_layer_norm_6[grid(4)](primals_1, buf12,
buf13, buf14, 4, XBLOCK=4, num_warps=1, num_stages=1)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_7[grid(16)](primals_1, buf12,
buf13, buf14, primals_8, primals_9, buf15, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf13
del buf14
del primals_9
buf16 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf15, reinterpret_tensor(primals_10, (4, 4), (1,
4), 0), out=buf16)
buf17 = buf16
del buf16
triton_poi_fused_relu_8[grid(16)](buf17, primals_11, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_11
buf18 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf17, reinterpret_tensor(primals_12, (4, 4), (1,
4), 0), out=buf18)
buf19 = buf18
del buf18
triton_poi_fused_add_9[grid(16)](buf19, primals_1, buf12,
primals_13, 16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_13
return (buf19, primals_1, primals_8, buf2, buf9, reinterpret_tensor(
buf11, (4, 4), (4, 1), 0), buf12, buf15, buf17, primals_12,
primals_10, primals_6, reinterpret_tensor(buf5, (4, 1, 4), (1, 1, 4
), 0), reinterpret_tensor(buf6, (4, 1, 4), (1, 1, 4), 0),
reinterpret_tensor(buf4, (4, 4, 1), (1, 4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (4, 1), 32),
reinterpret_tensor(primals_4, (4, 4), (4, 1), 16),
reinterpret_tensor(primals_4, (4, 4), (4, 1), 0))
class TransformerEncoderLayerNew(nn.Module):
def __init__(self, embed_dim, num_heads, hidden_size, dropout=0.0,
attention_dropout=0.0, activation_dropout=0.0):
super().__init__()
self.embed_dim = embed_dim
self.self_attn = torch.nn.MultiheadAttention(embed_dim=self.
embed_dim, num_heads=num_heads, dropout=attention_dropout)
self.self_attn_layer_norm = torch.nn.LayerNorm(self.embed_dim)
self.dropout = dropout
self.activation_dropout = activation_dropout
self.normalize_before = True
self.fc1 = torch.nn.Linear(self.embed_dim, hidden_size)
self.fc2 = torch.nn.Linear(hidden_size, self.embed_dim)
self.layer_norm = torch.nn.LayerNorm(self.embed_dim)
self.init_parameters()
def init_parameters(self):
nn.init.xavier_uniform_(self.fc1.weight)
nn.init.constant_(self.fc1.bias, 0.0)
nn.init.xavier_uniform_(self.fc2.weight)
nn.init.constant_(self.fc2.bias, 0.0)
def forward(self, input_0):
primals_4 = self.self_attn.in_proj_weight
primals_5 = self.self_attn.in_proj_bias
primals_1 = self.self_attn.out_proj.weight
primals_2 = self.self_attn.out_proj.bias
primals_3 = self.self_attn_layer_norm.weight
primals_7 = self.self_attn_layer_norm.bias
primals_6 = self.fc1.weight
primals_8 = self.fc1.bias
primals_10 = self.fc2.weight
primals_9 = self.fc2.bias
primals_11 = self.layer_norm.weight
primals_13 = self.layer_norm.bias
primals_12 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13])
return output[0]
| Daetheys/Lazimpa | TransformerEncoderLayer | false | 7,997 | [
"MIT"
] | 15 | 21f4f4ebcdfa8b6a775b64673dd3001763c91cf1 | https://github.com/Daetheys/Lazimpa/tree/21f4f4ebcdfa8b6a775b64673dd3001763c91cf1 | import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.utils.data
import torch.distributions
class Model(nn.Module):
def __init__(self, embed_dim, num_heads, hidden_size, dropout=0.0,
attention_dropout=0.0, activation_dropout=0.0):
super().__init__()
self.embed_dim = embed_dim
self.self_attn = torch.nn.MultiheadAttention(embed_dim=self.
embed_dim, num_heads=num_heads, dropout=attention_dropout)
self.self_attn_layer_norm = torch.nn.LayerNorm(self.embed_dim)
self.dropout = dropout
self.activation_dropout = activation_dropout
self.normalize_before = True
self.fc1 = torch.nn.Linear(self.embed_dim, hidden_size)
self.fc2 = torch.nn.Linear(hidden_size, self.embed_dim)
self.layer_norm = torch.nn.LayerNorm(self.embed_dim)
self.init_parameters()
def forward(self, x, key_padding_mask=None, attn_mask=None):
residual = x
x = self.self_attn_layer_norm(x)
x, _att = self.self_attn(query=x, key=x, value=x, key_padding_mask=
key_padding_mask, attn_mask=attn_mask)
x = F.dropout(x, p=self.dropout, training=self.training)
x = residual + x
residual = x
x = self.layer_norm(x)
x = F.relu(self.fc1(x))
x = F.dropout(x, p=self.activation_dropout, training=self.training)
x = self.fc2(x)
x = F.dropout(x, p=self.dropout, training=self.training)
x = residual + x
return x
def init_parameters(self):
nn.init.xavier_uniform_(self.fc1.weight)
nn.init.constant_(self.fc1.bias, 0.0)
nn.init.xavier_uniform_(self.fc2.weight)
nn.init.constant_(self.fc2.bias, 0.0)
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
SelfAttn | # 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_1/inductor_cache/lq/clqobdukzpbkgeirgnvgulpmn2isxuyjfio7deusrshagkjvl4li.py
# Topologically Sorted Source Nodes: [sub, mul, sub_1, scores], Original ATen: [aten.rsub, aten.mul, aten.sub, aten._softmax]
# Source node to ATen node mapping:
# mul => mul
# scores => amax, exp, sub_2, sum_1
# sub => sub
# sub_1 => sub_1
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %primals_4), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, 1e+20), kwargs = {})
# %sub_1 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%squeeze, %mul), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%sub_1, [1], True), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_2,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
triton_poi_fused__softmax_mul_rsub_sub_0 = async_compile.triton('triton_poi_fused__softmax_mul_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.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mul_rsub_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mul_rsub_sub_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr2 + (4*x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr2 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr2 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr2 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tmp0 + tmp2
tmp5 = 1.0
tmp6 = tmp5 - tmp4
tmp7 = 1e+20
tmp8 = tmp6 * tmp7
tmp9 = tmp3 - tmp8
tmp11 = tmp10 + tmp2
tmp13 = tmp5 - tmp12
tmp14 = tmp13 * tmp7
tmp15 = tmp11 - tmp14
tmp16 = triton_helpers.maximum(tmp9, tmp15)
tmp18 = tmp17 + tmp2
tmp20 = tmp5 - tmp19
tmp21 = tmp20 * tmp7
tmp22 = tmp18 - tmp21
tmp23 = triton_helpers.maximum(tmp16, tmp22)
tmp25 = tmp24 + tmp2
tmp27 = tmp5 - tmp26
tmp28 = tmp27 * tmp7
tmp29 = tmp25 - tmp28
tmp30 = triton_helpers.maximum(tmp23, tmp29)
tmp31 = tmp9 - tmp30
tmp32 = tl_math.exp(tmp31)
tmp33 = tmp15 - tmp30
tmp34 = tl_math.exp(tmp33)
tmp35 = tmp32 + tmp34
tmp36 = tmp22 - tmp30
tmp37 = tl_math.exp(tmp36)
tmp38 = tmp35 + tmp37
tmp39 = tmp29 - tmp30
tmp40 = tl_math.exp(tmp39)
tmp41 = tmp38 + tmp40
tl.store(out_ptr0 + (x0), tmp30, xmask)
tl.store(out_ptr1 + (x0), tmp41, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/da/cdaw666cppfgx6wcwv4ckhvc524daustk2y2woennedczz4cimzv.py
# Topologically Sorted Source Nodes: [sub, mul, sub_1, scores], Original ATen: [aten.rsub, aten.mul, aten.sub, aten._softmax]
# Source node to ATen node mapping:
# mul => mul
# scores => div, exp, sub_2
# sub => sub
# sub_1 => sub_1
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %primals_4), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, 1e+20), kwargs = {})
# %sub_1 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%squeeze, %mul), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_2,), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_mul_rsub_sub_1 = async_compile.triton('triton_poi_fused__softmax_mul_rsub_sub_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mul_rsub_sub_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mul_rsub_sub_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_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_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr1 + (x2), xmask)
tmp10 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tmp0 + tmp2
tmp5 = 1.0
tmp6 = tmp5 - tmp4
tmp7 = 1e+20
tmp8 = tmp6 * tmp7
tmp9 = tmp3 - tmp8
tmp11 = tmp9 - tmp10
tmp12 = tl_math.exp(tmp11)
tmp14 = tmp12 / tmp13
tl.store(in_out_ptr0 + (x2), tmp14, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/m2/cm25t6aoec4n6sgspbcsocz27g2angvsw2gvg7feqwzkrgwwi6dq.py
# Topologically Sorted Source Nodes: [context], Original ATen: [aten.sum]
# Source node to ATen node mapping:
# context => sum_2
# Graph fragment:
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%expand, [1]), kwargs = {})
triton_poi_fused_sum_2 = async_compile.triton('triton_poi_fused_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sum_2(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
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
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), (16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
assert_size_stride(primals_3, (1, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 1), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf2 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [sub, mul, sub_1, scores], Original ATen: [aten.rsub, aten.mul, aten.sub, aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_mul_rsub_sub_0.run(buf0, primals_3, primals_4, buf1, buf2, 4, grid=grid(4), stream=stream0)
buf3 = reinterpret_tensor(buf0, (4, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [sub, mul, sub_1, scores], Original ATen: [aten.rsub, aten.mul, aten.sub, aten._softmax]
triton_poi_fused__softmax_mul_rsub_sub_1.run(buf3, primals_3, primals_4, buf1, buf2, 16, grid=grid(16), stream=stream0)
del buf1
del buf2
del primals_3
del primals_4
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [context], Original ATen: [aten.sum]
triton_poi_fused_sum_2.run(buf3, buf4, 16, grid=grid(16), stream=stream0)
return (buf4, buf3, reinterpret_tensor(primals_1, (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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 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
from torch import nn
from torch.nn import functional as F
class SelfAttn(nn.Module):
def __init__(self, denc, dropout=0, enc_trans=None, dec_trans=None):
super().__init__()
self.scorer = nn.Linear(denc, 1)
self.dropout = nn.Dropout(dropout)
self.enc_trans = enc_trans or nn.Identity()
def forward(self, enc, enc_mask):
enc_trans = self.dropout(self.enc_trans(enc))
raw_scores = self.scorer(enc).squeeze(2)
scores = F.softmax(raw_scores - (1 - enc_mask) * 1e+20, dim=1)
context = scores.unsqueeze(2).expand_as(enc_trans).sum(1)
return context, scores
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'denc': 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__softmax_mul_rsub_sub_0(in_ptr0, in_ptr1, in_ptr2,
out_ptr0, out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr2 + 4 * x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr2 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp17 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp19 = tl.load(in_ptr2 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp24 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp26 = tl.load(in_ptr2 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp3 = tmp0 + tmp2
tmp5 = 1.0
tmp6 = tmp5 - tmp4
tmp7 = 1e+20
tmp8 = tmp6 * tmp7
tmp9 = tmp3 - tmp8
tmp11 = tmp10 + tmp2
tmp13 = tmp5 - tmp12
tmp14 = tmp13 * tmp7
tmp15 = tmp11 - tmp14
tmp16 = triton_helpers.maximum(tmp9, tmp15)
tmp18 = tmp17 + tmp2
tmp20 = tmp5 - tmp19
tmp21 = tmp20 * tmp7
tmp22 = tmp18 - tmp21
tmp23 = triton_helpers.maximum(tmp16, tmp22)
tmp25 = tmp24 + tmp2
tmp27 = tmp5 - tmp26
tmp28 = tmp27 * tmp7
tmp29 = tmp25 - tmp28
tmp30 = triton_helpers.maximum(tmp23, tmp29)
tmp31 = tmp9 - tmp30
tmp32 = tl_math.exp(tmp31)
tmp33 = tmp15 - tmp30
tmp34 = tl_math.exp(tmp33)
tmp35 = tmp32 + tmp34
tmp36 = tmp22 - tmp30
tmp37 = tl_math.exp(tmp36)
tmp38 = tmp35 + tmp37
tmp39 = tmp29 - tmp30
tmp40 = tl_math.exp(tmp39)
tmp41 = tmp38 + tmp40
tl.store(out_ptr0 + x0, tmp30, xmask)
tl.store(out_ptr1 + x0, tmp41, xmask)
@triton.jit
def triton_poi_fused__softmax_mul_rsub_sub_1(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, in_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_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp4 = tl.load(in_ptr1 + x2, xmask)
tmp10 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp3 = tmp0 + tmp2
tmp5 = 1.0
tmp6 = tmp5 - tmp4
tmp7 = 1e+20
tmp8 = tmp6 * tmp7
tmp9 = tmp3 - tmp8
tmp11 = tmp9 - tmp10
tmp12 = tl_math.exp(tmp11)
tmp14 = tmp12 / tmp13
tl.store(in_out_ptr0 + x2, tmp14, xmask)
@triton.jit
def triton_poi_fused_sum_2(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
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
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), (16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
assert_size_stride(primals_3, (1,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 1), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf2 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_mul_rsub_sub_0[grid(4)](buf0, primals_3,
primals_4, buf1, buf2, 4, XBLOCK=4, num_warps=1, num_stages=1)
buf3 = reinterpret_tensor(buf0, (4, 4), (4, 1), 0)
del buf0
triton_poi_fused__softmax_mul_rsub_sub_1[grid(16)](buf3, primals_3,
primals_4, buf1, buf2, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf1
del buf2
del primals_3
del primals_4
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_sum_2[grid(16)](buf3, buf4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
return buf4, buf3, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), buf3
class SelfAttnNew(nn.Module):
def __init__(self, denc, dropout=0, enc_trans=None, dec_trans=None):
super().__init__()
self.scorer = nn.Linear(denc, 1)
self.dropout = nn.Dropout(dropout)
self.enc_trans = enc_trans or nn.Identity()
def forward(self, input_0, input_1):
primals_2 = self.scorer.weight
primals_3 = self.scorer.bias
primals_1 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0], output[1]
| DreamerDeo/gazp | SelfAttn | false | 7,998 | [
"MIT"
] | 18 | 5f823a447ffdf5176023a01516d2be7c383294d9 | https://github.com/DreamerDeo/gazp/tree/5f823a447ffdf5176023a01516d2be7c383294d9 | import torch
from torch import nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self, denc, dropout=0, enc_trans=None, dec_trans=None):
super().__init__()
self.scorer = nn.Linear(denc, 1)
self.dropout = nn.Dropout(dropout)
self.enc_trans = enc_trans or nn.Identity()
def forward(self, enc, enc_mask):
enc_trans = self.dropout(self.enc_trans(enc))
raw_scores = self.scorer(enc).squeeze(2)
scores = F.softmax(raw_scores - (1 - enc_mask) * 1e+20, dim=1)
context = scores.unsqueeze(2).expand_as(enc_trans).sum(1)
return context, scores
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [4]
|
SpatialAttentionGate | # AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/qz/cqzt2jcpw5eyyq4emv5yflvehemjealzawvfp4gtzmley5qjynad.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x => convolution
# x_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 16
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/fd/cfdejdfurpf2un4x3bgw4ukzgktg7sd5dxdymhhm7tlyetyrhhwq.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.convolution, aten.sigmoid]
# Source node to ATen node mapping:
# x_2 => convolution_1
# x_3 => sigmoid
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_sigmoid_1 = async_compile.triton('triton_poi_fused_convolution_sigmoid_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_sigmoid_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 16, 1, 1), (16, 1, 1, 1))
assert_size_stride(primals_5, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 1024, grid=grid(1024), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 4, 4), (16, 16, 4, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.convolution, aten.sigmoid]
triton_poi_fused_convolution_sigmoid_1.run(buf3, primals_5, 64, grid=grid(64), stream=stream0)
del primals_5
return (buf3, primals_1, primals_3, primals_4, buf1, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((16, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 16, 1, 1), (16, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
from torch.nn import functional as F
class SpatialAttentionGate(nn.Module):
def __init__(self, channel, reduction=16):
super(SpatialAttentionGate, self).__init__()
self.fc1 = nn.Conv2d(channel, reduction, kernel_size=1, padding=0)
self.fc2 = nn.Conv2d(reduction, 1, kernel_size=1, padding=0)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x, inplace=True)
x = self.fc2(x)
x = torch.sigmoid(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'channel': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 16
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, xmask)
@triton.jit
def triton_poi_fused_convolution_sigmoid_1(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 16, 1, 1), (16, 1, 1, 1))
assert_size_stride(primals_5, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(1024)](buf1, primals_2,
1024, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf1, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 4, 4), (16, 16, 4, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_sigmoid_1[grid(64)](buf3, primals_5,
64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_5
return buf3, primals_1, primals_3, primals_4, buf1, buf3
class SpatialAttentionGateNew(nn.Module):
def __init__(self, channel, reduction=16):
super(SpatialAttentionGateNew, self).__init__()
self.fc1 = nn.Conv2d(channel, reduction, kernel_size=1, padding=0)
self.fc2 = nn.Conv2d(reduction, 1, kernel_size=1, padding=0)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| Diyago/Automatic-salt-deposits-segmentation | SpatialAttentionGate | false | 7,999 | [
"MIT"
] | 20 | fedfc7f1d9878674382eeb16a820b5f16791f4ab | https://github.com/Diyago/Automatic-salt-deposits-segmentation/tree/fedfc7f1d9878674382eeb16a820b5f16791f4ab | import torch
from torch import nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self, channel, reduction=16):
super().__init__()
self.fc1 = nn.Conv2d(channel, reduction, kernel_size=1, padding=0)
self.fc2 = nn.Conv2d(reduction, 1, kernel_size=1, padding=0)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x, inplace=True)
x = self.fc2(x)
x = torch.sigmoid(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
Whitening2d | # 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_1/inductor_cache/lz/clzp2u5x7odof4624bplw5iwjqpu24ukxra2zq3cujrwqigql6xc.py
# Topologically Sorted Source Nodes: [xn], Original ATen: [aten.sub]
# Source node to ATen node mapping:
# xn => sub
# Graph fragment:
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%unsqueeze_1, %view_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = 1.0
tmp11 = tmp9 / tmp10
tmp12 = tmp0 - tmp11
tl.store(out_ptr0 + (x2), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/lu/cluhknhj7j2cqjddc6fycskajvfidkt4fac4yep5d5qa6dejhl5j.py
# Topologically Sorted Source Nodes: [contiguous, T], Original ATen: [aten.clone, aten.view]
# Source node to ATen node mapping:
# T => view_2
# contiguous => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute,), kwargs = {memory_format: torch.contiguous_format})
# %view_2 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone, [4, -1]), kwargs = {})
triton_poi_fused_clone_view_1 = async_compile.triton('triton_poi_fused_clone_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=[4, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_view_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_view_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
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 + (4*x1)), xmask & ymask)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/iz/ciz2wzehddvyei6bay7meigtt26gdd5j5kpopcfphm2bcyt3gpmw.py
# Topologically Sorted Source Nodes: [f_cov, mul, eye, eye_1, mul_1, f_cov_shrinked], Original ATen: [aten.div, aten.mul, aten.eye, aten._to_copy, aten.add]
# Source node to ATen node mapping:
# eye => eq, full_default, full_default_1, iota_1, where
# eye_1 => device_put
# f_cov => div
# f_cov_shrinked => add
# mul => mul
# mul_1 => mul_1
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mm, 3), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, 1.0), kwargs = {})
# %iota_1 : [num_users=1] = call_function[target=torch.ops.prims.iota.default](args = (4,), kwargs = {start: 0, step: 1, dtype: torch.int64, device: cpu, requires_grad: False})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_2, %iota_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %full_default_1 : [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})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %full_default_1), kwargs = {})
# %device_put : [num_users=2] = call_function[target=torch.ops.prims.device_put.default](args = (%where, cuda:0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%device_put, 0.0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
triton_poi_fused__to_copy_add_div_eye_mul_2 = async_compile.triton('triton_poi_fused__to_copy_add_div_eye_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_add_div_eye_mul_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_div_eye_mul_2(in_out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = 0.3333333333333333
tmp2 = tmp0 * tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tmp5 = x1
tmp6 = x0
tmp7 = tmp5 == tmp6
tmp8 = 0.0
tmp9 = tl.where(tmp7, tmp3, tmp8)
tmp10 = tmp9 * tmp8
tmp11 = tmp4 + tmp10
tl.store(in_out_ptr0 + (x2), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ly/clyi2jg2po37gz3zpumtku76hwseyb7ioz27h7gei2qanxvmlhmn.py
# Topologically Sorted Source Nodes: [eye, eye_1], Original ATen: [aten.eye, aten._to_copy]
# Source node to ATen node mapping:
# eye => eq, full_default, full_default_1, iota_1, where
# eye_1 => device_put
# Graph fragment:
# %iota_1 : [num_users=1] = call_function[target=torch.ops.prims.iota.default](args = (4,), kwargs = {start: 0, step: 1, dtype: torch.int64, device: cpu, requires_grad: False})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Tensor](args = (%unsqueeze_2, %iota_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([1], 1), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %full_default_1 : [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})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %full_default_1), kwargs = {})
# %device_put : [num_users=2] = call_function[target=torch.ops.prims.device_put.default](args = (%where, cuda:0), kwargs = {})
triton_poi_fused__to_copy_eye_3 = async_compile.triton('triton_poi_fused__to_copy_eye_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_eye_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_eye_3(out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = x1
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tl.store(out_ptr0 + (x2), tmp5, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 16, 16), torch.float32)
# Topologically Sorted Source Nodes: [xn], Original ATen: [aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_sub_0.run(arg0_1, buf0, 16, grid=grid(16), stream=stream0)
del arg0_1
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous, T], Original ATen: [aten.clone, aten.view]
triton_poi_fused_clone_view_1.run(buf0, buf1, 4, 4, grid=grid(4, 4), stream=stream0)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mm], Original ATen: [aten.mm]
extern_kernels.mm(buf1, reinterpret_tensor(buf1, (4, 4), (1, 4), 0), out=buf2)
del buf1
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [f_cov, mul, eye, eye_1, mul_1, f_cov_shrinked], Original ATen: [aten.div, aten.mul, aten.eye, aten._to_copy, aten.add]
triton_poi_fused__to_copy_add_div_eye_mul_2.run(buf3, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [f_cov, mul, eye, eye_1, mul_1, f_cov_shrinked, cholesky], Original ATen: [aten.div, aten.mul, aten.eye, aten._to_copy, aten.add, aten.cholesky]
buf4 = torch.ops.aten.cholesky.default(buf3)
buf5 = buf4
del buf4
buf6 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [eye, eye_1], Original ATen: [aten.eye, aten._to_copy]
triton_poi_fused__to_copy_eye_3.run(buf6, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [eye, eye_1, triangular_solve], Original ATen: [aten.eye, aten._to_copy, aten.triangular_solve]
buf7 = torch.ops.aten.triangular_solve.default(buf6, buf5, False)
del buf5
buf8 = buf7[0]
del buf7
buf10 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_view_1.run(buf8, buf10, 4, 4, grid=grid(4, 4), stream=stream0)
del buf8
# Topologically Sorted Source Nodes: [decorrelated], Original ATen: [aten.convolution]
buf11 = extern_kernels.convolution(buf0, reinterpret_tensor(buf10, (4, 4, 1, 1), (4, 1, 0, 0), 0), stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf11, (4, 4, 1, 1), (4, 1, 1, 1))
del buf0
del buf10
return (reinterpret_tensor(buf11, (4, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4), (4, 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
from torch.cuda.amp import custom_fwd
from torch.nn.functional import conv2d
class Whitening2d(nn.Module):
def __init__(self, output_dim: 'int', eps: 'float'=0.0):
"""Layer that computes hard whitening for W-MSE using the Cholesky decomposition.
Args:
output_dim (int): number of dimension of projected features.
eps (float, optional): eps for numerical stability in Cholesky decomposition. Defaults
to 0.0.
"""
super(Whitening2d, self).__init__()
self.output_dim = output_dim
self.eps = eps
@custom_fwd(cast_inputs=torch.float32)
def forward(self, x: 'torch.Tensor') ->torch.Tensor:
"""Performs whitening using the Cholesky decomposition.
Args:
x (torch.Tensor): a batch or slice of projected features.
Returns:
torch.Tensor: a batch or slice of whitened features.
"""
x = x.unsqueeze(2).unsqueeze(3)
m = x.mean(0).view(self.output_dim, -1).mean(-1).view(1, -1, 1, 1)
xn = x - m
T = xn.permute(1, 0, 2, 3).contiguous().view(self.output_dim, -1)
f_cov = torch.mm(T, T.permute(1, 0)) / (T.shape[-1] - 1)
eye = torch.eye(self.output_dim).type(f_cov.type())
f_cov_shrinked = (1 - self.eps) * f_cov + self.eps * eye
inv_sqrt = torch.triangular_solve(eye, torch.cholesky(
f_cov_shrinked), upper=False)[0]
inv_sqrt = inv_sqrt.contiguous().view(self.output_dim, self.
output_dim, 1, 1)
decorrelated = conv2d(xn, inv_sqrt)
return decorrelated.squeeze(2).squeeze(2)
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'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 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_sub_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
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = 1.0
tmp11 = tmp9 / tmp10
tmp12 = tmp0 - tmp11
tl.store(out_ptr0 + x2, tmp12, xmask)
@triton.jit
def triton_poi_fused_clone_view_1(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK:
tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 4
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 + 4 * x1), xmask & ymask)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused__to_copy_add_div_eye_mul_2(in_out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = 0.3333333333333333
tmp2 = tmp0 * tmp1
tmp3 = 1.0
tmp4 = tmp2 * tmp3
tmp5 = x1
tmp6 = x0
tmp7 = tmp5 == tmp6
tmp8 = 0.0
tmp9 = tl.where(tmp7, tmp3, tmp8)
tmp10 = tmp9 * tmp8
tmp11 = tmp4 + tmp10
tl.store(in_out_ptr0 + x2, tmp11, xmask)
@triton.jit
def triton_poi_fused__to_copy_eye_3(out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = x1
tmp1 = x0
tmp2 = tmp0 == tmp1
tmp3 = 1.0
tmp4 = 0.0
tmp5 = tl.where(tmp2, tmp3, tmp4)
tl.store(out_ptr0 + x2, tmp5, xmask)
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, 1, 1), (4, 1, 16, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_sub_0[grid(16)](arg0_1, buf0, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_clone_view_1[grid(4, 4)](buf0, buf1, 4, 4, XBLOCK=
4, YBLOCK=4, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf1, reinterpret_tensor(buf1, (4, 4), (1, 4), 0),
out=buf2)
del buf1
buf3 = buf2
del buf2
triton_poi_fused__to_copy_add_div_eye_mul_2[grid(16)](buf3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf4 = torch.ops.aten.cholesky.default(buf3)
buf5 = buf4
del buf4
buf6 = buf3
del buf3
triton_poi_fused__to_copy_eye_3[grid(16)](buf6, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf7 = torch.ops.aten.triangular_solve.default(buf6, buf5, False)
del buf5
buf8 = buf7[0]
del buf7
buf10 = buf6
del buf6
triton_poi_fused_clone_view_1[grid(4, 4)](buf8, buf10, 4, 4, XBLOCK
=4, YBLOCK=4, num_warps=1, num_stages=1)
del buf8
buf11 = extern_kernels.convolution(buf0, reinterpret_tensor(buf10,
(4, 4, 1, 1), (4, 1, 0, 0), 0), stride=(1, 1), padding=(0, 0),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=1, bias=None)
assert_size_stride(buf11, (4, 4, 1, 1), (4, 1, 1, 1))
del buf0
del buf10
return reinterpret_tensor(buf11, (4, 4), (4, 1), 0),
class Whitening2dNew(nn.Module):
def __init__(self, output_dim: 'int', eps: 'float'=0.0):
"""Layer that computes hard whitening for W-MSE using the Cholesky decomposition.
Args:
output_dim (int): number of dimension of projected features.
eps (float, optional): eps for numerical stability in Cholesky decomposition. Defaults
to 0.0.
"""
super(Whitening2dNew, self).__init__()
self.output_dim = output_dim
self.eps = eps
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DonkeyShot21/cassle | Whitening2d | false | 8,000 | [
"MIT"
] | 13 | d25f9c7cb5e822660dc1ef03e7fac09a33d0b1a8 | https://github.com/DonkeyShot21/cassle/tree/d25f9c7cb5e822660dc1ef03e7fac09a33d0b1a8 | import torch
from torch import nn
from torch.cuda.amp import custom_fwd
from torch.nn.functional import conv2d
class Model(nn.Module):
def __init__(self, output_dim: 'int', eps: 'float'=0.0):
"""Layer that computes hard whitening for W-MSE using the Cholesky decomposition.
Args:
output_dim (int): number of dimension of projected features.
eps (float, optional): eps for numerical stability in Cholesky decomposition. Defaults
to 0.0.
"""
super().__init__()
self.output_dim = output_dim
self.eps = eps
@custom_fwd(cast_inputs=torch.float32)
def forward(self, x: 'torch.Tensor') ->torch.Tensor:
"""Performs whitening using the Cholesky decomposition.
Args:
x (torch.Tensor): a batch or slice of projected features.
Returns:
torch.Tensor: a batch or slice of whitened features.
"""
x = x.unsqueeze(2).unsqueeze(3)
m = x.mean(0).view(self.output_dim, -1).mean(-1).view(1, -1, 1, 1)
xn = x - m
T = xn.permute(1, 0, 2, 3).contiguous().view(self.output_dim, -1)
f_cov = torch.mm(T, T.permute(1, 0)) / (T.shape[-1] - 1)
eye = torch.eye(self.output_dim).type(f_cov.type())
f_cov_shrinked = (1 - self.eps) * f_cov + self.eps * eye
inv_sqrt = torch.triangular_solve(eye, torch.cholesky(
f_cov_shrinked), upper=False)[0]
inv_sqrt = inv_sqrt.contiguous().view(self.output_dim, self.
output_dim, 1, 1)
decorrelated = conv2d(xn, inv_sqrt)
return decorrelated.squeeze(2).squeeze(2)
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [4]
|
GBlock | # 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_1/inductor_cache/ph/cphucpl6cmfcfbzckpaceqnpgsu2eio6xfmyhwyz7gpuknx6lkld.py
# Topologically Sorted Source Nodes: [pow_1, mean, add, rsqrt, h, h_1, x], Original ATen: [aten.pow, aten.mean, aten.add, aten.rsqrt, aten.mul, aten._unsafe_index]
# Source node to ATen node mapping:
# add => add
# h => mul
# h_1 => _unsafe_index
# mean => mean
# pow_1 => pow_1
# rsqrt => rsqrt
# x => _unsafe_index_1
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%primals_1, 2), kwargs = {})
# %mean : [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, 1e-10), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %rsqrt), kwargs = {})
# %_unsafe_index : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%mul, [None, None, %unsqueeze, %convert_element_type_1]), kwargs = {})
# %_unsafe_index_1 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%primals_1, [None, None, %unsqueeze, %convert_element_type_1]), kwargs = {})
triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0 = async_compile.triton('triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 0, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 8) % 8
x0 = xindex % 8
x4 = (xindex // 64)
x3 = (xindex // 256)
x7 = xindex
tmp0 = x1
tmp1 = tmp0.to(tl.float32)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tmp3.to(tl.int32)
tmp5 = x0
tmp6 = tmp5.to(tl.float32)
tmp7 = tmp6 * tmp2
tmp8 = tmp7.to(tl.int32)
tmp9 = tl.load(in_ptr0 + (tmp8 + (4*tmp4) + (16*x4)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (tmp8 + (4*tmp4) + (64*x3)), xmask, eviction_policy='evict_last')
tmp11 = tmp10 * tmp10
tmp12 = tl.load(in_ptr0 + (16 + tmp8 + (4*tmp4) + (64*x3)), xmask, eviction_policy='evict_last')
tmp13 = tmp12 * tmp12
tmp14 = tmp11 + tmp13
tmp15 = tl.load(in_ptr0 + (32 + tmp8 + (4*tmp4) + (64*x3)), xmask, eviction_policy='evict_last')
tmp16 = tmp15 * tmp15
tmp17 = tmp14 + tmp16
tmp18 = tl.load(in_ptr0 + (48 + tmp8 + (4*tmp4) + (64*x3)), xmask, eviction_policy='evict_last')
tmp19 = tmp18 * tmp18
tmp20 = tmp17 + tmp19
tmp21 = 4.0
tmp22 = tmp20 / tmp21
tmp23 = 1e-10
tmp24 = tmp22 + tmp23
tmp25 = libdevice.rsqrt(tmp24)
tmp26 = tmp9 * tmp25
tl.store(out_ptr0 + (x7), tmp26, xmask)
tl.store(out_ptr1 + (x7), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/q4/cq4ddsy7iixflpiikxg4xy7augzxz6ai5j7roltn46gbzixj3hs5.py
# Topologically Sorted Source Nodes: [mv, norm, add_1, truediv], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div]
# Source node to ATen node mapping:
# add_1 => add_9
# mv => mul_9, sum_1
# norm => pow_2, pow_3, sum_2
# truediv => div
# Graph fragment:
# %mul_9 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute, %primals_2), kwargs = {})
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_9, [1]), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_1, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_2, None), kwargs = {})
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 0.5), kwargs = {})
# %add_9 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_3, 1e-10), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, %add_9), kwargs = {})
triton_per_fused_add_div_linalg_vector_norm_mv_1 = async_compile.triton('triton_per_fused_add_div_linalg_vector_norm_mv_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 64],
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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {5: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=(5,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_linalg_vector_norm_mv_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_linalg_vector_norm_mv_1(in_out_ptr0, in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 36
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 = rindex < rnumel
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), rmask, other=0.0)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (36 + r0), rmask, other=0.0)
tmp5 = tl.load(in_ptr1 + (1))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (72 + r0), rmask, other=0.0)
tmp10 = tl.load(in_ptr1 + (2))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (108 + r0), rmask, other=0.0)
tmp15 = tl.load(in_ptr1 + (3))
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.where(rmask, tmp20, 0)
tmp23 = tl.sum(tmp22, 1)[:, None]
tmp24 = libdevice.sqrt(tmp23)
tmp25 = 1e-10
tmp26 = tmp24 + tmp25
tmp27 = tmp18 / tmp26
tl.store(out_ptr0 + (tl.broadcast_to(r0, [XBLOCK, RBLOCK])), tmp18, rmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp26, None)
tl.store(out_ptr1 + (tl.broadcast_to(r0, [XBLOCK, RBLOCK])), tmp27, rmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zs/czsl7thppzzfo4ubzeysosbh2ltoeuasripyv7ajtv4vooqgdh4f.py
# Topologically Sorted Source Nodes: [truediv, mv_1], Original ATen: [aten.div, aten.mv]
# Source node to ATen node mapping:
# mv_1 => mul_10, sum_3
# truediv => div
# Graph fragment:
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, %add_9), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %div), kwargs = {})
# %sum_3 : [num_users=3] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_10, [1]), kwargs = {})
triton_per_fused_div_mv_2 = async_compile.triton('triton_per_fused_div_mv_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_div_mv_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mv_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 36
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (36*x0)), rmask & xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + (r1), rmask, eviction_policy='evict_last', other=0.0)
tmp2 = tl.load(in_ptr2 + (0))
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp4 = tmp1 / tmp3
tmp5 = tmp0 * tmp4
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp8 = tl.where(rmask & xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tl.store(out_ptr0 + (x0), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ua/cuasdguvgpp5eqyfvllfrdwhflimf3wa3aodnyyxc7mfz7mtpfou.py
# Topologically Sorted Source Nodes: [norm_1, add_2, truediv_1], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div]
# Source node to ATen node mapping:
# add_2 => add_10
# norm_1 => pow_4, pow_5, sum_4
# truediv_1 => div_1
# Graph fragment:
# %pow_4 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_3, 2), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_4, None), kwargs = {})
# %pow_5 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_4, 0.5), kwargs = {})
# %add_10 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_5, 1e-10), kwargs = {})
# %div_1 : [num_users=3] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_3, %add_10), kwargs = {})
triton_per_fused_add_div_linalg_vector_norm_3 = async_compile.triton('triton_per_fused_add_div_linalg_vector_norm_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_div_linalg_vector_norm_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_linalg_vector_norm_3(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)
tmp1 = tmp0 * tmp0
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.sum(tmp2, 1)[:, None]
tmp5 = libdevice.sqrt(tmp4)
tmp6 = 1e-10
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr1 + (tl.broadcast_to(r0, [XBLOCK, RBLOCK])), tmp8, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/42/c42xvpe3gjffzqc4bn7tge2apfou5yfc5nmclihgugpos6r75cug.py
# Topologically Sorted Source Nodes: [sigma], Original ATen: [aten.dot]
# Source node to ATen node mapping:
# sigma => mul_12, sum_6
# Graph fragment:
# %mul_12 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_1, %sum_3), kwargs = {})
# %sum_6 : [num_users=2] = call_function[target=torch.ops.aten.sum.default](args = (%mul_12,), kwargs = {})
triton_per_fused_dot_4 = async_compile.triton('triton_per_fused_dot_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 4],
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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_dot_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_dot_4(in_ptr0, in_ptr1, out_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 + (r0), None)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = tmp0 * tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.sum(tmp3, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp5, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wl/cwl54csp546dqc3ifocpzsnhgl42yo7j7rkk7rys5odmaa42bhvc.py
# Topologically Sorted Source Nodes: [truediv_2], Original ATen: [aten.div]
# Source node to ATen node mapping:
# truediv_2 => div_2
# Graph fragment:
# %div_2 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_4, %expand), kwargs = {})
triton_poi_fused_div_5 = async_compile.triton('triton_poi_fused_div_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_5(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 / tmp2
tl.store(out_ptr0 + (x0), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/xo/cxogtso5gdemfbsy2c2zxjk4b3czinoq5amdagqrqdt7u5uglv3k.py
# Topologically Sorted Source Nodes: [h_2], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# h_2 => convolution
# Graph fragment:
# %convolution : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index, %div_2, %primals_5, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_6 = async_compile.triton('triton_poi_fused_convolution_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=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_6', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_6(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 64) % 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_1/inductor_cache/u7/cu7swroexofbzso6rklbdu4ki5w7iicuimzeufcsufadjx2xwea4.py
# Topologically Sorted Source Nodes: [pow_2, mean_1, add_3, rsqrt_1, mul_1, h_3], Original ATen: [aten.pow, aten.mean, aten.add, aten.rsqrt, aten.mul, aten.relu]
# Source node to ATen node mapping:
# add_3 => add_11
# h_3 => relu
# mean_1 => mean_1
# mul_1 => mul_13
# pow_2 => pow_6
# rsqrt_1 => rsqrt_1
# Graph fragment:
# %pow_6 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%convolution, 2), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%pow_6, [1], True), kwargs = {})
# %add_11 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean_1, 1e-10), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_11,), kwargs = {})
# %mul_13 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, %rsqrt_1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%mul_13,), kwargs = {})
triton_poi_fused_add_mean_mul_pow_relu_rsqrt_7 = async_compile.triton('triton_poi_fused_add_mean_mul_pow_relu_rsqrt_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=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mean_mul_pow_relu_rsqrt_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_pow_relu_rsqrt_7(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 64
x2 = (xindex // 256)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (64 + x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (128 + x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (192 + x0 + (256*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = 1e-10
tmp15 = tmp13 + tmp14
tmp16 = libdevice.rsqrt(tmp15)
tmp17 = tmp0 * tmp16
tmp18 = tl.full([1], 0, tl.int32)
tmp19 = triton_helpers.maximum(tmp18, tmp17)
tl.store(out_ptr0 + (x3), tmp19, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/rb/crblahaeguv4mnz4gsme6klhcd44csiy65iq5kudsf6n7bam4wcg.py
# Topologically Sorted Source Nodes: [mv_6, norm_4, add_6, truediv_6], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div]
# Source node to ATen node mapping:
# add_6 => add_14
# mv_6 => mul_18, sum_13
# norm_4 => pow_11, pow_12, sum_14
# truediv_6 => div_6
# Graph fragment:
# %mul_18 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_2, %primals_10), kwargs = {})
# %sum_13 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_18, [1]), kwargs = {})
# %pow_11 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_13, 2), kwargs = {})
# %sum_14 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_11, None), kwargs = {})
# %pow_12 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_14, 0.5), kwargs = {})
# %add_14 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_12, 1e-10), kwargs = {})
# %div_6 : [num_users=3] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_13, %add_14), kwargs = {})
triton_per_fused_add_div_linalg_vector_norm_mv_8 = async_compile.triton('triton_per_fused_add_div_linalg_vector_norm_mv_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: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_add_div_linalg_vector_norm_mv_8', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_linalg_vector_norm_mv_8(in_ptr0, in_ptr1, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (4 + r0), None)
tmp5 = tl.load(in_ptr1 + (1))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (8 + r0), None)
tmp10 = tl.load(in_ptr1 + (2))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (12 + r0), None)
tmp15 = tl.load(in_ptr1 + (3))
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.sum(tmp20, 1)[:, None]
tmp23 = libdevice.sqrt(tmp22)
tmp24 = 1e-10
tmp25 = tmp23 + tmp24
tmp26 = tmp18 / tmp25
tl.store(out_ptr2 + (tl.broadcast_to(r0, [XBLOCK, RBLOCK])), tmp26, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/r4/cr44fhokthdrthyq4p3ocnqedadbndn4noyptpil5624jgrn3non.py
# Topologically Sorted Source Nodes: [mv_7, norm_5, add_7, truediv_7, sigma_2], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div, aten.dot]
# Source node to ATen node mapping:
# add_7 => add_15
# mv_7 => mul_19, sum_15
# norm_5 => pow_13, pow_14, sum_16
# sigma_2 => mul_21, sum_18
# truediv_7 => div_7
# Graph fragment:
# %mul_19 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_6, %div_6), kwargs = {})
# %sum_15 : [num_users=3] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_19, [1]), kwargs = {})
# %pow_13 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_15, 2), kwargs = {})
# %sum_16 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_13, None), kwargs = {})
# %pow_14 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_16, 0.5), kwargs = {})
# %add_15 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_14, 1e-10), kwargs = {})
# %div_7 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_15, %add_15), kwargs = {})
# %mul_21 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_7, %sum_15), kwargs = {})
# %sum_18 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%mul_21,), kwargs = {})
triton_per_fused_add_div_dot_linalg_vector_norm_mv_9 = async_compile.triton('triton_per_fused_add_div_dot_linalg_vector_norm_mv_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {4: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=(4,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_dot_linalg_vector_norm_mv_9', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 2, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_div_dot_linalg_vector_norm_mv_9(in_ptr0, in_ptr1, out_ptr2, out_ptr3, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (4*r0), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (1))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (2))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + (3))
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.sum(tmp20, 1)[:, None]
tmp23 = libdevice.sqrt(tmp22)
tmp24 = 1e-10
tmp25 = tmp23 + tmp24
tmp26 = tmp18 / tmp25
tmp27 = tmp26 * tmp18
tmp28 = tl.broadcast_to(tmp27, [XBLOCK, RBLOCK])
tmp30 = tl.sum(tmp28, 1)[:, None]
tl.store(out_ptr3 + (tl.broadcast_to(r0, [XBLOCK, RBLOCK])), tmp26, None)
tl.store(out_ptr2 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp30, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ih/cihgregdmczbgo2kp7nqzxc3af2wjini3h7eaqvzcn7bwlzr7rna.py
# Topologically Sorted Source Nodes: [truediv_8], Original ATen: [aten.div]
# Source node to ATen node mapping:
# truediv_8 => div_8
# Graph fragment:
# %div_8 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_12, %expand_2), kwargs = {})
triton_poi_fused_div_10 = async_compile.triton('triton_poi_fused_div_10', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_10', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_10(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 / tmp2
tl.store(out_ptr0 + (x0), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/fp/cfp3i5n73mhezh7n62gowlj7z4b6g3khqt5gqppyjxnpsljoext2.py
# Topologically Sorted Source Nodes: [h_4, x_1, add_8], Original ATen: [aten.convolution, aten.add]
# Source node to ATen node mapping:
# add_8 => add_16
# h_4 => convolution_1
# x_1 => convolution_2
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %div_5, %primals_9, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %convolution_2 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_1, %div_8, %primals_13, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %add_16 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_1, %convolution_2), kwargs = {})
triton_poi_fused_add_convolution_11 = async_compile.triton('triton_poi_fused_add_convolution_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=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_convolution_11', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_11(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 64) % 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)
tmp4 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + 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, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13 = 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, (36, ), (1, ))
assert_size_stride(primals_4, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (36, ), (1, ))
assert_size_stride(primals_8, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, ), (1, ))
assert_size_stride(primals_12, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_13, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 8, 8), (256, 64, 8, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 8, 8), (256, 64, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [pow_1, mean, add, rsqrt, h, h_1, x], Original ATen: [aten.pow, aten.mean, aten.add, aten.rsqrt, aten.mul, aten._unsafe_index]
stream0 = get_raw_stream(0)
triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0.run(primals_1, buf0, buf1, 1024, grid=grid(1024), stream=stream0)
del primals_1
buf2 = empty_strided_cuda((36, ), (1, ), torch.float32)
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3; del buf3 # reuse
buf36 = empty_strided_cuda((36, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mv, norm, add_1, truediv], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div]
triton_per_fused_add_div_linalg_vector_norm_mv_1.run(buf4, primals_4, primals_2, buf2, buf36, 1, 36, grid=grid(1), stream=stream0)
buf5 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [truediv, mv_1], Original ATen: [aten.div, aten.mv]
triton_per_fused_div_mv_2.run(primals_4, buf2, buf4, buf5, 4, 36, grid=grid(4), stream=stream0)
buf7 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [norm_1, add_2, truediv_1], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div]
triton_per_fused_add_div_linalg_vector_norm_3.run(buf5, buf7, 1, 4, grid=grid(1), stream=stream0)
buf8 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [sigma], Original ATen: [aten.dot]
triton_per_fused_dot_4.run(buf7, buf5, buf8, 1, 4, grid=grid(1), stream=stream0)
buf9 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv_2], Original ATen: [aten.div]
triton_poi_fused_div_5.run(primals_4, buf8, buf9, 144, grid=grid(144), stream=stream0)
# Topologically Sorted Source Nodes: [h_2], Original ATen: [aten.convolution]
buf10 = extern_kernels.convolution(buf0, buf9, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 4, 8, 8), (256, 64, 8, 1))
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [h_2], Original ATen: [aten.convolution]
triton_poi_fused_convolution_6.run(buf11, primals_5, 1024, grid=grid(1024), stream=stream0)
del primals_5
buf12 = empty_strided_cuda((4, 4, 8, 8), (256, 64, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [pow_2, mean_1, add_3, rsqrt_1, mul_1, h_3], Original ATen: [aten.pow, aten.mean, aten.add, aten.rsqrt, aten.mul, aten.relu]
triton_poi_fused_add_mean_mul_pow_relu_rsqrt_7.run(buf11, buf12, 1024, grid=grid(1024), stream=stream0)
buf13 = empty_strided_cuda((36, ), (1, ), torch.float32)
buf14 = empty_strided_cuda((), (), torch.float32)
buf15 = buf14; del buf14 # reuse
buf45 = empty_strided_cuda((36, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mv_3, norm_2, add_4, truediv_3], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div]
triton_per_fused_add_div_linalg_vector_norm_mv_1.run(buf15, primals_8, primals_6, buf13, buf45, 1, 36, grid=grid(1), stream=stream0)
buf16 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [truediv_3, mv_4], Original ATen: [aten.div, aten.mv]
triton_per_fused_div_mv_2.run(primals_8, buf13, buf15, buf16, 4, 36, grid=grid(4), stream=stream0)
buf18 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [norm_3, add_5, truediv_4], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div]
triton_per_fused_add_div_linalg_vector_norm_3.run(buf16, buf18, 1, 4, grid=grid(1), stream=stream0)
buf19 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [sigma_1], Original ATen: [aten.dot]
triton_per_fused_dot_4.run(buf18, buf16, buf19, 1, 4, grid=grid(1), stream=stream0)
buf20 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv_5], Original ATen: [aten.div]
triton_poi_fused_div_5.run(primals_8, buf19, buf20, 144, grid=grid(144), stream=stream0)
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.convolution]
buf21 = extern_kernels.convolution(buf12, buf20, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf21, (4, 4, 8, 8), (256, 64, 8, 1))
buf24 = buf16; del buf16 # reuse
# Topologically Sorted Source Nodes: [mv_6, norm_4, add_6, truediv_6], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div]
triton_per_fused_add_div_linalg_vector_norm_mv_8.run(primals_12, primals_10, buf24, 1, 4, grid=grid(1), stream=stream0)
buf27 = empty_strided_cuda((), (), torch.float32)
buf49 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mv_7, norm_5, add_7, truediv_7, sigma_2], Original ATen: [aten.mv, aten.linalg_vector_norm, aten.add, aten.div, aten.dot]
triton_per_fused_add_div_dot_linalg_vector_norm_mv_9.run(primals_12, buf24, buf27, buf49, 1, 4, grid=grid(1), stream=stream0)
buf28 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [truediv_8], Original ATen: [aten.div]
triton_poi_fused_div_10.run(primals_12, buf27, buf28, 16, grid=grid(16), stream=stream0)
del buf27
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.convolution]
buf29 = extern_kernels.convolution(buf1, buf28, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf29, (4, 4, 8, 8), (256, 64, 8, 1))
buf30 = buf21; del buf21 # reuse
# Topologically Sorted Source Nodes: [h_4, x_1, add_8], Original ATen: [aten.convolution, aten.add]
triton_poi_fused_add_convolution_11.run(buf30, primals_9, buf29, primals_13, 1024, grid=grid(1024), stream=stream0)
del buf29
del primals_13
del primals_9
# Topologically Sorted Source Nodes: [], Original ATen: []
buf31 = torch.ops.aten.set_.source_Tensor(primals_2, buf7)
assert_size_stride(buf31, (4, ), (1, ))
del buf2
# Topologically Sorted Source Nodes: [truediv], Original ATen: [aten.div]
buf37 = torch.ops.aten.set_.source_Tensor(primals_3, buf36)
assert_size_stride(buf37, (36, ), (1, ))
del primals_3
# Topologically Sorted Source Nodes: [], Original ATen: []
buf40 = torch.ops.aten.set_.source_Tensor(primals_6, buf18)
assert_size_stride(buf40, (4, ), (1, ))
del buf13
# Topologically Sorted Source Nodes: [truediv_3], Original ATen: [aten.div]
buf46 = torch.ops.aten.set_.source_Tensor(primals_7, buf45)
assert_size_stride(buf46, (36, ), (1, ))
del primals_7
# Topologically Sorted Source Nodes: [norm_5, add_7, truediv_7], Original ATen: [aten.linalg_vector_norm, aten.add, aten.div]
buf50 = torch.ops.aten.set_.source_Tensor(primals_10, buf49)
assert_size_stride(buf50, (4, ), (1, ))
del primals_10
# Topologically Sorted Source Nodes: [], Original ATen: []
buf54 = torch.ops.aten.set_.source_Tensor(primals_11, buf24)
assert_size_stride(buf54, (4, ), (1, ))
del primals_11
return (buf30, buf9, buf20, buf28, primals_2, primals_4, primals_6, primals_8, primals_12, buf0, buf1, buf4, buf7, buf8, buf9, buf11, buf12, buf15, buf18, buf19, buf20, buf24, 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, 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((36, ), (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)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((36, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_13 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import functools
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import Parameter
def l2_norm(v, eps=1e-10):
"""
L2 normalization
:param v:
:param eps:
:return:
"""
return v / (v.norm() + eps)
def pixel_norm(x, eps=1e-10):
"""
Pixel normalization
:param x:
:param eps:
:return:
"""
return x * torch.rsqrt(torch.mean(torch.pow(x, 2), dim=1, keepdim=True) +
eps)
class SpectralNorm(nn.Module):
"""
Spectral Normalization
"""
def __init__(self, module, name='weight', power_iterations=1):
super(SpectralNorm, self).__init__()
self.module = module
self.name = name
self.power_iterations = power_iterations
if not self._made_params():
self._make_params()
def _update_u_v(self):
u = getattr(self.module, self.name + '_u')
v = getattr(self.module, self.name + '_v')
w = getattr(self.module, self.name + '_bar')
height = w.data.shape[0]
for _ in range(self.power_iterations):
v.data = l2_norm(torch.mv(torch.t(w.view(height, -1).data), u.data)
)
u.data = l2_norm(torch.mv(w.view(height, -1).data, v.data))
sigma = u.dot(w.view(height, -1).mv(v))
setattr(self.module, self.name, w / sigma.expand_as(w))
del u, v, w
def _made_params(self):
try:
getattr(self.module, self.name + '_u')
getattr(self.module, self.name + '_v')
getattr(self.module, self.name + '_bar')
return True
except AttributeError:
return False
def _make_params(self):
w = getattr(self.module, self.name)
height = w.data.shape[0]
width = w.view(height, -1).data.shape[1]
u = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
v = Parameter(w.data.new(width).normal_(0, 1), requires_grad=False)
u.data = l2_norm(u.data)
v.data = l2_norm(v.data)
w_bar = Parameter(w.data)
del self.module._parameters[self.name]
self.module.register_parameter(self.name + '_u', u)
self.module.register_parameter(self.name + '_v', v)
self.module.register_parameter(self.name + '_bar', w_bar)
def forward(self, *args):
self._update_u_v()
return self.module.forward(*args)
class GBlock(nn.Module):
def __init__(self, in_channels, out_channels, activation=nn.ReLU(
inplace=False), upsample=functools.partial(F.interpolate,
scale_factor=2)):
super(GBlock, self).__init__()
self.in_channels, self.out_channels = in_channels, out_channels
self.activation = activation
self.upsample = upsample
self.conv1 = SpectralNorm(nn.Conv2d(self.in_channels, self.
out_channels, kernel_size=3, padding=1))
self.conv2 = SpectralNorm(nn.Conv2d(self.out_channels, self.
out_channels, kernel_size=3, padding=1))
self.learnable_sc = in_channels != out_channels or upsample
if self.learnable_sc:
self.conv_sc = SpectralNorm(nn.Conv2d(in_channels, out_channels,
kernel_size=1, padding=0))
self.upsample = upsample
def forward(self, x):
h = pixel_norm(x)
if self.upsample:
h = self.upsample(h)
x = self.upsample(x)
h = self.conv1(h)
h = self.activation(pixel_norm(h))
h = self.conv2(h)
if self.learnable_sc:
x = self.conv_sc(x)
return h + x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import functools
import torch.nn as nn
import torch.nn.functional as F
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
@triton.jit
def triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0(in_ptr0,
out_ptr0, out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 8 % 8
x0 = xindex % 8
x4 = xindex // 64
x3 = xindex // 256
x7 = xindex
tmp0 = x1
tmp1 = tmp0.to(tl.float32)
tmp2 = 0.5
tmp3 = tmp1 * tmp2
tmp4 = tmp3.to(tl.int32)
tmp5 = x0
tmp6 = tmp5.to(tl.float32)
tmp7 = tmp6 * tmp2
tmp8 = tmp7.to(tl.int32)
tmp9 = tl.load(in_ptr0 + (tmp8 + 4 * tmp4 + 16 * x4), xmask,
eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (tmp8 + 4 * tmp4 + 64 * x3), xmask,
eviction_policy='evict_last')
tmp11 = tmp10 * tmp10
tmp12 = tl.load(in_ptr0 + (16 + tmp8 + 4 * tmp4 + 64 * x3), xmask,
eviction_policy='evict_last')
tmp13 = tmp12 * tmp12
tmp14 = tmp11 + tmp13
tmp15 = tl.load(in_ptr0 + (32 + tmp8 + 4 * tmp4 + 64 * x3), xmask,
eviction_policy='evict_last')
tmp16 = tmp15 * tmp15
tmp17 = tmp14 + tmp16
tmp18 = tl.load(in_ptr0 + (48 + tmp8 + 4 * tmp4 + 64 * x3), xmask,
eviction_policy='evict_last')
tmp19 = tmp18 * tmp18
tmp20 = tmp17 + tmp19
tmp21 = 4.0
tmp22 = tmp20 / tmp21
tmp23 = 1e-10
tmp24 = tmp22 + tmp23
tmp25 = libdevice.rsqrt(tmp24)
tmp26 = tmp9 * tmp25
tl.store(out_ptr0 + x7, tmp26, xmask)
tl.store(out_ptr1 + x7, tmp9, xmask)
@triton.jit
def triton_per_fused_add_div_linalg_vector_norm_mv_1(in_out_ptr0, in_ptr0,
in_ptr1, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK: tl.constexpr):
rnumel = 36
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, :]
rmask = rindex < rnumel
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, rmask, other=0.0)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (36 + r0), rmask, other=0.0)
tmp5 = tl.load(in_ptr1 + 1)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (72 + r0), rmask, other=0.0)
tmp10 = tl.load(in_ptr1 + 2)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (108 + r0), rmask, other=0.0)
tmp15 = tl.load(in_ptr1 + 3)
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.where(rmask, tmp20, 0)
tmp23 = tl.sum(tmp22, 1)[:, None]
tmp24 = libdevice.sqrt(tmp23)
tmp25 = 1e-10
tmp26 = tmp24 + tmp25
tmp27 = tmp18 / tmp26
tl.store(out_ptr0 + tl.broadcast_to(r0, [XBLOCK, RBLOCK]), tmp18, rmask)
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp26, None)
tl.store(out_ptr1 + tl.broadcast_to(r0, [XBLOCK, RBLOCK]), tmp27, rmask)
@triton.jit
def triton_per_fused_div_mv_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
rnumel, XBLOCK: tl.constexpr):
xnumel = 4
rnumel = 36
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
rmask = rindex < rnumel
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 36 * x0), rmask & xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + r1, rmask, eviction_policy='evict_last', other=0.0
)
tmp2 = tl.load(in_ptr2 + 0)
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp4 = tmp1 / tmp3
tmp5 = tmp0 * tmp4
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp8 = tl.where(rmask & xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tl.store(out_ptr0 + x0, tmp9, xmask)
@triton.jit
def triton_per_fused_add_div_linalg_vector_norm_3(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)
tmp1 = tmp0 * tmp0
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.sum(tmp2, 1)[:, None]
tmp5 = libdevice.sqrt(tmp4)
tmp6 = 1e-10
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr1 + tl.broadcast_to(r0, [XBLOCK, RBLOCK]), tmp8, None)
@triton.jit
def triton_per_fused_dot_4(in_ptr0, in_ptr1, out_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 + r0, None)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = tmp0 * tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5 = tl.sum(tmp3, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp5, None)
@triton.jit
def triton_poi_fused_div_5(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 / tmp2
tl.store(out_ptr0 + x0, tmp3, xmask)
@triton.jit
def triton_poi_fused_convolution_6(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 64 % 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_mean_mul_pow_relu_rsqrt_7(in_ptr0, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 64
x2 = xindex // 256
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 256 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (64 + x0 + 256 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (128 + x0 + 256 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (192 + x0 + 256 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = 1e-10
tmp15 = tmp13 + tmp14
tmp16 = libdevice.rsqrt(tmp15)
tmp17 = tmp0 * tmp16
tmp18 = tl.full([1], 0, tl.int32)
tmp19 = triton_helpers.maximum(tmp18, tmp17)
tl.store(out_ptr0 + x3, tmp19, xmask)
@triton.jit
def triton_per_fused_add_div_linalg_vector_norm_mv_8(in_ptr0, in_ptr1,
out_ptr2, xnumel, rnumel, XBLOCK: tl.constexpr):
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (4 + r0), None)
tmp5 = tl.load(in_ptr1 + 1)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (8 + r0), None)
tmp10 = tl.load(in_ptr1 + 2)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (12 + r0), None)
tmp15 = tl.load(in_ptr1 + 3)
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.sum(tmp20, 1)[:, None]
tmp23 = libdevice.sqrt(tmp22)
tmp24 = 1e-10
tmp25 = tmp23 + tmp24
tmp26 = tmp18 / tmp25
tl.store(out_ptr2 + tl.broadcast_to(r0, [XBLOCK, RBLOCK]), tmp26, None)
@triton.jit
def triton_per_fused_add_div_dot_linalg_vector_norm_mv_9(in_ptr0, in_ptr1,
out_ptr2, out_ptr3, xnumel, rnumel, XBLOCK: tl.constexpr):
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + 4 * r0, None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp4 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + 1)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp9 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + 2)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp14 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr1 + 3)
tmp16 = tl.broadcast_to(tmp15, [XBLOCK, RBLOCK])
tmp3 = tmp0 * tmp2
tmp7 = tmp4 * tmp6
tmp8 = tmp3 + tmp7
tmp12 = tmp9 * tmp11
tmp13 = tmp8 + tmp12
tmp17 = tmp14 * tmp16
tmp18 = tmp13 + tmp17
tmp19 = tmp18 * tmp18
tmp20 = tl.broadcast_to(tmp19, [XBLOCK, RBLOCK])
tmp22 = tl.sum(tmp20, 1)[:, None]
tmp23 = libdevice.sqrt(tmp22)
tmp24 = 1e-10
tmp25 = tmp23 + tmp24
tmp26 = tmp18 / tmp25
tmp27 = tmp26 * tmp18
tmp28 = tl.broadcast_to(tmp27, [XBLOCK, RBLOCK])
tmp30 = tl.sum(tmp28, 1)[:, None]
tl.store(out_ptr3 + tl.broadcast_to(r0, [XBLOCK, RBLOCK]), tmp26, None)
tl.store(out_ptr2 + tl.full([XBLOCK, 1], 0, tl.int32), tmp30, None)
@triton.jit
def triton_poi_fused_div_10(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 / tmp2
tl.store(out_ptr0 + x0, tmp3, xmask)
@triton.jit
def triton_poi_fused_add_convolution_11(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, xnumel, XBLOCK: tl.constexpr):
xnumel = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 64 % 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)
tmp4 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + 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, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13) = 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, (36,), (1,))
assert_size_stride(primals_4, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (36,), (1,))
assert_size_stride(primals_8, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4,), (1,))
assert_size_stride(primals_12, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_13, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 8, 8), (256, 64, 8, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 8, 8), (256, 64, 8, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__unsafe_index_add_mean_mul_pow_rsqrt_0[grid(1024)](
primals_1, buf0, buf1, 1024, XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
buf2 = empty_strided_cuda((36,), (1,), torch.float32)
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3
del buf3
buf36 = empty_strided_cuda((36,), (1,), torch.float32)
triton_per_fused_add_div_linalg_vector_norm_mv_1[grid(1)](buf4,
primals_4, primals_2, buf2, buf36, 1, 36, XBLOCK=1, num_warps=2,
num_stages=1)
buf5 = empty_strided_cuda((4,), (1,), torch.float32)
triton_per_fused_div_mv_2[grid(4)](primals_4, buf2, buf4, buf5, 4,
36, XBLOCK=1, num_warps=2, num_stages=1)
buf7 = empty_strided_cuda((4,), (1,), torch.float32)
triton_per_fused_add_div_linalg_vector_norm_3[grid(1)](buf5, buf7,
1, 4, XBLOCK=1, num_warps=2, num_stages=1)
buf8 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_dot_4[grid(1)](buf7, buf5, buf8, 1, 4, XBLOCK=1,
num_warps=2, num_stages=1)
buf9 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
triton_poi_fused_div_5[grid(144)](primals_4, buf8, buf9, 144,
XBLOCK=128, num_warps=4, num_stages=1)
buf10 = extern_kernels.convolution(buf0, buf9, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 4, 8, 8), (256, 64, 8, 1))
buf11 = buf10
del buf10
triton_poi_fused_convolution_6[grid(1024)](buf11, primals_5, 1024,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf12 = empty_strided_cuda((4, 4, 8, 8), (256, 64, 8, 1), torch.float32
)
triton_poi_fused_add_mean_mul_pow_relu_rsqrt_7[grid(1024)](buf11,
buf12, 1024, XBLOCK=256, num_warps=4, num_stages=1)
buf13 = empty_strided_cuda((36,), (1,), torch.float32)
buf14 = empty_strided_cuda((), (), torch.float32)
buf15 = buf14
del buf14
buf45 = empty_strided_cuda((36,), (1,), torch.float32)
triton_per_fused_add_div_linalg_vector_norm_mv_1[grid(1)](buf15,
primals_8, primals_6, buf13, buf45, 1, 36, XBLOCK=1, num_warps=
2, num_stages=1)
buf16 = buf5
del buf5
triton_per_fused_div_mv_2[grid(4)](primals_8, buf13, buf15, buf16,
4, 36, XBLOCK=1, num_warps=2, num_stages=1)
buf18 = empty_strided_cuda((4,), (1,), torch.float32)
triton_per_fused_add_div_linalg_vector_norm_3[grid(1)](buf16, buf18,
1, 4, XBLOCK=1, num_warps=2, num_stages=1)
buf19 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_dot_4[grid(1)](buf18, buf16, buf19, 1, 4, XBLOCK=1,
num_warps=2, num_stages=1)
buf20 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
triton_poi_fused_div_5[grid(144)](primals_8, buf19, buf20, 144,
XBLOCK=128, num_warps=4, num_stages=1)
buf21 = extern_kernels.convolution(buf12, buf20, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf21, (4, 4, 8, 8), (256, 64, 8, 1))
buf24 = buf16
del buf16
triton_per_fused_add_div_linalg_vector_norm_mv_8[grid(1)](primals_12,
primals_10, buf24, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
buf27 = empty_strided_cuda((), (), torch.float32)
buf49 = empty_strided_cuda((4,), (1,), torch.float32)
triton_per_fused_add_div_dot_linalg_vector_norm_mv_9[grid(1)](
primals_12, buf24, buf27, buf49, 1, 4, XBLOCK=1, num_warps=2,
num_stages=1)
buf28 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
triton_poi_fused_div_10[grid(16)](primals_12, buf27, buf28, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del buf27
buf29 = extern_kernels.convolution(buf1, buf28, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf29, (4, 4, 8, 8), (256, 64, 8, 1))
buf30 = buf21
del buf21
triton_poi_fused_add_convolution_11[grid(1024)](buf30, primals_9,
buf29, primals_13, 1024, XBLOCK=256, num_warps=4, num_stages=1)
del buf29
del primals_13
del primals_9
buf31 = torch.ops.aten.set_.source_Tensor(primals_2, buf7)
assert_size_stride(buf31, (4,), (1,))
del buf2
buf37 = torch.ops.aten.set_.source_Tensor(primals_3, buf36)
assert_size_stride(buf37, (36,), (1,))
del primals_3
buf40 = torch.ops.aten.set_.source_Tensor(primals_6, buf18)
assert_size_stride(buf40, (4,), (1,))
del buf13
buf46 = torch.ops.aten.set_.source_Tensor(primals_7, buf45)
assert_size_stride(buf46, (36,), (1,))
del primals_7
buf50 = torch.ops.aten.set_.source_Tensor(primals_10, buf49)
assert_size_stride(buf50, (4,), (1,))
del primals_10
buf54 = torch.ops.aten.set_.source_Tensor(primals_11, buf24)
assert_size_stride(buf54, (4,), (1,))
del primals_11
return (buf30, buf9, buf20, buf28, primals_2, primals_4, primals_6,
primals_8, primals_12, buf0, buf1, buf4, buf7, buf8, buf9, buf11,
buf12, buf15, buf18, buf19, buf20, buf24, buf28)
def l2_norm(v, eps=1e-10):
"""
L2 normalization
:param v:
:param eps:
:return:
"""
return v / (v.norm() + eps)
def pixel_norm(x, eps=1e-10):
"""
Pixel normalization
:param x:
:param eps:
:return:
"""
return x * torch.rsqrt(torch.mean(torch.pow(x, 2), dim=1, keepdim=True) +
eps)
class SpectralNorm(nn.Module):
"""
Spectral Normalization
"""
def __init__(self, module, name='weight', power_iterations=1):
super(SpectralNorm, self).__init__()
self.module = module
self.name = name
self.power_iterations = power_iterations
if not self._made_params():
self._make_params()
def _update_u_v(self):
u = getattr(self.module, self.name + '_u')
v = getattr(self.module, self.name + '_v')
w = getattr(self.module, self.name + '_bar')
height = w.data.shape[0]
for _ in range(self.power_iterations):
v.data = l2_norm(torch.mv(torch.t(w.view(height, -1).data), u.data)
)
u.data = l2_norm(torch.mv(w.view(height, -1).data, v.data))
sigma = u.dot(w.view(height, -1).mv(v))
setattr(self.module, self.name, w / sigma.expand_as(w))
del u, v, w
def _made_params(self):
try:
getattr(self.module, self.name + '_u')
getattr(self.module, self.name + '_v')
getattr(self.module, self.name + '_bar')
return True
except AttributeError:
return False
def _make_params(self):
w = getattr(self.module, self.name)
height = w.data.shape[0]
width = w.view(height, -1).data.shape[1]
u = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
v = Parameter(w.data.new(width).normal_(0, 1), requires_grad=False)
u.data = l2_norm(u.data)
v.data = l2_norm(v.data)
w_bar = Parameter(w.data)
del self.module._parameters[self.name]
self.module.register_parameter(self.name + '_u', u)
self.module.register_parameter(self.name + '_v', v)
self.module.register_parameter(self.name + '_bar', w_bar)
def forward(self, *args):
self._update_u_v()
return self.module.forward(*args)
class GBlockNew(nn.Module):
def __init__(self, in_channels, out_channels, activation=nn.ReLU(
inplace=False), upsample=functools.partial(F.interpolate,
scale_factor=2)):
super(GBlockNew, self).__init__()
self.in_channels, self.out_channels = in_channels, out_channels
self.activation = activation
self.upsample = upsample
self.conv1 = SpectralNorm(nn.Conv2d(self.in_channels, self.
out_channels, kernel_size=3, padding=1))
self.conv2 = SpectralNorm(nn.Conv2d(self.out_channels, self.
out_channels, kernel_size=3, padding=1))
self.learnable_sc = in_channels != out_channels or upsample
if self.learnable_sc:
self.conv_sc = SpectralNorm(nn.Conv2d(in_channels, out_channels,
kernel_size=1, padding=0))
self.upsample = upsample
def forward(self, input_0):
primals_2 = self.conv1.module.bias
primals_5 = self.conv1.module.weight_u
primals_3 = self.conv1.module.weight_v
primals_4 = self.conv1.module.weight_bar
primals_6 = self.conv2.module.bias
primals_9 = self.conv2.module.weight_u
primals_7 = self.conv2.module.weight_v
primals_8 = self.conv2.module.weight_bar
primals_10 = self.conv_sc.module.bias
primals_11 = self.conv_sc.module.weight_u
primals_13 = self.conv_sc.module.weight_v
primals_12 = self.conv_sc.module.weight_bar
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13])
return output[0]
| Dong-Jie-Chen/GS-WGAN | GBlock | false | 8,001 | [
"MIT"
] | 32 | 5f33f21249431e53f44167da3ae7587e0dc695d9 | https://github.com/Dong-Jie-Chen/GS-WGAN/tree/5f33f21249431e53f44167da3ae7587e0dc695d9 | import functools
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import Parameter
def l2_norm(v, eps=1e-10):
"""
L2 normalization
:param v:
:param eps:
:return:
"""
return v / (v.norm() + eps)
def pixel_norm(x, eps=1e-10):
"""
Pixel normalization
:param x:
:param eps:
:return:
"""
return x * torch.rsqrt(torch.mean(torch.pow(x, 2), dim=1, keepdim=True) +
eps)
class SpectralNorm(nn.Module):
"""
Spectral Normalization
"""
def __init__(self, module, name='weight', power_iterations=1):
super().__init__()
self.module = module
self.name = name
self.power_iterations = power_iterations
if not self._made_params():
self._make_params()
def _update_u_v(self):
u = getattr(self.module, self.name + '_u')
v = getattr(self.module, self.name + '_v')
w = getattr(self.module, self.name + '_bar')
height = w.data.shape[0]
for _ in range(self.power_iterations):
v.data = l2_norm(torch.mv(torch.t(w.view(height, -1).data), u.data)
)
u.data = l2_norm(torch.mv(w.view(height, -1).data, v.data))
sigma = u.dot(w.view(height, -1).mv(v))
setattr(self.module, self.name, w / sigma.expand_as(w))
del u, v, w
def _made_params(self):
try:
getattr(self.module, self.name + '_u')
getattr(self.module, self.name + '_v')
getattr(self.module, self.name + '_bar')
return True
except AttributeError:
return False
def _make_params(self):
w = getattr(self.module, self.name)
height = w.data.shape[0]
width = w.view(height, -1).data.shape[1]
u = Parameter(w.data.new(height).normal_(0, 1), requires_grad=False)
v = Parameter(w.data.new(width).normal_(0, 1), requires_grad=False)
u.data = l2_norm(u.data)
v.data = l2_norm(v.data)
w_bar = Parameter(w.data)
del self.module._parameters[self.name]
self.module.register_parameter(self.name + '_u', u)
self.module.register_parameter(self.name + '_v', v)
self.module.register_parameter(self.name + '_bar', w_bar)
def forward(self, *args):
self._update_u_v()
return self.module.forward(*args)
class Model(nn.Module):
def __init__(self, in_channels, out_channels, activation=nn.ReLU(
inplace=False), upsample=functools.partial(F.interpolate,
scale_factor=2)):
super().__init__()
self.in_channels, self.out_channels = in_channels, out_channels
self.activation = activation
self.upsample = upsample
self.conv1 = SpectralNorm(nn.Conv2d(self.in_channels, self.
out_channels, kernel_size=3, padding=1))
self.conv2 = SpectralNorm(nn.Conv2d(self.out_channels, self.
out_channels, kernel_size=3, padding=1))
self.learnable_sc = in_channels != out_channels or upsample
if self.learnable_sc:
self.conv_sc = SpectralNorm(nn.Conv2d(in_channels, out_channels,
kernel_size=1, padding=0))
self.upsample = upsample
def forward(self, x):
h = pixel_norm(x)
if self.upsample:
h = self.upsample(h)
x = self.upsample(x)
h = self.conv1(h)
h = self.activation(pixel_norm(h))
h = self.conv2(h)
if self.learnable_sc:
x = self.conv_sc(x)
return h + x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
HardSigmoid | # 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_1/inductor_cache/pv/cpvskaiim5vykh7nqsyym2bx2wtr7uojtgyax7zjvey6cu6xawkr.py
# Topologically Sorted Source Nodes: [ge, type_as, le, type_as_1, mul, add, where, where_1], Original ATen: [aten.ge, aten._to_copy, aten.le, aten.mul, aten.add, aten.where]
# Source node to ATen node mapping:
# add => add
# ge => ge
# le => le
# mul => mul
# type_as => full_default
# type_as_1 => full_default_1
# where => where
# where_1 => where_1
# Graph fragment:
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%arg0_1, 3), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 1.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%arg0_1, -3), kwargs = {})
# %full_default_1 : [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})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 0.1666667), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, 0.5), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%le, %full_default_1, %add), kwargs = {})
# %where_1 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%ge, %full_default, %where), kwargs = {})
triton_poi_fused__to_copy_add_ge_le_mul_where_0 = async_compile.triton('triton_poi_fused__to_copy_add_ge_le_mul_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.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_ge_le_mul_where_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_ge_le_mul_where_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 3.0
tmp2 = tmp0 >= tmp1
tmp3 = -3.0
tmp4 = tmp0 <= tmp3
tmp5 = 0.1666667
tmp6 = tmp0 * tmp5
tmp7 = 0.5
tmp8 = tmp6 + tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = 1.0
tmp12 = tl.where(tmp2, tmp11, tmp10)
tl.store(out_ptr0 + (x0), tmp12, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [ge, type_as, le, type_as_1, mul, add, where, where_1], Original ATen: [aten.ge, aten._to_copy, aten.le, aten.mul, aten.add, aten.where]
stream0 = get_raw_stream(0)
triton_poi_fused__to_copy_add_ge_le_mul_where_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class HardSigmoid(nn.Module):
def __init__(self, slope=0.1666667, offset=0.5):
super().__init__()
self.slope = slope
self.offset = offset
def forward(self, input):
return torch.where(input >= 3, torch.tensor(1.0, dtype=torch.float)
.type_as(input), torch.where(input <= -3, torch.tensor(0.0,
dtype=torch.float).type_as(input), torch.add(torch.mul(input,
self.slope), self.offset)))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
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_add_ge_le_mul_where_0(in_ptr0, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 3.0
tmp2 = tmp0 >= tmp1
tmp3 = -3.0
tmp4 = tmp0 <= tmp3
tmp5 = 0.1666667
tmp6 = tmp0 * tmp5
tmp7 = 0.5
tmp8 = tmp6 + tmp7
tmp9 = 0.0
tmp10 = tl.where(tmp4, tmp9, tmp8)
tmp11 = 1.0
tmp12 = tl.where(tmp2, tmp11, tmp10)
tl.store(out_ptr0 + x0, tmp12, 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__to_copy_add_ge_le_mul_where_0[grid(256)](arg0_1,
buf0, 256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class HardSigmoidNew(nn.Module):
def __init__(self, slope=0.1666667, offset=0.5):
super().__init__()
self.slope = slope
self.offset = offset
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DocYard-ai/UCR | HardSigmoid | false | 8,002 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, slope=0.1666667, offset=0.5):
super().__init__()
self.slope = slope
self.offset = offset
def forward(self, input):
return torch.where(input >= 3, torch.tensor(1.0, dtype=torch.float)
.type_as(input), torch.where(input <= -3, torch.tensor(0.0,
dtype=torch.float).type_as(input), torch.add(torch.mul(input,
self.slope), self.offset)))
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
PixelNorm | # 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_1/inductor_cache/ln/clnknwdlswmj3l56wk4ertm47scosxi4dtpndhkl7cvrs2tmbcj5.py
# Topologically Sorted Source Nodes: [pow_1, mean, add, rsqrt, mul], Original ATen: [aten.pow, aten.mean, aten.add, aten.rsqrt, aten.mul]
# Source node to ATen node mapping:
# add => add
# mean => mean
# mul => mul
# pow_1 => pow_1
# rsqrt => rsqrt
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%arg0_1, 2), kwargs = {})
# %mean : [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, 1e-08), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %rsqrt), kwargs = {})
triton_poi_fused_add_mean_mul_pow_rsqrt_0 = async_compile.triton('triton_poi_fused_add_mean_mul_pow_rsqrt_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mean_mul_pow_rsqrt_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_pow_rsqrt_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = 1e-08
tmp15 = tmp13 + tmp14
tmp16 = libdevice.rsqrt(tmp15)
tmp17 = tmp0 * tmp16
tl.store(out_ptr0 + (x3), tmp17, 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, mean, add, rsqrt, mul], Original ATen: [aten.pow, aten.mean, aten.add, aten.rsqrt, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mean_mul_pow_rsqrt_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class PixelNorm(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input):
return input * torch.rsqrt(torch.mean(input ** 2, dim=1, keepdim=
True) + 1e-08)
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
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_mean_mul_pow_rsqrt_0(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = 1e-08
tmp15 = tmp13 + tmp14
tmp16 = libdevice.rsqrt(tmp15)
tmp17 = tmp0 * tmp16
tl.store(out_ptr0 + x3, tmp17, 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_mean_mul_pow_rsqrt_0[grid(256)](arg0_1, buf0,
256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class PixelNormNew(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Dolorousrtur/style-people | PixelNorm | false | 8,003 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input):
return input * torch.rsqrt(torch.mean(input ** 2, dim=1, keepdim=
True) + 1e-08)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
MAECriterion | # 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_1/inductor_cache/yb/cybq3njxownvbq6jt4cqsgu7yvx2slzbhd5rfjwp453u6giiuqh3.py
# Topologically Sorted Source Nodes: [sub, abs_1, mean], Original ATen: [aten.sub, aten.abs, aten.mean]
# Source node to ATen node mapping:
# abs_1 => abs_1
# mean => mean
# sub => sub
# 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 = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%abs_1,), kwargs = {})
triton_per_fused_abs_mean_sub_0 = async_compile.triton('triton_per_fused_abs_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mean_sub_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mean_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 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = triton_helpers.promote_to_tensor(tl.sum(tmp4, 0))
tmp7 = 256.0
tmp8 = tmp6 / 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: [sub, abs_1, mean], Original ATen: [aten.sub, aten.abs, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_abs_mean_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
class MAECriterion(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, pred, target):
return torch.mean(torch.abs(pred - 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 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_per_fused_abs_mean_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 = tl.broadcast_to(tmp3, [RBLOCK])
tmp6 = triton_helpers.promote_to_tensor(tl.sum(tmp4, 0))
tmp7 = 256.0
tmp8 = tmp6 / 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_abs_mean_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 MAECriterionNew(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| Dolorousrtur/style-people | MAECriterion | false | 8,004 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | import torch
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, pred, target):
return torch.mean(torch.abs(pred - target))
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
StackingNNet | # 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_1/inductor_cache/wk/cwkgtiw4rsyedr3g7k3dha5xhwfl3k5ccryskhhubg4wicitzwjh.py
# Topologically Sorted Source Nodes: [mul, wrapped_sqrt, pow_1, mul_1, add, mul_2, tanh, add_1, mul_3], Original ATen: [aten.mul, aten.sqrt, 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
# wrapped_sqrt => full_default
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, 0.5), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.7978845608028654), kwargs = {dtype: torch.float64, layout: torch.strided, device: cpu, pin_memory: False})
# %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 = (%full_default, %add), 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_sqrt_tanh_0 = async_compile.triton('triton_poi_fused_add_mul_pow_sqrt_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_sqrt_tanh_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_sqrt_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 = tmp8 * tmp7
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 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, wrapped_sqrt, pow_1, mul_1, add, mul_2, tanh, add_1, mul_3], Original ATen: [aten.mul, aten.sqrt, aten.pow, aten.add, aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_pow_sqrt_tanh_0.run(buf0, buf1, 256, grid=grid(256), stream=stream0)
return (buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), 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, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import numpy as np
import torch.nn as nn
from torch.nn import CrossEntropyLoss
import torch.utils.data.distributed
import torch.utils.data
def gelu(x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2.0 / np.pi) * (x + 0.044715 *
torch.pow(x, 3))))
class StackingNNet(nn.Module):
def __init__(self, input_size, output_size):
super(StackingNNet, self).__init__()
self.fc1 = nn.Linear(input_size, output_size)
def forward(self, x, labels=None):
x = self.fc1(x)
outputs = gelu(x),
if labels is not None:
loss_fct = CrossEntropyLoss()
loss = loss_fct(outputs[0], labels)
outputs = (loss,) + outputs
return outputs
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4, 'output_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import numpy as np
import torch.nn as nn
import torch.utils.data.distributed
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_add_mul_pow_sqrt_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 = tmp8 * tmp7
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 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_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_sqrt_tanh_0[grid(256)](buf0, buf1, 256,
XBLOCK=128, num_warps=4, num_stages=1)
return buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf0
def gelu(x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2.0 / np.pi) * (x + 0.044715 *
torch.pow(x, 3))))
class StackingNNetNew(nn.Module):
def __init__(self, input_size, output_size):
super(StackingNNetNew, self).__init__()
self.fc1 = nn.Linear(input_size, output_size)
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]
| ECNU-ICA/ECNU-SenseMaker | StackingNNet | false | 8,005 | [
"MIT"
] | 16 | 24f829c3dfefccea5fecbbe75904858ec1fefffb | https://github.com/ECNU-ICA/ECNU-SenseMaker/tree/24f829c3dfefccea5fecbbe75904858ec1fefffb | import torch
import numpy as np
import torch.nn as nn
from torch.nn import CrossEntropyLoss
import torch.utils.data.distributed
import torch.utils.data
def gelu(x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2.0 / np.pi) * (x + 0.044715 *
torch.pow(x, 3))))
class Model(nn.Module):
def __init__(self, input_size, output_size):
super().__init__()
self.fc1 = nn.Linear(input_size, output_size)
def forward(self, x, labels=None):
x = self.fc1(x)
outputs = gelu(x),
if labels is not None:
loss_fct = CrossEntropyLoss()
loss = loss_fct(outputs[0], labels)
outputs = (loss,) + outputs
return outputs
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
CosineSimilarityLoss | # 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_1/inductor_cache/n4/cn4zrd3wpvd7mzjgupqv5joml2alazaddigv5r6cxcueasaj6rlv.py
# Topologically Sorted Source Nodes: [cosine_similarity], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
# Source node to ATen node mapping:
# cosine_similarity => clamp_min, clamp_min_1, div, div_1, mul, pow_1, pow_2, pow_3, pow_4, sum_1, sum_2
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%arg1_1, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [1], True), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_1, 0.5), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_2, 1e-08), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg1_1, %clamp_min), kwargs = {})
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%arg0_1, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_3, [1], True), kwargs = {})
# %pow_4 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_2, 0.5), kwargs = {})
# %clamp_min_1 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_4, 1e-08), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%arg0_1, %clamp_min_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_1, %div), kwargs = {})
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_0 = async_compile.triton('triton_poi_fused_clamp_min_div_linalg_vector_norm_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 10, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_div_linalg_vector_norm_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
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (x3), xmask)
tmp17 = tl.load(in_ptr1 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-08
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tmp18 = tmp17 * tmp17
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = libdevice.sqrt(tmp27)
tmp29 = triton_helpers.maximum(tmp28, tmp13)
tmp30 = tmp16 / tmp29
tmp31 = tmp15 * tmp30
tl.store(out_ptr0 + (x3), tmp31, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/tg/ctgevin3weduxtpagbpxycetjjx3kvxjub3can24wb6du3qma2ey.py
# Topologically Sorted Source Nodes: [cosine_similarity, mul, sub], Original ATen: [aten.sum, aten.mul, aten.rsub]
# Source node to ATen node mapping:
# cosine_similarity => sum_3
# mul => mul_1
# sub => sub
# Graph fragment:
# %sum_3 : [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 = (%sum_3, 0.5), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.5, %mul_1), kwargs = {})
triton_poi_fused_mul_rsub_sum_1 = async_compile.triton('triton_poi_fused_mul_rsub_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_rsub_sum_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_rsub_sum_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 % 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 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.5
tmp8 = tmp6 * tmp7
tmp9 = tmp7 - tmp8
tl.store(out_ptr0 + (x2), tmp9, 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: [cosine_similarity], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_0.run(arg1_1, arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cosine_similarity, mul, sub], Original ATen: [aten.sum, aten.mul, aten.rsub]
triton_poi_fused_mul_rsub_sum_1.run(buf0, buf1, 64, grid=grid(64), 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)
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 CosineSimilarityLoss(nn.Module):
def __init__(self):
super(CosineSimilarityLoss, self).__init__()
def forward(self, x1, x2):
return 0.5 - 0.5 * torch.cosine_similarity(x1, x2)
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
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_clamp_min_div_linalg_vector_norm_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
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + x3, xmask)
tmp17 = tl.load(in_ptr1 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr1 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr1 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tmp1 * tmp1
tmp4 = tmp3 * tmp3
tmp5 = tmp2 + tmp4
tmp7 = tmp6 * tmp6
tmp8 = tmp5 + tmp7
tmp10 = tmp9 * tmp9
tmp11 = tmp8 + tmp10
tmp12 = libdevice.sqrt(tmp11)
tmp13 = 1e-08
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tmp18 = tmp17 * tmp17
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = libdevice.sqrt(tmp27)
tmp29 = triton_helpers.maximum(tmp28, tmp13)
tmp30 = tmp16 / tmp29
tmp31 = tmp15 * tmp30
tl.store(out_ptr0 + x3, tmp31, xmask)
@triton.jit
def triton_poi_fused_mul_rsub_sum_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 % 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 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.5
tmp8 = tmp6 * tmp7
tmp9 = tmp7 - tmp8
tl.store(out_ptr0 + x2, tmp9, 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_clamp_min_div_linalg_vector_norm_mul_0[grid(256)](
arg1_1, arg0_1, buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_mul_rsub_sum_1[grid(64)](buf0, buf1, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf0
return buf1,
class CosineSimilarityLossNew(nn.Module):
def __init__(self):
super(CosineSimilarityLossNew, 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]
| DunZhang/KnowledgeDistillation | CosineSimilarityLoss | false | 8,006 | [
"MIT"
] | 31 | 47a9dd0f51021001b53e3a76c9347eb3131f1f72 | https://github.com/DunZhang/KnowledgeDistillation/tree/47a9dd0f51021001b53e3a76c9347eb3131f1f72 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x1, x2):
return 0.5 - 0.5 * torch.cosine_similarity(x1, x2)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SEModule | # 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_1/inductor_cache/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.py
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# outputs => 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zc/czchmrm2eextseghnduyy73m4ncqtihxgmkgkbsyoxi4kty2ql7p.py
# Topologically Sorted Source Nodes: [outputs_1, outputs_2], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# outputs_1 => convolution
# outputs_2 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%mean, %primals_2, %primals_3, [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_1 = async_compile.triton('triton_poi_fused_convolution_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tl.store(in_out_ptr0 + (x0), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zf/czfilubyh45axw63qmrwbffauwo43vtl4fcotx6sqfpmcu2uwboi.py
# Topologically Sorted Source Nodes: [outputs_3, outputs_4, mul], Original ATen: [aten.convolution, aten.hardsigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# outputs_3 => convolution_1
# outputs_4 => add, clamp_max, clamp_min, div
# Graph fragment:
# %convolution_1 : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_1, 3), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%add, 0), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 6), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%clamp_max, 6), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %div), kwargs = {})
triton_poi_fused_convolution_hardsigmoid_mul_2 = async_compile.triton('triton_poi_fused_convolution_hardsigmoid_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_hardsigmoid_mul_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_hardsigmoid_mul_2(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
x3 = xindex
x4 = (xindex // 16)
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x4), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = 3.0
tmp5 = tmp3 + tmp4
tmp6 = 0.0
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = 6.0
tmp9 = triton_helpers.minimum(tmp7, tmp8)
tmp10 = 0.16666666666666666
tmp11 = tmp9 * tmp10
tmp12 = tmp0 * tmp11
tl.store(out_ptr0 + (x3), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mj/cmjkau4fysisskqjsgv7xcndprjkcqzfuhopf4hpcex3axyysyov.py
# Topologically Sorted Source Nodes: [outputs_3], Original ATen: [aten.convolution, aten.hardsigmoid_backward]
# Source node to ATen node mapping:
# outputs_3 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=3] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%convolution_1, -3.0), kwargs = {})
# %lt : [num_users=1] = call_function[target=torch.ops.aten.lt.Scalar](args = (%convolution_1, 3.0), kwargs = {})
# %bitwise_and : [num_users=1] = call_function[target=torch.ops.aten.bitwise_and.Tensor](args = (%gt, %lt), kwargs = {})
triton_poi_fused_convolution_hardsigmoid_backward_3 = async_compile.triton('triton_poi_fused_convolution_hardsigmoid_backward_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_hardsigmoid_backward_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_hardsigmoid_backward_3(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 = -3.0
tmp4 = tmp2 > tmp3
tmp5 = 3.0
tmp6 = tmp2 < tmp5
tmp7 = tmp4 & 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, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (1, ), (1, ))
assert_size_stride(primals_4, (4, 1, 1, 1), (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, 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: [outputs], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 16, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [outputs_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 1, 1), (1, 1, 1, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [outputs_1, outputs_2], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_1.run(buf3, primals_3, 4, grid=grid(4), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [outputs_3], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf3, 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, 4, 1, 1), (4, 1, 1, 1))
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [outputs_3, outputs_4, mul], Original ATen: [aten.convolution, aten.hardsigmoid, aten.mul]
triton_poi_fused_convolution_hardsigmoid_mul_2.run(primals_1, buf4, primals_5, buf5, 256, grid=grid(256), stream=stream0)
buf6 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [outputs_3], Original ATen: [aten.convolution, aten.hardsigmoid_backward]
triton_poi_fused_convolution_hardsigmoid_backward_3.run(buf4, primals_5, buf6, 16, grid=grid(16), stream=stream0)
del buf4
del primals_5
return (buf5, primals_1, primals_2, primals_4, buf1, buf3, 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((1, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 1, 1, 1), (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 torch import nn
class SEModule(nn.Module):
def __init__(self, in_channels, reduction=4):
super(SEModule, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.conv1 = nn.Conv2d(in_channels=in_channels, out_channels=
in_channels // reduction, kernel_size=1, stride=1, padding=0)
self.conv2 = nn.Conv2d(in_channels=in_channels // reduction,
out_channels=in_channels, kernel_size=1, stride=1, padding=0)
self.relu = nn.ReLU(inplace=True)
self.hardsigmoid = nn.Hardsigmoid(inplace=True)
def forward(self, inputs):
outputs = self.avg_pool(inputs)
outputs = self.conv1(outputs)
outputs = self.relu(outputs)
outputs = self.conv2(outputs)
outputs = self.hardsigmoid(outputs)
return inputs * outputs
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tl.store(in_out_ptr0 + x0, tmp5, xmask)
@triton.jit
def triton_poi_fused_convolution_hardsigmoid_mul_2(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
x3 = xindex
x4 = xindex // 16
x1 = xindex // 16 % 4
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x4, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = 3.0
tmp5 = tmp3 + tmp4
tmp6 = 0.0
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = 6.0
tmp9 = triton_helpers.minimum(tmp7, tmp8)
tmp10 = 0.16666666666666666
tmp11 = tmp9 * tmp10
tmp12 = tmp0 * tmp11
tl.store(out_ptr0 + x3, tmp12, xmask)
@triton.jit
def triton_poi_fused_convolution_hardsigmoid_backward_3(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 = -3.0
tmp4 = tmp2 > tmp3
tmp5 = 3.0
tmp6 = tmp2 < tmp5
tmp7 = tmp4 & tmp6
tl.store(out_ptr0 + x2, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (1,), (1,))
assert_size_stride(primals_4, (4, 1, 1, 1), (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, 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=8,
num_warps=2, num_stages=1)
buf2 = extern_kernels.convolution(buf1, primals_2, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 1, 1, 1), (1, 1, 1, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_relu_1[grid(4)](buf3, primals_3, 4,
XBLOCK=4, num_warps=1, num_stages=1)
del primals_3
buf4 = extern_kernels.convolution(buf3, 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, 4, 1, 1), (4, 1, 1, 1))
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_convolution_hardsigmoid_mul_2[grid(256)](primals_1,
buf4, primals_5, buf5, 256, XBLOCK=128, num_warps=4, num_stages=1)
buf6 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.bool)
triton_poi_fused_convolution_hardsigmoid_backward_3[grid(16)](buf4,
primals_5, buf6, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf4
del primals_5
return buf5, primals_1, primals_2, primals_4, buf1, buf3, buf6
class SEModuleNew(nn.Module):
def __init__(self, in_channels, reduction=4):
super(SEModuleNew, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.conv1 = nn.Conv2d(in_channels=in_channels, out_channels=
in_channels // reduction, kernel_size=1, stride=1, padding=0)
self.conv2 = nn.Conv2d(in_channels=in_channels // reduction,
out_channels=in_channels, kernel_size=1, stride=1, padding=0)
self.relu = nn.ReLU(inplace=True)
self.hardsigmoid = nn.Hardsigmoid(inplace=True)
def forward(self, input_0):
primals_2 = self.conv1.weight
primals_3 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DocYard-ai/UCR | SEModule | false | 8,007 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, in_channels, reduction=4):
super().__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.conv1 = nn.Conv2d(in_channels=in_channels, out_channels=
in_channels // reduction, kernel_size=1, stride=1, padding=0)
self.conv2 = nn.Conv2d(in_channels=in_channels // reduction,
out_channels=in_channels, kernel_size=1, stride=1, padding=0)
self.relu = nn.ReLU(inplace=True)
self.hardsigmoid = nn.Hardsigmoid(inplace=True)
def forward(self, inputs):
outputs = self.avg_pool(inputs)
outputs = self.conv1(outputs)
outputs = self.relu(outputs)
outputs = self.conv2(outputs)
outputs = self.hardsigmoid(outputs)
return inputs * outputs
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
ModulatedConv2d | # 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_1/inductor_cache/ag/caggqh3nrqirp3azykpm6daivhojwggbhob6pnubsheittrjeiky.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, 0.5), 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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qw/cqw2ahskucrmwkbadqswj76db4kfcxn5w76kji7peia7kwu4zbkq.py
# Topologically Sorted Source Nodes: [mul_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_1 => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, 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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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')
# kernel path: runs/run_shard_1/inductor_cache/sn/csn7adwcrtimy326lupk2anpklflbjwcebi4icerax26g54j67hv.py
# Topologically Sorted Source Nodes: [mul_2, weight, pow_1, sum_1, add, demod, weight_1], Original ATen: [aten.mul, aten.pow, aten.sum, aten.add, aten.rsqrt]
# Source node to ATen node mapping:
# add => add
# demod => rsqrt
# mul_2 => mul_2
# pow_1 => pow_1
# sum_1 => sum_1
# weight => mul_3
# weight_1 => mul_4
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_5, 0.125), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %view), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%mul_3, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [2, 3, 4]), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_1, 1e-08), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %view_1), kwargs = {})
triton_per_fused_add_mul_pow_rsqrt_sum_2 = async_compile.triton('triton_per_fused_add_mul_pow_rsqrt_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[16, 64],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_mul_pow_rsqrt_sum_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mul_pow_rsqrt_sum_2(in_out_ptr0, in_ptr0, in_ptr1, out_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)
r5 = rindex
x0 = xindex % 4
r3 = (rindex // 16)
x1 = (xindex // 4)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (r5 + (64*x0)), xmask, eviction_policy='evict_last', other=0.0)
tmp3 = tl.load(in_ptr1 + (r3 + (4*x1)), xmask, eviction_policy='evict_last', other=0.0)
tmp1 = 0.125
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp5 = tmp4 * tmp4
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp8 = tl.where(xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tmp10 = 1e-08
tmp11 = tmp9 + tmp10
tmp12 = libdevice.rsqrt(tmp11)
tmp13 = tmp4 * tmp12
tl.debug_barrier()
tl.store(in_out_ptr0 + (x4), tmp12, xmask)
tl.store(out_ptr0 + (r5 + (64*x4)), tmp13, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (1, 4, 4, 4, 4), (256, 64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_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_1], 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_1, out], Original ATen: [aten.mul, aten.addmm]
extern_kernels.addmm(buf1, primals_4, reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
buf3 = buf0; del buf0 # reuse
buf4 = buf3; del buf3 # reuse
buf5 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_2, weight, pow_1, sum_1, add, demod, weight_1], Original ATen: [aten.mul, aten.pow, aten.sum, aten.add, aten.rsqrt]
triton_per_fused_add_mul_pow_rsqrt_sum_2.run(buf4, primals_5, buf2, buf5, 16, 64, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.convolution]
buf6 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf5, (16, 4, 4, 4), (64, 16, 4, 1), 0), stride=(1, 1), padding=(2, 2), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=4, bias=None)
assert_size_stride(buf6, (1, 16, 5, 5), (400, 25, 5, 1))
return (reinterpret_tensor(buf6, (4, 4, 5, 5), (100, 25, 5, 1), 0), primals_4, primals_5, buf2, buf4, reinterpret_tensor(buf5, (16, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((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((1, 4, 4, 4, 4), (256, 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)
| from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
def make_kernel(k):
k = torch.tensor(k, dtype=torch.float32)
if k.ndim == 1:
k = k[None, :] * k[:, None]
k /= k.sum()
return k
def upfirdn2d(input, kernel, up=1, down=1, pad=(0, 0)):
out = UpFirDn2d.apply(input, kernel, (up, up), (down, down), (pad[0],
pad[1], pad[0], pad[1]))
return out
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class UpFirDn2dBackward(Function):
@staticmethod
def forward(ctx, grad_output, kernel, grad_kernel, up, down, pad, g_pad,
in_size, out_size):
up_x, up_y = up
down_x, down_y = down
g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1 = g_pad
grad_output = grad_output.reshape(-1, out_size[0], out_size[1], 1)
grad_input = upfirdn2d_op.upfirdn2d(grad_output, grad_kernel,
down_x, down_y, up_x, up_y, g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1)
grad_input = grad_input.view(in_size[0], in_size[1], in_size[2],
in_size[3])
ctx.save_for_backward(kernel)
pad_x0, pad_x1, pad_y0, pad_y1 = pad
ctx.up_x = up_x
ctx.up_y = up_y
ctx.down_x = down_x
ctx.down_y = down_y
ctx.pad_x0 = pad_x0
ctx.pad_x1 = pad_x1
ctx.pad_y0 = pad_y0
ctx.pad_y1 = pad_y1
ctx.in_size = in_size
ctx.out_size = out_size
return grad_input
@staticmethod
def backward(ctx, gradgrad_input):
kernel, = ctx.saved_tensors
gradgrad_input = gradgrad_input.reshape(-1, ctx.in_size[2], ctx.
in_size[3], 1)
gradgrad_out = upfirdn2d_op.upfirdn2d(gradgrad_input, kernel, ctx.
up_x, ctx.up_y, ctx.down_x, ctx.down_y, ctx.pad_x0, ctx.pad_x1,
ctx.pad_y0, ctx.pad_y1)
gradgrad_out = gradgrad_out.view(ctx.in_size[0], ctx.in_size[1],
ctx.out_size[0], ctx.out_size[1])
return gradgrad_out, None, None, None, None, None, None, None, None
class UpFirDn2d(Function):
@staticmethod
def forward(ctx, input, kernel, up, down, pad):
if input.dtype != kernel.dtype:
input = input.type(kernel.dtype)
up_x, up_y = up
down_x, down_y = down
pad_x0, pad_x1, pad_y0, pad_y1 = pad
kernel_h, kernel_w = kernel.shape
_batch, channel, in_h, in_w = input.shape
ctx.in_size = input.shape
input = input.reshape(-1, in_h, in_w, 1)
ctx.save_for_backward(kernel, torch.flip(kernel, [0, 1]))
out_h = (in_h * up_y + pad_y0 + pad_y1 - kernel_h) // down_y + 1
out_w = (in_w * up_x + pad_x0 + pad_x1 - kernel_w) // down_x + 1
ctx.out_size = out_h, out_w
ctx.up = up_x, up_y
ctx.down = down_x, down_y
ctx.pad = pad_x0, pad_x1, pad_y0, pad_y1
g_pad_x0 = kernel_w - pad_x0 - 1
g_pad_y0 = kernel_h - pad_y0 - 1
g_pad_x1 = in_w * up_x - out_w * down_x + pad_x0 - up_x + 1
g_pad_y1 = in_h * up_y - out_h * down_y + pad_y0 - up_y + 1
ctx.g_pad = g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1
out = upfirdn2d_op.upfirdn2d(input, kernel, up_x, up_y, down_x,
down_y, pad_x0, pad_x1, pad_y0, pad_y1)
out = out.view(-1, channel, out_h, out_w)
return out
@staticmethod
def backward(ctx, grad_output):
kernel, grad_kernel = ctx.saved_tensors
grad_input = UpFirDn2dBackward.apply(grad_output, kernel,
grad_kernel, ctx.up, ctx.down, ctx.pad, ctx.g_pad, ctx.in_size,
ctx.out_size)
return grad_input, None, None, None, None
class Blur(nn.Module):
def __init__(self, kernel, pad, upsample_factor=1):
super().__init__()
kernel = make_kernel(kernel)
if upsample_factor > 1:
kernel = kernel * upsample_factor ** 2
self.register_buffer('kernel', kernel)
self.pad = pad
def forward(self, input):
out = upfirdn2d(input, self.kernel, pad=self.pad)
return out
class ModulatedConv2d(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, style_dim,
demodulate=True, upsample=False, downsample=False, blur_kernel=[1,
3, 3, 1]):
super().__init__()
self.eps = 1e-08
self.kernel_size = kernel_size
self.in_channel = in_channel
self.out_channel = out_channel
self.upsample = upsample
self.downsample = downsample
if upsample:
factor = 2
p = len(blur_kernel) - factor - (kernel_size - 1)
pad0 = (p + 1) // 2 + factor - 1
pad1 = p // 2 + 1
self.blur = Blur(blur_kernel, pad=(pad0, pad1), upsample_factor
=factor)
if downsample:
factor = 2
p = len(blur_kernel) - factor + (kernel_size - 1)
pad0 = (p + 1) // 2
pad1 = p // 2
self.blur = Blur(blur_kernel, pad=(pad0, pad1))
fan_in = in_channel * kernel_size ** 2
self.scale = 1 / math.sqrt(fan_in)
self.padding = kernel_size // 2
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
kernel_size, kernel_size))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel}, {self.kernel_size}, upsample={self.upsample}, downsample={self.downsample})'
)
def forward(self, input, style):
batch, in_channel, height, width = input.shape
style = self.modulation(style)
style = style.view(batch, 1, in_channel, 1, 1)
weight = self.scale * self.weight * style
if self.demodulate:
demod = torch.rsqrt(weight.pow(2).sum([2, 3, 4]) + 1e-08)
weight = weight * demod.view(batch, self.out_channel, 1, 1, 1)
weight = weight.view(batch * self.out_channel, in_channel, self.
kernel_size, self.kernel_size)
if self.upsample:
input = input.view(1, batch * in_channel, height, width)
weight = weight.view(batch, self.out_channel, in_channel, self.
kernel_size, self.kernel_size)
weight = weight.transpose(1, 2).reshape(batch * in_channel,
self.out_channel, self.kernel_size, self.kernel_size)
out = F.conv_transpose2d(input, weight, padding=0, stride=2,
groups=batch)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
out = self.blur(out)
elif self.downsample:
input = self.blur(input)
_, _, height, width = input.shape
input = input.view(1, batch * in_channel, height, width)
out = F.conv2d(input, weight, padding=0, stride=2, groups=batch)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
else:
input = input.view(1, batch * in_channel, height, width)
out = F.conv2d(input, weight, padding=self.padding, groups=batch)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'out_channel': 4, 'kernel_size': 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
from torch.autograd import Function
import 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_mul_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
tl.store(out_ptr0 + x0, tmp2, 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)
@triton.jit
def triton_per_fused_add_mul_pow_rsqrt_sum_2(in_out_ptr0, in_ptr0, in_ptr1,
out_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)
r5 = rindex
x0 = xindex % 4
r3 = rindex // 16
x1 = xindex // 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (r5 + 64 * x0), xmask, eviction_policy=
'evict_last', other=0.0)
tmp3 = tl.load(in_ptr1 + (r3 + 4 * x1), xmask, eviction_policy=
'evict_last', other=0.0)
tmp1 = 0.125
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp5 = tmp4 * tmp4
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, RBLOCK])
tmp8 = tl.where(xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tmp10 = 1e-08
tmp11 = tmp9 + tmp10
tmp12 = libdevice.rsqrt(tmp11)
tmp13 = tmp4 * tmp12
tl.debug_barrier()
tl.store(in_out_ptr0 + x4, tmp12, xmask)
tl.store(out_ptr0 + (r5 + 64 * x4), tmp13, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (1, 4, 4, 4, 4), (256, 64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_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_4, reinterpret_tensor(buf0, (4,
4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
buf3 = buf0
del buf0
buf4 = buf3
del buf3
buf5 = empty_strided_cuda((4, 4, 4, 4, 4), (256, 64, 16, 4, 1),
torch.float32)
triton_per_fused_add_mul_pow_rsqrt_sum_2[grid(16)](buf4, primals_5,
buf2, buf5, 16, 64, XBLOCK=8, num_warps=4, num_stages=1)
buf6 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1,
16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf5, (16, 4,
4, 4), (64, 16, 4, 1), 0), stride=(1, 1), padding=(2, 2),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=4, bias=None)
assert_size_stride(buf6, (1, 16, 5, 5), (400, 25, 5, 1))
return reinterpret_tensor(buf6, (4, 4, 5, 5), (100, 25, 5, 1), 0
), primals_4, primals_5, buf2, buf4, reinterpret_tensor(buf5, (16,
4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_1, (1, 16,
4, 4), (256, 16, 4, 1), 0)
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
def make_kernel(k):
k = torch.tensor(k, dtype=torch.float32)
if k.ndim == 1:
k = k[None, :] * k[:, None]
k /= k.sum()
return k
def upfirdn2d(input, kernel, up=1, down=1, pad=(0, 0)):
out = UpFirDn2d.apply(input, kernel, (up, up), (down, down), (pad[0],
pad[1], pad[0], pad[1]))
return out
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class UpFirDn2dBackward(Function):
@staticmethod
def forward(ctx, grad_output, kernel, grad_kernel, up, down, pad, g_pad,
in_size, out_size):
up_x, up_y = up
down_x, down_y = down
g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1 = g_pad
grad_output = grad_output.reshape(-1, out_size[0], out_size[1], 1)
grad_input = upfirdn2d_op.upfirdn2d(grad_output, grad_kernel,
down_x, down_y, up_x, up_y, g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1)
grad_input = grad_input.view(in_size[0], in_size[1], in_size[2],
in_size[3])
ctx.save_for_backward(kernel)
pad_x0, pad_x1, pad_y0, pad_y1 = pad
ctx.up_x = up_x
ctx.up_y = up_y
ctx.down_x = down_x
ctx.down_y = down_y
ctx.pad_x0 = pad_x0
ctx.pad_x1 = pad_x1
ctx.pad_y0 = pad_y0
ctx.pad_y1 = pad_y1
ctx.in_size = in_size
ctx.out_size = out_size
return grad_input
@staticmethod
def backward(ctx, gradgrad_input):
kernel, = ctx.saved_tensors
gradgrad_input = gradgrad_input.reshape(-1, ctx.in_size[2], ctx.
in_size[3], 1)
gradgrad_out = upfirdn2d_op.upfirdn2d(gradgrad_input, kernel, ctx.
up_x, ctx.up_y, ctx.down_x, ctx.down_y, ctx.pad_x0, ctx.pad_x1,
ctx.pad_y0, ctx.pad_y1)
gradgrad_out = gradgrad_out.view(ctx.in_size[0], ctx.in_size[1],
ctx.out_size[0], ctx.out_size[1])
return gradgrad_out, None, None, None, None, None, None, None, None
class UpFirDn2d(Function):
@staticmethod
def forward(ctx, input, kernel, up, down, pad):
if input.dtype != kernel.dtype:
input = input.type(kernel.dtype)
up_x, up_y = up
down_x, down_y = down
pad_x0, pad_x1, pad_y0, pad_y1 = pad
kernel_h, kernel_w = kernel.shape
_batch, channel, in_h, in_w = input.shape
ctx.in_size = input.shape
input = input.reshape(-1, in_h, in_w, 1)
ctx.save_for_backward(kernel, torch.flip(kernel, [0, 1]))
out_h = (in_h * up_y + pad_y0 + pad_y1 - kernel_h) // down_y + 1
out_w = (in_w * up_x + pad_x0 + pad_x1 - kernel_w) // down_x + 1
ctx.out_size = out_h, out_w
ctx.up = up_x, up_y
ctx.down = down_x, down_y
ctx.pad = pad_x0, pad_x1, pad_y0, pad_y1
g_pad_x0 = kernel_w - pad_x0 - 1
g_pad_y0 = kernel_h - pad_y0 - 1
g_pad_x1 = in_w * up_x - out_w * down_x + pad_x0 - up_x + 1
g_pad_y1 = in_h * up_y - out_h * down_y + pad_y0 - up_y + 1
ctx.g_pad = g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1
out = upfirdn2d_op.upfirdn2d(input, kernel, up_x, up_y, down_x,
down_y, pad_x0, pad_x1, pad_y0, pad_y1)
out = out.view(-1, channel, out_h, out_w)
return out
@staticmethod
def backward(ctx, grad_output):
kernel, grad_kernel = ctx.saved_tensors
grad_input = UpFirDn2dBackward.apply(grad_output, kernel,
grad_kernel, ctx.up, ctx.down, ctx.pad, ctx.g_pad, ctx.in_size,
ctx.out_size)
return grad_input, None, None, None, None
class Blur(nn.Module):
def __init__(self, kernel, pad, upsample_factor=1):
super().__init__()
kernel = make_kernel(kernel)
if upsample_factor > 1:
kernel = kernel * upsample_factor ** 2
self.register_buffer('kernel', kernel)
self.pad = pad
def forward(self, input):
out = upfirdn2d(input, self.kernel, pad=self.pad)
return out
class ModulatedConv2dNew(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, style_dim,
demodulate=True, upsample=False, downsample=False, blur_kernel=[1,
3, 3, 1]):
super().__init__()
self.eps = 1e-08
self.kernel_size = kernel_size
self.in_channel = in_channel
self.out_channel = out_channel
self.upsample = upsample
self.downsample = downsample
if upsample:
factor = 2
p = len(blur_kernel) - factor - (kernel_size - 1)
pad0 = (p + 1) // 2 + factor - 1
pad1 = p // 2 + 1
self.blur = Blur(blur_kernel, pad=(pad0, pad1), upsample_factor
=factor)
if downsample:
factor = 2
p = len(blur_kernel) - factor + (kernel_size - 1)
pad0 = (p + 1) // 2
pad1 = p // 2
self.blur = Blur(blur_kernel, pad=(pad0, pad1))
fan_in = in_channel * kernel_size ** 2
self.scale = 1 / math.sqrt(fan_in)
self.padding = kernel_size // 2
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
kernel_size, kernel_size))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel}, {self.kernel_size}, upsample={self.upsample}, downsample={self.downsample})'
)
def forward(self, input_0, input_1):
primals_5 = self.weight
primals_2 = self.modulation.weight
primals_3 = self.modulation.bias
primals_1 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| Dolorousrtur/style-people | ModulatedConv2d | false | 8,008 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
def make_kernel(k):
k = torch.tensor(k, dtype=torch.float32)
if k.ndim == 1:
k = k[None, :] * k[:, None]
k /= k.sum()
return k
def upfirdn2d(input, kernel, up=1, down=1, pad=(0, 0)):
out = UpFirDn2d.apply(input, kernel, (up, up), (down, down), (pad[0],
pad[1], pad[0], pad[1]))
return out
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class UpFirDn2dBackward(Function):
@staticmethod
def forward(ctx, grad_output, kernel, grad_kernel, up, down, pad, g_pad,
in_size, out_size):
up_x, up_y = up
down_x, down_y = down
g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1 = g_pad
grad_output = grad_output.reshape(-1, out_size[0], out_size[1], 1)
grad_input = upfirdn2d_op.upfirdn2d(grad_output, grad_kernel,
down_x, down_y, up_x, up_y, g_pad_x0, g_pad_x1, g_pad_y0, g_pad_y1)
grad_input = grad_input.view(in_size[0], in_size[1], in_size[2],
in_size[3])
ctx.save_for_backward(kernel)
pad_x0, pad_x1, pad_y0, pad_y1 = pad
ctx.up_x = up_x
ctx.up_y = up_y
ctx.down_x = down_x
ctx.down_y = down_y
ctx.pad_x0 = pad_x0
ctx.pad_x1 = pad_x1
ctx.pad_y0 = pad_y0
# ... truncated (>4000 chars) for memory efficiency |
HexaLinearScore | # 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_1/inductor_cache/cy/ccyfkccfxbnfkjvwfjzaxsw5jvq2nydk37m6cncatguka3ts2rlm.py
# Topologically Sorted Source Nodes: [g1], Original ATen: [aten.clone, aten._unsafe_view]
# Source node to ATen node mapping:
# g1 => clone, view
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%slice_2,), kwargs = {memory_format: torch.contiguous_format})
# %view : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone, [8, 4]), kwargs = {})
triton_poi_fused__unsafe_view_clone_0 = async_compile.triton('triton_poi_fused__unsafe_view_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=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_view_clone_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_view_clone_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 % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*(x1 % 2)) + (16*(x1 // 2))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/rt/crtsvl4vcapoyr4zrx2mmyplgvhsm4q3hiqv6yxodqxhxdzlvten.py
# Topologically Sorted Source Nodes: [g2], Original ATen: [aten.clone, aten._unsafe_view]
# Source node to ATen node mapping:
# g2 => clone_1, view_2
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%slice_4,), kwargs = {memory_format: torch.contiguous_format})
# %view_2 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone_1, [8, 4]), kwargs = {})
triton_poi_fused__unsafe_view_clone_1 = async_compile.triton('triton_poi_fused__unsafe_view_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_view_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_view_clone_1(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 % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4 + x0 + (4*(x1 % 2)) + (16*(x1 // 2))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ol/colxstasdar3rgy6o73kzqubr2qazpoaebxabxnrh5inl76dnz3w.py
# Topologically Sorted Source Nodes: [g3], Original ATen: [aten.clone, aten._unsafe_view]
# Source node to ATen node mapping:
# g3 => clone_2, view_4
# Graph fragment:
# %clone_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%slice_6,), kwargs = {memory_format: torch.contiguous_format})
# %view_4 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone_2, [8, 4]), kwargs = {})
triton_poi_fused__unsafe_view_clone_2 = async_compile.triton('triton_poi_fused__unsafe_view_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=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__unsafe_view_clone_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__unsafe_view_clone_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
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (8 + x0 + (4*(x1 % 2)) + (16*(x1 // 2))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/au/cauo53uhysnuofbqpimuv5u5cbfcljpuepjqgooaez6xw5l6tj3c.py
# Topologically Sorted Source Nodes: [mul, temp01], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# temp01 => mul_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, %view_3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %view_5), kwargs = {})
triton_poi_fused_mul_3 = async_compile.triton('triton_poi_fused_mul_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_3(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 3168
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_ptr2 + (x0), xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/un/cun7hyyxvq2lmcw62zo6gtp47cgasrerutclwzulpnojehzpihfy.py
# Topologically Sorted Source Nodes: [temp02], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# temp02 => mul_2, mul_3
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute, %permute_1), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_2, %mul_2), 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=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 25344
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x5 = xindex % 1584
x0 = xindex % 396
x3 = (xindex // 6336)
x2 = (xindex // 1584) % 4
x4 = (xindex // 1584)
tmp0 = tl.load(in_ptr0 + (x5), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + (396*x3)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + (x0 + (396*x2)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 * tmp2
tmp4 = tmp0 * tmp3
tl.store(out_ptr0 + (x5 + (1600*x4)), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/uu/cuum6xwufvzm2y2iooq5cf3tfsn77du5nnc57dtqh3rhu2vlpwfj.py
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.bmm, aten.transpose]
# Source node to ATen node mapping:
# score => bmm
# Graph fragment:
# %bmm : [num_users=1] = call_function[target=torch.ops.aten.bmm.default](args = (%view_6, %view_7), kwargs = {})
# %permute_10 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%view_7, [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=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 25344
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 396
x1 = (xindex // 396)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (396*(x1 % 4)) + (1600*(x1 // 4))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
tl.store(out_ptr1 + (x2), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/2k/c2kz22lnv73jl7bp4bvurxlinhg5sxwjhpigj6pak5c4mf74v3fr.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten.div]
# Source node to ATen node mapping:
# score_1 => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_9, 19.8997487421324), kwargs = {})
triton_poi_fused_div_6 = async_compile.triton('triton_poi_fused_div_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_6', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_6(in_out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = 0.050251890762960605
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 396), (396, 1))
assert_size_stride(primals_3, (4, 396), (396, 1))
assert_size_stride(primals_4, (4, 396), (396, 1))
assert_size_stride(primals_5, (4, 20), (20, 1))
assert_size_stride(primals_6, (20, 396), (396, 1))
assert_size_stride(primals_7, (20, 396), (396, 1))
assert_size_stride(primals_8, (20, 396), (396, 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: [g1], Original ATen: [aten.clone, aten._unsafe_view]
stream0 = get_raw_stream(0)
triton_poi_fused__unsafe_view_clone_0.run(primals_1, buf0, 32, grid=grid(32), stream=stream0)
buf1 = empty_strided_cuda((8, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g1], Original ATen: [aten.mm]
extern_kernels.mm(buf0, primals_2, out=buf1)
del primals_2
buf2 = empty_strided_cuda((8, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [g2], Original ATen: [aten.clone, aten._unsafe_view]
triton_poi_fused__unsafe_view_clone_1.run(primals_1, buf2, 32, grid=grid(32), stream=stream0)
buf3 = empty_strided_cuda((8, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g2], Original ATen: [aten.mm]
extern_kernels.mm(buf2, primals_3, out=buf3)
del primals_3
buf4 = empty_strided_cuda((8, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [g3], Original ATen: [aten.clone, aten._unsafe_view]
triton_poi_fused__unsafe_view_clone_2.run(primals_1, buf4, 32, grid=grid(32), stream=stream0)
del primals_1
buf5 = empty_strided_cuda((8, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g3], Original ATen: [aten.mm]
extern_kernels.mm(buf4, primals_4, out=buf5)
del primals_4
buf6 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g4], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, primals_6, out=buf6)
buf7 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g5], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, primals_7, out=buf7)
buf8 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
# Topologically Sorted Source Nodes: [g6], Original ATen: [aten.mm]
extern_kernels.mm(primals_5, primals_8, out=buf8)
buf9 = empty_strided_cuda((4, 2, 396), (792, 396, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, temp01], Original ATen: [aten.mul]
triton_poi_fused_mul_3.run(buf1, buf3, buf5, buf9, 3168, grid=grid(3168), stream=stream0)
buf10 = empty_strided_cuda((4, 4, 4, 396), (6400, 1600, 396, 1), torch.float32)
# Topologically Sorted Source Nodes: [temp02], Original ATen: [aten.mul]
triton_poi_fused_mul_4.run(buf8, buf6, buf7, buf10, 25344, grid=grid(25344), stream=stream0)
buf11 = empty_strided_cuda((1, 396, 64), (25344, 1, 396), torch.float32)
buf14 = empty_strided_cuda((1, 64, 396), (25344, 396, 1), torch.float32)
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.bmm, aten.transpose]
triton_poi_fused_bmm_transpose_5.run(buf10, buf11, buf14, 25344, grid=grid(25344), stream=stream0)
del buf10
buf12 = empty_strided_cuda((1, 8, 64), (512, 64, 1), torch.float32)
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf9, (1, 8, 396), (0, 396, 1), 0), buf11, out=buf12)
del buf11
buf13 = reinterpret_tensor(buf12, (4, 2, 4, 4, 4), (128, 64, 16, 4, 1), 0); del buf12 # reuse
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten.div]
triton_poi_fused_div_6.run(buf13, 512, grid=grid(512), stream=stream0)
return (buf13, buf1, buf3, buf5, buf6, buf7, buf8, reinterpret_tensor(buf9, (1, 396, 8), (3168, 1, 396), 0), buf14, reinterpret_tensor(primals_5, (20, 4), (1, 20), 0), reinterpret_tensor(primals_8, (396, 20), (1, 396), 0), reinterpret_tensor(primals_7, (396, 20), (1, 396), 0), reinterpret_tensor(primals_6, (396, 20), (1, 396), 0), reinterpret_tensor(buf4, (4, 8), (1, 4), 0), reinterpret_tensor(buf2, (4, 8), (1, 4), 0), reinterpret_tensor(buf0, (4, 8), (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, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 20), (20, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((20, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((20, 396), (396, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((20, 396), (396, 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 math
import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class HexaLinearScore(nn.Module):
"""
Outer product version of hexalinear function for sequence labeling.
"""
def __init__(self, wemb_size, tagset_size, temb_size=20, rank=396, std=
0.1545, normalization=True, **kwargs):
"""
Args:
wemb_size: word embedding hidden size
tagset_size: tag set size
temb_size: tag embedding size
rank: rank of the weight tensor
std: standard deviation of the tensor
"""
super(HexaLinearScore, self).__init__()
self.wemb_size = wemb_size
self.tagset_size = tagset_size
self.temb_size = temb_size
self.rank = rank
self.std = std
self.normalization = normalization
self.tag_emd = nn.Parameter(torch.Tensor(self.tagset_size, self.
temb_size))
self.W1 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.W2 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.W3 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.T1 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.T2 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.T3 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.rand_init()
self
def rand_init(self):
"""random initialization
"""
nn.init.uniform_(self.tag_emd, a=math.sqrt(6 / self.temb_size), b=
math.sqrt(6 / self.temb_size))
nn.init.normal_(self.T1, std=self.std)
nn.init.normal_(self.T2, std=self.std)
nn.init.normal_(self.T3, std=self.std)
nn.init.normal_(self.W1, std=self.std)
nn.init.normal_(self.W2, std=self.std)
nn.init.normal_(self.W3, std=self.std)
def forward(self, word_emb):
"""
Args:
word_emb: [batch, sent_length, wemb_size]
Returns: Tensor
[batch, sent_length-window_size, tagset_size, tagset_size]
"""
assert word_emb.size(2
) == self.wemb_size, 'batch sizes of encoder and decoder are requires to be equal.'
g1 = torch.matmul(word_emb[:, :-2], self.W1)
g2 = torch.matmul(word_emb[:, 1:-1], self.W2)
g3 = torch.matmul(word_emb[:, 2:], self.W3)
g4 = torch.matmul(self.tag_emd, self.T1)
g5 = torch.matmul(self.tag_emd, self.T2)
g6 = torch.matmul(self.tag_emd, self.T3)
temp01 = g1 * g2 * g3
temp02 = torch.einsum('ak,bk,ck->abck', [g4, g5, g6])
score = torch.einsum('nmk,abck->nmabc', [temp01, temp02])
if self.normalization:
score = score / math.sqrt(self.rank)
return score
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'wemb_size': 4, 'tagset_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import math
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused__unsafe_view_clone_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 % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * (x1 % 2) + 16 * (x1 // 2)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused__unsafe_view_clone_1(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 % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4 + x0 + 4 * (x1 % 2) + 16 * (x1 // 2)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused__unsafe_view_clone_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
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (8 + x0 + 4 * (x1 % 2) + 16 * (x1 // 2)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_mul_3(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 3168
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_ptr2 + x0, xmask)
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_poi_fused_mul_4(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 25344
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x5 = xindex % 1584
x0 = xindex % 396
x3 = xindex // 6336
x2 = xindex // 1584 % 4
x4 = xindex // 1584
tmp0 = tl.load(in_ptr0 + x5, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + 396 * x3), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr2 + (x0 + 396 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tmp1 * tmp2
tmp4 = tmp0 * tmp3
tl.store(out_ptr0 + (x5 + 1600 * x4), tmp4, xmask)
@triton.jit
def triton_poi_fused_bmm_transpose_5(in_ptr0, out_ptr0, out_ptr1, xnumel,
XBLOCK: tl.constexpr):
xnumel = 25344
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 396
x1 = xindex // 396
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 396 * (x1 % 4) + 1600 * (x1 // 4)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
tl.store(out_ptr1 + x2, tmp0, xmask)
@triton.jit
def triton_poi_fused_div_6(in_out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = 0.050251890762960605
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + x0, tmp2, 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), (16, 4, 1))
assert_size_stride(primals_2, (4, 396), (396, 1))
assert_size_stride(primals_3, (4, 396), (396, 1))
assert_size_stride(primals_4, (4, 396), (396, 1))
assert_size_stride(primals_5, (4, 20), (20, 1))
assert_size_stride(primals_6, (20, 396), (396, 1))
assert_size_stride(primals_7, (20, 396), (396, 1))
assert_size_stride(primals_8, (20, 396), (396, 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__unsafe_view_clone_0[grid(32)](primals_1, buf0, 32,
XBLOCK=32, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((8, 396), (396, 1), torch.float32)
extern_kernels.mm(buf0, primals_2, out=buf1)
del primals_2
buf2 = empty_strided_cuda((8, 4), (4, 1), torch.float32)
triton_poi_fused__unsafe_view_clone_1[grid(32)](primals_1, buf2, 32,
XBLOCK=32, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((8, 396), (396, 1), torch.float32)
extern_kernels.mm(buf2, primals_3, out=buf3)
del primals_3
buf4 = empty_strided_cuda((8, 4), (4, 1), torch.float32)
triton_poi_fused__unsafe_view_clone_2[grid(32)](primals_1, buf4, 32,
XBLOCK=32, num_warps=1, num_stages=1)
del primals_1
buf5 = empty_strided_cuda((8, 396), (396, 1), torch.float32)
extern_kernels.mm(buf4, primals_4, out=buf5)
del primals_4
buf6 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
extern_kernels.mm(primals_5, primals_6, out=buf6)
buf7 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
extern_kernels.mm(primals_5, primals_7, out=buf7)
buf8 = empty_strided_cuda((4, 396), (396, 1), torch.float32)
extern_kernels.mm(primals_5, primals_8, out=buf8)
buf9 = empty_strided_cuda((4, 2, 396), (792, 396, 1), torch.float32)
triton_poi_fused_mul_3[grid(3168)](buf1, buf3, buf5, buf9, 3168,
XBLOCK=256, num_warps=4, num_stages=1)
buf10 = empty_strided_cuda((4, 4, 4, 396), (6400, 1600, 396, 1),
torch.float32)
triton_poi_fused_mul_4[grid(25344)](buf8, buf6, buf7, buf10, 25344,
XBLOCK=256, num_warps=4, num_stages=1)
buf11 = empty_strided_cuda((1, 396, 64), (25344, 1, 396), torch.float32
)
buf14 = empty_strided_cuda((1, 64, 396), (25344, 396, 1), torch.float32
)
triton_poi_fused_bmm_transpose_5[grid(25344)](buf10, buf11, buf14,
25344, XBLOCK=256, num_warps=4, num_stages=1)
del buf10
buf12 = empty_strided_cuda((1, 8, 64), (512, 64, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf9, (1, 8, 396), (0, 396, 1
), 0), buf11, out=buf12)
del buf11
buf13 = reinterpret_tensor(buf12, (4, 2, 4, 4, 4), (128, 64, 16, 4,
1), 0)
del buf12
triton_poi_fused_div_6[grid(512)](buf13, 512, XBLOCK=128, num_warps
=4, num_stages=1)
return buf13, buf1, buf3, buf5, buf6, buf7, buf8, reinterpret_tensor(buf9,
(1, 396, 8), (3168, 1, 396), 0), buf14, reinterpret_tensor(primals_5,
(20, 4), (1, 20), 0), reinterpret_tensor(primals_8, (396, 20), (1,
396), 0), reinterpret_tensor(primals_7, (396, 20), (1, 396), 0
), reinterpret_tensor(primals_6, (396, 20), (1, 396), 0
), reinterpret_tensor(buf4, (4, 8), (1, 4), 0), reinterpret_tensor(buf2
, (4, 8), (1, 4), 0), reinterpret_tensor(buf0, (4, 8), (1, 4), 0)
class HexaLinearScoreNew(nn.Module):
"""
Outer product version of hexalinear function for sequence labeling.
"""
def __init__(self, wemb_size, tagset_size, temb_size=20, rank=396, std=
0.1545, normalization=True, **kwargs):
"""
Args:
wemb_size: word embedding hidden size
tagset_size: tag set size
temb_size: tag embedding size
rank: rank of the weight tensor
std: standard deviation of the tensor
"""
super(HexaLinearScoreNew, self).__init__()
self.wemb_size = wemb_size
self.tagset_size = tagset_size
self.temb_size = temb_size
self.rank = rank
self.std = std
self.normalization = normalization
self.tag_emd = nn.Parameter(torch.Tensor(self.tagset_size, self.
temb_size))
self.W1 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.W2 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.W3 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.T1 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.T2 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.T3 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.rand_init()
self
def rand_init(self):
"""random initialization
"""
nn.init.uniform_(self.tag_emd, a=math.sqrt(6 / self.temb_size), b=
math.sqrt(6 / self.temb_size))
nn.init.normal_(self.T1, std=self.std)
nn.init.normal_(self.T2, std=self.std)
nn.init.normal_(self.T3, std=self.std)
nn.init.normal_(self.W1, std=self.std)
nn.init.normal_(self.W2, std=self.std)
nn.init.normal_(self.W3, std=self.std)
def forward(self, input_0):
primals_5 = self.tag_emd
primals_2 = self.W1
primals_3 = self.W2
primals_4 = self.W3
primals_6 = self.T1
primals_7 = self.T2
primals_8 = self.T3
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0]
| Dadmatech/DadmaTools | HexaLinearScore | false | 8,009 | [
"Apache-2.0"
] | 25 | c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | https://github.com/Dadmatech/DadmaTools/tree/c1b7add5c33544f69c1ba1c5250a5ea07caf9aa2 | import math
import torch
import torch.utils.data.dataloader
import torch.nn as nn
import torch.nn
class Model(nn.Module):
"""
Outer product version of hexalinear function for sequence labeling.
"""
def __init__(self, wemb_size, tagset_size, temb_size=20, rank=396, std=
0.1545, normalization=True, **kwargs):
"""
Args:
wemb_size: word embedding hidden size
tagset_size: tag set size
temb_size: tag embedding size
rank: rank of the weight tensor
std: standard deviation of the tensor
"""
super().__init__()
self.wemb_size = wemb_size
self.tagset_size = tagset_size
self.temb_size = temb_size
self.rank = rank
self.std = std
self.normalization = normalization
self.tag_emd = nn.Parameter(torch.Tensor(self.tagset_size, self.
temb_size))
self.W1 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.W2 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.W3 = nn.Parameter(torch.Tensor(self.wemb_size, self.rank))
self.T1 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.T2 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.T3 = nn.Parameter(torch.Tensor(self.temb_size, self.rank))
self.rand_init()
self
def rand_init(self):
"""random initialization
"""
nn.init.uniform_(self.tag_emd, a=math.sqrt(6 / self.temb_size), b=
math.sqrt(6 / self.temb_size))
nn.init.normal_(self.T1, std=self.std)
nn.init.normal_(self.T2, std=self.std)
nn.init.normal_(self.T3, std=self.std)
nn.init.normal_(self.W1, std=self.std)
nn.init.normal_(self.W2, std=self.std)
nn.init.normal_(self.W3, std=self.std)
def forward(self, word_emb):
"""
Args:
word_emb: [batch, sent_length, wemb_size]
Returns: Tensor
[batch, sent_length-window_size, tagset_size, tagset_size]
"""
assert word_emb.size(2
) == self.wemb_size, 'batch sizes of encoder and decoder are requires to be equal.'
g1 = torch.matmul(word_emb[:, :-2], self.W1)
g2 = torch.matmul(word_emb[:, 1:-1], self.W2)
g3 = torch.matmul(word_emb[:, 2:], self.W3)
g4 = torch.matmul(self.tag_emd, self.T1)
g5 = torch.matmul(self.tag_emd, self.T2)
g6 = torch.matmul(self.tag_emd, self.T3)
temp01 = g1 * g2 * g3
temp02 = torch.einsum('ak,bk,ck->abck', [g4, g5, g6])
score = torch.einsum('nmk,abck->nmabc', [temp01, temp02])
if self.normalization:
score = score / math.sqrt(self.rank)
return score
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
FFN | # 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_1/inductor_cache/4g/c4guhk7x6skkidedvs2gxz2kcu6gb76l3ig5crjjvjtzvnjlhlte.py
# Topologically Sorted Source Nodes: [hidden_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# hidden_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [hidden_1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf3, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 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, 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), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_4, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.nn.functional as F
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super(FFN, self).__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'d_inner_hid': 4, 'd_model': 4, 'dropout_rate': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 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)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf3 = 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, buf3, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf2)
del primals_5
return reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_4, buf3
class FFNNew(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super(FFNNew, self).__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DocYard-ai/UCR | FFN | false | 8,010 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super().__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 0.5]
|
AdaptiveConcatPool2d | # 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_1/inductor_cache/pq/cpqgsju5p2eyjjmy3imid4jf33v3d47eyjlf6k2xstq6mzcillsk.py
# Topologically Sorted Source Nodes: [adaptive_max_pool2d], Original ATen: [aten.adaptive_max_pool2d]
# Source node to ATen node mapping:
# adaptive_max_pool2d => adaptive_max_pool2d
# Graph fragment:
# %adaptive_max_pool2d : [num_users=1] = call_function[target=torch.ops.aten.adaptive_max_pool2d.default](args = (%arg0_1, [1, 1]), kwargs = {})
triton_poi_fused_adaptive_max_pool2d_0 = async_compile.triton('triton_poi_fused_adaptive_max_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 16, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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
x2 = xindex
x0 = xindex % 4
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (16*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (16*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (16*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (4 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (5 + (16*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (6 + (16*x2)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (7 + (16*x2)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (8 + (16*x2)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (9 + (16*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (10 + (16*x2)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (11 + (16*x2)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (12 + (16*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (13 + (16*x2)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (14 + (16*x2)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr0 + (15 + (16*x2)), xmask, eviction_policy='evict_last')
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 + (x0 + (8*x1)), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mq/cmqjuv33rly4whmxqjz5w75pgp36j44mgyzzrxfythdirwrwqper.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=1] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [-1, -2], True), kwargs = {})
triton_per_fused_mean_1 = async_compile.triton('triton_per_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mean_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_1(in_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)
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_ptr1 + (x2 + (8*x3)), 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)
buf3 = empty_strided_cuda((4, 8, 1, 1), (8, 1, 1, 1), torch.float32)
buf0 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 0) # alias
# Topologically Sorted Source Nodes: [adaptive_max_pool2d], Original ATen: [aten.adaptive_max_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_adaptive_max_pool2d_0.run(arg0_1, buf0, 16, grid=grid(16), stream=stream0)
buf2 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 4) # alias
# Topologically Sorted Source Nodes: [adaptive_avg_pool2d], Original ATen: [aten.mean]
triton_per_fused_mean_1.run(arg0_1, buf2, 16, 16, grid=grid(16), 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
from torch import nn
import torch.optim
class AdaptiveConcatPool2d(nn.Module):
def __init__(self, sz=1):
super().__init__()
self.output_size = sz or 1
self.ap = nn.AdaptiveAvgPool2d(self.output_size)
self.mp = nn.AdaptiveMaxPool2d(self.output_size)
def forward(self, x):
return torch.cat([self.mp(x), self.ap(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
from torch._inductor.runtime import triton_helpers
from torch import nn
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_adaptive_max_pool2d_0(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + 16 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr0 + (2 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr0 + (3 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp7 = tl.load(in_ptr0 + (4 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp9 = tl.load(in_ptr0 + (5 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (6 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp13 = tl.load(in_ptr0 + (7 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp15 = tl.load(in_ptr0 + (8 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp17 = tl.load(in_ptr0 + (9 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr0 + (10 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (11 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr0 + (12 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr0 + (13 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp27 = tl.load(in_ptr0 + (14 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr0 + (15 + 16 * x2), xmask, eviction_policy=
'evict_last')
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 + (x0 + 8 * x1), tmp30, xmask)
@triton.jit
def triton_per_fused_mean_1(in_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)
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_ptr1 + (x2 + 8 * x3), 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)
buf3 = empty_strided_cuda((4, 8, 1, 1), (8, 1, 1, 1), torch.float32)
buf0 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 0)
get_raw_stream(0)
triton_poi_fused_adaptive_max_pool2d_0[grid(16)](arg0_1, buf0, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf2 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 4)
triton_per_fused_mean_1[grid(16)](arg0_1, buf2, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
del arg0_1
return buf3,
class AdaptiveConcatPool2dNew(nn.Module):
def __init__(self, sz=1):
super().__init__()
self.output_size = sz or 1
self.ap = nn.AdaptiveAvgPool2d(self.output_size)
self.mp = nn.AdaptiveMaxPool2d(self.output_size)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DrHB/PANDA-2nd-place-solution | AdaptiveConcatPool2d | false | 8,011 | [
"MIT"
] | 17 | 44ab72780f9e3594811f2a7520456de7a9e677db | https://github.com/DrHB/PANDA-2nd-place-solution/tree/44ab72780f9e3594811f2a7520456de7a9e677db | import torch
from torch import nn
import torch.optim
class Model(nn.Module):
def __init__(self, sz=1):
super().__init__()
self.output_size = sz or 1
self.ap = nn.AdaptiveAvgPool2d(self.output_size)
self.mp = nn.AdaptiveMaxPool2d(self.output_size)
def forward(self, x):
return torch.cat([self.mp(x), self.ap(x)], 1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
StyledSiren | # 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_1/inductor_cache/ag/caggqh3nrqirp3azykpm6daivhojwggbhob6pnubsheittrjeiky.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, 0.5), 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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qw/cqw2ahskucrmwkbadqswj76db4kfcxn5w76kji7peia7kwu4zbkq.py
# Topologically Sorted Source Nodes: [mul_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_1 => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, 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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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')
# kernel path: runs/run_shard_1/inductor_cache/lo/clooq2sxzea2fl7hxiqrjcdxc7anstn74sfzrmxvn6fvyetbqgas.py
# Topologically Sorted Source Nodes: [mul_2, weight, pow_1, sum_1, mul_4, add, demod], Original ATen: [aten.mul, aten.pow, aten.sum, aten.add, aten.rsqrt]
# Source node to ATen node mapping:
# add => add
# demod => rsqrt
# mul_2 => mul_2
# mul_4 => mul_4
# pow_1 => pow_1
# sum_1 => sum_1
# weight => mul_3
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_5, 0.5), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %view), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%mul_3, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [2, 3, 4]), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 450.0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, 1e-08), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_add_mul_pow_rsqrt_sum_2 = async_compile.triton('triton_poi_fused_add_mul_pow_rsqrt_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_pow_rsqrt_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_rsqrt_sum_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp5 = tmp4 * tmp4
tmp7 = tmp6 * tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp5 + tmp10
tmp13 = tmp12 * tmp1
tmp15 = tmp13 * tmp14
tmp16 = tmp15 * tmp15
tmp17 = tmp11 + tmp16
tmp19 = tmp18 * tmp1
tmp21 = tmp19 * tmp20
tmp22 = tmp21 * tmp21
tmp23 = tmp17 + tmp22
tmp24 = 450.0
tmp25 = tmp23 * tmp24
tmp26 = 1e-08
tmp27 = tmp25 + tmp26
tmp28 = libdevice.rsqrt(tmp27)
tl.store(out_ptr0 + (x2), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kb/ckbodo65ksrjdz7utqztuporiv5akgtcmyczja2zigsos2ayoxlr.py
# Topologically Sorted Source Nodes: [mul_2, weight, weight_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_2 => mul_2
# weight => mul_3
# weight_1 => mul_5
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_5, 0.5), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %view), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %view_1), kwargs = {})
triton_poi_fused_mul_3 = async_compile.triton('triton_poi_fused_mul_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_3(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
x3 = xindex % 16
x0 = xindex % 4
x2 = (xindex // 16)
x4 = (xindex // 4)
x5 = xindex
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tl.store(out_ptr0 + (x5), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/no/cnodgesumng6y2yeeivcqt6dlp7xrcmuprgfv74nrzgrvczfkyar.py
# Topologically Sorted Source Nodes: [out_3, mul_7, out_4, out_5], Original ATen: [aten.mul, aten.add, aten.sin]
# Source node to ATen node mapping:
# mul_7 => mul_7
# out_3 => mul_6
# out_4 => add_1
# out_5 => sin
# Graph fragment:
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_4, 30.0), kwargs = {})
# %mul_7 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_6, %normal_functional), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_6, %mul_7), kwargs = {})
# %sin : [num_users=1] = call_function[target=torch.ops.aten.sin.default](args = (%add_1,), kwargs = {})
triton_poi_fused_add_mul_sin_4 = async_compile.triton('triton_poi_fused_add_mul_sin_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: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_sin_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_sin_4(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
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp3 = tl.load(in_ptr1 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp5 = tl.load(in_ptr2 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp1 = 30.0
tmp2 = tmp0 * tmp1
tmp6 = tmp4 * tmp5
tmp7 = tmp2 + tmp6
tmp8 = tl_math.sin(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 = 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, (1, 4, 4, 1, 1), (16, 4, 1, 1, 1))
assert_size_stride(primals_6, (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: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_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_1], 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_1, out], Original ATen: [aten.mul, aten.addmm]
extern_kernels.addmm(buf1, primals_4, reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul_2, weight, pow_1, sum_1, mul_4, add, demod], Original ATen: [aten.mul, aten.pow, aten.sum, aten.add, aten.rsqrt]
triton_poi_fused_add_mul_pow_rsqrt_sum_2.run(primals_5, buf2, buf3, 16, grid=grid(16), stream=stream0)
buf4 = empty_strided_cuda((4, 4, 4, 1, 1), (16, 4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_2, weight, weight_1], Original ATen: [aten.mul]
triton_poi_fused_mul_3.run(primals_5, buf2, buf3, buf4, 64, grid=grid(64), stream=stream0)
del buf3
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf4, (16, 4, 1, 1), (4, 1, 0, 0), 0), stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=4, bias=None)
assert_size_stride(buf5, (1, 16, 4, 4), (256, 16, 4, 1))
buf6 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [noise], Original ATen: [aten.normal_functional]
buf7 = torch.ops.aten.normal_functional.default(buf6)
del buf6
buf8 = buf7
del buf7
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_3, mul_7, out_4, out_5], Original ATen: [aten.mul, aten.add, aten.sin]
triton_poi_fused_add_mul_sin_4.run(buf5, primals_6, buf8, buf9, 256, grid=grid(256), stream=stream0)
return (buf9, primals_4, primals_5, primals_6, buf2, reinterpret_tensor(buf4, (16, 4, 1, 1), (4, 1, 1, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), buf5, buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((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((1, 4, 4, 1, 1), (16, 4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_6 = 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])
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 torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class NoiseInjection(nn.Module):
def __init__(self):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1))
def forward(self, image, noise=None):
if noise is None:
batch, _, height, width = image.shape
noise = image.new_empty(batch, 1, height, width).normal_()
return image + self.weight * noise
class ModulatedSiren2d(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30):
super().__init__()
self.eps = 1e-08
self.kernel_size = 1
self.in_channel = in_channel
self.out_channel = out_channel
fan_in = in_channel
self.scale = 1 / math.sqrt(fan_in)
if is_first:
self.demod_scale = 3 * fan_in
else:
self.demod_scale = omega_0 ** 2 / 2
self.omega0 = omega_0
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
1, 1))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel})'
)
def forward(self, input, style):
batch, in_channel, height, width = input.shape
style = self.modulation(style)
style = style.view(batch, 1, in_channel, 1, 1)
weight = self.scale * self.weight * style
if self.demodulate:
demod = torch.rsqrt(weight.pow(2).sum([2, 3, 4]) * self.
demod_scale + 1e-08)
weight = weight * demod.view(batch, self.out_channel, 1, 1, 1)
weight = weight.view(batch * self.out_channel, in_channel, 1, 1)
input = input.view(1, batch * in_channel, height, width)
out = F.conv2d(input, weight, groups=batch)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
out = out * self.omega0
return out
class StyledSiren(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30.0):
super().__init__()
self.conv = ModulatedSiren2d(in_channel, out_channel, style_dim,
demodulate=demodulate, is_first=is_first, omega_0=omega_0)
self.noise = NoiseInjection()
def forward(self, input, style, noise=None):
out = self.conv(input, style)
out = self.noise(out, noise=noise)
out = torch.sin(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'out_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, math as tl_math
from torch.autograd import Function
import 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_mul_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
tl.store(out_ptr0 + x0, tmp2, 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)
@triton.jit
def triton_poi_fused_add_mul_pow_rsqrt_sum_2(in_ptr0, in_ptr1, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp14 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp20 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp5 = tmp4 * tmp4
tmp7 = tmp6 * tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp5 + tmp10
tmp13 = tmp12 * tmp1
tmp15 = tmp13 * tmp14
tmp16 = tmp15 * tmp15
tmp17 = tmp11 + tmp16
tmp19 = tmp18 * tmp1
tmp21 = tmp19 * tmp20
tmp22 = tmp21 * tmp21
tmp23 = tmp17 + tmp22
tmp24 = 450.0
tmp25 = tmp23 * tmp24
tmp26 = 1e-08
tmp27 = tmp25 + tmp26
tmp28 = libdevice.rsqrt(tmp27)
tl.store(out_ptr0 + x2, tmp28, xmask)
@triton.jit
def triton_poi_fused_mul_3(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
x3 = xindex % 16
x0 = xindex % 4
x2 = xindex // 16
x4 = xindex // 4
x5 = xindex
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tl.store(out_ptr0 + x5, tmp6, xmask)
@triton.jit
def triton_poi_fused_add_mul_sin_4(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
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp3 = tl.load(in_ptr1 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp5 = tl.load(in_ptr2 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = 30.0
tmp2 = tmp0 * tmp1
tmp6 = tmp4 * tmp5
tmp7 = tmp2 + tmp6
tmp8 = tl_math.sin(tmp7)
tl.store(out_ptr0 + x3, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 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, (1, 4, 4, 1, 1), (16, 4, 1, 1, 1))
assert_size_stride(primals_6, (1,), (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_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_4, reinterpret_tensor(buf0, (4,
4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
buf3 = buf0
del buf0
triton_poi_fused_add_mul_pow_rsqrt_sum_2[grid(16)](primals_5, buf2,
buf3, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4, 4, 1, 1), (16, 4, 1, 1, 1), torch.
float32)
triton_poi_fused_mul_3[grid(64)](primals_5, buf2, buf3, buf4, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del buf3
buf5 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1,
16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf4, (16, 4,
1, 1), (4, 1, 0, 0), 0), stride=(1, 1), padding=(0, 0),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=4, bias=None)
assert_size_stride(buf5, (1, 16, 4, 4), (256, 16, 4, 1))
buf6 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
buf7 = torch.ops.aten.normal_functional.default(buf6)
del buf6
buf8 = buf7
del buf7
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_mul_sin_4[grid(256)](buf5, primals_6, buf8,
buf9, 256, XBLOCK=128, num_warps=4, num_stages=1)
return buf9, primals_4, primals_5, primals_6, buf2, reinterpret_tensor(buf4
, (16, 4, 1, 1), (4, 1, 1, 1), 0), reinterpret_tensor(primals_1, (1,
16, 4, 4), (256, 16, 4, 1), 0), buf5, buf8
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class NoiseInjection(nn.Module):
def __init__(self):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1))
def forward(self, image, noise=None):
if noise is None:
batch, _, height, width = image.shape
noise = image.new_empty(batch, 1, height, width).normal_()
return image + self.weight * noise
class ModulatedSiren2d(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30):
super().__init__()
self.eps = 1e-08
self.kernel_size = 1
self.in_channel = in_channel
self.out_channel = out_channel
fan_in = in_channel
self.scale = 1 / math.sqrt(fan_in)
if is_first:
self.demod_scale = 3 * fan_in
else:
self.demod_scale = omega_0 ** 2 / 2
self.omega0 = omega_0
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
1, 1))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel})'
)
def forward(self, input, style):
batch, in_channel, height, width = input.shape
style = self.modulation(style)
style = style.view(batch, 1, in_channel, 1, 1)
weight = self.scale * self.weight * style
if self.demodulate:
demod = torch.rsqrt(weight.pow(2).sum([2, 3, 4]) * self.
demod_scale + 1e-08)
weight = weight * demod.view(batch, self.out_channel, 1, 1, 1)
weight = weight.view(batch * self.out_channel, in_channel, 1, 1)
input = input.view(1, batch * in_channel, height, width)
out = F.conv2d(input, weight, groups=batch)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
out = out * self.omega0
return out
class StyledSirenNew(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30.0):
super().__init__()
self.conv = ModulatedSiren2d(in_channel, out_channel, style_dim,
demodulate=demodulate, is_first=is_first, omega_0=omega_0)
self.noise = NoiseInjection()
def forward(self, input_0, input_1):
primals_5 = self.conv.weight
primals_2 = self.conv.modulation.weight
primals_3 = self.conv.modulation.bias
primals_6 = self.noise.weight
primals_1 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
| Dolorousrtur/style-people | StyledSiren | false | 8,012 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class NoiseInjection(nn.Module):
def __init__(self):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1))
def forward(self, image, noise=None):
if noise is None:
batch, _, height, width = image.shape
noise = image.new_empty(batch, 1, height, width).normal_()
return image + self.weight * noise
class ModulatedSiren2d(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30):
super().__init__()
self.eps = 1e-08
self.kernel_size = 1
self.in_channel = in_channel
self.out_channel = out_channel
fan_in = in_channel
self.scale = 1 / math.sqrt(fan_in)
if is_first:
self.demod_scale = 3 * fan_in
else:
self.demod_scale = omega_0 ** 2 / 2
self.omega0 = omega_0
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
1, 1))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.i
# ... truncated (>4000 chars) for memory efficiency |
DotAttn | # 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_1/inductor_cache/ak/cak2jvwrrs3dupx437xgjs26hfkszjf3ad7of7btisy5e4l2yqym.py
# Topologically Sorted Source Nodes: [mul, raw_scores_exp, sub, mul_1, raw_scores_exp_1], Original ATen: [aten.mul, aten.sum, aten.rsub, aten.sub]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# raw_scores_exp => sum_1
# raw_scores_exp_1 => sub_1
# sub => sub
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%expand, %expand_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [3]), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %expand_2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, 1e+20), kwargs = {})
# %sub_1 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sum_1, %mul_1), kwargs = {})
triton_poi_fused_mul_rsub_sub_sum_0 = async_compile.triton('triton_poi_fused_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_rsub_sub_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_rsub_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 % 4
x2 = (xindex // 16)
x3 = (xindex // 4)
x4 = xindex
tmp0 = tl.load(in_ptr0 + ((4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x3), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x3)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x3)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0) + (16*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x3)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr2 + (x3), 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
tmp16 = 1.0
tmp17 = tmp16 - tmp15
tmp18 = 1e+20
tmp19 = tmp17 * tmp18
tmp20 = tmp14 - tmp19
tl.store(out_ptr0 + (x4), tmp20, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/so/csof3x4y7lrn7on2ocdmxal3ggsfnqhrpinejjswa64wfco7qk6s.py
# Topologically Sorted Source Nodes: [scores_exp], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# scores_exp => amax, exp, sub_2
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%sub_1, [1], True), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_2,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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')
tmp2 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/n2/cn2d7m5uq6db3dfhz3ze6y5xji4hq7paw66kfee3iwqh6vc2tgis.py
# Topologically Sorted Source Nodes: [scores_exp], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# scores_exp => div, sum_2
# Graph fragment:
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_2), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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')
tmp2 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + (16*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')
# kernel path: runs/run_shard_1/inductor_cache/xs/cxsgrcou4vp4e3jiac2h3gu6wahpi3hloocqhs667i3zdmrnhuug.py
# Topologically Sorted Source Nodes: [mul_2, context_exp], Original ATen: [aten.mul, aten.sum]
# Source node to ATen node mapping:
# context_exp => sum_3
# mul_2 => mul_2
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%expand_3, %expand_1), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_2, [1]), kwargs = {})
triton_poi_fused_mul_sum_3 = async_compile.triton('triton_poi_fused_mul_sum_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_sum_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_sum_3(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
x1 = (xindex // 4) % 4
x2 = (xindex // 16)
x0 = xindex % 4
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x1 + (16*x2)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (4 + x1 + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (8 + x1 + (16*x2)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (12 + x1 + (16*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (12 + x0 + (16*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
tl.store(out_ptr0 + (x3), tmp14, 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), (16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(arg2_1, (4, 4), (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, raw_scores_exp, sub, mul_1, raw_scores_exp_1], Original ATen: [aten.mul, aten.sum, aten.rsub, aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_rsub_sub_sum_0.run(arg1_1, arg0_1, arg2_1, buf0, 64, grid=grid(64), stream=stream0)
del arg1_1
del arg2_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [scores_exp], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [scores_exp], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [mul_2, context_exp], Original ATen: [aten.mul, aten.sum]
triton_poi_fused_mul_sum_3.run(buf2, arg0_1, buf3, 64, grid=grid(64), stream=stream0)
del arg0_1
return (buf3, 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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
arg2_1 = rand_strided((4, 4), (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
from torch import nn
from torch.nn import functional as F
class DotAttn(nn.Module):
def __init__(self, dropout=0, enc_trans=None, dec_trans=None):
super().__init__()
self.dropout = nn.Dropout(dropout)
self.dec_trans = dec_trans or nn.Identity()
self.enc_trans = enc_trans or nn.Identity()
def forward(self, enc, enc_mask, dec):
enc_trans = self.dropout(self.enc_trans(enc))
dec_trans = self.dropout(self.dec_trans(dec))
Bdec, Tdec, Ddec = dec_trans.size()
Benc, Tenc, Denc = enc_trans.size()
dec_trans_exp = dec_trans.unsqueeze(1).expand(Bdec, Tenc, Tdec, Ddec)
enc_trans_exp = enc_trans.unsqueeze(2).expand(Benc, Tenc, Tdec, Denc)
mask_exp = enc_mask.unsqueeze(2).expand(Benc, Tenc, Tdec)
raw_scores_exp = dec_trans_exp.mul(enc_trans_exp).sum(3)
raw_scores_exp -= (1 - mask_exp) * 1e+20
scores_exp = F.softmax(raw_scores_exp, dim=1)
context_exp = scores_exp.unsqueeze(3).expand_as(enc_trans_exp).mul(
enc_trans_exp).sum(1)
return context_exp, scores_exp
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
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
@triton.jit
def triton_poi_fused_mul_rsub_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 % 4
x2 = xindex // 16
x3 = xindex // 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (4 * x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x3, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0 + 16 * x2), xmask, eviction_policy
='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x3), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0 + 16 * x2), xmask, eviction_policy
='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x3), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0 + 16 * x2), xmask,
eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + 4 * x3), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr2 + x3, 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
tmp16 = 1.0
tmp17 = tmp16 - tmp15
tmp18 = 1e+20
tmp19 = tmp17 * tmp18
tmp20 = tmp14 - tmp19
tl.store(out_ptr0 + x4, tmp20, xmask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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')
tmp2 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 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')
tmp2 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + 16 * 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)
@triton.jit
def triton_poi_fused_mul_sum_3(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
x1 = xindex // 4 % 4
x2 = xindex // 16
x0 = xindex % 4
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x1 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (4 + x1 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr1 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp7 = tl.load(in_ptr0 + (8 + x1 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp8 = tl.load(in_ptr1 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp11 = tl.load(in_ptr0 + (12 + x1 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp12 = tl.load(in_ptr1 + (12 + x0 + 16 * 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
tl.store(out_ptr0 + x3, tmp14, xmask)
def call(args):
arg0_1, arg1_1, arg2_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(arg2_1, (4, 4), (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_rsub_sub_sum_0[grid(64)](arg1_1, arg0_1,
arg2_1, buf0, 64, XBLOCK=64, num_warps=1, num_stages=1)
del arg1_1
del arg2_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(64)](buf0, buf1, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf2 = buf0
del buf0
triton_poi_fused__softmax_2[grid(64)](buf1, buf2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf3 = buf1
del buf1
triton_poi_fused_mul_sum_3[grid(64)](buf2, arg0_1, buf3, 64, XBLOCK
=64, num_warps=1, num_stages=1)
del arg0_1
return buf3, buf2
class DotAttnNew(nn.Module):
def __init__(self, dropout=0, enc_trans=None, dec_trans=None):
super().__init__()
self.dropout = nn.Dropout(dropout)
self.dec_trans = dec_trans or nn.Identity()
self.enc_trans = enc_trans or nn.Identity()
def forward(self, input_0, input_1, input_2):
arg0_1 = input_0
arg2_1 = input_1
arg1_1 = input_2
output = call([arg0_1, arg1_1, arg2_1])
return output[0], output[1]
| DreamerDeo/gazp | DotAttn | false | 8,013 | [
"MIT"
] | 18 | 5f823a447ffdf5176023a01516d2be7c383294d9 | https://github.com/DreamerDeo/gazp/tree/5f823a447ffdf5176023a01516d2be7c383294d9 | import torch
from torch import nn
from torch.nn import functional as F
class Model(nn.Module):
def __init__(self, dropout=0, enc_trans=None, dec_trans=None):
super().__init__()
self.dropout = nn.Dropout(dropout)
self.dec_trans = dec_trans or nn.Identity()
self.enc_trans = enc_trans or nn.Identity()
def forward(self, enc, enc_mask, dec):
enc_trans = self.dropout(self.enc_trans(enc))
dec_trans = self.dropout(self.dec_trans(dec))
Bdec, Tdec, Ddec = dec_trans.size()
Benc, Tenc, Denc = enc_trans.size()
dec_trans_exp = dec_trans.unsqueeze(1).expand(Bdec, Tenc, Tdec, Ddec)
enc_trans_exp = enc_trans.unsqueeze(2).expand(Benc, Tenc, Tdec, Denc)
mask_exp = enc_mask.unsqueeze(2).expand(Benc, Tenc, Tdec)
raw_scores_exp = dec_trans_exp.mul(enc_trans_exp).sum(3)
raw_scores_exp -= (1 - mask_exp) * 1e+20
scores_exp = F.softmax(raw_scores_exp, dim=1)
context_exp = scores_exp.unsqueeze(3).expand_as(enc_trans_exp).mul(
enc_trans_exp).sum(1)
return context_exp, scores_exp
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return []
|
GeM | # 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_1/inductor_cache/rq/crq7jtx7tfmjsxxgrv45sgqhhnh6jipryykxxzc5kxd24py3qg6z.py
# Topologically Sorted Source Nodes: [clamp, pow_1], Original ATen: [aten.clamp, aten.pow]
# Source node to ATen node mapping:
# clamp => clamp_min
# pow_1 => pow_1
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%primals_2, 1e-06), kwargs = {})
# %pow_1 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Tensor](args = (%clamp_min, %primals_1), kwargs = {})
triton_poi_fused_clamp_pow_0 = async_compile.triton('triton_poi_fused_clamp_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clamp_pow_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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)
tmp3 = tl.load(in_ptr1 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp1 = 1e-06
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp5 = libdevice.pow(tmp2, tmp4)
tl.store(out_ptr0 + (x0), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ml/cml4fbhpk4n4s7nylm7j6amagl4nd5bf2fmbnouiq6lnakskb75o.py
# Topologically Sorted Source Nodes: [avg_pool2d, truediv, pow_2], Original ATen: [aten.avg_pool2d, aten.reciprocal, aten.mul, aten.pow]
# Source node to ATen node mapping:
# avg_pool2d => avg_pool2d
# pow_2 => pow_2
# truediv => mul, reciprocal
# Graph fragment:
# %avg_pool2d : [num_users=2] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%pow_1, [4, 4]), kwargs = {})
# %reciprocal : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%primals_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%reciprocal, 1.0), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Tensor](args = (%avg_pool2d, %mul), kwargs = {})
triton_poi_fused_avg_pool2d_mul_pow_reciprocal_1 = async_compile.triton('triton_poi_fused_avg_pool2d_mul_pow_reciprocal_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_avg_pool2d_mul_pow_reciprocal_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 17, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_avg_pool2d_mul_pow_reciprocal_1(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 + (16*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (16*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (16*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (16*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (4 + (16*x0)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (5 + (16*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (6 + (16*x0)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (7 + (16*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (8 + (16*x0)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (9 + (16*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (10 + (16*x0)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (11 + (16*x0)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (12 + (16*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (13 + (16*x0)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (14 + (16*x0)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr0 + (15 + (16*x0)), xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr1 + (0))
tmp34 = tl.broadcast_to(tmp33, [XBLOCK])
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp8 = tmp7 + tmp6
tmp10 = tmp9 + tmp8
tmp12 = tmp11 + tmp10
tmp14 = tmp13 + tmp12
tmp16 = tmp15 + tmp14
tmp18 = tmp17 + tmp16
tmp20 = tmp19 + tmp18
tmp22 = tmp21 + tmp20
tmp24 = tmp23 + tmp22
tmp26 = tmp25 + tmp24
tmp28 = tmp27 + tmp26
tmp30 = tmp29 + tmp28
tmp31 = 0.0625
tmp32 = tmp30 * tmp31
tmp35 = tl.full([1], 1, tl.int32)
tmp36 = tmp35 / tmp34
tmp37 = 1.0
tmp38 = tmp36 * tmp37
tmp39 = libdevice.pow(tmp32, tmp38)
tl.store(out_ptr0 + (x0), tmp32, xmask)
tl.store(out_ptr1 + (x0), tmp39, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (1, ), (1, ))
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: [clamp, pow_1], Original ATen: [aten.clamp, aten.pow]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_pow_0.run(primals_2, primals_1, buf0, 256, grid=grid(256), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [avg_pool2d, truediv, pow_2], Original ATen: [aten.avg_pool2d, aten.reciprocal, aten.mul, aten.pow]
triton_poi_fused_avg_pool2d_mul_pow_reciprocal_1.run(buf0, primals_1, buf1, buf2, 16, grid=grid(16), stream=stream0)
return (buf2, primals_1, primals_2, buf0, 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((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.functional as F
from torch import nn
from torch.nn.parameter import Parameter
import torch.optim
def gem(x, p=3, eps=1e-06):
return F.avg_pool2d(x.clamp(min=eps).pow(p), (x.size(-2), x.size(-1))).pow(
1.0 / p)
class GeM(nn.Module):
def __init__(self, p=3, eps=1e-06):
super(GeM, self).__init__()
self.p = Parameter(torch.ones(1) * p)
self.eps = eps
def forward(self, x):
return gem(x, p=self.p, eps=self.eps)
def __repr__(self):
return self.__class__.__name__ + '(' + 'p=' + '{:.4f}'.format(self.
p.data.tolist()[0]) + ', ' + 'eps=' + str(self.eps) + ')'
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 libdevice
import torch.nn.functional as F
from torch import nn
from torch.nn.parameter import Parameter
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_clamp_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)
tmp3 = tl.load(in_ptr1 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp1 = 1e-06
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp5 = libdevice.pow(tmp2, tmp4)
tl.store(out_ptr0 + x0, tmp5, xmask)
@triton.jit
def triton_poi_fused_avg_pool2d_mul_pow_reciprocal_1(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 + 16 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr0 + (2 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr0 + (3 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp7 = tl.load(in_ptr0 + (4 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp9 = tl.load(in_ptr0 + (5 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (6 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp13 = tl.load(in_ptr0 + (7 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp15 = tl.load(in_ptr0 + (8 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp17 = tl.load(in_ptr0 + (9 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr0 + (10 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (11 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr0 + (12 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr0 + (13 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp27 = tl.load(in_ptr0 + (14 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr0 + (15 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp33 = tl.load(in_ptr1 + 0)
tmp34 = tl.broadcast_to(tmp33, [XBLOCK])
tmp2 = tmp1 + tmp0
tmp4 = tmp3 + tmp2
tmp6 = tmp5 + tmp4
tmp8 = tmp7 + tmp6
tmp10 = tmp9 + tmp8
tmp12 = tmp11 + tmp10
tmp14 = tmp13 + tmp12
tmp16 = tmp15 + tmp14
tmp18 = tmp17 + tmp16
tmp20 = tmp19 + tmp18
tmp22 = tmp21 + tmp20
tmp24 = tmp23 + tmp22
tmp26 = tmp25 + tmp24
tmp28 = tmp27 + tmp26
tmp30 = tmp29 + tmp28
tmp31 = 0.0625
tmp32 = tmp30 * tmp31
tmp35 = tl.full([1], 1, tl.int32)
tmp36 = tmp35 / tmp34
tmp37 = 1.0
tmp38 = tmp36 * tmp37
tmp39 = libdevice.pow(tmp32, tmp38)
tl.store(out_ptr0 + x0, tmp32, xmask)
tl.store(out_ptr1 + x0, tmp39, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_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)
get_raw_stream(0)
triton_poi_fused_clamp_pow_0[grid(256)](primals_2, primals_1, buf0,
256, XBLOCK=128, num_warps=4, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
triton_poi_fused_avg_pool2d_mul_pow_reciprocal_1[grid(16)](buf0,
primals_1, buf1, buf2, 16, XBLOCK=16, num_warps=1, num_stages=1)
return buf2, primals_1, primals_2, buf0, buf1, buf2
def gem(x, p=3, eps=1e-06):
return F.avg_pool2d(x.clamp(min=eps).pow(p), (x.size(-2), x.size(-1))).pow(
1.0 / p)
class GeMNew(nn.Module):
def __init__(self, p=3, eps=1e-06):
super(GeMNew, self).__init__()
self.p = Parameter(torch.ones(1) * p)
self.eps = eps
def __repr__(self):
return self.__class__.__name__ + '(' + 'p=' + '{:.4f}'.format(self.
p.data.tolist()[0]) + ', ' + 'eps=' + str(self.eps) + ')'
def forward(self, input_0):
primals_1 = self.p
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
| DrHB/PANDA-2nd-place-solution | GeM | false | 8,014 | [
"MIT"
] | 17 | 44ab72780f9e3594811f2a7520456de7a9e677db | https://github.com/DrHB/PANDA-2nd-place-solution/tree/44ab72780f9e3594811f2a7520456de7a9e677db | import torch
import torch.nn.functional as F
from torch import nn
from torch.nn.parameter import Parameter
import torch.optim
def gem(x, p=3, eps=1e-06):
return F.avg_pool2d(x.clamp(min=eps).pow(p), (x.size(-2), x.size(-1))).pow(
1.0 / p)
class Model(nn.Module):
def __init__(self, p=3, eps=1e-06):
super().__init__()
self.p = Parameter(torch.ones(1) * p)
self.eps = eps
def forward(self, x):
return gem(x, p=self.p, eps=self.eps)
def __repr__(self):
return self.__class__.__name__ + '(' + 'p=' + '{:.4f}'.format(self.
p.data.tolist()[0]) + ', ' + 'eps=' + str(self.eps) + ')'
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ScaledLeakyReLU | # 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_1/inductor_cache/5b/c5b7hvkiyuokw7mc3s2d6egjzhpcsd6nlzqh4xmcri6kpq3dp7qi.py
# Topologically Sorted Source Nodes: [out, mul], Original ATen: [aten.leaky_relu, aten.mul]
# Source node to ATen node mapping:
# mul => mul_1
# out => gt, mul, where
# Graph fragment:
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%arg0_1, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 0.2), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %arg0_1, %mul), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where, 1.4142135623730951), kwargs = {})
triton_poi_fused_leaky_relu_mul_0 = async_compile.triton('triton_poi_fused_leaky_relu_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_leaky_relu_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_leaky_relu_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.0
tmp2 = tmp0 > tmp1
tmp3 = 0.2
tmp4 = tmp0 * tmp3
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = 1.4142135623730951
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: [out, mul], Original ATen: [aten.leaky_relu, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_leaky_relu_mul_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
from torch import nn
import torch.nn.functional as F
class ScaledLeakyReLU(nn.Module):
def __init__(self, negative_slope=0.2):
super().__init__()
self.negative_slope = negative_slope
def forward(self, input):
out = F.leaky_relu(input, negative_slope=self.negative_slope)
return out * math.sqrt(2)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
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_leaky_relu_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.0
tmp2 = tmp0 > tmp1
tmp3 = 0.2
tmp4 = tmp0 * tmp3
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = 1.4142135623730951
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_leaky_relu_mul_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class ScaledLeakyReLUNew(nn.Module):
def __init__(self, negative_slope=0.2):
super().__init__()
self.negative_slope = negative_slope
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Dolorousrtur/style-people | ScaledLeakyReLU | false | 8,015 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | import math
import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, negative_slope=0.2):
super().__init__()
self.negative_slope = negative_slope
def forward(self, input):
out = F.leaky_relu(input, negative_slope=self.negative_slope)
return out * math.sqrt(2)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
SinkhornKnopp | # 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_1/inductor_cache/ot/cot3q6kdk327rc4nujhxu3jnaqb5yclmpeeo6mljfhxnobszysch.py
# Topologically Sorted Source Nodes: [sum_Q], Original ATen: [aten.sum]
# Source node to ATen node mapping:
# sum_Q => sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%permute,), kwargs = {})
triton_per_fused_sum_0 = async_compile.triton('triton_per_fused_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_sum_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 1
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp6 = tl.sum(tmp4, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/6i/c6iyv4mgrroqbu2xddobbrfubdn7cyt6htjdr4y5p3wjtd7k2g5o.py
# Topologically Sorted Source Nodes: [sum_of_rows], Original ATen: [aten.sum]
# Source node to ATen node mapping:
# sum_of_rows => sum_2
# Graph fragment:
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%permute_2, [1], True), kwargs = {})
triton_poi_fused_sum_1 = async_compile.triton('triton_poi_fused_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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sum_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sum_1(in_ptr0, in_ptr1, 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)
tmp4 = tl.load(in_ptr1 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp12 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp17 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp8 = tmp7 * tmp1
tmp9 = tl_math.exp(tmp8)
tmp10 = tmp9 / tmp5
tmp11 = tmp6 + tmp10
tmp13 = tmp12 * tmp1
tmp14 = tl_math.exp(tmp13)
tmp15 = tmp14 / tmp5
tmp16 = tmp11 + tmp15
tmp18 = tmp17 * tmp1
tmp19 = tl_math.exp(tmp18)
tmp20 = tmp19 / tmp5
tmp21 = tmp16 + tmp20
tl.store(out_ptr0 + (x0), tmp21, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bh/cbhdsf3pq3airew4hlmzhwihblmvjkeods3bzqnaanw6dghxs3n6.py
# Topologically Sorted Source Nodes: [sum_3], Original ATen: [aten.sum]
# Source node to ATen node mapping:
# sum_3 => sum_3
# Graph fragment:
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%permute_6, [0], True), kwargs = {})
triton_poi_fused_sum_2 = async_compile.triton('triton_poi_fused_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 9, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sum_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr2 + (0))
tmp8 = tl.broadcast_to(tmp7, [XBLOCK])
tmp12 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr2 + (1))
tmp17 = tl.broadcast_to(tmp16, [XBLOCK])
tmp21 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + (2))
tmp26 = tl.broadcast_to(tmp25, [XBLOCK])
tmp30 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp34 = tl.load(in_ptr2 + (3))
tmp35 = tl.broadcast_to(tmp34, [XBLOCK])
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp9 = tmp6 / tmp8
tmp10 = 0.25
tmp11 = tmp9 * tmp10
tmp13 = tmp12 * tmp1
tmp14 = tl_math.exp(tmp13)
tmp15 = tmp14 / tmp5
tmp18 = tmp15 / tmp17
tmp19 = tmp18 * tmp10
tmp20 = tmp11 + tmp19
tmp22 = tmp21 * tmp1
tmp23 = tl_math.exp(tmp22)
tmp24 = tmp23 / tmp5
tmp27 = tmp24 / tmp26
tmp28 = tmp27 * tmp10
tmp29 = tmp20 + tmp28
tmp31 = tmp30 * tmp1
tmp32 = tl_math.exp(tmp31)
tmp33 = tmp32 / tmp5
tmp36 = tmp33 / tmp35
tmp37 = tmp36 * tmp10
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + (x0), tmp38, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/sn/csnoh5a4abnlt3xezs2jmakqtsswrgeelx7z34kzrdihbw3mxhvi.py
# Topologically Sorted Source Nodes: [sum_of_rows_1], Original ATen: [aten.sum]
# Source node to ATen node mapping:
# sum_of_rows_1 => sum_4
# Graph fragment:
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%permute_10, [1], True), kwargs = {})
triton_poi_fused_sum_3 = async_compile.triton('triton_poi_fused_sum_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sum_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 10, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_sum_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp4 = tl.load(in_ptr1 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr2 + (x0), xmask)
tmp11 = tl.load(in_ptr3 + (0))
tmp12 = tl.broadcast_to(tmp11, [XBLOCK])
tmp15 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp21 = tl.load(in_ptr3 + (1))
tmp22 = tl.broadcast_to(tmp21, [XBLOCK])
tmp26 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp32 = tl.load(in_ptr3 + (2))
tmp33 = tl.broadcast_to(tmp32, [XBLOCK])
tmp37 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp43 = tl.load(in_ptr3 + (3))
tmp44 = tl.broadcast_to(tmp43, [XBLOCK])
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp8 = tmp6 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tmp13 = tmp10 / tmp12
tmp14 = tmp13 * tmp9
tmp16 = tmp15 * tmp1
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 / tmp5
tmp19 = tmp18 / tmp7
tmp20 = tmp19 * tmp9
tmp23 = tmp20 / tmp22
tmp24 = tmp23 * tmp9
tmp25 = tmp14 + tmp24
tmp27 = tmp26 * tmp1
tmp28 = tl_math.exp(tmp27)
tmp29 = tmp28 / tmp5
tmp30 = tmp29 / tmp7
tmp31 = tmp30 * tmp9
tmp34 = tmp31 / tmp33
tmp35 = tmp34 * tmp9
tmp36 = tmp25 + tmp35
tmp38 = tmp37 * tmp1
tmp39 = tl_math.exp(tmp38)
tmp40 = tmp39 / tmp5
tmp41 = tmp40 / tmp7
tmp42 = tmp41 * tmp9
tmp45 = tmp42 / tmp44
tmp46 = tmp45 * tmp9
tmp47 = tmp36 + tmp46
tl.store(out_ptr0 + (x0), tmp47, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ft/cft4foizpuibgcetwxkpfzz5hebn5pk7gatfdd5hbugkslahkyy6.py
# Topologically Sorted Source Nodes: [Q_7], Original ATen: [aten.div]
# Source node to ATen node mapping:
# Q_7 => div_7
# Graph fragment:
# %div_7 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%permute_12, 4), kwargs = {})
triton_poi_fused_div_4 = async_compile.triton('triton_poi_fused_div_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_div_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_4(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp4 = tl.load(in_ptr1 + (0))
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp8 = tmp6 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tmp12 = tmp10 / tmp11
tmp13 = tmp12 * tmp9
tmp15 = tmp13 / tmp14
tmp16 = tmp15 * tmp9
tl.store(out_ptr0 + (x2), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pj/cpjqhv3xbscdjmk5rngespx6fykcht6pgwkt2rkp5o6a7gw7hcfo.py
# Topologically Sorted Source Nodes: [Q_9], Original ATen: [aten.div]
# Source node to ATen node mapping:
# Q_9 => div_9
# Graph fragment:
# %div_9 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%permute_16, 4), kwargs = {})
triton_poi_fused_div_5 = async_compile.triton('triton_poi_fused_div_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_5(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gj/cgjjc657eblowghmtslpqitcqus576lhldw4vdt3ojzfrwylnqqk.py
# Topologically Sorted Source Nodes: [Q_11], Original ATen: [aten.div]
# Source node to ATen node mapping:
# Q_11 => div_11
# Graph fragment:
# %div_11 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%permute_20, 4), kwargs = {})
triton_poi_fused_div_6 = async_compile.triton('triton_poi_fused_div_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_div_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/xg/cxgjax3624cq5biejiwl7j3auut62dmpvt4ucszwvebhxgwnhbtz.py
# Topologically Sorted Source Nodes: [Q_14], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# Q_14 => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%permute_26, 4), kwargs = {})
triton_poi_fused_mul_7 = async_compile.triton('triton_poi_fused_mul_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, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_mul_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_7(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
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 + (4*x1)), xmask & ymask)
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
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tmp11 = 4.0
tmp12 = tmp10 * tmp11
tl.store(out_ptr0 + (x1 + (4*y0)), tmp12, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/3b/c3bzkt2ovtq24wsdmetecher526pusomuyzekrzacgnqvfttvoyh.py
# Topologically Sorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
# Graph fragment:
# %permute_31 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%permute_30, [1, 0]), kwargs = {})
triton_poi_fused_8 = async_compile.triton('triton_poi_fused_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=[4, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_8', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_8(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
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 + (4*x1)), xmask & ymask)
tl.store(out_ptr0 + (x1 + (4*y0)), 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, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [sum_Q], Original ATen: [aten.sum]
stream0 = get_raw_stream(0)
triton_per_fused_sum_0.run(arg0_1, buf0, 1, 16, grid=grid(1), stream=stream0)
buf1 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [sum_of_rows], Original ATen: [aten.sum]
triton_poi_fused_sum_1.run(arg0_1, buf0, buf1, 4, grid=grid(4), stream=stream0)
buf2 = empty_strided_cuda((1, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sum_3], Original ATen: [aten.sum]
triton_poi_fused_sum_2.run(arg0_1, buf0, buf1, buf2, 4, grid=grid(4), stream=stream0)
buf3 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [sum_of_rows_1], Original ATen: [aten.sum]
triton_poi_fused_sum_3.run(arg0_1, buf0, buf1, buf2, buf3, 4, grid=grid(4), stream=stream0)
buf4 = empty_strided_cuda((4, 4), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [Q_7], Original ATen: [aten.div]
triton_poi_fused_div_4.run(arg0_1, buf0, buf1, buf2, buf3, buf4, 16, grid=grid(16), stream=stream0)
del arg0_1
del buf0
del buf1
del buf2
del buf3
buf5 = empty_strided_cuda((4, 4), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [Q_9], Original ATen: [aten.div]
triton_poi_fused_div_5.run(buf4, buf5, 16, grid=grid(16), stream=stream0)
buf6 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [Q_11], Original ATen: [aten.div]
triton_poi_fused_div_6.run(buf5, buf6, 16, grid=grid(16), stream=stream0)
buf7 = reinterpret_tensor(buf5, (4, 4), (4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [Q_14], Original ATen: [aten.mul]
triton_poi_fused_mul_7.run(buf6, buf7, 4, 4, grid=grid(4, 4), stream=stream0)
buf8 = reinterpret_tensor(buf6, (4, 4), (4, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
triton_poi_fused_8.run(buf7, buf8, 4, 4, grid=grid(4, 4), stream=stream0)
del buf7
return (buf8, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4), (4, 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.distributed as dist
class SinkhornKnopp(torch.nn.Module):
def __init__(self, num_iters: 'int'=3, epsilon: 'float'=0.05,
world_size: 'int'=1):
"""Approximates optimal transport using the Sinkhorn-Knopp algorithm.
A simple iterative method to approach the double stochastic matrix is to alternately rescale
rows and columns of the matrix to sum to 1.
Args:
num_iters (int, optional): number of times to perform row and column normalization.
Defaults to 3.
epsilon (float, optional): weight for the entropy regularization term. Defaults to 0.05.
world_size (int, optional): number of nodes for distributed training. Defaults to 1.
"""
super().__init__()
self.num_iters = num_iters
self.epsilon = epsilon
self.world_size = world_size
@torch.no_grad()
def forward(self, Q: 'torch.Tensor') ->torch.Tensor:
"""Produces assignments using Sinkhorn-Knopp algorithm.
Applies the entropy regularization, normalizes the Q matrix and then normalizes rows and
columns in an alternating fashion for num_iter times. Before returning it normalizes again
the columns in order for the output to be an assignment of samples to prototypes.
Args:
Q (torch.Tensor): cosine similarities between the features of the
samples and the prototypes.
Returns:
torch.Tensor: assignment of samples to prototypes according to optimal transport.
"""
Q = torch.exp(Q / self.epsilon).t()
B = Q.shape[1] * self.world_size
K = Q.shape[0]
sum_Q = torch.sum(Q)
if dist.is_available() and dist.is_initialized():
dist.all_reduce(sum_Q)
Q /= sum_Q
for it in range(self.num_iters):
sum_of_rows = torch.sum(Q, dim=1, keepdim=True)
if dist.is_available() and dist.is_initialized():
dist.all_reduce(sum_of_rows)
Q /= sum_of_rows
Q /= K
Q /= torch.sum(Q, dim=0, keepdim=True)
Q /= B
Q *= B
return Q.t()
def get_inputs():
return [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.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_per_fused_sum_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK: tl.
constexpr):
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK, RBLOCK])
tmp6 = tl.sum(tmp4, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp6, None)
@triton.jit
def triton_poi_fused_sum_1(in_ptr0, in_ptr1, 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)
tmp4 = tl.load(in_ptr1 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp12 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp17 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp8 = tmp7 * tmp1
tmp9 = tl_math.exp(tmp8)
tmp10 = tmp9 / tmp5
tmp11 = tmp6 + tmp10
tmp13 = tmp12 * tmp1
tmp14 = tl_math.exp(tmp13)
tmp15 = tmp14 / tmp5
tmp16 = tmp11 + tmp15
tmp18 = tmp17 * tmp1
tmp19 = tl_math.exp(tmp18)
tmp20 = tmp19 / tmp5
tmp21 = tmp16 + tmp20
tl.store(out_ptr0 + x0, tmp21, xmask)
@triton.jit
def triton_poi_fused_sum_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr2 + 0)
tmp8 = tl.broadcast_to(tmp7, [XBLOCK])
tmp12 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp16 = tl.load(in_ptr2 + 1)
tmp17 = tl.broadcast_to(tmp16, [XBLOCK])
tmp21 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr2 + 2)
tmp26 = tl.broadcast_to(tmp25, [XBLOCK])
tmp30 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp34 = tl.load(in_ptr2 + 3)
tmp35 = tl.broadcast_to(tmp34, [XBLOCK])
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp9 = tmp6 / tmp8
tmp10 = 0.25
tmp11 = tmp9 * tmp10
tmp13 = tmp12 * tmp1
tmp14 = tl_math.exp(tmp13)
tmp15 = tmp14 / tmp5
tmp18 = tmp15 / tmp17
tmp19 = tmp18 * tmp10
tmp20 = tmp11 + tmp19
tmp22 = tmp21 * tmp1
tmp23 = tl_math.exp(tmp22)
tmp24 = tmp23 / tmp5
tmp27 = tmp24 / tmp26
tmp28 = tmp27 * tmp10
tmp29 = tmp20 + tmp28
tmp31 = tmp30 * tmp1
tmp32 = tl_math.exp(tmp31)
tmp33 = tmp32 / tmp5
tmp36 = tmp33 / tmp35
tmp37 = tmp36 * tmp10
tmp38 = tmp29 + tmp37
tl.store(out_ptr0 + x0, tmp38, xmask)
@triton.jit
def triton_poi_fused_sum_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp4 = tl.load(in_ptr1 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr2 + x0, xmask)
tmp11 = tl.load(in_ptr3 + 0)
tmp12 = tl.broadcast_to(tmp11, [XBLOCK])
tmp15 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp21 = tl.load(in_ptr3 + 1)
tmp22 = tl.broadcast_to(tmp21, [XBLOCK])
tmp26 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp32 = tl.load(in_ptr3 + 2)
tmp33 = tl.broadcast_to(tmp32, [XBLOCK])
tmp37 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp43 = tl.load(in_ptr3 + 3)
tmp44 = tl.broadcast_to(tmp43, [XBLOCK])
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp8 = tmp6 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tmp13 = tmp10 / tmp12
tmp14 = tmp13 * tmp9
tmp16 = tmp15 * tmp1
tmp17 = tl_math.exp(tmp16)
tmp18 = tmp17 / tmp5
tmp19 = tmp18 / tmp7
tmp20 = tmp19 * tmp9
tmp23 = tmp20 / tmp22
tmp24 = tmp23 * tmp9
tmp25 = tmp14 + tmp24
tmp27 = tmp26 * tmp1
tmp28 = tl_math.exp(tmp27)
tmp29 = tmp28 / tmp5
tmp30 = tmp29 / tmp7
tmp31 = tmp30 * tmp9
tmp34 = tmp31 / tmp33
tmp35 = tmp34 * tmp9
tmp36 = tmp25 + tmp35
tmp38 = tmp37 * tmp1
tmp39 = tl_math.exp(tmp38)
tmp40 = tmp39 / tmp5
tmp41 = tmp40 / tmp7
tmp42 = tmp41 * tmp9
tmp45 = tmp42 / tmp44
tmp46 = tmp45 * tmp9
tmp47 = tmp36 + tmp46
tl.store(out_ptr0 + x0, tmp47, xmask)
@triton.jit
def triton_poi_fused_div_4(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp4 = tl.load(in_ptr1 + 0)
tmp5 = tl.broadcast_to(tmp4, [XBLOCK])
tmp7 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp1 = 20.0
tmp2 = tmp0 * tmp1
tmp3 = tl_math.exp(tmp2)
tmp6 = tmp3 / tmp5
tmp8 = tmp6 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tmp12 = tmp10 / tmp11
tmp13 = tmp12 * tmp9
tmp15 = tmp13 / tmp14
tmp16 = tmp15 * tmp9
tl.store(out_ptr0 + x2, tmp16, xmask)
@triton.jit
def triton_poi_fused_div_5(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_div_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
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_mul_7(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 4
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 + 4 * x1), xmask & ymask)
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
tmp9 = 0.25
tmp10 = tmp8 * tmp9
tmp11 = 4.0
tmp12 = tmp10 * tmp11
tl.store(out_ptr0 + (x1 + 4 * y0), tmp12, xmask & ymask)
@triton.jit
def triton_poi_fused_8(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 4
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 + 4 * x1), xmask & ymask)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
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((), (), torch.float32)
get_raw_stream(0)
triton_per_fused_sum_0[grid(1)](arg0_1, buf0, 1, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf1 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
triton_poi_fused_sum_1[grid(4)](arg0_1, buf0, buf1, 4, XBLOCK=4,
num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((1, 4), (4, 1), torch.float32)
triton_poi_fused_sum_2[grid(4)](arg0_1, buf0, buf1, buf2, 4, XBLOCK
=4, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
triton_poi_fused_sum_3[grid(4)](arg0_1, buf0, buf1, buf2, buf3, 4,
XBLOCK=4, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4), (1, 4), torch.float32)
triton_poi_fused_div_4[grid(16)](arg0_1, buf0, buf1, buf2, buf3,
buf4, 16, XBLOCK=16, num_warps=1, num_stages=1)
del arg0_1
del buf0
del buf1
del buf2
del buf3
buf5 = empty_strided_cuda((4, 4), (1, 4), torch.float32)
triton_poi_fused_div_5[grid(16)](buf4, buf5, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf6 = buf4
del buf4
triton_poi_fused_div_6[grid(16)](buf5, buf6, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf7 = reinterpret_tensor(buf5, (4, 4), (4, 1), 0)
del buf5
triton_poi_fused_mul_7[grid(4, 4)](buf6, buf7, 4, 4, XBLOCK=4,
YBLOCK=4, num_warps=1, num_stages=1)
buf8 = reinterpret_tensor(buf6, (4, 4), (4, 1), 0)
del buf6
triton_poi_fused_8[grid(4, 4)](buf7, buf8, 4, 4, XBLOCK=4, YBLOCK=4,
num_warps=1, num_stages=1)
del buf7
return buf8,
class SinkhornKnoppNew(torch.nn.Module):
def __init__(self, num_iters: 'int'=3, epsilon: 'float'=0.05,
world_size: 'int'=1):
"""Approximates optimal transport using the Sinkhorn-Knopp algorithm.
A simple iterative method to approach the double stochastic matrix is to alternately rescale
rows and columns of the matrix to sum to 1.
Args:
num_iters (int, optional): number of times to perform row and column normalization.
Defaults to 3.
epsilon (float, optional): weight for the entropy regularization term. Defaults to 0.05.
world_size (int, optional): number of nodes for distributed training. Defaults to 1.
"""
super().__init__()
self.num_iters = num_iters
self.epsilon = epsilon
self.world_size = world_size
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DonkeyShot21/cassle | SinkhornKnopp | false | 8,016 | [
"MIT"
] | 13 | d25f9c7cb5e822660dc1ef03e7fac09a33d0b1a8 | https://github.com/DonkeyShot21/cassle/tree/d25f9c7cb5e822660dc1ef03e7fac09a33d0b1a8 | import torch
import torch.distributed as dist
class Model(torch.nn.Module):
def __init__(self, num_iters: 'int'=3, epsilon: 'float'=0.05,
world_size: 'int'=1):
"""Approximates optimal transport using the Sinkhorn-Knopp algorithm.
A simple iterative method to approach the double stochastic matrix is to alternately rescale
rows and columns of the matrix to sum to 1.
Args:
num_iters (int, optional): number of times to perform row and column normalization.
Defaults to 3.
epsilon (float, optional): weight for the entropy regularization term. Defaults to 0.05.
world_size (int, optional): number of nodes for distributed training. Defaults to 1.
"""
super().__init__()
self.num_iters = num_iters
self.epsilon = epsilon
self.world_size = world_size
@torch.no_grad()
def forward(self, Q: 'torch.Tensor') ->torch.Tensor:
"""Produces assignments using Sinkhorn-Knopp algorithm.
Applies the entropy regularization, normalizes the Q matrix and then normalizes rows and
columns in an alternating fashion for num_iter times. Before returning it normalizes again
the columns in order for the output to be an assignment of samples to prototypes.
Args:
Q (torch.Tensor): cosine similarities between the features of the
samples and the prototypes.
Returns:
torch.Tensor: assignment of samples to prototypes according to optimal transport.
"""
Q = torch.exp(Q / self.epsilon).t()
B = Q.shape[1] * self.world_size
K = Q.shape[0]
sum_Q = torch.sum(Q)
if dist.is_available() and dist.is_initialized():
dist.all_reduce(sum_Q)
Q /= sum_Q
for it in range(self.num_iters):
sum_of_rows = torch.sum(Q, dim=1, keepdim=True)
if dist.is_available() and dist.is_initialized():
dist.all_reduce(sum_of_rows)
Q /= sum_of_rows
Q /= K
Q /= torch.sum(Q, dim=0, keepdim=True)
Q /= B
Q *= B
return Q.t()
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return []
|
HighwayLayer | # 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_1/inductor_cache/tz/ctznu4we6tcnccl44eq2h2ckxwd5kj64dzxpr2vge672k2gfve57.py
# Topologically Sorted Source Nodes: [gate, nlin, mul, sub, mul_1, res], Original ATen: [aten.sigmoid, aten.tanh, aten.mul, aten.rsub, aten.add]
# Source node to ATen node mapping:
# gate => sigmoid
# mul => mul
# mul_1 => mul_1
# nlin => tanh
# res => add
# sub => sub
# Graph fragment:
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_1,), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%view_3,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %tanh), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %primals_3), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
triton_poi_fused_add_mul_rsub_sigmoid_tanh_0 = async_compile.triton('triton_poi_fused_add_mul_rsub_sigmoid_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_rsub_sigmoid_tanh_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_rsub_sigmoid_tanh_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask)
tmp7 = tl.load(in_ptr2 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = libdevice.tanh(tmp2)
tmp4 = tmp1 * tmp3
tmp5 = 1.0
tmp6 = tmp5 - tmp1
tmp8 = tmp6 * tmp7
tmp9 = tmp4 + tmp8
tl.store(out_ptr0 + (x0), tmp9, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = 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((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_3, (64, 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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gate, nlin, mul, sub, mul_1, res], Original ATen: [aten.sigmoid, aten.tanh, aten.mul, aten.rsub, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_rsub_sigmoid_tanh_0.run(buf0, buf1, primals_3, buf2, 256, grid=grid(256), stream=stream0)
return (buf2, primals_3, buf0, buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
import torch.utils.data
import torch.utils.data.distributed
import torch.utils.checkpoint
def my_xavier_init(m, gain=1):
"""Xavier initialization: weights initialization that tries to make variance of outputs
of a layer equal to variance of its inputs.
"""
for p in m.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p, gain)
else:
nn.init.constant_(p, 0)
class HighwayLayer(torch.nn.Module):
"""Highway transformation used in span prediction."""
def __init__(self, dim):
super(HighwayLayer, self).__init__()
self.gate_proj = nn.Linear(dim, dim, bias=True)
self.nlin_proj = nn.Linear(dim, dim, bias=True)
my_xavier_init(self.nlin_proj)
my_xavier_init(self.gate_proj)
nn.init.constant_(self.gate_proj.bias, -1)
def forward(self, x):
gate = torch.sigmoid(self.gate_proj(x))
nlin = torch.tanh(self.nlin_proj(x))
res = gate * nlin + (1 - gate) * x
return res
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
import torch.utils.data
import torch.utils.data.distributed
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_mul_rsub_sigmoid_tanh_0(in_ptr0, in_ptr1, in_ptr2,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask)
tmp7 = tl.load(in_ptr2 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = libdevice.tanh(tmp2)
tmp4 = tmp1 * tmp3
tmp5 = 1.0
tmp6 = tmp5 - tmp1
tmp8 = tmp6 * tmp7
tmp9 = tmp4 + tmp8
tl.store(out_ptr0 + x0, tmp9, 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((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_3, (64,
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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mul_rsub_sigmoid_tanh_0[grid(256)](buf0, buf1,
primals_3, buf2, 256, XBLOCK=128, num_warps=4, num_stages=1)
return buf2, primals_3, buf0, buf1
def my_xavier_init(m, gain=1):
"""Xavier initialization: weights initialization that tries to make variance of outputs
of a layer equal to variance of its inputs.
"""
for p in m.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p, gain)
else:
nn.init.constant_(p, 0)
class HighwayLayerNew(torch.nn.Module):
"""Highway transformation used in span prediction."""
def __init__(self, dim):
super(HighwayLayerNew, self).__init__()
self.gate_proj = nn.Linear(dim, dim, bias=True)
self.nlin_proj = nn.Linear(dim, dim, bias=True)
my_xavier_init(self.nlin_proj)
my_xavier_init(self.gate_proj)
nn.init.constant_(self.gate_proj.bias, -1)
def forward(self, input_0):
primals_1 = self.gate_proj.weight
primals_2 = self.gate_proj.bias
primals_4 = self.nlin_proj.weight
primals_5 = self.nlin_proj.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| CowherdChris/droidlet | HighwayLayer | false | 8,017 | [
"MIT"
] | 26 | 8d965c1ebc38eceb6f8083c52b1146c1bc17d5e1 | https://github.com/CowherdChris/droidlet/tree/8d965c1ebc38eceb6f8083c52b1146c1bc17d5e1 | import torch
from torch import nn
import torch.utils.data
import torch.utils.data.distributed
import torch.utils.checkpoint
def my_xavier_init(m, gain=1):
"""Xavier initialization: weights initialization that tries to make variance of outputs
of a layer equal to variance of its inputs.
"""
for p in m.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p, gain)
else:
nn.init.constant_(p, 0)
class Model(torch.nn.Module):
"""Highway transformation used in span prediction."""
def __init__(self, dim):
super().__init__()
self.gate_proj = nn.Linear(dim, dim, bias=True)
self.nlin_proj = nn.Linear(dim, dim, bias=True)
my_xavier_init(self.nlin_proj)
my_xavier_init(self.gate_proj)
nn.init.constant_(self.gate_proj.bias, -1)
def forward(self, x):
gate = torch.sigmoid(self.gate_proj(x))
nlin = torch.tanh(self.nlin_proj(x))
res = gate * nlin + (1 - gate) * x
return res
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
PostGCN | # 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_1/inductor_cache/qo/cqohgshabkxbfecv5yze7hpj7fgq55zcx3d52r5tyzac2rqzwjbv.py
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# output => 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/yy/cyyqxa7owbligvibo6shuiwpo6qn3y5r4hoxta4khaadodv53dfs.py
# Topologically Sorted Source Nodes: [output, y], Original ATen: [aten.clone, aten.add]
# Source node to ATen node mapping:
# output => clone_1
# y => add
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_2,), kwargs = {memory_format: torch.contiguous_format})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%clone_1, %primals_4), kwargs = {})
triton_poi_fused_add_clone_1 = async_compile.triton('triton_poi_fused_add_clone_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_clone_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4*y3)), tmp2, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (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: [support], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_2, (64, 4), (4, 1), 0), primals_1, out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, buf1, 64, 4, grid=grid(64, 4), stream=stream0)
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output, y], Original ATen: [aten.clone, aten.add]
triton_poi_fused_add_clone_1.run(buf2, primals_4, buf3, 64, 4, grid=grid(64, 4), stream=stream0)
del buf2
del primals_4
return (buf3, reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_3, reinterpret_tensor(primals_2, (4, 64), (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, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (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 math
import torch
import torch.nn as nn
from torch.nn.parameter import Parameter
class GraphConvolution(nn.Module):
"""
adapted from : https://github.com/tkipf/gcn/blob/92600c39797c2bfb61a508e52b88fb554df30177/gcn/layers.py#L132
"""
def __init__(self, in_features, out_features, bias=True, node_n=48):
super(GraphConvolution, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.weight = Parameter(torch.FloatTensor(in_features, out_features))
self.att = Parameter(torch.FloatTensor(node_n, node_n))
if bias:
self.bias = Parameter(torch.FloatTensor(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.weight.data.uniform_(-stdv, stdv)
self.att.data.uniform_(-stdv, stdv)
if self.bias is not None:
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
support = torch.matmul(input, self.weight)
output = torch.matmul(self.att, support)
if self.bias is not None:
return output + self.bias
else:
return output
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class PostGCN(nn.Module):
def __init__(self, input_feature, hidden_feature, node_n):
super(PostGCN, self).__init__()
self.input_feature = input_feature
self.hidden_feature = hidden_feature
self.node_n = node_n
self.gcn = GraphConvolution(input_feature, hidden_feature, node_n=
node_n)
def forward(self, x):
y = self.gcn(x)
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_feature': 4, 'hidden_feature': 4, 'node_n': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import math
import torch.nn as nn
from torch.nn.parameter 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_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 % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4 * y3), tmp2, xmask & ymask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (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_2, (64, 4), (4, 1), 0),
primals_1, out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64, 4)](buf0, buf1, 64, 4, XBLOCK=4,
YBLOCK=32, num_warps=4, num_stages=1)
buf2 = buf0
del buf0
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_clone_1[grid(64, 4)](buf2, primals_4, buf3, 64,
4, XBLOCK=4, YBLOCK=64, num_warps=4, num_stages=1)
del buf2
del primals_4
return buf3, reinterpret_tensor(buf1, (64, 4), (4, 1), 0
), primals_3, reinterpret_tensor(primals_2, (4, 64), (1, 4), 0)
class GraphConvolution(nn.Module):
"""
adapted from : https://github.com/tkipf/gcn/blob/92600c39797c2bfb61a508e52b88fb554df30177/gcn/layers.py#L132
"""
def __init__(self, in_features, out_features, bias=True, node_n=48):
super(GraphConvolution, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.weight = Parameter(torch.FloatTensor(in_features, out_features))
self.att = Parameter(torch.FloatTensor(node_n, node_n))
if bias:
self.bias = Parameter(torch.FloatTensor(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.weight.data.uniform_(-stdv, stdv)
self.att.data.uniform_(-stdv, stdv)
if self.bias is not None:
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
support = torch.matmul(input, self.weight)
output = torch.matmul(self.att, support)
if self.bias is not None:
return output + self.bias
else:
return output
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class PostGCNNew(nn.Module):
def __init__(self, input_feature, hidden_feature, node_n):
super(PostGCNNew, self).__init__()
self.input_feature = input_feature
self.hidden_feature = hidden_feature
self.node_n = node_n
self.gcn = GraphConvolution(input_feature, hidden_feature, node_n=
node_n)
def forward(self, input_0):
primals_1 = self.gcn.weight
primals_3 = self.gcn.att
primals_4 = self.gcn.bias
primals_2 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| Droliven/MSRGCN | PostGCN | false | 8,018 | [
"MIT"
] | 28 | 5d8d8e3365d3b23ca2ac734ace7e84135a6e3a9e | https://github.com/Droliven/MSRGCN/tree/5d8d8e3365d3b23ca2ac734ace7e84135a6e3a9e | import math
import torch
import torch.nn as nn
from torch.nn.parameter import Parameter
class GraphConvolution(nn.Module):
"""
adapted from : https://github.com/tkipf/gcn/blob/92600c39797c2bfb61a508e52b88fb554df30177/gcn/layers.py#L132
"""
def __init__(self, in_features, out_features, bias=True, node_n=48):
super().__init__()
self.in_features = in_features
self.out_features = out_features
self.weight = Parameter(torch.FloatTensor(in_features, out_features))
self.att = Parameter(torch.FloatTensor(node_n, node_n))
if bias:
self.bias = Parameter(torch.FloatTensor(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.weight.data.uniform_(-stdv, stdv)
self.att.data.uniform_(-stdv, stdv)
if self.bias is not None:
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
support = torch.matmul(input, self.weight)
output = torch.matmul(self.att, support)
if self.bias is not None:
return output + self.bias
else:
return output
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features
) + ' -> ' + str(self.out_features) + ')'
class Model(nn.Module):
def __init__(self, input_feature, hidden_feature, node_n):
super().__init__()
self.input_feature = input_feature
self.hidden_feature = hidden_feature
self.node_n = node_n
self.gcn = GraphConvolution(input_feature, hidden_feature, node_n=
node_n)
def forward(self, x):
y = self.gcn(x)
return y
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
DiscountBlackLoss | # 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_1/inductor_cache/3e/c3erk7rja6zjyzz2d55ojwzz7i2xrwgrg2i7wjxozpegdsshjth3.py
# Topologically Sorted Source Nodes: [mask, setitem, setitem_1], Original ATen: [aten.sum, aten.lift_fresh, aten.index_put]
# Source node to ATen node mapping:
# mask => sum_1
# setitem => full_default, index_put
# setitem_1 => full_default_1, index_put_1
# Graph fragment:
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%arg0_1, [1], True), kwargs = {})
# %full_default : [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 : [num_users=2] = call_function[target=torch.ops.aten.index_put_.default](args = (%sum_1, [%gt], %full_default), kwargs = {})
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 9.999999974752427e-07), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %index_put_1 : [num_users=2] = call_function[target=torch.ops.aten.index_put_.default](args = (%index_put, [%eq], %full_default_1), kwargs = {})
triton_poi_fused_index_put_lift_fresh_sum_0 = async_compile.triton('triton_poi_fused_index_put_lift_fresh_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_sum_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.0
tmp8 = tmp6 > tmp7
tmp9 = 1.0
tmp10 = tl.where(tmp8, tmp9, tmp6)
tmp11 = tmp10 == tmp7
tmp12 = 9.999999974752427e-07
tmp13 = tl.where(tmp11, tmp12, tmp10)
tl.store(in_out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vg/cvgsgpniipq5ap3aq6f6clznbei5yb5dpps3stoga4rjyyjvn7yj.py
# Topologically Sorted Source Nodes: [loss, loss_1, mean], Original ATen: [aten.mse_loss, aten.mul, aten.mean]
# Source node to ATen node mapping:
# loss => pow_1, sub
# loss_1 => mul
# mean => mean
# 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 = {})
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_1, %index_put_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul,), kwargs = {})
triton_per_fused_mean_mse_loss_mul_1 = async_compile.triton('triton_per_fused_mean_mse_loss_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=[1, 256],
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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {5: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 6), equal_to_1=(5,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mean_mse_loss_mul_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 3, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mse_loss_mul_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, rnumel):
xnumel = 1
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r3 = rindex
r0 = rindex % 16
r2 = (rindex // 64)
tmp0 = tl.load(in_ptr0 + (r3), None)
tmp1 = tl.load(in_ptr1 + (r3), None)
tmp4 = tl.load(in_ptr2 + (r0 + (16*r2)), None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp5 = tmp3 * tmp4
tmp6 = tl.broadcast_to(tmp5, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = 256.0
tmp10 = tmp8 / tmp9
tl.store(out_ptr0 + (tl.broadcast_to(r3, [RBLOCK])), tmp5, None)
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((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf1 = reinterpret_tensor(buf0, (4, 1, 4, 4), (16, 16, 4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [mask, setitem, setitem_1], Original ATen: [aten.sum, aten.lift_fresh, aten.index_put]
stream0 = get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_sum_0.run(buf1, arg0_1, 64, grid=grid(64), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [loss, loss_1, mean], Original ATen: [aten.mse_loss, aten.mul, aten.mean]
triton_per_fused_mean_mse_loss_mul_1.run(buf4, arg1_1, arg0_1, buf1, buf2, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
return (buf4, buf2, 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
import torch.utils.data
import torch.random
import torch.nn.functional as F
class DiscountBlackLoss(nn.Module):
def __init__(self):
super().__init__()
def forward(self, source, target):
"""
Ignores pixels with all channels set to zero
Probably Not suitable for greyscale images
:param source: image from network
:param target: ground truth
:return: the loss
"""
loss = F.mse_loss(source, target, reduction='none')
mask = torch.sum(target, dim=1, keepdim=True)
mask[mask > 0.0] = 1.0
mask[mask == 0.0] = 1e-06
loss = loss * mask
return torch.mean(loss), loss, mask
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
import torch.utils.data
import torch.random
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_index_put_lift_fresh_sum_0(in_out_ptr0, in_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.0
tmp8 = tmp6 > tmp7
tmp9 = 1.0
tmp10 = tl.where(tmp8, tmp9, tmp6)
tmp11 = tmp10 == tmp7
tmp12 = 9.999999974752427e-07
tmp13 = tl.where(tmp11, tmp12, tmp10)
tl.store(in_out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_per_fused_mean_mse_loss_mul_1(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, out_ptr0, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r3 = rindex
r0 = rindex % 16
r2 = rindex // 64
tmp0 = tl.load(in_ptr0 + r3, None)
tmp1 = tl.load(in_ptr1 + r3, None)
tmp4 = tl.load(in_ptr2 + (r0 + 16 * r2), None, eviction_policy='evict_last'
)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp5 = tmp3 * tmp4
tmp6 = tl.broadcast_to(tmp5, [RBLOCK])
tmp8 = triton_helpers.promote_to_tensor(tl.sum(tmp6, 0))
tmp9 = 256.0
tmp10 = tmp8 / tmp9
tl.store(out_ptr0 + tl.broadcast_to(r3, [RBLOCK]), tmp5, None)
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((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf1 = reinterpret_tensor(buf0, (4, 1, 4, 4), (16, 16, 4, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_sum_0[grid(64)](buf1, arg0_1,
64, XBLOCK=64, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf3 = empty_strided_cuda((), (), torch.float32)
buf4 = buf3
del buf3
triton_per_fused_mean_mse_loss_mul_1[grid(1)](buf4, arg1_1, arg0_1,
buf1, buf2, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf4, buf2, buf1
class DiscountBlackLossNew(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0], output[1], output[2]
| DuaneNielsen/keypoints | DiscountBlackLoss | false | 8,019 | [
"MIT"
] | 42 | 302fa02966d4372ac9b5aaa3d8dc24684be0b252 | https://github.com/DuaneNielsen/keypoints/tree/302fa02966d4372ac9b5aaa3d8dc24684be0b252 | import torch
import torch.nn as nn
import torch.utils.data
import torch.random
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, source, target):
"""
Ignores pixels with all channels set to zero
Probably Not suitable for greyscale images
:param source: image from network
:param target: ground truth
:return: the loss
"""
loss = F.mse_loss(source, target, reduction='none')
mask = torch.sum(target, dim=1, keepdim=True)
mask[mask > 0.0] = 1.0
mask[mask == 0.0] = 1e-06
loss = loss * mask
return torch.mean(loss), loss, mask
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
EqualConv2dSame | # 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_1/inductor_cache/kl/cklozoj2jqpjpxiw63467cdardq77ypqaelvulktzmextpwu3xri.py
# Topologically Sorted Source Nodes: [input_1], Original ATen: [aten.replication_pad2d]
# Source node to ATen node mapping:
# input_1 => _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, %clamp_max, None]), kwargs = {})
# %_unsafe_index_1 : [num_users=2] = call_function[target=torch.ops.aten._unsafe_index.Tensor](args = (%_unsafe_index, [None, None, None, %clamp_max]), kwargs = {})
triton_poi_fused_replication_pad2d_0 = async_compile.triton('triton_poi_fused_replication_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_replication_pad2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_replication_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 + ((4*((3) * ((3) <= (x1)) + (x1) * ((x1) < (3)))) + (16*x2) + ((3) * ((3) <= (x0)) + (x0) * ((x0) < (3)))), xmask)
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kn/cknia4ynremi77dthoc4ot5wwbcximcfom4jnkidvm7x6dzwuynh.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, 0.125), kwargs = {})
triton_poi_fused_mul_1 = async_compile.triton('triton_poi_fused_mul_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), 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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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 = 0.125
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/x7/cx7evvvm7te22h7xf3yh7pnjatqie5vy54vyorfffrtctztd4wn5.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# out => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%_unsafe_index_1, %mul, %primals_3, [1, 1], [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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 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: [input_1], Original ATen: [aten.replication_pad2d]
stream0 = get_raw_stream(0)
triton_poi_fused_replication_pad2d_0.run(primals_1, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
triton_poi_fused_mul_1.run(primals_2, buf1, 256, grid=grid(256), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf0, buf1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 1, 1), (4, 1, 1, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf3, primals_3, 16, grid=grid(16), stream=stream0)
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, 4, 4, 4), (64, 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)
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 math
import torch
from torch import nn
import torch.nn.functional as F
class EqualConv2dSame(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, stride=1,
padding=0, bias=True):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_channel, in_channel,
kernel_size, kernel_size))
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
def forward(self, input):
pad = self.padding
input = F.pad(input, (pad, pad, pad, pad), mode='replicate')
out = F.conv2d(input, self.weight * self.scale, bias=self.bias,
stride=self.stride)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]}, {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})'
)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'out_channel': 4, 'kernel_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import 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
@triton.jit
def triton_poi_fused_replication_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 + (4 * (3 * (3 <= x1) + x1 * (x1 < 3)) + 16 * x2 +
(3 * (3 <= x0) + x0 * (x0 < 3))), xmask)
tl.store(out_ptr0 + x3, tmp0, xmask)
@triton.jit
def triton_poi_fused_mul_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 = 0.125
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_convolution_2(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
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 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_replication_pad2d_0[grid(256)](primals_1, buf0,
256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_1[grid(256)](primals_2, buf1, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(buf0, buf1, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 1, 1), (4, 1, 1, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_2[grid(16)](buf3, primals_3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
return buf3, buf0, buf1
class EqualConv2dSameNew(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, stride=1,
padding=0, bias=True):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_channel, in_channel,
kernel_size, kernel_size))
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]}, {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})'
)
def forward(self, input_0):
primals_1 = self.weight
primals_3 = self.bias
primals_2 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| Dolorousrtur/style-people | EqualConv2dSame | false | 8,020 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | import math
import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, stride=1,
padding=0, bias=True):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_channel, in_channel,
kernel_size, kernel_size))
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
def forward(self, input):
pad = self.padding
input = F.pad(input, (pad, pad, pad, pad), mode='replicate')
out = F.conv2d(input, self.weight * self.scale, bias=self.bias,
stride=self.stride)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]}, {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})'
)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
EqualLinear | # 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_1/inductor_cache/ag/caggqh3nrqirp3azykpm6daivhojwggbhob6pnubsheittrjeiky.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, 0.5), 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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qw/cqw2ahskucrmwkbadqswj76db4kfcxn5w76kji7peia7kwu4zbkq.py
# Topologically Sorted Source Nodes: [mul_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_1 => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, 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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, ), (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((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(primals_1, buf0, 16, grid=grid(16), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mul_1], Original ATen: [aten.mul]
triton_poi_fused_mul_1.run(primals_2, buf1, 4, grid=grid(4), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_1, out], Original ATen: [aten.mul, aten.addmm]
extern_kernels.addmm(buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf0
del buf1
return (reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_dim': 4, 'out_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.autograd import Function
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_mul_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
tl.store(out_ptr0 + x0, tmp2, 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,), (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((4, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(16)](primals_1, buf0, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
triton_poi_fused_mul_1[grid(4)](primals_2, buf1, 4, XBLOCK=4,
num_warps=1, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(buf1, reinterpret_tensor(primals_3, (64, 4), (
4, 1), 0), reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1,
beta=1, out=buf2)
del buf0
del buf1
return reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0)
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinearNew(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
def forward(self, input_0):
primals_1 = self.weight
primals_2 = self.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| Dolorousrtur/style-people | EqualLinear | false | 8,021 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class Model(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
ModulatedSiren2d | # 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_1/inductor_cache/ag/caggqh3nrqirp3azykpm6daivhojwggbhob6pnubsheittrjeiky.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, 0.5), 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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qw/cqw2ahskucrmwkbadqswj76db4kfcxn5w76kji7peia7kwu4zbkq.py
# Topologically Sorted Source Nodes: [mul_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_1 => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_3, 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=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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')
# kernel path: runs/run_shard_1/inductor_cache/lo/clooq2sxzea2fl7hxiqrjcdxc7anstn74sfzrmxvn6fvyetbqgas.py
# Topologically Sorted Source Nodes: [mul_2, weight, pow_1, sum_1, mul_4, add, demod], Original ATen: [aten.mul, aten.pow, aten.sum, aten.add, aten.rsqrt]
# Source node to ATen node mapping:
# add => add
# demod => rsqrt
# mul_2 => mul_2
# mul_4 => mul_4
# pow_1 => pow_1
# sum_1 => sum_1
# weight => mul_3
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_5, 0.5), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %view), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%mul_3, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [2, 3, 4]), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 450.0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, 1e-08), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_add_mul_pow_rsqrt_sum_2 = async_compile.triton('triton_poi_fused_add_mul_pow_rsqrt_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_pow_rsqrt_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_rsqrt_sum_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp5 = tmp4 * tmp4
tmp7 = tmp6 * tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp5 + tmp10
tmp13 = tmp12 * tmp1
tmp15 = tmp13 * tmp14
tmp16 = tmp15 * tmp15
tmp17 = tmp11 + tmp16
tmp19 = tmp18 * tmp1
tmp21 = tmp19 * tmp20
tmp22 = tmp21 * tmp21
tmp23 = tmp17 + tmp22
tmp24 = 450.0
tmp25 = tmp23 * tmp24
tmp26 = 1e-08
tmp27 = tmp25 + tmp26
tmp28 = libdevice.rsqrt(tmp27)
tl.store(out_ptr0 + (x2), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kb/ckbodo65ksrjdz7utqztuporiv5akgtcmyczja2zigsos2ayoxlr.py
# Topologically Sorted Source Nodes: [mul_2, weight, weight_1], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul_2 => mul_2
# weight => mul_3
# weight_1 => mul_5
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_5, 0.5), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %view), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %view_1), kwargs = {})
triton_poi_fused_mul_3 = async_compile.triton('triton_poi_fused_mul_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_3(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
x3 = xindex % 16
x0 = xindex % 4
x2 = (xindex // 16)
x4 = (xindex // 4)
x5 = xindex
tmp0 = tl.load(in_ptr0 + (x3), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tl.store(out_ptr0 + (x5), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bn/cbnyt3sqpf3bwvqfujmzzmtpazilphbcmuz7kwqc6x4xtppgbc5n.py
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# out_3 => mul_6
# Graph fragment:
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_4, 30), 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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1), 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': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, 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)
tmp1 = 30.0
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 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, (1, 4, 4, 1, 1), (16, 4, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_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_1], 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_1, out], Original ATen: [aten.mul, aten.addmm]
extern_kernels.addmm(buf1, primals_4, reinterpret_tensor(buf0, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul_2, weight, pow_1, sum_1, mul_4, add, demod], Original ATen: [aten.mul, aten.pow, aten.sum, aten.add, aten.rsqrt]
triton_poi_fused_add_mul_pow_rsqrt_sum_2.run(primals_5, buf2, buf3, 16, grid=grid(16), stream=stream0)
buf4 = empty_strided_cuda((4, 4, 4, 1, 1), (16, 4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_2, weight, weight_1], Original ATen: [aten.mul]
triton_poi_fused_mul_3.run(primals_5, buf2, buf3, buf4, 64, grid=grid(64), stream=stream0)
del buf3
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf4, (16, 4, 1, 1), (4, 1, 0, 0), 0), stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=4, bias=None)
assert_size_stride(buf5, (1, 16, 4, 4), (256, 16, 4, 1))
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.mul]
triton_poi_fused_mul_4.run(buf6, 256, grid=grid(256), stream=stream0)
return (buf6, primals_4, primals_5, buf2, reinterpret_tensor(buf4, (16, 4, 1, 1), (4, 1, 1, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4, 4), (256, 16, 4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((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((1, 4, 4, 1, 1), (16, 4, 1, 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)
| from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class ModulatedSiren2d(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30):
super().__init__()
self.eps = 1e-08
self.kernel_size = 1
self.in_channel = in_channel
self.out_channel = out_channel
fan_in = in_channel
self.scale = 1 / math.sqrt(fan_in)
if is_first:
self.demod_scale = 3 * fan_in
else:
self.demod_scale = omega_0 ** 2 / 2
self.omega0 = omega_0
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
1, 1))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel})'
)
def forward(self, input, style):
batch, in_channel, height, width = input.shape
style = self.modulation(style)
style = style.view(batch, 1, in_channel, 1, 1)
weight = self.scale * self.weight * style
if self.demodulate:
demod = torch.rsqrt(weight.pow(2).sum([2, 3, 4]) * self.
demod_scale + 1e-08)
weight = weight * demod.view(batch, self.out_channel, 1, 1, 1)
weight = weight.view(batch * self.out_channel, in_channel, 1, 1)
input = input.view(1, batch * in_channel, height, width)
out = F.conv2d(input, weight, groups=batch)
_, _, height, width = out.shape
out = out.view(batch, self.out_channel, height, width)
out = out * self.omega0
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'out_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
from torch.autograd import Function
import 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_mul_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
tl.store(out_ptr0 + x0, tmp2, 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)
@triton.jit
def triton_poi_fused_add_mul_pow_rsqrt_sum_2(in_ptr0, in_ptr1, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp14 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp18 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp20 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp5 = tmp4 * tmp4
tmp7 = tmp6 * tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp9 * tmp9
tmp11 = tmp5 + tmp10
tmp13 = tmp12 * tmp1
tmp15 = tmp13 * tmp14
tmp16 = tmp15 * tmp15
tmp17 = tmp11 + tmp16
tmp19 = tmp18 * tmp1
tmp21 = tmp19 * tmp20
tmp22 = tmp21 * tmp21
tmp23 = tmp17 + tmp22
tmp24 = 450.0
tmp25 = tmp23 * tmp24
tmp26 = 1e-08
tmp27 = tmp25 + tmp26
tmp28 = libdevice.rsqrt(tmp27)
tl.store(out_ptr0 + x2, tmp28, xmask)
@triton.jit
def triton_poi_fused_mul_3(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
x3 = xindex % 16
x0 = xindex % 4
x2 = xindex // 16
x4 = xindex // 4
x5 = xindex
tmp0 = tl.load(in_ptr0 + x3, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tl.store(out_ptr0 + x5, tmp6, xmask)
@triton.jit
def triton_poi_fused_mul_4(in_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_out_ptr0 + x0, xmask)
tmp1 = 30.0
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + x0, tmp2, 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, (1, 4, 4, 1, 1), (16, 4, 1, 1, 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_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_4, reinterpret_tensor(buf0, (4,
4), (1, 4), 0), alpha=1, beta=1, out=buf2)
del buf1
buf3 = buf0
del buf0
triton_poi_fused_add_mul_pow_rsqrt_sum_2[grid(16)](primals_5, buf2,
buf3, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4, 4, 1, 1), (16, 4, 1, 1, 1), torch.
float32)
triton_poi_fused_mul_3[grid(64)](primals_5, buf2, buf3, buf4, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del buf3
buf5 = extern_kernels.convolution(reinterpret_tensor(primals_1, (1,
16, 4, 4), (256, 16, 4, 1), 0), reinterpret_tensor(buf4, (16, 4,
1, 1), (4, 1, 0, 0), 0), stride=(1, 1), padding=(0, 0),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=4, bias=None)
assert_size_stride(buf5, (1, 16, 4, 4), (256, 16, 4, 1))
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused_mul_4[grid(256)](buf6, 256, XBLOCK=256, num_warps=
4, num_stages=1)
return buf6, primals_4, primals_5, buf2, reinterpret_tensor(buf4, (16,
4, 1, 1), (4, 1, 1, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4,
4), (256, 16, 4, 1), 0)
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class ModulatedSiren2dNew(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30):
super().__init__()
self.eps = 1e-08
self.kernel_size = 1
self.in_channel = in_channel
self.out_channel = out_channel
fan_in = in_channel
self.scale = 1 / math.sqrt(fan_in)
if is_first:
self.demod_scale = 3 * fan_in
else:
self.demod_scale = omega_0 ** 2 / 2
self.omega0 = omega_0
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
1, 1))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel})'
)
def forward(self, input_0, input_1):
primals_5 = self.weight
primals_2 = self.modulation.weight
primals_3 = self.modulation.bias
primals_1 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| Dolorousrtur/style-people | ModulatedSiren2d | false | 8,022 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class Model(nn.Module):
def __init__(self, in_channel, out_channel, style_dim, demodulate=True,
is_first=False, omega_0=30):
super().__init__()
self.eps = 1e-08
self.kernel_size = 1
self.in_channel = in_channel
self.out_channel = out_channel
fan_in = in_channel
self.scale = 1 / math.sqrt(fan_in)
if is_first:
self.demod_scale = 3 * fan_in
else:
self.demod_scale = omega_0 ** 2 / 2
self.omega0 = omega_0
self.weight = nn.Parameter(torch.randn(1, out_channel, in_channel,
1, 1))
self.modulation = EqualLinear(style_dim, in_channel, bias_init=1)
self.demodulate = demodulate
def __repr__(self):
return (
f'{self.__class__.__name__}({self.in_channel}, {self.out_channel})'
)
def forward(self, input, style):
batch, in_channel, height, width = input.shape
style = self.modulation(style)
style = style.view(batch, 1, in_channel, 1, 1)
weight = self.scale * self.weight * style
if self.demodulate:
demod = torch.rsqrt(weight.pow(2).sum([2, 3, 4]) * self.
# ... truncated (>4000 chars) for memory efficiency |
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_1/inductor_cache/d2/cd2onkfnnqqav3af2iyomd6l7n3rqdwwvxsyw3ac7zfntydugnbi.py
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add => add
# x => relu
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_3), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_add_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_relu_threshold_backward_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_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_1/inductor_cache/k3/ck3vn263mrulbbfpwhyvmpfmjvy7o6jnv3xicbldp4dkgtnp2lkx.py
# Topologically Sorted Source Nodes: [add_1, log_softmax], Original ATen: [aten.add, aten._log_softmax]
# Source node to ATen node mapping:
# add_1 => add_1
# log_softmax => amax, exp, log, sub, sum_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_5), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_1, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_1, %amax), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_1,), 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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_add_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 16)
x3 = xindex % 16
x0 = xindex % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x3 + (64*x2)), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x3 + (64*x2)), xmask)
tmp6 = tl.load(in_ptr0 + (32 + x3 + (64*x2)), xmask)
tmp9 = tl.load(in_ptr0 + (48 + x3 + (64*x2)), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = triton_helpers.maximum(tmp2, tmp4)
tmp7 = tmp6 + tmp1
tmp8 = triton_helpers.maximum(tmp5, tmp7)
tmp10 = tmp9 + tmp1
tmp11 = triton_helpers.maximum(tmp8, tmp10)
tmp12 = tmp2 - tmp11
tmp13 = tl_math.exp(tmp12)
tmp14 = tmp4 - tmp11
tmp15 = tl_math.exp(tmp14)
tmp16 = tmp13 + tmp15
tmp17 = tmp7 - tmp11
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp10 - tmp11
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp23 = tl_math.log(tmp22)
tl.store(out_ptr0 + (x4), tmp11, xmask)
tl.store(out_ptr1 + (x4), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/m2/cm2wuyciui3pvxqh5sq2x4kd5p6wolgpwpfhmhz2qdlwf5xy2kdu.py
# Topologically Sorted Source Nodes: [add_1, log_softmax], Original ATen: [aten.add, aten._log_softmax]
# Source node to ATen node mapping:
# add_1 => add_1
# log_softmax => amax, exp, log, sub, sub_1, sum_1
# Graph fragment:
# %add_1 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_5), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_1, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_1, %amax), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %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_add_2 = async_compile.triton('triton_poi_fused__log_softmax_add_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_add_2', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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
x4 = xindex
x0 = xindex % 4
x3 = (xindex // 64)
x5 = xindex % 16
tmp0 = tl.load(in_out_ptr0 + (x4), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x5 + (16*x3)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x5 + (16*x3)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = tmp4 - tmp5
tl.store(in_out_ptr0 + (x4), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), primals_2, out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [add, x], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0.run(buf1, primals_3, buf6, 256, grid=grid(256), stream=stream0)
del primals_3
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), primals_4, out=buf2)
buf3 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf4 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add_1, log_softmax], Original ATen: [aten.add, aten._log_softmax]
triton_poi_fused__log_softmax_add_1.run(buf2, primals_5, buf3, buf4, 64, grid=grid(64), stream=stream0)
buf5 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [add_1, log_softmax], Original ATen: [aten.add, aten._log_softmax]
triton_poi_fused__log_softmax_add_2.run(buf5, primals_5, buf3, buf4, 256, grid=grid(256), stream=stream0)
del buf3
del buf4
del primals_5
return (buf5, buf5, reinterpret_tensor(buf1, (4, 64), (1, 4), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), buf6, reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import math
import torch
import torch.nn as nn
import torch.nn.functional as F
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super(Linear, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class MLP(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super(MLP, self).__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
def forward(self, x):
x = torch.relu(self.Linear1(x))
return torch.log_softmax(self.Linear2(x), dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nfeat': 4, 'nhid': 4, 'nclass': 4, 'dropout': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import math
import torch.nn as nn
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_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_add_1(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 16
x3 = xindex % 16
x0 = xindex % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x3 + 64 * x2), xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x3 + 64 * x2), xmask)
tmp6 = tl.load(in_ptr0 + (32 + x3 + 64 * x2), xmask)
tmp9 = tl.load(in_ptr0 + (48 + x3 + 64 * x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = triton_helpers.maximum(tmp2, tmp4)
tmp7 = tmp6 + tmp1
tmp8 = triton_helpers.maximum(tmp5, tmp7)
tmp10 = tmp9 + tmp1
tmp11 = triton_helpers.maximum(tmp8, tmp10)
tmp12 = tmp2 - tmp11
tmp13 = tl_math.exp(tmp12)
tmp14 = tmp4 - tmp11
tmp15 = tl_math.exp(tmp14)
tmp16 = tmp13 + tmp15
tmp17 = tmp7 - tmp11
tmp18 = tl_math.exp(tmp17)
tmp19 = tmp16 + tmp18
tmp20 = tmp10 - tmp11
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp19 + tmp21
tmp23 = tl_math.log(tmp22)
tl.store(out_ptr0 + x4, tmp11, xmask)
tl.store(out_ptr1 + x4, tmp23, xmask)
@triton.jit
def triton_poi_fused__log_softmax_add_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
x4 = xindex
x0 = xindex % 4
x3 = xindex // 64
x5 = xindex % 16
tmp0 = tl.load(in_out_ptr0 + x4, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x5 + 16 * x3), xmask, eviction_policy=
'evict_last')
tmp5 = tl.load(in_ptr2 + (x5 + 16 * x3), xmask, eviction_policy=
'evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = tmp4 - tmp5
tl.store(in_out_ptr0 + x4, tmp6, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (64, 4), (4, 1), 0),
primals_2, out=buf0)
del primals_2
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0[grid(256)](buf1,
primals_3, buf6, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_3
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
primals_4, out=buf2)
buf3 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
buf4 = empty_strided_cuda((4, 1, 4, 4), (16, 64, 4, 1), torch.float32)
triton_poi_fused__log_softmax_add_1[grid(64)](buf2, primals_5, buf3,
buf4, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf5 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
triton_poi_fused__log_softmax_add_2[grid(256)](buf5, primals_5,
buf3, buf4, 256, XBLOCK=256, num_warps=4, num_stages=1)
del buf3
del buf4
del primals_5
return buf5, buf5, reinterpret_tensor(buf1, (4, 64), (1, 4), 0
), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0
), buf6, reinterpret_tensor(primals_1, (4, 64), (1, 4), 0)
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super(Linear, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class MLPNew(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super(MLPNew, self).__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
def forward(self, input_0):
primals_2 = self.Linear1.weight
primals_3 = self.Linear1.bias
primals_4 = self.Linear2.weight
primals_5 = self.Linear2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
| DongHande/PT_propagation_then_training | MLP | false | 8,023 | [
"MIT"
] | 21 | 3f346ff161d2a0b807e3c0269ad26a7266305cc3 | https://github.com/DongHande/PT_propagation_then_training/tree/3f346ff161d2a0b807e3c0269ad26a7266305cc3 | import math
import torch
import torch.nn as nn
import torch.nn.functional as F
class Linear(nn.Module):
def __init__(self, in_features, out_features, dropout, bias=False):
super().__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.randn(in_features, out_features))
if bias:
self.bias = nn.Parameter(torch.randn(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.weight, mode='fan_out', a=math.sqrt(5))
if self.bias is not None:
stdv = 1.0 / math.sqrt(self.weight.size(1))
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input):
input = F.dropout(input, self.dropout, training=self.training)
output = torch.matmul(input, self.weight)
if self.bias is not None:
return output + self.bias
else:
return output
class Model(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super().__init__()
self.Linear1 = Linear(nfeat, nhid, dropout, bias=True)
self.Linear2 = Linear(nhid, nclass, dropout, bias=True)
def forward(self, x):
x = torch.relu(self.Linear1(x))
return torch.log_softmax(self.Linear2(x), dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4, 0.5]
|
SEModule | # 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_1/inductor_cache/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.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 = (%view, [-1]), 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bq/cbqs25ilz2wluqikclvslctdlvul4rcbcdk3m2pcvrxffj6hdrw5.py
# Topologically Sorted Source Nodes: [x_se2, x_se2_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x_se2 => convolution
# x_se2_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%view_1, %primals_2, %primals_3, [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_1 = async_compile.triton('triton_poi_fused_convolution_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_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_1/inductor_cache/x7/cx7evvvm7te22h7xf3yh7pnjatqie5vy54vyorfffrtctztd4wn5.py
# Topologically Sorted Source Nodes: [x_se_1], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_se_1 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%relu, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vz/cvzlcxmuowtstgqrxzb5hcsechd32n3vjbzbpf457cjjvsojtkea.py
# Topologically Sorted Source Nodes: [x_se_2, mul], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# x_se_2 => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 // 16)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, 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, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
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: [mean], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 16, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [x_se2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (4, 4, 1, 1), (4, 1, 0, 0), 0), primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 1, 1), (4, 1, 1, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_se2, x_se2_1], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_1.run(buf3, primals_3, 16, grid=grid(16), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [x_se_1], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf3, 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, 4, 1, 1), (4, 1, 1, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [x_se_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf5, primals_5, 16, grid=grid(16), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_se_2, 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, (4, 4, 1, 1), (4, 1, 1, 1), 0), buf3, buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
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 torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class FastAvgPool2d(nn.Module):
def __init__(self, flatten=False):
super(FastAvgPool2d, self).__init__()
self.flatten = flatten
def forward(self, x):
if self.flatten:
in_size = x.size()
return x.view((in_size[0], in_size[1], -1)).mean(dim=2)
else:
return x.view(x.size(0), x.size(1), -1).mean(-1).view(x.size(0),
x.size(1), 1, 1)
class SEModule(nn.Module):
def __init__(self, channels, reduction_channels, inplace=True):
super(SEModule, self).__init__()
self.avg_pool = FastAvgPool2d()
self.fc1 = nn.Conv2d(channels, reduction_channels, kernel_size=1,
padding=0, bias=True)
self.relu = nn.ReLU(inplace=inplace)
self.fc2 = nn.Conv2d(reduction_channels, channels, kernel_size=1,
padding=0, bias=True)
self.activation = nn.Sigmoid()
def forward(self, x):
x_se = self.avg_pool(x)
x_se2 = self.fc1(x_se)
x_se2 = self.relu(x_se2)
x_se = self.fc2(x_se2)
x_se = self.activation(x_se)
return x * x_se
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'channels': 4, 'reduction_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torchvision import datasets as datasets
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 = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_convolution_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_convolution_2(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
tl.store(in_out_ptr0 + x2, 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 // 16
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, 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, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
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, 16, XBLOCK=8,
num_warps=2, num_stages=1)
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (4, 4, 1,
1), (4, 1, 0, 0), 0), primals_2, stride=(1, 1), padding=(0, 0),
dilation=(1, 1), transposed=False, output_padding=(0, 0),
groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 1, 1), (4, 1, 1, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_relu_1[grid(16)](buf3, primals_3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf4 = extern_kernels.convolution(buf3, 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, 4, 1, 1), (4, 1, 1, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_2[grid(16)](buf5, primals_5, 16,
XBLOCK=16, 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=128, num_warps=4, num_stages=1)
return buf6, primals_1, primals_2, primals_4, reinterpret_tensor(buf1,
(4, 4, 1, 1), (4, 1, 1, 1), 0), buf3, buf5
class FastAvgPool2d(nn.Module):
def __init__(self, flatten=False):
super(FastAvgPool2d, self).__init__()
self.flatten = flatten
def forward(self, x):
if self.flatten:
in_size = x.size()
return x.view((in_size[0], in_size[1], -1)).mean(dim=2)
else:
return x.view(x.size(0), x.size(1), -1).mean(-1).view(x.size(0),
x.size(1), 1, 1)
class SEModuleNew(nn.Module):
def __init__(self, channels, reduction_channels, inplace=True):
super(SEModuleNew, self).__init__()
self.avg_pool = FastAvgPool2d()
self.fc1 = nn.Conv2d(channels, reduction_channels, kernel_size=1,
padding=0, bias=True)
self.relu = nn.ReLU(inplace=inplace)
self.fc2 = nn.Conv2d(reduction_channels, channels, kernel_size=1,
padding=0, bias=True)
self.activation = 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]
| Alibaba-MIIL/ZS_SDL | SEModule | false | 8,024 | [
"MIT"
] | 20 | 769fe4f57d2d458a7c4b5468a6395c9b296b1dad | https://github.com/Alibaba-MIIL/ZS_SDL/tree/769fe4f57d2d458a7c4b5468a6395c9b296b1dad | import torch
from torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
class FastAvgPool2d(nn.Module):
def __init__(self, flatten=False):
super().__init__()
self.flatten = flatten
def forward(self, x):
if self.flatten:
in_size = x.size()
return x.view((in_size[0], in_size[1], -1)).mean(dim=2)
else:
return x.view(x.size(0), x.size(1), -1).mean(-1).view(x.size(0),
x.size(1), 1, 1)
class Model(nn.Module):
def __init__(self, channels, reduction_channels, inplace=True):
super().__init__()
self.avg_pool = FastAvgPool2d()
self.fc1 = nn.Conv2d(channels, reduction_channels, kernel_size=1,
padding=0, bias=True)
self.relu = nn.ReLU(inplace=inplace)
self.fc2 = nn.Conv2d(reduction_channels, channels, kernel_size=1,
padding=0, bias=True)
self.activation = nn.Sigmoid()
def forward(self, x):
x_se = self.avg_pool(x)
x_se2 = self.fc1(x_se)
x_se2 = self.relu(x_se2)
x_se = self.fc2(x_se2)
x_se = self.activation(x_se)
return x * x_se
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
Encoder | # 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_1/inductor_cache/24/c24qsgijonbiqjcskkesmr6djddhrqjlc6pskdyvv3cj26t4733k.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x => add, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [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=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_native_layer_norm_0 = async_compile.triton('triton_poi_fused_native_layer_norm_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/wk/cwkmrcckbwmqrnn75bcrj6x53nm4p3l2vitrgxgtbfaftyuxfsme.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [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=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_3, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_2), kwargs = {})
triton_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/al/caliettpjjnnqg6p2uhziboizxtsw5rgrltfbzvndr37hmhbnoiz.py
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# product => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1) + (64*x3)), xmask)
tl.store(out_ptr0 + (x4), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/44/c44udt7qyfjbhr63snwdnvww3k3vzms334kz5lteswpyl475z5jc.py
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# product => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_3 = async_compile.triton('triton_poi_fused_clone_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, 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')
# kernel path: runs/run_shard_1/inductor_cache/lh/clhxgnq4cafirtz47wl6jzcd3h5dig7oyfpzue44vcylnn7f323m.py
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
# Source node to ATen node mapping:
# product_1 => mul_2
# product_2 => add_2
# weights => amax, exp, sub_1, sum_1
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_11, 0.5), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_7), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_2, [-1], True), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %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_add_mul_4 = async_compile.triton('triton_poi_fused__softmax_add_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_mul_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_4(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp5 * tmp1
tmp8 = tmp6 + tmp7
tmp9 = triton_helpers.maximum(tmp4, tmp8)
tmp11 = tmp10 * tmp1
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp9, tmp13)
tmp16 = tmp15 * tmp1
tmp18 = tmp16 + tmp17
tmp19 = triton_helpers.maximum(tmp14, tmp18)
tmp20 = tmp4 - tmp19
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp8 - tmp19
tmp23 = tl_math.exp(tmp22)
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp19
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp24 + tmp26
tmp28 = tmp18 - tmp19
tmp29 = tl_math.exp(tmp28)
tmp30 = tmp27 + tmp29
tl.store(out_ptr0 + (x2), tmp19, xmask)
tl.store(out_ptr1 + (x2), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/x6/cx6c4ok5m7sbmxirupirmab2aktopchha7fb7a34wbincnrln762.py
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
# Source node to ATen node mapping:
# product_1 => mul_2
# product_2 => add_2
# weights => amax, div, exp, sub_1
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_11, 0.5), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_7), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_2, [-1], True), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_add_mul_5 = async_compile.triton('triton_poi_fused__softmax_add_mul_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_mul_5', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_5(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
x4 = xindex % 64
x5 = (xindex // 4)
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp3 = tl.load(in_ptr0 + (x4), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (x5), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + (x5), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tl.store(in_out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/eq/ceqkt22etynsgdiire3npctu6ae2jyxstmwvc6an3vghfqql2yu2.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_3
# x_3 => var_mean_1
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_3, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/6z/c6zahf5ye65saqqmps3f6q6adeffviegsxonkuv73ylppfyxij3s.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_3
# x_3 => add_4, add_5, mul_3, mul_4, rsqrt_1, sub_2
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %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_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, %rsqrt_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %primals_9), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %primals_10), kwargs = {})
triton_poi_fused_add_native_layer_norm_7 = async_compile.triton('triton_poi_fused_add_native_layer_norm_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cu/ccuzxb2hwpelvwivkigx4lpsgwl5usrjfvjoxv7imz5d36525vcs.py
# Topologically Sorted Source Nodes: [hidden_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# hidden_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_19,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_8 = async_compile.triton('triton_poi_fused_relu_threshold_backward_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_8', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_8(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/cm/ccmfj63xgs56ptvu7q76u66udb6hv4yzprftbodghvdyo62b2lwe.py
# Topologically Sorted Source Nodes: [x_2, x_5], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x_2 => add_3
# x_5 => add_6
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %add_6 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_21, %add_3), kwargs = {})
triton_poi_fused_add_9 = async_compile.triton('triton_poi_fused_add_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_9(in_out_ptr0, in_ptr0, in_ptr1, 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_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp4 = tl.load(in_ptr2 + (x2), xmask)
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 = 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), (16, 4, 1))
assert_size_stride(primals_4, (16, 4), (4, 1))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16, 4), (4, 1))
assert_size_stride(primals_7, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_8, (4, 16), (16, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, 4), (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, ), (1, ))
assert_size_stride(primals_16, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf1 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_0.run(primals_3, buf0, buf1, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(primals_3, buf0, buf1, primals_1, primals_2, buf2, 64, grid=grid(64), stream=stream0)
del primals_1
del primals_2
buf3 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [q], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 16), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf4)
buf5 = empty_strided_cuda((16, 16), (16, 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_6, (4, 16), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf3, buf6, 256, grid=grid(256), stream=stream0)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf4, buf7, 64, 4, grid=grid(64, 4), stream=stream0)
buf8 = reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf6, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), out=buf8)
buf9 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
triton_poi_fused__softmax_add_mul_4.run(buf8, primals_7, buf9, buf10, 64, grid=grid(64), stream=stream0)
buf11 = reinterpret_tensor(buf8, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
triton_poi_fused__softmax_add_mul_5.run(buf11, primals_7, buf9, buf10, 256, grid=grid(256), stream=stream0)
del primals_7
buf12 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf5, buf12, 256, grid=grid(256), stream=stream0)
buf13 = reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf11, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf12, (16, 4, 4), (16, 4, 1), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf13, buf14, 256, grid=grid(256), stream=stream0)
del buf13
buf15 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf14, (16, 16), (16, 1), 0), reinterpret_tensor(primals_8, (16, 4), (1, 16), 0), out=buf15)
buf16 = buf1; del buf1 # reuse
buf17 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(buf15, primals_3, buf16, buf17, 16, grid=grid(16), stream=stream0)
buf18 = reinterpret_tensor(buf10, (4, 4, 4), (16, 4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_7.run(buf15, primals_3, buf16, buf17, primals_9, primals_10, buf18, 64, grid=grid(64), stream=stream0)
del primals_10
buf19 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf18, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf19)
buf20 = reinterpret_tensor(buf19, (4, 4, 4), (16, 4, 1), 0); del buf19 # reuse
buf26 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [hidden_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_8.run(buf20, primals_12, buf26, 64, grid=grid(64), stream=stream0)
del primals_12
buf21 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf20, (16, 4), (4, 1), 0), reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf21)
buf22 = reinterpret_tensor(buf21, (4, 4, 4), (16, 4, 1), 0); del buf21 # reuse
# Topologically Sorted Source Nodes: [x_2, x_5], Original ATen: [aten.add]
triton_poi_fused_add_9.run(buf22, primals_14, buf15, primals_3, 64, grid=grid(64), stream=stream0)
del primals_14
buf23 = buf17; del buf17 # reuse
buf24 = buf16; del buf16 # reuse
# Topologically Sorted Source Nodes: [x_6], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_0.run(buf22, buf23, buf24, 16, grid=grid(16), stream=stream0)
buf25 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_6], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(buf22, buf23, buf24, primals_15, primals_16, buf25, 64, grid=grid(64), stream=stream0)
del buf23
del buf24
del primals_16
return (buf25, primals_3, primals_9, primals_15, reinterpret_tensor(buf2, (16, 4), (4, 1), 0), buf11, reinterpret_tensor(buf14, (16, 16), (16, 1), 0), buf15, reinterpret_tensor(buf18, (16, 4), (4, 1), 0), reinterpret_tensor(buf20, (16, 4), (4, 1), 0), buf22, primals_13, buf26, primals_11, primals_8, reinterpret_tensor(buf12, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf6, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf7, (16, 4, 4), (16, 1, 4), 0), primals_6, primals_5, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 4), (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, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_16 = 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])
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
from functools import partial
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super(FFN, self).__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
class MultiHeadAttention(nn.Module):
"""
Multi-Head Attention
"""
def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.d_key = d_key
self.d_value = d_value
self.d_model = d_model
self.dropout_rate = dropout_rate
self.q_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.k_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.v_fc = nn.Linear(in_features=d_model, out_features=d_value *
n_head, bias=False)
self.proj_fc = nn.Linear(in_features=d_value * n_head, out_features
=d_model, bias=False)
def _prepare_qkv(self, queries, keys, values, cache=None):
if keys is None:
keys, values = queries, queries
static_kv = False
else:
static_kv = True
q = self.q_fc(queries)
q = torch.reshape(q, shape=(q.shape[0], -1, self.n_head, self.d_key))
q = q.permute(0, 2, 1, 3)
if cache is not None and static_kv and 'static_k' in cache:
k = cache['static_k']
v = cache['static_v']
else:
k = self.k_fc(keys)
v = self.v_fc(values)
k = torch.reshape(k, shape=(k.shape[0], -1, self.n_head, self.
d_key))
k = k.permute(0, 2, 1, 3)
v = torch.reshape(v, shape=(v.shape[0], -1, self.n_head, self.
d_value))
v = v.permute(0, 2, 1, 3)
if cache is not None:
if static_kv and 'static_k' not in cache:
cache['static_k'], cache['static_v'] = k, v
elif not static_kv:
cache_k, cache_v = cache['k'], cache['v']
k = torch.cat((cache_k, k), dim=2)
v = torch.cat((cache_v, v), dim=2)
cache['k'], cache['v'] = k, v
return q, k, v
def forward(self, queries, keys, values, attn_bias, cache=None):
keys = queries if keys is None else keys
values = keys if values is None else values
q, k, v = self._prepare_qkv(queries, keys, values, cache)
product = torch.matmul(q, k.transpose(-2, -1))
product = product * self.d_model ** -0.5
if attn_bias is not None:
product += attn_bias
weights = F.softmax(product, dim=-1)
if self.dropout_rate:
weights = F.dropout(weights, p=self.dropout_rate, training=False)
out = torch.matmul(weights, v)
out = out.permute(0, 2, 1, 3)
out = torch.reshape(out, shape=(out.shape[0], -1, out.shape[2] *
out.shape[3]))
out = self.proj_fc(out)
return out
class PrePostProcessLayer(nn.Module):
"""
PrePostProcessLayer
"""
def __init__(self, process_cmd, d_model, dropout_rate):
super(PrePostProcessLayer, self).__init__()
self.process_cmd = process_cmd
self.functors = []
self.g = partial(self.f)
self.g1 = partial(self.f1, dropout_rate)
for cmd in self.process_cmd:
if cmd == 'a':
self.functors.append(self.g)
elif cmd == 'n':
num = sum(1 for line in self.children())
layernorm = nn.LayerNorm(normalized_shape=d_model)
self.add_module('layer_norm_%d' % num, layernorm)
self.functors.append(layernorm)
elif cmd == 'd':
self.functors.append(self.g1)
@staticmethod
def f(x, y):
if y is not None:
return x + y
else:
return x
@staticmethod
def f1(p, x):
if p:
return F.dropout(x, p, training=False)
else:
return x
def forward(self, x, residual=None):
for i, cmd in enumerate(self.process_cmd):
if cmd == 'a':
x = self.functors[i](x, residual)
else:
x = self.functors[i](x)
return x
class EncoderLayer(nn.Module):
"""
EncoderLayer
"""
def __init__(self, n_head, d_key, d_value, d_model, d_inner_hid,
prepostprocess_dropout, attention_dropout, relu_dropout,
preprocess_cmd='n', postprocess_cmd='da'):
super(EncoderLayer, self).__init__()
self.preprocesser1 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.self_attn = MultiHeadAttention(d_key, d_value, d_model, n_head,
attention_dropout)
self.postprocesser1 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
self.preprocesser2 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.ffn = FFN(d_inner_hid, d_model, relu_dropout)
self.postprocesser2 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
def forward(self, enc_input, attn_bias):
attn_output = self.self_attn(self.preprocesser1(enc_input), None,
None, attn_bias)
attn_output = self.postprocesser1(attn_output, enc_input)
ffn_output = self.ffn(self.preprocesser2(attn_output))
ffn_output = self.postprocesser2(ffn_output, attn_output)
return ffn_output
class Encoder(nn.Module):
"""
encoder
"""
def __init__(self, n_layer, n_head, d_key, d_value, d_model,
d_inner_hid, prepostprocess_dropout, attention_dropout,
relu_dropout, preprocess_cmd='n', postprocess_cmd='da'):
super(Encoder, self).__init__()
self.encoder_layers = list()
for i in range(n_layer):
encoder = EncoderLayer(n_head, d_key, d_value, d_model,
d_inner_hid, prepostprocess_dropout, attention_dropout,
relu_dropout, preprocess_cmd, postprocess_cmd)
self.add_module('layer_%d' % i, encoder)
self.encoder_layers.append(encoder)
self.processer = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
def forward(self, enc_input, attn_bias):
for encoder_layer in self.encoder_layers:
enc_output = encoder_layer(enc_input, attn_bias)
enc_input = enc_output
enc_output = self.processer(enc_output)
return enc_output
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'n_layer': 1, 'n_head': 4, 'd_key': 4, 'd_value': 4,
'd_model': 4, 'd_inner_hid': 4, 'prepostprocess_dropout': 0.5,
'attention_dropout': 0.5, 'relu_dropout': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime 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
from functools import partial
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 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_1(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_clone_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
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1 + 64 * x3), xmask)
tl.store(out_ptr0 + x4, tmp0, xmask)
@triton.jit
def triton_poi_fused_clone_3(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)
@triton.jit
def triton_poi_fused__softmax_add_mul_4(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp17 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp5 * tmp1
tmp8 = tmp6 + tmp7
tmp9 = triton_helpers.maximum(tmp4, tmp8)
tmp11 = tmp10 * tmp1
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp9, tmp13)
tmp16 = tmp15 * tmp1
tmp18 = tmp16 + tmp17
tmp19 = triton_helpers.maximum(tmp14, tmp18)
tmp20 = tmp4 - tmp19
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp8 - tmp19
tmp23 = tl_math.exp(tmp22)
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp19
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp24 + tmp26
tmp28 = tmp18 - tmp19
tmp29 = tl_math.exp(tmp28)
tmp30 = tmp27 + tmp29
tl.store(out_ptr0 + x2, tmp19, xmask)
tl.store(out_ptr1 + x2, tmp30, xmask)
@triton.jit
def triton_poi_fused__softmax_add_mul_5(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
x4 = xindex % 64
x5 = xindex // 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp3 = tl.load(in_ptr0 + x4, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + x5, xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + x5, xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tl.store(in_out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_8(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
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_9(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_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp4 = tl.load(in_ptr2 + x2, xmask)
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) = 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), (16, 4, 1))
assert_size_stride(primals_4, (16, 4), (4, 1))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16, 4), (4, 1))
assert_size_stride(primals_7, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_8, (4, 16), (16, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4, 4), (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,), (1,))
assert_size_stride(primals_16, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf1 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_native_layer_norm_0[grid(16)](primals_3, buf0,
buf1, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(64)](primals_3, buf0,
buf1, primals_1, primals_2, buf2, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del primals_1
del primals_2
buf3 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 16), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf4)
buf5 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 16), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(256)](buf3, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused_clone_3[grid(64, 4)](buf4, buf7, 64, 4, XBLOCK=4,
YBLOCK=32, num_warps=4, num_stages=1)
buf8 = reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0)
del buf4
extern_kernels.bmm(reinterpret_tensor(buf6, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), out=buf8)
buf9 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused__softmax_add_mul_4[grid(64)](buf8, primals_7, buf9,
buf10, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf8, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf8
triton_poi_fused__softmax_add_mul_5[grid(256)](buf11, primals_7,
buf9, buf10, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
buf12 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(256)](buf5, buf12, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf13 = reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1), 0)
del buf5
extern_kernels.bmm(reinterpret_tensor(buf11, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf12, (16, 4, 4), (16, 4, 1), 0), out=buf13
)
buf14 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(256)](buf13, buf14, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf13
buf15 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.mm(reinterpret_tensor(buf14, (16, 16), (16, 1), 0),
reinterpret_tensor(primals_8, (16, 4), (1, 16), 0), out=buf15)
buf16 = buf1
del buf1
buf17 = buf0
del buf0
triton_poi_fused_add_native_layer_norm_6[grid(16)](buf15, primals_3,
buf16, buf17, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf18 = reinterpret_tensor(buf10, (4, 4, 4), (16, 4, 1), 0)
del buf10
triton_poi_fused_add_native_layer_norm_7[grid(64)](buf15, primals_3,
buf16, buf17, primals_9, primals_10, buf18, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_10
buf19 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf18, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf19)
buf20 = reinterpret_tensor(buf19, (4, 4, 4), (16, 4, 1), 0)
del buf19
buf26 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_8[grid(64)](buf20,
primals_12, buf26, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_12
buf21 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf20, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf21)
buf22 = reinterpret_tensor(buf21, (4, 4, 4), (16, 4, 1), 0)
del buf21
triton_poi_fused_add_9[grid(64)](buf22, primals_14, buf15,
primals_3, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_14
buf23 = buf17
del buf17
buf24 = buf16
del buf16
triton_poi_fused_native_layer_norm_0[grid(16)](buf22, buf23, buf24,
16, XBLOCK=16, num_warps=1, num_stages=1)
buf25 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(64)](buf22, buf23, buf24,
primals_15, primals_16, buf25, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf23
del buf24
del primals_16
return buf25, primals_3, primals_9, primals_15, reinterpret_tensor(buf2,
(16, 4), (4, 1), 0), buf11, reinterpret_tensor(buf14, (16, 16), (16,
1), 0), buf15, reinterpret_tensor(buf18, (16, 4), (4, 1), 0
), reinterpret_tensor(buf20, (16, 4), (4, 1), 0
), buf22, primals_13, buf26, primals_11, primals_8, reinterpret_tensor(
buf12, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf6, (16, 4,
4), (16, 1, 4), 0), reinterpret_tensor(buf7, (16, 4, 4), (16, 1, 4), 0
), primals_6, primals_5, primals_4
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super(FFN, self).__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
class MultiHeadAttention(nn.Module):
"""
Multi-Head Attention
"""
def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.d_key = d_key
self.d_value = d_value
self.d_model = d_model
self.dropout_rate = dropout_rate
self.q_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.k_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.v_fc = nn.Linear(in_features=d_model, out_features=d_value *
n_head, bias=False)
self.proj_fc = nn.Linear(in_features=d_value * n_head, out_features
=d_model, bias=False)
def _prepare_qkv(self, queries, keys, values, cache=None):
if keys is None:
keys, values = queries, queries
static_kv = False
else:
static_kv = True
q = self.q_fc(queries)
q = torch.reshape(q, shape=(q.shape[0], -1, self.n_head, self.d_key))
q = q.permute(0, 2, 1, 3)
if cache is not None and static_kv and 'static_k' in cache:
k = cache['static_k']
v = cache['static_v']
else:
k = self.k_fc(keys)
v = self.v_fc(values)
k = torch.reshape(k, shape=(k.shape[0], -1, self.n_head, self.
d_key))
k = k.permute(0, 2, 1, 3)
v = torch.reshape(v, shape=(v.shape[0], -1, self.n_head, self.
d_value))
v = v.permute(0, 2, 1, 3)
if cache is not None:
if static_kv and 'static_k' not in cache:
cache['static_k'], cache['static_v'] = k, v
elif not static_kv:
cache_k, cache_v = cache['k'], cache['v']
k = torch.cat((cache_k, k), dim=2)
v = torch.cat((cache_v, v), dim=2)
cache['k'], cache['v'] = k, v
return q, k, v
def forward(self, queries, keys, values, attn_bias, cache=None):
keys = queries if keys is None else keys
values = keys if values is None else values
q, k, v = self._prepare_qkv(queries, keys, values, cache)
product = torch.matmul(q, k.transpose(-2, -1))
product = product * self.d_model ** -0.5
if attn_bias is not None:
product += attn_bias
weights = F.softmax(product, dim=-1)
if self.dropout_rate:
weights = F.dropout(weights, p=self.dropout_rate, training=False)
out = torch.matmul(weights, v)
out = out.permute(0, 2, 1, 3)
out = torch.reshape(out, shape=(out.shape[0], -1, out.shape[2] *
out.shape[3]))
out = self.proj_fc(out)
return out
class PrePostProcessLayer(nn.Module):
"""
PrePostProcessLayer
"""
def __init__(self, process_cmd, d_model, dropout_rate):
super(PrePostProcessLayer, self).__init__()
self.process_cmd = process_cmd
self.functors = []
self.g = partial(self.f)
self.g1 = partial(self.f1, dropout_rate)
for cmd in self.process_cmd:
if cmd == 'a':
self.functors.append(self.g)
elif cmd == 'n':
num = sum(1 for line in self.children())
layernorm = nn.LayerNorm(normalized_shape=d_model)
self.add_module('layer_norm_%d' % num, layernorm)
self.functors.append(layernorm)
elif cmd == 'd':
self.functors.append(self.g1)
@staticmethod
def f(x, y):
if y is not None:
return x + y
else:
return x
@staticmethod
def f1(p, x):
if p:
return F.dropout(x, p, training=False)
else:
return x
def forward(self, x, residual=None):
for i, cmd in enumerate(self.process_cmd):
if cmd == 'a':
x = self.functors[i](x, residual)
else:
x = self.functors[i](x)
return x
class EncoderLayer(nn.Module):
"""
EncoderLayer
"""
def __init__(self, n_head, d_key, d_value, d_model, d_inner_hid,
prepostprocess_dropout, attention_dropout, relu_dropout,
preprocess_cmd='n', postprocess_cmd='da'):
super(EncoderLayer, self).__init__()
self.preprocesser1 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.self_attn = MultiHeadAttention(d_key, d_value, d_model, n_head,
attention_dropout)
self.postprocesser1 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
self.preprocesser2 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.ffn = FFN(d_inner_hid, d_model, relu_dropout)
self.postprocesser2 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
def forward(self, enc_input, attn_bias):
attn_output = self.self_attn(self.preprocesser1(enc_input), None,
None, attn_bias)
attn_output = self.postprocesser1(attn_output, enc_input)
ffn_output = self.ffn(self.preprocesser2(attn_output))
ffn_output = self.postprocesser2(ffn_output, attn_output)
return ffn_output
class EncoderNew(nn.Module):
"""
encoder
"""
def __init__(self, n_layer, n_head, d_key, d_value, d_model,
d_inner_hid, prepostprocess_dropout, attention_dropout,
relu_dropout, preprocess_cmd='n', postprocess_cmd='da'):
super(EncoderNew, self).__init__()
self.encoder_layers = list()
for i in range(n_layer):
encoder = EncoderLayer(n_head, d_key, d_value, d_model,
d_inner_hid, prepostprocess_dropout, attention_dropout,
relu_dropout, preprocess_cmd, postprocess_cmd)
self.add_module('layer_%d' % i, encoder)
self.encoder_layers.append(encoder)
self.processer = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
def forward(self, input_0, input_1):
primals_1 = self.layer_0.preprocesser1.layer_norm_0.weight
primals_2 = self.layer_0.preprocesser1.layer_norm_0.bias
primals_4 = self.layer_0.self_attn.q_fc.weight
primals_5 = self.layer_0.self_attn.k_fc.weight
primals_6 = self.layer_0.self_attn.v_fc.weight
primals_8 = self.layer_0.self_attn.proj_fc.weight
primals_9 = self.layer_0.preprocesser2.layer_norm_0.weight
primals_10 = self.layer_0.preprocesser2.layer_norm_0.bias
primals_11 = self.layer_0.ffn.fc1.weight
primals_12 = self.layer_0.ffn.fc1.bias
primals_13 = self.layer_0.ffn.fc2.weight
primals_14 = self.layer_0.ffn.fc2.bias
primals_15 = self.processer.layer_norm_0.weight
primals_16 = self.processer.layer_norm_0.bias
primals_3 = input_0
primals_7 = 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])
return output[0]
| DocYard-ai/UCR | Encoder | false | 8,025 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
import torch.nn.functional as F
from functools import partial
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super().__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
class MultiHeadAttention(nn.Module):
"""
Multi-Head Attention
"""
def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
super().__init__()
self.n_head = n_head
self.d_key = d_key
self.d_value = d_value
self.d_model = d_model
self.dropout_rate = dropout_rate
self.q_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.k_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.v_fc = nn.Linear(in_features=d_model, out_features=d_value *
n_head, bias=False)
self.proj_fc = nn.Linear(in_features=d_value * n_head, out_features
=d_model, bias=False)
def _prepare_qkv(self, queries, keys, values, cache=None):
if keys is None:
keys, values = queries, queries
static_kv = False
else:
static_kv = True
q = self.q_fc(queries)
q = torch.reshape(q, shape=(q.shape[0], -1, self.n_head, self.d_key))
q = q.permute(0, 2, 1, 3)
if cache is not None and static_kv and 'static_k' in cache:
k = cache['static_k']
v = cache['static_v']
else:
k = self.k_fc(keys)
v = self.v_fc(values)
k = torch.reshape(k, shape=(k.shape[0], -1, self.n_head, self.
d_key))
k = k.permute(0, 2, 1, 3)
v = torch.reshape(v, shape=(v.shape[0], -1, self.n_head, self.
d_value))
v = v.permute(0, 2, 1, 3)
if cache is not None:
if static_kv and 'static_k' not in cache:
cache['static_k'], cache['static_v'] = k, v
elif not static_kv:
cache_k, cache_v = cache['k'], cache['v']
k = torch.cat((cache_k, k), dim=2)
v = torch.cat((cache_v, v), dim=2)
cache['k'], cache['v'] = k, v
return q, k, v
def forward(self, queries, keys, values, attn_bias, cache=None):
keys = queries if keys is None else keys
values = keys if values is None else values
q, k, v = self._prepare_qkv(queries, keys, values, cache)
product = torch.matmul(q, k.transpose(-2, -1))
product = product * self.d_model ** -0.5
if attn_bias is not None:
product += attn_bias
weights = F.softmax(product, dim=-1)
if self.dropout_rate:
weights = F.dropout(weights, p=self.dropout_rate, training=False)
out = torch.matmul(weights, v)
out = out.permute(0, 2, 1, 3)
out = torch.reshape(out, shape=(out.shape[0], -1, out.shape[2] *
out.shape[3]))
out = self.proj_fc(out)
return out
class PrePostProcessLayer(nn.Module):
"""
PrePostProcessLayer
"""
def __init__(self, process_cmd, d_model, dropout_rate):
super().__init__()
self.process_cmd = process_cmd
self.functors = []
self.g = partial(self.f)
self.g1 = partial(self.f1, dropout_rate)
for cmd in self.process_cmd:
if cmd == 'a':
self.functors.append(self.g)
elif cmd == 'n':
num = sum(1 for line in self.children())
# ... truncated (>4000 chars) for memory efficiency |
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_1/inductor_cache/24/c24qsgijonbiqjcskkesmr6djddhrqjlc6pskdyvv3cj26t4733k.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x => add, rsqrt, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [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=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_poi_fused_native_layer_norm_0 = async_compile.triton('triton_poi_fused_native_layer_norm_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/wk/cwkmrcckbwmqrnn75bcrj6x53nm4p3l2vitrgxgtbfaftyuxfsme.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# x => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_3, [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=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_3, %getitem_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %primals_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %primals_2), kwargs = {})
triton_poi_fused_native_layer_norm_1 = async_compile.triton('triton_poi_fused_native_layer_norm_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_native_layer_norm_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_native_layer_norm_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/al/caliettpjjnnqg6p2uhziboizxtsw5rgrltfbzvndr37hmhbnoiz.py
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# product => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x2 = (xindex // 16) % 4
x3 = (xindex // 64)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1) + (64*x3)), xmask)
tl.store(out_ptr0 + (x4), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/44/c44udt7qyfjbhr63snwdnvww3k3vzms334kz5lteswpyl475z5jc.py
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# product => clone_1
# Graph fragment:
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_3 = async_compile.triton('triton_poi_fused_clone_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_clone_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, 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')
# kernel path: runs/run_shard_1/inductor_cache/lh/clhxgnq4cafirtz47wl6jzcd3h5dig7oyfpzue44vcylnn7f323m.py
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
# Source node to ATen node mapping:
# product_1 => mul_2
# product_2 => add_2
# weights => amax, exp, sub_1, sum_1
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_11, 0.5), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_7), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_2, [-1], True), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %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_add_mul_4 = async_compile.triton('triton_poi_fused__softmax_add_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_mul_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_4(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp5 * tmp1
tmp8 = tmp6 + tmp7
tmp9 = triton_helpers.maximum(tmp4, tmp8)
tmp11 = tmp10 * tmp1
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp9, tmp13)
tmp16 = tmp15 * tmp1
tmp18 = tmp16 + tmp17
tmp19 = triton_helpers.maximum(tmp14, tmp18)
tmp20 = tmp4 - tmp19
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp8 - tmp19
tmp23 = tl_math.exp(tmp22)
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp19
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp24 + tmp26
tmp28 = tmp18 - tmp19
tmp29 = tl_math.exp(tmp28)
tmp30 = tmp27 + tmp29
tl.store(out_ptr0 + (x2), tmp19, xmask)
tl.store(out_ptr1 + (x2), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/x6/cx6c4ok5m7sbmxirupirmab2aktopchha7fb7a34wbincnrln762.py
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
# Source node to ATen node mapping:
# product_1 => mul_2
# product_2 => add_2
# weights => amax, div, exp, sub_1
# Graph fragment:
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_11, 0.5), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_7), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%add_2, [-1], True), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_add_mul_5 = async_compile.triton('triton_poi_fused__softmax_add_mul_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: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_add_mul_5', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mul_5(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
x4 = xindex % 64
x5 = (xindex // 4)
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp3 = tl.load(in_ptr0 + (x4), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (x5), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + (x5), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tl.store(in_out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/eq/ceqkt22etynsgdiire3npctu6ae2jyxstmwvc6an3vghfqql2yu2.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_3
# x_3 => var_mean_1
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_3, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_6', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/6z/c6zahf5ye65saqqmps3f6q6adeffviegsxonkuv73ylppfyxij3s.py
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# x_2 => add_3
# x_3 => add_4, add_5, mul_3, mul_4, rsqrt_1, sub_2
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %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_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_2, %rsqrt_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_3, %primals_9), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_4, %primals_10), kwargs = {})
triton_poi_fused_add_native_layer_norm_7 = async_compile.triton('triton_poi_fused_add_native_layer_norm_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6, 7), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 6, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + (x2), tmp13, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cu/ccuzxb2hwpelvwivkigx4lpsgwl5usrjfvjoxv7imz5d36525vcs.py
# Topologically Sorted Source Nodes: [hidden_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# hidden_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_19,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_8 = async_compile.triton('triton_poi_fused_relu_threshold_backward_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_relu_threshold_backward_8', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_relu_threshold_backward_8(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 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_1/inductor_cache/cm/ccmfj63xgs56ptvu7q76u66udb6hv4yzprftbodghvdyo62b2lwe.py
# Topologically Sorted Source Nodes: [x_2, x_5], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x_2 => add_3
# x_5 => add_6
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_17, %primals_3), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_21, %add_3), kwargs = {})
triton_poi_fused_add_9 = async_compile.triton('triton_poi_fused_add_9', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_9', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_9(in_out_ptr0, in_ptr0, in_ptr1, 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_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp4 = tl.load(in_ptr2 + (x2), xmask)
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 = 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), (16, 4, 1))
assert_size_stride(primals_4, (16, 4), (4, 1))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16, 4), (4, 1))
assert_size_stride(primals_7, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_8, (4, 16), (16, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, 4), (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, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf1 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_0.run(primals_3, buf0, buf1, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_1.run(primals_3, buf0, buf1, primals_1, primals_2, buf2, 64, grid=grid(64), stream=stream0)
del primals_1
del primals_2
buf3 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [q], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 16), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [k], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf4)
buf5 = empty_strided_cuda((16, 16), (16, 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_6, (4, 16), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf3, buf6, 256, grid=grid(256), stream=stream0)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf4, buf7, 64, 4, grid=grid(64, 4), stream=stream0)
buf8 = reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [product], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf6, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), out=buf8)
buf9 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
triton_poi_fused__softmax_add_mul_4.run(buf8, primals_7, buf9, buf10, 64, grid=grid(64), stream=stream0)
buf11 = reinterpret_tensor(buf8, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf8 # reuse
# Topologically Sorted Source Nodes: [product_1, product_2, weights], Original ATen: [aten.mul, aten.add, aten._softmax]
triton_poi_fused__softmax_add_mul_5.run(buf11, primals_7, buf9, buf10, 256, grid=grid(256), stream=stream0)
del primals_7
buf12 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf5, buf12, 256, grid=grid(256), stream=stream0)
buf13 = reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf11, (16, 4, 4), (16, 4, 1), 0), reinterpret_tensor(buf12, (16, 4, 4), (16, 4, 1), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf13, buf14, 256, grid=grid(256), stream=stream0)
del buf13
buf15 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [out_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf14, (16, 16), (16, 1), 0), reinterpret_tensor(primals_8, (16, 4), (1, 16), 0), out=buf15)
buf16 = buf1; del buf1 # reuse
buf17 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(buf15, primals_3, buf16, buf17, 16, grid=grid(16), stream=stream0)
buf18 = reinterpret_tensor(buf10, (4, 4, 4), (16, 4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [x_2, x_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_7.run(buf15, primals_3, buf16, buf17, primals_9, primals_10, buf18, 64, grid=grid(64), stream=stream0)
del buf16
del buf17
del primals_10
buf19 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf18, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf19)
buf20 = reinterpret_tensor(buf19, (4, 4, 4), (16, 4, 1), 0); del buf19 # reuse
buf23 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [hidden_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_8.run(buf20, primals_12, buf23, 64, grid=grid(64), stream=stream0)
del primals_12
buf21 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf20, (16, 4), (4, 1), 0), reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf21)
buf22 = reinterpret_tensor(buf21, (4, 4, 4), (16, 4, 1), 0); del buf21 # reuse
# Topologically Sorted Source Nodes: [x_2, x_5], Original ATen: [aten.add]
triton_poi_fused_add_9.run(buf22, primals_14, buf15, primals_3, 64, grid=grid(64), stream=stream0)
del primals_14
return (buf22, primals_3, primals_9, reinterpret_tensor(buf2, (16, 4), (4, 1), 0), buf11, reinterpret_tensor(buf14, (16, 16), (16, 1), 0), buf15, reinterpret_tensor(buf18, (16, 4), (4, 1), 0), reinterpret_tensor(buf20, (16, 4), (4, 1), 0), primals_13, buf23, primals_11, primals_8, reinterpret_tensor(buf12, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf6, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf7, (16, 4, 4), (16, 1, 4), 0), primals_6, primals_5, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 4), (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)
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])
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
from functools import partial
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super(FFN, self).__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
class MultiHeadAttention(nn.Module):
"""
Multi-Head Attention
"""
def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.d_key = d_key
self.d_value = d_value
self.d_model = d_model
self.dropout_rate = dropout_rate
self.q_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.k_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.v_fc = nn.Linear(in_features=d_model, out_features=d_value *
n_head, bias=False)
self.proj_fc = nn.Linear(in_features=d_value * n_head, out_features
=d_model, bias=False)
def _prepare_qkv(self, queries, keys, values, cache=None):
if keys is None:
keys, values = queries, queries
static_kv = False
else:
static_kv = True
q = self.q_fc(queries)
q = torch.reshape(q, shape=(q.shape[0], -1, self.n_head, self.d_key))
q = q.permute(0, 2, 1, 3)
if cache is not None and static_kv and 'static_k' in cache:
k = cache['static_k']
v = cache['static_v']
else:
k = self.k_fc(keys)
v = self.v_fc(values)
k = torch.reshape(k, shape=(k.shape[0], -1, self.n_head, self.
d_key))
k = k.permute(0, 2, 1, 3)
v = torch.reshape(v, shape=(v.shape[0], -1, self.n_head, self.
d_value))
v = v.permute(0, 2, 1, 3)
if cache is not None:
if static_kv and 'static_k' not in cache:
cache['static_k'], cache['static_v'] = k, v
elif not static_kv:
cache_k, cache_v = cache['k'], cache['v']
k = torch.cat((cache_k, k), dim=2)
v = torch.cat((cache_v, v), dim=2)
cache['k'], cache['v'] = k, v
return q, k, v
def forward(self, queries, keys, values, attn_bias, cache=None):
keys = queries if keys is None else keys
values = keys if values is None else values
q, k, v = self._prepare_qkv(queries, keys, values, cache)
product = torch.matmul(q, k.transpose(-2, -1))
product = product * self.d_model ** -0.5
if attn_bias is not None:
product += attn_bias
weights = F.softmax(product, dim=-1)
if self.dropout_rate:
weights = F.dropout(weights, p=self.dropout_rate, training=False)
out = torch.matmul(weights, v)
out = out.permute(0, 2, 1, 3)
out = torch.reshape(out, shape=(out.shape[0], -1, out.shape[2] *
out.shape[3]))
out = self.proj_fc(out)
return out
class PrePostProcessLayer(nn.Module):
"""
PrePostProcessLayer
"""
def __init__(self, process_cmd, d_model, dropout_rate):
super(PrePostProcessLayer, self).__init__()
self.process_cmd = process_cmd
self.functors = []
self.g = partial(self.f)
self.g1 = partial(self.f1, dropout_rate)
for cmd in self.process_cmd:
if cmd == 'a':
self.functors.append(self.g)
elif cmd == 'n':
num = sum(1 for line in self.children())
layernorm = nn.LayerNorm(normalized_shape=d_model)
self.add_module('layer_norm_%d' % num, layernorm)
self.functors.append(layernorm)
elif cmd == 'd':
self.functors.append(self.g1)
@staticmethod
def f(x, y):
if y is not None:
return x + y
else:
return x
@staticmethod
def f1(p, x):
if p:
return F.dropout(x, p, training=False)
else:
return x
def forward(self, x, residual=None):
for i, cmd in enumerate(self.process_cmd):
if cmd == 'a':
x = self.functors[i](x, residual)
else:
x = self.functors[i](x)
return x
class EncoderLayer(nn.Module):
"""
EncoderLayer
"""
def __init__(self, n_head, d_key, d_value, d_model, d_inner_hid,
prepostprocess_dropout, attention_dropout, relu_dropout,
preprocess_cmd='n', postprocess_cmd='da'):
super(EncoderLayer, self).__init__()
self.preprocesser1 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.self_attn = MultiHeadAttention(d_key, d_value, d_model, n_head,
attention_dropout)
self.postprocesser1 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
self.preprocesser2 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.ffn = FFN(d_inner_hid, d_model, relu_dropout)
self.postprocesser2 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
def forward(self, enc_input, attn_bias):
attn_output = self.self_attn(self.preprocesser1(enc_input), None,
None, attn_bias)
attn_output = self.postprocesser1(attn_output, enc_input)
ffn_output = self.ffn(self.preprocesser2(attn_output))
ffn_output = self.postprocesser2(ffn_output, attn_output)
return ffn_output
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'n_head': 4, 'd_key': 4, 'd_value': 4, 'd_model': 4,
'd_inner_hid': 4, 'prepostprocess_dropout': 0.5,
'attention_dropout': 0.5, 'relu_dropout': 0.5}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime 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
from functools import partial
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_native_layer_norm_0(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 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_1(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_clone_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
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1 + 64 * x3), xmask)
tl.store(out_ptr0 + x4, tmp0, xmask)
@triton.jit
def triton_poi_fused_clone_3(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)
@triton.jit
def triton_poi_fused__softmax_add_mul_4(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp17 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp5 * tmp1
tmp8 = tmp6 + tmp7
tmp9 = triton_helpers.maximum(tmp4, tmp8)
tmp11 = tmp10 * tmp1
tmp13 = tmp11 + tmp12
tmp14 = triton_helpers.maximum(tmp9, tmp13)
tmp16 = tmp15 * tmp1
tmp18 = tmp16 + tmp17
tmp19 = triton_helpers.maximum(tmp14, tmp18)
tmp20 = tmp4 - tmp19
tmp21 = tl_math.exp(tmp20)
tmp22 = tmp8 - tmp19
tmp23 = tl_math.exp(tmp22)
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp19
tmp26 = tl_math.exp(tmp25)
tmp27 = tmp24 + tmp26
tmp28 = tmp18 - tmp19
tmp29 = tl_math.exp(tmp28)
tmp30 = tmp27 + tmp29
tl.store(out_ptr0 + x2, tmp19, xmask)
tl.store(out_ptr1 + x2, tmp30, xmask)
@triton.jit
def triton_poi_fused__softmax_add_mul_5(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
x4 = xindex % 64
x5 = xindex // 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp3 = tl.load(in_ptr0 + x4, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + x5, xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr2 + x5, xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 - tmp5
tmp7 = tl_math.exp(tmp6)
tmp9 = tmp7 / tmp8
tl.store(in_out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_7(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_8(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
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_9(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_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp4 = tl.load(in_ptr2 + x2, xmask)
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) = 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), (16, 4, 1))
assert_size_stride(primals_4, (16, 4), (4, 1))
assert_size_stride(primals_5, (16, 4), (4, 1))
assert_size_stride(primals_6, (16, 4), (4, 1))
assert_size_stride(primals_7, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_8, (4, 16), (16, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4, 4), (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,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf1 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_native_layer_norm_0[grid(16)](primals_3, buf0,
buf1, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_1[grid(64)](primals_3, buf0,
buf1, primals_1, primals_2, buf2, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del primals_1
del primals_2
buf3 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 16), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 16), (1, 4), 0), out=buf4)
buf5 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 16), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(256)](buf3, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf7 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused_clone_3[grid(64, 4)](buf4, buf7, 64, 4, XBLOCK=4,
YBLOCK=32, num_warps=4, num_stages=1)
buf8 = reinterpret_tensor(buf4, (16, 4, 4), (16, 4, 1), 0)
del buf4
extern_kernels.bmm(reinterpret_tensor(buf6, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf7, (16, 4, 4), (16, 4, 1), 0), out=buf8)
buf9 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 64), torch.float32)
triton_poi_fused__softmax_add_mul_4[grid(64)](buf8, primals_7, buf9,
buf10, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf8, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf8
triton_poi_fused__softmax_add_mul_5[grid(256)](buf11, primals_7,
buf9, buf10, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_7
buf12 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(256)](buf5, buf12, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf13 = reinterpret_tensor(buf5, (16, 4, 4), (16, 4, 1), 0)
del buf5
extern_kernels.bmm(reinterpret_tensor(buf11, (16, 4, 4), (16, 4, 1),
0), reinterpret_tensor(buf12, (16, 4, 4), (16, 4, 1), 0), out=buf13
)
buf14 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_2[grid(256)](buf13, buf14, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf13
buf15 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.mm(reinterpret_tensor(buf14, (16, 16), (16, 1), 0),
reinterpret_tensor(primals_8, (16, 4), (1, 16), 0), out=buf15)
buf16 = buf1
del buf1
buf17 = buf0
del buf0
triton_poi_fused_add_native_layer_norm_6[grid(16)](buf15, primals_3,
buf16, buf17, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf18 = reinterpret_tensor(buf10, (4, 4, 4), (16, 4, 1), 0)
del buf10
triton_poi_fused_add_native_layer_norm_7[grid(64)](buf15, primals_3,
buf16, buf17, primals_9, primals_10, buf18, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf16
del buf17
del primals_10
buf19 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf18, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf19)
buf20 = reinterpret_tensor(buf19, (4, 4, 4), (16, 4, 1), 0)
del buf19
buf23 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_8[grid(64)](buf20,
primals_12, buf23, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_12
buf21 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf20, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf21)
buf22 = reinterpret_tensor(buf21, (4, 4, 4), (16, 4, 1), 0)
del buf21
triton_poi_fused_add_9[grid(64)](buf22, primals_14, buf15,
primals_3, 64, XBLOCK=64, num_warps=1, num_stages=1)
del primals_14
return buf22, primals_3, primals_9, reinterpret_tensor(buf2, (16, 4), (
4, 1), 0), buf11, reinterpret_tensor(buf14, (16, 16), (16, 1), 0
), buf15, reinterpret_tensor(buf18, (16, 4), (4, 1), 0
), reinterpret_tensor(buf20, (16, 4), (4, 1), 0
), primals_13, buf23, primals_11, primals_8, reinterpret_tensor(buf12,
(16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf6, (16, 4, 4), (
16, 1, 4), 0), reinterpret_tensor(buf7, (16, 4, 4), (16, 1, 4), 0
), primals_6, primals_5, primals_4
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super(FFN, self).__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
class MultiHeadAttention(nn.Module):
"""
Multi-Head Attention
"""
def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
super(MultiHeadAttention, self).__init__()
self.n_head = n_head
self.d_key = d_key
self.d_value = d_value
self.d_model = d_model
self.dropout_rate = dropout_rate
self.q_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.k_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.v_fc = nn.Linear(in_features=d_model, out_features=d_value *
n_head, bias=False)
self.proj_fc = nn.Linear(in_features=d_value * n_head, out_features
=d_model, bias=False)
def _prepare_qkv(self, queries, keys, values, cache=None):
if keys is None:
keys, values = queries, queries
static_kv = False
else:
static_kv = True
q = self.q_fc(queries)
q = torch.reshape(q, shape=(q.shape[0], -1, self.n_head, self.d_key))
q = q.permute(0, 2, 1, 3)
if cache is not None and static_kv and 'static_k' in cache:
k = cache['static_k']
v = cache['static_v']
else:
k = self.k_fc(keys)
v = self.v_fc(values)
k = torch.reshape(k, shape=(k.shape[0], -1, self.n_head, self.
d_key))
k = k.permute(0, 2, 1, 3)
v = torch.reshape(v, shape=(v.shape[0], -1, self.n_head, self.
d_value))
v = v.permute(0, 2, 1, 3)
if cache is not None:
if static_kv and 'static_k' not in cache:
cache['static_k'], cache['static_v'] = k, v
elif not static_kv:
cache_k, cache_v = cache['k'], cache['v']
k = torch.cat((cache_k, k), dim=2)
v = torch.cat((cache_v, v), dim=2)
cache['k'], cache['v'] = k, v
return q, k, v
def forward(self, queries, keys, values, attn_bias, cache=None):
keys = queries if keys is None else keys
values = keys if values is None else values
q, k, v = self._prepare_qkv(queries, keys, values, cache)
product = torch.matmul(q, k.transpose(-2, -1))
product = product * self.d_model ** -0.5
if attn_bias is not None:
product += attn_bias
weights = F.softmax(product, dim=-1)
if self.dropout_rate:
weights = F.dropout(weights, p=self.dropout_rate, training=False)
out = torch.matmul(weights, v)
out = out.permute(0, 2, 1, 3)
out = torch.reshape(out, shape=(out.shape[0], -1, out.shape[2] *
out.shape[3]))
out = self.proj_fc(out)
return out
class PrePostProcessLayer(nn.Module):
"""
PrePostProcessLayer
"""
def __init__(self, process_cmd, d_model, dropout_rate):
super(PrePostProcessLayer, self).__init__()
self.process_cmd = process_cmd
self.functors = []
self.g = partial(self.f)
self.g1 = partial(self.f1, dropout_rate)
for cmd in self.process_cmd:
if cmd == 'a':
self.functors.append(self.g)
elif cmd == 'n':
num = sum(1 for line in self.children())
layernorm = nn.LayerNorm(normalized_shape=d_model)
self.add_module('layer_norm_%d' % num, layernorm)
self.functors.append(layernorm)
elif cmd == 'd':
self.functors.append(self.g1)
@staticmethod
def f(x, y):
if y is not None:
return x + y
else:
return x
@staticmethod
def f1(p, x):
if p:
return F.dropout(x, p, training=False)
else:
return x
def forward(self, x, residual=None):
for i, cmd in enumerate(self.process_cmd):
if cmd == 'a':
x = self.functors[i](x, residual)
else:
x = self.functors[i](x)
return x
class EncoderLayerNew(nn.Module):
"""
EncoderLayer
"""
def __init__(self, n_head, d_key, d_value, d_model, d_inner_hid,
prepostprocess_dropout, attention_dropout, relu_dropout,
preprocess_cmd='n', postprocess_cmd='da'):
super(EncoderLayerNew, self).__init__()
self.preprocesser1 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.self_attn = MultiHeadAttention(d_key, d_value, d_model, n_head,
attention_dropout)
self.postprocesser1 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
self.preprocesser2 = PrePostProcessLayer(preprocess_cmd, d_model,
prepostprocess_dropout)
self.ffn = FFN(d_inner_hid, d_model, relu_dropout)
self.postprocesser2 = PrePostProcessLayer(postprocess_cmd, d_model,
prepostprocess_dropout)
def forward(self, input_0, input_1):
primals_1 = self.preprocesser1.layer_norm_0.weight
primals_2 = self.preprocesser1.layer_norm_0.bias
primals_4 = self.self_attn.q_fc.weight
primals_5 = self.self_attn.k_fc.weight
primals_6 = self.self_attn.v_fc.weight
primals_8 = self.self_attn.proj_fc.weight
primals_9 = self.preprocesser2.layer_norm_0.weight
primals_10 = self.preprocesser2.layer_norm_0.bias
primals_11 = self.ffn.fc1.weight
primals_12 = self.ffn.fc1.bias
primals_13 = self.ffn.fc2.weight
primals_14 = self.ffn.fc2.bias
primals_3 = input_0
primals_7 = 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])
return output[0]
| DocYard-ai/UCR | EncoderLayer | false | 8,026 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
import torch.nn.functional as F
from functools import partial
class FFN(nn.Module):
"""
Feed-Forward Network
"""
def __init__(self, d_inner_hid, d_model, dropout_rate):
super().__init__()
self.dropout_rate = dropout_rate
self.fc1 = nn.Linear(in_features=d_model, out_features=d_inner_hid)
self.fc2 = nn.Linear(in_features=d_inner_hid, out_features=d_model)
def forward(self, x):
hidden = self.fc1(x)
hidden = F.relu(hidden)
if self.dropout_rate:
hidden = F.dropout(hidden, p=self.dropout_rate, training=False)
out = self.fc2(hidden)
return out
class MultiHeadAttention(nn.Module):
"""
Multi-Head Attention
"""
def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
super().__init__()
self.n_head = n_head
self.d_key = d_key
self.d_value = d_value
self.d_model = d_model
self.dropout_rate = dropout_rate
self.q_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.k_fc = nn.Linear(in_features=d_model, out_features=d_key *
n_head, bias=False)
self.v_fc = nn.Linear(in_features=d_model, out_features=d_value *
n_head, bias=False)
self.proj_fc = nn.Linear(in_features=d_value * n_head, out_features
=d_model, bias=False)
def _prepare_qkv(self, queries, keys, values, cache=None):
if keys is None:
keys, values = queries, queries
static_kv = False
else:
static_kv = True
q = self.q_fc(queries)
q = torch.reshape(q, shape=(q.shape[0], -1, self.n_head, self.d_key))
q = q.permute(0, 2, 1, 3)
if cache is not None and static_kv and 'static_k' in cache:
k = cache['static_k']
v = cache['static_v']
else:
k = self.k_fc(keys)
v = self.v_fc(values)
k = torch.reshape(k, shape=(k.shape[0], -1, self.n_head, self.
d_key))
k = k.permute(0, 2, 1, 3)
v = torch.reshape(v, shape=(v.shape[0], -1, self.n_head, self.
d_value))
v = v.permute(0, 2, 1, 3)
if cache is not None:
if static_kv and 'static_k' not in cache:
cache['static_k'], cache['static_v'] = k, v
elif not static_kv:
cache_k, cache_v = cache['k'], cache['v']
k = torch.cat((cache_k, k), dim=2)
v = torch.cat((cache_v, v), dim=2)
cache['k'], cache['v'] = k, v
return q, k, v
def forward(self, queries, keys, values, attn_bias, cache=None):
keys = queries if keys is None else keys
values = keys if values is None else values
q, k, v = self._prepare_qkv(queries, keys, values, cache)
product = torch.matmul(q, k.transpose(-2, -1))
product = product * self.d_model ** -0.5
if attn_bias is not None:
product += attn_bias
weights = F.softmax(product, dim=-1)
if self.dropout_rate:
weights = F.dropout(weights, p=self.dropout_rate, training=False)
out = torch.matmul(weights, v)
out = out.permute(0, 2, 1, 3)
out = torch.reshape(out, shape=(out.shape[0], -1, out.shape[2] *
out.shape[3]))
out = self.proj_fc(out)
return out
class PrePostProcessLayer(nn.Module):
"""
PrePostProcessLayer
"""
def __init__(self, process_cmd, d_model, dropout_rate):
super().__init__()
self.process_cmd = process_cmd
self.functors = []
self.g = partial(self.f)
self.g1 = partial(self.f1, dropout_rate)
for cmd in self.process_cmd:
if cmd == 'a':
self.functors.append(self.g)
elif cmd == 'n':
num = sum(1 for line in self.children())
# ... truncated (>4000 chars) for memory efficiency |
SpatialLogSoftmax | # 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_1/inductor_cache/hw/chw3eev3so6cgjavghhlmuimvqx7kttdwf4y5ocackdwuxwr7vmo.py
# Topologically Sorted Source Nodes: [marginal, sm, marginal_1, sm_1], Original ATen: [aten.mean, aten._log_softmax]
# Source node to ATen node mapping:
# marginal => mean
# marginal_1 => mean_1
# sm => amax
# sm_1 => amax_1
# Graph fragment:
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [3]), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mean, [2], True), kwargs = {})
# %mean_1 : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [2]), kwargs = {})
# %amax_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mean_1, [2], True), kwargs = {})
triton_poi_fused__log_softmax_mean_0 = async_compile.triton('triton_poi_fused__log_softmax_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mean_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 16, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mean_0(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 + (16*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (16*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (16*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (16*x0)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (4 + (16*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (5 + (16*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (6 + (16*x0)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (7 + (16*x0)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (8 + (16*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (9 + (16*x0)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (10 + (16*x0)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (11 + (16*x0)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (12 + (16*x0)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr0 + (13 + (16*x0)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr0 + (14 + (16*x0)), xmask, eviction_policy='evict_last')
tmp32 = tl.load(in_ptr0 + (15 + (16*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp11 = tmp9 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tmp13 + tmp14
tmp16 = tmp15 / tmp7
tmp17 = triton_helpers.maximum(tmp8, tmp16)
tmp20 = tmp18 + tmp19
tmp22 = tmp20 + tmp21
tmp24 = tmp22 + tmp23
tmp25 = tmp24 / tmp7
tmp26 = triton_helpers.maximum(tmp17, tmp25)
tmp29 = tmp27 + tmp28
tmp31 = tmp29 + tmp30
tmp33 = tmp31 + tmp32
tmp34 = tmp33 / tmp7
tmp35 = triton_helpers.maximum(tmp26, tmp34)
tmp36 = tmp0 + tmp9
tmp37 = tmp36 + tmp18
tmp38 = tmp37 + tmp27
tmp39 = tmp38 / tmp7
tmp40 = tmp1 + tmp10
tmp41 = tmp40 + tmp19
tmp42 = tmp41 + tmp28
tmp43 = tmp42 / tmp7
tmp44 = triton_helpers.maximum(tmp39, tmp43)
tmp45 = tmp3 + tmp12
tmp46 = tmp45 + tmp21
tmp47 = tmp46 + tmp30
tmp48 = tmp47 / tmp7
tmp49 = triton_helpers.maximum(tmp44, tmp48)
tmp50 = tmp5 + tmp14
tmp51 = tmp50 + tmp23
tmp52 = tmp51 + tmp32
tmp53 = tmp52 / tmp7
tmp54 = triton_helpers.maximum(tmp49, tmp53)
tl.store(out_ptr0 + (x0), tmp35, xmask)
tl.store(out_ptr1 + (x0), tmp54, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/5u/c5uwzm4jwygmr3lewuhx2phahw4nd72lwnsljfhyqju52dzgpaxu.py
# Topologically Sorted Source Nodes: [marginal, sm], Original ATen: [aten.mean, aten._log_softmax]
# Source node to ATen node mapping:
# marginal => mean
# sm => sub
# Graph fragment:
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [3]), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mean, %amax), kwargs = {})
triton_poi_fused__log_softmax_mean_1 = async_compile.triton('triton_poi_fused__log_softmax_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mean_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mean_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/dr/cdrhgtgnf3uts3skt4d3dgxxnhj6xyyalvz25tlvdmttvx2tfe4n.py
# Topologically Sorted Source Nodes: [sm, add], Original ATen: [aten._log_softmax, aten.add]
# Source node to ATen node mapping:
# add => add_1
# sm => 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, [2], 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 = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_1, %expand), kwargs = {})
triton_poi_fused__log_softmax_add_2 = async_compile.triton('triton_poi_fused__log_softmax_add_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__log_softmax_add_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = 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
tmp14 = x0
tmp15 = tmp14.to(tl.float32)
tmp16 = 2.0
tmp17 = tmp15 < tmp16
tmp18 = 0.3333333333333333
tmp19 = tmp15 * tmp18
tmp20 = 0.0
tmp21 = tmp19 + tmp20
tmp22 = 3 + ((-1)*x0)
tmp23 = tmp22.to(tl.float32)
tmp24 = tmp23 * tmp18
tmp25 = 1.0
tmp26 = tmp25 - tmp24
tmp27 = tl.where(tmp17, tmp21, tmp26)
tmp28 = tl_math.log(tmp27)
tmp29 = tmp13 + tmp28
tl.store(out_ptr0 + (x2), tmp29, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zx/czxmwae6zioiifsix7ejcwb4yspjodqymp2rqpzfe2efg6ayen6z.py
# Topologically Sorted Source Nodes: [marginal_1, sm_1], Original ATen: [aten.mean, aten._log_softmax]
# Source node to ATen node mapping:
# marginal_1 => mean_1
# sm_1 => sub_2
# Graph fragment:
# %mean_1 : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [2]), kwargs = {})
# %sub_2 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mean_1, %amax_1), kwargs = {})
triton_poi_fused__log_softmax_mean_3 = async_compile.triton('triton_poi_fused__log_softmax_mean_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mean_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_mean_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + (16*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + (16*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + (16*x1)), xmask)
tmp9 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/we/cweortkzvuk3ueqabs5edhx2rs7jgqjahomswfsjye7fi733l3gp.py
# Topologically Sorted Source Nodes: [stack], Original ATen: [aten.stack]
# Source node to ATen node mapping:
# stack => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze, %unsqueeze_1], 2), kwargs = {})
triton_poi_fused_stack_4 = async_compile.triton('triton_poi_fused_stack_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=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_stack_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_stack_4(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 2
x1 = (xindex // 2)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4*x1), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.load(in_ptr0 + (1 + (4*x1)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp8 = tl_math.exp(tmp7)
tmp9 = tmp6 + tmp8
tmp10 = tl.load(in_ptr0 + (2 + (4*x1)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tl_math.exp(tmp10)
tmp12 = tmp9 + tmp11
tmp13 = tl.load(in_ptr0 + (3 + (4*x1)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp14 = tl_math.exp(tmp13)
tmp15 = tmp12 + tmp14
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp4, tmp15, tmp16)
tmp18 = tmp0 >= tmp3
tmp19 = tl.full([1], 2, tl.int64)
tmp20 = tmp0 < tmp19
tmp21 = tl.load(in_ptr1 + (4*x1), tmp18 & xmask, eviction_policy='evict_last', other=0.0)
tmp22 = tl_math.exp(tmp21)
tmp23 = tl.load(in_ptr1 + (1 + (4*x1)), tmp18 & xmask, eviction_policy='evict_last', other=0.0)
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tl.load(in_ptr1 + (2 + (4*x1)), tmp18 & xmask, eviction_policy='evict_last', other=0.0)
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tl.load(in_ptr1 + (3 + (4*x1)), tmp18 & xmask, eviction_policy='evict_last', other=0.0)
tmp30 = tl_math.exp(tmp29)
tmp31 = tmp28 + tmp30
tmp32 = tl.full(tmp31.shape, 0.0, tmp31.dtype)
tmp33 = tl.where(tmp18, tmp31, tmp32)
tmp34 = tl.where(tmp4, tmp17, tmp33)
tl.store(out_ptr0 + (x2), tmp34, 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, 1), (4, 1, 16), torch.float32)
buf3 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [marginal, sm, marginal_1, sm_1], Original ATen: [aten.mean, aten._log_softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__log_softmax_mean_0.run(arg0_1, buf0, buf3, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [marginal, sm], Original ATen: [aten.mean, aten._log_softmax]
triton_poi_fused__log_softmax_mean_1.run(arg0_1, buf0, buf1, 64, grid=grid(64), stream=stream0)
del buf0
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sm, add], Original ATen: [aten._log_softmax, aten.add]
triton_poi_fused__log_softmax_add_2.run(buf1, buf2, 64, grid=grid(64), stream=stream0)
buf4 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [marginal_1, sm_1], Original ATen: [aten.mean, aten._log_softmax]
triton_poi_fused__log_softmax_mean_3.run(arg0_1, buf3, buf4, 64, grid=grid(64), stream=stream0)
del arg0_1
del buf3
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sm_1, add_1], Original ATen: [aten._log_softmax, aten.add]
triton_poi_fused__log_softmax_add_2.run(buf4, buf5, 64, grid=grid(64), stream=stream0)
del buf4
buf6 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [stack], Original ATen: [aten.stack]
triton_poi_fused_stack_4.run(buf2, buf5, buf6, 32, grid=grid(32), stream=stream0)
del buf2
del buf5
return (buf6, )
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.utils.data
import torch.random
import torch.nn.functional as F
def logprob_to_keypoints(prob, length):
ruler = torch.log(torch.linspace(0, 1, length, device=prob.device)
).type_as(prob).expand(1, 1, -1)
return torch.sum(torch.exp(prob + ruler), dim=2, keepdim=True).squeeze(2)
def marginal_logsoftmax(heatmap, dim):
marginal = torch.mean(heatmap, dim=dim)
sm = F.log_softmax(marginal, dim=2)
return sm
def spacial_logsoftmax(heatmap, probs=False):
height, width = heatmap.size(2), heatmap.size(3)
hp, wp = marginal_logsoftmax(heatmap, dim=3), marginal_logsoftmax(heatmap,
dim=2)
hk, wk = logprob_to_keypoints(hp, height), logprob_to_keypoints(wp, width)
if probs:
return torch.stack((hk, wk), dim=2), (torch.exp(hp), torch.exp(wp))
else:
return torch.stack((hk, wk), dim=2)
class SpatialLogSoftmax(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, heatmap, probs=False):
return spacial_logsoftmax(heatmap, probs)
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
import torch.utils.data
import torch.random
import torch.nn.functional as F
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused__log_softmax_mean_0(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 + 16 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr0 + (2 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr0 + (3 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp9 = tl.load(in_ptr0 + (4 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp10 = tl.load(in_ptr0 + (5 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp12 = tl.load(in_ptr0 + (6 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp14 = tl.load(in_ptr0 + (7 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp18 = tl.load(in_ptr0 + (8 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr0 + (9 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (10 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr0 + (11 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp27 = tl.load(in_ptr0 + (12 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp28 = tl.load(in_ptr0 + (13 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp30 = tl.load(in_ptr0 + (14 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp32 = tl.load(in_ptr0 + (15 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp11 = tmp9 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tmp13 + tmp14
tmp16 = tmp15 / tmp7
tmp17 = triton_helpers.maximum(tmp8, tmp16)
tmp20 = tmp18 + tmp19
tmp22 = tmp20 + tmp21
tmp24 = tmp22 + tmp23
tmp25 = tmp24 / tmp7
tmp26 = triton_helpers.maximum(tmp17, tmp25)
tmp29 = tmp27 + tmp28
tmp31 = tmp29 + tmp30
tmp33 = tmp31 + tmp32
tmp34 = tmp33 / tmp7
tmp35 = triton_helpers.maximum(tmp26, tmp34)
tmp36 = tmp0 + tmp9
tmp37 = tmp36 + tmp18
tmp38 = tmp37 + tmp27
tmp39 = tmp38 / tmp7
tmp40 = tmp1 + tmp10
tmp41 = tmp40 + tmp19
tmp42 = tmp41 + tmp28
tmp43 = tmp42 / tmp7
tmp44 = triton_helpers.maximum(tmp39, tmp43)
tmp45 = tmp3 + tmp12
tmp46 = tmp45 + tmp21
tmp47 = tmp46 + tmp30
tmp48 = tmp47 / tmp7
tmp49 = triton_helpers.maximum(tmp44, tmp48)
tmp50 = tmp5 + tmp14
tmp51 = tmp50 + tmp23
tmp52 = tmp51 + tmp32
tmp53 = tmp52 / tmp7
tmp54 = triton_helpers.maximum(tmp49, tmp53)
tl.store(out_ptr0 + x0, tmp35, xmask)
tl.store(out_ptr1 + x0, tmp54, xmask)
@triton.jit
def triton_poi_fused__log_softmax_mean_1(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused__log_softmax_add_2(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = 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
tmp14 = x0
tmp15 = tmp14.to(tl.float32)
tmp16 = 2.0
tmp17 = tmp15 < tmp16
tmp18 = 0.3333333333333333
tmp19 = tmp15 * tmp18
tmp20 = 0.0
tmp21 = tmp19 + tmp20
tmp22 = 3 + -1 * x0
tmp23 = tmp22.to(tl.float32)
tmp24 = tmp23 * tmp18
tmp25 = 1.0
tmp26 = tmp25 - tmp24
tmp27 = tl.where(tmp17, tmp21, tmp26)
tmp28 = tl_math.log(tmp27)
tmp29 = tmp13 + tmp28
tl.store(out_ptr0 + x2, tmp29, xmask)
@triton.jit
def triton_poi_fused__log_softmax_mean_3(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + 16 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + 16 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + 16 * x1), xmask)
tmp9 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_stack_4(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 2
x1 = xindex // 2
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + 4 * x1, tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.load(in_ptr0 + (1 + 4 * x1), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp8 = tl_math.exp(tmp7)
tmp9 = tmp6 + tmp8
tmp10 = tl.load(in_ptr0 + (2 + 4 * x1), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp11 = tl_math.exp(tmp10)
tmp12 = tmp9 + tmp11
tmp13 = tl.load(in_ptr0 + (3 + 4 * x1), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp14 = tl_math.exp(tmp13)
tmp15 = tmp12 + tmp14
tmp16 = tl.full(tmp15.shape, 0.0, tmp15.dtype)
tmp17 = tl.where(tmp4, tmp15, tmp16)
tmp18 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp21 = tl.load(in_ptr1 + 4 * x1, tmp18 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp22 = tl_math.exp(tmp21)
tmp23 = tl.load(in_ptr1 + (1 + 4 * x1), tmp18 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tl.load(in_ptr1 + (2 + 4 * x1), tmp18 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp27 = tl_math.exp(tmp26)
tmp28 = tmp25 + tmp27
tmp29 = tl.load(in_ptr1 + (3 + 4 * x1), tmp18 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp30 = tl_math.exp(tmp29)
tmp31 = tmp28 + tmp30
tmp32 = tl.full(tmp31.shape, 0.0, tmp31.dtype)
tmp33 = tl.where(tmp18, tmp31, tmp32)
tmp34 = tl.where(tmp4, tmp17, tmp33)
tl.store(out_ptr0 + x2, tmp34, 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, 1), (4, 1, 16), torch.float32)
buf3 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused__log_softmax_mean_0[grid(16)](arg0_1, buf0, buf3,
16, XBLOCK=16, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__log_softmax_mean_1[grid(64)](arg0_1, buf0, buf1,
64, XBLOCK=64, num_warps=1, num_stages=1)
del buf0
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__log_softmax_add_2[grid(64)](buf1, buf2, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf4 = buf1
del buf1
triton_poi_fused__log_softmax_mean_3[grid(64)](arg0_1, buf3, buf4,
64, XBLOCK=64, num_warps=1, num_stages=1)
del arg0_1
del buf3
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__log_softmax_add_2[grid(64)](buf4, buf5, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del buf4
buf6 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
triton_poi_fused_stack_4[grid(32)](buf2, buf5, buf6, 32, XBLOCK=32,
num_warps=1, num_stages=1)
del buf2
del buf5
return buf6,
def logprob_to_keypoints(prob, length):
ruler = torch.log(torch.linspace(0, 1, length, device=prob.device)
).type_as(prob).expand(1, 1, -1)
return torch.sum(torch.exp(prob + ruler), dim=2, keepdim=True).squeeze(2)
def marginal_logsoftmax(heatmap, dim):
marginal = torch.mean(heatmap, dim=dim)
sm = F.log_softmax(marginal, dim=2)
return sm
def spacial_logsoftmax(heatmap, probs=False):
height, width = heatmap.size(2), heatmap.size(3)
hp, wp = marginal_logsoftmax(heatmap, dim=3), marginal_logsoftmax(heatmap,
dim=2)
hk, wk = logprob_to_keypoints(hp, height), logprob_to_keypoints(wp, width)
if probs:
return torch.stack((hk, wk), dim=2), (torch.exp(hp), torch.exp(wp))
else:
return torch.stack((hk, wk), dim=2)
class SpatialLogSoftmaxNew(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DuaneNielsen/keypoints | SpatialLogSoftmax | false | 8,027 | [
"MIT"
] | 42 | 302fa02966d4372ac9b5aaa3d8dc24684be0b252 | https://github.com/DuaneNielsen/keypoints/tree/302fa02966d4372ac9b5aaa3d8dc24684be0b252 | import torch
import torch.utils.data
import torch.random
import torch.nn.functional as F
def logprob_to_keypoints(prob, length):
ruler = torch.log(torch.linspace(0, 1, length, device=prob.device)
).type_as(prob).expand(1, 1, -1)
return torch.sum(torch.exp(prob + ruler), dim=2, keepdim=True).squeeze(2)
def marginal_logsoftmax(heatmap, dim):
marginal = torch.mean(heatmap, dim=dim)
sm = F.log_softmax(marginal, dim=2)
return sm
def spacial_logsoftmax(heatmap, probs=False):
height, width = heatmap.size(2), heatmap.size(3)
hp, wp = marginal_logsoftmax(heatmap, dim=3), marginal_logsoftmax(heatmap,
dim=2)
hk, wk = logprob_to_keypoints(hp, height), logprob_to_keypoints(wp, width)
if probs:
return torch.stack((hk, wk), dim=2), (torch.exp(hp), torch.exp(wp))
else:
return torch.stack((hk, wk), dim=2)
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, heatmap, probs=False):
return spacial_logsoftmax(heatmap, probs)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
EqualConv2d | # 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_1/inductor_cache/l6/cl6bc7pdn46tqb3csxvust3qvusxlvnmm7fro7xbgig23fwkvu2x.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, 0.125), 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.125
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vb/cvbno3dccglzmlbisnwicoai3aocrgweun3buh6avsdqdjjhjczh.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# out => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %mul, %primals_2, [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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((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(primals_1, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(primals_3, buf0, 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, 1, 1), (4, 1, 1, 1))
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf2, primals_2, 16, grid=grid(16), stream=stream0)
del primals_2
return (buf2, primals_3, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((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, 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 math
import torch
from torch import nn
import torch.nn.functional as F
class EqualConv2d(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, stride=1,
padding=0, bias=True, replicate_pad=False):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_channel, in_channel,
kernel_size, kernel_size))
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
self.replicate_pad = replicate_pad
def forward(self, input):
pad = self.padding
if self.replicate_pad:
input = F.pad(input, (pad, pad, pad, pad), mode='replicate')
pad = 0
out = F.conv2d(input, self.weight * self.scale, bias=self.bias,
stride=self.stride, padding=pad)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]}, {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})'
)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'out_channel': 4, 'kernel_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import 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
@triton.jit
def triton_poi_fused_mul_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.125
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(256)](primals_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_1
buf1 = extern_kernels.convolution(primals_3, buf0, 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, 1, 1), (4, 1, 1, 1))
buf2 = buf1
del buf1
triton_poi_fused_convolution_1[grid(16)](buf2, primals_2, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_2
return buf2, primals_3, buf0
class EqualConv2dNew(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, stride=1,
padding=0, bias=True, replicate_pad=False):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_channel, in_channel,
kernel_size, kernel_size))
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
self.replicate_pad = replicate_pad
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]}, {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})'
)
def forward(self, input_0):
primals_1 = self.weight
primals_2 = self.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| Dolorousrtur/style-people | EqualConv2d | false | 8,028 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | import math
import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, stride=1,
padding=0, bias=True, replicate_pad=False):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_channel, in_channel,
kernel_size, kernel_size))
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
self.replicate_pad = replicate_pad
def forward(self, input):
pad = self.padding
if self.replicate_pad:
input = F.pad(input, (pad, pad, pad, pad), mode='replicate')
pad = 0
out = F.conv2d(input, self.weight * self.scale, bias=self.bias,
stride=self.stride, padding=pad)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]}, {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})'
)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
AdaptiveAvgMaxPool2d | # 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_1/inductor_cache/sm/csmyxxnkficg6lg6sqykwigmxevsw6zlqryjbycdou4l3ckh5akm.py
# Topologically Sorted Source Nodes: [x_max, x_avg, add, mul], Original ATen: [aten.adaptive_max_pool2d, aten.mean, aten.add, aten.mul]
# Source node to ATen node mapping:
# add => add
# mul => mul
# x_avg => mean
# x_max => adaptive_max_pool2d
# Graph fragment:
# %adaptive_max_pool2d : [num_users=1] = call_function[target=torch.ops.aten.adaptive_max_pool2d.default](args = (%arg0_1, [1, 1]), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [-1, -2], True), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mean, %getitem), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, 0.5), kwargs = {})
triton_per_fused_adaptive_max_pool2d_add_mean_mul_0 = async_compile.triton('triton_per_fused_adaptive_max_pool2d_add_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=[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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_adaptive_max_pool2d_add_mean_mul_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 17, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_adaptive_max_pool2d_add_mean_mul_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)
tmp5 = tl.load(in_ptr0 + (16*x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + (16*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (2 + (16*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (3 + (16*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (4 + (16*x0)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (5 + (16*x0)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr0 + (6 + (16*x0)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (7 + (16*x0)), xmask, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr0 + (8 + (16*x0)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr0 + (9 + (16*x0)), xmask, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr0 + (10 + (16*x0)), xmask, eviction_policy='evict_last')
tmp26 = tl.load(in_ptr0 + (11 + (16*x0)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr0 + (12 + (16*x0)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr0 + (13 + (16*x0)), xmask, eviction_policy='evict_last')
tmp32 = tl.load(in_ptr0 + (14 + (16*x0)), xmask, eviction_policy='evict_last')
tmp34 = tl.load(in_ptr0 + (15 + (16*x0)), xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp7 = triton_helpers.maximum(tmp6, tmp5)
tmp9 = triton_helpers.maximum(tmp8, tmp7)
tmp11 = triton_helpers.maximum(tmp10, tmp9)
tmp13 = triton_helpers.maximum(tmp12, tmp11)
tmp15 = triton_helpers.maximum(tmp14, tmp13)
tmp17 = triton_helpers.maximum(tmp16, tmp15)
tmp19 = triton_helpers.maximum(tmp18, tmp17)
tmp21 = triton_helpers.maximum(tmp20, tmp19)
tmp23 = triton_helpers.maximum(tmp22, tmp21)
tmp25 = triton_helpers.maximum(tmp24, tmp23)
tmp27 = triton_helpers.maximum(tmp26, tmp25)
tmp29 = triton_helpers.maximum(tmp28, tmp27)
tmp31 = triton_helpers.maximum(tmp30, tmp29)
tmp33 = triton_helpers.maximum(tmp32, tmp31)
tmp35 = triton_helpers.maximum(tmp34, tmp33)
tmp36 = 16.0
tmp37 = tmp4 / tmp36
tmp38 = tmp37 + tmp35
tmp39 = 0.5
tmp40 = tmp38 * tmp39
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp40, 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, 1, 1), (4, 1, 16, 16), torch.float32)
buf2 = reinterpret_tensor(buf0, (4, 4, 1, 1), (4, 1, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [x_max, x_avg, add, mul], Original ATen: [aten.adaptive_max_pool2d, aten.mean, aten.add, aten.mul]
stream0 = get_raw_stream(0)
triton_per_fused_adaptive_max_pool2d_add_mean_mul_0.run(buf2, arg0_1, 16, 16, grid=grid(16), stream=stream0)
del arg0_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, 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 torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
def adaptive_avgmax_pool2d(x, output_size=1):
x_avg = F.adaptive_avg_pool2d(x, output_size)
x_max = F.adaptive_max_pool2d(x, output_size)
return 0.5 * (x_avg + x_max)
class AdaptiveAvgMaxPool2d(nn.Module):
def __init__(self, output_size=1):
super(AdaptiveAvgMaxPool2d, self).__init__()
self.output_size = output_size
def forward(self, x):
return adaptive_avgmax_pool2d(x, self.output_size)
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 torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
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_adaptive_max_pool2d_add_mean_mul_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)
tmp5 = tl.load(in_ptr0 + 16 * x0, xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (1 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp8 = tl.load(in_ptr0 + (2 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp10 = tl.load(in_ptr0 + (3 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp12 = tl.load(in_ptr0 + (4 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp14 = tl.load(in_ptr0 + (5 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr0 + (6 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp18 = tl.load(in_ptr0 + (7 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp20 = tl.load(in_ptr0 + (8 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr0 + (9 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp24 = tl.load(in_ptr0 + (10 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp26 = tl.load(in_ptr0 + (11 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp28 = tl.load(in_ptr0 + (12 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp30 = tl.load(in_ptr0 + (13 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp32 = tl.load(in_ptr0 + (14 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp34 = tl.load(in_ptr0 + (15 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp7 = triton_helpers.maximum(tmp6, tmp5)
tmp9 = triton_helpers.maximum(tmp8, tmp7)
tmp11 = triton_helpers.maximum(tmp10, tmp9)
tmp13 = triton_helpers.maximum(tmp12, tmp11)
tmp15 = triton_helpers.maximum(tmp14, tmp13)
tmp17 = triton_helpers.maximum(tmp16, tmp15)
tmp19 = triton_helpers.maximum(tmp18, tmp17)
tmp21 = triton_helpers.maximum(tmp20, tmp19)
tmp23 = triton_helpers.maximum(tmp22, tmp21)
tmp25 = triton_helpers.maximum(tmp24, tmp23)
tmp27 = triton_helpers.maximum(tmp26, tmp25)
tmp29 = triton_helpers.maximum(tmp28, tmp27)
tmp31 = triton_helpers.maximum(tmp30, tmp29)
tmp33 = triton_helpers.maximum(tmp32, tmp31)
tmp35 = triton_helpers.maximum(tmp34, tmp33)
tmp36 = 16.0
tmp37 = tmp4 / tmp36
tmp38 = tmp37 + tmp35
tmp39 = 0.5
tmp40 = tmp38 * tmp39
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp40, 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, 1, 1), (4, 1, 16, 16), torch.float32)
buf2 = reinterpret_tensor(buf0, (4, 4, 1, 1), (4, 1, 1, 1), 0)
del buf0
get_raw_stream(0)
triton_per_fused_adaptive_max_pool2d_add_mean_mul_0[grid(16)](buf2,
arg0_1, 16, 16, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
return buf2,
def adaptive_avgmax_pool2d(x, output_size=1):
x_avg = F.adaptive_avg_pool2d(x, output_size)
x_max = F.adaptive_max_pool2d(x, output_size)
return 0.5 * (x_avg + x_max)
class AdaptiveAvgMaxPool2dNew(nn.Module):
def __init__(self, output_size=1):
super(AdaptiveAvgMaxPool2dNew, self).__init__()
self.output_size = output_size
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Alibaba-MIIL/ZS_SDL | AdaptiveAvgMaxPool2d | false | 8,029 | [
"MIT"
] | 20 | 769fe4f57d2d458a7c4b5468a6395c9b296b1dad | https://github.com/Alibaba-MIIL/ZS_SDL/tree/769fe4f57d2d458a7c4b5468a6395c9b296b1dad | import torch
from torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
def adaptive_avgmax_pool2d(x, output_size=1):
x_avg = F.adaptive_avg_pool2d(x, output_size)
x_max = F.adaptive_max_pool2d(x, output_size)
return 0.5 * (x_avg + x_max)
class Model(nn.Module):
def __init__(self, output_size=1):
super().__init__()
self.output_size = output_size
def forward(self, x):
return adaptive_avgmax_pool2d(x, self.output_size)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ClsHead | # 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_1/inductor_cache/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.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], 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/oo/coobrm7cytaeij2eyoctjbuz6lnjrauzvpxipvlrrepxba5rhlgq.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_3 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%addmm, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%addmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wp/cwpr2jp7bpqwkbaxgttcl2ekjdtkf6lezuspoiunutzd3j2qgeyk.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# x_3 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
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 = 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, 16, 16, grid=grid(16), stream=stream0)
del primals_1
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], 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
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf2, buf3, 16, grid=grid(16), stream=stream0)
buf4 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf3, buf4, 16, grid=grid(16), stream=stream0)
del buf3
return (buf4, reinterpret_tensor(buf1, (4, 4), (4, 1), 0), 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, 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
import torch.nn.functional as F
class ClsHead(nn.Module):
"""
Class orientation
Args:
params(dict): super parameters for build Class network
"""
def __init__(self, in_channels, class_dim, **kwargs):
super(ClsHead, self).__init__()
self.pool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Linear(in_channels, class_dim)
def forward(self, x):
x = self.pool(x)
x = torch.reshape(x, shape=(x.shape[0], x.shape[1]))
x = self.fc(x)
if not self.training:
x = F.softmax(x, dim=1)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'class_dim': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 16.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused__softmax_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
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
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 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(16)](buf1, primals_1, 16, 16, XBLOCK=8,
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
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(16)](buf2, buf3, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf4 = buf2
del buf2
triton_poi_fused__softmax_2[grid(16)](buf3, buf4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf3
return buf4, reinterpret_tensor(buf1, (4, 4), (4, 1), 0), buf4
class ClsHeadNew(nn.Module):
"""
Class orientation
Args:
params(dict): super parameters for build Class network
"""
def __init__(self, in_channels, class_dim, **kwargs):
super(ClsHeadNew, self).__init__()
self.pool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Linear(in_channels, class_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]
| DocYard-ai/UCR | ClsHead | false | 8,030 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
import torch.nn.functional as F
class Model(nn.Module):
"""
Class orientation
Args:
params(dict): super parameters for build Class network
"""
def __init__(self, in_channels, class_dim, **kwargs):
super().__init__()
self.pool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Linear(in_channels, class_dim)
def forward(self, x):
x = self.pool(x)
x = torch.reshape(x, shape=(x.shape[0], x.shape[1]))
x = self.fc(x)
if not self.training:
x = F.softmax(x, dim=1)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
AdaIn | # 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_1/inductor_cache/nl/cnlg43n52wivf3lj76722xh6tab7r4xsgpl4ke5anegxckv76d7x.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten._native_batch_norm_legit]
# Source node to ATen node mapping:
# out => add, rsqrt, var_mean
# 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, 0.0001), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
triton_per_fused__native_batch_norm_legit_0 = async_compile.triton('triton_per_fused__native_batch_norm_legit_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: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__native_batch_norm_legit_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 4, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_0(in_out_ptr0, 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)
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.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 = 0.0001
tmp20 = tmp18 + tmp19
tmp21 = libdevice.rsqrt(tmp20)
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp21, xmask)
tl.store(out_ptr0 + (x0), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mo/cmocvew4ra7imn3ezcqq3xaqk6k6k477oupfmxtz4zz2423fettk.py
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# mul => mul_1
# Graph fragment:
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, 0.5), 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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), 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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 = 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
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4j/c4jtutufnwlinse7a37pyibz6segbjuwwtcymngd6u2mhqcdvjg7.py
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# out_2 => mul_3
# Graph fragment:
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, %unsqueeze_1), kwargs = {})
triton_poi_fused_mul_2 = async_compile.triton('triton_poi_fused_mul_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mul_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x4 = xindex % 256
x5 = (xindex // 16) % 16
x6 = (xindex // 16)
x1 = (xindex // 16) % 4
x7 = xindex
tmp0 = tl.load(in_ptr0 + (x4), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x5), None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x5), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x6), None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp7 = 1.0
tmp8 = tmp6 * tmp7
tmp9 = tmp5 + tmp8
tmp10 = tmp4 * tmp9
tl.store(out_ptr0 + (x7), tmp10, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 1, 1), torch.float32)
buf1 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 16, 16), torch.float32)
buf3 = reinterpret_tensor(buf1, (1, 16, 1, 1), (16, 1, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten._native_batch_norm_legit]
stream0 = get_raw_stream(0)
triton_per_fused__native_batch_norm_legit_0.run(buf3, primals_1, buf0, 16, 16, grid=grid(16), stream=stream0)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
triton_poi_fused_mul_1.run(primals_2, buf4, 16, grid=grid(16), stream=stream0)
del primals_2
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), reinterpret_tensor(buf4, (4, 4), (1, 4), 0), out=buf5)
del buf4
buf6 = empty_strided_cuda((4, 4, 4, 4, 4, 4), (1024, 256, 64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(primals_1, buf0, buf3, buf5, primals_3, buf6, 4096, grid=grid(4096), stream=stream0)
del buf5
del primals_3
return (buf6, primals_1, buf0, buf3, reinterpret_tensor(primals_4, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 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, 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)
| from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class AdaIn(nn.Module):
def __init__(self, num_features, style_dim, eps=0.0001):
super().__init__()
self.num_features = num_features
self.weight = self.bias = None
self.norm_layer = nn.InstanceNorm2d(num_features, eps=eps, affine=False
)
self.modulation = EqualLinear(style_dim, num_features, bias_init=1)
def forward(self, input, style):
_B, _C, _H, _W = input.shape
out = self.norm_layer(input)
gammas = self.modulation(style)
gammas = gammas[..., None, None]
out = out * gammas
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_features': 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
from torch.autograd import Function
import 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_per_fused__native_batch_norm_legit_0(in_out_ptr0, 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)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
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 = 0.0001
tmp20 = tmp18 + tmp19
tmp21 = libdevice.rsqrt(tmp20)
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp21, xmask)
tl.store(out_ptr0 + x0, tmp10, xmask)
@triton.jit
def triton_poi_fused_mul_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x4 = xindex % 256
x5 = xindex // 16 % 16
x6 = xindex // 16
x1 = xindex // 16 % 4
x7 = xindex
tmp0 = tl.load(in_ptr0 + x4, None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x5, None, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x5, None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x6, None, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr4 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp7 = 1.0
tmp8 = tmp6 * tmp7
tmp9 = tmp5 + tmp8
tmp10 = tmp4 * tmp9
tl.store(out_ptr0 + x7, tmp10, None)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 1, 1), torch.float32)
buf1 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 16, 16), torch.float32
)
buf3 = reinterpret_tensor(buf1, (1, 16, 1, 1), (16, 1, 1, 1), 0)
del buf1
get_raw_stream(0)
triton_per_fused__native_batch_norm_legit_0[grid(16)](buf3,
primals_1, buf0, 16, 16, XBLOCK=8, num_warps=2, num_stages=1)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_mul_1[grid(16)](primals_2, buf4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_2
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_4, (64, 4), (4, 1), 0),
reinterpret_tensor(buf4, (4, 4), (1, 4), 0), out=buf5)
del buf4
buf6 = empty_strided_cuda((4, 4, 4, 4, 4, 4), (1024, 256, 64, 16, 4,
1), torch.float32)
triton_poi_fused_mul_2[grid(4096)](primals_1, buf0, buf3, buf5,
primals_3, buf6, 4096, XBLOCK=256, num_warps=4, num_stages=1)
del buf5
del primals_3
return buf6, primals_1, buf0, buf3, reinterpret_tensor(primals_4, (64,
4), (4, 1), 0)
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class AdaInNew(nn.Module):
def __init__(self, num_features, style_dim, eps=0.0001):
super().__init__()
self.num_features = num_features
self.weight = self.bias = None
self.norm_layer = nn.InstanceNorm2d(num_features, eps=eps, affine=False
)
self.modulation = EqualLinear(style_dim, num_features, bias_init=1)
def forward(self, input_0, input_1):
primals_2 = self.modulation.weight
primals_3 = self.modulation.bias
primals_1 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
| Dolorousrtur/style-people | AdaIn | false | 8,031 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | from torch.autograd import Function
import math
import torch
from torch import nn
import torch.nn.functional as F
def fused_leaky_relu(input, bias, negative_slope=0.2, scale=2 ** 0.5):
return FusedLeakyReLUFunction.apply(input, bias, negative_slope, scale)
class FusedLeakyReLUFunctionBackward(Function):
@staticmethod
def forward(ctx, grad_output, out, negative_slope, scale):
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
empty = grad_output.new_empty(0)
grad_input = fused.fused_bias_act(grad_output, empty, out, 3, 1,
negative_slope, scale)
dim = [0]
if grad_input.ndim > 2:
dim += list(range(2, grad_input.ndim))
grad_bias = grad_input.sum(dim).detach()
return grad_input, grad_bias
@staticmethod
def backward(ctx, gradgrad_input, gradgrad_bias):
out, = ctx.saved_tensors
gradgrad_out = fused.fused_bias_act(gradgrad_input, gradgrad_bias,
out, 3, 1, ctx.negative_slope, ctx.scale)
return gradgrad_out, None, None, None
class FusedLeakyReLUFunction(Function):
@staticmethod
def forward(ctx, input, bias, negative_slope, scale):
if input.dtype != bias.dtype:
input = input.type(bias.dtype)
empty = input.new_empty(0)
out = fused.fused_bias_act(input, bias, empty, 3, 0, negative_slope,
scale)
ctx.save_for_backward(out)
ctx.negative_slope = negative_slope
ctx.scale = scale
return out
@staticmethod
def backward(ctx, grad_output):
out, = ctx.saved_tensors
grad_input, grad_bias = FusedLeakyReLUFunctionBackward.apply(
grad_output, out, ctx.negative_slope, ctx.scale)
return grad_input, grad_bias, None, None
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1,
activation=None):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = 1 / math.sqrt(in_dim) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(input, self.weight * self.scale, bias=self.bias *
self.lr_mul)
return out
def __repr__(self):
return (
f'{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})'
)
class Model(nn.Module):
def __init__(self, num_features, style_dim, eps=0.0001):
super().__init__()
self.num_features = num_features
self.weight = self.bias = None
self.norm_layer = nn.InstanceNorm2d(num_features, eps=eps, affine=False
)
self.modulation = EqualLinear(style_dim, num_features, bias_init=1)
def forward(self, input, style):
_B, _C, _H, _W = input.shape
out = self.norm_layer(input)
gammas = self.modulation(style)
gammas = gammas[..., None, None]
out = out * gammas
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
SpatialSoftmax | # 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_1/inductor_cache/e7/ce7mremlbshviguvqvzmsjmhwhjodb3laibhsl43wvnng5oobu53.py
# Topologically Sorted Source Nodes: [marginal, sm, marginal_1, sm_1], Original ATen: [aten.mean, aten._softmax]
# Source node to ATen node mapping:
# marginal => mean
# marginal_1 => mean_1
# sm => amax
# sm_1 => amax_1
# Graph fragment:
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [3]), kwargs = {})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mean, [2], True), kwargs = {})
# %mean_1 : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [2]), kwargs = {})
# %amax_1 : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mean_1, [2], True), kwargs = {})
triton_poi_fused__softmax_mean_0 = async_compile.triton('triton_poi_fused__softmax_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_mean_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 16, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mean_0(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 + (16*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (16*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (16*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (16*x0)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (4 + (16*x0)), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (5 + (16*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (6 + (16*x0)), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (7 + (16*x0)), xmask, eviction_policy='evict_last')
tmp18 = tl.load(in_ptr0 + (8 + (16*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (9 + (16*x0)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (10 + (16*x0)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (11 + (16*x0)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (12 + (16*x0)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr0 + (13 + (16*x0)), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr0 + (14 + (16*x0)), xmask, eviction_policy='evict_last')
tmp32 = tl.load(in_ptr0 + (15 + (16*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp11 = tmp9 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tmp13 + tmp14
tmp16 = tmp15 / tmp7
tmp17 = triton_helpers.maximum(tmp8, tmp16)
tmp20 = tmp18 + tmp19
tmp22 = tmp20 + tmp21
tmp24 = tmp22 + tmp23
tmp25 = tmp24 / tmp7
tmp26 = triton_helpers.maximum(tmp17, tmp25)
tmp29 = tmp27 + tmp28
tmp31 = tmp29 + tmp30
tmp33 = tmp31 + tmp32
tmp34 = tmp33 / tmp7
tmp35 = triton_helpers.maximum(tmp26, tmp34)
tmp36 = tmp0 + tmp9
tmp37 = tmp36 + tmp18
tmp38 = tmp37 + tmp27
tmp39 = tmp38 / tmp7
tmp40 = tmp1 + tmp10
tmp41 = tmp40 + tmp19
tmp42 = tmp41 + tmp28
tmp43 = tmp42 / tmp7
tmp44 = triton_helpers.maximum(tmp39, tmp43)
tmp45 = tmp3 + tmp12
tmp46 = tmp45 + tmp21
tmp47 = tmp46 + tmp30
tmp48 = tmp47 / tmp7
tmp49 = triton_helpers.maximum(tmp44, tmp48)
tmp50 = tmp5 + tmp14
tmp51 = tmp50 + tmp23
tmp52 = tmp51 + tmp32
tmp53 = tmp52 / tmp7
tmp54 = triton_helpers.maximum(tmp49, tmp53)
tl.store(out_ptr0 + (x0), tmp35, xmask)
tl.store(out_ptr1 + (x0), tmp54, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/uf/cuf4anpobe4oogg4xkjr6iypnlyahcecuzfrixcunnn5mrpbbz4x.py
# Topologically Sorted Source Nodes: [marginal, sm], Original ATen: [aten.mean, aten._softmax]
# Source node to ATen node mapping:
# marginal => mean
# sm => exp, sub
# Graph fragment:
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [3]), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mean, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_mean_1 = async_compile.triton('triton_poi_fused__softmax_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=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_mean_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mean_1(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (4*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tmp11 = tl_math.exp(tmp10)
tl.store(out_ptr0 + (x2), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/yu/cyue6xqmai2s22r6qjefxtmpo4swxbazl2oajxdnyw77eda5ymdt.py
# Topologically Sorted Source Nodes: [sm, mul, sum_1], Original ATen: [aten._softmax, aten.mul, aten.sum]
# Source node to ATen node mapping:
# mul => mul_2
# sm => div, sum_1
# sum_1 => sum_3
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [2], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, %expand), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_2, [2], True), kwargs = {})
triton_poi_fused__softmax_mul_sum_2 = async_compile.triton('triton_poi_fused__softmax_mul_sum_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_mul_sum_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 4, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mul_sum_2(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = tmp0 / tmp6
tmp8 = 0.0
tmp9 = 2.0
tmp10 = tmp8 < tmp9
tmp11 = tl.where(tmp10, tmp8, tmp8)
tmp12 = tmp7 * tmp11
tmp13 = tmp1 / tmp6
tmp14 = 1.0
tmp15 = tmp14 < tmp9
tmp16 = 0.3333333333333333
tmp17 = 0.33333333333333337
tmp18 = tl.where(tmp15, tmp16, tmp17)
tmp19 = tmp13 * tmp18
tmp20 = tmp12 + tmp19
tmp21 = tmp3 / tmp6
tmp22 = tmp9 < tmp9
tmp23 = 0.6666666666666666
tmp24 = 0.6666666666666667
tmp25 = tl.where(tmp22, tmp23, tmp24)
tmp26 = tmp21 * tmp25
tmp27 = tmp20 + tmp26
tmp28 = tmp5 / tmp6
tmp29 = 3.0
tmp30 = tmp29 < tmp9
tmp31 = tl.where(tmp30, tmp14, tmp14)
tmp32 = tmp28 * tmp31
tmp33 = tmp27 + tmp32
tl.store(out_ptr0 + (x0), tmp33, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ia/ciavixtvykpxzi5wiobrblt7ppzuez2fidpz6kwbeygqfs3jkuiy.py
# Topologically Sorted Source Nodes: [marginal_1, sm_1], Original ATen: [aten.mean, aten._softmax]
# Source node to ATen node mapping:
# marginal_1 => mean_1
# sm_1 => exp_1, sub_1
# Graph fragment:
# %mean_1 : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [2]), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mean_1, %amax_1), kwargs = {})
# %exp_1 : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
triton_poi_fused__softmax_mean_3 = async_compile.triton('triton_poi_fused__softmax_mean_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__softmax_mean_3', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused__softmax_mean_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + (16*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + (16*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + (16*x1)), xmask)
tmp9 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tmp11 = tl_math.exp(tmp10)
tl.store(out_ptr0 + (x2), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/5p/c5p43prdoadksd4vcmn4uyq2d3o6vtv4szk4bhqbm544wfpdpw7c.py
# Topologically Sorted Source Nodes: [stack], Original ATen: [aten.stack]
# Source node to ATen node mapping:
# stack => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze, %unsqueeze_1], 2), kwargs = {})
triton_poi_fused_stack_4 = async_compile.triton('triton_poi_fused_stack_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=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_stack_4', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_stack_4(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 2
x1 = (xindex // 2)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x1), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 2, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x1), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
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, 1), (4, 1, 16), torch.float32)
buf3 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [marginal, sm, marginal_1, sm_1], Original ATen: [aten.mean, aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_mean_0.run(arg0_1, buf0, buf3, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [marginal, sm], Original ATen: [aten.mean, aten._softmax]
triton_poi_fused__softmax_mean_1.run(arg0_1, buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = reinterpret_tensor(buf0, (4, 4, 1), (4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [sm, mul, sum_1], Original ATen: [aten._softmax, aten.mul, aten.sum]
triton_poi_fused__softmax_mul_sum_2.run(buf1, buf2, 16, grid=grid(16), stream=stream0)
buf4 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [marginal_1, sm_1], Original ATen: [aten.mean, aten._softmax]
triton_poi_fused__softmax_mean_3.run(arg0_1, buf3, buf4, 64, grid=grid(64), stream=stream0)
del arg0_1
buf5 = reinterpret_tensor(buf3, (4, 4, 1), (4, 1, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [sm_1, mul_1, sum_2], Original ATen: [aten._softmax, aten.mul, aten.sum]
triton_poi_fused__softmax_mul_sum_2.run(buf4, buf5, 16, grid=grid(16), stream=stream0)
del buf4
buf6 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [stack], Original ATen: [aten.stack]
triton_poi_fused_stack_4.run(buf2, buf5, buf6, 32, grid=grid(32), stream=stream0)
del buf2
del buf5
return (buf6, )
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.utils.data
import torch.random
import torch.nn.functional as F
def marginal_softmax(heatmap, dim):
marginal = torch.mean(heatmap, dim=dim)
sm = F.softmax(marginal, dim=2)
return sm
def prob_to_keypoints(prob, length):
ruler = torch.linspace(0, 1, length).type_as(prob).expand(1, 1, -1)
return torch.sum(prob * ruler, dim=2, keepdim=True).squeeze(2)
def spacial_softmax(heatmap, probs=False):
height, width = heatmap.size(2), heatmap.size(3)
hp, wp = marginal_softmax(heatmap, dim=3), marginal_softmax(heatmap, dim=2)
hk, wk = prob_to_keypoints(hp, height), prob_to_keypoints(wp, width)
if probs:
return torch.stack((hk, wk), dim=2), (hp, wp)
else:
return torch.stack((hk, wk), dim=2)
class SpatialSoftmax(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, heatmap, probs=False):
return spacial_softmax(heatmap, probs)
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
import torch.utils.data
import torch.random
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__softmax_mean_0(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 + 16 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr0 + (2 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr0 + (3 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp9 = tl.load(in_ptr0 + (4 + 16 * x0), xmask, eviction_policy='evict_last'
)
tmp10 = tl.load(in_ptr0 + (5 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp12 = tl.load(in_ptr0 + (6 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp14 = tl.load(in_ptr0 + (7 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp18 = tl.load(in_ptr0 + (8 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr0 + (9 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (10 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr0 + (11 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp27 = tl.load(in_ptr0 + (12 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp28 = tl.load(in_ptr0 + (13 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp30 = tl.load(in_ptr0 + (14 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp32 = tl.load(in_ptr0 + (15 + 16 * x0), xmask, eviction_policy=
'evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp11 = tmp9 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tmp13 + tmp14
tmp16 = tmp15 / tmp7
tmp17 = triton_helpers.maximum(tmp8, tmp16)
tmp20 = tmp18 + tmp19
tmp22 = tmp20 + tmp21
tmp24 = tmp22 + tmp23
tmp25 = tmp24 / tmp7
tmp26 = triton_helpers.maximum(tmp17, tmp25)
tmp29 = tmp27 + tmp28
tmp31 = tmp29 + tmp30
tmp33 = tmp31 + tmp32
tmp34 = tmp33 / tmp7
tmp35 = triton_helpers.maximum(tmp26, tmp34)
tmp36 = tmp0 + tmp9
tmp37 = tmp36 + tmp18
tmp38 = tmp37 + tmp27
tmp39 = tmp38 / tmp7
tmp40 = tmp1 + tmp10
tmp41 = tmp40 + tmp19
tmp42 = tmp41 + tmp28
tmp43 = tmp42 / tmp7
tmp44 = triton_helpers.maximum(tmp39, tmp43)
tmp45 = tmp3 + tmp12
tmp46 = tmp45 + tmp21
tmp47 = tmp46 + tmp30
tmp48 = tmp47 / tmp7
tmp49 = triton_helpers.maximum(tmp44, tmp48)
tmp50 = tmp5 + tmp14
tmp51 = tmp50 + tmp23
tmp52 = tmp51 + tmp32
tmp53 = tmp52 / tmp7
tmp54 = triton_helpers.maximum(tmp49, tmp53)
tl.store(out_ptr0 + x0, tmp35, xmask)
tl.store(out_ptr1 + x0, tmp54, xmask)
@triton.jit
def triton_poi_fused__softmax_mean_1(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + 4 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x2), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x2), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x2), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tmp11 = tl_math.exp(tmp10)
tl.store(out_ptr0 + x2, tmp11, xmask)
@triton.jit
def triton_poi_fused__softmax_mul_sum_2(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = tmp0 / tmp6
tmp8 = 0.0
tmp9 = 2.0
tmp10 = tmp8 < tmp9
tmp11 = tl.where(tmp10, tmp8, tmp8)
tmp12 = tmp7 * tmp11
tmp13 = tmp1 / tmp6
tmp14 = 1.0
tmp15 = tmp14 < tmp9
tmp16 = 0.3333333333333333
tmp17 = 0.33333333333333337
tmp18 = tl.where(tmp15, tmp16, tmp17)
tmp19 = tmp13 * tmp18
tmp20 = tmp12 + tmp19
tmp21 = tmp3 / tmp6
tmp22 = tmp9 < tmp9
tmp23 = 0.6666666666666666
tmp24 = 0.6666666666666667
tmp25 = tl.where(tmp22, tmp23, tmp24)
tmp26 = tmp21 * tmp25
tmp27 = tmp20 + tmp26
tmp28 = tmp5 / tmp6
tmp29 = 3.0
tmp30 = tmp29 < tmp9
tmp31 = tl.where(tmp30, tmp14, tmp14)
tmp32 = tmp28 * tmp31
tmp33 = tmp27 + tmp32
tl.store(out_ptr0 + x0, tmp33, xmask)
@triton.jit
def triton_poi_fused__softmax_mean_3(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0 + 16 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0 + 16 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0 + 16 * x1), xmask)
tmp9 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tmp10 = tmp8 - tmp9
tmp11 = tl_math.exp(tmp10)
tl.store(out_ptr0 + x2, tmp11, xmask)
@triton.jit
def triton_poi_fused_stack_4(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 2
x1 = xindex // 2
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + x1, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp9 = tl.load(in_ptr1 + x1, tmp6 & xmask, eviction_policy='evict_last',
other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x2, tmp10, 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, 1), (4, 1, 16), torch.float32)
buf3 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused__softmax_mean_0[grid(16)](arg0_1, buf0, buf3, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_mean_1[grid(64)](arg0_1, buf0, buf1, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf2 = reinterpret_tensor(buf0, (4, 4, 1), (4, 1, 1), 0)
del buf0
triton_poi_fused__softmax_mul_sum_2[grid(16)](buf1, buf2, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf4 = buf1
del buf1
triton_poi_fused__softmax_mean_3[grid(64)](arg0_1, buf3, buf4, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del arg0_1
buf5 = reinterpret_tensor(buf3, (4, 4, 1), (4, 1, 1), 0)
del buf3
triton_poi_fused__softmax_mul_sum_2[grid(16)](buf4, buf5, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del buf4
buf6 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
triton_poi_fused_stack_4[grid(32)](buf2, buf5, buf6, 32, XBLOCK=32,
num_warps=1, num_stages=1)
del buf2
del buf5
return buf6,
def marginal_softmax(heatmap, dim):
marginal = torch.mean(heatmap, dim=dim)
sm = F.softmax(marginal, dim=2)
return sm
def prob_to_keypoints(prob, length):
ruler = torch.linspace(0, 1, length).type_as(prob).expand(1, 1, -1)
return torch.sum(prob * ruler, dim=2, keepdim=True).squeeze(2)
def spacial_softmax(heatmap, probs=False):
height, width = heatmap.size(2), heatmap.size(3)
hp, wp = marginal_softmax(heatmap, dim=3), marginal_softmax(heatmap, dim=2)
hk, wk = prob_to_keypoints(hp, height), prob_to_keypoints(wp, width)
if probs:
return torch.stack((hk, wk), dim=2), (hp, wp)
else:
return torch.stack((hk, wk), dim=2)
class SpatialSoftmaxNew(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DuaneNielsen/keypoints | SpatialSoftmax | false | 8,032 | [
"MIT"
] | 42 | 302fa02966d4372ac9b5aaa3d8dc24684be0b252 | https://github.com/DuaneNielsen/keypoints/tree/302fa02966d4372ac9b5aaa3d8dc24684be0b252 | import torch
import torch.utils.data
import torch.random
import torch.nn.functional as F
def marginal_softmax(heatmap, dim):
marginal = torch.mean(heatmap, dim=dim)
sm = F.softmax(marginal, dim=2)
return sm
def prob_to_keypoints(prob, length):
ruler = torch.linspace(0, 1, length).type_as(prob).expand(1, 1, -1)
return torch.sum(prob * ruler, dim=2, keepdim=True).squeeze(2)
def spacial_softmax(heatmap, probs=False):
height, width = heatmap.size(2), heatmap.size(3)
hp, wp = marginal_softmax(heatmap, dim=3), marginal_softmax(heatmap, dim=2)
hk, wk = prob_to_keypoints(hp, height), prob_to_keypoints(wp, width)
if probs:
return torch.stack((hk, wk), dim=2), (hp, wp)
else:
return torch.stack((hk, wk), dim=2)
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
def forward(self, heatmap, probs=False):
return spacial_softmax(heatmap, probs)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
GRUCell | # 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_1/inductor_cache/zn/czn6cztle4peyy4pa7mkag53s34sjkn6wpenptu6ttfuvhgzzrup.py
# Topologically Sorted Source Nodes: [rz_input], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# rz_input => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_1, %primals_2], -1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 512
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + ((4*x1) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/or/cornvoiibm2dj5uevwstcy73xsnr3istcs65mdwxrvwduvq2jjdw.py
# Topologically Sorted Source Nodes: [r, z, mul, add, u, mul_1, sub, mul_2, new_h], Original ATen: [aten.sigmoid, aten.mul, aten.add, aten.tanh, aten.rsub]
# Source node to ATen node mapping:
# add => add
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# new_h => add_1
# r => sigmoid
# sub => sub
# u => tanh
# z => sigmoid_1
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_1,), kwargs = {})
# %sigmoid_1 : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_3,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %view_7), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_5, %mul), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid_1, %primals_2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %tanh), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, %mul_2), kwargs = {})
triton_poi_fused_add_mul_rsub_sigmoid_tanh_1 = async_compile.triton('triton_poi_fused_add_mul_rsub_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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_rsub_sigmoid_tanh_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_rsub_sigmoid_tanh_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask)
tmp6 = tl.load(in_ptr2 + (x0), xmask)
tmp7 = tl.load(in_ptr3 + (x0), xmask)
tmp9 = tl.load(in_ptr4 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = 1.0
tmp5 = tmp4 - tmp1
tmp8 = tl.sigmoid(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = tmp6 + tmp10
tmp12 = libdevice.tanh(tmp11)
tmp13 = tmp5 * tmp12
tmp14 = tmp3 + tmp13
tl.store(out_ptr0 + (x0), tmp14, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 8), (8, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, 8), (8, 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, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [rz_input], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, primals_2, buf0, 512, grid=grid(512), stream=stream0)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_4, reinterpret_tensor(buf0, (64, 8), (8, 1), 0), reinterpret_tensor(primals_3, (8, 4), (1, 8), 0), alpha=1, beta=1, out=buf1)
del primals_3
del primals_4
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_6, reinterpret_tensor(buf0, (64, 8), (8, 1), 0), reinterpret_tensor(primals_5, (8, 4), (1, 8), 0), alpha=1, beta=1, out=buf2)
del primals_5
del primals_6
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_7
del primals_8
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_10, reinterpret_tensor(primals_2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_10
del primals_9
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [r, z, mul, add, u, mul_1, sub, mul_2, new_h], Original ATen: [aten.sigmoid, aten.mul, aten.add, aten.tanh, aten.rsub]
triton_poi_fused_add_mul_rsub_sigmoid_tanh_1.run(buf2, primals_2, buf3, buf1, buf4, buf5, 256, grid=grid(256), stream=stream0)
return (buf5, primals_2, reinterpret_tensor(buf0, (64, 8), (8, 1), 0), buf1, buf2, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0), 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, 4), (64, 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)
primals_3 = rand_strided((4, 8), (8, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((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((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)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.utils.checkpoint
class GRUCell(nn.Module):
def __init__(self, x_dim, h_dim):
super(GRUCell, self).__init__()
self.r = nn.Linear(x_dim + h_dim, h_dim, True)
self.z = nn.Linear(x_dim + h_dim, h_dim, True)
self.c = nn.Linear(x_dim, h_dim, True)
self.u = nn.Linear(h_dim, h_dim, True)
def forward(self, x, h):
rz_input = torch.cat((x, h), -1)
r = torch.sigmoid(self.r(rz_input))
z = torch.sigmoid(self.z(rz_input))
u = torch.tanh(self.c(x) + r * self.u(h))
new_h = z * h + (1 - z) * u
return new_h
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'x_dim': 4, 'h_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
import torch.utils.checkpoint
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(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
x0 = xindex % 8
x1 = xindex // 8
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (4 * x1 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_mul_rsub_sigmoid_tanh_1(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask)
tmp6 = tl.load(in_ptr2 + x0, xmask)
tmp7 = tl.load(in_ptr3 + x0, xmask)
tmp9 = tl.load(in_ptr4 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tmp4 = 1.0
tmp5 = tmp4 - tmp1
tmp8 = tl.sigmoid(tmp7)
tmp10 = tmp8 * tmp9
tmp11 = tmp6 + tmp10
tmp12 = libdevice.tanh(tmp11)
tmp13 = tmp5 * tmp12
tmp14 = tmp3 + tmp13
tl.store(out_ptr0 + x0, tmp14, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (4, 8), (8, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4, 8), (8, 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,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 8), (128, 32, 8, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(512)](primals_1, primals_2, buf0, 512,
XBLOCK=128, num_warps=4, num_stages=1)
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_4, reinterpret_tensor(buf0, (64, 8), (
8, 1), 0), reinterpret_tensor(primals_3, (8, 4), (1, 8), 0),
alpha=1, beta=1, out=buf1)
del primals_3
del primals_4
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_6, reinterpret_tensor(buf0, (64, 8), (
8, 1), 0), reinterpret_tensor(primals_5, (8, 4), (1, 8), 0),
alpha=1, beta=1, out=buf2)
del primals_5
del primals_6
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_1, (64,
4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf3)
del primals_7
del primals_8
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_10, reinterpret_tensor(primals_2, (64,
4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf4)
del primals_10
del primals_9
buf5 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_mul_rsub_sigmoid_tanh_1[grid(256)](buf2,
primals_2, buf3, buf1, buf4, buf5, 256, XBLOCK=256, num_warps=4,
num_stages=1)
return buf5, primals_2, reinterpret_tensor(buf0, (64, 8), (8, 1), 0
), buf1, buf2, reinterpret_tensor(primals_1, (64, 4), (4, 1), 0
), buf3, buf4
class GRUCellNew(nn.Module):
def __init__(self, x_dim, h_dim):
super(GRUCellNew, self).__init__()
self.r = nn.Linear(x_dim + h_dim, h_dim, True)
self.z = nn.Linear(x_dim + h_dim, h_dim, True)
self.c = nn.Linear(x_dim, h_dim, True)
self.u = nn.Linear(h_dim, h_dim, True)
def forward(self, input_0, input_1):
primals_3 = self.r.weight
primals_4 = self.r.bias
primals_5 = self.z.weight
primals_6 = self.z.bias
primals_7 = self.c.weight
primals_8 = self.c.bias
primals_9 = self.u.weight
primals_10 = self.u.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return output[0]
| DeepLearnXMU/IRSEG | GRUCell | false | 8,033 | [
"Apache-2.0"
] | 14 | 6027c9601dbcb626d4adaf429c4bed07febb1034 | https://github.com/DeepLearnXMU/IRSEG/tree/6027c9601dbcb626d4adaf429c4bed07febb1034 | import torch
import torch.nn as nn
import torch.utils.checkpoint
class Model(nn.Module):
def __init__(self, x_dim, h_dim):
super().__init__()
self.r = nn.Linear(x_dim + h_dim, h_dim, True)
self.z = nn.Linear(x_dim + h_dim, h_dim, True)
self.c = nn.Linear(x_dim, h_dim, True)
self.u = nn.Linear(h_dim, h_dim, True)
def forward(self, x, h):
rz_input = torch.cat((x, h), -1)
r = torch.sigmoid(self.r(rz_input))
z = torch.sigmoid(self.z(rz_input))
u = torch.tanh(self.c(x) + r * self.u(h))
new_h = z * h + (1 - z) * u
return new_h
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
GroupNorm | # 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_1/inductor_cache/j5/cj53qan5ptxzet4xou5gs6bf5vxvuzy4blgs4rgvihmthn37ij2j.py
# Topologically Sorted Source Nodes: [mul, add_1], Original ATen: [aten.mul, aten.add]
# Source node to ATen node mapping:
# add_1 => add_1
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, %primals_2), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_3), kwargs = {})
triton_poi_fused_add_mul_0 = async_compile.triton('triton_poi_fused_add_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 7, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x5 = (xindex // 4)
x2 = (xindex // 16) % 4
tmp0 = tl.load(in_ptr0 + (x4), xmask)
tmp1 = tl.load(in_ptr0 + (4*x5), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x5)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x5)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x5)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (x2), xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = tmp0 - tmp9
tmp11 = tmp1 - tmp9
tmp12 = tmp11 * tmp11
tmp13 = tmp2 - tmp9
tmp14 = tmp13 * tmp13
tmp15 = tmp12 + tmp14
tmp16 = tmp4 - tmp9
tmp17 = tmp16 * tmp16
tmp18 = tmp15 + tmp17
tmp19 = tmp6 - tmp9
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = 3.0
tmp23 = tmp21 / tmp22
tmp24 = libdevice.sqrt(tmp23)
tmp25 = 1e-10
tmp26 = tmp24 + tmp25
tmp27 = tmp10 / tmp26
tmp29 = tmp27 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + (x4), 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, 1), (1, 1, 1))
assert_size_stride(primals_3, (4, 1, 1), (1, 1, 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_1], Original ATen: [aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_0.run(primals_1, primals_2, 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, 1), (1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 1, 1), (1, 1, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class GroupNorm(nn.Module):
def __init__(self, c_num, group_num=16, eps=1e-10):
"""
The groupnorm layer from https://arxiv.org/abs/1803.08494
Args:
c_num (int): Number of input channels
group_num (int): Number of group by which to divide the channels
eps (float): Epsilon
"""
super(GroupNorm, self).__init__()
self.group_num = group_num
self.gamma = nn.Parameter(torch.ones(c_num, 1, 1))
self.beta = nn.Parameter(torch.zeros(c_num, 1, 1))
self.eps = eps
def forward(self, x):
batch_size, channels, height, width = x.size()
x = x.view(batch_size, self.group_num, -1)
mean = x.mean(dim=2, keepdim=True)
std = x.std(dim=2, keepdim=True)
x = (x - mean) / (std + self.eps)
x = x.view(batch_size, channels, height, width)
return x * self.gamma + self.beta
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'c_num': 4}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x4 = xindex
x5 = xindex // 4
x2 = xindex // 16 % 4
tmp0 = tl.load(in_ptr0 + x4, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x5, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x5), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x5), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x5), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + x2, xmask, eviction_policy='evict_last')
tmp30 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = 4.0
tmp9 = tmp7 / tmp8
tmp10 = tmp0 - tmp9
tmp11 = tmp1 - tmp9
tmp12 = tmp11 * tmp11
tmp13 = tmp2 - tmp9
tmp14 = tmp13 * tmp13
tmp15 = tmp12 + tmp14
tmp16 = tmp4 - tmp9
tmp17 = tmp16 * tmp16
tmp18 = tmp15 + tmp17
tmp19 = tmp6 - tmp9
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = 3.0
tmp23 = tmp21 / tmp22
tmp24 = libdevice.sqrt(tmp23)
tmp25 = 1e-10
tmp26 = tmp24 + tmp25
tmp27 = tmp10 / tmp26
tmp29 = tmp27 * tmp28
tmp31 = tmp29 + tmp30
tl.store(out_ptr0 + x4, 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, 1), (1, 1, 1))
assert_size_stride(primals_3, (4, 1, 1), (1, 1, 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_3, buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
del primals_3
return buf0, primals_1
class GroupNormNew(nn.Module):
def __init__(self, c_num, group_num=16, eps=1e-10):
"""
The groupnorm layer from https://arxiv.org/abs/1803.08494
Args:
c_num (int): Number of input channels
group_num (int): Number of group by which to divide the channels
eps (float): Epsilon
"""
super(GroupNormNew, self).__init__()
self.group_num = group_num
self.gamma = nn.Parameter(torch.ones(c_num, 1, 1))
self.beta = nn.Parameter(torch.zeros(c_num, 1, 1))
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]
| EKami/EzeeML | GroupNorm | false | 8,034 | [
"MIT"
] | 35 | 21753a0ede7cc1dc675a2dcd09b6306cea2cad56 | https://github.com/EKami/EzeeML/tree/21753a0ede7cc1dc675a2dcd09b6306cea2cad56 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, c_num, group_num=16, eps=1e-10):
"""
The groupnorm layer from https://arxiv.org/abs/1803.08494
Args:
c_num (int): Number of input channels
group_num (int): Number of group by which to divide the channels
eps (float): Epsilon
"""
super().__init__()
self.group_num = group_num
self.gamma = nn.Parameter(torch.ones(c_num, 1, 1))
self.beta = nn.Parameter(torch.zeros(c_num, 1, 1))
self.eps = eps
def forward(self, x):
batch_size, channels, height, width = x.size()
x = x.view(batch_size, self.group_num, -1)
mean = x.mean(dim=2, keepdim=True)
std = x.std(dim=2, keepdim=True)
x = (x - mean) / (std + self.eps)
x = x.view(batch_size, channels, height, width)
return x * self.gamma + self.beta
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4]
|
CTCHead | # 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_1/inductor_cache/ai/caivsorbpc5aa3it57vpn6elo7hp6bayj2fn3k6vrqjvtfxip7lp.py
# Topologically Sorted Source Nodes: [predicts_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# predicts_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [2], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_0 = async_compile.triton('triton_poi_fused__softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ze/czehbz6hskvah4rwfpjgwn5voib6lo6n3dqua2e2eiytox46bycf.py
# Topologically Sorted Source Nodes: [predicts_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# predicts_1 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [2], True), kwargs = {})
# %div : [num_users=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=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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 % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + (16*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 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [predicts], 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: [predicts_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: [predicts_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
del buf1
return (buf2, 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((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
from torch import nn
class CTCHead(nn.Module):
def __init__(self, in_channels, out_channels, fc_decay=0.0004, **kwargs):
super(CTCHead, self).__init__()
self.fc = nn.Linear(in_channels, out_channels)
self.softmax = nn.Softmax(dim=2)
self.out_channels = out_channels
def forward(self, x, labels=None):
predicts = self.fc(x)
if not self.training:
predicts = self.softmax(predicts)
return predicts
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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 % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_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 % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (12 + x0 + 16 * 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 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_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__softmax_0[grid(256)](buf0, buf1, 256, XBLOCK=128,
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=256,
num_warps=4, num_stages=1)
del buf1
return buf2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf2
class CTCHeadNew(nn.Module):
def __init__(self, in_channels, out_channels, fc_decay=0.0004, **kwargs):
super(CTCHeadNew, self).__init__()
self.fc = nn.Linear(in_channels, out_channels)
self.softmax = nn.Softmax(dim=2)
self.out_channels = out_channels
def forward(self, input_0):
primals_1 = self.fc.weight
primals_2 = self.fc.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| DocYard-ai/UCR | CTCHead | false | 8,035 | [
"Apache-2.0"
] | 10 | 7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | https://github.com/DocYard-ai/UCR/tree/7618aa336f56e71d9fd8cdc2d591e3d138e3dc68 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self, in_channels, out_channels, fc_decay=0.0004, **kwargs):
super().__init__()
self.fc = nn.Linear(in_channels, out_channels)
self.softmax = nn.Softmax(dim=2)
self.out_channels = out_channels
def forward(self, x, labels=None):
predicts = self.fc(x)
if not self.training:
predicts = self.softmax(predicts)
return predicts
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4]
|
FlowfieldDiscountedLoss | # 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_1/inductor_cache/po/cpo4jioqqsx6wrh2ytpfysxsofmfg4luq5vbmzgnf677l7cgknzi.py
# Topologically Sorted Source Nodes: [bad_bits, setitem, setitem_1], Original ATen: [aten.pow, aten.lift_fresh, aten.index_put]
# Source node to ATen node mapping:
# bad_bits => pow_2
# setitem => full_default, index_put
# setitem_1 => full_default_1, index_put_1
# Graph fragment:
# %pow_2 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%arg2_1, 2), kwargs = {})
# %full_default : [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 : [num_users=2] = call_function[target=torch.ops.aten.index_put_.default](args = (%pow_2, [%le], %full_default), kwargs = {})
# %full_default_1 : [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_1 : [num_users=1] = call_function[target=torch.ops.aten.index_put_.default](args = (%index_put, [%gt], %full_default_1), kwargs = {})
triton_poi_fused_index_put_lift_fresh_pow_0 = async_compile.triton('triton_poi_fused_index_put_lift_fresh_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_pow_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_pow_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 = tmp0 * tmp0
tmp2 = 1.0
tmp3 = tmp1 <= tmp2
tmp4 = tl.where(tmp3, tmp2, tmp1)
tmp5 = tmp4 > tmp2
tmp6 = 0.0
tmp7 = tl.where(tmp5, tmp6, tmp4)
tl.store(in_out_ptr0 + (x0), tmp7, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zu/czu6l3u5bq2pnskg5g3tcbjbvc5zthz2en6dt4e2vcxx7acdmqpq.py
# Topologically Sorted Source Nodes: [loss, loss_1, sum_1], Original ATen: [aten.mse_loss, aten.mul, aten.sum]
# Source node to ATen node mapping:
# loss => pow_1, sub
# loss_1 => mul
# sum_1 => sum_1
# 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 = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_1, %permute), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%mul,), kwargs = {})
triton_per_fused_mse_loss_mul_sum_1 = async_compile.triton('triton_per_fused_mse_loss_mul_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, 256],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {4: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 5), equal_to_1=(4,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mse_loss_mul_sum_1', 'mutated_arg_names': [], 'no_x_dim': True, 'num_load': 6, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_mse_loss_mul_sum_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, rnumel):
xnumel = 1
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r3 = rindex
r0 = rindex % 16
r2 = (rindex // 64)
tmp0 = tl.load(in_ptr0 + (r3), None)
tmp1 = tl.load(in_ptr1 + (r3), None)
tmp4 = tl.load(in_ptr2 + ((4*r0) + (64*r2)), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (1 + (4*r0) + (64*r2)), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr2 + (2 + (4*r0) + (64*r2)), None, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (3 + (4*r0) + (64*r2)), None, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 * tmp5
tmp8 = tmp6 * tmp7
tmp10 = tmp8 * tmp9
tmp11 = tmp3 * tmp10
tmp12 = tl.broadcast_to(tmp11, [RBLOCK])
tmp14 = triton_helpers.promote_to_tensor(tl.sum(tmp12, 0))
tl.store(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, 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, 4), (64, 16, 4, 1), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [bad_bits, setitem, setitem_1], Original ATen: [aten.pow, aten.lift_fresh, aten.index_put]
stream0 = get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_pow_0.run(buf1, arg2_1, 256, grid=grid(256), stream=stream0)
del arg2_1
buf2 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [loss, loss_1, sum_1], Original ATen: [aten.mse_loss, aten.mul, aten.sum]
triton_per_fused_mse_loss_mul_sum_1.run(arg1_1, arg0_1, buf1, buf2, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_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)
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
import torch.utils.data
import torch.random
import torch.nn.functional as F
class FlowfieldDiscountedLoss(nn.Module):
def __init__(self):
super().__init__()
def forward(self, source, target, flowfield):
loss = F.mse_loss(source, target, reduction='none')
bad_bits = flowfield ** 2
bad_bits[bad_bits <= 1.0] = 1.0
bad_bits[bad_bits > 1.0] = 0
mask = torch.prod(bad_bits, 3).expand(1, -1, -1, -1).permute(1, 0, 2, 3
)
loss = loss * mask
return torch.sum(loss)
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
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.utils.data
import torch.random
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_pow_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 = tmp0 * tmp0
tmp2 = 1.0
tmp3 = tmp1 <= tmp2
tmp4 = tl.where(tmp3, tmp2, tmp1)
tmp5 = tmp4 > tmp2
tmp6 = 0.0
tmp7 = tl.where(tmp5, tmp6, tmp4)
tl.store(in_out_ptr0 + x0, tmp7, xmask)
@triton.jit
def triton_per_fused_mse_loss_mul_sum_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0,
xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r3 = rindex
r0 = rindex % 16
r2 = rindex // 64
tmp0 = tl.load(in_ptr0 + r3, None)
tmp1 = tl.load(in_ptr1 + r3, None)
tmp4 = tl.load(in_ptr2 + (4 * r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp5 = tl.load(in_ptr2 + (1 + 4 * r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp7 = tl.load(in_ptr2 + (2 + 4 * r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr2 + (3 + 4 * r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 * tmp5
tmp8 = tmp6 * tmp7
tmp10 = tmp8 * tmp9
tmp11 = tmp3 * tmp10
tmp12 = tl.broadcast_to(tmp11, [RBLOCK])
tmp14 = triton_helpers.promote_to_tensor(tl.sum(tmp12, 0))
tl.store(out_ptr0 + tl.full([1], 0, tl.int32), tmp14, None)
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, 4), (64, 16, 4, 1), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_pow_0[grid(256)](buf1, arg2_1,
256, XBLOCK=256, num_warps=4, num_stages=1)
del arg2_1
buf2 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_mse_loss_mul_sum_1[grid(1)](arg1_1, arg0_1, buf1,
buf2, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
del buf1
return buf2,
class FlowfieldDiscountedLossNew(nn.Module):
def __init__(self):
super().__init__()
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]
| DuaneNielsen/keypoints | FlowfieldDiscountedLoss | false | 8,036 | [
"MIT"
] | 42 | 302fa02966d4372ac9b5aaa3d8dc24684be0b252 | https://github.com/DuaneNielsen/keypoints/tree/302fa02966d4372ac9b5aaa3d8dc24684be0b252 | import torch
import torch.nn as nn
import torch.utils.data
import torch.random
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, source, target, flowfield):
loss = F.mse_loss(source, target, reduction='none')
bad_bits = flowfield ** 2
bad_bits[bad_bits <= 1.0] = 1.0
bad_bits[bad_bits > 1.0] = 0
mask = torch.prod(bad_bits, 3).expand(1, -1, -1, -1).permute(1, 0, 2, 3
)
loss = loss * mask
return torch.sum(loss)
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 []
|
Coords | # 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_1/inductor_cache/ui/cuibnt5qzvcwk5rggp4ncnio4mlgphkmsoebwnnkn6v3qad5gcxc.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, %repeat_1], 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=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_cat_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 16) % 6
x3 = (xindex // 96)
x4 = xindex % 16
x1 = (xindex // 4) % 4
x0 = xindex % 4
x7 = 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 + (x4 + (16*x2) + (64*x3)), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 5, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = x1
tmp11 = tmp10.to(tl.float32)
tmp12 = 2.0
tmp13 = tmp11 < tmp12
tmp14 = 0.3333333333333333
tmp15 = tmp11 * tmp14
tmp16 = 0.0
tmp17 = tmp15 + tmp16
tmp18 = 3 + ((-1)*x1)
tmp19 = tmp18.to(tl.float32)
tmp20 = tmp19 * tmp14
tmp21 = 1.0
tmp22 = tmp21 - tmp20
tmp23 = tl.where(tmp13, tmp17, tmp22)
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp9, tmp23, tmp24)
tmp26 = tmp0 >= tmp7
tmp27 = tl.full([1], 6, tl.int64)
tmp28 = tmp0 < tmp27
tmp29 = x0
tmp30 = tmp29.to(tl.float32)
tmp31 = tmp30 < tmp12
tmp32 = tmp30 * tmp14
tmp33 = tmp32 + tmp16
tmp34 = 3 + ((-1)*x0)
tmp35 = tmp34.to(tl.float32)
tmp36 = tmp35 * tmp14
tmp37 = tmp21 - tmp36
tmp38 = tl.where(tmp31, tmp33, tmp37)
tmp39 = tl.full(tmp38.shape, 0.0, tmp38.dtype)
tmp40 = tl.where(tmp26, tmp38, tmp39)
tmp41 = tl.where(tmp9, tmp25, tmp40)
tmp42 = tl.where(tmp4, tmp5, tmp41)
tl.store(out_ptr0 + (x7), tmp42, 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, 6, 4, 4), (96, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(arg0_1, buf0, 384, grid=grid(384), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
import torch.utils.data
import torch.random
class Coords(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
""" adds 2 channels that carry co-ordinate information """
b, h, w = x.size(0), x.size(2), x.size(3)
hm = torch.linspace(0, 1, h, dtype=x.dtype, device=x.device).reshape(
1, 1, h, 1).repeat(b, 1, 1, w)
wm = torch.linspace(0, 1, w, dtype=x.dtype, device=x.device).reshape(
1, 1, 1, w).repeat(b, 1, h, 1)
return torch.cat((x, hm, wm), 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
import torch.nn as nn
import torch.utils.data
import torch.random
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 16 % 6
x3 = xindex // 96
x4 = xindex % 16
x1 = xindex // 4 % 4
x0 = xindex % 4
x7 = xindex
tmp0 = x2
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x4 + 16 * x2 + 64 * x3), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 5, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = x1
tmp11 = tmp10.to(tl.float32)
tmp12 = 2.0
tmp13 = tmp11 < tmp12
tmp14 = 0.3333333333333333
tmp15 = tmp11 * tmp14
tmp16 = 0.0
tmp17 = tmp15 + tmp16
tmp18 = 3 + -1 * x1
tmp19 = tmp18.to(tl.float32)
tmp20 = tmp19 * tmp14
tmp21 = 1.0
tmp22 = tmp21 - tmp20
tmp23 = tl.where(tmp13, tmp17, tmp22)
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp9, tmp23, tmp24)
tmp26 = tmp0 >= tmp7
tl.full([1], 6, tl.int64)
tmp29 = x0
tmp30 = tmp29.to(tl.float32)
tmp31 = tmp30 < tmp12
tmp32 = tmp30 * tmp14
tmp33 = tmp32 + tmp16
tmp34 = 3 + -1 * x0
tmp35 = tmp34.to(tl.float32)
tmp36 = tmp35 * tmp14
tmp37 = tmp21 - tmp36
tmp38 = tl.where(tmp31, tmp33, tmp37)
tmp39 = tl.full(tmp38.shape, 0.0, tmp38.dtype)
tmp40 = tl.where(tmp26, tmp38, tmp39)
tmp41 = tl.where(tmp9, tmp25, tmp40)
tmp42 = tl.where(tmp4, tmp5, tmp41)
tl.store(out_ptr0 + x7, tmp42, 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, 6, 4, 4), (96, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(384)](arg0_1, buf0, 384, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class CoordsNew(nn.Module):
def __init__(self):
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| DuaneNielsen/keypoints | Coords | false | 8,037 | [
"MIT"
] | 42 | 302fa02966d4372ac9b5aaa3d8dc24684be0b252 | https://github.com/DuaneNielsen/keypoints/tree/302fa02966d4372ac9b5aaa3d8dc24684be0b252 | import torch
import torch.nn as nn
import torch.utils.data
import torch.random
class Model(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
""" adds 2 channels that carry co-ordinate information """
b, h, w = x.size(0), x.size(2), x.size(3)
hm = torch.linspace(0, 1, h, dtype=x.dtype, device=x.device).reshape(
1, 1, h, 1).repeat(b, 1, 1, w)
wm = torch.linspace(0, 1, w, dtype=x.dtype, device=x.device).reshape(
1, 1, 1, w).repeat(b, 1, h, 1)
return torch.cat((x, hm, wm), dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
ContrastiveLoss | # 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_1/inductor_cache/v2/cv2ohk4vpat6jimblbxldlcmj6xwbdtpffu3zhdixmnvm47unics.py
# Topologically Sorted Source Nodes: [sub, pow_1, distances], Original ATen: [aten.sub, aten.pow, aten.sum]
# Source node to ATen node mapping:
# distances => sum_1
# pow_1 => pow_1
# sub => sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %sum_1 : [num_users=2] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [1]), kwargs = {})
triton_poi_fused_pow_sub_sum_0 = async_compile.triton('triton_poi_fused_pow_sub_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_pow_sub_sum_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 8, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_sub_sum_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (64*x1)), xmask)
tmp4 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp5 = tl.load(in_ptr1 + (16 + x0 + (64*x1)), xmask)
tmp9 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp10 = tl.load(in_ptr1 + (32 + x0 + (64*x1)), xmask)
tmp14 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp15 = tl.load(in_ptr1 + (48 + x0 + (64*x1)), xmask)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp6
tmp8 = tmp3 + tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tl.store(out_ptr0 + (x2), tmp18, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/zr/czr2vrtmp2zzmoldsxojgyuiyfeueaymvvtixpuppz45iypjtjb3.py
# Topologically Sorted Source Nodes: [mul, mul_1, add, add_1, sqrt, sub_1, relu, pow_2, mul_2, add_2, losses, mean], Original ATen: [aten.mul, aten.add, aten.sqrt, aten.rsub, aten.relu, aten.pow, aten.mean]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# losses => mul_3
# mean => mean
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# pow_2 => pow_2
# relu => relu
# sqrt => sqrt
# sub_1 => sub_1
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg2_1, %sum_1), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg2_1, -1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_1, 1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_1, 1e-09), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (4, %sqrt), kwargs = {})
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%sub_1,), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%relu, 2), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, %pow_2), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_2), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_2, 0.5), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul_3,), kwargs = {})
triton_per_fused_add_mean_mul_pow_relu_rsub_sqrt_1 = async_compile.triton('triton_per_fused_add_mean_mul_pow_relu_rsub_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_mean_mul_pow_relu_rsub_sqrt_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_pow_relu_rsub_sqrt_1(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)
r2 = rindex
r0 = rindex % 64
tmp0 = tl.load(in_ptr0 + (r2), None)
tmp1 = tl.load(in_ptr1 + (r0), None, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tmp3 = -1.0
tmp4 = tmp0 * tmp3
tmp5 = 1.0
tmp6 = tmp4 + tmp5
tmp7 = 1e-09
tmp8 = tmp1 + tmp7
tmp9 = libdevice.sqrt(tmp8)
tmp10 = 4.0
tmp11 = tmp10 - tmp9
tmp12 = tl.full([1], 0, tl.int32)
tmp13 = triton_helpers.maximum(tmp12, tmp11)
tmp14 = tmp13 * tmp13
tmp15 = tmp6 * tmp14
tmp16 = tmp2 + tmp15
tmp17 = 0.5
tmp18 = tmp16 * tmp17
tmp19 = tl.broadcast_to(tmp18, [RBLOCK])
tmp21 = triton_helpers.promote_to_tensor(tl.sum(tmp19, 0))
tmp22 = 256.0
tmp23 = tmp21 / tmp22
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp23, None)
''', 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: [sub, pow_1, distances], Original ATen: [aten.sub, aten.pow, aten.sum]
stream0 = get_raw_stream(0)
triton_poi_fused_pow_sub_sum_0.run(arg0_1, arg1_1, buf0, 64, grid=grid(64), stream=stream0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [mul, mul_1, add, add_1, sqrt, sub_1, relu, pow_2, mul_2, add_2, losses, mean], Original ATen: [aten.mul, aten.add, aten.sqrt, aten.rsub, aten.relu, aten.pow, aten.mean]
triton_per_fused_add_mean_mul_pow_relu_rsub_sqrt_1.run(buf2, arg2_1, buf0, 1, 256, grid=grid(1), stream=stream0)
del arg2_1
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, 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
from torch.nn import functional as F
class ContrastiveLoss(nn.Module):
"""
Contrastive loss
Takes embeddings of two samples and a target label == 1 if samples are from the same class and label == 0 otherwise
"""
def __init__(self, margin):
super(ContrastiveLoss, self).__init__()
self.margin = margin
self.eps = 1e-09
def forward(self, output1, output2, target, size_average=True):
distances = (output2 - output1).pow(2).sum(1)
losses = 0.5 * (target.float() * distances + (1 + -1 * target).
float() * F.relu(self.margin - (distances + self.eps).sqrt()).
pow(2))
return losses.mean() if size_average else losses.sum()
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 [[], {'margin': 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.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_pow_sub_sum_0(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 64 * x1), xmask)
tmp4 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp5 = tl.load(in_ptr1 + (16 + x0 + 64 * x1), xmask)
tmp9 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp10 = tl.load(in_ptr1 + (32 + x0 + 64 * x1), xmask)
tmp14 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp15 = tl.load(in_ptr1 + (48 + x0 + 64 * x1), xmask)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp6
tmp8 = tmp3 + tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tl.store(out_ptr0 + x2, tmp18, xmask)
@triton.jit
def triton_per_fused_add_mean_mul_pow_relu_rsub_sqrt_1(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)
r2 = rindex
r0 = rindex % 64
tmp0 = tl.load(in_ptr0 + r2, None)
tmp1 = tl.load(in_ptr1 + r0, None, eviction_policy='evict_last')
tmp2 = tmp0 * tmp1
tmp3 = -1.0
tmp4 = tmp0 * tmp3
tmp5 = 1.0
tmp6 = tmp4 + tmp5
tmp7 = 1e-09
tmp8 = tmp1 + tmp7
tmp9 = libdevice.sqrt(tmp8)
tmp10 = 4.0
tmp11 = tmp10 - tmp9
tmp12 = tl.full([1], 0, tl.int32)
tmp13 = triton_helpers.maximum(tmp12, tmp11)
tmp14 = tmp13 * tmp13
tmp15 = tmp6 * tmp14
tmp16 = tmp2 + tmp15
tmp17 = 0.5
tmp18 = tmp16 * tmp17
tmp19 = tl.broadcast_to(tmp18, [RBLOCK])
tmp21 = triton_helpers.promote_to_tensor(tl.sum(tmp19, 0))
tmp22 = 256.0
tmp23 = tmp21 / tmp22
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp23, None)
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_pow_sub_sum_0[grid(64)](arg0_1, arg1_1, buf0, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
triton_per_fused_add_mean_mul_pow_relu_rsub_sqrt_1[grid(1)](buf2,
arg2_1, buf0, 1, 256, num_warps=2, num_stages=1)
del arg2_1
del buf0
return buf2,
class ContrastiveLossNew(nn.Module):
"""
Contrastive loss
Takes embeddings of two samples and a target label == 1 if samples are from the same class and label == 0 otherwise
"""
def __init__(self, margin):
super(ContrastiveLossNew, self).__init__()
self.margin = margin
self.eps = 1e-09
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]
| EmmaW8/EISNet | ContrastiveLoss | false | 8,038 | [
"MIT"
] | 40 | 97c420d6763c5f825e88ed732edee4e75f3b947e | https://github.com/EmmaW8/EISNet/tree/97c420d6763c5f825e88ed732edee4e75f3b947e | import torch
import torch.nn as nn
from torch.nn import functional as F
class Model(nn.Module):
"""
Contrastive loss
Takes embeddings of two samples and a target label == 1 if samples are from the same class and label == 0 otherwise
"""
def __init__(self, margin):
super().__init__()
self.margin = margin
self.eps = 1e-09
def forward(self, output1, output2, target, size_average=True):
distances = (output2 - output1).pow(2).sum(1)
losses = 0.5 * (target.float() * distances + (1 + -1 * target).
float() * F.relu(self.margin - (distances + self.eps).sqrt()).
pow(2))
return losses.mean() if size_average else losses.sum()
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 [4]
|
GlobalAveragePool | # 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_1/inductor_cache/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.py
# Topologically Sorted Source Nodes: [avg], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# avg => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [-2, -1], 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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):
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, 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: [avg], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, arg0_1, 16, 16, grid=grid(16), 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
from torch import nn
class GlobalAveragePool(nn.Module):
"""
Average pooling in an equivariant network
"""
def __init__(self):
"""
"""
super().__init__()
def forward(self, x):
"""
"""
avg = torch.mean(x, dim=[-2, -1], keepdim=True)
return avg
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
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):
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, 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, arg0_1, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
del arg0_1
return buf1,
class GlobalAveragePoolNew(nn.Module):
"""
Average pooling in an equivariant network
"""
def __init__(self):
"""
"""
super().__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| ElisevanderPol/symmetrizer | GlobalAveragePool | false | 8,039 | [
"MIT"
] | 16 | 8dae02bee2ba7132ae4fb07e07020767d280842c | https://github.com/ElisevanderPol/symmetrizer/tree/8dae02bee2ba7132ae4fb07e07020767d280842c | import torch
from torch import nn
class Model(nn.Module):
"""
Average pooling in an equivariant network
"""
def __init__(self):
"""
"""
super().__init__()
def forward(self, x):
"""
"""
avg = torch.mean(x, dim=[-2, -1], keepdim=True)
return avg
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
TripletLoss | # 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_1/inductor_cache/5d/c5dtcurhbtazyithtpfbdetxra6jy6ksi52gzmuuc2ffkwgnlw4w.py
# Topologically Sorted Source Nodes: [sub, pow_1, distance_positive, sub_1, pow_2, distance_negative, sub_2, add, losses, mean], Original ATen: [aten.sub, aten.pow, aten.sum, aten.add, aten.relu, aten.mean]
# Source node to ATen node mapping:
# add => add
# distance_negative => sum_2
# distance_positive => sum_1
# losses => relu
# mean => mean
# pow_1 => pow_1
# pow_2 => pow_2
# sub => sub
# sub_1 => sub_1
# sub_2 => sub_2
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [1]), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %arg2_1), kwargs = {})
# %pow_2 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_1, 2), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_2, [1]), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sum_1, %sum_2), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_2, 4), kwargs = {})
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%add,), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%relu,), kwargs = {})
triton_per_fused_add_mean_pow_relu_sub_sum_0 = async_compile.triton('triton_per_fused_add_mean_pow_relu_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: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {4: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 5), equal_to_1=(4,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_mean_pow_relu_sub_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 12, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_relu_sub_sum_0(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, 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 % 16
r1 = (rindex // 16)
r2 = rindex
tmp0 = tl.load(in_ptr0 + (r0 + (64*r1)), None)
tmp1 = tl.load(in_ptr1 + (r0 + (64*r1)), None)
tmp4 = tl.load(in_ptr0 + (16 + r0 + (64*r1)), None)
tmp5 = tl.load(in_ptr1 + (16 + r0 + (64*r1)), None)
tmp9 = tl.load(in_ptr0 + (32 + r0 + (64*r1)), None)
tmp10 = tl.load(in_ptr1 + (32 + r0 + (64*r1)), None)
tmp14 = tl.load(in_ptr0 + (48 + r0 + (64*r1)), None)
tmp15 = tl.load(in_ptr1 + (48 + r0 + (64*r1)), None)
tmp19 = tl.load(in_ptr2 + (r0 + (64*r1)), None)
tmp22 = tl.load(in_ptr2 + (16 + r0 + (64*r1)), None)
tmp26 = tl.load(in_ptr2 + (32 + r0 + (64*r1)), None)
tmp30 = tl.load(in_ptr2 + (48 + r0 + (64*r1)), None)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp6
tmp8 = tmp3 + tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tmp20 = tmp0 - tmp19
tmp21 = tmp20 * tmp20
tmp23 = tmp4 - tmp22
tmp24 = tmp23 * tmp23
tmp25 = tmp21 + tmp24
tmp27 = tmp9 - tmp26
tmp28 = tmp27 * tmp27
tmp29 = tmp25 + tmp28
tmp31 = tmp14 - tmp30
tmp32 = tmp31 * tmp31
tmp33 = tmp29 + tmp32
tmp34 = tmp18 - tmp33
tmp35 = 4.0
tmp36 = tmp34 + tmp35
tmp37 = tl.full([1, 1], 0, tl.int32)
tmp38 = triton_helpers.maximum(tmp37, tmp36)
tmp39 = tl.broadcast_to(tmp38, [XBLOCK, RBLOCK])
tmp41 = tl.sum(tmp39, 1)[:, None]
tmp42 = 64.0
tmp43 = tmp41 / tmp42
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp43, None)
''', 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)
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [sub, pow_1, distance_positive, sub_1, pow_2, distance_negative, sub_2, add, losses, mean], Original ATen: [aten.sub, aten.pow, aten.sum, aten.add, aten.relu, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_add_mean_pow_relu_sub_sum_0.run(buf2, arg0_1, arg1_1, arg2_1, 1, 64, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
del arg2_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, 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
from torch.nn import functional as F
class TripletLoss(nn.Module):
"""
Triplet loss
Takes embeddings of an anchor sample, a positive sample and a negative sample
"""
def __init__(self, margin):
super(TripletLoss, self).__init__()
self.margin = margin
def forward(self, anchor, positive, negative, size_average=True):
distance_positive = (anchor - positive).pow(2).sum(1)
distance_negative = (anchor - negative).pow(2).sum(1)
losses = F.relu(distance_positive - distance_negative + self.margin)
return losses.mean() if size_average else losses.sum()
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 [[], {'margin': 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_per_fused_add_mean_pow_relu_sub_sum_0(in_out_ptr0, in_ptr0,
in_ptr1, in_ptr2, 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 % 16
r1 = rindex // 16
tmp0 = tl.load(in_ptr0 + (r0 + 64 * r1), None)
tmp1 = tl.load(in_ptr1 + (r0 + 64 * r1), None)
tmp4 = tl.load(in_ptr0 + (16 + r0 + 64 * r1), None)
tmp5 = tl.load(in_ptr1 + (16 + r0 + 64 * r1), None)
tmp9 = tl.load(in_ptr0 + (32 + r0 + 64 * r1), None)
tmp10 = tl.load(in_ptr1 + (32 + r0 + 64 * r1), None)
tmp14 = tl.load(in_ptr0 + (48 + r0 + 64 * r1), None)
tmp15 = tl.load(in_ptr1 + (48 + r0 + 64 * r1), None)
tmp19 = tl.load(in_ptr2 + (r0 + 64 * r1), None)
tmp22 = tl.load(in_ptr2 + (16 + r0 + 64 * r1), None)
tmp26 = tl.load(in_ptr2 + (32 + r0 + 64 * r1), None)
tmp30 = tl.load(in_ptr2 + (48 + r0 + 64 * r1), None)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp6 = tmp4 - tmp5
tmp7 = tmp6 * tmp6
tmp8 = tmp3 + tmp7
tmp11 = tmp9 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tmp8 + tmp12
tmp16 = tmp14 - tmp15
tmp17 = tmp16 * tmp16
tmp18 = tmp13 + tmp17
tmp20 = tmp0 - tmp19
tmp21 = tmp20 * tmp20
tmp23 = tmp4 - tmp22
tmp24 = tmp23 * tmp23
tmp25 = tmp21 + tmp24
tmp27 = tmp9 - tmp26
tmp28 = tmp27 * tmp27
tmp29 = tmp25 + tmp28
tmp31 = tmp14 - tmp30
tmp32 = tmp31 * tmp31
tmp33 = tmp29 + tmp32
tmp34 = tmp18 - tmp33
tmp35 = 4.0
tmp36 = tmp34 + tmp35
tmp37 = tl.full([1, 1], 0, tl.int32)
tmp38 = triton_helpers.maximum(tmp37, tmp36)
tmp39 = tl.broadcast_to(tmp38, [XBLOCK, RBLOCK])
tmp41 = tl.sum(tmp39, 1)[:, None]
tmp42 = 64.0
tmp43 = tmp41 / tmp42
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp43, None)
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)
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
get_raw_stream(0)
triton_per_fused_add_mean_pow_relu_sub_sum_0[grid(1)](buf2, arg0_1,
arg1_1, arg2_1, 1, 64, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
del arg2_1
return buf2,
class TripletLossNew(nn.Module):
"""
Triplet loss
Takes embeddings of an anchor sample, a positive sample and a negative sample
"""
def __init__(self, margin):
super(TripletLossNew, self).__init__()
self.margin = margin
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]
| EmmaW8/EISNet | TripletLoss | false | 8,040 | [
"MIT"
] | 40 | 97c420d6763c5f825e88ed732edee4e75f3b947e | https://github.com/EmmaW8/EISNet/tree/97c420d6763c5f825e88ed732edee4e75f3b947e | import torch
import torch.nn as nn
from torch.nn import functional as F
class Model(nn.Module):
"""
Triplet loss
Takes embeddings of an anchor sample, a positive sample and a negative sample
"""
def __init__(self, margin):
super().__init__()
self.margin = margin
def forward(self, anchor, positive, negative, size_average=True):
distance_positive = (anchor - positive).pow(2).sum(1)
distance_negative = (anchor - negative).pow(2).sum(1)
losses = F.relu(distance_positive - distance_negative + self.margin)
return losses.mean() if size_average else losses.sum()
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 [4]
|
AdaptiveCatAvgMaxPool2d | # 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_1/inductor_cache/64/c64w46a7i4izwmvmm5ywpkpxddqyczhqj5il24twnnockwwjrev5.py
# Topologically Sorted Source Nodes: [x_max], Original ATen: [aten.adaptive_max_pool2d]
# Source node to ATen node mapping:
# x_max => adaptive_max_pool2d
# Graph fragment:
# %adaptive_max_pool2d : [num_users=1] = call_function[target=torch.ops.aten.adaptive_max_pool2d.default](args = (%arg0_1, [1, 1]), kwargs = {})
triton_poi_fused_adaptive_max_pool2d_0 = async_compile.triton('triton_poi_fused_adaptive_max_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 2), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_adaptive_max_pool2d_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 16, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_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
x2 = xindex
x0 = xindex % 4
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (16*x2), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (2 + (16*x2)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (3 + (16*x2)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (4 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (5 + (16*x2)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (6 + (16*x2)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (7 + (16*x2)), xmask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr0 + (8 + (16*x2)), xmask, eviction_policy='evict_last')
tmp17 = tl.load(in_ptr0 + (9 + (16*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (10 + (16*x2)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (11 + (16*x2)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr0 + (12 + (16*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (13 + (16*x2)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr0 + (14 + (16*x2)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr0 + (15 + (16*x2)), xmask, eviction_policy='evict_last')
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 + (x0 + (8*x1)), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/eb/cebjmouk3zywsk4q2f3nfc4le7kmjjf6suapzs6bh5k5zq36xtun.py
# Topologically Sorted Source Nodes: [x_avg], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# x_avg => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%arg0_1, [-1, -2], True), kwargs = {})
triton_per_fused_mean_1 = async_compile.triton('triton_per_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.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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_mean_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_1(in_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)
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_ptr1 + (x2 + (8*x3)), 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)
buf3 = empty_strided_cuda((4, 8, 1, 1), (8, 1, 1, 1), torch.float32)
buf0 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 4) # alias
# Topologically Sorted Source Nodes: [x_max], Original ATen: [aten.adaptive_max_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_adaptive_max_pool2d_0.run(arg0_1, buf0, 16, grid=grid(16), stream=stream0)
buf2 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 0) # alias
# Topologically Sorted Source Nodes: [x_avg], Original ATen: [aten.mean]
triton_per_fused_mean_1.run(arg0_1, buf2, 16, 16, grid=grid(16), 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
from torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
def adaptive_catavgmax_pool2d(x, output_size=1):
x_avg = F.adaptive_avg_pool2d(x, output_size)
x_max = F.adaptive_max_pool2d(x, output_size)
return torch.cat((x_avg, x_max), 1)
class AdaptiveCatAvgMaxPool2d(nn.Module):
def __init__(self, output_size=1):
super(AdaptiveCatAvgMaxPool2d, self).__init__()
self.output_size = output_size
def forward(self, x):
return adaptive_catavgmax_pool2d(x, self.output_size)
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 torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
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_poi_fused_adaptive_max_pool2d_0(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + 16 * x2, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr0 + (2 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp5 = tl.load(in_ptr0 + (3 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp7 = tl.load(in_ptr0 + (4 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp9 = tl.load(in_ptr0 + (5 + 16 * x2), xmask, eviction_policy='evict_last'
)
tmp11 = tl.load(in_ptr0 + (6 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp13 = tl.load(in_ptr0 + (7 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp15 = tl.load(in_ptr0 + (8 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp17 = tl.load(in_ptr0 + (9 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr0 + (10 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp21 = tl.load(in_ptr0 + (11 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr0 + (12 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr0 + (13 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp27 = tl.load(in_ptr0 + (14 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr0 + (15 + 16 * x2), xmask, eviction_policy=
'evict_last')
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 + (x0 + 8 * x1), tmp30, xmask)
@triton.jit
def triton_per_fused_mean_1(in_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)
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_ptr1 + (x2 + 8 * x3), 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)
buf3 = empty_strided_cuda((4, 8, 1, 1), (8, 1, 1, 1), torch.float32)
buf0 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 4)
get_raw_stream(0)
triton_poi_fused_adaptive_max_pool2d_0[grid(16)](arg0_1, buf0, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf2 = reinterpret_tensor(buf3, (4, 4, 1, 1), (8, 1, 1, 1), 0)
triton_per_fused_mean_1[grid(16)](arg0_1, buf2, 16, 16, XBLOCK=1,
num_warps=2, num_stages=1)
del arg0_1
return buf3,
def adaptive_catavgmax_pool2d(x, output_size=1):
x_avg = F.adaptive_avg_pool2d(x, output_size)
x_max = F.adaptive_max_pool2d(x, output_size)
return torch.cat((x_avg, x_max), 1)
class AdaptiveCatAvgMaxPool2dNew(nn.Module):
def __init__(self, output_size=1):
super(AdaptiveCatAvgMaxPool2dNew, self).__init__()
self.output_size = output_size
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
| Alibaba-MIIL/ZS_SDL | AdaptiveCatAvgMaxPool2d | false | 8,041 | [
"MIT"
] | 20 | 769fe4f57d2d458a7c4b5468a6395c9b296b1dad | https://github.com/Alibaba-MIIL/ZS_SDL/tree/769fe4f57d2d458a7c4b5468a6395c9b296b1dad | import torch
from torchvision import datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.parallel
import torch.optim
import torch.utils.data.distributed
def adaptive_catavgmax_pool2d(x, output_size=1):
x_avg = F.adaptive_avg_pool2d(x, output_size)
x_max = F.adaptive_max_pool2d(x, output_size)
return torch.cat((x_avg, x_max), 1)
class Model(nn.Module):
def __init__(self, output_size=1):
super().__init__()
self.output_size = output_size
def forward(self, x):
return adaptive_catavgmax_pool2d(x, self.output_size)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
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_1/inductor_cache/nn/cnnz6kyugk4ex7z2gna45k3yovarjt3y33yp6vb6ho7xpwkopvda.py
# Topologically Sorted Source Nodes: [invprobs, neg_3, mul_1, sub_2, mul_2, mul_3, exp_2, mul, sub, neg, max_val, add, neg_1, exp, neg_2, sub_1, exp_1, add_1, log, loss, loss_1, mean], Original ATen: [aten.log_sigmoid_forward, aten.neg, aten.mul, aten.sub, aten.exp, aten.clamp, aten.add, aten.log, aten.mean]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# exp => exp
# exp_1 => exp_1
# exp_2 => exp_3
# invprobs => abs_1, exp_2, full_default, log1p, minimum, neg_4, sub_3
# log => log
# loss => add_2
# loss_1 => mul_4
# max_val => clamp_min
# mean => mean
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# mul_3 => mul_3
# neg => neg
# neg_1 => neg_1
# neg_2 => neg_2
# neg_3 => neg_3
# sub => sub
# sub_1 => sub_1
# sub_2 => 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})
# %neg_3 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%arg1_1,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 2.0), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_1, 1.0), kwargs = {})
# %mul_2 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%neg_3, %sub_2), 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_4 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp_2 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg_4,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp_2,), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_3, 2), kwargs = {})
# %exp_3 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%mul_3,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg1_1, %arg0_1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %mul), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%arg1_1,), kwargs = {})
# %clamp_min : [num_users=3] = call_function[target=torch.ops.aten.clamp_min.default](args = (%neg, 0), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub, %clamp_min), kwargs = {})
# %neg_1 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%clamp_min,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg_1,), kwargs = {})
# %neg_2 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%arg1_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%neg_2, %clamp_min), kwargs = {})
# %exp_1 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%exp, %exp_1), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add_1,), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %log), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%exp_3, %add_2), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul_4,), kwargs = {})
triton_per_fused_add_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_sub_0 = async_compile.triton('triton_per_fused_add_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_sub_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': True, 'num_load': 2, 'num_reduction': 1, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_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_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_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)
tmp2 = tl.load(in_ptr1 + (r0), None)
tmp1 = -tmp0
tmp3 = 2.0
tmp4 = tmp2 * tmp3
tmp5 = 1.0
tmp6 = tmp4 - tmp5
tmp7 = tmp1 * tmp6
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 * tmp3
tmp16 = tl_math.exp(tmp15)
tmp17 = tmp0 * tmp2
tmp18 = tmp0 - tmp17
tmp19 = triton_helpers.maximum(tmp1, tmp8)
tmp20 = tmp18 + tmp19
tmp21 = -tmp19
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp1 - tmp19
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tl_math.log(tmp25)
tmp27 = tmp20 + tmp26
tmp28 = tmp16 * tmp27
tmp29 = tl.broadcast_to(tmp28, [RBLOCK])
tmp31 = triton_helpers.promote_to_tensor(tl.sum(tmp29, 0))
tmp32 = 256.0
tmp33 = tmp31 / tmp32
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp33, 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: [invprobs, neg_3, mul_1, sub_2, mul_2, mul_3, exp_2, mul, sub, neg, max_val, add, neg_1, exp, neg_2, sub_1, exp_1, add_1, log, loss, loss_1, mean], Original ATen: [aten.log_sigmoid_forward, aten.neg, aten.mul, aten.sub, aten.exp, aten.clamp, aten.add, aten.log, aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_add_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_sub_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
import torch.nn.functional as F
class FocalLoss(nn.Module):
def __init__(self, gamma=2):
super().__init__()
self.gamma = gamma
def forward(self, logit, target):
target = target.float()
max_val = (-logit).clamp(min=0)
loss = logit - logit * target + max_val + ((-max_val).exp() + (-
logit - max_val).exp()).log()
invprobs = F.logsigmoid(-logit * (target * 2.0 - 1.0))
loss = (invprobs * self.gamma).exp() * loss
if len(loss.size()) == 2:
loss = loss.sum(dim=1)
return loss.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
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_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_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)
tmp2 = tl.load(in_ptr1 + r0, None)
tmp1 = -tmp0
tmp3 = 2.0
tmp4 = tmp2 * tmp3
tmp5 = 1.0
tmp6 = tmp4 - tmp5
tmp7 = tmp1 * tmp6
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 * tmp3
tmp16 = tl_math.exp(tmp15)
tmp17 = tmp0 * tmp2
tmp18 = tmp0 - tmp17
tmp19 = triton_helpers.maximum(tmp1, tmp8)
tmp20 = tmp18 + tmp19
tmp21 = -tmp19
tmp22 = tl_math.exp(tmp21)
tmp23 = tmp1 - tmp19
tmp24 = tl_math.exp(tmp23)
tmp25 = tmp22 + tmp24
tmp26 = tl_math.log(tmp25)
tmp27 = tmp20 + tmp26
tmp28 = tmp16 * tmp27
tmp29 = tl.broadcast_to(tmp28, [RBLOCK])
tmp31 = triton_helpers.promote_to_tensor(tl.sum(tmp29, 0))
tmp32 = 256.0
tmp33 = tmp31 / tmp32
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp33, 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_clamp_exp_log_log_sigmoid_forward_mean_mul_neg_sub_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 FocalLossNew(nn.Module):
def __init__(self, gamma=2):
super().__init__()
self.gamma = gamma
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| EMUNES/Auto-Subtitle-File-Generation | FocalLoss | false | 8,042 | [
"Apache-2.0"
] | 33 | 535a6351f450b1970da50bbbf4cc6d2f442ec335 | https://github.com/EMUNES/Auto-Subtitle-File-Generation/tree/535a6351f450b1970da50bbbf4cc6d2f442ec335 | import torch
import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
def __init__(self, gamma=2):
super().__init__()
self.gamma = gamma
def forward(self, logit, target):
target = target.float()
max_val = (-logit).clamp(min=0)
loss = logit - logit * target + max_val + ((-max_val).exp() + (-
logit - max_val).exp()).log()
invprobs = F.logsigmoid(-logit * (target * 2.0 - 1.0))
loss = (invprobs * self.gamma).exp() * loss
if len(loss.size()) == 2:
loss = loss.sum(dim=1)
return loss.mean()
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Conv2d | # 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_1/inductor_cache/ho/cho72zpf3yxrxsxjoqgiu5dmii3lj4efdjr3rk7fhvnzcxosbxxn.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => convolution
# x_1 => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_convolution_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_convolution_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=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), '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_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_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_0(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
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 1, 1), (4, 1, 1, 1))
buf1 = buf0; del buf0 # reuse
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_threshold_backward_0.run(buf1, primals_2, buf2, 16, grid=grid(16), stream=stream0)
del primals_2
return (buf1, primals_1, primals_3, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), 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 Conv2d(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride=1, NL
='relu', same_padding=False, bn=False, dilation=1):
super(Conv2d, self).__init__()
padding = int((kernel_size - 1) / 2) if same_padding else 0
self.conv = []
if dilation == 1:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size,
stride, padding=padding, dilation=dilation)
else:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size,
stride, padding=dilation, dilation=dilation)
self.bn = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0,
affine=True) if bn else None
if NL == 'relu':
self.relu = nn.ReLU(inplace=True)
elif NL == 'prelu':
self.relu = nn.PReLU()
else:
self.relu = None
def forward(self, x):
x = self.conv(x)
if self.bn is not None:
x = self.bn(x)
if self.relu is not None:
x = self.relu(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4, 'kernel_size': 4}]
| import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_convolution_relu_threshold_backward_0(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
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 1, 1), (4, 1, 1, 1))
buf1 = buf0
del buf0
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_convolution_relu_threshold_backward_0[grid(16)](buf1,
primals_2, buf2, 16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_2
return buf1, primals_1, primals_3, buf2
class Conv2dNew(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride=1, NL
='relu', same_padding=False, bn=False, dilation=1):
super(Conv2dNew, self).__init__()
padding = int((kernel_size - 1) / 2) if same_padding else 0
self.conv = []
if dilation == 1:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size,
stride, padding=padding, dilation=dilation)
else:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size,
stride, padding=dilation, dilation=dilation)
self.bn = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0,
affine=True) if bn else None
if NL == 'relu':
self.relu = nn.ReLU(inplace=True)
elif NL == 'prelu':
self.relu = nn.PReLU()
else:
self.relu = None
def forward(self, input_0):
primals_1 = self.conv.weight
primals_2 = self.conv.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
| Elin24/DCL-CrowdCounting | Conv2d | false | 8,043 | [
"MIT"
] | 12 | 2f8e68a2d29a8599e795b502f21b4de778e6214c | https://github.com/Elin24/DCL-CrowdCounting/tree/2f8e68a2d29a8599e795b502f21b4de778e6214c | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, stride=1, NL
='relu', same_padding=False, bn=False, dilation=1):
super().__init__()
padding = int((kernel_size - 1) / 2) if same_padding else 0
self.conv = []
if dilation == 1:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size,
stride, padding=padding, dilation=dilation)
else:
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size,
stride, padding=dilation, dilation=dilation)
self.bn = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0,
affine=True) if bn else None
if NL == 'relu':
self.relu = nn.ReLU(inplace=True)
elif NL == 'prelu':
self.relu = nn.PReLU()
else:
self.relu = None
def forward(self, x):
x = self.conv(x)
if self.bn is not None:
x = self.bn(x)
if self.relu is not None:
x = self.relu(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [4, 4, 4]
|
NoiseInjection | # 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_1/inductor_cache/5i/c5itgadrrjqedncsqmt7gkyi2u4rtmc72mygkdrubora2tjkucz3.py
# Topologically Sorted Source Nodes: [mul, add], Original ATen: [aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %normal_functional), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %mul), kwargs = {})
triton_poi_fused_add_mul_0 = async_compile.triton('triton_poi_fused_add_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tl.load(in_ptr2 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp4 = tmp2 * tmp3
tmp5 = tmp0 + tmp4
tl.store(out_ptr0 + (x3), 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, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, ), (1, ))
buf0 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [noise], Original ATen: [aten.normal_functional]
buf1 = torch.ops.aten.normal_functional.default(buf0)
del buf0
buf2 = buf1
del buf1
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, add], Original ATen: [aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_0.run(primals_1, primals_2, buf2, buf3, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_2
return (buf3, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, ), (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
from torch import nn
class NoiseInjection(nn.Module):
def __init__(self):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1))
def forward(self, image, noise=None):
if noise is None:
batch, _, height, width = image.shape
noise = image.new_empty(batch, 1, height, width).normal_()
return image + self.weight * noise
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
| import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_mul_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tl.load(in_ptr2 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tmp2 * tmp3
tmp5 = tmp0 + tmp4
tl.store(out_ptr0 + x3, tmp5, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1,), (1,))
buf0 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf1 = torch.ops.aten.normal_functional.default(buf0)
del buf0
buf2 = buf1
del buf1
buf3 = 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, buf2,
buf3, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
del primals_2
return buf3, buf2
class NoiseInjectionNew(nn.Module):
def __init__(self):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1))
def forward(self, input_0):
primals_2 = self.weight
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
| Dolorousrtur/style-people | NoiseInjection | false | 8,044 | [
"MIT"
] | 15 | c48b12b245cc50f8230c0654dffe40016f2a69f1 | https://github.com/Dolorousrtur/style-people/tree/c48b12b245cc50f8230c0654dffe40016f2a69f1 | import torch
from torch import nn
class Model(nn.Module):
def __init__(self):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1))
def forward(self, image, noise=None):
if noise is None:
batch, _, height, width = image.shape
noise = image.new_empty(batch, 1, height, width).normal_()
return image + self.weight * noise
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
DiceLoss | # AOT ID: ['0_inference']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/m6/cm6wzurewzqnksq5u7c7eenepi63vxbk7q7xtmblsrhi2aja4o7a.py
# Topologically Sorted Source Nodes: [mul, a, mul_3, mul_1, sum_2, b, mul_2, sum_3, c, add_2, d, sub], Original ATen: [aten.mul, aten.sum, aten.add, aten.div, aten.rsub]
# Source node to ATen node mapping:
# a => sum_1
# add_2 => add_2
# b => add
# c => add_1
# d => div
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# mul_3 => mul_3
# sub => sub
# sum_2 => sum_2
# sum_3 => sum_3
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %view_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1]), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %view), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_1, [1]), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, 0.001), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_1, %view_1), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul_2, [1]), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_3, 0.001), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %add_1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul_3, %add_2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %div), kwargs = {})
triton_per_fused_add_div_mul_rsub_sum_0 = async_compile.triton('triton_per_fused_add_div_mul_rsub_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 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=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_mul_rsub_sum_0', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 3, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused_add_div_mul_rsub_sum_0(in_out_ptr0, in_ptr0, in_ptr1, xnumel, rnumel, 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.load(in_ptr1 + (r1 + (64*x0)), 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]
tmp7 = tmp0 * tmp0
tmp8 = tl.broadcast_to(tmp7, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tmp1 * tmp1
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = 2.0
tmp18 = tmp6 * tmp17
tmp19 = 0.001
tmp20 = tmp11 + tmp19
tmp21 = tmp16 + tmp19
tmp22 = tmp20 + tmp21
tmp23 = tmp18 / tmp22
tmp24 = 1.0
tmp25 = tmp24 - tmp23
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp25, 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, ), (1, ), torch.float32)
buf3 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [mul, a, mul_3, mul_1, sum_2, b, mul_2, sum_3, c, add_2, d, sub], Original ATen: [aten.mul, aten.sum, aten.add, aten.div, aten.rsub]
stream0 = get_raw_stream(0)
triton_per_fused_add_div_mul_rsub_sum_0.run(buf3, arg0_1, arg1_1, 4, 64, grid=grid(4), stream=stream0)
del arg0_1
del arg1_1
return (buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
| import torch
import torch.nn as nn
class DiceLoss(nn.Module):
def __init__(self):
super(DiceLoss, self).__init__()
pass
def forward(self, input, target):
input = input.contiguous().view(input.size()[0], -1)
target = target.contiguous().view(target.size()[0], -1).float()
a = torch.sum(input * target, 1)
b = torch.sum(input * input, 1) + 0.001
c = torch.sum(target * target, 1) + 0.001
d = 2 * a / (b + c)
return 1 - d
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_per_fused_add_div_mul_rsub_sum_0(in_out_ptr0, in_ptr0, in_ptr1,
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.load(in_ptr1 + (r1 + 64 * x0), 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]
tmp7 = tmp0 * tmp0
tmp8 = tl.broadcast_to(tmp7, [XBLOCK, RBLOCK])
tmp10 = tl.where(xmask, tmp8, 0)
tmp11 = tl.sum(tmp10, 1)[:, None]
tmp12 = tmp1 * tmp1
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = 2.0
tmp18 = tmp6 * tmp17
tmp19 = 0.001
tmp20 = tmp11 + tmp19
tmp21 = tmp16 + tmp19
tmp22 = tmp20 + tmp21
tmp23 = tmp18 / tmp22
tmp24 = 1.0
tmp25 = tmp24 - tmp23
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp25, 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,), (1,), torch.float32)
buf3 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_add_div_mul_rsub_sum_0[grid(4)](buf3, arg0_1,
arg1_1, 4, 64, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf3,
class DiceLossNew(nn.Module):
def __init__(self):
super(DiceLossNew, self).__init__()
pass
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
| EryiXie/PlaneRecNet | DiceLoss | false | 8,045 | [
"MIT"
] | 34 | 534e23e6c5db2235ab1e5a9419fb4bfec3ffa943 | https://github.com/EryiXie/PlaneRecNet/tree/534e23e6c5db2235ab1e5a9419fb4bfec3ffa943 | import torch
import torch.nn as nn
class Model(nn.Module):
def __init__(self):
super().__init__()
pass
def forward(self, input, target):
input = input.contiguous().view(input.size()[0], -1)
target = target.contiguous().view(target.size()[0], -1).float()
a = torch.sum(input * target, 1)
b = torch.sum(input * input, 1) + 0.001
c = torch.sum(target * target, 1) + 0.001
d = 2 * a / (b + c)
return 1 - d
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return []
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.