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
listlengths 1
6
| stars
int64 0
19.8k
| sha
stringlengths 40
40
| repo_link
stringlengths 72
180
|
|---|---|---|---|---|---|---|---|---|---|---|---|
IDPredictor
|
# 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: [pred_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# pred_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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/cq/ccqbfbzeetpsc2d2hbydlb6zh44oakzviqnnh5rue727xy45hkgn.py
# Topologically Sorted Source Nodes: [pred_3], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# pred_3 => relu_1
# Graph fragment:
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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 = 8192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (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, (128, 256), (256, 1))
assert_size_stride(primals_5, (128, ), (1, ))
assert_size_stride(primals_6, (64, 128), (128, 1))
assert_size_stride(primals_7, (64, ), (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
buf6 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1), torch.bool)
# Topologically Sorted Source Nodes: [pred_1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf6, 16384, grid=grid(16384), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 256), (256, 1), 0), reinterpret_tensor(primals_4, (256, 128), (1, 256), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 128), (2048, 512, 128, 1), 0); del buf2 # reuse
buf5 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1), torch.bool)
# Topologically Sorted Source Nodes: [pred_3], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_5, buf5, 8192, grid=grid(8192), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [pred_4], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 128), (128, 1), 0), reinterpret_tensor(primals_6, (128, 64), (1, 128), 0), alpha=1, beta=1, out=buf4)
del primals_7
return (reinterpret_tensor(buf4, (4, 4, 4, 64), (1024, 256, 64, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 256), (256, 1), 0), reinterpret_tensor(buf3, (64, 128), (128, 1), 0), primals_6, buf5, primals_4, buf6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((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((128, 256), (256, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((64, 128), (128, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn.functional as F
from torch import nn
class IDPredictor(nn.Module):
def __init__(self, nz_feat, n_dim=5):
super(IDPredictor, self).__init__()
self.pred_layer = nn.Linear(nz_feat, 256)
self.sc_layer = nn.Linear(256, 128)
self.sc_layer2 = nn.Linear(128, 64)
def forward(self, feat):
pred = self.pred_layer.forward(feat)
pred = F.relu(pred)
pred = self.sc_layer.forward(pred)
pred = F.relu(pred)
pred = self.sc_layer2.forward(pred)
return pred
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nz_feat': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch 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):
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_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (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, (128, 256), (256, 1))
assert_size_stride(primals_5, (128,), (1,))
assert_size_stride(primals_6, (64, 128), (128, 1))
assert_size_stride(primals_7, (64,), (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
buf6 = 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, buf6, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 256), (256, 1), 0),
reinterpret_tensor(primals_4, (256, 128), (1, 256), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 128), (2048, 512, 128, 1), 0)
del buf2
buf5 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(8192)](buf3,
primals_5, buf5, 8192, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 128),
(128, 1), 0), reinterpret_tensor(primals_6, (128, 64), (1, 128),
0), alpha=1, beta=1, out=buf4)
del primals_7
return reinterpret_tensor(buf4, (4, 4, 4, 64), (1024, 256, 64, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 256), (256, 1), 0
), reinterpret_tensor(buf3, (64, 128), (128, 1), 0
), primals_6, buf5, primals_4, buf6
class IDPredictorNew(nn.Module):
def __init__(self, nz_feat, n_dim=5):
super(IDPredictorNew, self).__init__()
self.pred_layer = nn.Linear(nz_feat, 256)
self.sc_layer = nn.Linear(256, 128)
self.sc_layer2 = nn.Linear(128, 64)
def forward(self, input_0):
primals_1 = self.pred_layer.weight
primals_2 = self.pred_layer.bias
primals_4 = self.sc_layer.weight
primals_5 = self.sc_layer.bias
primals_6 = self.sc_layer2.weight
primals_7 = self.sc_layer2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
|
JasonQSY/Associative3D
|
IDPredictor
| false
| 8,350
|
[
"MIT"
] | 25
|
c50818b593ec48c38ed7ee3e109c23531089da32
|
https://github.com/JasonQSY/Associative3D/tree/c50818b593ec48c38ed7ee3e109c23531089da32
|
ConvModule
|
# 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: [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_1, %primals_2, %primals_3, [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, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_3, (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_1, primals_2, 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: [x], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_3, 256, grid=grid(256), stream=stream0)
del primals_3
return (buf1, primals_1, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.utils.data
import torch.nn as nn
def _get_activation(activation):
valid = ['relu', 'leaky_relu', 'lrelu', 'tanh', 'sigmoid']
assert activation in valid, 'activation should be one of {}'.format(valid)
if activation == 'relu':
return nn.ReLU(inplace=True)
if activation == 'leaky_relu' or activation == 'lrelu':
return nn.LeakyReLU(inplace=True)
if activation == 'sigmoid':
return nn.Sigmoid()
if activation == 'tanh':
return nn.Tanh()
return None
def _init_fc_or_conv(fc_conv, activation):
gain = 1.0
if activation is not None:
gain = nn.init.calculate_gain(activation)
nn.init.xavier_uniform_(fc_conv.weight, gain)
if fc_conv.bias is not None:
nn.init.constant_(fc_conv.bias, 0.0)
def _get_norm_layer(norm_layer, channels):
valid = ['instance', 'batch']
assert norm_layer in valid, 'norm_layer should be one of {}'.format(valid)
if norm_layer == 'instance':
layer = nn.InstanceNorm2d(channels, affine=True)
elif norm_layer == 'batch':
layer = nn.BatchNorm2d(channels, affine=True)
nn.init.constant_(layer.bias, 0.0)
nn.init.constant_(layer.weight, 1.0)
return layer
class ConvModule(nn.Module):
"""Basic convolution module with conv + norm(optional) + activation(optional).
Args:
n_in(int): number of input channels.
n_out(int): number of output channels.
ksize(int): size of the convolution kernel (square).
stride(int): downsampling factor
pad(bool): if True, pad the convolutions to maintain a constant size.
padding_mode(str): 'zero' or 'reflection'.
activation(str): nonlinear activation function between convolutions.
norm_layer(str): normalization to apply between the convolution modules.
"""
def __init__(self, n_in, n_out, ksize=3, stride=1, pad=True,
padding_mode='zero', activation=None, norm_layer=None):
super(ConvModule, self).__init__()
assert isinstance(n_in, int
) and n_in > 0, 'Input channels should be a positive integer got {}'.format(
n_in)
assert isinstance(n_out, int
) and n_out > 0, 'Output channels should be a positive integer got {}'.format(
n_out)
assert isinstance(ksize, int
) and ksize > 0, 'Kernel size should be a positive integer got {}'.format(
ksize)
assert padding_mode in ['zero', 'reflection'], 'Invalid padding mode'
padding = (ksize - 1) // 2 if pad else 0
use_bias_in_conv = norm_layer is None
conv_pad = padding
if padding_mode == 'reflection' and padding > 0:
self.add_module('reflection_pad', nn.ReflectionPad2d(padding))
conv_pad = 0
else:
self.add_module('no-op', nn.Identity())
self.add_module('conv', nn.Conv2d(n_in, n_out, ksize, stride=stride,
padding=conv_pad, bias=use_bias_in_conv))
if norm_layer is not None:
self.add_module('norm', _get_norm_layer(norm_layer, n_out))
if activation is not None:
self.add_module('activation', _get_activation(activation))
_init_fc_or_conv(self.conv, activation)
def forward(self, x):
for c in self.children():
x = c(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_in': 4, 'n_out': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.utils.data
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, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,
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_3, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_3
return buf1, primals_1, primals_2
def _get_activation(activation):
valid = ['relu', 'leaky_relu', 'lrelu', 'tanh', 'sigmoid']
assert activation in valid, 'activation should be one of {}'.format(valid)
if activation == 'relu':
return nn.ReLU(inplace=True)
if activation == 'leaky_relu' or activation == 'lrelu':
return nn.LeakyReLU(inplace=True)
if activation == 'sigmoid':
return nn.Sigmoid()
if activation == 'tanh':
return nn.Tanh()
return None
def _init_fc_or_conv(fc_conv, activation):
gain = 1.0
if activation is not None:
gain = nn.init.calculate_gain(activation)
nn.init.xavier_uniform_(fc_conv.weight, gain)
if fc_conv.bias is not None:
nn.init.constant_(fc_conv.bias, 0.0)
def _get_norm_layer(norm_layer, channels):
valid = ['instance', 'batch']
assert norm_layer in valid, 'norm_layer should be one of {}'.format(valid)
if norm_layer == 'instance':
layer = nn.InstanceNorm2d(channels, affine=True)
elif norm_layer == 'batch':
layer = nn.BatchNorm2d(channels, affine=True)
nn.init.constant_(layer.bias, 0.0)
nn.init.constant_(layer.weight, 1.0)
return layer
class ConvModuleNew(nn.Module):
"""Basic convolution module with conv + norm(optional) + activation(optional).
Args:
n_in(int): number of input channels.
n_out(int): number of output channels.
ksize(int): size of the convolution kernel (square).
stride(int): downsampling factor
pad(bool): if True, pad the convolutions to maintain a constant size.
padding_mode(str): 'zero' or 'reflection'.
activation(str): nonlinear activation function between convolutions.
norm_layer(str): normalization to apply between the convolution modules.
"""
def __init__(self, n_in, n_out, ksize=3, stride=1, pad=True,
padding_mode='zero', activation=None, norm_layer=None):
super(ConvModuleNew, self).__init__()
assert isinstance(n_in, int
) and n_in > 0, 'Input channels should be a positive integer got {}'.format(
n_in)
assert isinstance(n_out, int
) and n_out > 0, 'Output channels should be a positive integer got {}'.format(
n_out)
assert isinstance(ksize, int
) and ksize > 0, 'Kernel size should be a positive integer got {}'.format(
ksize)
assert padding_mode in ['zero', 'reflection'], 'Invalid padding mode'
padding = (ksize - 1) // 2 if pad else 0
use_bias_in_conv = norm_layer is None
conv_pad = padding
if padding_mode == 'reflection' and padding > 0:
self.add_module('reflection_pad', nn.ReflectionPad2d(padding))
conv_pad = 0
else:
self.add_module('no-op', nn.Identity())
self.add_module('conv', nn.Conv2d(n_in, n_out, ksize, stride=stride,
padding=conv_pad, bias=use_bias_in_conv))
if norm_layer is not None:
self.add_module('norm', _get_norm_layer(norm_layer, n_out))
if activation is not None:
self.add_module('activation', _get_activation(activation))
_init_fc_or_conv(self.conv, activation)
def forward(self, input_0):
primals_2 = self.conv.weight
primals_3 = self.conv.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
IlyaBizyaev/ttools
|
ConvModule
| false
| 8,351
|
[
"MIT"
] | 11
|
b1435b19f397ce1baff9daed3cb287e52a029fdb
|
https://github.com/IlyaBizyaev/ttools/tree/b1435b19f397ce1baff9daed3cb287e52a029fdb
|
Normalize
|
# 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/wz/cwzcah4g5b3ejt55evsx6ffyi2xbfvkkf2onb5k7hy6quvpcd2du.py
# Topologically Sorted Source Nodes: [normalize], Original ATen: [aten.div]
# Source node to ATen node mapping:
# normalize => 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
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 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tl.store(out_ptr0 + (x3), tmp15, 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: [normalize], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data.distributed
class Normalize(nn.Module):
def __init__(self, p=2):
super(Normalize, self).__init__()
self.p = p
def forward(self, x):
return F.normalize(x, p=self.p, dim=1)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_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
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 = libdevice.sqrt(tmp11)
tmp13 = 1e-12
tmp14 = triton_helpers.maximum(tmp12, tmp13)
tmp15 = tmp0 / tmp14
tl.store(out_ptr0 + x3, tmp15, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_div_0[grid(256)](arg0_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class NormalizeNew(nn.Module):
def __init__(self, p=2):
super(NormalizeNew, self).__init__()
self.p = p
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
JindongGu/SimDis
|
Normalize
| false
| 8,352
|
[
"MIT"
] | 12
|
0871a217a756acc268f35f802e35b01b12817f0d
|
https://github.com/JindongGu/SimDis/tree/0871a217a756acc268f35f802e35b01b12817f0d
|
MultiHeadAttn
|
# 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/ln/clnt3erhf2vizhltt5ni3tzbxdrqzb24w3hxrxz3felmuz3ea257.py
# Topologically Sorted Source Nodes: [attn_score], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# attn_score => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_5,), 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 % 16
y1 = (yindex // 16)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (32*y1) + (128*x2)), 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/xg/cxgg5ur4hlkqy2n7nqe4avbkgxrnmrwp72sf3zsbul3hqhwb763z.py
# Topologically Sorted Source Nodes: [attn_prob], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_prob => amax, clone_1, exp, sub
# Graph fragment:
# %clone_1 : [num_users=2] = call_function[target=torch.ops.aten.clone.default](args = (%view_15,), kwargs = {memory_format: torch.contiguous_format})
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%clone_1, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%clone_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp3 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = tl_math.exp(tmp14)
tl.store(out_ptr0 + (x2), tmp15, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ai/caii3ty3hjop5wbrwicwqfsequgiktb7f3sryyqan43mr2gzy37v.py
# Topologically Sorted Source Nodes: [attn_prob], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_prob => 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=[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__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, 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
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
tmp0 = tl.load(in_ptr0 + (y3 + (16*x2)), xmask & ymask)
tmp1 = tl.load(in_ptr0 + ((4*y1) + (16*x2)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*y1) + (16*x2)), xmask & ymask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*y1) + (16*x2)), xmask & ymask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*y1) + (16*x2)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2 + (16*y3)), tmp8, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/nh/cnhcwfhpz65jk72mibqqs7c4vsafhgq7u5vwyjuv2twqui6fijel.py
# Topologically Sorted Source Nodes: [attn_vec], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# attn_vec => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_14,), 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=[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_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, 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 + (16 + x0 + (4*x2) + (32*x3) + (128*x1)), xmask)
tl.store(out_ptr0 + (x4), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/xf/cxft7od37hyrfwlrp6j7j6fh4zyeacsalgxq2xf3bneipbygbkkb.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone_4
# Graph fragment:
# %clone_4 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%view_21,), 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) % 16
x2 = (xindex // 64)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2) + (16*x1)), xmask)
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/yv/cyvu7b655f7w4y6fs3cr3d3vawpnn3vmcirao3tw5zgpuuobc2mb.py
# Topologically Sorted Source Nodes: [add, output], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# output => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_24), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [2]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_5', '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_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (4*x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + (x0), tmp16, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/rp/crp6yznjcr5keantuusvl77ssv2xcxe4iqpzesafqd5zf32kmhfv.py
# Topologically Sorted Source Nodes: [add, output], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add => add
# output => add_1, add_2, mul_1, mul_2, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_24), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_2, 1e-05), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, %primals_5), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_6), kwargs = {})
triton_poi_fused_add_native_layer_norm_6 = async_compile.triton('triton_poi_fused_add_native_layer_norm_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_6', '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_6(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-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')
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, (16, 4), (4, 1))
assert_size_stride(primals_3, (32, 4), (4, 1))
assert_size_stride(primals_4, (4, 16), (16, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [head_q], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 16), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((16, 32), (32, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 32), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4, 4, 4, 1), (64, 16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_score], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf1, buf2, 64, 4, grid=grid(64, 4), stream=stream0)
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_score], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 4), (4, 64, 1), 0), reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (4, 1, 64, 16), torch.float32)
# Topologically Sorted Source Nodes: [attn_prob], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf3, buf4, 256, grid=grid(256), stream=stream0)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [attn_prob], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf4, buf5, 16, 16, grid=grid(16, 16), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4, 1), (64, 16, 4, 1, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [attn_vec], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf1, buf6, 256, grid=grid(256), stream=stream0)
del buf1
buf7 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_vec], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf5, (16, 4, 4), (1, 64, 16), 0), reinterpret_tensor(buf6, (16, 4, 4), (16, 4, 1), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf7, buf8, 256, grid=grid(256), stream=stream0)
del buf7
buf9 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_out], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf8, (16, 16), (16, 1), 0), reinterpret_tensor(primals_4, (16, 4), (1, 16), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [add, output], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(primals_1, buf9, buf10, buf11, 16, grid=grid(16), stream=stream0)
buf12 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, output], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_6.run(primals_1, buf9, buf10, buf11, primals_5, primals_6, buf12, 64, grid=grid(64), stream=stream0)
del buf10
del buf11
del primals_6
return (buf12, primals_1, primals_5, buf5, reinterpret_tensor(buf8, (16, 16), (16, 1), 0), buf9, primals_4, reinterpret_tensor(buf6, (16, 4, 4), (16, 1, 4), 0), reinterpret_tensor(buf0, (16, 4, 4), (4, 1, 64), 0), reinterpret_tensor(buf2, (16, 4, 4), (16, 1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((32, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
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 torch
import torch.nn.functional as F
from torch import nn
class MultiHeadAttn(nn.Module):
def __init__(self, n_head, d_model, d_head, dropout, dropatt=0,
pre_lnorm=False):
super(MultiHeadAttn, self).__init__()
self.n_head = n_head
self.d_model = d_model
self.d_head = d_head
self.dropout = dropout
self.q_net = nn.Linear(d_model, n_head * d_head, bias=False)
self.kv_net = nn.Linear(d_model, 2 * n_head * d_head, bias=False)
self.drop = nn.Dropout(dropout)
self.dropatt = nn.Dropout(dropatt)
self.o_net = nn.Linear(n_head * d_head, d_model, bias=False)
self.layer_norm = nn.LayerNorm(d_model)
self.scale = 1 / d_head ** 0.5
self.pre_lnorm = pre_lnorm
def forward(self, h, attn_mask=None, mems=None):
if mems is not None:
c = torch.cat([mems, h], 0)
else:
c = h
if self.pre_lnorm:
c = self.layer_norm(c)
head_q = self.q_net(h)
head_k, head_v = torch.chunk(self.kv_net(c), 2, -1)
head_q = head_q.view(h.size(0), h.size(1), self.n_head, self.d_head)
head_k = head_k.view(c.size(0), c.size(1), self.n_head, self.d_head)
head_v = head_v.view(c.size(0), c.size(1), self.n_head, self.d_head)
attn_score = torch.einsum('ibnd,jbnd->ijbn', (head_q, head_k))
attn_score.mul_(self.scale)
if attn_mask is not None and attn_mask.any().item():
if attn_mask.dim() == 2:
attn_score.masked_fill_(attn_mask[None, :, :, None], -float
('inf'))
elif attn_mask.dim() == 3:
attn_score.masked_fill_(attn_mask[:, :, :, None], -float('inf')
)
attn_prob = F.softmax(attn_score, dim=1)
attn_prob = self.dropatt(attn_prob)
attn_vec = torch.einsum('ijbn,jbnd->ibnd', (attn_prob, head_v))
attn_vec = attn_vec.contiguous().view(attn_vec.size(0), attn_vec.
size(1), self.n_head * self.d_head)
attn_out = self.o_net(attn_vec)
attn_out = self.drop(attn_out)
if self.pre_lnorm:
output = h + attn_out
else:
output = self.layer_norm(h + attn_out)
return output
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'n_head': 4, 'd_model': 4, 'd_head': 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 libdevice, math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clone_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 64
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 16
y1 = yindex // 16
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 32 * y1 + 128 * x2), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp3 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp1 = 0.5
tmp2 = tmp0 * tmp1
tmp4 = tmp3 * tmp1
tmp6 = tmp5 * tmp1
tmp7 = triton_helpers.maximum(tmp4, tmp6)
tmp9 = tmp8 * tmp1
tmp10 = triton_helpers.maximum(tmp7, tmp9)
tmp12 = tmp11 * tmp1
tmp13 = triton_helpers.maximum(tmp10, tmp12)
tmp14 = tmp2 - tmp13
tmp15 = tl_math.exp(tmp14)
tl.store(out_ptr0 + x2, tmp15, xmask)
@triton.jit
def triton_poi_fused__softmax_2(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
x2 = xindex
y3 = yindex
y1 = yindex // 4
tmp0 = tl.load(in_ptr0 + (y3 + 16 * x2), xmask & ymask)
tmp1 = tl.load(in_ptr0 + (4 * y1 + 16 * x2), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * y1 + 16 * x2), xmask & ymask,
eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * y1 + 16 * x2), xmask & ymask,
eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * y1 + 16 * x2), xmask & ymask,
eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2 + 16 * y3), tmp8, xmask & ymask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x2 = xindex // 16 % 4
x3 = xindex // 64
x4 = xindex
tmp0 = tl.load(in_ptr0 + (16 + x0 + 4 * x2 + 32 * x3 + 128 * x1), xmask)
tl.store(out_ptr0 + x4, tmp0, 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 % 16
x2 = xindex // 64
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2 + 16 * x1), xmask)
tl.store(out_ptr0 + x3, tmp0, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp11 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp12 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp9 = tmp7 + tmp8
tmp10 = tmp6 + tmp9
tmp13 = tmp11 + tmp12
tmp14 = tmp10 + tmp13
tmp15 = 4.0
tmp16 = tmp14 / tmp15
tmp17 = tmp2 - tmp16
tmp18 = tmp17 * tmp17
tmp19 = tmp5 - tmp16
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp9 - tmp16
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp13 - tmp16
tmp26 = tmp25 * tmp25
tmp27 = tmp24 + tmp26
tmp28 = tmp27 / tmp15
tl.store(out_ptr0 + x0, tmp16, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_add_native_layer_norm_6(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr5 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = 1e-05
tmp7 = tmp5 + tmp6
tmp8 = libdevice.rsqrt(tmp7)
tmp9 = tmp4 * tmp8
tmp11 = tmp9 * tmp10
tmp13 = tmp11 + tmp12
tl.store(out_ptr0 + x2, tmp13, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (16, 4), (4, 1))
assert_size_stride(primals_3, (32, 4), (4, 1))
assert_size_stride(primals_4, (4, 16), (16, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 16), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((16, 32), (32, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 32), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((4, 4, 4, 4, 1), (64, 16, 4, 1, 1), torch
.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64, 4)](buf1, buf2, 64, 4, XBLOCK=4,
YBLOCK=32, num_warps=4, num_stages=1)
buf3 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf0, (16, 4, 4), (4, 64, 1),
0), reinterpret_tensor(buf2, (16, 4, 4), (16, 4, 1), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4, 4, 4), (4, 1, 64, 16), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf3, buf4, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf5 = reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf3
triton_poi_fused__softmax_2[grid(16, 16)](buf4, buf5, 16, 16,
XBLOCK=16, YBLOCK=16, num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 4, 4, 4, 1), (64, 16, 4, 1, 1), 0)
del buf4
triton_poi_fused_clone_3[grid(256)](buf1, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf1
buf7 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf5, (16, 4, 4), (1, 64, 16),
0), reinterpret_tensor(buf6, (16, 4, 4), (16, 4, 1), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_clone_4[grid(256)](buf7, buf8, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf7
buf9 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf8, (16, 16), (16, 1), 0),
reinterpret_tensor(primals_4, (16, 4), (1, 16), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
buf11 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](primals_1, buf9,
buf10, buf11, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf12 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_6[grid(64)](primals_1, buf9,
buf10, buf11, primals_5, primals_6, buf12, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf10
del buf11
del primals_6
return buf12, primals_1, primals_5, buf5, reinterpret_tensor(buf8, (16,
16), (16, 1), 0), buf9, primals_4, reinterpret_tensor(buf6, (16, 4,
4), (16, 1, 4), 0), reinterpret_tensor(buf0, (16, 4, 4), (4, 1, 64), 0
), reinterpret_tensor(buf2, (16, 4, 4), (16, 1, 4), 0)
class MultiHeadAttnNew(nn.Module):
def __init__(self, n_head, d_model, d_head, dropout, dropatt=0,
pre_lnorm=False):
super(MultiHeadAttnNew, self).__init__()
self.n_head = n_head
self.d_model = d_model
self.d_head = d_head
self.dropout = dropout
self.q_net = nn.Linear(d_model, n_head * d_head, bias=False)
self.kv_net = nn.Linear(d_model, 2 * n_head * d_head, bias=False)
self.drop = nn.Dropout(dropout)
self.dropatt = nn.Dropout(dropatt)
self.o_net = nn.Linear(n_head * d_head, d_model, bias=False)
self.layer_norm = nn.LayerNorm(d_model)
self.scale = 1 / d_head ** 0.5
self.pre_lnorm = pre_lnorm
def forward(self, input_0):
primals_2 = self.q_net.weight
primals_3 = self.kv_net.weight
primals_4 = self.o_net.weight
primals_5 = self.layer_norm.weight
primals_6 = self.layer_norm.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
|
JasonBenn/duet
|
MultiHeadAttn
| false
| 8,353
|
[
"Apache-2.0"
] | 11
|
0d6f1f66fad097023b022f2a361a1587d0f740ba
|
https://github.com/JasonBenn/duet/tree/0d6f1f66fad097023b022f2a361a1587d0f740ba
|
PositionalWiseFeedForward
|
# 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/wx/cwxfsgfgbdsczjgwdefn66lhdznwlofx4nu63vcfcyg45oomz6zu.py
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv1d => convolution
# Graph fragment:
# %convolution : [num_users=4] = call_function[target=torch.ops.aten.convolution.default](args = (%permute, %primals_2, %primals_3, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048, 512], 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_convolution_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_convolution_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 2048
xnumel = 512
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 512
y1 = (yindex // 512)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (512*x2) + (262144*y1)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (512*y3)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/va/cvaocwkuykjfxf7ngodp4om3hlrge6v3r4h6f6go7isdlo7nnxs7.py
# Topologically Sorted Source Nodes: [conv1d, mul, pow_1, mul_1, add, mul_2, tanh, add_1, mul_3], Original ATen: [aten.convolution, aten.mul, aten.pow, aten.add, aten.tanh]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# conv1d => convolution
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# mul_3 => mul_3
# pow_1 => pow_1
# tanh => tanh
# Graph fragment:
# %convolution : [num_users=4] = call_function[target=torch.ops.aten.convolution.default](args = (%permute, %primals_2, %primals_3, [1], [0], [1], False, [0], 1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, 0.5), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%convolution, 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 = (%convolution, %mul_1), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, 0.7978845608028654), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%mul_2,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%tanh, 1), kwargs = {})
# %mul_3 : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, %add_1), kwargs = {})
triton_poi_fused_add_convolution_mul_pow_tanh_1 = async_compile.triton('triton_poi_fused_add_convolution_mul_pow_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4194304],
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_convolution_mul_pow_tanh_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_convolution_mul_pow_tanh_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4194304
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 512) % 2048
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.5
tmp4 = tmp2 * tmp3
tmp5 = tmp2 * tmp2
tmp6 = tmp5 * tmp2
tmp7 = 0.044715
tmp8 = tmp6 * tmp7
tmp9 = tmp2 + tmp8
tmp10 = 0.7978845608028654
tmp11 = tmp9 * tmp10
tmp12 = libdevice.tanh(tmp11)
tmp13 = 1.0
tmp14 = tmp12 + tmp13
tmp15 = tmp4 * tmp14
tl.store(in_out_ptr0 + (x3), tmp2, None)
tl.store(out_ptr0 + (x3), tmp15, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/7m/c7mr3nxsdcjlcp7bsx2ryn2k52bavp5ypqkbc6u2zuwnmuh2ehop.py
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# output_1 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%mul_3, %primals_4, %primals_5, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_2 = async_compile.triton('triton_poi_fused_convolution_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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 = 1048576
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = (xindex // 512) % 512
tmp0 = tl.load(in_out_ptr0 + (x3), None)
tmp1 = tl.load(in_ptr0 + (x1), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ta/ctayft4emgpqxi444nzfv6hwbh2mc4hdmoxhfmq5ufyxz6raxfqo.py
# Topologically Sorted Source Nodes: [add_2, output_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add_2 => add_2
# output_3 => var_mean
# Graph fragment:
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %permute_1), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_2, [2]), kwargs = {correction: 0, keepdim: True})
triton_red_fused_add_native_layer_norm_3 = async_compile.triton('triton_red_fused_add_native_layer_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.reduction(
size_hints=[8192, 128],
reduction_hint=ReductionHint.OUTER,
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, 5, 6), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_red_fused_add_native_layer_norm_3', '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_red_fused_add_native_layer_norm_3(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
xnumel = 8192
rnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rbase = tl.arange(0, RBLOCK)[None, :]
x4 = xindex
x0 = xindex % 4
x1 = (xindex // 4) % 512
x2 = (xindex // 2048)
tmp4_mean = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_m2 = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_weight = tl.zeros([XBLOCK, RBLOCK], tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r3 = rindex
tmp0 = tl.load(in_ptr0 + (r3 + (128*x4)), rmask, eviction_policy='evict_first', other=0.0)
tmp1 = tl.load(in_ptr1 + (x1 + (512*r3) + (65536*x0) + (262144*x2)), rmask, eviction_policy='evict_last', other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp4_mean_next, tmp4_m2_next, tmp4_weight_next = triton_helpers.welford_reduce(
tmp3, tmp4_mean, tmp4_m2, tmp4_weight, roffset == 0
)
tmp4_mean = tl.where(rmask, tmp4_mean_next, tmp4_mean)
tmp4_m2 = tl.where(rmask, tmp4_m2_next, tmp4_m2)
tmp4_weight = tl.where(rmask, tmp4_weight_next, tmp4_weight)
tmp4_tmp, tmp5_tmp, tmp6_tmp = triton_helpers.welford(
tmp4_mean, tmp4_m2, tmp4_weight, 1
)
tmp4 = tmp4_tmp[:, None]
tmp5 = tmp5_tmp[:, None]
tmp6 = tmp6_tmp[:, None]
tl.store(out_ptr0 + (x4), tmp4, None)
tl.store(out_ptr1 + (x4), tmp5, None)
tl.store(out_ptr2 + (x4), tmp6, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/sd/csdf46zwe54qx66x3edbjjw7hq4dfpbrhphpzzp46ptloxaljs2s.py
# Topologically Sorted Source Nodes: [add_2, output_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add_2 => add_2
# output_3 => add_3, rsqrt, var_mean
# Graph fragment:
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %permute_1), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_2, [2]), kwargs = {correction: 0, keepdim: True})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-05), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_3,), kwargs = {})
triton_per_fused_add_native_layer_norm_4 = async_compile.triton('triton_per_fused_add_native_layer_norm_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[2048, 4],
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, 5), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_native_layer_norm_4', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 3, '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_native_layer_norm_4(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 2048
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)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (4*x0)), None)
tmp1 = tl.load(in_ptr1 + (r1 + (4*x0)), None)
tmp2 = tl.load(in_ptr2 + (r1 + (4*x0)), None)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp5 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp7, tmp8, tmp9 = triton_helpers.welford(tmp3, tmp4, tmp5, 1)
tmp10 = tmp7[:, None]
tmp11 = tmp8[:, None]
tmp12 = tmp9[:, None]
tmp13 = 512.0
tmp14 = tmp11 / tmp13
tmp15 = 1e-05
tmp16 = tmp14 + tmp15
tmp17 = libdevice.rsqrt(tmp16)
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp17, None)
tl.store(out_ptr0 + (x0), tmp10, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/yr/cyruwyid4b75e7lq7sldqmuslk636bzi4r7yuwue3atsnf7zebj6.py
# Topologically Sorted Source Nodes: [add_2, output_3], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# add_2 => add_2
# output_3 => add_4, mul_4, mul_5, sub
# Graph fragment:
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %permute_1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %getitem_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %rsqrt), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_4, %primals_6), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_5, %primals_7), kwargs = {})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048, 512], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32', 8: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, 8), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_native_layer_norm_5', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 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_5(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 2048
xnumel = 512
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = (yindex // 512)
tmp0 = tl.load(in_ptr0 + (x2 + (512*y3)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0 + (512*x2) + (262144*y1)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y3), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (y3), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x2), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr5 + (x2), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 * tmp7
tmp10 = tmp8 + tmp9
tl.store(out_ptr0 + (x2 + (512*y3)), tmp10, 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, 512, 512), (262144, 512, 1))
assert_size_stride(primals_2, (2048, 512, 1), (512, 1, 1))
assert_size_stride(primals_3, (2048, ), (1, ))
assert_size_stride(primals_4, (512, 2048, 1), (2048, 1, 1))
assert_size_stride(primals_5, (512, ), (1, ))
assert_size_stride(primals_6, (512, ), (1, ))
assert_size_stride(primals_7, (512, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 512, 512), (262144, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(primals_1, buf0, 2048, 512, grid=grid(2048, 512), stream=stream0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf1, (4, 2048, 512), (1048576, 512, 1))
buf2 = buf1; del buf1 # reuse
buf3 = empty_strided_cuda((4, 2048, 512), (1048576, 512, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv1d, mul, pow_1, mul_1, add, mul_2, tanh, add_1, mul_3], Original ATen: [aten.convolution, aten.mul, aten.pow, aten.add, aten.tanh]
triton_poi_fused_add_convolution_mul_pow_tanh_1.run(buf2, primals_3, buf3, 4194304, grid=grid(4194304), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf3, primals_4, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf4, (4, 512, 512), (262144, 512, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [output_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf5, primals_5, 1048576, grid=grid(1048576), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 512, 1, 4), (2048, 4, 8192, 1), torch.float32)
buf7 = empty_strided_cuda((4, 512, 1, 4), (2048, 4, 8192, 1), torch.float32)
buf8 = empty_strided_cuda((4, 512, 1, 4), (2048, 4, 8192, 1), torch.float32)
# Topologically Sorted Source Nodes: [add_2, output_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_red_fused_add_native_layer_norm_3.run(primals_1, buf5, buf6, buf7, buf8, 8192, 128, grid=grid(8192), stream=stream0)
buf9 = empty_strided_cuda((4, 512, 1), (512, 1, 1), torch.float32)
buf10 = empty_strided_cuda((4, 512, 1), (512, 1, 2048), torch.float32)
buf12 = reinterpret_tensor(buf10, (4, 512, 1), (512, 1, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [add_2, output_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_per_fused_add_native_layer_norm_4.run(buf12, buf6, buf7, buf8, buf9, 2048, 4, grid=grid(2048), stream=stream0)
del buf6
del buf7
del buf8
buf13 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [add_2, output_3], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(primals_1, buf5, buf9, buf12, primals_6, primals_7, buf13, 2048, 512, grid=grid(2048, 512), stream=stream0)
del primals_7
return (buf13, primals_1, primals_2, primals_4, primals_6, buf2, buf3, buf5, buf9, buf12, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 512, 512), (262144, 512, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((2048, 512, 1), (512, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((2048, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((512, 2048, 1), (2048, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((512, ), (1, ), device='cuda:0', dtype=torch.float32)
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 math
import torch
import torch.nn as nn
class GELU(nn.Module):
"""
This is a smoother version of the RELU.
Original paper: https://arxiv.org/abs/1606.08415
"""
def forward(self, x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x +
0.044715 * torch.pow(x, 3))))
class PositionalWiseFeedForward(nn.Module):
def __init__(self, model_dim=512, ffn_dim=2048, dropout=0.0):
super(PositionalWiseFeedForward, self).__init__()
self.w1 = nn.Conv1d(model_dim, ffn_dim, 1)
self.w2 = nn.Conv1d(ffn_dim, model_dim, 1)
self.dropout = nn.Dropout(dropout)
self.layer_norm = nn.LayerNorm(model_dim)
self.gelu = GELU()
def forward(self, x):
"""
:param x: [b, t, d*h]
:return:
"""
output = x.transpose(1, 2)
output = self.w2(self.gelu(self.w1(output)))
output = self.dropout(output.transpose(1, 2))
output = self.layer_norm(x + output)
return output
def get_inputs():
return [torch.rand([4, 512, 512])]
def get_init_inputs():
return [[], {}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import 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_convolution_0(in_ptr0, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
xnumel = 512
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 512
y1 = yindex // 512
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 512 * x2 + 262144 * y1), xmask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 512 * y3), tmp0, xmask)
@triton.jit
def triton_poi_fused_add_convolution_mul_pow_tanh_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 512 % 2048
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.5
tmp4 = tmp2 * tmp3
tmp5 = tmp2 * tmp2
tmp6 = tmp5 * tmp2
tmp7 = 0.044715
tmp8 = tmp6 * tmp7
tmp9 = tmp2 + tmp8
tmp10 = 0.7978845608028654
tmp11 = tmp9 * tmp10
tmp12 = libdevice.tanh(tmp11)
tmp13 = 1.0
tmp14 = tmp12 + tmp13
tmp15 = tmp4 * tmp14
tl.store(in_out_ptr0 + x3, tmp2, None)
tl.store(out_ptr0 + x3, tmp15, None)
@triton.jit
def triton_poi_fused_convolution_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x3 = xindex
x1 = xindex // 512 % 512
tmp0 = tl.load(in_out_ptr0 + x3, None)
tmp1 = tl.load(in_ptr0 + x1, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, None)
@triton.jit
def triton_red_fused_add_native_layer_norm_3(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, out_ptr2, xnumel, rnumel, XBLOCK: tl.constexpr, RBLOCK: tl.
constexpr):
rnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rbase = tl.arange(0, RBLOCK)[None, :]
x4 = xindex
x0 = xindex % 4
x1 = xindex // 4 % 512
x2 = xindex // 2048
tmp4_mean = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_m2 = tl.zeros([XBLOCK, RBLOCK], tl.float32)
tmp4_weight = tl.zeros([XBLOCK, RBLOCK], tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r3 = rindex
tmp0 = tl.load(in_ptr0 + (r3 + 128 * x4), rmask, eviction_policy=
'evict_first', other=0.0)
tmp1 = tl.load(in_ptr1 + (x1 + 512 * r3 + 65536 * x0 + 262144 * x2),
rmask, eviction_policy='evict_last', other=0.0)
tmp2 = tmp0 + tmp1
tmp3 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp4_mean_next, tmp4_m2_next, tmp4_weight_next = (triton_helpers.
welford_reduce(tmp3, tmp4_mean, tmp4_m2, tmp4_weight, roffset == 0)
)
tmp4_mean = tl.where(rmask, tmp4_mean_next, tmp4_mean)
tmp4_m2 = tl.where(rmask, tmp4_m2_next, tmp4_m2)
tmp4_weight = tl.where(rmask, tmp4_weight_next, tmp4_weight)
tmp4_tmp, tmp5_tmp, tmp6_tmp = triton_helpers.welford(tmp4_mean,
tmp4_m2, tmp4_weight, 1)
tmp4 = tmp4_tmp[:, None]
tmp5 = tmp5_tmp[:, None]
tmp6 = tmp6_tmp[:, None]
tl.store(out_ptr0 + x4, tmp4, None)
tl.store(out_ptr1 + x4, tmp5, None)
tl.store(out_ptr2 + x4, tmp6, None)
@triton.jit
def triton_per_fused_add_native_layer_norm_4(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, out_ptr0, xnumel, rnumel, XBLOCK: tl.constexpr):
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = 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)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 4 * x0), None)
tmp1 = tl.load(in_ptr1 + (r1 + 4 * x0), None)
tmp2 = tl.load(in_ptr2 + (r1 + 4 * x0), None)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp5 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp7, tmp8, tmp9 = triton_helpers.welford(tmp3, tmp4, tmp5, 1)
tmp10 = tmp7[:, None]
tmp11 = tmp8[:, None]
tmp9[:, None]
tmp13 = 512.0
tmp14 = tmp11 / tmp13
tmp15 = 1e-05
tmp16 = tmp14 + tmp15
tmp17 = libdevice.rsqrt(tmp16)
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp17, None)
tl.store(out_ptr0 + x0, tmp10, None)
@triton.jit
def triton_poi_fused_add_native_layer_norm_5(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 512
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 512
y1 = yindex // 512
tmp0 = tl.load(in_ptr0 + (x2 + 512 * y3), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + (y0 + 512 * x2 + 262144 * y1), xmask,
eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + y3, None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + y3, None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x2, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr5 + x2, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 - tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 * tmp7
tmp10 = tmp8 + tmp9
tl.store(out_ptr0 + (x2 + 512 * y3), tmp10, 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, 512, 512), (262144, 512, 1))
assert_size_stride(primals_2, (2048, 512, 1), (512, 1, 1))
assert_size_stride(primals_3, (2048,), (1,))
assert_size_stride(primals_4, (512, 2048, 1), (2048, 1, 1))
assert_size_stride(primals_5, (512,), (1,))
assert_size_stride(primals_6, (512,), (1,))
assert_size_stride(primals_7, (512,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 512, 512), (262144, 512, 1), torch.
float32)
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(2048, 512)](primals_1, buf0,
2048, 512, XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf1, (4, 2048, 512), (1048576, 512, 1))
buf2 = buf1
del buf1
buf3 = empty_strided_cuda((4, 2048, 512), (1048576, 512, 1), torch.
float32)
triton_poi_fused_add_convolution_mul_pow_tanh_1[grid(4194304)](buf2,
primals_3, buf3, 4194304, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_3
buf4 = extern_kernels.convolution(buf3, primals_4, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf4, (4, 512, 512), (262144, 512, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_2[grid(1048576)](buf5, primals_5,
1048576, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_5
buf6 = empty_strided_cuda((4, 512, 1, 4), (2048, 4, 8192, 1), torch
.float32)
buf7 = empty_strided_cuda((4, 512, 1, 4), (2048, 4, 8192, 1), torch
.float32)
buf8 = empty_strided_cuda((4, 512, 1, 4), (2048, 4, 8192, 1), torch
.float32)
triton_red_fused_add_native_layer_norm_3[grid(8192)](primals_1,
buf5, buf6, buf7, buf8, 8192, 128, XBLOCK=64, RBLOCK=8,
num_warps=4, num_stages=1)
buf9 = empty_strided_cuda((4, 512, 1), (512, 1, 1), torch.float32)
buf10 = empty_strided_cuda((4, 512, 1), (512, 1, 2048), torch.float32)
buf12 = reinterpret_tensor(buf10, (4, 512, 1), (512, 1, 1), 0)
del buf10
triton_per_fused_add_native_layer_norm_4[grid(2048)](buf12, buf6,
buf7, buf8, buf9, 2048, 4, XBLOCK=32, num_warps=2, num_stages=1)
del buf6
del buf7
del buf8
buf13 = buf0
del buf0
triton_poi_fused_add_native_layer_norm_5[grid(2048, 512)](primals_1,
buf5, buf9, buf12, primals_6, primals_7, buf13, 2048, 512,
XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
del primals_7
return (buf13, primals_1, primals_2, primals_4, primals_6, buf2, buf3,
buf5, buf9, buf12)
class GELU(nn.Module):
"""
This is a smoother version of the RELU.
Original paper: https://arxiv.org/abs/1606.08415
"""
def forward(self, x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x +
0.044715 * torch.pow(x, 3))))
class PositionalWiseFeedForwardNew(nn.Module):
def __init__(self, model_dim=512, ffn_dim=2048, dropout=0.0):
super(PositionalWiseFeedForwardNew, self).__init__()
self.w1 = nn.Conv1d(model_dim, ffn_dim, 1)
self.w2 = nn.Conv1d(ffn_dim, model_dim, 1)
self.dropout = nn.Dropout(dropout)
self.layer_norm = nn.LayerNorm(model_dim)
self.gelu = GELU()
def forward(self, input_0):
primals_2 = self.w1.weight
primals_3 = self.w1.bias
primals_4 = self.w2.weight
primals_5 = self.w2.bias
primals_6 = self.layer_norm.weight
primals_7 = self.layer_norm.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
|
JiaweiSheng/FAAN
|
PositionalWiseFeedForward
| false
| 8,354
|
[
"MIT"
] | 41
|
b439b829506c4e2e9044a6b2ab7f3d844f445a95
|
https://github.com/JiaweiSheng/FAAN/tree/b439b829506c4e2e9044a6b2ab7f3d844f445a95
|
SelfAttention
|
# 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/sv/csvq3ckvp44ay4sw4dwiwogcyjauztqawdlj4czrg4o64nxotgqb.py
# Topologically Sorted Source Nodes: [tanh], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# tanh => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%mm,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_tanh_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_tanh_0(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 = libdevice.tanh(tmp0)
tl.store(in_out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/oo/coobrm7cytaeij2eyoctjbuz6lnjrauzvpxipvlrrepxba5rhlgq.py
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => 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), (16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 1), (1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mm], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), primals_2, out=buf0)
del primals_2
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [tanh], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, 64, grid=grid(64), stream=stream0)
buf2 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [mm_1], Original ATen: [aten.mm]
extern_kernels.mm(buf1, primals_3, out=buf2)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf2, buf3, 16, grid=grid(16), stream=stream0)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf3, buf4, 16, grid=grid(16), stream=stream0)
buf5 = reinterpret_tensor(buf3, (4, 1, 4), (4, 4, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [bmm], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf4, (4, 1, 4), (4, 0, 1), 0), primals_1, out=buf5)
del buf4
return (reinterpret_tensor(buf5, (4, 4), (4, 1), 0), primals_1, buf1, buf2, reinterpret_tensor(primals_3, (1, 4), (1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (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), (1, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import nn
import torch.nn.modules.loss
from scipy.sparse import *
class SelfAttention(nn.Module):
def __init__(self, input_size, hidden_size):
super(SelfAttention, self).__init__()
self.W1 = torch.Tensor(input_size, hidden_size)
self.W1 = nn.Parameter(nn.init.xavier_uniform_(self.W1))
self.W2 = torch.Tensor(hidden_size, 1)
self.W2 = nn.Parameter(nn.init.xavier_uniform_(self.W2))
def forward(self, x, attention_mask=None):
attention = torch.mm(torch.tanh(torch.mm(x.view(-1, x.size(-1)),
self.W1)), self.W2).view(x.size(0), -1)
if attention_mask is not None:
attention = attention.masked_fill_(1 - attention_mask.byte(), -INF)
probs = torch.softmax(attention, dim=-1).unsqueeze(1)
weighted_x = torch.bmm(probs, x).squeeze(1)
return weighted_x
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4, 'hidden_size': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch import nn
import torch.nn.modules.loss
from scipy.sparse import *
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, 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 = libdevice.tanh(tmp0)
tl.store(in_out_ptr0 + x0, tmp1, 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), (16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 1), (1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
primals_2, out=buf0)
del primals_2
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(64)](buf1, 64, XBLOCK=64, num_warps=1,
num_stages=1)
buf2 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf1, primals_3, out=buf2)
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 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(16)](buf3, buf4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf5 = reinterpret_tensor(buf3, (4, 1, 4), (4, 4, 1), 0)
del buf3
extern_kernels.bmm(reinterpret_tensor(buf4, (4, 1, 4), (4, 0, 1), 0
), primals_1, out=buf5)
del buf4
return reinterpret_tensor(buf5, (4, 4), (4, 1), 0
), primals_1, buf1, buf2, reinterpret_tensor(primals_3, (1, 4), (1,
1), 0)
class SelfAttentionNew(nn.Module):
def __init__(self, input_size, hidden_size):
super(SelfAttentionNew, self).__init__()
self.W1 = torch.Tensor(input_size, hidden_size)
self.W1 = nn.Parameter(nn.init.xavier_uniform_(self.W1))
self.W2 = torch.Tensor(hidden_size, 1)
self.W2 = nn.Parameter(nn.init.xavier_uniform_(self.W2))
def forward(self, input_0):
primals_2 = self.W1
primals_3 = self.W2
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
IBM/graph4nlp
|
SelfAttention
| false
| 8,355
|
[
"Apache-2.0"
] | 18
|
a9bf20b23fa1ec368d9bd40cc8c557f86a9f8297
|
https://github.com/IBM/graph4nlp/tree/a9bf20b23fa1ec368d9bd40cc8c557f86a9f8297
|
ScaledDotProductAttention
|
# 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/wz/cwzlgmghy6nxuchbiog4puo46i4tq7yhd3qu6ftkgjf3gwib6hxn.py
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [2], 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_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: [attn_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_1 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [2], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_1 = async_compile.triton('triton_poi_fused__softmax_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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')
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), (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: [attn], Original ATen: [aten.bmm]
extern_kernels.bmm(arg1_1, reinterpret_tensor(arg0_1, (4, 4, 4), (16, 1, 4), 0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__softmax_0.run(buf0, buf1, 64, grid=grid(64), stream=stream0)
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [attn_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.bmm]
extern_kernels.bmm(buf2, arg2_1, out=buf3)
del arg2_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), (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 numpy as np
import torch.nn as nn
class ScaledDotProductAttention(nn.Module):
""" Scaled Dot-Product Attention
"""
def __init__(self, attn_dropout=0.0):
super(ScaledDotProductAttention, self).__init__()
self.dropout = nn.Dropout(attn_dropout)
self.softmax = nn.Softmax(dim=2)
def forward(self, q, k, v, scale=None, attn_mask=None):
"""
:param attn_mask: [batch, time]
:param scale:
:param q: [batch, time, dim]
:param k: [batch, time, dim]
:param v: [batch, time, dim]
:return:
"""
attn = torch.bmm(q, k.transpose(1, 2))
if scale:
attn = attn * scale
if attn_mask:
attn = attn.masked_fill_(attn_mask, -np.inf)
attn = self.softmax(attn)
attn = self.dropout(attn)
output = torch.bmm(attn, v)
return output, attn
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4, 4, 4])
]
def get_init_inputs():
return [[], {}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime 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__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)
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), (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(arg1_1, reinterpret_tensor(arg0_1, (4, 4, 4), (
16, 1, 4), 0), out=buf0)
del arg0_1
del arg1_1
buf1 = empty_strided_cuda((4, 4, 4), (16, 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 = buf0
del buf0
triton_poi_fused__softmax_1[grid(64)](buf1, buf2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf3 = buf1
del buf1
extern_kernels.bmm(buf2, arg2_1, out=buf3)
del arg2_1
return buf3, buf2
class ScaledDotProductAttentionNew(nn.Module):
""" Scaled Dot-Product Attention
"""
def __init__(self, attn_dropout=0.0):
super(ScaledDotProductAttentionNew, self).__init__()
self.dropout = nn.Dropout(attn_dropout)
self.softmax = nn.Softmax(dim=2)
def forward(self, input_0, input_1, input_2):
arg0_1 = input_0
arg1_1 = input_1
arg2_1 = input_2
output = call([arg0_1, arg1_1, arg2_1])
return output[0], output[1]
|
JiaweiSheng/FAAN
|
ScaledDotProductAttention
| false
| 8,356
|
[
"MIT"
] | 41
|
b439b829506c4e2e9044a6b2ab7f3d844f445a95
|
https://github.com/JiaweiSheng/FAAN/tree/b439b829506c4e2e9044a6b2ab7f3d844f445a95
|
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/ng/cngjwaj32ulse46l45qzlmxoqdt4xu62eznv6tfk7uhocabugjkd.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 8192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x2), tmp4, None)
tl.store(out_ptr0 + (x2), tmp6, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (128, 4), (4, 1))
assert_size_stride(primals_2, (128, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (128, 128), (128, 1))
assert_size_stride(primals_5, (128, ), (1, ))
assert_size_stride(primals_6, (4, 128), (128, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 128), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 128), (2048, 512, 128, 1), 0); del buf0 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf6, 8192, grid=grid(8192), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 128), (128, 1), 0), reinterpret_tensor(primals_4, (128, 128), (1, 128), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 128), (2048, 512, 128, 1), 0); del buf2 # reuse
buf5 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf5, 8192, grid=grid(8192), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 128), (128, 1), 0), reinterpret_tensor(primals_6, (128, 4), (1, 128), 0), alpha=1, beta=1, out=buf4)
del primals_7
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 128), (128, 1), 0), reinterpret_tensor(buf3, (64, 128), (128, 1), 0), primals_6, buf5, primals_4, buf6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((128, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((128, 128), (128, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((128, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 128), (128, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
class MLP(nn.Module):
def __init__(self, input_dim, output_dim, hidden_dim=128):
""" 初始化q网络，为全连接网络
input_dim: 输入的特征数即环境的状态维度
output_dim: 输出的动作维度
"""
super(MLP, self).__init__()
self.fc1 = nn.Linear(input_dim, hidden_dim)
self.fc2 = nn.Linear(hidden_dim, hidden_dim)
self.fc3 = nn.Linear(hidden_dim, output_dim)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
return self.fc3(x)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'output_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, None)
tl.store(out_ptr0 + x2, tmp6, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (128, 4), (4, 1))
assert_size_stride(primals_2, (128,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (128, 128), (128, 1))
assert_size_stride(primals_5, (128,), (1,))
assert_size_stride(primals_6, (4, 128), (128, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 128), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 128), (2048, 512, 128, 1), 0)
del buf0
buf6 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1),
torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(8192)](buf1,
primals_2, buf6, 8192, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 128), (128, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 128), (128, 1), 0),
reinterpret_tensor(primals_4, (128, 128), (1, 128), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 128), (2048, 512, 128, 1), 0)
del buf2
buf5 = empty_strided_cuda((4, 4, 4, 128), (2048, 512, 128, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(8192)](buf3,
primals_5, buf5, 8192, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 128),
(128, 1), 0), reinterpret_tensor(primals_6, (128, 4), (1, 128),
0), alpha=1, beta=1, out=buf4)
del primals_7
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 128), (128, 1), 0
), reinterpret_tensor(buf3, (64, 128), (128, 1), 0
), primals_6, buf5, primals_4, buf6
class MLPNew(nn.Module):
def __init__(self, input_dim, output_dim, hidden_dim=128):
""" 初始化q网络，为全连接网络
input_dim: 输入的特征数即环境的状态维度
output_dim: 输出的动作维度
"""
super(MLPNew, self).__init__()
self.fc1 = nn.Linear(input_dim, hidden_dim)
self.fc2 = nn.Linear(hidden_dim, hidden_dim)
self.fc3 = nn.Linear(hidden_dim, output_dim)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
|
JohnJim0816/rl-tutorials
|
MLP
| false
| 8,357
|
[
"MIT"
] | 16
|
e99daea815da85f9f25dff2d01b030249a203d22
|
https://github.com/JohnJim0816/rl-tutorials/tree/e99daea815da85f9f25dff2d01b030249a203d22
|
FixupBasicBlock
|
# 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/lw/clwmhv3ard4vibrwx3znjaja7nsm4gxiuq5s4gjf3wv47kclvepv.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.add]
# Source node to ATen node mapping:
# x => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %primals_2), kwargs = {})
triton_poi_fused_add_0 = async_compile.triton('triton_poi_fused_add_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_add_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
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/fx/cfxyl7fil5puh5xllnvnypodhp2566clml7xolta3dccawk6glj2.py
# Topologically Sorted Source Nodes: [x_1, add_1, out, x_2], Original ATen: [aten.convolution, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add_1 => add_1
# out => relu
# x_1 => convolution
# x_2 => add_2
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%add, %primals_3, %primals_4, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution, %primals_5), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_1,), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%relu, %primals_6), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_convolution_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_add_convolution_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*i1', 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_convolution_relu_threshold_backward_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_add_convolution_relu_threshold_backward_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp8 = tl.load(in_ptr3 + (0))
tmp9 = tl.broadcast_to(tmp8, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp5 = tmp2 + tmp4
tmp6 = tl.full([1], 0, tl.int32)
tmp7 = triton_helpers.maximum(tmp6, tmp5)
tmp10 = tmp7 + tmp9
tmp11 = 0.0
tmp12 = tmp7 <= tmp11
tl.store(out_ptr0 + (x3), tmp10, xmask)
tl.store(out_ptr1 + (x3), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ki/ckicbc5amfeyao6o6reux5pjkp6hdqwsvkwip4tonvuecjc3pmll.py
# Topologically Sorted Source Nodes: [x_3, mul, out_1, out_2, out_3], Original ATen: [aten.convolution, aten.mul, aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# mul => mul
# out_1 => add_3
# out_2 => add_4
# out_3 => relu_1
# x_3 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%add_2, %primals_7, %primals_8, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution_1, %primals_9), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_10), kwargs = {})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_3, %primals_1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_4,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu_1, 0), kwargs = {})
triton_poi_fused_add_convolution_mul_relu_threshold_backward_2 = async_compile.triton('triton_poi_fused_add_convolution_mul_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: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*i1', 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_convolution_mul_relu_threshold_backward_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_add_convolution_mul_relu_threshold_backward_2(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (0))
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp6 = tl.load(in_ptr2 + (0))
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp9 = tl.load(in_ptr3 + (x3), xmask)
tmp2 = tmp0 + tmp1
tmp5 = tmp2 * tmp4
tmp8 = tmp5 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.full([1], 0, tl.int32)
tmp12 = triton_helpers.maximum(tmp11, tmp10)
tmp13 = 0.0
tmp14 = tmp12 <= tmp13
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
tl.store(out_ptr0 + (x3), tmp12, xmask)
tl.store(out_ptr1 + (x3), 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, (1, ), (1, ))
assert_size_stride(primals_3, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (1, ), (1, ))
assert_size_stride(primals_6, (1, ), (1, ))
assert_size_stride(primals_7, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (1, ), (1, ))
assert_size_stride(primals_10, (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: [x], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(primals_1, primals_2, buf0, 256, grid=grid(256), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_3, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 4, 4), (64, 16, 4, 1))
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1, add_1, out, x_2], Original ATen: [aten.convolution, aten.add, aten.relu, aten.threshold_backward]
triton_poi_fused_add_convolution_relu_threshold_backward_1.run(buf1, primals_4, primals_5, primals_6, buf2, buf7, 256, grid=grid(256), stream=stream0)
del primals_4
del primals_5
del primals_6
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf3 = extern_kernels.convolution(buf2, primals_7, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 4, 4), (64, 16, 4, 1))
buf4 = buf3; del buf3 # reuse
buf5 = buf1; del buf1 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_3, mul, out_1, out_2, out_3], Original ATen: [aten.convolution, aten.mul, aten.add, aten.relu, aten.threshold_backward]
triton_poi_fused_add_convolution_mul_relu_threshold_backward_2.run(buf4, primals_8, primals_9, primals_10, primals_1, buf5, buf6, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_10
del primals_8
return (buf5, primals_3, primals_7, primals_9, buf0, buf2, buf4, buf6, buf7, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch as th
import torch.utils.data
import torch.nn as nn
def _get_activation(activation):
valid = ['relu', 'leaky_relu', 'lrelu', 'tanh', 'sigmoid']
assert activation in valid, 'activation should be one of {}'.format(valid)
if activation == 'relu':
return nn.ReLU(inplace=True)
if activation == 'leaky_relu' or activation == 'lrelu':
return nn.LeakyReLU(inplace=True)
if activation == 'sigmoid':
return nn.Sigmoid()
if activation == 'tanh':
return nn.Tanh()
return None
def _init_fc_or_conv(fc_conv, activation):
gain = 1.0
if activation is not None:
gain = nn.init.calculate_gain(activation)
nn.init.xavier_uniform_(fc_conv.weight, gain)
if fc_conv.bias is not None:
nn.init.constant_(fc_conv.bias, 0.0)
def _get_norm_layer(norm_layer, channels):
valid = ['instance', 'batch']
assert norm_layer in valid, 'norm_layer should be one of {}'.format(valid)
if norm_layer == 'instance':
layer = nn.InstanceNorm2d(channels, affine=True)
elif norm_layer == 'batch':
layer = nn.BatchNorm2d(channels, affine=True)
nn.init.constant_(layer.bias, 0.0)
nn.init.constant_(layer.weight, 1.0)
return layer
class ConvModule(nn.Module):
"""Basic convolution module with conv + norm(optional) + activation(optional).
Args:
n_in(int): number of input channels.
n_out(int): number of output channels.
ksize(int): size of the convolution kernel (square).
stride(int): downsampling factor
pad(bool): if True, pad the convolutions to maintain a constant size.
padding_mode(str): 'zero' or 'reflection'.
activation(str): nonlinear activation function between convolutions.
norm_layer(str): normalization to apply between the convolution modules.
"""
def __init__(self, n_in, n_out, ksize=3, stride=1, pad=True,
padding_mode='zero', activation=None, norm_layer=None):
super(ConvModule, self).__init__()
assert isinstance(n_in, int
) and n_in > 0, 'Input channels should be a positive integer got {}'.format(
n_in)
assert isinstance(n_out, int
) and n_out > 0, 'Output channels should be a positive integer got {}'.format(
n_out)
assert isinstance(ksize, int
) and ksize > 0, 'Kernel size should be a positive integer got {}'.format(
ksize)
assert padding_mode in ['zero', 'reflection'], 'Invalid padding mode'
padding = (ksize - 1) // 2 if pad else 0
use_bias_in_conv = norm_layer is None
conv_pad = padding
if padding_mode == 'reflection' and padding > 0:
self.add_module('reflection_pad', nn.ReflectionPad2d(padding))
conv_pad = 0
else:
self.add_module('no-op', nn.Identity())
self.add_module('conv', nn.Conv2d(n_in, n_out, ksize, stride=stride,
padding=conv_pad, bias=use_bias_in_conv))
if norm_layer is not None:
self.add_module('norm', _get_norm_layer(norm_layer, n_out))
if activation is not None:
self.add_module('activation', _get_activation(activation))
_init_fc_or_conv(self.conv, activation)
def forward(self, x):
for c in self.children():
x = c(x)
return x
class FixupBasicBlock(nn.Module):
expansion = 1
def __init__(self, n_features, ksize=3, pad=True, padding_mode='zero',
activation='relu'):
super(FixupBasicBlock, self).__init__()
self.bias1a = nn.Parameter(th.zeros(1))
self.conv1 = ConvModule(n_features, n_features, ksize=ksize, stride
=1, pad=pad, activation=None, norm_layer=None, padding_mode=
padding_mode)
self.bias1b = nn.Parameter(th.zeros(1))
self.activation = _get_activation(activation)
self.bias2a = nn.Parameter(th.zeros(1))
self.conv2 = ConvModule(n_features, n_features, ksize=ksize, stride
=1, pad=pad, activation=None, norm_layer=None, padding_mode=
padding_mode)
self.scale = nn.Parameter(th.ones(1))
self.bias2b = nn.Parameter(th.zeros(1))
self.activation2 = _get_activation(activation)
self.ksize = 3
self.pad = pad
def forward(self, x):
identity = x
out = self.conv1(x + self.bias1a)
out = self.activation(out + self.bias1b)
out = self.conv2(out + self.bias2a)
out = out * self.scale + self.bias2b
crop = (self.ksize - 1) // 2 * 2
if crop > 0 and not self.pad:
identity = identity[:, :, crop:-crop, crop:-crop]
out += identity
out = self.activation2(out)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_features': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch as th
import torch.utils.data
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
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_relu_threshold_backward_1(in_ptr0,
in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 4
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp8 = tl.load(in_ptr3 + 0)
tmp9 = tl.broadcast_to(tmp8, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp5 = tmp2 + tmp4
tmp6 = tl.full([1], 0, tl.int32)
tmp7 = triton_helpers.maximum(tmp6, tmp5)
tmp10 = tmp7 + tmp9
tmp11 = 0.0
tmp12 = tmp7 <= tmp11
tl.store(out_ptr0 + x3, tmp10, xmask)
tl.store(out_ptr1 + x3, tmp12, xmask)
@triton.jit
def triton_poi_fused_add_convolution_mul_relu_threshold_backward_2(in_out_ptr0,
in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, out_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 16 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + 0)
tmp4 = tl.broadcast_to(tmp3, [XBLOCK])
tmp6 = tl.load(in_ptr2 + 0)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK])
tmp9 = tl.load(in_ptr3 + x3, xmask)
tmp2 = tmp0 + tmp1
tmp5 = tmp2 * tmp4
tmp8 = tmp5 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = tl.full([1], 0, tl.int32)
tmp12 = triton_helpers.maximum(tmp11, tmp10)
tmp13 = 0.0
tmp14 = tmp12 <= tmp13
tl.store(in_out_ptr0 + x3, tmp2, xmask)
tl.store(out_ptr0 + x3, tmp12, xmask)
tl.store(out_ptr1 + x3, 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, (1,), (1,))
assert_size_stride(primals_3, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (1,), (1,))
assert_size_stride(primals_6, (1,), (1,))
assert_size_stride(primals_7, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (1,), (1,))
assert_size_stride(primals_10, (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_0[grid(256)](primals_1, primals_2, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf1 = extern_kernels.convolution(buf0, primals_3, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 4, 4), (64, 16, 4, 1))
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_add_convolution_relu_threshold_backward_1[grid(256)](
buf1, primals_4, primals_5, primals_6, buf2, buf7, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_4
del primals_5
del primals_6
buf3 = extern_kernels.convolution(buf2, primals_7, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf3, (4, 4, 4, 4), (64, 16, 4, 1))
buf4 = buf3
del buf3
buf5 = buf1
del buf1
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_add_convolution_mul_relu_threshold_backward_2[grid
(256)](buf4, primals_8, primals_9, primals_10, primals_1, buf5,
buf6, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
del primals_10
del primals_8
return buf5, primals_3, primals_7, primals_9, buf0, buf2, buf4, buf6, buf7
def _get_activation(activation):
valid = ['relu', 'leaky_relu', 'lrelu', 'tanh', 'sigmoid']
assert activation in valid, 'activation should be one of {}'.format(valid)
if activation == 'relu':
return nn.ReLU(inplace=True)
if activation == 'leaky_relu' or activation == 'lrelu':
return nn.LeakyReLU(inplace=True)
if activation == 'sigmoid':
return nn.Sigmoid()
if activation == 'tanh':
return nn.Tanh()
return None
def _init_fc_or_conv(fc_conv, activation):
gain = 1.0
if activation is not None:
gain = nn.init.calculate_gain(activation)
nn.init.xavier_uniform_(fc_conv.weight, gain)
if fc_conv.bias is not None:
nn.init.constant_(fc_conv.bias, 0.0)
def _get_norm_layer(norm_layer, channels):
valid = ['instance', 'batch']
assert norm_layer in valid, 'norm_layer should be one of {}'.format(valid)
if norm_layer == 'instance':
layer = nn.InstanceNorm2d(channels, affine=True)
elif norm_layer == 'batch':
layer = nn.BatchNorm2d(channels, affine=True)
nn.init.constant_(layer.bias, 0.0)
nn.init.constant_(layer.weight, 1.0)
return layer
class ConvModule(nn.Module):
"""Basic convolution module with conv + norm(optional) + activation(optional).
Args:
n_in(int): number of input channels.
n_out(int): number of output channels.
ksize(int): size of the convolution kernel (square).
stride(int): downsampling factor
pad(bool): if True, pad the convolutions to maintain a constant size.
padding_mode(str): 'zero' or 'reflection'.
activation(str): nonlinear activation function between convolutions.
norm_layer(str): normalization to apply between the convolution modules.
"""
def __init__(self, n_in, n_out, ksize=3, stride=1, pad=True,
padding_mode='zero', activation=None, norm_layer=None):
super(ConvModule, self).__init__()
assert isinstance(n_in, int
) and n_in > 0, 'Input channels should be a positive integer got {}'.format(
n_in)
assert isinstance(n_out, int
) and n_out > 0, 'Output channels should be a positive integer got {}'.format(
n_out)
assert isinstance(ksize, int
) and ksize > 0, 'Kernel size should be a positive integer got {}'.format(
ksize)
assert padding_mode in ['zero', 'reflection'], 'Invalid padding mode'
padding = (ksize - 1) // 2 if pad else 0
use_bias_in_conv = norm_layer is None
conv_pad = padding
if padding_mode == 'reflection' and padding > 0:
self.add_module('reflection_pad', nn.ReflectionPad2d(padding))
conv_pad = 0
else:
self.add_module('no-op', nn.Identity())
self.add_module('conv', nn.Conv2d(n_in, n_out, ksize, stride=stride,
padding=conv_pad, bias=use_bias_in_conv))
if norm_layer is not None:
self.add_module('norm', _get_norm_layer(norm_layer, n_out))
if activation is not None:
self.add_module('activation', _get_activation(activation))
_init_fc_or_conv(self.conv, activation)
def forward(self, x):
for c in self.children():
x = c(x)
return x
class FixupBasicBlockNew(nn.Module):
expansion = 1
def __init__(self, n_features, ksize=3, pad=True, padding_mode='zero',
activation='relu'):
super(FixupBasicBlockNew, self).__init__()
self.bias1a = nn.Parameter(th.zeros(1))
self.conv1 = ConvModule(n_features, n_features, ksize=ksize, stride
=1, pad=pad, activation=None, norm_layer=None, padding_mode=
padding_mode)
self.bias1b = nn.Parameter(th.zeros(1))
self.activation = _get_activation(activation)
self.bias2a = nn.Parameter(th.zeros(1))
self.conv2 = ConvModule(n_features, n_features, ksize=ksize, stride
=1, pad=pad, activation=None, norm_layer=None, padding_mode=
padding_mode)
self.scale = nn.Parameter(th.ones(1))
self.bias2b = nn.Parameter(th.zeros(1))
self.activation2 = _get_activation(activation)
self.ksize = 3
self.pad = pad
def forward(self, input_0):
primals_2 = self.bias1a
primals_5 = self.bias1b
primals_6 = self.bias2a
primals_9 = self.scale
primals_10 = self.bias2b
primals_3 = self.conv1.conv.weight
primals_4 = self.conv1.conv.bias
primals_7 = self.conv2.conv.weight
primals_8 = self.conv2.conv.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9, primals_10])
return output[0]
|
IlyaBizyaev/ttools
|
FixupBasicBlock
| false
| 8,358
|
[
"MIT"
] | 11
|
b1435b19f397ce1baff9daed3cb287e52a029fdb
|
https://github.com/IlyaBizyaev/ttools/tree/b1435b19f397ce1baff9daed3cb287e52a029fdb
|
LabelPredictor
|
# 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/u7/cu7el5okpylwtytr7ulyozevinrghb53q53bmqwswo2zwf3hvhr3.py
# Topologically Sorted Source Nodes: [pred_1], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# pred_1 => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_1,), kwargs = {})
triton_poi_fused_sigmoid_0 = async_compile.triton('triton_poi_fused_sigmoid_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_sigmoid_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_sigmoid_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
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (0))
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 4), (4, 1))
assert_size_stride(primals_2, (1, ), (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, 1), (1, 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, 1), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [pred_1], Original ATen: [aten.sigmoid]
stream0 = get_raw_stream(0)
triton_poi_fused_sigmoid_0.run(buf1, primals_2, 64, grid=grid(64), stream=stream0)
del primals_2
return (buf1, 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((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, ), (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 LabelPredictor(nn.Module):
def __init__(self, nz_feat, classify_rot=True):
super(LabelPredictor, self).__init__()
self.pred_layer = nn.Linear(nz_feat, 1)
def forward(self, feat):
pred = self.pred_layer.forward(feat)
pred = torch.sigmoid(pred)
return pred
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'nz_feat': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch 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_sigmoid_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
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 0)
tmp2 = tl.broadcast_to(tmp1, [XBLOCK])
tmp3 = tmp0 + tmp2
tmp4 = tl.sigmoid(tmp3)
tl.store(in_out_ptr0 + x0, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 4), (4, 1))
assert_size_stride(primals_2, (1,), (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, 1), (1, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 1), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_sigmoid_0[grid(64)](buf1, primals_2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_2
return buf1, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1
class LabelPredictorNew(nn.Module):
def __init__(self, nz_feat, classify_rot=True):
super(LabelPredictorNew, self).__init__()
self.pred_layer = nn.Linear(nz_feat, 1)
def forward(self, input_0):
primals_1 = self.pred_layer.weight
primals_2 = self.pred_layer.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
JasonQSY/Associative3D
|
LabelPredictor
| false
| 8,359
|
[
"MIT"
] | 25
|
c50818b593ec48c38ed7ee3e109c23531089da32
|
https://github.com/JasonQSY/Associative3D/tree/c50818b593ec48c38ed7ee3e109c23531089da32
|
MultimodalHead
|
# 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: [xy_cat], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# xy_cat => 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/om/com5a2jdtzsd2gxozlfy64n7thpakjehaajnnx7weffss4gxj2sy.py
# Topologically Sorted Source Nodes: [xy_cat_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# xy_cat_2 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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
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/5m/c5mma4y56ura3imiphserxkqyervoqe3bptp4i4swvp3yenvzn36.py
# Topologically Sorted Source Nodes: [xy_cat_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# xy_cat_2 => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = 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, ))
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: [xy_cat], 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)
del primals_1
del primals_2
buf1 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [xy_cat_1], 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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [xy_cat_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [xy_cat_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf2, buf3, 256, grid=grid(256), stream=stream0)
del buf2
return (buf3, reinterpret_tensor(buf0, (64, 8), (8, 1), 0), buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 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)
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
class MultimodalHead(nn.Module):
"""
Multimodal head for the conv net outputs.
This layer concatenate the outputs of audio and visual convoluational nets
and performs a fully-connected projection
"""
def __init__(self, dim_in, num_classes, dropout_rate=0.0, act_func=
'softmax'):
"""
Args:
dim_in (int): the channel dimensions of the visual/audio inputs.
num_classes (int): the channel dimension of the output.
dropout_rate (float): dropout rate. If equal to 0.0, perform no
dropout.
act_func (string): activation function to use. 'softmax': applies
softmax on the output. 'sigmoid': applies sigmoid on the output.
"""
super(MultimodalHead, self).__init__()
if dropout_rate > 0.0:
self.dropout = nn.Dropout(dropout_rate)
self.projection = nn.Linear(sum(dim_in), num_classes, bias=True)
if act_func == 'softmax':
self.act = nn.Softmax(dim=-1)
elif act_func == 'sigmoid':
self.act = nn.Sigmoid()
else:
raise NotImplementedError(
'{} is not supported as an activationfunction.'.format(
act_func))
def forward(self, x, y):
xy_cat = torch.cat((x, y), dim=-1)
if hasattr(self, 'dropout'):
xy_cat = self.dropout(xy_cat)
xy_cat = self.projection(xy_cat)
if not self.training:
xy_cat = self.act(xy_cat)
return xy_cat
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'dim_in': [4, 4], 'num_classes': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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_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__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = 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,))
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)
del primals_1
del primals_2
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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf1, buf2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf1
triton_poi_fused__softmax_2[grid(256)](buf2, buf3, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf2
return buf3, reinterpret_tensor(buf0, (64, 8), (8, 1), 0), buf3
class MultimodalHeadNew(nn.Module):
"""
Multimodal head for the conv net outputs.
This layer concatenate the outputs of audio and visual convoluational nets
and performs a fully-connected projection
"""
def __init__(self, dim_in, num_classes, dropout_rate=0.0, act_func=
'softmax'):
"""
Args:
dim_in (int): the channel dimensions of the visual/audio inputs.
num_classes (int): the channel dimension of the output.
dropout_rate (float): dropout rate. If equal to 0.0, perform no
dropout.
act_func (string): activation function to use. 'softmax': applies
softmax on the output. 'sigmoid': applies sigmoid on the output.
"""
super(MultimodalHeadNew, self).__init__()
if dropout_rate > 0.0:
self.dropout = nn.Dropout(dropout_rate)
self.projection = nn.Linear(sum(dim_in), num_classes, bias=True)
if act_func == 'softmax':
self.act = nn.Softmax(dim=-1)
elif act_func == 'sigmoid':
self.act = nn.Sigmoid()
else:
raise NotImplementedError(
'{} is not supported as an activationfunction.'.format(
act_func))
def forward(self, input_0, input_1):
primals_3 = self.projection.weight
primals_4 = self.projection.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
JiwanChung/acav100m
|
MultimodalHead
| false
| 8,360
|
[
"MIT"
] | 27
|
51cb948d5682da69334a8d05d2df631971b60215
|
https://github.com/JiwanChung/acav100m/tree/51cb948d5682da69334a8d05d2df631971b60215
|
CNN_small
|
# 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/x6/cx6gv6zuidd6x6ruyylvvviaeho3kngissg76xu3rlub5tsz3ztv.py
# Topologically Sorted Source Nodes: [conv2d, relu], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d => convolution
# relu => relu
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_convolution_relu_0 = async_compile.triton('triton_poi_fused_convolution_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32768],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 18816
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 784) % 6
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/di/cdilhsraowpc4455uuolbkzkppabfggt7uygpzwf5542ztj6h4sc.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x => getitem, getitem_1
# Graph fragment:
# %getitem : [num_users=2] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 0), kwargs = {})
# %getitem_1 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_1 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i8', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_max_pool2d_with_indices_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_max_pool2d_with_indices_1(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 4704
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 14
x3 = (xindex // 14)
x2 = (xindex // 1176)
x4 = xindex % 1176
tmp0 = tl.load(in_ptr0 + ((2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (28 + (2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (29 + (2*x0) + (56*x3)), xmask, eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (x4 + (1184*x2)), tmp6, xmask)
tl.store(out_ptr1 + (x4 + (1280*x2)), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ee/ceeva6iv3mozgrawsjpvwa2bdz6tz5zrsscynyu5eixnhjzlgeej.py
# Topologically Sorted Source Nodes: [conv2d_1, relu_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d_1 => convolution_1
# relu_1 => relu_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%getitem, %primals_4, %primals_5, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_relu_2 = async_compile.triton('triton_poi_fused_convolution_relu_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_relu_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 6400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 100) % 16
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x3), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/22/c22xudorbl4i7pd6ur5mccol5iaivhxawnvbofzq2iyziy74773n.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x_1 => _low_memory_max_pool2d_with_offsets_1, getitem_3
# Graph fragment:
# %_low_memory_max_pool2d_with_offsets_1 : [num_users=2] = call_function[target=torch.ops.prims._low_memory_max_pool2d_with_offsets.default](args = (%relu_1, [2, 2], [2, 2], [0, 0], [1, 1], False), kwargs = {})
# %getitem_3 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets_1, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_3 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i8', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_max_pool2d_with_indices_3', '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_pool2d_with_indices_3(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 1600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = (xindex // 5)
x2 = xindex
tmp0 = tl.load(in_ptr0 + ((2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (1 + (2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr0 + (10 + (2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (11 + (2*x0) + (20*x1)), xmask, eviction_policy='evict_last')
tmp2 = tmp1 > tmp0
tmp3 = tl.full([1], 1, tl.int8)
tmp4 = tl.full([1], 0, tl.int8)
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = triton_helpers.maximum(tmp1, tmp0)
tmp8 = tmp7 > tmp6
tmp9 = tl.full([1], 2, tl.int8)
tmp10 = tl.where(tmp8, tmp9, tmp5)
tmp11 = triton_helpers.maximum(tmp7, tmp6)
tmp13 = tmp12 > tmp11
tmp14 = tl.full([1], 3, tl.int8)
tmp15 = tl.where(tmp13, tmp14, tmp10)
tmp16 = triton_helpers.maximum(tmp12, tmp11)
tl.store(out_ptr0 + (x2), tmp15, xmask)
tl.store(out_ptr1 + (x2), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ru/cruawdvfvxrhsbe7qff46sjr6zw27jrt5kifvimhizevclrhnjxs.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_3 => relu_2
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_7), kwargs = {})
# %relu_2 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_1,), kwargs = {})
triton_poi_fused_relu_4 = async_compile.triton('triton_poi_fused_relu_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_4', '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_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 480
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 120
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/w3/cw3dl3pqvlgwpje5dexejrnijkmtxsxgsenqkvj2lymiezzeghmf.py
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_4 => relu_3
# Graph fragment:
# %add_tensor : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default, %primals_9), kwargs = {})
# %relu_3 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor,), kwargs = {})
triton_poi_fused_relu_5 = async_compile.triton('triton_poi_fused_relu_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_5', '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_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 336
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 84
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (6, 3, 5, 5), (75, 25, 5, 1))
assert_size_stride(primals_2, (6, ), (1, ))
assert_size_stride(primals_3, (4, 3, 32, 32), (3072, 1024, 32, 1))
assert_size_stride(primals_4, (16, 6, 5, 5), (150, 25, 5, 1))
assert_size_stride(primals_5, (16, ), (1, ))
assert_size_stride(primals_6, (120, 400), (400, 1))
assert_size_stride(primals_7, (120, ), (1, ))
assert_size_stride(primals_8, (84, 120), (120, 1))
assert_size_stride(primals_9, (84, ), (1, ))
assert_size_stride(primals_10, (10, 84), (84, 1))
assert_size_stride(primals_11, (10, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 6, 28, 28), (4704, 784, 28, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv2d, relu], Original ATen: [aten.convolution, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_relu_0.run(buf1, primals_2, 18816, grid=grid(18816), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((4, 6, 14, 14), (1184, 196, 14, 1), torch.float32)
buf3 = empty_strided_cuda((4, 6, 14, 14), (1280, 196, 14, 1), torch.int8)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_1.run(buf1, buf2, buf3, 4704, grid=grid(4704), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 16, 10, 10), (1600, 100, 10, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [conv2d_1, relu_1], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf5, primals_5, 6400, grid=grid(6400), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.int8)
buf7 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_3.run(buf5, buf6, buf7, 1600, grid=grid(1600), stream=stream0)
buf8 = empty_strided_cuda((4, 120), (120, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf7, (4, 400), (400, 1), 0), reinterpret_tensor(primals_6, (400, 120), (1, 400), 0), out=buf8)
buf9 = buf8; del buf8 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.relu]
triton_poi_fused_relu_4.run(buf9, primals_7, 480, grid=grid(480), stream=stream0)
del primals_7
buf10 = empty_strided_cuda((4, 84), (84, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf9, reinterpret_tensor(primals_8, (120, 84), (1, 120), 0), out=buf10)
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [x_4], Original ATen: [aten.relu]
triton_poi_fused_relu_5.run(buf11, primals_9, 336, grid=grid(336), stream=stream0)
del primals_9
buf12 = empty_strided_cuda((4, 10), (10, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, buf11, reinterpret_tensor(primals_10, (84, 10), (1, 84), 0), alpha=1, beta=1, out=buf12)
del primals_11
return (buf12, primals_1, primals_3, primals_4, buf1, buf2, buf3, buf5, buf6, reinterpret_tensor(buf7, (4, 400), (400, 1), 0), buf9, buf11, primals_10, primals_8, primals_6, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((6, 3, 5, 5), (75, 25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((6, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 3, 32, 32), (3072, 1024, 32, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, 6, 5, 5), (150, 25, 5, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((120, 400), (400, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((120, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((84, 120), (120, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((84, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((10, 84), (84, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((10, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
from torch.nn import functional as F
import torch.utils.data
class CNN_small(nn.Module):
def __init__(self, num_classes=10):
super(CNN_small, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, num_classes)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = x.view(-1, 16 * 5 * 5)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
def get_inputs():
return [torch.rand([4, 3, 32, 32])]
def get_init_inputs():
return [[], {}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.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_convolution_relu_0(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 18816
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 784 % 6
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, xmask)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_1(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 4704
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 14
x3 = xindex // 14
x2 = xindex // 1176
x4 = xindex % 1176
tmp0 = tl.load(in_ptr0 + (2 * x0 + 56 * x3), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 56 * x3), xmask, eviction_policy
='evict_last')
tmp3 = tl.load(in_ptr0 + (28 + 2 * x0 + 56 * x3), xmask,
eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (29 + 2 * x0 + 56 * x3), xmask,
eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (x4 + 1184 * x2), tmp6, xmask)
tl.store(out_ptr1 + (x4 + 1280 * x2), tmp16, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_2(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 6400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 100 % 16
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x3, tmp4, xmask)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_3(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 1600
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 5
x1 = xindex // 5
x2 = xindex
tmp0 = tl.load(in_ptr0 + (2 * x0 + 20 * x1), xmask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr0 + (1 + 2 * x0 + 20 * x1), xmask, eviction_policy
='evict_last')
tmp7 = tl.load(in_ptr0 + (10 + 2 * x0 + 20 * x1), xmask,
eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (11 + 2 * x0 + 20 * x1), xmask,
eviction_policy='evict_last')
tmp2 = tmp1 > tmp0
tmp3 = tl.full([1], 1, tl.int8)
tmp4 = tl.full([1], 0, tl.int8)
tmp5 = tl.where(tmp2, tmp3, tmp4)
tmp6 = triton_helpers.maximum(tmp1, tmp0)
tmp8 = tmp7 > tmp6
tmp9 = tl.full([1], 2, tl.int8)
tmp10 = tl.where(tmp8, tmp9, tmp5)
tmp11 = triton_helpers.maximum(tmp7, tmp6)
tmp13 = tmp12 > tmp11
tmp14 = tl.full([1], 3, tl.int8)
tmp15 = tl.where(tmp13, tmp14, tmp10)
tmp16 = triton_helpers.maximum(tmp12, tmp11)
tl.store(out_ptr0 + x2, tmp15, xmask)
tl.store(out_ptr1 + x2, tmp16, xmask)
@triton.jit
def triton_poi_fused_relu_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 480
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 120
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_relu_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 336
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 84
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (6, 3, 5, 5), (75, 25, 5, 1))
assert_size_stride(primals_2, (6,), (1,))
assert_size_stride(primals_3, (4, 3, 32, 32), (3072, 1024, 32, 1))
assert_size_stride(primals_4, (16, 6, 5, 5), (150, 25, 5, 1))
assert_size_stride(primals_5, (16,), (1,))
assert_size_stride(primals_6, (120, 400), (400, 1))
assert_size_stride(primals_7, (120,), (1,))
assert_size_stride(primals_8, (84, 120), (120, 1))
assert_size_stride(primals_9, (84,), (1,))
assert_size_stride(primals_10, (10, 84), (84, 1))
assert_size_stride(primals_11, (10,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 6, 28, 28), (4704, 784, 28, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_relu_0[grid(18816)](buf1, primals_2,
18816, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((4, 6, 14, 14), (1184, 196, 14, 1), torch
.float32)
buf3 = empty_strided_cuda((4, 6, 14, 14), (1280, 196, 14, 1), torch
.int8)
triton_poi_fused_max_pool2d_with_indices_1[grid(4704)](buf1, buf2,
buf3, 4704, XBLOCK=128, num_warps=4, num_stages=1)
buf4 = extern_kernels.convolution(buf2, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 16, 10, 10), (1600, 100, 10, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_relu_2[grid(6400)](buf5, primals_5,
6400, XBLOCK=256, num_warps=4, num_stages=1)
del primals_5
buf6 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.int8)
buf7 = empty_strided_cuda((4, 16, 5, 5), (400, 25, 5, 1), torch.float32
)
triton_poi_fused_max_pool2d_with_indices_3[grid(1600)](buf5, buf6,
buf7, 1600, XBLOCK=128, num_warps=4, num_stages=1)
buf8 = empty_strided_cuda((4, 120), (120, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf7, (4, 400), (400, 1), 0),
reinterpret_tensor(primals_6, (400, 120), (1, 400), 0), out=buf8)
buf9 = buf8
del buf8
triton_poi_fused_relu_4[grid(480)](buf9, primals_7, 480, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_7
buf10 = empty_strided_cuda((4, 84), (84, 1), torch.float32)
extern_kernels.mm(buf9, reinterpret_tensor(primals_8, (120, 84), (1,
120), 0), out=buf10)
buf11 = buf10
del buf10
triton_poi_fused_relu_5[grid(336)](buf11, primals_9, 336, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_9
buf12 = empty_strided_cuda((4, 10), (10, 1), torch.float32)
extern_kernels.addmm(primals_11, buf11, reinterpret_tensor(
primals_10, (84, 10), (1, 84), 0), alpha=1, beta=1, out=buf12)
del primals_11
return (buf12, primals_1, primals_3, primals_4, buf1, buf2, buf3, buf5,
buf6, reinterpret_tensor(buf7, (4, 400), (400, 1), 0), buf9, buf11,
primals_10, primals_8, primals_6)
class CNN_smallNew(nn.Module):
def __init__(self, num_classes=10):
super(CNN_smallNew, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, num_classes)
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_6 = self.fc1.weight
primals_7 = self.fc1.bias
primals_8 = self.fc2.weight
primals_9 = self.fc2.bias
primals_10 = self.fc3.weight
primals_11 = self.fc3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
|
JiarunLiu/Co-correcting
|
CNN_small
| false
| 8,361
|
[
"Apache-2.0"
] | 19
|
4e3ca4951de5d73ca812bbbcfe666273082ff2fd
|
https://github.com/JiarunLiu/Co-correcting/tree/4e3ca4951de5d73ca812bbbcfe666273082ff2fd
|
CRFLoss
|
# 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/t6/ct63xddmvrwrurfmzfkl2onel7brbvegrcdo7jl4mmzcpzcc76ah.py
# Topologically Sorted Source Nodes: [add_1, cur, logsumexp], Original ATen: [aten.add, aten.logsumexp]
# Source node to ATen node mapping:
# add_1 => add_1
# cur => add_2
# logsumexp => abs_1, amax, eq, exp, full_default, sub, sum_1, where
# Graph fragment:
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%unsqueeze, %unsqueeze_1), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %permute_2), kwargs = {})
# %amax : [num_users=2] = call_function[target=torch.ops.aten.amax.default](args = (%add_2, [1], True), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%amax,), kwargs = {})
# %eq : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%abs_1, inf), kwargs = {})
# %full_default : [num_users=4] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %full_default, %amax), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_2, %where), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1]), kwargs = {})
triton_poi_fused_add_logsumexp_0 = async_compile.triton('triton_poi_fused_add_logsumexp_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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: '*i64', 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_add_logsumexp_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 13, '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_logsumexp_0(in_ptr0, in_ptr1, in_ptr2, 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
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (0))
tmp1 = tl.broadcast_to(tmp0, [XBLOCK])
tmp2 = tl.load(in_ptr1 + (16*x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (4 + x0 + (16*x1)), xmask)
tmp8 = tl.load(in_ptr2 + (4*x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + (1))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp12 = tl.load(in_ptr1 + (1 + (16*x1)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr2 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr0 + (2))
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp21 = tl.load(in_ptr1 + (2 + (16*x1)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr2 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr0 + (3))
tmp29 = tl.broadcast_to(tmp28, [XBLOCK])
tmp30 = tl.load(in_ptr1 + (3 + (16*x1)), xmask, eviction_policy='evict_last')
tmp34 = tl.load(in_ptr2 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp3 = tmp2.to(tl.float32)
tmp4 = tmp1 + tmp3
tmp6 = tmp5.to(tl.float32)
tmp7 = tmp4 + tmp6
tmp9 = tmp7 + tmp8
tmp13 = tmp12.to(tl.float32)
tmp14 = tmp11 + tmp13
tmp15 = tmp14 + tmp6
tmp17 = tmp15 + tmp16
tmp18 = triton_helpers.maximum(tmp9, tmp17)
tmp22 = tmp21.to(tl.float32)
tmp23 = tmp20 + tmp22
tmp24 = tmp23 + tmp6
tmp26 = tmp24 + tmp25
tmp27 = triton_helpers.maximum(tmp18, tmp26)
tmp31 = tmp30.to(tl.float32)
tmp32 = tmp29 + tmp31
tmp33 = tmp32 + tmp6
tmp35 = tmp33 + tmp34
tmp36 = triton_helpers.maximum(tmp27, tmp35)
tmp37 = tl_math.abs(tmp36)
tmp38 = float("inf")
tmp39 = tmp37 == tmp38
tmp40 = 0.0
tmp41 = tl.where(tmp39, tmp40, tmp36)
tmp42 = tmp9 - tmp41
tmp43 = tl_math.exp(tmp42)
tmp44 = tmp17 - tmp41
tmp45 = tl_math.exp(tmp44)
tmp46 = tmp43 + tmp45
tmp47 = tmp26 - tmp41
tmp48 = tl_math.exp(tmp47)
tmp49 = tmp46 + tmp48
tmp50 = tmp35 - tmp41
tmp51 = tl_math.exp(tmp50)
tmp52 = tmp49 + tmp51
tl.store(out_ptr0 + (x2), tmp36, xmask)
tl.store(out_ptr1 + (x2), tmp52, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ip/cip4b45ktm4voqrpp53advyy56sjyrnwf2eiklzwxplbmwnudd6o.py
# Topologically Sorted Source Nodes: [logsumexp, add_3, cur_1, logsumexp_1], Original ATen: [aten.logsumexp, aten.add]
# Source node to ATen node mapping:
# add_3 => add_4
# cur_1 => add_5
# logsumexp => full_default
# logsumexp_1 => abs_2, amax_1, eq_1, exp_1, sub_1, sum_2, where_1
# Graph fragment:
# %full_default : [num_users=4] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%unsqueeze_2, %unsqueeze_3), kwargs = {})
# %add_5 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_4, %permute_2), kwargs = {})
# %amax_1 : [num_users=2] = call_function[target=torch.ops.aten.amax.default](args = (%add_5, [1], True), kwargs = {})
# %abs_2 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%amax_1,), kwargs = {})
# %eq_1 : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%abs_2, inf), kwargs = {})
# %where_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_1, %full_default, %amax_1), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_5, %where_1), kwargs = {})
# %exp_1 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_1,), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_1, [1]), kwargs = {})
triton_poi_fused_add_logsumexp_1 = async_compile.triton('triton_poi_fused_add_logsumexp_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: '*i64', 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_logsumexp_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 13, '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_logsumexp_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, 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
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (8 + x0 + (16*x1)), xmask)
tmp12 = tl.load(in_ptr3 + (4*x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr3 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr3 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp36 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp38 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp44 = tl.load(in_ptr3 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = tl_math.log(tmp0)
tmp3 = tl_math.abs(tmp2)
tmp4 = float("inf")
tmp5 = tmp3 == tmp4
tmp6 = 0.0
tmp7 = tl.where(tmp5, tmp6, tmp2)
tmp8 = tmp1 + tmp7
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tl_math.log(tmp14)
tmp17 = tl_math.abs(tmp16)
tmp18 = tmp17 == tmp4
tmp19 = tl.where(tmp18, tmp6, tmp16)
tmp20 = tmp15 + tmp19
tmp21 = tmp20 + tmp10
tmp23 = tmp21 + tmp22
tmp24 = triton_helpers.maximum(tmp13, tmp23)
tmp26 = tl_math.log(tmp25)
tmp28 = tl_math.abs(tmp27)
tmp29 = tmp28 == tmp4
tmp30 = tl.where(tmp29, tmp6, tmp27)
tmp31 = tmp26 + tmp30
tmp32 = tmp31 + tmp10
tmp34 = tmp32 + tmp33
tmp35 = triton_helpers.maximum(tmp24, tmp34)
tmp37 = tl_math.log(tmp36)
tmp39 = tl_math.abs(tmp38)
tmp40 = tmp39 == tmp4
tmp41 = tl.where(tmp40, tmp6, tmp38)
tmp42 = tmp37 + tmp41
tmp43 = tmp42 + tmp10
tmp45 = tmp43 + tmp44
tmp46 = triton_helpers.maximum(tmp35, tmp45)
tmp47 = tl_math.abs(tmp46)
tmp48 = tmp47 == tmp4
tmp49 = tl.where(tmp48, tmp6, tmp46)
tmp50 = tmp13 - tmp49
tmp51 = tl_math.exp(tmp50)
tmp52 = tmp23 - tmp49
tmp53 = tl_math.exp(tmp52)
tmp54 = tmp51 + tmp53
tmp55 = tmp34 - tmp49
tmp56 = tl_math.exp(tmp55)
tmp57 = tmp54 + tmp56
tmp58 = tmp45 - tmp49
tmp59 = tl_math.exp(tmp58)
tmp60 = tmp57 + tmp59
tl.store(out_ptr0 + (x2), tmp46, xmask)
tl.store(out_ptr1 + (x2), tmp60, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/dy/cdy57vwzdvcdqwbotoym6jkngdeywqpl7c6kmxpvecoo7otey4bd.py
# Topologically Sorted Source Nodes: [logsumexp, add_5, cur_2, logsumexp_2], Original ATen: [aten.logsumexp, aten.add]
# Source node to ATen node mapping:
# add_5 => add_7
# cur_2 => add_8
# logsumexp => full_default
# logsumexp_2 => abs_3, amax_2, eq_2, exp_2, sub_2, sum_3, where_2
# Graph fragment:
# %full_default : [num_users=4] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %add_7 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%unsqueeze_4, %unsqueeze_5), kwargs = {})
# %add_8 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_7, %permute_2), kwargs = {})
# %amax_2 : [num_users=2] = call_function[target=torch.ops.aten.amax.default](args = (%add_8, [1], True), kwargs = {})
# %abs_3 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%amax_2,), kwargs = {})
# %eq_2 : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%abs_3, inf), kwargs = {})
# %where_2 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_2, %full_default, %amax_2), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_8, %where_2), kwargs = {})
# %exp_2 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_2,), kwargs = {})
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_2, [1]), kwargs = {})
triton_poi_fused_add_logsumexp_2 = async_compile.triton('triton_poi_fused_add_logsumexp_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: '*i64', 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_logsumexp_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 13, '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_logsumexp_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, 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
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (12 + x0 + (16*x1)), xmask)
tmp12 = tl.load(in_ptr3 + (4*x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr3 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp27 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr3 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp36 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp38 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp44 = tl.load(in_ptr3 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = tl_math.log(tmp0)
tmp3 = tl_math.abs(tmp2)
tmp4 = float("inf")
tmp5 = tmp3 == tmp4
tmp6 = 0.0
tmp7 = tl.where(tmp5, tmp6, tmp2)
tmp8 = tmp1 + tmp7
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tl_math.log(tmp14)
tmp17 = tl_math.abs(tmp16)
tmp18 = tmp17 == tmp4
tmp19 = tl.where(tmp18, tmp6, tmp16)
tmp20 = tmp15 + tmp19
tmp21 = tmp20 + tmp10
tmp23 = tmp21 + tmp22
tmp24 = triton_helpers.maximum(tmp13, tmp23)
tmp26 = tl_math.log(tmp25)
tmp28 = tl_math.abs(tmp27)
tmp29 = tmp28 == tmp4
tmp30 = tl.where(tmp29, tmp6, tmp27)
tmp31 = tmp26 + tmp30
tmp32 = tmp31 + tmp10
tmp34 = tmp32 + tmp33
tmp35 = triton_helpers.maximum(tmp24, tmp34)
tmp37 = tl_math.log(tmp36)
tmp39 = tl_math.abs(tmp38)
tmp40 = tmp39 == tmp4
tmp41 = tl.where(tmp40, tmp6, tmp38)
tmp42 = tmp37 + tmp41
tmp43 = tmp42 + tmp10
tmp45 = tmp43 + tmp44
tmp46 = triton_helpers.maximum(tmp35, tmp45)
tmp47 = tl_math.abs(tmp46)
tmp48 = tmp47 == tmp4
tmp49 = tl.where(tmp48, tmp6, tmp46)
tmp50 = tmp13 - tmp49
tmp51 = tl_math.exp(tmp50)
tmp52 = tmp23 - tmp49
tmp53 = tl_math.exp(tmp52)
tmp54 = tmp51 + tmp53
tmp55 = tmp34 - tmp49
tmp56 = tl_math.exp(tmp55)
tmp57 = tmp54 + tmp56
tmp58 = tmp45 - tmp49
tmp59 = tl_math.exp(tmp58)
tmp60 = tmp57 + tmp59
tl.store(out_ptr0 + (x2), tmp46, xmask)
tl.store(out_ptr1 + (x2), tmp60, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/nq/cnqhciwfke5eb5tu7d76qfs4bmxzvzqvaqtr5kz5ywnopvcq7jfb.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 = ([%gather_1, %index, %index_1, %index_2, %gather_2],), kwargs = {})
triton_poi_fused_stack_3 = async_compile.triton('triton_poi_fused_stack_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=[32],
filename=__file__,
triton_meta={'signature': {0: '*i64', 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_stack_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_stack_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 20
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4*x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tl.full([XBLOCK], 4, tl.int32)
tmp7 = tmp5 + tmp6
tmp8 = tmp5 < 0
tmp9 = tl.where(tmp8, tmp7, tmp5)
tl.device_assert(((0 <= tl.broadcast_to(tmp9, [XBLOCK])) & (tl.broadcast_to(tmp9, [XBLOCK]) < 4)) | ~(tmp4 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp9, [XBLOCK]) < 4")
tmp11 = tl.load(in_ptr1 + (tl.broadcast_to(tmp9, [XBLOCK])), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp12 = tmp0 >= tmp3
tmp13 = tl.full([1], 8, tl.int64)
tmp14 = tmp0 < tmp13
tmp15 = tmp12 & tmp14
tmp16 = tl.load(in_ptr0 + (1 + (4*((-4) + x0))), tmp15 & xmask, eviction_policy='evict_last', other=0.0)
tmp17 = tmp16 + tmp6
tmp18 = tmp16 < 0
tmp19 = tl.where(tmp18, tmp17, tmp16)
tl.device_assert(((0 <= tl.broadcast_to(tmp19, [XBLOCK])) & (tl.broadcast_to(tmp19, [XBLOCK]) < 4)) | ~(tmp15 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp19, [XBLOCK]) < 4")
tmp21 = tl.load(in_ptr0 + (4*((-4) + x0)), tmp15 & xmask, eviction_policy='evict_last', other=0.0)
tmp22 = tmp21 + tmp6
tmp23 = tmp21 < 0
tmp24 = tl.where(tmp23, tmp22, tmp21)
tl.device_assert(((0 <= tl.broadcast_to(tmp24, [XBLOCK])) & (tl.broadcast_to(tmp24, [XBLOCK]) < 4)) | ~(tmp15 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp24, [XBLOCK]) < 4")
tmp26 = tl.load(in_ptr2 + (tl.broadcast_to(tmp24 + (4*tmp19), [XBLOCK])), tmp15 & xmask, eviction_policy='evict_last', other=0.0)
tmp27 = tmp0 >= tmp13
tmp28 = tl.full([1], 12, tl.int64)
tmp29 = tmp0 < tmp28
tmp30 = tmp27 & tmp29
tmp31 = tl.load(in_ptr0 + (2 + (4*((-8) + x0))), tmp30 & xmask, eviction_policy='evict_last', other=0.0)
tmp32 = tmp31 + tmp6
tmp33 = tmp31 < 0
tmp34 = tl.where(tmp33, tmp32, tmp31)
tl.device_assert(((0 <= tl.broadcast_to(tmp34, [XBLOCK])) & (tl.broadcast_to(tmp34, [XBLOCK]) < 4)) | ~(tmp30 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp34, [XBLOCK]) < 4")
tmp36 = tl.load(in_ptr0 + (1 + (4*((-8) + x0))), tmp30 & xmask, eviction_policy='evict_last', other=0.0)
tmp37 = tmp36 + tmp6
tmp38 = tmp36 < 0
tmp39 = tl.where(tmp38, tmp37, tmp36)
tl.device_assert(((0 <= tl.broadcast_to(tmp39, [XBLOCK])) & (tl.broadcast_to(tmp39, [XBLOCK]) < 4)) | ~(tmp30 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp39, [XBLOCK]) < 4")
tmp41 = tl.load(in_ptr2 + (tl.broadcast_to(tmp39 + (4*tmp34), [XBLOCK])), tmp30 & xmask, eviction_policy='evict_last', other=0.0)
tmp42 = tmp0 >= tmp28
tmp43 = tl.full([1], 16, tl.int64)
tmp44 = tmp0 < tmp43
tmp45 = tmp42 & tmp44
tmp46 = tl.load(in_ptr0 + (3 + (4*((-12) + x0))), tmp45 & xmask, eviction_policy='evict_last', other=0.0)
tmp47 = tmp46 + tmp6
tmp48 = tmp46 < 0
tmp49 = tl.where(tmp48, tmp47, tmp46)
tl.device_assert(((0 <= tl.broadcast_to(tmp49, [XBLOCK])) & (tl.broadcast_to(tmp49, [XBLOCK]) < 4)) | ~(tmp45 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp49, [XBLOCK]) < 4")
tmp51 = tl.load(in_ptr0 + (2 + (4*((-12) + x0))), tmp45 & xmask, eviction_policy='evict_last', other=0.0)
tmp52 = tmp51 + tmp6
tmp53 = tmp51 < 0
tmp54 = tl.where(tmp53, tmp52, tmp51)
tl.device_assert(((0 <= tl.broadcast_to(tmp54, [XBLOCK])) & (tl.broadcast_to(tmp54, [XBLOCK]) < 4)) | ~(tmp45 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp54, [XBLOCK]) < 4")
tmp56 = tl.load(in_ptr2 + (tl.broadcast_to(tmp54 + (4*tmp49), [XBLOCK])), tmp45 & xmask, eviction_policy='evict_last', other=0.0)
tmp57 = tmp0 >= tmp43
tmp58 = tl.full([1], 20, tl.int64)
tmp59 = tmp0 < tmp58
tmp60 = tl.load(in_ptr0 + (3 + (4*((-16) + x0))), tmp57 & xmask, eviction_policy='evict_last', other=0.0)
tmp61 = tmp60 + tmp6
tmp62 = tmp60 < 0
tmp63 = tl.where(tmp62, tmp61, tmp60)
tl.device_assert(((0 <= tl.broadcast_to(tmp63, [XBLOCK])) & (tl.broadcast_to(tmp63, [XBLOCK]) < 4)) | ~(tmp57 & xmask), "index out of bounds: 0 <= tl.broadcast_to(tmp63, [XBLOCK]) < 4")
tmp65 = tl.load(in_ptr3 + (tl.broadcast_to(tmp63, [XBLOCK])), tmp57 & xmask, eviction_policy='evict_last', other=0.0)
tmp66 = tl.where(tmp45, tmp56, tmp65)
tmp67 = tl.where(tmp30, tmp41, tmp66)
tmp68 = tl.where(tmp15, tmp26, tmp67)
tmp69 = tl.where(tmp4, tmp11, tmp68)
tl.store(out_ptr0 + (x0), tmp69, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/jo/cjof6wkwqqyuqnshhtxfbhtrtthq7x2eqwfhv2hmeqmry3wo3epy.py
# Topologically Sorted Source Nodes: [logsumexp, logsumexp_2, prev_4, normalizers, emits, trans, target_scores, sub, loss], Original ATen: [aten.logsumexp, aten.add, aten.sum, aten.sub, aten.mean]
# Source node to ATen node mapping:
# emits => sum_5
# logsumexp => full_default
# logsumexp_2 => add_9, log_2
# loss => mean
# normalizers => abs_4, add_11, amax_3, eq_3, exp_3, log_3, sub_3, sum_4, where_3
# prev_4 => add_10
# sub => sub_4
# target_scores => add_12
# trans => sum_6
# Graph fragment:
# %full_default : [num_users=4] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %log_2 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_3,), kwargs = {})
# %add_9 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%log_2, %squeeze_2), kwargs = {})
# %add_10 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_9, %primals_4), kwargs = {})
# %amax_3 : [num_users=2] = call_function[target=torch.ops.aten.amax.default](args = (%add_10, [1], True), kwargs = {})
# %abs_4 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%amax_3,), kwargs = {})
# %eq_3 : [num_users=1] = call_function[target=torch.ops.aten.eq.Scalar](args = (%abs_4, inf), kwargs = {})
# %where_3 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_3, %full_default, %amax_3), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_10, %where_3), kwargs = {})
# %exp_3 : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub_3,), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp_3, [1]), kwargs = {})
# %log_3 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sum_4,), kwargs = {})
# %add_11 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%log_3, %squeeze_3), kwargs = {})
# %sum_5 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%squeeze_4, [1]), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%view, [0]), kwargs = {})
# %add_12 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_5, %sum_6), kwargs = {})
# %sub_4 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_11, %add_12), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sub_4,), kwargs = {})
triton_per_fused_add_logsumexp_mean_sub_sum_4 = async_compile.triton('triton_per_fused_add_logsumexp_mean_sub_sum_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 4],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*i64', 2: '*i64', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32', 8: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=90, major=9, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, multi_processor_count=132), 'constants': {7: 1}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3, 4, 5, 6), equal_to_1=(7,))]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_logsumexp_mean_sub_sum_4', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 21, '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_logsumexp_mean_sub_sum_4(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, 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')
tmp7 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr2 + (r0), None)
tmp30 = tl.load(in_ptr2 + (4 + r0), None)
tmp32 = tl.load(in_ptr2 + (8 + r0), None)
tmp34 = tl.load(in_ptr2 + (12 + r0), None)
tmp36 = tl.load(in_ptr2 + (16 + r0), None)
tmp39 = tl.load(in_ptr3 + (4*r0), None, eviction_policy='evict_last')
tmp41 = tl.load(in_ptr4 + (4*r0), None, eviction_policy='evict_last')
tmp48 = tl.load(in_ptr5 + (0))
tmp49 = tl.broadcast_to(tmp48, [XBLOCK, RBLOCK])
tmp51 = tl.load(in_ptr3 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp53 = tl.load(in_ptr4 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp58 = tl.load(in_ptr5 + (1))
tmp59 = tl.broadcast_to(tmp58, [XBLOCK, RBLOCK])
tmp62 = tl.load(in_ptr3 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp64 = tl.load(in_ptr4 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp69 = tl.load(in_ptr5 + (2))
tmp70 = tl.broadcast_to(tmp69, [XBLOCK, RBLOCK])
tmp73 = tl.load(in_ptr3 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp75 = tl.load(in_ptr4 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp80 = tl.load(in_ptr5 + (3))
tmp81 = tl.broadcast_to(tmp80, [XBLOCK, RBLOCK])
tmp1 = tl.full([XBLOCK, RBLOCK], 4, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert((0 <= tmp4) & (tmp4 < 4), "index out of bounds: 0 <= tmp4 < 4")
tmp6 = tl.load(in_ptr1 + (tmp4 + (16*r0)), None, eviction_policy='evict_last')
tmp8 = tmp7 + tmp1
tmp9 = tmp7 < 0
tmp10 = tl.where(tmp9, tmp8, tmp7)
tl.device_assert((0 <= tmp10) & (tmp10 < 4), "index out of bounds: 0 <= tmp10 < 4")
tmp12 = tl.load(in_ptr1 + (4 + tmp10 + (16*r0)), None, eviction_policy='evict_last')
tmp13 = tmp6 + tmp12
tmp15 = tmp14 + tmp1
tmp16 = tmp14 < 0
tmp17 = tl.where(tmp16, tmp15, tmp14)
tl.device_assert((0 <= tmp17) & (tmp17 < 4), "index out of bounds: 0 <= tmp17 < 4")
tmp19 = tl.load(in_ptr1 + (8 + tmp17 + (16*r0)), None, eviction_policy='evict_last')
tmp20 = tmp13 + tmp19
tmp22 = tmp21 + tmp1
tmp23 = tmp21 < 0
tmp24 = tl.where(tmp23, tmp22, tmp21)
tl.device_assert((0 <= tmp24) & (tmp24 < 4), "index out of bounds: 0 <= tmp24 < 4")
tmp26 = tl.load(in_ptr1 + (12 + tmp24 + (16*r0)), None, eviction_policy='evict_last')
tmp27 = tmp20 + tmp26
tmp28 = tmp27.to(tl.float32)
tmp31 = tmp29 + tmp30
tmp33 = tmp31 + tmp32
tmp35 = tmp33 + tmp34
tmp37 = tmp35 + tmp36
tmp38 = tmp28 + tmp37
tmp40 = tl_math.log(tmp39)
tmp42 = tl_math.abs(tmp41)
tmp43 = float("inf")
tmp44 = tmp42 == tmp43
tmp45 = 0.0
tmp46 = tl.where(tmp44, tmp45, tmp41)
tmp47 = tmp40 + tmp46
tmp50 = tmp47 + tmp49
tmp52 = tl_math.log(tmp51)
tmp54 = tl_math.abs(tmp53)
tmp55 = tmp54 == tmp43
tmp56 = tl.where(tmp55, tmp45, tmp53)
tmp57 = tmp52 + tmp56
tmp60 = tmp57 + tmp59
tmp61 = triton_helpers.maximum(tmp50, tmp60)
tmp63 = tl_math.log(tmp62)
tmp65 = tl_math.abs(tmp64)
tmp66 = tmp65 == tmp43
tmp67 = tl.where(tmp66, tmp45, tmp64)
tmp68 = tmp63 + tmp67
tmp71 = tmp68 + tmp70
tmp72 = triton_helpers.maximum(tmp61, tmp71)
tmp74 = tl_math.log(tmp73)
tmp76 = tl_math.abs(tmp75)
tmp77 = tmp76 == tmp43
tmp78 = tl.where(tmp77, tmp45, tmp75)
tmp79 = tmp74 + tmp78
tmp82 = tmp79 + tmp81
tmp83 = triton_helpers.maximum(tmp72, tmp82)
tmp84 = tl_math.abs(tmp83)
tmp85 = tmp84 == tmp43
tmp86 = tl.where(tmp85, tmp45, tmp83)
tmp87 = tmp50 - tmp86
tmp88 = tl_math.exp(tmp87)
tmp89 = tmp60 - tmp86
tmp90 = tl_math.exp(tmp89)
tmp91 = tmp88 + tmp90
tmp92 = tmp71 - tmp86
tmp93 = tl_math.exp(tmp92)
tmp94 = tmp91 + tmp93
tmp95 = tmp82 - tmp86
tmp96 = tl_math.exp(tmp95)
tmp97 = tmp94 + tmp96
tmp98 = tl_math.log(tmp97)
tmp99 = tmp98 + tmp86
tmp100 = tmp99 - tmp38
tmp101 = tl.broadcast_to(tmp100, [XBLOCK, RBLOCK])
tmp103 = tl.sum(tmp101, 1)[:, None]
tmp104 = 4.0
tmp105 = tmp103 / tmp104
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp105, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (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))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add_1, cur, logsumexp], Original ATen: [aten.add, aten.logsumexp]
stream0 = get_raw_stream(0)
triton_poi_fused_add_logsumexp_0.run(primals_2, primals_1, primals_3, buf0, buf1, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [logsumexp, add_3, cur_1, logsumexp_1], Original ATen: [aten.logsumexp, aten.add]
triton_poi_fused_add_logsumexp_1.run(buf1, buf0, primals_1, primals_3, buf2, buf3, 16, grid=grid(16), stream=stream0)
buf4 = reinterpret_tensor(buf1, (4, 1, 4), (4, 16, 1), 0); del buf1 # reuse
buf5 = reinterpret_tensor(buf0, (4, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [logsumexp, add_5, cur_2, logsumexp_2], Original ATen: [aten.logsumexp, aten.add]
triton_poi_fused_add_logsumexp_2.run(buf3, buf2, primals_1, primals_3, buf4, buf5, 16, grid=grid(16), stream=stream0)
del buf2
del buf3
buf8 = empty_strided_cuda((20, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [stack], Original ATen: [aten.stack]
triton_poi_fused_stack_3.run(primals_5, primals_2, primals_3, primals_4, buf8, 20, grid=grid(20), stream=stream0)
buf10 = empty_strided_cuda((), (), torch.float32)
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [logsumexp, logsumexp_2, prev_4, normalizers, emits, trans, target_scores, sub, loss], Original ATen: [aten.logsumexp, aten.add, aten.sum, aten.sub, aten.mean]
triton_per_fused_add_logsumexp_mean_sub_sum_4.run(buf11, primals_5, primals_1, buf8, buf5, buf4, primals_4, 1, 4, grid=grid(1), stream=stream0)
del buf4
del buf5
del buf8
return (buf11, primals_1, primals_2, primals_3, primals_4, reinterpret_tensor(primals_5, (4, ), (4, ), 0), reinterpret_tensor(primals_5, (4, ), (4, ), 1), reinterpret_tensor(primals_5, (4, ), (4, ), 2), reinterpret_tensor(primals_5, (4, ), (4, ), 3), reinterpret_tensor(primals_5, (4, ), (4, ), 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, 4), (16, 4, 1), device='cuda:0', dtype=torch.int64)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.int64)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class CRFLoss(nn.Module):
def __init__(self, L, init):
super(CRFLoss, self).__init__()
self.start = nn.Parameter(torch.Tensor(L).uniform_(-init, init))
self.T = nn.Parameter(torch.Tensor(L, L).uniform_(-init, init))
self.end = nn.Parameter(torch.Tensor(L).uniform_(-init, init))
def forward(self, scores, targets):
normalizers = self.compute_normalizers(scores)
target_scores = self.score_targets(scores, targets)
loss = (normalizers - target_scores).mean()
return loss
def decode(self, scores):
_B, T, _L = scores.size()
prev = self.start.unsqueeze(0) + scores[:, 0]
back = []
for i in range(1, T):
cur = prev.unsqueeze(2) + scores.transpose(0, 1)[i].unsqueeze(1
) + self.T.transpose(0, 1)
prev, indices = cur.max(dim=1)
back.append(indices)
prev += self.end
max_scores, indices = prev.max(dim=1)
tape = [indices]
back = list(reversed(back))
for i in range(T - 1):
indices = torch.gather(back[i], 1, indices.unsqueeze(1)).squeeze(1)
tape.append(indices)
return max_scores, torch.stack(tape[::-1], dim=1)
def compute_normalizers(self, scores):
_B, T, _L = scores.size()
prev = self.start + scores.transpose(0, 1)[0]
for i in range(1, T):
cur = prev.unsqueeze(2) + scores.transpose(0, 1)[i].unsqueeze(1
) + self.T.transpose(0, 1)
prev = torch.logsumexp(cur, dim=1).clone()
prev += self.end
normalizers = torch.logsumexp(prev, 1)
return normalizers
def score_targets(self, scores, targets):
_B, T, _L = scores.size()
emits = scores.gather(2, targets.unsqueeze(2)).squeeze(2).sum(1)
trans = torch.stack([self.start.gather(0, targets[:, 0])] + [self.T
[targets[:, i], targets[:, i - 1]] for i in range(1, T)] + [
self.end.gather(0, targets[:, -1])]).sum(0)
return emits + trans
def get_inputs():
return [torch.ones([4, 4, 4], dtype=torch.int64), torch.ones([4, 4],
dtype=torch.int64)]
def get_init_inputs():
return [[], {'L': 4, 'init': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn 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_add_logsumexp_0(in_ptr0, in_ptr1, in_ptr2, 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
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK])
tmp2 = tl.load(in_ptr1 + 16 * x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (4 + x0 + 16 * x1), xmask)
tmp8 = tl.load(in_ptr2 + 4 * x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr0 + 1)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp12 = tl.load(in_ptr1 + (1 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr2 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp19 = tl.load(in_ptr0 + 2)
tmp20 = tl.broadcast_to(tmp19, [XBLOCK])
tmp21 = tl.load(in_ptr1 + (2 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr2 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp28 = tl.load(in_ptr0 + 3)
tmp29 = tl.broadcast_to(tmp28, [XBLOCK])
tmp30 = tl.load(in_ptr1 + (3 + 16 * x1), xmask, eviction_policy=
'evict_last')
tmp34 = tl.load(in_ptr2 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp3 = tmp2.to(tl.float32)
tmp4 = tmp1 + tmp3
tmp6 = tmp5.to(tl.float32)
tmp7 = tmp4 + tmp6
tmp9 = tmp7 + tmp8
tmp13 = tmp12.to(tl.float32)
tmp14 = tmp11 + tmp13
tmp15 = tmp14 + tmp6
tmp17 = tmp15 + tmp16
tmp18 = triton_helpers.maximum(tmp9, tmp17)
tmp22 = tmp21.to(tl.float32)
tmp23 = tmp20 + tmp22
tmp24 = tmp23 + tmp6
tmp26 = tmp24 + tmp25
tmp27 = triton_helpers.maximum(tmp18, tmp26)
tmp31 = tmp30.to(tl.float32)
tmp32 = tmp29 + tmp31
tmp33 = tmp32 + tmp6
tmp35 = tmp33 + tmp34
tmp36 = triton_helpers.maximum(tmp27, tmp35)
tmp37 = tl_math.abs(tmp36)
tmp38 = float('inf')
tmp39 = tmp37 == tmp38
tmp40 = 0.0
tmp41 = tl.where(tmp39, tmp40, tmp36)
tmp42 = tmp9 - tmp41
tmp43 = tl_math.exp(tmp42)
tmp44 = tmp17 - tmp41
tmp45 = tl_math.exp(tmp44)
tmp46 = tmp43 + tmp45
tmp47 = tmp26 - tmp41
tmp48 = tl_math.exp(tmp47)
tmp49 = tmp46 + tmp48
tmp50 = tmp35 - tmp41
tmp51 = tl_math.exp(tmp50)
tmp52 = tmp49 + tmp51
tl.store(out_ptr0 + x2, tmp36, xmask)
tl.store(out_ptr1 + x2, tmp52, xmask)
@triton.jit
def triton_poi_fused_add_logsumexp_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
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
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (8 + x0 + 16 * x1), xmask)
tmp12 = tl.load(in_ptr3 + 4 * x0, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp16 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp22 = tl.load(in_ptr3 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp27 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp33 = tl.load(in_ptr3 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp36 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp38 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp44 = tl.load(in_ptr3 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = tl_math.log(tmp0)
tmp3 = tl_math.abs(tmp2)
tmp4 = float('inf')
tmp5 = tmp3 == tmp4
tmp6 = 0.0
tmp7 = tl.where(tmp5, tmp6, tmp2)
tmp8 = tmp1 + tmp7
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tl_math.log(tmp14)
tmp17 = tl_math.abs(tmp16)
tmp18 = tmp17 == tmp4
tmp19 = tl.where(tmp18, tmp6, tmp16)
tmp20 = tmp15 + tmp19
tmp21 = tmp20 + tmp10
tmp23 = tmp21 + tmp22
tmp24 = triton_helpers.maximum(tmp13, tmp23)
tmp26 = tl_math.log(tmp25)
tmp28 = tl_math.abs(tmp27)
tmp29 = tmp28 == tmp4
tmp30 = tl.where(tmp29, tmp6, tmp27)
tmp31 = tmp26 + tmp30
tmp32 = tmp31 + tmp10
tmp34 = tmp32 + tmp33
tmp35 = triton_helpers.maximum(tmp24, tmp34)
tmp37 = tl_math.log(tmp36)
tmp39 = tl_math.abs(tmp38)
tmp40 = tmp39 == tmp4
tmp41 = tl.where(tmp40, tmp6, tmp38)
tmp42 = tmp37 + tmp41
tmp43 = tmp42 + tmp10
tmp45 = tmp43 + tmp44
tmp46 = triton_helpers.maximum(tmp35, tmp45)
tmp47 = tl_math.abs(tmp46)
tmp48 = tmp47 == tmp4
tmp49 = tl.where(tmp48, tmp6, tmp46)
tmp50 = tmp13 - tmp49
tmp51 = tl_math.exp(tmp50)
tmp52 = tmp23 - tmp49
tmp53 = tl_math.exp(tmp52)
tmp54 = tmp51 + tmp53
tmp55 = tmp34 - tmp49
tmp56 = tl_math.exp(tmp55)
tmp57 = tmp54 + tmp56
tmp58 = tmp45 - tmp49
tmp59 = tl_math.exp(tmp58)
tmp60 = tmp57 + tmp59
tl.store(out_ptr0 + x2, tmp46, xmask)
tl.store(out_ptr1 + x2, tmp60, xmask)
@triton.jit
def triton_poi_fused_add_logsumexp_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
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
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr2 + (12 + x0 + 16 * x1), xmask)
tmp12 = tl.load(in_ptr3 + 4 * x0, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp16 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp22 = tl.load(in_ptr3 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp27 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp33 = tl.load(in_ptr3 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp36 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp38 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp44 = tl.load(in_ptr3 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = tl_math.log(tmp0)
tmp3 = tl_math.abs(tmp2)
tmp4 = float('inf')
tmp5 = tmp3 == tmp4
tmp6 = 0.0
tmp7 = tl.where(tmp5, tmp6, tmp2)
tmp8 = tmp1 + tmp7
tmp10 = tmp9.to(tl.float32)
tmp11 = tmp8 + tmp10
tmp13 = tmp11 + tmp12
tmp15 = tl_math.log(tmp14)
tmp17 = tl_math.abs(tmp16)
tmp18 = tmp17 == tmp4
tmp19 = tl.where(tmp18, tmp6, tmp16)
tmp20 = tmp15 + tmp19
tmp21 = tmp20 + tmp10
tmp23 = tmp21 + tmp22
tmp24 = triton_helpers.maximum(tmp13, tmp23)
tmp26 = tl_math.log(tmp25)
tmp28 = tl_math.abs(tmp27)
tmp29 = tmp28 == tmp4
tmp30 = tl.where(tmp29, tmp6, tmp27)
tmp31 = tmp26 + tmp30
tmp32 = tmp31 + tmp10
tmp34 = tmp32 + tmp33
tmp35 = triton_helpers.maximum(tmp24, tmp34)
tmp37 = tl_math.log(tmp36)
tmp39 = tl_math.abs(tmp38)
tmp40 = tmp39 == tmp4
tmp41 = tl.where(tmp40, tmp6, tmp38)
tmp42 = tmp37 + tmp41
tmp43 = tmp42 + tmp10
tmp45 = tmp43 + tmp44
tmp46 = triton_helpers.maximum(tmp35, tmp45)
tmp47 = tl_math.abs(tmp46)
tmp48 = tmp47 == tmp4
tmp49 = tl.where(tmp48, tmp6, tmp46)
tmp50 = tmp13 - tmp49
tmp51 = tl_math.exp(tmp50)
tmp52 = tmp23 - tmp49
tmp53 = tl_math.exp(tmp52)
tmp54 = tmp51 + tmp53
tmp55 = tmp34 - tmp49
tmp56 = tl_math.exp(tmp55)
tmp57 = tmp54 + tmp56
tmp58 = tmp45 - tmp49
tmp59 = tl_math.exp(tmp58)
tmp60 = tmp57 + tmp59
tl.store(out_ptr0 + x2, tmp46, xmask)
tl.store(out_ptr1 + x2, tmp60, xmask)
@triton.jit
def triton_poi_fused_stack_3(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 20
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + 4 * x0, tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tl.full([XBLOCK], 4, tl.int32)
tmp7 = tmp5 + tmp6
tmp8 = tmp5 < 0
tmp9 = tl.where(tmp8, tmp7, tmp5)
tl.device_assert((0 <= tl.broadcast_to(tmp9, [XBLOCK])) & (tl.
broadcast_to(tmp9, [XBLOCK]) < 4) | ~(tmp4 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp9, [XBLOCK]) < 4')
tmp11 = tl.load(in_ptr1 + tl.broadcast_to(tmp9, [XBLOCK]), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp12 = tmp0 >= tmp3
tmp13 = tl.full([1], 8, tl.int64)
tmp14 = tmp0 < tmp13
tmp15 = tmp12 & tmp14
tmp16 = tl.load(in_ptr0 + (1 + 4 * (-4 + x0)), tmp15 & xmask,
eviction_policy='evict_last', other=0.0)
tmp17 = tmp16 + tmp6
tmp18 = tmp16 < 0
tmp19 = tl.where(tmp18, tmp17, tmp16)
tl.device_assert((0 <= tl.broadcast_to(tmp19, [XBLOCK])) & (tl.
broadcast_to(tmp19, [XBLOCK]) < 4) | ~(tmp15 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp19, [XBLOCK]) < 4')
tmp21 = tl.load(in_ptr0 + 4 * (-4 + x0), tmp15 & xmask, eviction_policy
='evict_last', other=0.0)
tmp22 = tmp21 + tmp6
tmp23 = tmp21 < 0
tmp24 = tl.where(tmp23, tmp22, tmp21)
tl.device_assert((0 <= tl.broadcast_to(tmp24, [XBLOCK])) & (tl.
broadcast_to(tmp24, [XBLOCK]) < 4) | ~(tmp15 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp24, [XBLOCK]) < 4')
tmp26 = tl.load(in_ptr2 + tl.broadcast_to(tmp24 + 4 * tmp19, [XBLOCK]),
tmp15 & xmask, eviction_policy='evict_last', other=0.0)
tmp27 = tmp0 >= tmp13
tmp28 = tl.full([1], 12, tl.int64)
tmp29 = tmp0 < tmp28
tmp30 = tmp27 & tmp29
tmp31 = tl.load(in_ptr0 + (2 + 4 * (-8 + x0)), tmp30 & xmask,
eviction_policy='evict_last', other=0.0)
tmp32 = tmp31 + tmp6
tmp33 = tmp31 < 0
tmp34 = tl.where(tmp33, tmp32, tmp31)
tl.device_assert((0 <= tl.broadcast_to(tmp34, [XBLOCK])) & (tl.
broadcast_to(tmp34, [XBLOCK]) < 4) | ~(tmp30 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp34, [XBLOCK]) < 4')
tmp36 = tl.load(in_ptr0 + (1 + 4 * (-8 + x0)), tmp30 & xmask,
eviction_policy='evict_last', other=0.0)
tmp37 = tmp36 + tmp6
tmp38 = tmp36 < 0
tmp39 = tl.where(tmp38, tmp37, tmp36)
tl.device_assert((0 <= tl.broadcast_to(tmp39, [XBLOCK])) & (tl.
broadcast_to(tmp39, [XBLOCK]) < 4) | ~(tmp30 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp39, [XBLOCK]) < 4')
tmp41 = tl.load(in_ptr2 + tl.broadcast_to(tmp39 + 4 * tmp34, [XBLOCK]),
tmp30 & xmask, eviction_policy='evict_last', other=0.0)
tmp42 = tmp0 >= tmp28
tmp43 = tl.full([1], 16, tl.int64)
tmp44 = tmp0 < tmp43
tmp45 = tmp42 & tmp44
tmp46 = tl.load(in_ptr0 + (3 + 4 * (-12 + x0)), tmp45 & xmask,
eviction_policy='evict_last', other=0.0)
tmp47 = tmp46 + tmp6
tmp48 = tmp46 < 0
tmp49 = tl.where(tmp48, tmp47, tmp46)
tl.device_assert((0 <= tl.broadcast_to(tmp49, [XBLOCK])) & (tl.
broadcast_to(tmp49, [XBLOCK]) < 4) | ~(tmp45 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp49, [XBLOCK]) < 4')
tmp51 = tl.load(in_ptr0 + (2 + 4 * (-12 + x0)), tmp45 & xmask,
eviction_policy='evict_last', other=0.0)
tmp52 = tmp51 + tmp6
tmp53 = tmp51 < 0
tmp54 = tl.where(tmp53, tmp52, tmp51)
tl.device_assert((0 <= tl.broadcast_to(tmp54, [XBLOCK])) & (tl.
broadcast_to(tmp54, [XBLOCK]) < 4) | ~(tmp45 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp54, [XBLOCK]) < 4')
tmp56 = tl.load(in_ptr2 + tl.broadcast_to(tmp54 + 4 * tmp49, [XBLOCK]),
tmp45 & xmask, eviction_policy='evict_last', other=0.0)
tmp57 = tmp0 >= tmp43
tl.full([1], 20, tl.int64)
tmp60 = tl.load(in_ptr0 + (3 + 4 * (-16 + x0)), tmp57 & xmask,
eviction_policy='evict_last', other=0.0)
tmp61 = tmp60 + tmp6
tmp62 = tmp60 < 0
tmp63 = tl.where(tmp62, tmp61, tmp60)
tl.device_assert((0 <= tl.broadcast_to(tmp63, [XBLOCK])) & (tl.
broadcast_to(tmp63, [XBLOCK]) < 4) | ~(tmp57 & xmask),
'index out of bounds: 0 <= tl.broadcast_to(tmp63, [XBLOCK]) < 4')
tmp65 = tl.load(in_ptr3 + tl.broadcast_to(tmp63, [XBLOCK]), tmp57 &
xmask, eviction_policy='evict_last', other=0.0)
tmp66 = tl.where(tmp45, tmp56, tmp65)
tmp67 = tl.where(tmp30, tmp41, tmp66)
tmp68 = tl.where(tmp15, tmp26, tmp67)
tmp69 = tl.where(tmp4, tmp11, tmp68)
tl.store(out_ptr0 + x0, tmp69, xmask)
@triton.jit
def triton_per_fused_add_logsumexp_mean_sub_sum_4(in_out_ptr0, in_ptr0,
in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, 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')
tmp7 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr2 + r0, None)
tmp30 = tl.load(in_ptr2 + (4 + r0), None)
tmp32 = tl.load(in_ptr2 + (8 + r0), None)
tmp34 = tl.load(in_ptr2 + (12 + r0), None)
tmp36 = tl.load(in_ptr2 + (16 + r0), None)
tmp39 = tl.load(in_ptr3 + 4 * r0, None, eviction_policy='evict_last')
tmp41 = tl.load(in_ptr4 + 4 * r0, None, eviction_policy='evict_last')
tmp48 = tl.load(in_ptr5 + 0)
tmp49 = tl.broadcast_to(tmp48, [XBLOCK, RBLOCK])
tmp51 = tl.load(in_ptr3 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp53 = tl.load(in_ptr4 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp58 = tl.load(in_ptr5 + 1)
tmp59 = tl.broadcast_to(tmp58, [XBLOCK, RBLOCK])
tmp62 = tl.load(in_ptr3 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp64 = tl.load(in_ptr4 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp69 = tl.load(in_ptr5 + 2)
tmp70 = tl.broadcast_to(tmp69, [XBLOCK, RBLOCK])
tmp73 = tl.load(in_ptr3 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp75 = tl.load(in_ptr4 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp80 = tl.load(in_ptr5 + 3)
tmp81 = tl.broadcast_to(tmp80, [XBLOCK, RBLOCK])
tmp1 = tl.full([XBLOCK, RBLOCK], 4, tl.int32)
tmp2 = tmp0 + tmp1
tmp3 = tmp0 < 0
tmp4 = tl.where(tmp3, tmp2, tmp0)
tl.device_assert((0 <= tmp4) & (tmp4 < 4),
'index out of bounds: 0 <= tmp4 < 4')
tmp6 = tl.load(in_ptr1 + (tmp4 + 16 * r0), None, eviction_policy=
'evict_last')
tmp8 = tmp7 + tmp1
tmp9 = tmp7 < 0
tmp10 = tl.where(tmp9, tmp8, tmp7)
tl.device_assert((0 <= tmp10) & (tmp10 < 4),
'index out of bounds: 0 <= tmp10 < 4')
tmp12 = tl.load(in_ptr1 + (4 + tmp10 + 16 * r0), None, eviction_policy=
'evict_last')
tmp13 = tmp6 + tmp12
tmp15 = tmp14 + tmp1
tmp16 = tmp14 < 0
tmp17 = tl.where(tmp16, tmp15, tmp14)
tl.device_assert((0 <= tmp17) & (tmp17 < 4),
'index out of bounds: 0 <= tmp17 < 4')
tmp19 = tl.load(in_ptr1 + (8 + tmp17 + 16 * r0), None, eviction_policy=
'evict_last')
tmp20 = tmp13 + tmp19
tmp22 = tmp21 + tmp1
tmp23 = tmp21 < 0
tmp24 = tl.where(tmp23, tmp22, tmp21)
tl.device_assert((0 <= tmp24) & (tmp24 < 4),
'index out of bounds: 0 <= tmp24 < 4')
tmp26 = tl.load(in_ptr1 + (12 + tmp24 + 16 * r0), None, eviction_policy
='evict_last')
tmp27 = tmp20 + tmp26
tmp28 = tmp27.to(tl.float32)
tmp31 = tmp29 + tmp30
tmp33 = tmp31 + tmp32
tmp35 = tmp33 + tmp34
tmp37 = tmp35 + tmp36
tmp38 = tmp28 + tmp37
tmp40 = tl_math.log(tmp39)
tmp42 = tl_math.abs(tmp41)
tmp43 = float('inf')
tmp44 = tmp42 == tmp43
tmp45 = 0.0
tmp46 = tl.where(tmp44, tmp45, tmp41)
tmp47 = tmp40 + tmp46
tmp50 = tmp47 + tmp49
tmp52 = tl_math.log(tmp51)
tmp54 = tl_math.abs(tmp53)
tmp55 = tmp54 == tmp43
tmp56 = tl.where(tmp55, tmp45, tmp53)
tmp57 = tmp52 + tmp56
tmp60 = tmp57 + tmp59
tmp61 = triton_helpers.maximum(tmp50, tmp60)
tmp63 = tl_math.log(tmp62)
tmp65 = tl_math.abs(tmp64)
tmp66 = tmp65 == tmp43
tmp67 = tl.where(tmp66, tmp45, tmp64)
tmp68 = tmp63 + tmp67
tmp71 = tmp68 + tmp70
tmp72 = triton_helpers.maximum(tmp61, tmp71)
tmp74 = tl_math.log(tmp73)
tmp76 = tl_math.abs(tmp75)
tmp77 = tmp76 == tmp43
tmp78 = tl.where(tmp77, tmp45, tmp75)
tmp79 = tmp74 + tmp78
tmp82 = tmp79 + tmp81
tmp83 = triton_helpers.maximum(tmp72, tmp82)
tmp84 = tl_math.abs(tmp83)
tmp85 = tmp84 == tmp43
tmp86 = tl.where(tmp85, tmp45, tmp83)
tmp87 = tmp50 - tmp86
tmp88 = tl_math.exp(tmp87)
tmp89 = tmp60 - tmp86
tmp90 = tl_math.exp(tmp89)
tmp91 = tmp88 + tmp90
tmp92 = tmp71 - tmp86
tmp93 = tl_math.exp(tmp92)
tmp94 = tmp91 + tmp93
tmp95 = tmp82 - tmp86
tmp96 = tl_math.exp(tmp95)
tmp97 = tmp94 + tmp96
tmp98 = tl_math.log(tmp97)
tmp99 = tmp98 + tmp86
tmp100 = tmp99 - tmp38
tmp101 = tl.broadcast_to(tmp100, [XBLOCK, RBLOCK])
tmp103 = tl.sum(tmp101, 1)[:, None]
tmp104 = 4.0
tmp105 = tmp103 / tmp104
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp105, None)
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,), (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))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_logsumexp_0[grid(16)](primals_2, primals_1,
primals_3, buf0, buf1, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_logsumexp_1[grid(16)](buf1, buf0, primals_1,
primals_3, buf2, buf3, 16, XBLOCK=16, num_warps=1, num_stages=1)
buf4 = reinterpret_tensor(buf1, (4, 1, 4), (4, 16, 1), 0)
del buf1
buf5 = reinterpret_tensor(buf0, (4, 4), (4, 1), 0)
del buf0
triton_poi_fused_add_logsumexp_2[grid(16)](buf3, buf2, primals_1,
primals_3, buf4, buf5, 16, XBLOCK=16, num_warps=1, num_stages=1)
del buf2
del buf3
buf8 = empty_strided_cuda((20,), (1,), torch.float32)
triton_poi_fused_stack_3[grid(20)](primals_5, primals_2, primals_3,
primals_4, buf8, 20, XBLOCK=32, num_warps=1, num_stages=1)
buf10 = empty_strided_cuda((), (), torch.float32)
buf11 = buf10
del buf10
triton_per_fused_add_logsumexp_mean_sub_sum_4[grid(1)](buf11,
primals_5, primals_1, buf8, buf5, buf4, primals_4, 1, 4, XBLOCK
=1, num_warps=2, num_stages=1)
del buf4
del buf5
del buf8
return (buf11, primals_1, primals_2, primals_3, primals_4,
reinterpret_tensor(primals_5, (4,), (4,), 0), reinterpret_tensor(
primals_5, (4,), (4,), 1), reinterpret_tensor(primals_5, (4,), (4,),
2), reinterpret_tensor(primals_5, (4,), (4,), 3),
reinterpret_tensor(primals_5, (4,), (4,), 3))
class CRFLossNew(nn.Module):
def __init__(self, L, init):
super(CRFLossNew, self).__init__()
self.start = nn.Parameter(torch.Tensor(L).uniform_(-init, init))
self.T = nn.Parameter(torch.Tensor(L, L).uniform_(-init, init))
self.end = nn.Parameter(torch.Tensor(L).uniform_(-init, init))
def decode(self, scores):
_B, T, _L = scores.size()
prev = self.start.unsqueeze(0) + scores[:, 0]
back = []
for i in range(1, T):
cur = prev.unsqueeze(2) + scores.transpose(0, 1)[i].unsqueeze(1
) + self.T.transpose(0, 1)
prev, indices = cur.max(dim=1)
back.append(indices)
prev += self.end
max_scores, indices = prev.max(dim=1)
tape = [indices]
back = list(reversed(back))
for i in range(T - 1):
indices = torch.gather(back[i], 1, indices.unsqueeze(1)).squeeze(1)
tape.append(indices)
return max_scores, torch.stack(tape[::-1], dim=1)
def compute_normalizers(self, scores):
_B, T, _L = scores.size()
prev = self.start + scores.transpose(0, 1)[0]
for i in range(1, T):
cur = prev.unsqueeze(2) + scores.transpose(0, 1)[i].unsqueeze(1
) + self.T.transpose(0, 1)
prev = torch.logsumexp(cur, dim=1).clone()
prev += self.end
normalizers = torch.logsumexp(prev, 1)
return normalizers
def score_targets(self, scores, targets):
_B, T, _L = scores.size()
emits = scores.gather(2, targets.unsqueeze(2)).squeeze(2).sum(1)
trans = torch.stack([self.start.gather(0, targets[:, 0])] + [self.T
[targets[:, i], targets[:, i - 1]] for i in range(1, T)] + [
self.end.gather(0, targets[:, -1])]).sum(0)
return emits + trans
def forward(self, input_0, input_1):
primals_2 = self.start
primals_3 = self.T
primals_4 = self.end
primals_1 = input_0
primals_5 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
Johannes0Horn/mtl-dts
|
CRFLoss
| false
| 8,362
|
[
"MIT"
] | 19
|
ae50253c808bbb77af3b1117f69f08d2268099e9
|
https://github.com/Johannes0Horn/mtl-dts/tree/ae50253c808bbb77af3b1117f69f08d2268099e9
|
NonLocalBlock
|
# 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/op/copz2tw3m6344xrasrh5hzk6yutylalzjzib33mb27nfr4jplthm.py
# Topologically Sorted Source Nodes: [x_theta], Original ATen: [aten.clone, aten.transpose]
# Source node to ATen node mapping:
# x_theta => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute,), kwargs = {memory_format: torch.contiguous_format})
# %permute_6 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%expand, [0, 2, 1]), kwargs = {})
triton_poi_fused_clone_transpose_0 = async_compile.triton('triton_poi_fused_clone_transpose_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 8], 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_transpose_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_clone_transpose_0(in_out_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 8
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
y2 = yindex % 4
y3 = (yindex // 4)
tmp0 = tl.load(in_out_ptr0 + (x1 + (8*y0)), xmask & ymask, eviction_policy='evict_last')
tl.debug_barrier()
tl.store(in_out_ptr0 + (x1 + (8*y0)), tmp0, xmask & ymask)
tl.store(out_ptr0 + (y2 + (4*x1) + (32*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/tn/ctnh7fhjwh7vjosr4dycfzlc5g2xwfxmpyix66g5k5by26kf46z5.py
# Topologically Sorted Source Nodes: [mul_theta_phi_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# mul_theta_phi_1 => amax, exp, sub, sum_1
# 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 = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
triton_per_fused__softmax_1 = async_compile.triton('triton_per_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.persistent_reduction(
size_hints=[32, 8],
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': {}, 'configs': [AttrsDescriptor(divisible_by_16=(0, 1, 2, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused__softmax_1', '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_1(in_ptr0, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 32
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)
r2 = rindex
x0 = xindex % 8
x1 = (xindex // 8)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (8*r2) + (64*x1)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, float("-inf"))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp5 = tmp0 - tmp4
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = tl.where(xmask, tmp7, 0)
tmp10 = tl.sum(tmp9, 1)[:, None]
tl.store(out_ptr0 + (x3), tmp4, xmask)
tl.store(out_ptr1 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/oy/coyiujg4k3scmhojckibo5zfgl772p5irbnhsc5zsdnvzc674zyb.py
# Topologically Sorted Source Nodes: [mul_theta_phi_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# mul_theta_phi_1 => div, exp, sub
# Graph fragment:
# %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 = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': 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_2', '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__softmax_2(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
x3 = xindex
x0 = xindex % 8
x2 = (xindex // 64)
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (8*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (x0 + (8*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tl.store(in_out_ptr0 + (x3), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wt/cwtcww3tbo4exhrcd7ybuajqbsewmltofgu5pmc6bbds7mr6iw7j.py
# Topologically Sorted Source Nodes: [contiguous_2, mul_theta_phi_g_1], Original ATen: [aten.clone, aten.view]
# Source node to ATen node mapping:
# contiguous_2 => clone_2
# mul_theta_phi_g_1 => view_9
# Graph fragment:
# %clone_2 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_2,), kwargs = {memory_format: torch.contiguous_format})
# %view_9 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%clone_2, [4, 2, 4, 4]), kwargs = {})
triton_poi_fused_clone_view_3 = async_compile.triton('triton_poi_fused_clone_view_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, 8], 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_view_3', '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_clone_view_3(in_out_ptr0, in_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 8
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) + (32*y1)), xmask & ymask, eviction_policy='evict_last')
tl.debug_barrier()
tl.store(in_out_ptr0 + (x2 + (8*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/3a/c3afomcyr2z37ah5vsar7fpsgyizsx3rpx5pu7cc7lh3rlyhy7c2.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.add]
# Source node to ATen node mapping:
# out => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_3, %primals_1), 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=[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_4', '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_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
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, (2, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (2, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_4, (2, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4, 2, 1, 1), (2, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 2, 4, 4), (32, 16, 4, 1))
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(primals_1, primals_3, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 2, 4, 4), (32, 16, 4, 1))
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(primals_1, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 2, 4, 4), (32, 16, 4, 1))
buf3 = reinterpret_tensor(buf1, (4, 8, 4), (32, 1, 8), 0); del buf1 # reuse
buf15 = empty_strided_cuda((4, 4, 8), (32, 1, 4), torch.float32)
# Topologically Sorted Source Nodes: [x_theta], Original ATen: [aten.clone, aten.transpose]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_transpose_0.run(buf3, buf15, 16, 8, grid=grid(16, 8), stream=stream0)
buf4 = empty_strided_cuda((4, 8, 8), (64, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_theta, mul_theta_phi], Original ATen: [aten.clone, aten.bmm]
extern_kernels.bmm(buf3, reinterpret_tensor(buf0, (4, 4, 8), (32, 8, 1), 0), out=buf4)
buf5 = empty_strided_cuda((4, 1, 8), (8, 32, 1), torch.float32)
buf6 = empty_strided_cuda((4, 1, 8), (8, 32, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_theta_phi_1], Original ATen: [aten._softmax]
triton_per_fused__softmax_1.run(buf4, buf5, buf6, 32, 8, grid=grid(32), stream=stream0)
buf7 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [mul_theta_phi_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf7, buf5, buf6, 256, grid=grid(256), stream=stream0)
del buf5
del buf6
buf8 = reinterpret_tensor(buf2, (4, 8, 4), (32, 1, 8), 0); del buf2 # reuse
buf14 = reinterpret_tensor(buf3, (4, 4, 8), (32, 1, 4), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [x_g], Original ATen: [aten.clone, aten.transpose]
triton_poi_fused_clone_transpose_0.run(buf8, buf14, 16, 8, grid=grid(16, 8), stream=stream0)
buf9 = empty_strided_cuda((4, 8, 4), (32, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_g, mul_theta_phi_g], Original ATen: [aten.clone, aten.bmm]
extern_kernels.bmm(buf7, buf8, out=buf9)
buf10 = reinterpret_tensor(buf8, (4, 4, 8), (32, 8, 1), 0); del buf8 # reuse
buf11 = reinterpret_tensor(buf10, (4, 2, 4, 4), (32, 16, 4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [contiguous_2, mul_theta_phi_g_1], Original ATen: [aten.clone, aten.view]
triton_poi_fused_clone_view_3.run(buf11, buf9, 16, 8, grid=grid(16, 8), stream=stream0)
del buf9
# Topologically Sorted Source Nodes: [mask], Original ATen: [aten.convolution]
buf12 = extern_kernels.convolution(buf11, primals_5, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 4, 4, 4), (64, 16, 4, 1))
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.add]
triton_poi_fused_add_4.run(buf13, primals_1, 256, grid=grid(256), stream=stream0)
return (buf13, primals_1, primals_2, primals_3, primals_4, primals_5, buf7, buf11, buf14, buf15, reinterpret_tensor(buf0, (4, 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, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((2, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((2, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((2, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 2, 1, 1), (2, 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)
|
import torch
import torch.nn as nn
from time import *
class NonLocalBlock(nn.Module):
def __init__(self, channel):
super(NonLocalBlock, self).__init__()
self.inter_channel = channel // 2
self.conv_phi = nn.Conv2d(in_channels=channel, out_channels=self.
inter_channel, kernel_size=1, stride=1, padding=0, bias=False)
self.conv_theta = nn.Conv2d(in_channels=channel, out_channels=self.
inter_channel, kernel_size=1, stride=1, padding=0, bias=False)
self.conv_g = nn.Conv2d(in_channels=channel, out_channels=self.
inter_channel, kernel_size=1, stride=1, padding=0, bias=False)
self.softmax = nn.Softmax(dim=1)
self.conv_mask = nn.Conv2d(in_channels=self.inter_channel,
out_channels=channel, kernel_size=1, stride=1, padding=0, bias=
False)
def forward(self, x):
b, c, h, w = x.size()
x_phi = self.conv_phi(x).view(b, c, -1)
x_theta = self.conv_theta(x).view(b, c, -1).permute(0, 2, 1
).contiguous()
x_g = self.conv_g(x).view(b, c, -1).permute(0, 2, 1).contiguous()
mul_theta_phi = torch.matmul(x_theta, x_phi)
mul_theta_phi = self.softmax(mul_theta_phi)
mul_theta_phi_g = torch.matmul(mul_theta_phi, x_g)
mul_theta_phi_g = mul_theta_phi_g.permute(0, 2, 1).contiguous().view(b,
self.inter_channel, h, w)
mask = self.conv_mask(mul_theta_phi_g)
out = mask + x
return out
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._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
from time import *
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_transpose_0(in_out_ptr0, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 8
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
y2 = yindex % 4
y3 = yindex // 4
tmp0 = tl.load(in_out_ptr0 + (x1 + 8 * y0), xmask & ymask,
eviction_policy='evict_last')
tl.debug_barrier()
tl.store(in_out_ptr0 + (x1 + 8 * y0), tmp0, xmask & ymask)
tl.store(out_ptr0 + (y2 + 4 * x1 + 32 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_per_fused__softmax_1(in_ptr0, out_ptr0, out_ptr1, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 32
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)
r2 = rindex
x0 = xindex % 8
x1 = xindex // 8
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 8 * r2 + 64 * x1), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, float('-inf'))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp5 = tmp0 - tmp4
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = tl.where(xmask, tmp7, 0)
tmp10 = tl.sum(tmp9, 1)[:, None]
tl.store(out_ptr0 + x3, tmp4, xmask)
tl.store(out_ptr1 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused__softmax_2(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
x3 = xindex
x0 = xindex % 8
x2 = xindex // 64
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 8 * x2), xmask, eviction_policy='evict_last'
)
tmp4 = tl.load(in_ptr1 + (x0 + 8 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tl.store(in_out_ptr0 + x3, tmp5, xmask)
@triton.jit
def triton_poi_fused_clone_view_3(in_out_ptr0, in_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 8
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 + 32 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.debug_barrier()
tl.store(in_out_ptr0 + (x2 + 8 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
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, (2, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (2, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_4, (2, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_5, (4, 2, 1, 1), (2, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 2, 4, 4), (32, 16, 4, 1))
buf1 = extern_kernels.convolution(primals_1, primals_3, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 2, 4, 4), (32, 16, 4, 1))
buf2 = extern_kernels.convolution(primals_1, primals_4, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 2, 4, 4), (32, 16, 4, 1))
buf3 = reinterpret_tensor(buf1, (4, 8, 4), (32, 1, 8), 0)
del buf1
buf15 = empty_strided_cuda((4, 4, 8), (32, 1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_transpose_0[grid(16, 8)](buf3, buf15, 16, 8,
XBLOCK=8, YBLOCK=16, num_warps=4, num_stages=1)
buf4 = empty_strided_cuda((4, 8, 8), (64, 8, 1), torch.float32)
extern_kernels.bmm(buf3, reinterpret_tensor(buf0, (4, 4, 8), (32, 8,
1), 0), out=buf4)
buf5 = empty_strided_cuda((4, 1, 8), (8, 32, 1), torch.float32)
buf6 = empty_strided_cuda((4, 1, 8), (8, 32, 1), torch.float32)
triton_per_fused__softmax_1[grid(32)](buf4, buf5, buf6, 32, 8,
XBLOCK=1, num_warps=2, num_stages=1)
buf7 = buf4
del buf4
triton_poi_fused__softmax_2[grid(256)](buf7, buf5, buf6, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del buf5
del buf6
buf8 = reinterpret_tensor(buf2, (4, 8, 4), (32, 1, 8), 0)
del buf2
buf14 = reinterpret_tensor(buf3, (4, 4, 8), (32, 1, 4), 0)
del buf3
triton_poi_fused_clone_transpose_0[grid(16, 8)](buf8, buf14, 16, 8,
XBLOCK=8, YBLOCK=16, num_warps=4, num_stages=1)
buf9 = empty_strided_cuda((4, 8, 4), (32, 4, 1), torch.float32)
extern_kernels.bmm(buf7, buf8, out=buf9)
buf10 = reinterpret_tensor(buf8, (4, 4, 8), (32, 8, 1), 0)
del buf8
buf11 = reinterpret_tensor(buf10, (4, 2, 4, 4), (32, 16, 4, 1), 0)
del buf10
triton_poi_fused_clone_view_3[grid(16, 8)](buf11, buf9, 16, 8,
XBLOCK=8, YBLOCK=16, num_warps=4, num_stages=1)
del buf9
buf12 = extern_kernels.convolution(buf11, primals_5, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 4, 4, 4), (64, 16, 4, 1))
buf13 = buf12
del buf12
triton_poi_fused_add_4[grid(256)](buf13, primals_1, 256, XBLOCK=256,
num_warps=4, num_stages=1)
return (buf13, primals_1, primals_2, primals_3, primals_4, primals_5,
buf7, buf11, buf14, buf15, reinterpret_tensor(buf0, (4, 8, 4), (32,
1, 8), 0))
class NonLocalBlockNew(nn.Module):
def __init__(self, channel):
super(NonLocalBlockNew, self).__init__()
self.inter_channel = channel // 2
self.conv_phi = nn.Conv2d(in_channels=channel, out_channels=self.
inter_channel, kernel_size=1, stride=1, padding=0, bias=False)
self.conv_theta = nn.Conv2d(in_channels=channel, out_channels=self.
inter_channel, kernel_size=1, stride=1, padding=0, bias=False)
self.conv_g = nn.Conv2d(in_channels=channel, out_channels=self.
inter_channel, kernel_size=1, stride=1, padding=0, bias=False)
self.softmax = nn.Softmax(dim=1)
self.conv_mask = nn.Conv2d(in_channels=self.inter_channel,
out_channels=channel, kernel_size=1, stride=1, padding=0, bias=
False)
def forward(self, input_0):
primals_2 = self.conv_phi.weight
primals_3 = self.conv_theta.weight
primals_4 = self.conv_g.weight
primals_5 = self.conv_mask.weight
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
Jinming-Su/SGNet
|
NonLocalBlock
| false
| 8,363
|
[
"MIT"
] | 13
|
fcf35edaf332c1a4e2713acad5a0fc0e21509c3e
|
https://github.com/Jinming-Su/SGNet/tree/fcf35edaf332c1a4e2713acad5a0fc0e21509c3e
|
SoftCrossEntropyLoss
|
# 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/ia/ciayqeqzx5aa7nfdwxamntj3iy4xmzooazzst6o6kdmglfu2uckl.py
# Topologically Sorted Source Nodes: [max_1, logit], Original ATen: [aten.max, aten.sub]
# Source node to ATen node mapping:
# logit => sub
# max_1 => max_1
# Graph fragment:
# %max_1 : [num_users=1] = call_function[target=torch.ops.aten.max.dim](args = (%arg0_1, 3, True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg0_1, %getitem), kwargs = {})
triton_poi_fused_max_sub_0 = async_compile.triton('triton_poi_fused_max_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_max_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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/66/c66il2i3eecgrwjjvtmjmavgkqkvozokg43embnqmual5yqwin4k.py
# Topologically Sorted Source Nodes: [exp_logit, exp_sum, log_exp_sum, neg_log_prob, mul, out, mean], Original ATen: [aten.exp, aten.sum, aten.log, aten.sub, aten.mul, aten.mean]
# Source node to ATen node mapping:
# exp_logit => exp
# exp_sum => sum_1
# log_exp_sum => log
# mean => mean
# mul => mul
# neg_log_prob => sub_1
# out => sum_2
# 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, [3], 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 = (%log, %sub), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %arg1_1), kwargs = {})
# %sum_2 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [3]), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sum_2,), kwargs = {})
triton_per_fused_exp_log_mean_mul_sub_sum_1 = async_compile.triton('triton_per_fused_exp_log_mean_mul_sub_sum_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[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_exp_log_mean_mul_sub_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_exp_log_mean_mul_sub_sum_1(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
tmp0 = tl.load(in_ptr0 + (4*r0), None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + (4*r0), None, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (1 + (4*r0)), None, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + (2 + (4*r0)), None, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr1 + (3 + (4*r0)), None, eviction_policy='evict_last')
tmp1 = tl_math.exp(tmp0)
tmp3 = tl_math.exp(tmp2)
tmp4 = tmp1 + tmp3
tmp6 = tl_math.exp(tmp5)
tmp7 = tmp4 + tmp6
tmp9 = tl_math.exp(tmp8)
tmp10 = tmp7 + tmp9
tmp11 = tl_math.log(tmp10)
tmp12 = tmp11 - tmp0
tmp14 = tmp12 * tmp13
tmp15 = tmp11 - tmp2
tmp17 = tmp15 * tmp16
tmp18 = tmp14 + tmp17
tmp19 = tmp11 - tmp5
tmp21 = tmp19 * tmp20
tmp22 = tmp18 + tmp21
tmp23 = tmp11 - tmp8
tmp25 = tmp23 * tmp24
tmp26 = tmp22 + tmp25
tmp27 = tl.broadcast_to(tmp26, [XBLOCK, RBLOCK])
tmp29 = tl.sum(tmp27, 1)[:, None]
tmp30 = 64.0
tmp31 = tmp29 / tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp31, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [max_1, logit], Original ATen: [aten.max, aten.sub]
stream0 = get_raw_stream(0)
triton_poi_fused_max_sub_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [exp_logit, exp_sum, log_exp_sum, neg_log_prob, mul, out, mean], Original ATen: [aten.exp, aten.sum, aten.log, aten.sub, aten.mul, aten.mean]
triton_per_fused_exp_log_mean_mul_sub_sum_1.run(buf2, buf0, arg1_1, 1, 64, grid=grid(1), stream=stream0)
del arg1_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)
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
def soft_cross_entropy(logit, label, weight=None, reduce=None, reduction='mean'
):
if weight is not None and weight.requires_grad:
raise RuntimeError('gradient for weight is not supported')
losses = SoftCrossEntropyFunction.apply(logit, label, weight)
reduction = {(True): 'mean', (False): 'none', None: reduction}[reduce]
if reduction == 'mean':
return losses.mean()
elif reduction == 'sum':
return losses.sum()
elif reduction == 'none':
return losses
else:
raise ValueError('invalid value for reduction: {}'.format(reduction))
class SoftCrossEntropyFunction(torch.autograd.Function):
@staticmethod
def forward(ctx, logit, label, weight=None):
assert logit.size() == label.size(), 'logit.size() != label.size()'
dim = logit.dim()
max_logit = logit.max(dim - 1, keepdim=True)[0]
logit = logit - max_logit
exp_logit = logit.exp()
exp_sum = exp_logit.sum(dim - 1, keepdim=True)
prob = exp_logit / exp_sum
log_exp_sum = exp_sum.log()
neg_log_prob = log_exp_sum - logit
if weight is None:
weighted_label = label
else:
if weight.size() != (logit.size(-1),):
raise ValueError(
'since logit.size() = {}, weight.size() should be ({},), but got {}'
.format(logit.size(), logit.size(-1), weight.size()))
size = [1] * label.dim()
size[-1] = label.size(-1)
weighted_label = label * weight.view(size)
ctx.save_for_backward(weighted_label, prob)
out = (neg_log_prob * weighted_label).sum(dim - 1)
return out
@staticmethod
def backward(ctx, grad_output):
weighted_label, prob = ctx.saved_tensors
old_size = weighted_label.size()
K = old_size[-1]
B = weighted_label.numel() // K
grad_output = grad_output.view(B, 1)
weighted_label = weighted_label.view(B, K)
prob = prob.view(B, K)
grad_input = grad_output * (prob * weighted_label.sum(1, True) -
weighted_label)
grad_input = grad_input.view(old_size)
return grad_input, None, None
class SoftCrossEntropyLoss(torch.nn.Module):
def __init__(self, weight=None, reduce=None, reduction='mean'):
super(SoftCrossEntropyLoss, self).__init__()
self.weight = weight
self.reduce = reduce
self.reduction = reduction
def forward(self, logit, label, weight=None):
if weight is None:
weight = self.weight
return soft_cross_entropy(logit, label, weight, self.reduce, self.
reduction)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_max_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
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_per_fused_exp_log_mean_mul_sub_sum_1(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
tmp0 = tl.load(in_ptr0 + 4 * r0, None, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr1 + 4 * r0, None, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (1 + 4 * r0), None, eviction_policy='evict_last')
tmp20 = tl.load(in_ptr1 + (2 + 4 * r0), None, eviction_policy='evict_last')
tmp24 = tl.load(in_ptr1 + (3 + 4 * r0), None, eviction_policy='evict_last')
tmp1 = tl_math.exp(tmp0)
tmp3 = tl_math.exp(tmp2)
tmp4 = tmp1 + tmp3
tmp6 = tl_math.exp(tmp5)
tmp7 = tmp4 + tmp6
tmp9 = tl_math.exp(tmp8)
tmp10 = tmp7 + tmp9
tmp11 = tl_math.log(tmp10)
tmp12 = tmp11 - tmp0
tmp14 = tmp12 * tmp13
tmp15 = tmp11 - tmp2
tmp17 = tmp15 * tmp16
tmp18 = tmp14 + tmp17
tmp19 = tmp11 - tmp5
tmp21 = tmp19 * tmp20
tmp22 = tmp18 + tmp21
tmp23 = tmp11 - tmp8
tmp25 = tmp23 * tmp24
tmp26 = tmp22 + tmp25
tmp27 = tl.broadcast_to(tmp26, [XBLOCK, RBLOCK])
tmp29 = tl.sum(tmp27, 1)[:, None]
tmp30 = 64.0
tmp31 = tmp29 / tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp31, None)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_max_sub_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
triton_per_fused_exp_log_mean_mul_sub_sum_1[grid(1)](buf2, buf0,
arg1_1, 1, 64, XBLOCK=1, num_warps=2, num_stages=1)
del arg1_1
del buf0
return buf2,
def soft_cross_entropy(logit, label, weight=None, reduce=None, reduction='mean'
):
if weight is not None and weight.requires_grad:
raise RuntimeError('gradient for weight is not supported')
losses = SoftCrossEntropyFunction.apply(logit, label, weight)
reduction = {(True): 'mean', (False): 'none', None: reduction}[reduce]
if reduction == 'mean':
return losses.mean()
elif reduction == 'sum':
return losses.sum()
elif reduction == 'none':
return losses
else:
raise ValueError('invalid value for reduction: {}'.format(reduction))
class SoftCrossEntropyFunction(torch.autograd.Function):
@staticmethod
def forward(ctx, logit, label, weight=None):
assert logit.size() == label.size(), 'logit.size() != label.size()'
dim = logit.dim()
max_logit = logit.max(dim - 1, keepdim=True)[0]
logit = logit - max_logit
exp_logit = logit.exp()
exp_sum = exp_logit.sum(dim - 1, keepdim=True)
prob = exp_logit / exp_sum
log_exp_sum = exp_sum.log()
neg_log_prob = log_exp_sum - logit
if weight is None:
weighted_label = label
else:
if weight.size() != (logit.size(-1),):
raise ValueError(
'since logit.size() = {}, weight.size() should be ({},), but got {}'
.format(logit.size(), logit.size(-1), weight.size()))
size = [1] * label.dim()
size[-1] = label.size(-1)
weighted_label = label * weight.view(size)
ctx.save_for_backward(weighted_label, prob)
out = (neg_log_prob * weighted_label).sum(dim - 1)
return out
@staticmethod
def backward(ctx, grad_output):
weighted_label, prob = ctx.saved_tensors
old_size = weighted_label.size()
K = old_size[-1]
B = weighted_label.numel() // K
grad_output = grad_output.view(B, 1)
weighted_label = weighted_label.view(B, K)
prob = prob.view(B, K)
grad_input = grad_output * (prob * weighted_label.sum(1, True) -
weighted_label)
grad_input = grad_input.view(old_size)
return grad_input, None, None
class SoftCrossEntropyLossNew(torch.nn.Module):
def __init__(self, weight=None, reduce=None, reduction='mean'):
super(SoftCrossEntropyLossNew, self).__init__()
self.weight = weight
self.reduce = reduce
self.reduction = reduction
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
Jingkang50/ICCV21_SCOOD
|
SoftCrossEntropyLoss
| false
| 8,364
|
[
"MIT"
] | 34
|
51204e3788a9e81aa334611072bef106fd9d13ad
|
https://github.com/Jingkang50/ICCV21_SCOOD/tree/51204e3788a9e81aa334611072bef106fd9d13ad
|
MaxPool2dSamePadding
|
# 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/ta/ctajmro4s3ivei3yjt7j34qft3mpkn36aqmrt7konkpig4v4ko5j.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x_1 => getitem
# Graph fragment:
# %getitem : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 0), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_0 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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_max_pool2d_with_indices_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_max_pool2d_with_indices_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4) % 4
x0 = xindex % 4
x4 = xindex
tmp0 = (-1) + x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = (-1) + x0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + ((-5) + x4), tmp10 & xmask, other=0.0)
tmp12 = x0
tmp13 = tmp12 >= tmp1
tmp14 = tmp12 < tmp3
tmp15 = tmp8 & tmp13
tmp16 = tmp15 & tmp14
tmp17 = tl.load(in_ptr0 + ((-4) + x4), tmp16 & xmask, other=0.0)
tmp18 = triton_helpers.maximum(tmp17, tmp11)
tmp19 = 1 + x0
tmp20 = tmp19 >= tmp1
tmp21 = tmp19 < tmp3
tmp22 = tmp8 & tmp20
tmp23 = tmp22 & tmp21
tmp24 = tl.load(in_ptr0 + ((-3) + x4), tmp23 & xmask, other=0.0)
tmp25 = triton_helpers.maximum(tmp24, tmp18)
tmp26 = 2 + x0
tmp27 = tmp26 >= tmp1
tmp28 = tmp26 < tmp3
tmp29 = tmp8 & tmp27
tmp30 = tmp29 & tmp28
tmp31 = tl.load(in_ptr0 + ((-2) + x4), tmp30 & xmask, other=0.0)
tmp32 = triton_helpers.maximum(tmp31, tmp25)
tmp33 = x1
tmp34 = tmp33 >= tmp1
tmp35 = tmp33 < tmp3
tmp36 = tmp34 & tmp35
tmp37 = tmp36 & tmp6
tmp38 = tmp37 & tmp7
tmp39 = tl.load(in_ptr0 + ((-1) + x4), tmp38 & xmask, other=0.0)
tmp40 = triton_helpers.maximum(tmp39, tmp32)
tmp41 = tmp36 & tmp13
tmp42 = tmp41 & tmp14
tmp43 = tl.load(in_ptr0 + (x4), tmp42 & xmask, other=0.0)
tmp44 = triton_helpers.maximum(tmp43, tmp40)
tmp45 = tmp36 & tmp20
tmp46 = tmp45 & tmp21
tmp47 = tl.load(in_ptr0 + (1 + x4), tmp46 & xmask, other=0.0)
tmp48 = triton_helpers.maximum(tmp47, tmp44)
tmp49 = tmp36 & tmp27
tmp50 = tmp49 & tmp28
tmp51 = tl.load(in_ptr0 + (2 + x4), tmp50 & xmask, other=0.0)
tmp52 = triton_helpers.maximum(tmp51, tmp48)
tmp53 = 1 + x1
tmp54 = tmp53 >= tmp1
tmp55 = tmp53 < tmp3
tmp56 = tmp54 & tmp55
tmp57 = tmp56 & tmp6
tmp58 = tmp57 & tmp7
tmp59 = tl.load(in_ptr0 + (3 + x4), tmp58 & xmask, other=0.0)
tmp60 = triton_helpers.maximum(tmp59, tmp52)
tmp61 = tmp56 & tmp13
tmp62 = tmp61 & tmp14
tmp63 = tl.load(in_ptr0 + (4 + x4), tmp62 & xmask, other=0.0)
tmp64 = triton_helpers.maximum(tmp63, tmp60)
tmp65 = tmp56 & tmp20
tmp66 = tmp65 & tmp21
tmp67 = tl.load(in_ptr0 + (5 + x4), tmp66 & xmask, other=0.0)
tmp68 = triton_helpers.maximum(tmp67, tmp64)
tmp69 = tmp56 & tmp27
tmp70 = tmp69 & tmp28
tmp71 = tl.load(in_ptr0 + (6 + x4), tmp70 & xmask, other=0.0)
tmp72 = triton_helpers.maximum(tmp71, tmp68)
tmp73 = 2 + x1
tmp74 = tmp73 >= tmp1
tmp75 = tmp73 < tmp3
tmp76 = tmp74 & tmp75
tmp77 = tmp76 & tmp6
tmp78 = tmp77 & tmp7
tmp79 = tl.load(in_ptr0 + (7 + x4), tmp78 & xmask, other=0.0)
tmp80 = triton_helpers.maximum(tmp79, tmp72)
tmp81 = tmp76 & tmp13
tmp82 = tmp81 & tmp14
tmp83 = tl.load(in_ptr0 + (8 + x4), tmp82 & xmask, other=0.0)
tmp84 = triton_helpers.maximum(tmp83, tmp80)
tmp85 = tmp76 & tmp20
tmp86 = tmp85 & tmp21
tmp87 = tl.load(in_ptr0 + (9 + x4), tmp86 & xmask, other=0.0)
tmp88 = triton_helpers.maximum(tmp87, tmp84)
tmp89 = tmp76 & tmp27
tmp90 = tmp89 & tmp28
tmp91 = tl.load(in_ptr0 + (10 + x4), tmp90 & xmask, other=0.0)
tmp92 = triton_helpers.maximum(tmp91, tmp88)
tl.store(out_ptr0 + (x4), tmp92, 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: [x_1], Original ATen: [aten.max_pool2d_with_indices]
stream0 = get_raw_stream(0)
triton_poi_fused_max_pool2d_with_indices_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
import torch.nn.functional as F
def get_same_padding(in_size, kernel_size, stride):
"""'Same 'same' operation with tensorflow
notice：padding=(0, 1, 0, 1) and padding=(1, 1, 1, 1) are different
padding=(1, 1, 1, 1):
out(H, W) = (in + [2 * padding] − kernel_size) // stride + 1
'same' padding=(0, 1, 0, 1):
out(H, W) = (in + [2 * padding] − kernel_size) / stride + 1
:param in_size: Union[int, tuple(in_h, in_w)]
:param kernel_size: Union[int, tuple(kernel_h, kernel_w)]
:param stride: Union[int, tuple(stride_h, stride_w)]
:return: padding: tuple(left, right, top, bottom)
"""
in_h, in_w = (in_size, in_size) if isinstance(in_size, int) else in_size
kernel_h, kernel_w = (kernel_size, kernel_size) if isinstance(kernel_size,
int) else kernel_size
stride_h, stride_w = (stride, stride) if isinstance(stride, int
) else stride
out_h, out_w = math.ceil(in_h / stride_h), math.ceil(in_w / stride_w)
pad_h = max((out_h - 1) * stride_h + kernel_h - in_h, 0)
pad_w = max((out_w - 1) * stride_w + kernel_w - in_w, 0)
return pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2
class MaxPool2dSamePadding(nn.MaxPool2d):
"""MaxPool2dDynamicSamePadding
由于输入大小都是128的倍数，所以动态池化和静态池化的结果是一致的。此处用动态池化代替静态池化，因为实现方便。
Since the input size is a multiple of 128,
the results of dynamic maxpool and static maxpool are consistent.
Here, dynamic maxpool is used instead of static maxpool,
because it is convenient to implement"""
def __init__(self, kernel_size, stride):
self.kernel_size = kernel_size
self.stride = stride
super(MaxPool2dSamePadding, self).__init__(kernel_size, stride)
def forward(self, x):
padding = get_same_padding(x.shape[-2:], self.kernel_size, self.stride)
x = F.pad(x, padding)
x = super().forward(x)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'kernel_size': 4, 'stride': 1}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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_max_pool2d_with_indices_0(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4 % 4
x0 = xindex % 4
x4 = xindex
tmp0 = -1 + x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = -1 + x0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (-5 + x4), tmp10 & xmask, other=0.0)
tmp12 = x0
tmp13 = tmp12 >= tmp1
tmp14 = tmp12 < tmp3
tmp15 = tmp8 & tmp13
tmp16 = tmp15 & tmp14
tmp17 = tl.load(in_ptr0 + (-4 + x4), tmp16 & xmask, other=0.0)
tmp18 = triton_helpers.maximum(tmp17, tmp11)
tmp19 = 1 + x0
tmp20 = tmp19 >= tmp1
tmp21 = tmp19 < tmp3
tmp22 = tmp8 & tmp20
tmp23 = tmp22 & tmp21
tmp24 = tl.load(in_ptr0 + (-3 + x4), tmp23 & xmask, other=0.0)
tmp25 = triton_helpers.maximum(tmp24, tmp18)
tmp26 = 2 + x0
tmp27 = tmp26 >= tmp1
tmp28 = tmp26 < tmp3
tmp29 = tmp8 & tmp27
tmp30 = tmp29 & tmp28
tmp31 = tl.load(in_ptr0 + (-2 + x4), tmp30 & xmask, other=0.0)
tmp32 = triton_helpers.maximum(tmp31, tmp25)
tmp33 = x1
tmp34 = tmp33 >= tmp1
tmp35 = tmp33 < tmp3
tmp36 = tmp34 & tmp35
tmp37 = tmp36 & tmp6
tmp38 = tmp37 & tmp7
tmp39 = tl.load(in_ptr0 + (-1 + x4), tmp38 & xmask, other=0.0)
tmp40 = triton_helpers.maximum(tmp39, tmp32)
tmp41 = tmp36 & tmp13
tmp42 = tmp41 & tmp14
tmp43 = tl.load(in_ptr0 + x4, tmp42 & xmask, other=0.0)
tmp44 = triton_helpers.maximum(tmp43, tmp40)
tmp45 = tmp36 & tmp20
tmp46 = tmp45 & tmp21
tmp47 = tl.load(in_ptr0 + (1 + x4), tmp46 & xmask, other=0.0)
tmp48 = triton_helpers.maximum(tmp47, tmp44)
tmp49 = tmp36 & tmp27
tmp50 = tmp49 & tmp28
tmp51 = tl.load(in_ptr0 + (2 + x4), tmp50 & xmask, other=0.0)
tmp52 = triton_helpers.maximum(tmp51, tmp48)
tmp53 = 1 + x1
tmp54 = tmp53 >= tmp1
tmp55 = tmp53 < tmp3
tmp56 = tmp54 & tmp55
tmp57 = tmp56 & tmp6
tmp58 = tmp57 & tmp7
tmp59 = tl.load(in_ptr0 + (3 + x4), tmp58 & xmask, other=0.0)
tmp60 = triton_helpers.maximum(tmp59, tmp52)
tmp61 = tmp56 & tmp13
tmp62 = tmp61 & tmp14
tmp63 = tl.load(in_ptr0 + (4 + x4), tmp62 & xmask, other=0.0)
tmp64 = triton_helpers.maximum(tmp63, tmp60)
tmp65 = tmp56 & tmp20
tmp66 = tmp65 & tmp21
tmp67 = tl.load(in_ptr0 + (5 + x4), tmp66 & xmask, other=0.0)
tmp68 = triton_helpers.maximum(tmp67, tmp64)
tmp69 = tmp56 & tmp27
tmp70 = tmp69 & tmp28
tmp71 = tl.load(in_ptr0 + (6 + x4), tmp70 & xmask, other=0.0)
tmp72 = triton_helpers.maximum(tmp71, tmp68)
tmp73 = 2 + x1
tmp74 = tmp73 >= tmp1
tmp75 = tmp73 < tmp3
tmp76 = tmp74 & tmp75
tmp77 = tmp76 & tmp6
tmp78 = tmp77 & tmp7
tmp79 = tl.load(in_ptr0 + (7 + x4), tmp78 & xmask, other=0.0)
tmp80 = triton_helpers.maximum(tmp79, tmp72)
tmp81 = tmp76 & tmp13
tmp82 = tmp81 & tmp14
tmp83 = tl.load(in_ptr0 + (8 + x4), tmp82 & xmask, other=0.0)
tmp84 = triton_helpers.maximum(tmp83, tmp80)
tmp85 = tmp76 & tmp20
tmp86 = tmp85 & tmp21
tmp87 = tl.load(in_ptr0 + (9 + x4), tmp86 & xmask, other=0.0)
tmp88 = triton_helpers.maximum(tmp87, tmp84)
tmp89 = tmp76 & tmp27
tmp90 = tmp89 & tmp28
tmp91 = tl.load(in_ptr0 + (10 + x4), tmp90 & xmask, other=0.0)
tmp92 = triton_helpers.maximum(tmp91, tmp88)
tl.store(out_ptr0 + x4, tmp92, 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_max_pool2d_with_indices_0[grid(256)](arg0_1, buf0,
256, XBLOCK=128, num_warps=4, num_stages=1)
del arg0_1
return buf0,
def get_same_padding(in_size, kernel_size, stride):
"""'Same 'same' operation with tensorflow
notice：padding=(0, 1, 0, 1) and padding=(1, 1, 1, 1) are different
padding=(1, 1, 1, 1):
out(H, W) = (in + [2 * padding] − kernel_size) // stride + 1
'same' padding=(0, 1, 0, 1):
out(H, W) = (in + [2 * padding] − kernel_size) / stride + 1
:param in_size: Union[int, tuple(in_h, in_w)]
:param kernel_size: Union[int, tuple(kernel_h, kernel_w)]
:param stride: Union[int, tuple(stride_h, stride_w)]
:return: padding: tuple(left, right, top, bottom)
"""
in_h, in_w = (in_size, in_size) if isinstance(in_size, int) else in_size
kernel_h, kernel_w = (kernel_size, kernel_size) if isinstance(kernel_size,
int) else kernel_size
stride_h, stride_w = (stride, stride) if isinstance(stride, int
) else stride
out_h, out_w = math.ceil(in_h / stride_h), math.ceil(in_w / stride_w)
pad_h = max((out_h - 1) * stride_h + kernel_h - in_h, 0)
pad_w = max((out_w - 1) * stride_w + kernel_w - in_w, 0)
return pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2
class MaxPool2dSamePaddingNew(nn.MaxPool2d):
"""MaxPool2dDynamicSamePadding
由于输入大小都是128的倍数，所以动态池化和静态池化的结果是一致的。此处用动态池化代替静态池化，因为实现方便。
Since the input size is a multiple of 128,
the results of dynamic maxpool and static maxpool are consistent.
Here, dynamic maxpool is used instead of static maxpool,
because it is convenient to implement"""
def __init__(self, kernel_size, stride):
self.kernel_size = kernel_size
self.stride = stride
super(MaxPool2dSamePaddingNew, self).__init__(kernel_size, stride)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Jintao-Huang/EfficientDet_PyTorch
|
MaxPool2dSamePadding
| false
| 8,365
|
[
"Apache-2.0"
] | 18
|
79616be397b7f57992cd43b772f65b58b5e25a8b
|
https://github.com/Jintao-Huang/EfficientDet_PyTorch/tree/79616be397b7f57992cd43b772f65b58b5e25a8b
|
SoftSelectPrototype
|
# 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/d7/cd7yhzyv6qomqlznstk6ctxa74hks6bg46x4onw73m5tseazikoy.py
# Topologically Sorted Source Nodes: [query_, support_, mul, sum_1], Original ATen: [aten.clone, aten.mul, aten.sum]
# Source node to ATen node mapping:
# mul => mul
# query_ => clone
# sum_1 => sum_1
# support_ => clone_1
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand,), kwargs = {memory_format: torch.contiguous_format})
# %clone_1 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%expand_1,), kwargs = {memory_format: torch.contiguous_format})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%clone, %clone_1), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [2]), kwargs = {})
# %mul_tensor : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_1, 1), kwargs = {})
triton_poi_fused_clone_mul_sum_0 = async_compile.triton('triton_poi_fused_clone_mul_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_clone_mul_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_clone_mul_sum_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4)
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + (4*x1)), 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*x1)), 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*x1)), 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 = 1.0
tmp16 = tmp14 * tmp15
tl.store(out_ptr0 + (x2), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gw/cgwpqvpipztpejbgkq3vb2qlu5uj6b7ltuukxzruggnwbma657ku.py
# Topologically Sorted Source Nodes: [att], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# att => exp
# Graph fragment:
# %amax_default : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%mul_tensor, [1], True), kwargs = {})
# %sub_tensor : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_tensor, %amax_default), kwargs = {})
# %mul_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_tensor, 0.5), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%mul_tensor_1,), kwargs = {})
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 = 0.5
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/wp/cwpr2jp7bpqwkbaxgttcl2ekjdtkf6lezuspoiunutzd3j2qgeyk.py
# Topologically Sorted Source Nodes: [att], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# att => 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=1] = 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=[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):
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: [query_, support_, mul, sum_1], Original ATen: [aten.clone, aten.mul, aten.sum]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_mul_sum_0.run(arg0_1, arg1_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: [att], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf0, buf1, 16, grid=grid(16), stream=stream0)
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [att], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf1, buf2, 16, grid=grid(16), stream=stream0)
buf3 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [att, center], Original ATen: [aten._softmax, aten.mm]
extern_kernels.mm(buf2, arg1_1, out=buf3)
del arg1_1
del buf2
return (buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class SoftSelectAttention(nn.Module):
def __init__(self, hidden_size):
super(SoftSelectAttention, self).__init__()
def forward(self, support, query):
"""
:param support: [few, dim]
:param query: [batch, dim]
:return:
"""
query_ = query.unsqueeze(1).expand(query.size(0), support.size(0),
query.size(1)).contiguous()
support_ = support.unsqueeze(0).expand_as(query_).contiguous()
scalar = support.size(1) ** -0.5
score = torch.sum(query_ * support_, dim=2) * scalar
att = torch.softmax(score, dim=1)
center = torch.mm(att, support)
return center
class SoftSelectPrototype(nn.Module):
def __init__(self, r_dim):
super(SoftSelectPrototype, self).__init__()
self.Attention = SoftSelectAttention(hidden_size=r_dim)
def forward(self, support, query):
center = self.Attention(support, query)
return center
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'r_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_clone_mul_sum_0(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4
x0 = xindex % 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x1), 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 * x1), 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 * x1), 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 = 1.0
tmp16 = tmp14 * tmp15
tl.store(out_ptr0 + x2, tmp16, 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 = 0.5
tmp10 = tmp8 * tmp9
tmp11 = tl_math.exp(tmp10)
tl.store(out_ptr0 + x2, tmp11, 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):
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_clone_mul_sum_0[grid(16)](arg0_1, arg1_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__softmax_1[grid(16)](buf0, buf1, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf2 = buf0
del buf0
triton_poi_fused__softmax_2[grid(16)](buf1, buf2, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf3 = buf1
del buf1
extern_kernels.mm(buf2, arg1_1, out=buf3)
del arg1_1
del buf2
return buf3,
class SoftSelectAttention(nn.Module):
def __init__(self, hidden_size):
super(SoftSelectAttention, self).__init__()
def forward(self, support, query):
"""
:param support: [few, dim]
:param query: [batch, dim]
:return:
"""
query_ = query.unsqueeze(1).expand(query.size(0), support.size(0),
query.size(1)).contiguous()
support_ = support.unsqueeze(0).expand_as(query_).contiguous()
scalar = support.size(1) ** -0.5
score = torch.sum(query_ * support_, dim=2) * scalar
att = torch.softmax(score, dim=1)
center = torch.mm(att, support)
return center
class SoftSelectPrototypeNew(nn.Module):
def __init__(self, r_dim):
super(SoftSelectPrototypeNew, self).__init__()
self.Attention = SoftSelectAttention(hidden_size=r_dim)
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
JiaweiSheng/FAAN
|
SoftSelectPrototype
| false
| 8,366
|
[
"MIT"
] | 41
|
b439b829506c4e2e9044a6b2ab7f3d844f445a95
|
https://github.com/JiaweiSheng/FAAN/tree/b439b829506c4e2e9044a6b2ab7f3d844f445a95
|
Critic
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/zm/czm6acrrgjryz6xi3wza7npycjuiqsdsygpfdo3lbzaquecrmeuj.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 = ([%primals_1, %primals_2], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[32],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((4*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + ((4*x1) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qk/cqk5cnmrffdyh6x4yseboivxypitwp75t6ywqxjlpiff6wcgokiw.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# x_1 => relu
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_4), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_1,), kwargs = {})
triton_poi_fused_relu_1 = async_compile.triton('triton_poi_fused_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_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')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 8), (8, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (1, 4), (4, 1))
assert_size_stride(primals_8, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 8), (8, 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, primals_2, buf0, 32, grid=grid(32), stream=stream0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf0, reinterpret_tensor(primals_3, (8, 4), (1, 8), 0), out=buf1)
del primals_3
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu]
triton_poi_fused_relu_1.run(buf2, primals_4, 16, grid=grid(16), stream=stream0)
del primals_4
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf2, reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf3)
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu]
triton_poi_fused_relu_1.run(buf4, primals_6, 16, grid=grid(16), stream=stream0)
del primals_6
buf6 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, buf4, reinterpret_tensor(primals_7, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf6)
del primals_8
return (buf6, buf0, buf2, buf4, primals_7, primals_5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
class Critic(nn.Module):
def __init__(self, n_obs, action_dim, hidden_size, init_w=0.003):
super(Critic, self).__init__()
self.linear1 = nn.Linear(n_obs + action_dim, hidden_size)
self.linear2 = nn.Linear(hidden_size, hidden_size)
self.linear3 = nn.Linear(hidden_size, 1)
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
def forward(self, state, action):
x = torch.cat([state, action], 1)
x = F.relu(self.linear1(x))
x = F.relu(self.linear2(x))
x = self.linear3(x)
return x
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'n_obs': 4, 'action_dim': 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
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = xindex // 8
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (4 * x1 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_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)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 8), (8, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (1, 4), (4, 1))
assert_size_stride(primals_8, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(32)](primals_1, primals_2, buf0, 32,
XBLOCK=32, num_warps=1, num_stages=1)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf0, reinterpret_tensor(primals_3, (8, 4), (1, 8
), 0), out=buf1)
del primals_3
buf2 = buf1
del buf1
triton_poi_fused_relu_1[grid(16)](buf2, primals_4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf2, reinterpret_tensor(primals_5, (4, 4), (1, 4
), 0), out=buf3)
buf4 = buf3
del buf3
triton_poi_fused_relu_1[grid(16)](buf4, primals_6, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_6
buf6 = empty_strided_cuda((4, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_8, buf4, reinterpret_tensor(primals_7,
(4, 1), (1, 4), 0), alpha=1, beta=1, out=buf6)
del primals_8
return buf6, buf0, buf2, buf4, primals_7, primals_5
class CriticNew(nn.Module):
def __init__(self, n_obs, action_dim, hidden_size, init_w=0.003):
super(CriticNew, self).__init__()
self.linear1 = nn.Linear(n_obs + action_dim, hidden_size)
self.linear2 = nn.Linear(hidden_size, hidden_size)
self.linear3 = nn.Linear(hidden_size, 1)
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
def forward(self, input_0, input_1):
primals_3 = self.linear1.weight
primals_4 = self.linear1.bias
primals_1 = self.linear2.weight
primals_6 = self.linear2.bias
primals_7 = self.linear3.weight
primals_8 = self.linear3.bias
primals_2 = input_0
primals_5 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0]
|
JohnJim0816/rl-tutorials
|
Critic
| false
| 8,367
|
[
"MIT"
] | 16
|
e99daea815da85f9f25dff2d01b030249a203d22
|
https://github.com/JohnJim0816/rl-tutorials/tree/e99daea815da85f9f25dff2d01b030249a203d22
|
GlobalAveragePooling
|
# 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: [outs], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# outs => 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_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 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [outs], 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 (reinterpret_tensor(buf1, (4, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
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 GlobalAveragePooling(nn.Module):
"""Global Average Pooling neck.
Note that we use `view` to remove extra channel after pooling. We do not
use `squeeze` as it will also remove the batch dimension when the tensor
has a batch dimension of size 1, which can lead to unexpected errors.
"""
def __init__(self):
super().__init__()
self.gap = nn.AdaptiveAvgPool2d((1, 1))
def init_weights(self):
pass
def forward(self, inputs):
if isinstance(inputs, tuple):
outs = tuple([self.gap(x) for x in inputs])
outs = tuple([out.view(x.size(0), -1) for out, x in zip(outs,
inputs)])
elif isinstance(inputs, list):
outs = [self.gap(x) for x in inputs]
outs = [out.view(x.size(0), -1) for out, x in zip(outs, inputs)]
elif isinstance(inputs, torch.Tensor):
outs = self.gap(inputs)
outs = outs.view(inputs.size(0), -1)
else:
raise TypeError('neck inputs should be tuple or torch.tensor')
return outs
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
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 = 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 reinterpret_tensor(buf1, (4, 4), (4, 1), 0),
class GlobalAveragePoolingNew(nn.Module):
"""Global Average Pooling neck.
Note that we use `view` to remove extra channel after pooling. We do not
use `squeeze` as it will also remove the batch dimension when the tensor
has a batch dimension of size 1, which can lead to unexpected errors.
"""
def __init__(self):
super().__init__()
self.gap = nn.AdaptiveAvgPool2d((1, 1))
def init_weights(self):
pass
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Jackqu/mmpose
|
GlobalAveragePooling
| false
| 8,368
|
[
"Apache-2.0"
] | 38
|
ad8acc5ff5da7993c6befdc4b1ced2c2ecb64533
|
https://github.com/Jackqu/mmpose/tree/ad8acc5ff5da7993c6befdc4b1ced2c2ecb64533
|
GlobalAttentionGeneral
|
# 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/gh/cghdn4pihhs4yx6f6a4ig3pjgfy5zh6dghclq5xeandkuwdvuk5c.py
# Topologically Sorted Source Nodes: [targetT], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# targetT => 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=[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_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 = 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)
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/om/com5a2jdtzsd2gxozlfy64n7thpakjehaajnnx7weffss4gxj2sy.py
# Topologically Sorted Source Nodes: [attn_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_2 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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
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/o6/co6rds2fnpdijj6xbnv2tmkoxmykgqu7idfvvwu22ylhgstr53gf.py
# Topologically Sorted Source Nodes: [attn_4], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# attn_4 => clone_1
# Graph fragment:
# %clone_1 : [num_users=2] = call_function[target=torch.ops.aten.clone.default](args = (%permute_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_2 = async_compile.triton('triton_poi_fused_clone_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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_clone_2(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
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (64*y1)), xmask & ymask)
tmp1 = tl.load(in_ptr0 + ((4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2 + (16*y3)), tmp8, xmask & ymask)
''', 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), (16, 4, 1))
assert_size_stride(arg2_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 16, 4), (64, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [targetT], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
buf1 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [targetT, attn], Original ATen: [aten.clone, aten.bmm]
extern_kernels.bmm(buf0, arg1_1, out=buf1)
del arg1_1
buf2 = reinterpret_tensor(buf0, (64, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [attn_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (4, 4, 16), (64, 16, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [attn_4], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf2, buf3, 16, 16, grid=grid(16, 16), stream=stream0)
buf4 = reinterpret_tensor(buf2, (4, 4, 16), (64, 16, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [weightedContext], Original ATen: [aten.bmm]
extern_kernels.bmm(arg2_1, buf3, out=buf4)
del arg2_1
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf3, (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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
arg2_1 = rand_strided((4, 4, 4), (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
class GlobalAttentionGeneral(nn.Module):
def __init__(self, idf, cdf):
super(GlobalAttentionGeneral, self).__init__()
self.sm = nn.Softmax()
self.mask = None
def applyMask(self, mask):
self.mask = mask
def forward(self, input, context_key, content_value):
"""
input: batch x idf x ih x iw (queryL=ihxiw)
context: batch x cdf x sourceL
"""
ih, iw = input.size(2), input.size(3)
queryL = ih * iw
batch_size, sourceL = context_key.size(0), context_key.size(2)
target = input.view(batch_size, -1, queryL)
targetT = torch.transpose(target, 1, 2).contiguous()
sourceT = context_key
attn = torch.bmm(targetT, sourceT)
attn = attn.view(batch_size * queryL, sourceL)
if self.mask is not None:
mask = self.mask.repeat(queryL, 1)
attn.data.masked_fill_(mask.data, -float('inf'))
attn = self.sm(attn)
attn = attn.view(batch_size, queryL, sourceL)
attn = torch.transpose(attn, 1, 2).contiguous()
weightedContext = torch.bmm(content_value, attn)
weightedContext = weightedContext.view(batch_size, -1, ih, iw)
attn = attn.view(batch_size, -1, ih, iw)
return weightedContext, attn
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4,
4, 4])]
def get_init_inputs():
return [[], {'idf': 4, 'cdf': 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
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 = 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)
tl.store(out_ptr0 + x0, tmp0, 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
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_clone_2(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
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 64 * y1), xmask & ymask)
tmp1 = tl.load(in_ptr0 + (4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2 + 16 * y3), tmp8, xmask & ymask)
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), (16, 4, 1))
assert_size_stride(arg2_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 16, 4), (64, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(256)](arg0_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
extern_kernels.bmm(buf0, arg1_1, out=buf1)
del arg1_1
buf2 = reinterpret_tensor(buf0, (64, 4), (4, 1), 0)
del buf0
triton_poi_fused__softmax_1[grid(256)](buf1, buf2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (4, 4, 16), (64, 16, 1), 0)
del buf1
triton_poi_fused_clone_2[grid(16, 16)](buf2, buf3, 16, 16, XBLOCK=
16, YBLOCK=16, num_warps=4, num_stages=1)
buf4 = reinterpret_tensor(buf2, (4, 4, 16), (64, 16, 1), 0)
del buf2
extern_kernels.bmm(arg2_1, buf3, out=buf4)
del arg2_1
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
class GlobalAttentionGeneralNew(nn.Module):
def __init__(self, idf, cdf):
super(GlobalAttentionGeneralNew, self).__init__()
self.sm = nn.Softmax()
self.mask = None
def applyMask(self, mask):
self.mask = mask
def forward(self, input_0, input_1, input_2):
arg0_1 = input_0
arg1_1 = input_1
arg2_1 = input_2
output = call([arg0_1, arg1_1, arg2_1])
return output[0], output[1]
|
JoonHong-Kim/T2I_CL
|
GlobalAttentionGeneral
| false
| 8,369
|
[
"MIT"
] | 35
|
c52aa73da903d6e4174eeef2663e5bc1163785b1
|
https://github.com/JoonHong-Kim/T2I_CL/tree/c52aa73da903d6e4174eeef2663e5bc1163785b1
|
PolicyNet
|
# 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: [x], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le_2 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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')
# kernel path: runs/run_shard_1/inductor_cache/cv/ccv5nvw6wyiix4ayfbjqnrimru7yfskf67vnmbigmvir4lnairer.py
# Topologically Sorted Source Nodes: [log_std_1], Original ATen: [aten.clamp, aten.ge, aten.le, aten.logical_and]
# Source node to ATen node mapping:
# log_std_1 => clamp_max, clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%view_7, -20), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 2), kwargs = {})
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%view_7, -20), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%view_7, 2), kwargs = {})
# %logical_and : [num_users=1] = call_function[target=torch.ops.aten.logical_and.default](args = (%ge, %le), kwargs = {})
triton_poi_fused_clamp_ge_le_logical_and_1 = async_compile.triton('triton_poi_fused_clamp_ge_le_logical_and_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_clamp_ge_le_logical_and_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_clamp_ge_le_logical_and_1(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = -20.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = 2.0
tmp6 = triton_helpers.minimum(tmp4, tmp5)
tmp7 = tmp2 >= tmp3
tmp8 = tmp2 <= tmp5
tmp9 = tmp7 & tmp8
tl.store(out_ptr0 + (x2), tmp6, xmask)
tl.store(out_ptr1 + (x2), tmp9, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf9, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf8, 256, grid=grid(256), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mean], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [log_std_1], Original ATen: [aten.clamp, aten.ge, aten.le, aten.logical_and]
triton_poi_fused_clamp_ge_le_logical_and_1.run(buf5, primals_9, buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf5
del primals_9
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(buf3, (64, 4), (4, 1), 0), buf7, primals_8, primals_6, buf8, primals_4, buf9, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.distributions import Normal
class PolicyNet(nn.Module):
def __init__(self, state_dim, action_dim, hidden_dim, init_w=0.003,
log_std_min=-20, log_std_max=2):
super(PolicyNet, self).__init__()
self.log_std_min = log_std_min
self.log_std_max = log_std_max
self.linear1 = nn.Linear(state_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
self.mean_linear = nn.Linear(hidden_dim, action_dim)
self.mean_linear.weight.data.uniform_(-init_w, init_w)
self.mean_linear.bias.data.uniform_(-init_w, init_w)
self.log_std_linear = nn.Linear(hidden_dim, action_dim)
self.log_std_linear.weight.data.uniform_(-init_w, init_w)
self.log_std_linear.bias.data.uniform_(-init_w, init_w)
def forward(self, state):
x = F.relu(self.linear1(state))
x = F.relu(self.linear2(x))
mean = self.mean_linear(x)
log_std = self.log_std_linear(x)
log_std = torch.clamp(log_std, self.log_std_min, self.log_std_max)
return mean, log_std
def evaluate(self, state, epsilon=1e-06):
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
log_prob = normal.log_prob(z) - torch.log(1 - action.pow(2) + epsilon)
log_prob = log_prob.sum(-1, keepdim=True)
return action, log_prob, z, mean, log_std
def get_action(self, state):
state = torch.FloatTensor(state).unsqueeze(0)
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
action = action.detach().cpu().numpy()
return action[0]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4, 'action_dim': 4, 'hidden_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
from torch.distributions import Normal
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr0 + x2, tmp6, xmask)
@triton.jit
def triton_poi_fused_clamp_ge_le_logical_and_1(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = -20.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = 2.0
tmp6 = triton_helpers.minimum(tmp4, tmp5)
tmp7 = tmp2 >= tmp3
tmp8 = tmp2 <= tmp5
tmp9 = tmp7 & tmp8
tl.store(out_ptr0 + x2, tmp6, xmask)
tl.store(out_ptr1 + x2, tmp9, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf1,
primals_2, buf9, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf3,
primals_5, buf8, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf4)
del primals_7
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf5)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_clamp_ge_le_logical_and_1[grid(256)](buf5,
primals_9, buf6, buf7, 256, XBLOCK=256, num_warps=4, num_stages=1)
del buf5
del primals_9
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), buf6, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(
buf3, (64, 4), (4, 1), 0
), buf7, primals_8, primals_6, buf8, primals_4, buf9
class PolicyNetNew(nn.Module):
def __init__(self, state_dim, action_dim, hidden_dim, init_w=0.003,
log_std_min=-20, log_std_max=2):
super(PolicyNetNew, self).__init__()
self.log_std_min = log_std_min
self.log_std_max = log_std_max
self.linear1 = nn.Linear(state_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
self.mean_linear = nn.Linear(hidden_dim, action_dim)
self.mean_linear.weight.data.uniform_(-init_w, init_w)
self.mean_linear.bias.data.uniform_(-init_w, init_w)
self.log_std_linear = nn.Linear(hidden_dim, action_dim)
self.log_std_linear.weight.data.uniform_(-init_w, init_w)
self.log_std_linear.bias.data.uniform_(-init_w, init_w)
def evaluate(self, state, epsilon=1e-06):
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
log_prob = normal.log_prob(z) - torch.log(1 - action.pow(2) + epsilon)
log_prob = log_prob.sum(-1, keepdim=True)
return action, log_prob, z, mean, log_std
def get_action(self, state):
state = torch.FloatTensor(state).unsqueeze(0)
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(mean, std)
z = normal.sample()
action = torch.tanh(z)
action = action.detach().cpu().numpy()
return action[0]
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_6 = self.mean_linear.weight
primals_7 = self.mean_linear.bias
primals_8 = self.log_std_linear.weight
primals_9 = self.log_std_linear.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0], output[1]
|
JohnJim0816/rl-tutorials
|
PolicyNet
| false
| 8,370
|
[
"MIT"
] | 16
|
e99daea815da85f9f25dff2d01b030249a203d22
|
https://github.com/JohnJim0816/rl-tutorials/tree/e99daea815da85f9f25dff2d01b030249a203d22
|
SE
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/yt/cytnwjfj2fwh3czq2vl4j7w7fjotaquc3crv6vgeaqdaw77o76vp.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# x => mean
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [-1, -2, -3], True), kwargs = {})
triton_per_fused_mean_0 = async_compile.triton('triton_per_fused_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 64],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, 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 = 4
rnumel = 64
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (64*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 64.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/5x/c5xvbbrbrcgfjcyatbfkvnueua5ae2wvjxm5atfvp25dbd2xeb7g.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_2 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%squeeze,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
tl.store(out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/va/cvahq7bwdf37kayfnung5geylifqe6iggx2yfk65vyjdptqkn76q.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_3 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%unsqueeze_1, %primals_4, %primals_5, [1, 1, 1], [0, 0, 0], [1, 1, 1], False, [0, 0, 0], 1), kwargs = {})
triton_poi_fused_convolution_2 = async_compile.triton('triton_poi_fused_convolution_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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 = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/za/czalxflwqycvo5d526kdekopuvio357kxbzo7c6fo3fwmp5s6a7k.py
# Topologically Sorted Source Nodes: [x_4, mul], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# mul => mul
# x_4 => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%squeeze_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %sigmoid), kwargs = {})
triton_poi_fused_mul_sigmoid_3 = async_compile.triton('triton_poi_fused_mul_sigmoid_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 // 64)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (16, 4, 1, 1, 1), (4, 1, 1, 1, 1))
assert_size_stride(primals_3, (16, ), (1, ))
assert_size_stride(primals_4, (4, 16, 1, 1, 1), (16, 1, 1, 1, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 1, 1), (1, 4, 4, 4), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 4, 64, grid=grid(4), stream=stream0)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 1, 1, 1), (0, 1, 0, 0, 0), 0), primals_2, stride=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False, output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf2, (1, 16, 1, 1, 1), (16, 1, 1, 1, 1))
buf3 = reinterpret_tensor(buf2, (16, 1, 1, 1), (1, 16, 16, 16), 0); del buf2 # reuse
buf7 = empty_strided_cuda((16, 1, 1, 1), (1, 1, 1, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_1.run(buf3, primals_3, buf7, 16, grid=grid(16), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(reinterpret_tensor(buf3, (1, 16, 1, 1, 1), (0, 1, 0, 0, 0), 0), primals_4, stride=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False, output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf4, (1, 4, 1, 1, 1), (4, 1, 1, 1, 1))
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf5, primals_5, 4, grid=grid(4), stream=stream0)
del primals_5
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_4, mul], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_3.run(primals_1, buf5, buf6, 256, grid=grid(256), stream=stream0)
return (buf6, primals_1, primals_2, primals_4, reinterpret_tensor(buf1, (1, 4, 1, 1, 1), (4, 1, 1, 1, 1), 0), reinterpret_tensor(buf3, (1, 16, 1, 1, 1), (16, 1, 1, 1, 1), 0), buf5, buf7, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, 4, 1, 1, 1), (4, 1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 16, 1, 1, 1), (16, 1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from itertools import chain as chain
import torch.utils.data
import torch.nn as nn
class SwishEfficient(torch.autograd.Function):
"""Swish activation function: x * sigmoid(x)."""
@staticmethod
def forward(ctx, x):
result = x * torch.sigmoid(x)
ctx.save_for_backward(x)
return result
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_variables[0]
sigmoid_x = torch.sigmoid(x)
return grad_output * (sigmoid_x * (1 + x * (1 - sigmoid_x)))
class Swish(nn.Module):
"""Swish activation function: x * sigmoid(x)."""
def __init__(self):
super(Swish, self).__init__()
def forward(self, x):
return SwishEfficient.apply(x)
class SE(nn.Module):
"""Squeeze-and-Excitation (SE) block w/ Swish: AvgPool, FC, Swish, FC, Sigmoid."""
def _round_width(self, width, multiplier, min_width=8, divisor=8):
"""
Round width of filters based on width multiplier
Args:
width (int): the channel dimensions of the input.
multiplier (float): the multiplication factor.
min_width (int): the minimum width after multiplication.
divisor (int): the new width should be dividable by divisor.
"""
if not multiplier:
return width
width *= multiplier
min_width = min_width or divisor
width_out = max(min_width, int(width + divisor / 2) // divisor *
divisor)
if width_out < 0.9 * width:
width_out += divisor
return int(width_out)
def __init__(self, dim_in, ratio, relu_act=True):
"""
Args:
dim_in (int): the channel dimensions of the input.
ratio (float): the channel reduction ratio for squeeze.
relu_act (bool): whether to use ReLU activation instead
of Swish (default).
divisor (int): the new width should be dividable by divisor.
"""
super(SE, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
dim_fc = self._round_width(dim_in, ratio)
self.fc1 = nn.Conv3d(dim_in, dim_fc, 1, bias=True)
self.fc1_act = nn.ReLU() if relu_act else Swish()
self.fc2 = nn.Conv3d(dim_fc, dim_in, 1, bias=True)
self.fc2_sig = nn.Sigmoid()
def forward(self, x):
x_in = x
for module in self.children():
x = module(x)
return x_in * x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'dim_in': 4, 'ratio': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from itertools import chain as chain
import torch.utils.data
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 4
RBLOCK: tl.constexpr = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 64 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp5 = 64.0
tmp6 = tmp4 / tmp5
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x0, tmp4, xmask)
tl.store(out_ptr0 + x0, tmp6, xmask)
@triton.jit
def triton_poi_fused_convolution_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_mul_sigmoid_3(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 64
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tl.sigmoid(tmp1)
tmp3 = tmp0 * tmp2
tl.store(out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (16, 4, 1, 1, 1), (4, 1, 1, 1, 1))
assert_size_stride(primals_3, (16,), (1,))
assert_size_stride(primals_4, (4, 16, 1, 1, 1), (16, 1, 1, 1, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 1, 1, 1), (1, 4, 4, 4), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_per_fused_mean_0[grid(4)](buf1, primals_1, 4, 64, XBLOCK=1,
num_warps=2, num_stages=1)
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 1,
1, 1), (0, 1, 0, 0, 0), 0), primals_2, stride=(1, 1, 1),
padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False,
output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf2, (1, 16, 1, 1, 1), (16, 1, 1, 1, 1))
buf3 = reinterpret_tensor(buf2, (16, 1, 1, 1), (1, 16, 16, 16), 0)
del buf2
buf7 = empty_strided_cuda((16, 1, 1, 1), (1, 1, 1, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_1[grid(16)](buf3,
primals_3, buf7, 16, XBLOCK=16, num_warps=1, num_stages=1)
del primals_3
buf4 = extern_kernels.convolution(reinterpret_tensor(buf3, (1, 16,
1, 1, 1), (0, 1, 0, 0, 0), 0), primals_4, stride=(1, 1, 1),
padding=(0, 0, 0), dilation=(1, 1, 1), transposed=False,
output_padding=(0, 0, 0), groups=1, bias=None)
assert_size_stride(buf4, (1, 4, 1, 1, 1), (4, 1, 1, 1, 1))
buf5 = buf4
del buf4
triton_poi_fused_convolution_2[grid(4)](buf5, primals_5, 4, XBLOCK=
4, num_warps=1, num_stages=1)
del primals_5
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_3[grid(256)](primals_1, buf5, buf6,
256, XBLOCK=128, num_warps=4, num_stages=1)
return buf6, primals_1, primals_2, primals_4, reinterpret_tensor(buf1,
(1, 4, 1, 1, 1), (4, 1, 1, 1, 1), 0), reinterpret_tensor(buf3, (1,
16, 1, 1, 1), (16, 1, 1, 1, 1), 0), buf5, buf7
class SwishEfficient(torch.autograd.Function):
"""Swish activation function: x * sigmoid(x)."""
@staticmethod
def forward(ctx, x):
result = x * torch.sigmoid(x)
ctx.save_for_backward(x)
return result
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_variables[0]
sigmoid_x = torch.sigmoid(x)
return grad_output * (sigmoid_x * (1 + x * (1 - sigmoid_x)))
class Swish(nn.Module):
"""Swish activation function: x * sigmoid(x)."""
def __init__(self):
super(Swish, self).__init__()
def forward(self, x):
return SwishEfficient.apply(x)
class SENew(nn.Module):
"""Squeeze-and-Excitation (SE) block w/ Swish: AvgPool, FC, Swish, FC, Sigmoid."""
def _round_width(self, width, multiplier, min_width=8, divisor=8):
"""
Round width of filters based on width multiplier
Args:
width (int): the channel dimensions of the input.
multiplier (float): the multiplication factor.
min_width (int): the minimum width after multiplication.
divisor (int): the new width should be dividable by divisor.
"""
if not multiplier:
return width
width *= multiplier
min_width = min_width or divisor
width_out = max(min_width, int(width + divisor / 2) // divisor *
divisor)
if width_out < 0.9 * width:
width_out += divisor
return int(width_out)
def __init__(self, dim_in, ratio, relu_act=True):
"""
Args:
dim_in (int): the channel dimensions of the input.
ratio (float): the channel reduction ratio for squeeze.
relu_act (bool): whether to use ReLU activation instead
of Swish (default).
divisor (int): the new width should be dividable by divisor.
"""
super(SENew, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
dim_fc = self._round_width(dim_in, ratio)
self.fc1 = nn.Conv3d(dim_in, dim_fc, 1, bias=True)
self.fc1_act = nn.ReLU() if relu_act else Swish()
self.fc2 = nn.Conv3d(dim_fc, dim_in, 1, bias=True)
self.fc2_sig = nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.fc1.weight
primals_3 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
JaywongWang/SlowFast
|
SE
| false
| 8,371
|
[
"Apache-2.0"
] | 43
|
366467aafc856712fdc3e9c4cce8e90969047ee6
|
https://github.com/JaywongWang/SlowFast/tree/366467aafc856712fdc3e9c4cce8e90969047ee6
|
WasLoss
|
# 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/ig/cigrcwiw3qaaurkftdct7zaeymdvdxbs6k43z7o6rbv6aogggdgl.py
# Topologically Sorted Source Nodes: [sort], Original ATen: [aten.sort]
# Source node to ATen node mapping:
# sort => sort
# Graph fragment:
# %sort : [num_users=1] = call_function[target=torch.ops.aten.sort.default](args = (%arg0_1, 0), kwargs = {})
triton_per_fused_sort_0 = async_compile.triton('triton_per_fused_sort_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[64, 4],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_sort_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}
)
@triton.jit
def triton_per_fused_sort_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 64
rnumel = 4
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*r1)), xmask, other=0.0)
tmp1 = r1
tmp2 = tmp1.to(tl.int16)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5, tmp6, = triton_helpers.sort_with_index(tmp3, tmp4, None, 1, stable=False, descending=False)
tl.store(out_ptr0 + (x0 + (64*r1)), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bs/cbsemzrkik2b22jicx2oigreiyzk4y72j3nlfm67n3hm5oazp2al.py
# Topologically Sorted Source Nodes: [binary_cross_entropy_with_logits, sub], Original ATen: [aten.binary_cross_entropy_with_logits, aten.sub]
# Source node to ATen node mapping:
# binary_cross_entropy_with_logits => abs_1, exp, full_default, full_default_1, log1p, mean, minimum, mul, neg, sub_2, sub_3
# sub => sub
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 4, 4, 4], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %sub : [num_users=3] = call_function[target=torch.ops.aten.sub.Tensor](args = (%getitem_2, %getitem), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%full_default, %sub), kwargs = {})
# %full_default_1 : [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_1, %sub), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%sub,), 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_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul, %sub_2), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sub_3,), kwargs = {})
triton_per_fused_binary_cross_entropy_with_logits_sub_1 = async_compile.triton('triton_per_fused_binary_cross_entropy_with_logits_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.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_sub_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_binary_cross_entropy_with_logits_sub_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)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp1 = tl.load(in_ptr1 + (r0), None)
tmp2 = tmp0 - tmp1
tmp3 = 0.0
tmp4 = tmp3 * tmp2
tmp5 = triton_helpers.minimum(tmp3, tmp2)
tmp6 = tl_math.abs(tmp2)
tmp7 = -tmp6
tmp8 = tl_math.exp(tmp7)
tmp9 = libdevice.log1p(tmp8)
tmp10 = tmp5 - tmp9
tmp11 = tmp4 - tmp10
tmp12 = tl.broadcast_to(tmp11, [RBLOCK])
tmp14 = triton_helpers.promote_to_tensor(tl.sum(tmp12, 0))
tmp15 = 256.0
tmp16 = tmp14 / tmp15
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([1], 0, tl.int32)), tmp16, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sort], Original ATen: [aten.sort]
stream0 = get_raw_stream(0)
triton_per_fused_sort_0.run(arg0_1, buf0, 64, 4, grid=grid(64), stream=stream0)
del arg0_1
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sort_1], Original ATen: [aten.sort]
triton_per_fused_sort_0.run(arg1_1, buf2, 64, 4, grid=grid(64), stream=stream0)
del arg1_1
buf4 = empty_strided_cuda((), (), torch.float32)
buf5 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [binary_cross_entropy_with_logits, sub], Original ATen: [aten.binary_cross_entropy_with_logits, aten.sub]
triton_per_fused_binary_cross_entropy_with_logits_sub_1.run(buf5, buf2, buf0, 1, 256, grid=grid(1), stream=stream0)
del buf0
del buf2
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
class WasLoss(nn.Module):
def __init__(self):
super(WasLoss, self).__init__()
self.MSEls = torch.nn.BCEWithLogitsLoss()
def forward(self, true_data, fake_data):
SLX, _ = torch.sort(true_data, 0)
SLG, _ = torch.sort(fake_data, 0)
return self.MSEls(SLG - SLX, torch.ones_like(SLX))
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_sort_0(in_ptr0, out_ptr0, xnumel, rnumel, XBLOCK: tl.
constexpr):
xnumel = 64
RBLOCK: tl.constexpr = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * r1), xmask, other=0.0)
tmp1 = r1
tmp2 = tmp1.to(tl.int16)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5, _tmp6 = triton_helpers.sort_with_index(tmp3, tmp4, None, 1,
stable=False, descending=False)
tl.store(out_ptr0 + (x0 + 64 * r1), tmp5, xmask)
@triton.jit
def triton_per_fused_binary_cross_entropy_with_logits_sub_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)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp1 = tl.load(in_ptr1 + r0, None)
tmp2 = tmp0 - tmp1
tmp3 = 0.0
tmp4 = tmp3 * tmp2
tmp5 = triton_helpers.minimum(tmp3, tmp2)
tmp6 = tl_math.abs(tmp2)
tmp7 = -tmp6
tmp8 = tl_math.exp(tmp7)
tmp9 = libdevice.log1p(tmp8)
tmp10 = tmp5 - tmp9
tmp11 = tmp4 - tmp10
tmp12 = tl.broadcast_to(tmp11, [RBLOCK])
tmp14 = triton_helpers.promote_to_tensor(tl.sum(tmp12, 0))
tmp15 = 256.0
tmp16 = tmp14 / tmp15
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([1], 0, tl.int32), tmp16, None)
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_sort_0[grid(64)](arg0_1, buf0, 64, 4, XBLOCK=8,
num_warps=2, num_stages=1)
del arg0_1
buf2 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_per_fused_sort_0[grid(64)](arg1_1, buf2, 64, 4, XBLOCK=8,
num_warps=2, num_stages=1)
del arg1_1
buf4 = empty_strided_cuda((), (), torch.float32)
buf5 = buf4
del buf4
triton_per_fused_binary_cross_entropy_with_logits_sub_1[grid(1)](buf5,
buf2, buf0, 1, 256, num_warps=2, num_stages=1)
del buf0
del buf2
return buf5,
class WasLossNew(nn.Module):
def __init__(self):
super(WasLossNew, self).__init__()
self.MSEls = torch.nn.BCEWithLogitsLoss()
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
Johnson-yue/RS-GAN
|
WasLoss
| false
| 8,372
|
[
"MIT"
] | 26
|
8e8723045d63d8f9a4b510800cd909e7a6e3d195
|
https://github.com/Johnson-yue/RS-GAN/tree/8e8723045d63d8f9a4b510800cd909e7a6e3d195
|
Actor
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/4g/c4guhk7x6skkidedvs2gxz2kcu6gb76l3ig5crjjvjtzvnjlhlte.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=[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')
# kernel path: runs/run_shard_1/inductor_cache/7m/c7mvqxmsbygbhwwz3nonbbvjqbbvum3cmwg6hr3pmkhq3cffqzsx.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x_2 => tanh
# Graph fragment:
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%view_5,), kwargs = {})
triton_poi_fused_tanh_1 = async_compile.triton('triton_poi_fused_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_tanh_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_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf0 # reuse
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf7, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf2 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf3, primals_5, buf6, 256, grid=grid(256), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.tanh]
triton_poi_fused_tanh_1.run(buf5, primals_7, 256, grid=grid(256), stream=stream0)
del primals_7
return (buf5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(buf3, (64, 4), (4, 1), 0), buf5, 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((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
class Actor(nn.Module):
def __init__(self, n_obs, action_dim, hidden_size, init_w=0.003):
super(Actor, self).__init__()
self.linear1 = nn.Linear(n_obs, hidden_size)
self.linear2 = nn.Linear(hidden_size, hidden_size)
self.linear3 = nn.Linear(hidden_size, action_dim)
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
def forward(self, x):
x = F.relu(self.linear1(x))
x = F.relu(self.linear2(x))
x = torch.tanh(self.linear3(x))
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_obs': 4, 'action_dim': 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
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 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_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf0
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf1,
primals_2, buf7, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf2
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(256)](buf3,
primals_5, buf6, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf4
triton_poi_fused_tanh_1[grid(256)](buf5, primals_7, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_7
return buf5, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 4), (4, 1), 0), reinterpret_tensor(
buf3, (64, 4), (4, 1), 0), buf5, primals_6, buf6, primals_4, buf7
class ActorNew(nn.Module):
def __init__(self, n_obs, action_dim, hidden_size, init_w=0.003):
super(ActorNew, self).__init__()
self.linear1 = nn.Linear(n_obs, hidden_size)
self.linear2 = nn.Linear(hidden_size, hidden_size)
self.linear3 = nn.Linear(hidden_size, action_dim)
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_6 = self.linear3.weight
primals_7 = self.linear3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
|
JohnJim0816/rl-tutorials
|
Actor
| false
| 8,373
|
[
"MIT"
] | 16
|
e99daea815da85f9f25dff2d01b030249a203d22
|
https://github.com/JohnJim0816/rl-tutorials/tree/e99daea815da85f9f25dff2d01b030249a203d22
|
Mish
|
# 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/dq/cdqlooiebjg7blxgexhesa4fqwpmlazofexpzvqxzlikwith45aw.py
# Topologically Sorted Source Nodes: [softplus, tanh, mul_], Original ATen: [aten.softplus, aten.tanh, aten.mul]
# Source node to ATen node mapping:
# mul_ => mul
# softplus => exp, gt, log1p, where
# tanh => tanh
# Graph fragment:
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%arg0_1, 20), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%arg0_1,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %arg0_1, %log1p), kwargs = {})
# %tanh : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%where,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %tanh), kwargs = {})
# %copy_ : [num_users=1] = call_function[target=torch.ops.aten.copy_.default](args = (%arg0_1, %mul), kwargs = {})
triton_poi_fused_mul_softplus_tanh_0 = async_compile.triton('triton_poi_fused_mul_softplus_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_mul_softplus_tanh_0', 'mutated_arg_names': ['in_ptr0', 'out_ptr1'], '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_softplus_tanh_0(in_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 20.0
tmp2 = tmp0 > tmp1
tmp3 = tl_math.exp(tmp0)
tmp4 = libdevice.log1p(tmp3)
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = libdevice.tanh(tmp5)
tmp7 = tmp0 * tmp6
tl.store(out_ptr1 + (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)
# Topologically Sorted Source Nodes: [softplus, tanh, mul_], Original ATen: [aten.softplus, aten.tanh, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_softplus_tanh_0.run(arg0_1, arg0_1, 256, grid=grid(256), stream=stream0)
return (arg0_1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.utils.data
from torchvision.transforms import functional as F
import torch.nn as nn
import torch.nn.functional as F
from math import sqrt as sqrt
from itertools import product as product
class Mish(nn.Module):
def forward(self, x):
return x.mul_(F.softplus(x).tanh())
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.utils.data
import torch.nn as nn
from math import sqrt as sqrt
from itertools import product as product
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
@triton.jit
def triton_poi_fused_mul_softplus_tanh_0(in_ptr0, out_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 20.0
tmp2 = tmp0 > tmp1
tmp3 = tl_math.exp(tmp0)
tmp4 = libdevice.log1p(tmp3)
tmp5 = tl.where(tmp2, tmp0, tmp4)
tmp6 = libdevice.tanh(tmp5)
tmp7 = tmp0 * tmp6
tl.store(out_ptr1 + 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)
get_raw_stream(0)
triton_poi_fused_mul_softplus_tanh_0[grid(256)](arg0_1, arg0_1, 256,
XBLOCK=128, num_warps=4, num_stages=1)
return arg0_1,
class MishNew(nn.Module):
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Het-Shah/Monk_Object_Detection
|
Mish
| false
| 8,374
|
[
"Apache-2.0"
] | 15
|
1d7a07193ea3455221caa41d07c33c81d50c6b3f
|
https://github.com/Het-Shah/Monk_Object_Detection/tree/1d7a07193ea3455221caa41d07c33c81d50c6b3f
|
AttentionPool2d
|
# 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/7f/c7fequtewyyumb6i6j3rrvf5f3u56e4mlexmunj3gqmbnqklfkfc.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], 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': [], '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_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.sum(tmp3, 1)[:, None]
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ut/cutvvo5gthqlxnth7x2gozzelpebm6mkfykvs3eea66jzcoftrid.py
# Topologically Sorted Source Nodes: [x_1, x_2], Original ATen: [aten.cat, aten.add]
# Source node to ATen node mapping:
# x_1 => cat
# x_2 => add
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%mean, %view], -1), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%cat, %unsqueeze), kwargs = {})
triton_poi_fused_add_cat_1 = async_compile.triton('triton_poi_fused_add_cat_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[512],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*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_cat_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_cat_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 17
x3 = (xindex // 17)
x4 = xindex % 68
x5 = xindex
tmp15 = tl.load(in_ptr2 + (x4), xmask, eviction_policy='evict_last')
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 + (x3), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = 16.0
tmp7 = tmp5 / tmp6
tmp8 = tl.full(tmp7.shape, 0.0, tmp7.dtype)
tmp9 = tl.where(tmp4, tmp7, tmp8)
tmp10 = tmp0 >= tmp3
tmp11 = tl.full([1], 17, tl.int64)
tmp12 = tmp0 < tmp11
tmp13 = tl.load(in_ptr1 + ((16*x3) + ((-1) + x0)), tmp10 & xmask, eviction_policy='evict_last', other=0.0)
tmp14 = tl.where(tmp4, tmp9, tmp13)
tmp16 = tmp14 + tmp15
tl.store(out_ptr0 + (x5), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/tb/ctbcuetqinne2ybb7dyjjonbo4g7eywxln2njl4yvaix5nhrmlqe.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 = (%getitem, 0.7071067811865475), 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=[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_mul_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_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 68)
x3 = xindex % 68
x1 = (xindex // 17) % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x3 + (204*x2)), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x4), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/u7/cu7lp447lusg7bljtnryaylzcwatkewad7pag72t62px64ruckb5.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 = (%getitem_1, 0.7071067811865475), 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=[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_mul_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_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 68)
x3 = xindex % 68
x1 = (xindex // 17) % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (68 + x3 + (204*x2)), xmask)
tmp1 = tl.load(in_ptr1 + (4 + x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + (x4), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/jx/cjxgbr5u5yrn4nmcouanmubdpdl5ucyhqcmf5viwou3bg3kq3hf5.py
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# softmax => amax, div, exp, sub, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%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 = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_per_fused__softmax_4 = async_compile.triton('triton_per_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.persistent_reduction(
size_hints=[128, 32],
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__softmax_4', '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_4(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 68
rnumel = 17
RBLOCK: tl.constexpr = 32
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 + (17*x0)), rmask & xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(rmask & xmask, tmp1, float("-inf"))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp5 = tmp0 - tmp4
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = tl.where(rmask & xmask, tmp7, 0)
tmp10 = tl.sum(tmp9, 1)[:, None]
tmp11 = tmp6 / tmp10
tl.store(out_ptr2 + (r1 + (17*x0)), tmp11, rmask & xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/de/cdecumlqx7ysledzlk3rgvm7fhgo53wpmdrm5b56wmh5fqhrlzn6.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_3 => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%add, %primals_3, %primals_4, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_5 = async_compile.triton('triton_poi_fused_convolution_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*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_5', '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_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 816
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 17) % 12
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/ql/cqlaazo6wjsgfps3dbx7jj6ckf5lakc55nzareoaextgqyt2ygiv.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_5 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%view_9, %primals_5, %primals_6, [1], [0], [1], False, [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=[16, 32], 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_convolution_6', '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_convolution_6(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 17
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) + (68*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (17*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/i6/ci6xrsjecnjlhtmozchzuhczqusydcb3qfupvpr6t5ejlx6eeo5u.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_5 => convolution_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%view_9, %primals_5, %primals_6, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_7 = async_compile.triton('triton_poi_fused_convolution_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=[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_7', '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_7(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 17) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 17), (17, 1))
assert_size_stride(primals_3, (12, 4, 1), (4, 1, 1))
assert_size_stride(primals_4, (12, ), (1, ))
assert_size_stride(primals_5, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_6, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [mean], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(primals_1, buf0, 16, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_1, x_2], Original ATen: [aten.cat, aten.add]
triton_poi_fused_add_cat_1.run(buf0, primals_1, primals_2, buf1, 272, grid=grid(272), stream=stream0)
del buf0
del primals_1
del primals_2
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_3, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 12, 17), (204, 17, 1))
buf3 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
triton_poi_fused_mul_2.run(buf2, primals_4, buf3, 272, grid=grid(272), stream=stream0)
buf4 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_1], Original ATen: [aten.mul]
triton_poi_fused_mul_3.run(buf2, primals_4, buf4, 272, grid=grid(272), stream=stream0)
buf5 = empty_strided_cuda((4, 17, 17), (289, 17, 1), torch.float32)
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 17, 4), (68, 1, 17), 0), buf4, out=buf5)
buf8 = empty_strided_cuda((4, 17, 17), (289, 17, 1), torch.float32)
# Topologically Sorted Source Nodes: [softmax], Original ATen: [aten._softmax]
triton_per_fused__softmax_4.run(buf5, buf8, 68, 17, grid=grid(68), stream=stream0)
del buf5
buf9 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
triton_poi_fused_convolution_5.run(buf9, primals_4, 816, grid=grid(816), stream=stream0)
del primals_4
buf10 = empty_strided_cuda((4, 17, 4), (68, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [a], Original ATen: [aten.bmm]
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (4, 17, 4), (204, 1, 17), 136), out=buf10)
buf11 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.convolution]
triton_poi_fused_convolution_6.run(buf10, buf11, 16, 17, grid=grid(16, 17), stream=stream0)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.convolution]
buf12 = extern_kernels.convolution(buf11, primals_5, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf12, (4, 4, 17), (68, 17, 1))
del buf11
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.convolution]
triton_poi_fused_convolution_7.run(buf13, primals_6, 272, grid=grid(272), stream=stream0)
del primals_6
return (reinterpret_tensor(buf13, (4, 4), (68, 17), 0), primals_3, primals_5, buf1, buf8, reinterpret_tensor(buf10, (4, 4, 17), (68, 1, 4), 0), reinterpret_tensor(buf9, (4, 4, 17), (204, 17, 1), 136), buf3, reinterpret_tensor(buf4, (4, 17, 4), (68, 1, 17), 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, 17), (17, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((12, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((12, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import math
import torch
from torch import nn
import torch as th
def conv_nd(dims, *args, **kwargs):
"""
Create a 1D, 2D, or 3D convolution module.
"""
if dims == 1:
return nn.Conv1d(*args, **kwargs)
elif dims == 2:
return nn.Conv2d(*args, **kwargs)
elif dims == 3:
return nn.Conv3d(*args, **kwargs)
raise ValueError(f'unsupported dimensions: {dims}')
class QKVAttention(nn.Module):
"""
A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
"""
def __init__(self, n_heads):
super().__init__()
self.n_heads = n_heads
def forward(self, qkv, encoder_kv=None):
"""
Apply QKV attention.
:param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
:return: an [N x (H * C) x T] tensor after attention.
"""
bs, width, length = qkv.shape
assert width % (3 * self.n_heads) == 0
ch = width // (3 * self.n_heads)
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch,
dim=1)
if encoder_kv is not None:
assert encoder_kv.shape[1] == self.n_heads * ch * 2
ek, ev = encoder_kv.reshape(bs * self.n_heads, ch * 2, -1).split(ch
, dim=1)
k = th.cat([ek, k], dim=-1)
v = th.cat([ev, v], dim=-1)
scale = 1 / math.sqrt(math.sqrt(ch))
weight = th.einsum('bct,bcs->bts', q * scale, k * scale)
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
a = th.einsum('bts,bcs->bct', weight, v)
return a.reshape(bs, -1, length)
class AttentionPool2d(nn.Module):
"""
Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
"""
def __init__(self, spacial_dim: 'int', embed_dim: 'int',
num_heads_channels: 'int', output_dim: 'int'=None):
super().__init__()
self.positional_embedding = nn.Parameter(th.randn(embed_dim,
spacial_dim ** 2 + 1) / embed_dim ** 0.5)
self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
self.num_heads = embed_dim // num_heads_channels
self.attention = QKVAttention(self.num_heads)
def forward(self, x):
b, c, *_spatial = x.shape
x = x.reshape(b, c, -1)
x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1)
x = x + self.positional_embedding[None, :, :]
x = self.qkv_proj(x)
x = self.attention(x)
x = self.c_proj(x)
return x[:, :, 0]
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'spacial_dim': 4, 'embed_dim': 4, 'num_heads_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
import math
from torch import nn
import torch as th
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_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])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_poi_fused_add_cat_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 17
x3 = xindex // 17
x4 = xindex % 68
x5 = xindex
tmp15 = tl.load(in_ptr2 + x4, xmask, eviction_policy='evict_last')
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + x3, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp6 = 16.0
tmp7 = tmp5 / tmp6
tmp8 = tl.full(tmp7.shape, 0.0, tmp7.dtype)
tmp9 = tl.where(tmp4, tmp7, tmp8)
tmp10 = tmp0 >= tmp3
tl.full([1], 17, tl.int64)
tmp13 = tl.load(in_ptr1 + (16 * x3 + (-1 + x0)), tmp10 & xmask,
eviction_policy='evict_last', other=0.0)
tmp14 = tl.where(tmp4, tmp9, tmp13)
tmp16 = tmp14 + tmp15
tl.store(out_ptr0 + x5, tmp16, xmask)
@triton.jit
def triton_poi_fused_mul_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 68
x3 = xindex % 68
x1 = xindex // 17 % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x3 + 204 * x2), xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + x4, tmp4, xmask)
@triton.jit
def triton_poi_fused_mul_3(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 68
x3 = xindex % 68
x1 = xindex // 17 % 4
x4 = xindex
tmp0 = tl.load(in_ptr0 + (68 + x3 + 204 * x2), xmask)
tmp1 = tl.load(in_ptr1 + (4 + x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.7071067811865475
tmp4 = tmp2 * tmp3
tl.store(out_ptr0 + x4, tmp4, xmask)
@triton.jit
def triton_per_fused__softmax_4(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 68
rnumel = 17
RBLOCK: tl.constexpr = 32
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 + 17 * x0), rmask & xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(rmask & xmask, tmp1, float('-inf'))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp5 = tmp0 - tmp4
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = tl.where(rmask & xmask, tmp7, 0)
tmp10 = tl.sum(tmp9, 1)[:, None]
tmp11 = tmp6 / tmp10
tl.store(out_ptr2 + (r1 + 17 * x0), tmp11, rmask & xmask)
@triton.jit
def triton_poi_fused_convolution_5(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 816
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 17 % 12
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_convolution_6(in_ptr0, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 17
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 + 68 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 17 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_convolution_7(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 272
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 17 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, 17), (17, 1))
assert_size_stride(primals_3, (12, 4, 1), (4, 1, 1))
assert_size_stride(primals_4, (12,), (1,))
assert_size_stride(primals_5, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_6, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
get_raw_stream(0)
triton_per_fused_mean_0[grid(16)](primals_1, buf0, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
buf1 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
triton_poi_fused_add_cat_1[grid(272)](buf0, primals_1, primals_2,
buf1, 272, XBLOCK=128, num_warps=4, num_stages=1)
del buf0
del primals_1
del primals_2
buf2 = extern_kernels.convolution(buf1, primals_3, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 12, 17), (204, 17, 1))
buf3 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
triton_poi_fused_mul_2[grid(272)](buf2, primals_4, buf3, 272,
XBLOCK=256, num_warps=4, num_stages=1)
buf4 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
triton_poi_fused_mul_3[grid(272)](buf2, primals_4, buf4, 272,
XBLOCK=256, num_warps=4, num_stages=1)
buf5 = empty_strided_cuda((4, 17, 17), (289, 17, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 17, 4), (68, 1, 17),
0), buf4, out=buf5)
buf8 = empty_strided_cuda((4, 17, 17), (289, 17, 1), torch.float32)
triton_per_fused__softmax_4[grid(68)](buf5, buf8, 68, 17, XBLOCK=1,
num_warps=2, num_stages=1)
del buf5
buf9 = buf2
del buf2
triton_poi_fused_convolution_5[grid(816)](buf9, primals_4, 816,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_4
buf10 = empty_strided_cuda((4, 17, 4), (68, 4, 1), torch.float32)
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (4, 17, 4), (204,
1, 17), 136), out=buf10)
buf11 = empty_strided_cuda((4, 4, 17), (68, 17, 1), torch.float32)
triton_poi_fused_convolution_6[grid(16, 17)](buf10, buf11, 16, 17,
XBLOCK=32, YBLOCK=16, num_warps=4, num_stages=1)
buf12 = extern_kernels.convolution(buf11, primals_5, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf12, (4, 4, 17), (68, 17, 1))
del buf11
buf13 = buf12
del buf12
triton_poi_fused_convolution_7[grid(272)](buf13, primals_6, 272,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_6
return reinterpret_tensor(buf13, (4, 4), (68, 17), 0
), primals_3, primals_5, buf1, buf8, reinterpret_tensor(buf10, (4,
4, 17), (68, 1, 4), 0), reinterpret_tensor(buf9, (4, 4, 17), (204,
17, 1), 136), buf3, reinterpret_tensor(buf4, (4, 17, 4), (68, 1, 17), 0
)
def conv_nd(dims, *args, **kwargs):
"""
Create a 1D, 2D, or 3D convolution module.
"""
if dims == 1:
return nn.Conv1d(*args, **kwargs)
elif dims == 2:
return nn.Conv2d(*args, **kwargs)
elif dims == 3:
return nn.Conv3d(*args, **kwargs)
raise ValueError(f'unsupported dimensions: {dims}')
class QKVAttention(nn.Module):
"""
A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
"""
def __init__(self, n_heads):
super().__init__()
self.n_heads = n_heads
def forward(self, qkv, encoder_kv=None):
"""
Apply QKV attention.
:param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
:return: an [N x (H * C) x T] tensor after attention.
"""
bs, width, length = qkv.shape
assert width % (3 * self.n_heads) == 0
ch = width // (3 * self.n_heads)
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch,
dim=1)
if encoder_kv is not None:
assert encoder_kv.shape[1] == self.n_heads * ch * 2
ek, ev = encoder_kv.reshape(bs * self.n_heads, ch * 2, -1).split(ch
, dim=1)
k = th.cat([ek, k], dim=-1)
v = th.cat([ev, v], dim=-1)
scale = 1 / math.sqrt(math.sqrt(ch))
weight = th.einsum('bct,bcs->bts', q * scale, k * scale)
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
a = th.einsum('bts,bcs->bct', weight, v)
return a.reshape(bs, -1, length)
class AttentionPool2dNew(nn.Module):
"""
Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
"""
def __init__(self, spacial_dim: 'int', embed_dim: 'int',
num_heads_channels: 'int', output_dim: 'int'=None):
super().__init__()
self.positional_embedding = nn.Parameter(th.randn(embed_dim,
spacial_dim ** 2 + 1) / embed_dim ** 0.5)
self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
self.num_heads = embed_dim // num_heads_channels
self.attention = QKVAttention(self.num_heads)
def forward(self, input_0):
primals_2 = self.positional_embedding
primals_3 = self.qkv_proj.weight
primals_4 = self.qkv_proj.bias
primals_5 = self.c_proj.weight
primals_6 = self.c_proj.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
|
Jack000/glid-3
|
AttentionPool2d
| false
| 8,375
|
[
"MIT"
] | 31
|
4a18efc2785339ebc743e149a7955e34fff436fb
|
https://github.com/Jack000/glid-3/tree/4a18efc2785339ebc743e149a7955e34fff436fb
|
GaussianKernel
|
# 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/r4/cr4iard5mldabudni5dsdweauytlr3bx7biikcndnzx42fwduxw5.py
# Topologically Sorted Source Nodes: [neg, mul, truediv, exp], Original ATen: [aten.neg, aten.mul, aten.div, aten.exp]
# Source node to ATen node mapping:
# exp => exp
# mul => mul
# neg => neg
# truediv => div
# Graph fragment:
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%arg0_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%neg, %arg0_1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul, -11.541560327111707), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%div,), kwargs = {})
triton_poi_fused_div_exp_mul_neg_0 = async_compile.triton('triton_poi_fused_div_exp_mul_neg_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_exp_mul_neg_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_div_exp_mul_neg_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = -tmp0
tmp2 = tmp1 * tmp0
tmp3 = -0.08664339756999316
tmp4 = tmp2 * tmp3
tmp5 = tl_math.exp(tmp4)
tl.store(out_ptr0 + (x0), tmp5, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [neg, mul, truediv, exp], Original ATen: [aten.neg, aten.mul, aten.div, aten.exp]
stream0 = get_raw_stream(0)
triton_poi_fused_div_exp_mul_neg_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
import torch.utils.data
class GaussianKernel(nn.Module):
def __init__(self, delta_var, pmaps_threshold):
super().__init__()
self.delta_var = delta_var
self.two_sigma = delta_var * delta_var / -math.log(pmaps_threshold)
def forward(self, dist_map):
return torch.exp(-dist_map * dist_map / self.two_sigma)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'delta_var': 4, 'pmaps_threshold': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
import math
import torch.nn as nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_div_exp_mul_neg_0(in_ptr0, out_ptr0, xnumel, XBLOCK:
tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = -tmp0
tmp2 = tmp1 * tmp0
tmp3 = -0.08664339756999316
tmp4 = tmp2 * tmp3
tmp5 = tl_math.exp(tmp4)
tl.store(out_ptr0 + x0, tmp5, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_div_exp_mul_neg_0[grid(256)](arg0_1, buf0, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class GaussianKernelNew(nn.Module):
def __init__(self, delta_var, pmaps_threshold):
super().__init__()
self.delta_var = delta_var
self.two_sigma = delta_var * delta_var / -math.log(pmaps_threshold)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
JonasHell/torch-em
|
GaussianKernel
| false
| 8,376
|
[
"MIT"
] | 13
|
2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
https://github.com/JonasHell/torch-em/tree/2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
Attention
|
# 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/5p/c5pxxtyrzkj54dkrsw6he2qmyvbmzkzejttdrjawolvlvpg3mxxx.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, %repeat], 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=[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_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 = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x3 = (xindex // 8)
x2 = (xindex // 32)
x4 = 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*x3) + 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*x2) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x4), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cv/ccvvrbl5xq2dop2s3ilfb3hyf275whlxxdyq7j2gky5t2pg6gozw.py
# Topologically Sorted Source Nodes: [tanh], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# tanh => tanh
# Graph fragment:
# %add_tensor : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default, %primals_4), kwargs = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_tensor,), kwargs = {})
triton_poi_fused_tanh_1 = async_compile.triton('triton_poi_fused_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_tanh_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_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/fp/cfp6aijow2g5is2smjrqj7kca6itit4xrxs2tseqvxiqv3qu5hs6.py
# Topologically Sorted Source Nodes: [alpha], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# alpha => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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 = 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/vk/cvkegrqrqvgc6l3q574j6alpmoropwpd76xcxahk7k7j2fjfbppc.py
# Topologically Sorted Source Nodes: [alpha], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# alpha => div, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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 = 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 = 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, 8), (8, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (1, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, primals_2, buf0, 128, grid=grid(128), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf0, (16, 8), (8, 1), 0), reinterpret_tensor(primals_3, (8, 4), (1, 8), 0), out=buf1)
del primals_3
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [tanh], Original ATen: [aten.tanh]
triton_poi_fused_tanh_1.run(buf2, primals_4, 64, grid=grid(64), stream=stream0)
del primals_4
buf3 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [o], Original ATen: [aten.mm]
extern_kernels.mm(buf2, reinterpret_tensor(primals_5, (4, 1), (1, 4), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [alpha], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf3, buf4, 16, grid=grid(16), stream=stream0)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [alpha], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf4, buf5, 16, grid=grid(16), stream=stream0)
buf6 = reinterpret_tensor(buf4, (4, 1, 4), (4, 4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [bmm], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf5, (4, 1, 4), (4, 0, 1), 0), primals_1, out=buf6)
del buf5
return (reinterpret_tensor(buf6, (4, 4), (4, 1), 0), reinterpret_tensor(buf0, (16, 8), (8, 1), 0), buf2, buf3, reinterpret_tensor(primals_1, (4, 4, 4), (16, 1, 4), 0), 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, 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((1, 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
from torch import nn
import torch.nn.functional as F
class Attention(nn.Module):
"""
Applies an attention mechanism on the output features from the decoder.
"""
def __init__(self, dim):
super(Attention, self).__init__()
self.dim = dim
self.linear1 = nn.Linear(dim * 2, dim)
self.linear2 = nn.Linear(dim, 1, bias=False)
def forward(self, hidden_state, encoder_outputs):
"""
Arguments:
hidden_state {Variable} -- batch_size x dim
encoder_outputs {Variable} -- batch_size x seq_len x dim
Returns:
Variable -- context vector of size batch_size x dim
"""
batch_size, seq_len, _ = encoder_outputs.size()
hidden_state = hidden_state.unsqueeze(1).repeat(1, seq_len, 1)
inputs = torch.cat((encoder_outputs, hidden_state), 2).view(-1,
self.dim * 2)
o = self.linear2(torch.tanh(self.linear1(inputs)))
e = o.view(batch_size, seq_len)
alpha = F.softmax(e, dim=1)
context = torch.bmm(alpha.unsqueeze(1), encoder_outputs).squeeze(1)
return context
def get_inputs():
return [torch.rand([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 libdevice, math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(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
x3 = xindex // 8
x2 = xindex // 32
x4 = 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 * x3 + 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 * x2 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x4, tmp10, xmask)
@triton.jit
def triton_poi_fused_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@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 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 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 = 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, 8), (8, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (1, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(128)](primals_1, primals_2, buf0, 128,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 8), (8, 1), 0),
reinterpret_tensor(primals_3, (8, 4), (1, 8), 0), out=buf1)
del primals_3
buf2 = buf1
del buf1
triton_poi_fused_tanh_1[grid(64)](buf2, primals_4, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(buf2, reinterpret_tensor(primals_5, (4, 1), (1, 4
), 0), out=buf3)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(16)](buf3, buf4, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused__softmax_3[grid(16)](buf4, buf5, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf6 = reinterpret_tensor(buf4, (4, 1, 4), (4, 4, 1), 0)
del buf4
extern_kernels.bmm(reinterpret_tensor(buf5, (4, 1, 4), (4, 0, 1), 0
), primals_1, out=buf6)
del buf5
return reinterpret_tensor(buf6, (4, 4), (4, 1), 0), reinterpret_tensor(buf0
, (16, 8), (8, 1), 0), buf2, buf3, reinterpret_tensor(primals_1, (4,
4, 4), (16, 1, 4), 0), primals_5
class AttentionNew(nn.Module):
"""
Applies an attention mechanism on the output features from the decoder.
"""
def __init__(self, dim):
super(AttentionNew, self).__init__()
self.dim = dim
self.linear1 = nn.Linear(dim * 2, dim)
self.linear2 = nn.Linear(dim, 1, bias=False)
def forward(self, input_0, input_1):
primals_3 = self.linear1.weight
primals_4 = self.linear1.bias
primals_5 = self.linear2.weight
primals_2 = input_0
primals_1 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
JiwanChung/tapm
|
Attention
| false
| 8,377
|
[
"MIT"
] | 14
|
ec42b139d1c012daccc55f85e67744488d526476
|
https://github.com/JiwanChung/tapm/tree/ec42b139d1c012daccc55f85e67744488d526476
|
Net
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/s5/cs5cii52tc3hbclol2cj5szknqlzoteg2pw4i5j57dyaz6f54ybl.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=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_1,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_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 = tl.sigmoid(tmp0)
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ox/coxzkq4kueijjjccjhdtvzhfquura4nj5lgusydc343b4la3telx.py
# Topologically Sorted Source Nodes: [input_2], Original ATen: [aten.view]
# Source node to ATen node mapping:
# input_2 => view_7
# Graph fragment:
# %view_7 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%view_6, [64, 4]), kwargs = {})
triton_poi_fused_view_1 = async_compile.triton('triton_poi_fused_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=[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_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_view_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 % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x1) + (16*((x1 % 4) // 4)) + (64*(((4*((x1 // 4) % 4)) + (x1 % 4)) // 16))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (3, 4), (4, 1))
assert_size_stride(primals_9, (3, ), (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: [input_1], 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: [sigmoid, mul_], Original ATen: [aten.sigmoid, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_sigmoid_0.run(buf0, buf1, 256, grid=grid(256), stream=stream0)
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [input_2], Original ATen: [aten.view]
triton_poi_fused_view_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (64, 4), (4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [input_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sigmoid_1, mul__1], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_0.run(buf3, buf4, 256, grid=grid(256), stream=stream0)
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [input_3], Original ATen: [aten.view]
triton_poi_fused_view_1.run(buf4, buf5, 256, grid=grid(256), stream=stream0)
buf6 = reinterpret_tensor(buf4, (64, 4), (4, 1), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [input_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, buf5, reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf6)
del primals_7
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sigmoid_2, mul__2], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_0.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
buf8 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [input_4], Original ATen: [aten.view]
triton_poi_fused_view_1.run(buf7, buf8, 256, grid=grid(256), stream=stream0)
del buf7
buf9 = empty_strided_cuda((64, 3), (3, 1), torch.float32)
# Topologically Sorted Source Nodes: [input_4], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_9, buf8, reinterpret_tensor(primals_8, (4, 3), (1, 4), 0), alpha=1, beta=1, out=buf9)
del primals_9
return (reinterpret_tensor(buf9, (4, 4, 4, 3), (48, 12, 3, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf0, buf2, buf3, buf5, buf6, buf8, primals_8, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((3, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((3, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, 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 Swish(nn.Module):
def __init__(self, inplace=True):
super(Swish, self).__init__()
self.inplace = inplace
def forward(self, x):
if self.inplace:
x.mul_(torch.sigmoid(x))
return x
else:
return x * torch.sigmoid(x)
class Net(torch.nn.Module):
def __init__(self, n_feature, n_hidden):
super(Net, self).__init__()
self.features = nn.Sequential()
self.features.add_module('hidden', torch.nn.Linear(n_feature, n_hidden)
)
self.features.add_module('active1', Swish())
self.features.add_module('hidden2', torch.nn.Linear(n_hidden, n_hidden)
)
self.features.add_module('active2', Swish())
self.features.add_module('hidden3', torch.nn.Linear(n_hidden, n_hidden)
)
self.features.add_module('active3', Swish())
self.features.add_module('predict', torch.nn.Linear(n_hidden, 3))
def forward(self, x):
return self.features(x)
def reset_parameters(self, verbose=False):
for module in self.modules():
if isinstance(module, self.__class__):
continue
if 'reset_parameters' in dir(module):
if callable(module.reset_parameters):
module.reset_parameters()
if verbose:
None
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_feature': 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.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_mul_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 = tl.sigmoid(tmp0)
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_view_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 % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x1 + 16 * (x1 % 4 // 4) + 64 * ((4 *
(x1 // 4 % 4) + x1 % 4) // 16)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (3, 4), (4, 1))
assert_size_stride(primals_9, (3,), (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_mul_sigmoid_0[grid(256)](buf0, buf1, 256, XBLOCK=
256, num_warps=4, num_stages=1)
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
triton_poi_fused_view_1[grid(256)](buf1, buf2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (64, 4), (4, 1), 0)
del buf1
extern_kernels.addmm(primals_5, buf2, reinterpret_tensor(primals_4,
(4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_0[grid(256)](buf3, buf4, 256, XBLOCK=
256, num_warps=4, num_stages=1)
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
triton_poi_fused_view_1[grid(256)](buf4, buf5, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf6 = reinterpret_tensor(buf4, (64, 4), (4, 1), 0)
del buf4
extern_kernels.addmm(primals_7, buf5, reinterpret_tensor(primals_6,
(4, 4), (1, 4), 0), alpha=1, beta=1, out=buf6)
del primals_7
buf7 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_0[grid(256)](buf6, buf7, 256, XBLOCK=
256, num_warps=4, num_stages=1)
buf8 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
triton_poi_fused_view_1[grid(256)](buf7, buf8, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf7
buf9 = empty_strided_cuda((64, 3), (3, 1), torch.float32)
extern_kernels.addmm(primals_9, buf8, reinterpret_tensor(primals_8,
(4, 3), (1, 4), 0), alpha=1, beta=1, out=buf9)
del primals_9
return reinterpret_tensor(buf9, (4, 4, 4, 3), (48, 12, 3, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf0, buf2, buf3, buf5, buf6, buf8, primals_8, primals_6, primals_4
class Swish(nn.Module):
def __init__(self, inplace=True):
super(Swish, self).__init__()
self.inplace = inplace
def forward(self, x):
if self.inplace:
x.mul_(torch.sigmoid(x))
return x
else:
return x * torch.sigmoid(x)
class NetNew(torch.nn.Module):
def __init__(self, n_feature, n_hidden):
super(NetNew, self).__init__()
self.features = nn.Sequential()
self.features.add_module('hidden', torch.nn.Linear(n_feature, n_hidden)
)
self.features.add_module('active1', Swish())
self.features.add_module('hidden2', torch.nn.Linear(n_hidden, n_hidden)
)
self.features.add_module('active2', Swish())
self.features.add_module('hidden3', torch.nn.Linear(n_hidden, n_hidden)
)
self.features.add_module('active3', Swish())
self.features.add_module('predict', torch.nn.Linear(n_hidden, 3))
def reset_parameters(self, verbose=False):
for module in self.modules():
if isinstance(module, self.__class__):
continue
if 'reset_parameters' in dir(module):
if callable(module.reset_parameters):
module.reset_parameters()
if verbose:
None
def forward(self, input_0):
primals_1 = self.features.hidden.weight
primals_2 = self.features.hidden.bias
primals_4 = self.features.hidden2.weight
primals_5 = self.features.hidden2.bias
primals_6 = self.features.hidden3.weight
primals_7 = self.features.hidden3.bias
primals_8 = self.features.predict.weight
primals_9 = self.features.predict.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
|
Jianxun-Wang/Physics-constrained-Bayesian-deep-learning
|
Net
| false
| 8,378
|
[
"MIT"
] | 24
|
cde0287f848f83c6def1fe409c67d7d4e14174da
|
https://github.com/Jianxun-Wang/Physics-constrained-Bayesian-deep-learning/tree/cde0287f848f83c6def1fe409c67d7d4e14174da
|
Block
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/qg/cqg7kxk2nitsge62at7xkllsb33tp5red44errkhdjtdt7osxswo.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.relu, aten.native_layer_norm]
# Source node to ATen node mapping:
# x => relu
# x_1 => add, rsqrt, var_mean
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%primals_1,), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%relu, [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_relu_0 = async_compile.triton('triton_poi_fused_native_layer_norm_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_relu_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_relu_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')
tmp3 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp1, tmp3)
tmp5 = tmp2 + tmp4
tmp7 = triton_helpers.maximum(tmp1, tmp6)
tmp8 = tmp5 + tmp7
tmp10 = triton_helpers.maximum(tmp1, tmp9)
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tmp2 - tmp13
tmp15 = tmp14 * tmp14
tmp16 = tmp4 - tmp13
tmp17 = tmp16 * tmp16
tmp18 = tmp15 + tmp17
tmp19 = tmp7 - tmp13
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp10 - tmp13
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp24 / tmp12
tmp26 = 1e-05
tmp27 = tmp25 + tmp26
tmp28 = libdevice.rsqrt(tmp27)
tl.store(out_ptr0 + (x0), tmp13, xmask)
tl.store(out_ptr1 + (x0), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/me/cmelpfzh2bnqoptwzovfjzudj7nxc3f5j65jzp52rbdpdznpxvvs.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.relu, aten.native_layer_norm]
# Source node to ATen node mapping:
# x => relu
# x_1 => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%primals_1,), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%relu, [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 = (%relu, %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_relu_1 = async_compile.triton('triton_poi_fused_native_layer_norm_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=[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_relu_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_relu_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)
tmp3 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = tmp2 - tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 * tmp7
tmp10 = tmp8 + tmp9
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/u3/cu3syptlgfek6uvujcqv7qkxhs257xmprzm64nz5ymayrh4qnxtj.py
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# x_3 => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%permute, %primals_4, %primals_5, [1], [1], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_2 = async_compile.triton('triton_poi_fused_convolution_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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': [], '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_convolution_2(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ae/cae5mpdh3qeemhrdtcccr7i77uuobw27a3v2ycnvmce3odjbmofa.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 = (%permute_1, %primals_1), 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=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 4
y1 = (yindex // 4)
tmp0 = tl.load(in_out_ptr0 + (x2 + (4*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (y0), ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.debug_barrier()
tl.store(in_out_ptr0 + (x2 + (4*y3)), tmp4, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4, 3), (12, 3, 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), (4, 1, 16), torch.float32)
buf1 = empty_strided_cuda((4, 4, 1), (4, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.relu, aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_relu_0.run(primals_1, 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, x_1], Original ATen: [aten.relu, aten.native_layer_norm]
triton_poi_fused_native_layer_norm_relu_1.run(primals_1, buf0, buf1, primals_2, primals_3, buf2, 64, grid=grid(64), stream=stream0)
del buf0
del buf1
del primals_2
del primals_3
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
triton_poi_fused_convolution_2.run(buf2, buf3, 16, 4, grid=grid(16, 4), stream=stream0)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf3, primals_4, stride=(1,), padding=(1,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf4, (4, 4, 4), (16, 4, 1))
del buf3
buf5 = reinterpret_tensor(buf4, (4, 4, 4), (16, 1, 4), 0); del buf4 # reuse
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
triton_poi_fused_add_3.run(buf5, primals_5, primals_1, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_5
return (buf5, primals_1, primals_4, reinterpret_tensor(buf2, (4, 4, 4), (16, 1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (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, 3), (12, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import nn
import torch.nn.functional as F
class Block(nn.Module):
def __init__(self, dim):
super(Block, self).__init__()
self.dim = dim
self.layer_norm = nn.LayerNorm(self.dim)
self.conv = nn.Conv1d(self.dim, self.dim, kernel_size=3, padding=1)
def forward(self, x):
x_orig = x
x = F.relu(x)
x = self.layer_norm(x)
x = x.transpose(1, 2)
x = self.conv(x)
x = x.transpose(1, 2)
return x + x_orig
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 libdevice
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_native_layer_norm_relu_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')
tmp3 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last')
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp1, tmp3)
tmp5 = tmp2 + tmp4
tmp7 = triton_helpers.maximum(tmp1, tmp6)
tmp8 = tmp5 + tmp7
tmp10 = triton_helpers.maximum(tmp1, tmp9)
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tmp2 - tmp13
tmp15 = tmp14 * tmp14
tmp16 = tmp4 - tmp13
tmp17 = tmp16 * tmp16
tmp18 = tmp15 + tmp17
tmp19 = tmp7 - tmp13
tmp20 = tmp19 * tmp19
tmp21 = tmp18 + tmp20
tmp22 = tmp10 - tmp13
tmp23 = tmp22 * tmp22
tmp24 = tmp21 + tmp23
tmp25 = tmp24 / tmp12
tmp26 = 1e-05
tmp27 = tmp25 + tmp26
tmp28 = libdevice.rsqrt(tmp27)
tl.store(out_ptr0 + x0, tmp13, xmask)
tl.store(out_ptr1 + x0, tmp28, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_relu_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)
tmp3 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = tmp2 - tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 * tmp7
tmp10 = tmp8 + tmp9
tl.store(out_ptr0 + x2, tmp10, xmask)
@triton.jit
def triton_poi_fused_convolution_2(in_ptr0, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_3(in_out_ptr0, in_ptr0, in_ptr1, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 4
y1 = yindex // 4
tmp0 = tl.load(in_out_ptr0 + (x2 + 4 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + y0, ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.debug_barrier()
tl.store(in_out_ptr0 + (x2 + 4 * y3), tmp4, xmask & ymask)
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,), (1,))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4, 3), (12, 3, 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), (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_relu_0[grid(16)](primals_1, 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_relu_1[grid(64)](primals_1, buf0,
buf1, primals_2, primals_3, buf2, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf0
del buf1
del primals_2
del primals_3
buf3 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_convolution_2[grid(16, 4)](buf2, buf3, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf4 = extern_kernels.convolution(buf3, primals_4, stride=(1,),
padding=(1,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf4, (4, 4, 4), (16, 4, 1))
del buf3
buf5 = reinterpret_tensor(buf4, (4, 4, 4), (16, 1, 4), 0)
del buf4
triton_poi_fused_add_3[grid(16, 4)](buf5, primals_5, primals_1, 16,
4, XBLOCK=2, YBLOCK=16, num_warps=1, num_stages=1)
del primals_5
return buf5, primals_1, primals_4, reinterpret_tensor(buf2, (4, 4, 4),
(16, 1, 4), 0)
class BlockNew(nn.Module):
def __init__(self, dim):
super(BlockNew, self).__init__()
self.dim = dim
self.layer_norm = nn.LayerNorm(self.dim)
self.conv = nn.Conv1d(self.dim, self.dim, kernel_size=3, padding=1)
def forward(self, input_0):
primals_2 = self.layer_norm.weight
primals_3 = self.layer_norm.bias
primals_4 = self.conv.weight
primals_5 = self.conv.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
JiwanChung/tapm
|
Block
| false
| 8,379
|
[
"MIT"
] | 14
|
ec42b139d1c012daccc55f85e67744488d526476
|
https://github.com/JiwanChung/tapm/tree/ec42b139d1c012daccc55f85e67744488d526476
|
RLFeatPreprocessNet
|
# 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/vs/cvs4eu6m5jgj773xwkaazwcptgesom3eznf44ursaoowub57ko6g.py
# Topologically Sorted Source Nodes: [output_feat], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# output_feat => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%view_1, %view_3, %view_5, %view_7], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 4) % 16
x0 = xindex % 4
x2 = (xindex // 64)
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 + (4*x1) + (16*x2)), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr1 + (x0 + (4*((-4) + x1)) + (16*x2)), tmp9 & xmask, other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1], 12, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tmp11 & tmp13
tmp15 = tl.load(in_ptr2 + (x0 + (4*((-8) + x1)) + (16*x2)), tmp14 & xmask, other=0.0)
tmp16 = tmp0 >= tmp12
tmp17 = tl.full([1], 16, tl.int64)
tmp18 = tmp0 < tmp17
tmp19 = tl.load(in_ptr3 + (x0 + (4*((-12) + x1)) + (16*x2)), tmp16 & xmask, 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')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (1, 4), (4, 1))
assert_size_stride(primals_2, (1, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_7, (1, 4), (4, 1))
assert_size_stride(primals_8, (1, ), (1, ))
assert_size_stride(primals_9, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_10, (1, 4), (4, 1))
assert_size_stride(primals_11, (1, ), (1, ))
assert_size_stride(primals_12, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf1 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [resized_obj], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_1
del primals_2
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [resized_embed], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_4
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [resized_box], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_9, (64, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf5)
del primals_7
del primals_8
buf7 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [resized_overlap], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, reinterpret_tensor(primals_12, (64, 4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 1), (1, 4), 0), alpha=1, beta=1, out=buf7)
del primals_10
del primals_11
buf8 = empty_strided_cuda((4, 16, 4, 1), (64, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [output_feat], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(buf1, buf3, buf5, buf7, buf8, 256, grid=grid(256), stream=stream0)
del buf1
del buf3
del buf5
del buf7
return (buf8, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), reinterpret_tensor(primals_9, (64, 4), (4, 1), 0), reinterpret_tensor(primals_12, (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((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.utils.data
class RLFeatPreprocessNet(nn.Module):
"""
Preprocess Features
1. visual feature
2. label prediction embed feature
3. box embed
4. overlap embed
"""
def __init__(self, feat_size, embed_size, bbox_size, overlap_size,
output_size):
super(RLFeatPreprocessNet, self).__init__()
self.feature_size = feat_size
self.embed_size = embed_size
self.box_info_size = bbox_size
self.overlap_info_size = overlap_size
self.output_size = output_size
self.resize_feat = nn.Linear(self.feature_size, int(output_size / 4))
self.resize_embed = nn.Linear(self.embed_size, int(output_size / 4))
self.resize_box = nn.Linear(self.box_info_size, int(output_size / 4))
self.resize_overlap = nn.Linear(self.overlap_info_size, int(
output_size / 4))
self.resize_feat.weight.data.normal_(0, 0.001)
self.resize_embed.weight.data.normal_(0, 0.01)
self.resize_box.weight.data.normal_(0, 1)
self.resize_overlap.weight.data.normal_(0, 1)
self.resize_feat.bias.data.zero_()
self.resize_embed.bias.data.zero_()
self.resize_box.bias.data.zero_()
self.resize_overlap.bias.data.zero_()
def forward(self, obj_feat, obj_embed, box_info, overlap_info):
resized_obj = self.resize_feat(obj_feat)
resized_embed = self.resize_embed(obj_embed)
resized_box = self.resize_box(box_info)
resized_overlap = self.resize_overlap(overlap_info)
output_feat = torch.cat((resized_obj, resized_embed, resized_box,
resized_overlap), 1)
return output_feat
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4]), torch.rand(
[4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'feat_size': 4, 'embed_size': 4, 'bbox_size': 4,
'overlap_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
import torch.nn as nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 4 % 16
x0 = xindex % 4
x2 = xindex // 64
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 + 4 * x1 + 16 * x2), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tmp6 & tmp8
tmp10 = tl.load(in_ptr1 + (x0 + 4 * (-4 + x1) + 16 * x2), tmp9 & xmask,
other=0.0)
tmp11 = tmp0 >= tmp7
tmp12 = tl.full([1], 12, tl.int64)
tmp13 = tmp0 < tmp12
tmp14 = tmp11 & tmp13
tmp15 = tl.load(in_ptr2 + (x0 + 4 * (-8 + x1) + 16 * x2), tmp14 & xmask,
other=0.0)
tmp16 = tmp0 >= tmp12
tl.full([1], 16, tl.int64)
tmp19 = tl.load(in_ptr3 + (x0 + 4 * (-12 + x1) + 16 * x2), tmp16 &
xmask, 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)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (1, 4), (4, 1))
assert_size_stride(primals_2, (1,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1,), (1,))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_7, (1, 4), (4, 1))
assert_size_stride(primals_8, (1,), (1,))
assert_size_stride(primals_9, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_10, (1, 4), (4, 1))
assert_size_stride(primals_11, (1,), (1,))
assert_size_stride(primals_12, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf1 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (64,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 1), (1, 4), 0
), alpha=1, beta=1, out=buf1)
del primals_1
del primals_2
buf3 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_6, (64,
4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0
), alpha=1, beta=1, out=buf3)
del primals_4
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_9, (64,
4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 1), (1, 4), 0
), alpha=1, beta=1, out=buf5)
del primals_7
del primals_8
buf7 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_11, reinterpret_tensor(primals_12, (64,
4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 1), (1, 4),
0), alpha=1, beta=1, out=buf7)
del primals_10
del primals_11
buf8 = empty_strided_cuda((4, 16, 4, 1), (64, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(256)](buf1, buf3, buf5, buf7, buf8, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del buf1
del buf3
del buf5
del buf7
return buf8, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(primals_6, (64, 4), (4, 1), 0
), reinterpret_tensor(primals_9, (64, 4), (4, 1), 0
), reinterpret_tensor(primals_12, (64, 4), (4, 1), 0)
class RLFeatPreprocessNetNew(nn.Module):
"""
Preprocess Features
1. visual feature
2. label prediction embed feature
3. box embed
4. overlap embed
"""
def __init__(self, feat_size, embed_size, bbox_size, overlap_size,
output_size):
super(RLFeatPreprocessNetNew, self).__init__()
self.feature_size = feat_size
self.embed_size = embed_size
self.box_info_size = bbox_size
self.overlap_info_size = overlap_size
self.output_size = output_size
self.resize_feat = nn.Linear(self.feature_size, int(output_size / 4))
self.resize_embed = nn.Linear(self.embed_size, int(output_size / 4))
self.resize_box = nn.Linear(self.box_info_size, int(output_size / 4))
self.resize_overlap = nn.Linear(self.overlap_info_size, int(
output_size / 4))
self.resize_feat.weight.data.normal_(0, 0.001)
self.resize_embed.weight.data.normal_(0, 0.01)
self.resize_box.weight.data.normal_(0, 1)
self.resize_overlap.weight.data.normal_(0, 1)
self.resize_feat.bias.data.zero_()
self.resize_embed.bias.data.zero_()
self.resize_box.bias.data.zero_()
self.resize_overlap.bias.data.zero_()
def forward(self, input_0, input_1, input_2, input_3):
primals_1 = self.resize_feat.weight
primals_2 = self.resize_feat.bias
primals_4 = self.resize_embed.weight
primals_5 = self.resize_embed.bias
primals_7 = self.resize_box.weight
primals_8 = self.resize_box.bias
primals_10 = self.resize_overlap.weight
primals_11 = self.resize_overlap.bias
primals_3 = input_0
primals_6 = input_1
primals_9 = input_2
primals_12 = input_3
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12])
return output[0]
|
KaihuaTang/VCTree-Visual-Question-Answering
|
RLFeatPreprocessNet
| false
| 8,380
|
[
"MIT"
] | 31
|
b6b0a8bdb01d45d36de3bded91db42544ad6a593
|
https://github.com/KaihuaTang/VCTree-Visual-Question-Answering/tree/b6b0a8bdb01d45d36de3bded91db42544ad6a593
|
ELUPlus
|
# 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/ts/ctspwovthq3dsi5jsbmf2jpq65oh65tdm6wwxrawn6eaigffxrfz.py
# Topologically Sorted Source Nodes: [elu, add], Original ATen: [aten.elu, aten.add]
# Source node to ATen node mapping:
# add => add
# elu => expm1, gt, mul, mul_1, mul_2, 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, 1.0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, 1.0), kwargs = {})
# %expm1 : [num_users=1] = call_function[target=torch.ops.aten.expm1.default](args = (%mul_1,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%expm1, 1.0), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %mul, %mul_2), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%where, 1.0), kwargs = {})
triton_poi_fused_add_elu_0 = async_compile.triton('triton_poi_fused_add_elu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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_elu_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_elu_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 = 1.0
tmp4 = tmp0 * tmp3
tmp5 = libdevice.expm1(tmp4)
tmp6 = tmp5 * tmp3
tmp7 = tl.where(tmp2, tmp4, tmp6)
tmp8 = tmp7 + tmp3
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: [elu, add], Original ATen: [aten.elu, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_elu_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import nn
import torch.nn
class ELUPlus(nn.Module):
def __init__(self):
super().__init__()
self.elu = nn.ELU()
def forward(self, x):
return self.elu(x) + 1.0
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
import torch.nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_elu_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 = 1.0
tmp4 = tmp0 * tmp3
tmp5 = libdevice.expm1(tmp4)
tmp6 = tmp5 * tmp3
tmp7 = tl.where(tmp2, tmp4, tmp6)
tmp8 = tmp7 + tmp3
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_elu_0[grid(256)](arg0_1, buf0, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class ELUPlusNew(nn.Module):
def __init__(self):
super().__init__()
self.elu = nn.ELU()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
KailinLi/nflows
|
ELUPlus
| false
| 8,381
|
[
"MIT"
] | 13
|
7c07a1d5e510beb681d1b11d6ffda95a086a8153
|
https://github.com/KailinLi/nflows/tree/7c07a1d5e510beb681d1b11d6ffda95a086a8153
|
Memory
|
# 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/gh/cghdn4pihhs4yx6f6a4ig3pjgfy5zh6dghclq5xeandkuwdvuk5c.py
# Topologically Sorted Source Nodes: [targetT], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# targetT => 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=[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_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 = 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)
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/om/com5a2jdtzsd2gxozlfy64n7thpakjehaajnnx7weffss4gxj2sy.py
# Topologically Sorted Source Nodes: [weight_2], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# weight_2 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_1, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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
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/o6/co6rds2fnpdijj6xbnv2tmkoxmykgqu7idfvvwu22ylhgstr53gf.py
# Topologically Sorted Source Nodes: [weight_4], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# weight_4 => clone_1
# Graph fragment:
# %clone_1 : [num_users=2] = call_function[target=torch.ops.aten.clone.default](args = (%permute_1,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_2 = async_compile.triton('triton_poi_fused_clone_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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_clone_2(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
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (64*y1)), xmask & ymask)
tmp1 = tl.load(in_ptr0 + ((4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x2) + (64*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2 + (16*y3)), tmp8, xmask & ymask)
''', 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), (16, 4, 1))
assert_size_stride(arg2_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 16, 4), (64, 1, 16), torch.float32)
# Topologically Sorted Source Nodes: [targetT], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
buf1 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [targetT, weight], Original ATen: [aten.clone, aten.bmm]
extern_kernels.bmm(buf0, arg1_1, out=buf1)
del arg1_1
buf2 = reinterpret_tensor(buf0, (64, 4), (4, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [weight_2], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (4, 4, 16), (64, 16, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [weight_4], Original ATen: [aten.clone]
triton_poi_fused_clone_2.run(buf2, buf3, 16, 16, grid=grid(16, 16), stream=stream0)
buf4 = reinterpret_tensor(buf2, (4, 4, 16), (64, 16, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [weightedContext], Original ATen: [aten.bmm]
extern_kernels.bmm(arg2_1, buf3, out=buf4)
del arg2_1
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(buf3, (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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
arg2_1 = rand_strided((4, 4, 4), (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
class Memory(nn.Module):
def __init__(self):
super(Memory, self).__init__()
self.sm = nn.Softmax()
self.mask = None
def applyMask(self, mask):
self.mask = mask
def forward(self, input, context_key, content_value):
"""
input: batch x idf x ih x iw (queryL=ihxiw)
context: batch x idf x sourceL
"""
ih, iw = input.size(2), input.size(3)
queryL = ih * iw
batch_size, sourceL = context_key.size(0), context_key.size(2)
target = input.view(batch_size, -1, queryL)
targetT = torch.transpose(target, 1, 2).contiguous()
sourceT = context_key
weight = torch.bmm(targetT, sourceT)
weight = weight.view(batch_size * queryL, sourceL)
if self.mask is not None:
mask = self.mask.repeat(queryL, 1)
weight.data.masked_fill_(mask.data, -float('inf'))
weight = torch.nn.functional.softmax(weight, dim=1)
weight = weight.view(batch_size, queryL, sourceL)
weight = torch.transpose(weight, 1, 2).contiguous()
weightedContext = torch.bmm(content_value, weight)
weightedContext = weightedContext.view(batch_size, -1, ih, iw)
weight = weight.view(batch_size, -1, ih, iw)
return weightedContext, weight
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4]), torch.rand([4,
4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import torch.nn as nn
import torch.nn.parallel
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 = 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)
tl.store(out_ptr0 + x0, tmp0, 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
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_clone_2(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
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 64 * y1), xmask & ymask)
tmp1 = tl.load(in_ptr0 + (4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x2 + 64 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2 + 16 * y3), tmp8, xmask & ymask)
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), (16, 4, 1))
assert_size_stride(arg2_1, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 16, 4), (64, 1, 16), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(256)](arg0_1, buf0, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
buf1 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
extern_kernels.bmm(buf0, arg1_1, out=buf1)
del arg1_1
buf2 = reinterpret_tensor(buf0, (64, 4), (4, 1), 0)
del buf0
triton_poi_fused__softmax_1[grid(256)](buf1, buf2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (4, 4, 16), (64, 16, 1), 0)
del buf1
triton_poi_fused_clone_2[grid(16, 16)](buf2, buf3, 16, 16, XBLOCK=
16, YBLOCK=16, num_warps=4, num_stages=1)
buf4 = reinterpret_tensor(buf2, (4, 4, 16), (64, 16, 1), 0)
del buf2
extern_kernels.bmm(arg2_1, buf3, out=buf4)
del arg2_1
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0)
class MemoryNew(nn.Module):
def __init__(self):
super(MemoryNew, self).__init__()
self.sm = nn.Softmax()
self.mask = None
def applyMask(self, mask):
self.mask = mask
def forward(self, input_0, input_1, input_2):
arg0_1 = input_0
arg1_1 = input_1
arg2_1 = input_2
output = call([arg0_1, arg1_1, arg2_1])
return output[0], output[1]
|
JoonHong-Kim/T2I_CL
|
Memory
| false
| 8,382
|
[
"MIT"
] | 35
|
c52aa73da903d6e4174eeef2663e5bc1163785b1
|
https://github.com/JoonHong-Kim/T2I_CL/tree/c52aa73da903d6e4174eeef2663e5bc1163785b1
|
DiceLossWithLogits
|
# 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/zw/czwkdr5kj5kvov6jgdxr47eylapl5isggzw76mkedkqzt26xbniw.py
# Topologically Sorted Source Nodes: [mul, numerator, mul_1, sum_2, mul_2, sum_3], Original ATen: [aten.mul, aten.sum]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# numerator => sum_1
# 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_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 = {})
# %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 = {})
triton_per_fused_mul_sum_0 = async_compile.triton('triton_per_fused_mul_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_0', '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_0(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 + ((16*x0) + (64*(r1 // 16)) + (r1 % 16)), xmask, other=0.0)
tmp2 = tl.load(in_ptr1 + ((16*x0) + (64*(r1 // 16)) + (r1 % 16)), 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 = tmp1 * tmp1
tmp9 = tl.broadcast_to(tmp8, [XBLOCK, RBLOCK])
tmp11 = tl.where(xmask, tmp9, 0)
tmp12 = tl.sum(tmp11, 1)[:, None]
tmp13 = tmp2 * tmp2
tmp14 = tl.broadcast_to(tmp13, [XBLOCK, RBLOCK])
tmp16 = tl.where(xmask, tmp14, 0)
tmp17 = tl.sum(tmp16, 1)[:, None]
tl.store(out_ptr0 + (x0), tmp7, xmask)
tl.store(out_ptr1 + (x0), tmp12, xmask)
tl.store(out_ptr2 + (x0), tmp17, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ow/cow45uy76wdortu6qm2wtsqlpzlmjty77mvzrpxh2y7u3t4lgfoy.py
# Topologically Sorted Source Nodes: [denominator, clamp, truediv, channelwise_score, channelwise_score_1, score], Original ATen: [aten.add, aten.clamp, aten.div, aten.mul, aten.rsub, aten.sum]
# Source node to ATen node mapping:
# channelwise_score => mul_3
# channelwise_score_1 => sub
# clamp => clamp_min
# denominator => add
# score => sum_4
# truediv => div
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, %sum_3), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%add, 1e-07), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, %clamp_min), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, 2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %mul_3), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%sub,), kwargs = {})
triton_per_fused_add_clamp_div_mul_rsub_sum_1 = async_compile.triton('triton_per_fused_add_clamp_div_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.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_clamp_div_mul_rsub_sum_1', '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_add_clamp_div_mul_rsub_sum_1(in_ptr0, in_ptr1, in_ptr2, 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 = tl.load(in_ptr2 + (r0), None)
tmp3 = tmp1 + tmp2
tmp4 = 1e-07
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp6 = tmp0 / tmp5
tmp7 = 2.0
tmp8 = tmp6 * tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp13 = tl.sum(tmp11, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp13, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, 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)
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf2 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mul, numerator, mul_1, sum_2, mul_2, sum_3], Original ATen: [aten.mul, aten.sum]
stream0 = get_raw_stream(0)
triton_per_fused_mul_sum_0.run(arg0_1, arg1_1, buf0, buf1, buf2, 4, 64, grid=grid(4), stream=stream0)
del arg0_1
del arg1_1
buf3 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [denominator, clamp, truediv, channelwise_score, channelwise_score_1, score], Original ATen: [aten.add, aten.clamp, aten.div, aten.mul, aten.rsub, aten.sum]
triton_per_fused_add_clamp_div_mul_rsub_sum_1.run(buf0, buf1, buf2, buf3, 1, 4, grid=grid(1), stream=stream0)
del buf0
del buf1
del buf2
return (buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.utils.data
def flatten_samples(input_):
"""
Flattens a tensor or a variable such that the channel axis is first and the sample axis
is second. The shapes are transformed as follows:
(N, C, H, W) --> (C, N * H * W)
(N, C, D, H, W) --> (C, N * D * H * W)
(N, C) --> (C, N)
The input must be atleast 2d.
"""
num_channels = input_.size(1)
permute_axes = list(range(input_.dim()))
permute_axes[0], permute_axes[1] = permute_axes[1], permute_axes[0]
permuted = input_.permute(*permute_axes).contiguous()
flattened = permuted.view(num_channels, -1)
return flattened
def dice_score(input_, target, invert=False, channelwise=True, eps=1e-07):
if channelwise:
input_ = flatten_samples(input_)
target = flatten_samples(target)
numerator = (input_ * target).sum(-1)
denominator = (input_ * input_).sum(-1) + (target * target).sum(-1)
channelwise_score = 2 * (numerator / denominator.clamp(min=eps))
if invert:
channelwise_score = 1.0 - channelwise_score
score = channelwise_score.sum()
else:
numerator = (input_ * target).sum()
denominator = (input_ * input_).sum() + (target * target).sum()
score = 2.0 * (numerator / denominator.clamp(min=eps))
if invert:
score = 1.0 - score
return score
class DiceLossWithLogits(nn.Module):
def __init__(self, channelwise=True, eps=1e-07):
super().__init__()
self.channelwise = channelwise
self.eps = eps
self.init_kwargs = {'channelwise': channelwise, 'eps': self.eps}
def forward(self, input_, target):
return dice_score(nn.functional.sigmoid(input_), target, invert=
True, channelwise=self.channelwise, eps=self.eps)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_per_fused_mul_sum_0(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 + (16 * x0 + 64 * (r1 // 16) + r1 % 16), xmask,
other=0.0)
tmp2 = tl.load(in_ptr1 + (16 * x0 + 64 * (r1 // 16) + r1 % 16), 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 = tmp1 * tmp1
tmp9 = tl.broadcast_to(tmp8, [XBLOCK, RBLOCK])
tmp11 = tl.where(xmask, tmp9, 0)
tmp12 = tl.sum(tmp11, 1)[:, None]
tmp13 = tmp2 * tmp2
tmp14 = tl.broadcast_to(tmp13, [XBLOCK, RBLOCK])
tmp16 = tl.where(xmask, tmp14, 0)
tmp17 = tl.sum(tmp16, 1)[:, None]
tl.store(out_ptr0 + x0, tmp7, xmask)
tl.store(out_ptr1 + x0, tmp12, xmask)
tl.store(out_ptr2 + x0, tmp17, xmask)
@triton.jit
def triton_per_fused_add_clamp_div_mul_rsub_sum_1(in_ptr0, in_ptr1, in_ptr2,
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 = tl.load(in_ptr2 + r0, None)
tmp3 = tmp1 + tmp2
tmp4 = 1e-07
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp6 = tmp0 / tmp5
tmp7 = 2.0
tmp8 = tmp6 * tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp13 = tl.sum(tmp11, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp13, 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,), torch.float32)
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
buf2 = empty_strided_cuda((4,), (1,), torch.float32)
get_raw_stream(0)
triton_per_fused_mul_sum_0[grid(4)](arg0_1, arg1_1, buf0, buf1,
buf2, 4, 64, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
buf3 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_add_clamp_div_mul_rsub_sum_1[grid(1)](buf0, buf1,
buf2, buf3, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del buf0
del buf1
del buf2
return buf3,
def flatten_samples(input_):
"""
Flattens a tensor or a variable such that the channel axis is first and the sample axis
is second. The shapes are transformed as follows:
(N, C, H, W) --> (C, N * H * W)
(N, C, D, H, W) --> (C, N * D * H * W)
(N, C) --> (C, N)
The input must be atleast 2d.
"""
num_channels = input_.size(1)
permute_axes = list(range(input_.dim()))
permute_axes[0], permute_axes[1] = permute_axes[1], permute_axes[0]
permuted = input_.permute(*permute_axes).contiguous()
flattened = permuted.view(num_channels, -1)
return flattened
def dice_score(input_, target, invert=False, channelwise=True, eps=1e-07):
if channelwise:
input_ = flatten_samples(input_)
target = flatten_samples(target)
numerator = (input_ * target).sum(-1)
denominator = (input_ * input_).sum(-1) + (target * target).sum(-1)
channelwise_score = 2 * (numerator / denominator.clamp(min=eps))
if invert:
channelwise_score = 1.0 - channelwise_score
score = channelwise_score.sum()
else:
numerator = (input_ * target).sum()
denominator = (input_ * input_).sum() + (target * target).sum()
score = 2.0 * (numerator / denominator.clamp(min=eps))
if invert:
score = 1.0 - score
return score
class DiceLossWithLogitsNew(nn.Module):
def __init__(self, channelwise=True, eps=1e-07):
super().__init__()
self.channelwise = channelwise
self.eps = eps
self.init_kwargs = {'channelwise': channelwise, 'eps': self.eps}
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
JonasHell/torch-em
|
DiceLossWithLogits
| false
| 8,383
|
[
"MIT"
] | 13
|
2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
https://github.com/JonasHell/torch-em/tree/2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
ActorNet
|
# 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_2 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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/ta/ctarkhd7gqmxsmyghlzwykngqas6dfclf7bk5ijucwm3q5stwfhi.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# out => sigmoid
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_7,), 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=[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_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_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = 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, (256, 256), (256, 1))
assert_size_stride(primals_7, (256, ), (1, ))
assert_size_stride(primals_8, (4, 256), (256, 1))
assert_size_stride(primals_9, (4, ), (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
buf10 = 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, buf10, 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
buf9 = 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, buf9, 16384, grid=grid(16384), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf3, (64, 256), (256, 1), 0), reinterpret_tensor(primals_6, (256, 256), (1, 256), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 256), (4096, 1024, 256, 1), 0); del buf4 # reuse
buf8 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.relu, aten.threshold_backward]
triton_poi_fused_relu_threshold_backward_0.run(buf5, primals_7, buf8, 16384, grid=grid(16384), stream=stream0)
del primals_7
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf5, (64, 256), (256, 1), 0), reinterpret_tensor(primals_8, (256, 4), (1, 256), 0), out=buf6)
buf7 = reinterpret_tensor(buf6, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf6 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_1.run(buf7, primals_9, 256, grid=grid(256), stream=stream0)
del primals_9
return (buf7, 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), reinterpret_tensor(buf5, (64, 256), (256, 1), 0), buf7, primals_8, buf8, primals_6, buf9, primals_4, buf10, )
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((256, 256), (256, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 256), (256, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class ActorNet(nn.Module):
""" Actor Network """
def __init__(self, state_num, action_num, hidden1=256, hidden2=256,
hidden3=256):
"""
:param state_num: number of states
:param action_num: number of actions
:param hidden1: hidden layer 1 dimension
:param hidden2: hidden layer 2 dimension
:param hidden3: hidden layer 3 dimension
"""
super(ActorNet, self).__init__()
self.fc1 = nn.Linear(state_num, hidden1)
self.fc2 = nn.Linear(hidden1, hidden2)
self.fc3 = nn.Linear(hidden2, hidden3)
self.fc4 = nn.Linear(hidden3, action_num)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, x):
x = self.relu(self.fc1(x))
x = self.relu(self.fc2(x))
x = self.relu(self.fc3(x))
out = self.sigmoid(self.fc4(x))
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_num': 4, 'action_num': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_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_sigmoid_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = 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, (256, 256), (256, 1))
assert_size_stride(primals_7, (256,), (1,))
assert_size_stride(primals_8, (4, 256), (256, 1))
assert_size_stride(primals_9, (4,), (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
buf10 = 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, buf10, 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
buf9 = 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, buf9, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 256), (256, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 256), (256, 1), 0),
reinterpret_tensor(primals_6, (256, 256), (1, 256), 0), out=buf4)
buf5 = reinterpret_tensor(buf4, (4, 4, 4, 256), (4096, 1024, 256, 1), 0
)
del buf4
buf8 = empty_strided_cuda((4, 4, 4, 256), (4096, 1024, 256, 1),
torch.bool)
triton_poi_fused_relu_threshold_backward_0[grid(16384)](buf5,
primals_7, buf8, 16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_7
buf6 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf5, (64, 256), (256, 1), 0),
reinterpret_tensor(primals_8, (256, 4), (1, 256), 0), out=buf6)
buf7 = reinterpret_tensor(buf6, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf6
triton_poi_fused_sigmoid_1[grid(256)](buf7, primals_9, 256, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_9
return buf7, 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
), reinterpret_tensor(buf5, (64, 256), (256, 1), 0
), buf7, primals_8, buf8, primals_6, buf9, primals_4, buf10
class ActorNetNew(nn.Module):
""" Actor Network """
def __init__(self, state_num, action_num, hidden1=256, hidden2=256,
hidden3=256):
"""
:param state_num: number of states
:param action_num: number of actions
:param hidden1: hidden layer 1 dimension
:param hidden2: hidden layer 2 dimension
:param hidden3: hidden layer 3 dimension
"""
super(ActorNetNew, self).__init__()
self.fc1 = nn.Linear(state_num, hidden1)
self.fc2 = nn.Linear(hidden1, hidden2)
self.fc3 = nn.Linear(hidden2, hidden3)
self.fc4 = nn.Linear(hidden3, action_num)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_6 = self.fc3.weight
primals_7 = self.fc3.bias
primals_8 = self.fc4.weight
primals_9 = self.fc4.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
|
Kanaderu/spiking-ddpg-mapless-navigation
|
ActorNet
| false
| 8,384
|
[
"MIT"
] | 29
|
2b5e7e67385dee4428b8036bc4ffe95e812b34e0
|
https://github.com/Kanaderu/spiking-ddpg-mapless-navigation/tree/2b5e7e67385dee4428b8036bc4ffe95e812b34e0
|
StochasticClassifier
|
# 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/ch/cchqa5hdpbx7rsa6qui2lc3gc2ibrjiz764rtewh4vyywumhvxek.py
# Topologically Sorted Source Nodes: [sub, sigma, mul, weight, weight_1], Original ATen: [aten.sub, aten.softplus, aten.mul, aten.add, aten.linalg_vector_norm]
# Source node to ATen node mapping:
# mul => mul
# sigma => exp, gt, log1p, where
# sub => sub
# weight => add
# weight_1 => pow_1, sum_1
# Graph fragment:
# %sub : [num_users=3] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_2, 4), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%sub, 20), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %sub, %log1p), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where, %randn), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_1), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [1], True), kwargs = {})
triton_poi_fused_add_linalg_vector_norm_mul_softplus_sub_0 = async_compile.triton('triton_poi_fused_add_linalg_vector_norm_mul_softplus_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_linalg_vector_norm_mul_softplus_sub_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 12, '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_linalg_vector_norm_mul_softplus_sub_0(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')
tmp8 = tl.load(in_ptr1 + (4*x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr2 + (4*x0), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr2 + (1 + (4*x0)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr0 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp31 = tl.load(in_ptr1 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp33 = tl.load(in_ptr2 + (2 + (4*x0)), xmask, eviction_policy='evict_last')
tmp37 = tl.load(in_ptr0 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp43 = tl.load(in_ptr1 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp45 = tl.load(in_ptr2 + (3 + (4*x0)), xmask, eviction_policy='evict_last')
tmp1 = 4.0
tmp2 = tmp0 - tmp1
tmp3 = 20.0
tmp4 = tmp2 > tmp3
tmp5 = tl_math.exp(tmp2)
tmp6 = libdevice.log1p(tmp5)
tmp7 = tl.where(tmp4, tmp2, tmp6)
tmp9 = tmp7 * tmp8
tmp11 = tmp9 + tmp10
tmp12 = tmp11 * tmp11
tmp14 = tmp13 - tmp1
tmp15 = tmp14 > tmp3
tmp16 = tl_math.exp(tmp14)
tmp17 = libdevice.log1p(tmp16)
tmp18 = tl.where(tmp15, tmp14, tmp17)
tmp20 = tmp18 * tmp19
tmp22 = tmp20 + tmp21
tmp23 = tmp22 * tmp22
tmp24 = tmp12 + tmp23
tmp26 = tmp25 - tmp1
tmp27 = tmp26 > tmp3
tmp28 = tl_math.exp(tmp26)
tmp29 = libdevice.log1p(tmp28)
tmp30 = tl.where(tmp27, tmp26, tmp29)
tmp32 = tmp30 * tmp31
tmp34 = tmp32 + tmp33
tmp35 = tmp34 * tmp34
tmp36 = tmp24 + tmp35
tmp38 = tmp37 - tmp1
tmp39 = tmp38 > tmp3
tmp40 = tl_math.exp(tmp38)
tmp41 = libdevice.log1p(tmp40)
tmp42 = tl.where(tmp39, tmp38, tmp41)
tmp44 = tmp42 * tmp43
tmp46 = tmp44 + tmp45
tmp47 = tmp46 * tmp46
tmp48 = tmp36 + tmp47
tl.store(out_ptr0 + (x0), tmp48, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/iy/ciyoam77rqlmo46awlouw6beqi42jrkvczuvflcc5gyko2m7oxvf.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.div]
# Source node to ATen node mapping:
# x => div_1
# Graph fragment:
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_3, %expand_1), kwargs = {})
triton_poi_fused_div_1 = async_compile.triton('triton_poi_fused_div_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_div_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_div_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = 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 + (x3), tmp15, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cu/ccudamigdjuge5lrohb5jywlfmakjinu7tfdwxtxrzyowiviiboc.py
# Topologically Sorted Source Nodes: [sub, sigma, mul, weight, weight_1], Original ATen: [aten.sub, aten.softplus, aten.mul, aten.add, aten.div]
# Source node to ATen node mapping:
# mul => mul
# sigma => exp, gt, log1p, where
# sub => sub
# weight => add
# weight_1 => div
# Graph fragment:
# %sub : [num_users=3] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_2, 4), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%sub, 20), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %sub, %log1p), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%where, %randn), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_1), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add, %expand), kwargs = {})
triton_poi_fused_add_div_mul_softplus_sub_2 = async_compile.triton('triton_poi_fused_add_div_mul_softplus_sub_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: 'i32'}, 'device': 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_add_div_mul_softplus_sub_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_add_div_mul_softplus_sub_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp8 = tl.load(in_ptr1 + (x2), xmask)
tmp10 = tl.load(in_ptr2 + (x2), xmask)
tmp12 = tl.load(in_ptr3 + (x1), xmask, eviction_policy='evict_last')
tmp1 = 4.0
tmp2 = tmp0 - tmp1
tmp3 = 20.0
tmp4 = tmp2 > tmp3
tmp5 = tl_math.exp(tmp2)
tmp6 = libdevice.log1p(tmp5)
tmp7 = tl.where(tmp4, tmp2, tmp6)
tmp9 = tmp7 * tmp8
tmp11 = tmp9 + tmp10
tmp13 = libdevice.sqrt(tmp12)
tmp14 = 1e-12
tmp15 = triton_helpers.maximum(tmp13, tmp14)
tmp16 = tmp11 / tmp15
tl.store(out_ptr0 + (x2), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ra/cras443tkv6z6hwlmatoqwgpcamttsk37ugx7fpwtni3jqua6ejo.py
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten.div]
# Source node to ATen node mapping:
# score_1 => div_2
# Graph fragment:
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%view_1, 0.05), kwargs = {})
triton_poi_fused_div_3 = async_compile.triton('triton_poi_fused_div_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: 'i32'}, 'device': 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_3', '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_3(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 = 20.0
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 = 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, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [randn_like], Original ATen: [aten.randn_like]
buf0 = torch.ops.aten.randn.default([4, 4], dtype=torch.float32, device=device(type='cuda', index=0), pin_memory=False)
buf1 = buf0
del buf0
buf2 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [sub, sigma, mul, weight, weight_1], Original ATen: [aten.sub, aten.softplus, aten.mul, aten.add, aten.linalg_vector_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_add_linalg_vector_norm_mul_softplus_sub_0.run(primals_2, buf1, primals_1, buf2, 4, grid=grid(4), stream=stream0)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.div]
triton_poi_fused_div_1.run(primals_3, buf3, 256, grid=grid(256), stream=stream0)
del primals_3
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sub, sigma, mul, weight, weight_1], Original ATen: [aten.sub, aten.softplus, aten.mul, aten.add, aten.div]
triton_poi_fused_add_div_mul_softplus_sub_2.run(primals_2, buf1, primals_1, buf2, buf4, 16, grid=grid(16), stream=stream0)
del buf2
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [score], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0), reinterpret_tensor(buf4, (4, 4), (1, 4), 0), out=buf5)
del buf4
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf5 # reuse
# Topologically Sorted Source Nodes: [score_1], Original ATen: [aten.div]
triton_poi_fused_div_3.run(buf6, 256, grid=grid(256), stream=stream0)
return (buf6, primals_1, primals_2, buf1, reinterpret_tensor(buf3, (64, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 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
from torch.nn import functional as F
class StochasticClassifier(nn.Module):
def __init__(self, num_features, num_classes, temp=0.05):
super().__init__()
self.mu = nn.Parameter(0.01 * torch.randn(num_classes, num_features))
self.sigma = nn.Parameter(torch.zeros(num_classes, num_features))
self.temp = temp
def forward(self, x, stochastic=True):
mu = self.mu
sigma = self.sigma
if stochastic:
sigma = F.softplus(sigma - 4)
weight = sigma * torch.randn_like(mu) + mu
else:
weight = mu
weight = F.normalize(weight, p=2, dim=1)
x = F.normalize(x, p=2, dim=1)
score = F.linear(x, weight)
score = score / self.temp
return score
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_features': 4, 'num_classes': 4}]
|
import torch
from torch import device
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_linalg_vector_norm_mul_softplus_sub_0(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')
tmp8 = tl.load(in_ptr1 + 4 * x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr2 + 4 * x0, xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr0 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp19 = tl.load(in_ptr1 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp21 = tl.load(in_ptr2 + (1 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp25 = tl.load(in_ptr0 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp31 = tl.load(in_ptr1 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp33 = tl.load(in_ptr2 + (2 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp37 = tl.load(in_ptr0 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp43 = tl.load(in_ptr1 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp45 = tl.load(in_ptr2 + (3 + 4 * x0), xmask, eviction_policy='evict_last'
)
tmp1 = 4.0
tmp2 = tmp0 - tmp1
tmp3 = 20.0
tmp4 = tmp2 > tmp3
tmp5 = tl_math.exp(tmp2)
tmp6 = libdevice.log1p(tmp5)
tmp7 = tl.where(tmp4, tmp2, tmp6)
tmp9 = tmp7 * tmp8
tmp11 = tmp9 + tmp10
tmp12 = tmp11 * tmp11
tmp14 = tmp13 - tmp1
tmp15 = tmp14 > tmp3
tmp16 = tl_math.exp(tmp14)
tmp17 = libdevice.log1p(tmp16)
tmp18 = tl.where(tmp15, tmp14, tmp17)
tmp20 = tmp18 * tmp19
tmp22 = tmp20 + tmp21
tmp23 = tmp22 * tmp22
tmp24 = tmp12 + tmp23
tmp26 = tmp25 - tmp1
tmp27 = tmp26 > tmp3
tmp28 = tl_math.exp(tmp26)
tmp29 = libdevice.log1p(tmp28)
tmp30 = tl.where(tmp27, tmp26, tmp29)
tmp32 = tmp30 * tmp31
tmp34 = tmp32 + tmp33
tmp35 = tmp34 * tmp34
tmp36 = tmp24 + tmp35
tmp38 = tmp37 - tmp1
tmp39 = tmp38 > tmp3
tmp40 = tl_math.exp(tmp38)
tmp41 = libdevice.log1p(tmp40)
tmp42 = tl.where(tmp39, tmp38, tmp41)
tmp44 = tmp42 * tmp43
tmp46 = tmp44 + tmp45
tmp47 = tmp46 * tmp46
tmp48 = tmp36 + tmp47
tl.store(out_ptr0 + x0, tmp48, xmask)
@triton.jit
def triton_poi_fused_div_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = 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 + x3, tmp15, xmask)
@triton.jit
def triton_poi_fused_add_div_mul_softplus_sub_2(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp8 = tl.load(in_ptr1 + x2, xmask)
tmp10 = tl.load(in_ptr2 + x2, xmask)
tmp12 = tl.load(in_ptr3 + x1, xmask, eviction_policy='evict_last')
tmp1 = 4.0
tmp2 = tmp0 - tmp1
tmp3 = 20.0
tmp4 = tmp2 > tmp3
tmp5 = tl_math.exp(tmp2)
tmp6 = libdevice.log1p(tmp5)
tmp7 = tl.where(tmp4, tmp2, tmp6)
tmp9 = tmp7 * tmp8
tmp11 = tmp9 + tmp10
tmp13 = libdevice.sqrt(tmp12)
tmp14 = 1e-12
tmp15 = triton_helpers.maximum(tmp13, tmp14)
tmp16 = tmp11 / tmp15
tl.store(out_ptr0 + x2, tmp16, xmask)
@triton.jit
def triton_poi_fused_div_3(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 = 20.0
tmp2 = tmp0 * tmp1
tl.store(in_out_ptr0 + x0, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = torch.ops.aten.randn.default([4, 4], dtype=torch.float32,
device=device(type='cuda', index=0), pin_memory=False)
buf1 = buf0
del buf0
buf2 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_linalg_vector_norm_mul_softplus_sub_0[grid(4)](
primals_2, buf1, primals_1, buf2, 4, XBLOCK=4, num_warps=1,
num_stages=1)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_div_1[grid(256)](primals_3, buf3, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_3
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_div_mul_softplus_sub_2[grid(16)](primals_2,
buf1, primals_1, buf2, buf4, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del buf2
buf5 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf3, (64, 4), (4, 1), 0),
reinterpret_tensor(buf4, (4, 4), (1, 4), 0), out=buf5)
del buf4
buf6 = reinterpret_tensor(buf5, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf5
triton_poi_fused_div_3[grid(256)](buf6, 256, XBLOCK=256, num_warps=
4, num_stages=1)
return buf6, primals_1, primals_2, buf1, reinterpret_tensor(buf3, (64,
4), (4, 1), 0)
class StochasticClassifierNew(nn.Module):
def __init__(self, num_features, num_classes, temp=0.05):
super().__init__()
self.mu = nn.Parameter(0.01 * torch.randn(num_classes, num_features))
self.sigma = nn.Parameter(torch.zeros(num_classes, num_features))
self.temp = temp
def forward(self, input_0):
primals_1 = self.mu
primals_2 = self.sigma
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
KaiyangZhou/ssdg-benchmark
|
StochasticClassifier
| false
| 8,385
|
[
"MIT"
] | 43
|
aaa48be4f93b77347fbadff649be6b3e0f7a8779
|
https://github.com/KaiyangZhou/ssdg-benchmark/tree/aaa48be4f93b77347fbadff649be6b3e0f7a8779
|
Highway
|
# 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/v7/cv7rclhzdb2c7qoumqndxevg24mioljdnzprrwaz2h2f5vnuzevx.py
# Topologically Sorted Source Nodes: [t, relu, mul, sub, mul_1, add], Original ATen: [aten.sigmoid, aten.relu, aten.mul, aten.rsub, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# mul_1 => mul_1
# relu => relu
# sub => sub
# t => sigmoid
# Graph fragment:
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view_1,), kwargs = {})
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%view_3,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %relu), 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_relu_rsub_sigmoid_0 = async_compile.triton('triton_poi_fused_add_mul_relu_rsub_sigmoid_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_relu_rsub_sigmoid_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_relu_rsub_sigmoid_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)
tmp8 = tl.load(in_ptr2 + (x0), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = tmp1 * tmp4
tmp6 = 1.0
tmp7 = tmp6 - tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp5 + tmp9
tl.store(out_ptr0 + (x0), tmp10, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, 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: [t, relu, mul, sub, mul_1, add], Original ATen: [aten.sigmoid, aten.relu, aten.mul, aten.rsub, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mul_relu_rsub_sigmoid_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
import torch.nn as nn
import torch.nn.functional as F
class Highway(nn.Module):
"""Highway network"""
def __init__(self, input_size):
super(Highway, self).__init__()
self.fc1 = nn.Linear(input_size, input_size, bias=True)
self.fc2 = nn.Linear(input_size, input_size, bias=True)
def forward(self, x):
t = F.sigmoid(self.fc1(x))
return torch.mul(t, F.relu(self.fc2(x))) + torch.mul(1 - t, x)
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
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_add_mul_relu_rsub_sigmoid_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)
tmp8 = tl.load(in_ptr2 + x0, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = tmp1 * tmp4
tmp6 = 1.0
tmp7 = tmp6 - tmp1
tmp9 = tmp7 * tmp8
tmp10 = tmp5 + tmp9
tl.store(out_ptr0 + x0, tmp10, 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_relu_rsub_sigmoid_0[grid(256)](buf0, buf1,
primals_3, buf2, 256, XBLOCK=128, num_warps=4, num_stages=1)
return buf2, primals_3, buf0, buf1
class HighwayNew(nn.Module):
"""Highway network"""
def __init__(self, input_size):
super(HighwayNew, self).__init__()
self.fc1 = nn.Linear(input_size, input_size, bias=True)
self.fc2 = nn.Linear(input_size, input_size, bias=True)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.fc2.weight
primals_5 = self.fc2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
Kailianghu/Character-Aware-Neural-Language-Model
|
Highway
| false
| 8,386
|
[
"MIT"
] | 35
|
6bd72ce00a3ac9eb152ba006bdae8a6922e0ad35
|
https://github.com/Kailianghu/Character-Aware-Neural-Language-Model/tree/6bd72ce00a3ac9eb152ba006bdae8a6922e0ad35
|
BCEDiceLossWithLogits
|
# 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/zw/czwkdr5kj5kvov6jgdxr47eylapl5isggzw76mkedkqzt26xbniw.py
# Topologically Sorted Source Nodes: [mul, numerator, mul_1, sum_2, mul_2, sum_3], Original ATen: [aten.mul, aten.sum]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# numerator => sum_1
# 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_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 = {})
# %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 = {})
triton_per_fused_mul_sum_0 = async_compile.triton('triton_per_fused_mul_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_0', '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_0(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 + ((16*x0) + (64*(r1 // 16)) + (r1 % 16)), xmask, other=0.0)
tmp2 = tl.load(in_ptr1 + ((16*x0) + (64*(r1 // 16)) + (r1 % 16)), 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 = tmp1 * tmp1
tmp9 = tl.broadcast_to(tmp8, [XBLOCK, RBLOCK])
tmp11 = tl.where(xmask, tmp9, 0)
tmp12 = tl.sum(tmp11, 1)[:, None]
tmp13 = tmp2 * tmp2
tmp14 = tl.broadcast_to(tmp13, [XBLOCK, RBLOCK])
tmp16 = tl.where(xmask, tmp14, 0)
tmp17 = tl.sum(tmp16, 1)[:, None]
tl.store(out_ptr0 + (x0), tmp7, xmask)
tl.store(out_ptr1 + (x0), tmp12, xmask)
tl.store(out_ptr2 + (x0), tmp17, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ow/cow45uy76wdortu6qm2wtsqlpzlmjty77mvzrpxh2y7u3t4lgfoy.py
# Topologically Sorted Source Nodes: [denominator, clamp, truediv, channelwise_score, channelwise_score_1, score], Original ATen: [aten.add, aten.clamp, aten.div, aten.mul, aten.rsub, aten.sum]
# Source node to ATen node mapping:
# channelwise_score => mul_3
# channelwise_score_1 => sub
# clamp => clamp_min
# denominator => add
# score => sum_4
# truediv => div
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, %sum_3), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%add, 1e-07), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, %clamp_min), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, 2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %mul_3), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%sub,), kwargs = {})
triton_per_fused_add_clamp_div_mul_rsub_sum_1 = async_compile.triton('triton_per_fused_add_clamp_div_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.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_clamp_div_mul_rsub_sum_1', '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_add_clamp_div_mul_rsub_sum_1(in_ptr0, in_ptr1, in_ptr2, 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 = tl.load(in_ptr2 + (r0), None)
tmp3 = tmp1 + tmp2
tmp4 = 1e-07
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp6 = tmp0 / tmp5
tmp7 = 2.0
tmp8 = tmp6 * tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp13 = tl.sum(tmp11, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp13, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pj/cpj665hzyutddfso57lgimdbgfbtaxy53oaowlpwtxp7qhdjcp5o.py
# Topologically Sorted Source Nodes: [mul_4, loss_bce, mul_5, add_1], Original ATen: [aten.mul, aten.binary_cross_entropy_with_logits, aten.add]
# Source node to ATen node mapping:
# add_1 => add_1
# loss_bce => abs_1, exp, full_default, log1p, mean, minimum, mul_4, neg, sub_1, sub_2, sub_3
# mul_4 => mul_5
# mul_5 => mul_6
# Graph fragment:
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sum_4, 1.0), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %arg1_1), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %arg0_1), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
# %minimum : [num_users=1] = call_function[target=torch.ops.aten.minimum.default](args = (%full_default, %arg0_1), kwargs = {})
# %abs_1 : [num_users=1] = call_function[target=torch.ops.aten.abs.default](args = (%arg0_1,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%abs_1,), kwargs = {})
# %exp : [num_users=1] = call_function[target=torch.ops.aten.exp.default](args = (%neg,), kwargs = {})
# %log1p : [num_users=1] = call_function[target=torch.ops.aten.log1p.default](args = (%exp,), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%minimum, %log1p), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%mul_4, %sub_2), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%sub_3,), kwargs = {})
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mean, 1.0), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_5, %mul_6), kwargs = {})
triton_per_fused_add_binary_cross_entropy_with_logits_mul_2 = async_compile.triton('triton_per_fused_add_binary_cross_entropy_with_logits_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.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_binary_cross_entropy_with_logits_mul_2', '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_add_binary_cross_entropy_with_logits_mul_2(in_out_ptr0, in_ptr0, in_ptr1, xnumel, rnumel):
xnumel = 1
XBLOCK: tl.constexpr = 1
rnumel = 256
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = tl.full([1], xoffset, tl.int32)
xmask = tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
roffset = 0
rmask = tl.full([RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + (r0), None)
tmp3 = tl.load(in_ptr1 + (r0), None)
tmp16 = tl.load(in_out_ptr0 + (0))
tmp17 = tl.broadcast_to(tmp16, [1])
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))
tmp18 = tmp17 * tmp1
tmp19 = 256.0
tmp20 = tmp15 / tmp19
tmp21 = tmp20 * tmp1
tmp22 = tmp18 + tmp21
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([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((4, ), (1, ), torch.float32)
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf2 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mul, numerator, mul_1, sum_2, mul_2, sum_3], Original ATen: [aten.mul, aten.sum]
stream0 = get_raw_stream(0)
triton_per_fused_mul_sum_0.run(arg0_1, arg1_1, buf0, buf1, buf2, 4, 64, grid=grid(4), stream=stream0)
buf3 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [denominator, clamp, truediv, channelwise_score, channelwise_score_1, score], Original ATen: [aten.add, aten.clamp, aten.div, aten.mul, aten.rsub, aten.sum]
triton_per_fused_add_clamp_div_mul_rsub_sum_1.run(buf0, buf1, buf2, buf3, 1, 4, grid=grid(1), stream=stream0)
del buf0
del buf1
del buf2
buf5 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [mul_4, loss_bce, mul_5, add_1], Original ATen: [aten.mul, aten.binary_cross_entropy_with_logits, aten.add]
triton_per_fused_add_binary_cross_entropy_with_logits_mul_2.run(buf5, arg1_1, arg0_1, 1, 256, grid=grid(1), stream=stream0)
del arg0_1
del arg1_1
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
import torch.utils.data
def flatten_samples(input_):
"""
Flattens a tensor or a variable such that the channel axis is first and the sample axis
is second. The shapes are transformed as follows:
(N, C, H, W) --> (C, N * H * W)
(N, C, D, H, W) --> (C, N * D * H * W)
(N, C) --> (C, N)
The input must be atleast 2d.
"""
num_channels = input_.size(1)
permute_axes = list(range(input_.dim()))
permute_axes[0], permute_axes[1] = permute_axes[1], permute_axes[0]
permuted = input_.permute(*permute_axes).contiguous()
flattened = permuted.view(num_channels, -1)
return flattened
def dice_score(input_, target, invert=False, channelwise=True, eps=1e-07):
if channelwise:
input_ = flatten_samples(input_)
target = flatten_samples(target)
numerator = (input_ * target).sum(-1)
denominator = (input_ * input_).sum(-1) + (target * target).sum(-1)
channelwise_score = 2 * (numerator / denominator.clamp(min=eps))
if invert:
channelwise_score = 1.0 - channelwise_score
score = channelwise_score.sum()
else:
numerator = (input_ * target).sum()
denominator = (input_ * input_).sum() + (target * target).sum()
score = 2.0 * (numerator / denominator.clamp(min=eps))
if invert:
score = 1.0 - score
return score
class BCEDiceLossWithLogits(nn.Module):
def __init__(self, alpha=1.0, beta=1.0, channelwise=True, eps=1e-07):
super().__init__()
self.alpha = alpha
self.beta = beta
self.channelwise = channelwise
self.eps = eps
self.init_kwargs = {'alpha': alpha, 'beta': beta, 'channelwise':
channelwise, 'eps': self.eps}
def forward(self, input_, target):
loss_dice = dice_score(nn.functional.sigmoid(input_), target,
invert=True, channelwise=self.channelwise, eps=self.eps)
loss_bce = nn.functional.binary_cross_entropy_with_logits(input_,
target)
return self.alpha * loss_dice + self.beta * loss_bce
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
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_per_fused_mul_sum_0(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 + (16 * x0 + 64 * (r1 // 16) + r1 % 16), xmask,
other=0.0)
tmp2 = tl.load(in_ptr1 + (16 * x0 + 64 * (r1 // 16) + r1 % 16), 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 = tmp1 * tmp1
tmp9 = tl.broadcast_to(tmp8, [XBLOCK, RBLOCK])
tmp11 = tl.where(xmask, tmp9, 0)
tmp12 = tl.sum(tmp11, 1)[:, None]
tmp13 = tmp2 * tmp2
tmp14 = tl.broadcast_to(tmp13, [XBLOCK, RBLOCK])
tmp16 = tl.where(xmask, tmp14, 0)
tmp17 = tl.sum(tmp16, 1)[:, None]
tl.store(out_ptr0 + x0, tmp7, xmask)
tl.store(out_ptr1 + x0, tmp12, xmask)
tl.store(out_ptr2 + x0, tmp17, xmask)
@triton.jit
def triton_per_fused_add_clamp_div_mul_rsub_sum_1(in_ptr0, in_ptr1, in_ptr2,
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 = tl.load(in_ptr2 + r0, None)
tmp3 = tmp1 + tmp2
tmp4 = 1e-07
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp6 = tmp0 / tmp5
tmp7 = 2.0
tmp8 = tmp6 * tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp13 = tl.sum(tmp11, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp13, None)
@triton.jit
def triton_per_fused_add_binary_cross_entropy_with_logits_mul_2(in_out_ptr0,
in_ptr0, in_ptr1, xnumel, rnumel):
XBLOCK: tl.constexpr = 1
RBLOCK: tl.constexpr = 256
xoffset = tl.program_id(0) * XBLOCK
tl.full([1], xoffset, tl.int32)
tl.full([RBLOCK], True, tl.int1)
rindex = tl.arange(0, RBLOCK)[:]
tl.full([RBLOCK], True, tl.int1)
r0 = rindex
tmp0 = tl.load(in_ptr0 + r0, None)
tmp3 = tl.load(in_ptr1 + r0, None)
tmp16 = tl.load(in_out_ptr0 + 0)
tmp17 = tl.broadcast_to(tmp16, [1])
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))
tmp18 = tmp17 * tmp1
tmp19 = 256.0
tmp20 = tmp15 / tmp19
tmp21 = tmp20 * tmp1
tmp22 = tmp18 + tmp21
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([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((4,), (1,), torch.float32)
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
buf2 = empty_strided_cuda((4,), (1,), torch.float32)
get_raw_stream(0)
triton_per_fused_mul_sum_0[grid(4)](arg0_1, arg1_1, buf0, buf1,
buf2, 4, 64, XBLOCK=1, num_warps=2, num_stages=1)
buf3 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_add_clamp_div_mul_rsub_sum_1[grid(1)](buf0, buf1,
buf2, buf3, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del buf0
del buf1
del buf2
buf5 = buf3
del buf3
triton_per_fused_add_binary_cross_entropy_with_logits_mul_2[grid(1)](
buf5, arg1_1, arg0_1, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
return buf5,
def flatten_samples(input_):
"""
Flattens a tensor or a variable such that the channel axis is first and the sample axis
is second. The shapes are transformed as follows:
(N, C, H, W) --> (C, N * H * W)
(N, C, D, H, W) --> (C, N * D * H * W)
(N, C) --> (C, N)
The input must be atleast 2d.
"""
num_channels = input_.size(1)
permute_axes = list(range(input_.dim()))
permute_axes[0], permute_axes[1] = permute_axes[1], permute_axes[0]
permuted = input_.permute(*permute_axes).contiguous()
flattened = permuted.view(num_channels, -1)
return flattened
def dice_score(input_, target, invert=False, channelwise=True, eps=1e-07):
if channelwise:
input_ = flatten_samples(input_)
target = flatten_samples(target)
numerator = (input_ * target).sum(-1)
denominator = (input_ * input_).sum(-1) + (target * target).sum(-1)
channelwise_score = 2 * (numerator / denominator.clamp(min=eps))
if invert:
channelwise_score = 1.0 - channelwise_score
score = channelwise_score.sum()
else:
numerator = (input_ * target).sum()
denominator = (input_ * input_).sum() + (target * target).sum()
score = 2.0 * (numerator / denominator.clamp(min=eps))
if invert:
score = 1.0 - score
return score
class BCEDiceLossWithLogitsNew(nn.Module):
def __init__(self, alpha=1.0, beta=1.0, channelwise=True, eps=1e-07):
super().__init__()
self.alpha = alpha
self.beta = beta
self.channelwise = channelwise
self.eps = eps
self.init_kwargs = {'alpha': alpha, 'beta': beta, 'channelwise':
channelwise, 'eps': self.eps}
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
JonasHell/torch-em
|
BCEDiceLossWithLogits
| false
| 8,387
|
[
"MIT"
] | 13
|
2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
https://github.com/JonasHell/torch-em/tree/2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
ResBlock
|
# 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/bm/cbmvwkhgioz63mnhrh3onxemouh4axyclce6ay7mypmzm62glj7h.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, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-05), kwargs = {})
# %rsqrt : [num_users=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-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/si/csimdaz64hjbvur6nmd43gt7k3bajb7m7ijryw3lzott7bamgbud.py
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.native_layer_norm, aten.tanh]
# Source node to ATen node mapping:
# x => add, add_1, mul, mul_1, rsqrt, sub, var_mean
# x_1 => tanh
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%primals_1, [3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem, 1e-05), kwargs = {})
# %rsqrt : [num_users=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 = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_1,), kwargs = {})
triton_poi_fused_native_layer_norm_tanh_1 = async_compile.triton('triton_poi_fused_native_layer_norm_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_native_layer_norm_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_native_layer_norm_tanh_1(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
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
tmp9 = libdevice.tanh(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/7c/c7cbrddc7pva2pbqkwckqqlewfgj7bbllxvduk5p3rrrl6lutgez.py
# Topologically Sorted Source Nodes: [x_3, x_4, x_5], Original ATen: [aten.native_layer_norm, aten.relu, aten.tanh]
# Source node to ATen node mapping:
# x_3 => add_2, add_3, mul_2, mul_3, rsqrt_1, sub_1, var_mean_1
# x_4 => relu
# x_5 => tanh_1
# Graph fragment:
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view_1, [3]), kwargs = {correction: 0, keepdim: True})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_2, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_2,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %getitem_3), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt_1), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_2, %primals_6), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_3, %primals_7), kwargs = {})
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%add_3,), kwargs = {})
# %tanh_1 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%relu,), kwargs = {})
triton_poi_fused_native_layer_norm_relu_tanh_2 = async_compile.triton('triton_poi_fused_native_layer_norm_relu_tanh_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_relu_tanh_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_native_layer_norm_relu_tanh_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + (x0), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tmp11 = libdevice.tanh(tmp10)
tl.store(out_ptr0 + (x2), tmp11, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/dx/cdximr72rpoz6ofltt664fzbyo6yjq7mz46rje3ectxq2n5t45t7.py
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
# Source node to ATen node mapping:
# add => add_4
# Graph fragment:
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %view_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_ptr0 + (x2), xmask)
tmp1 = tl.load(in_out_ptr0 + (x2), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((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: [x], Original ATen: [aten.native_layer_norm]
stream0 = get_raw_stream(0)
triton_poi_fused_native_layer_norm_0.run(primals_1, 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: [x, x_1], Original ATen: [aten.native_layer_norm, aten.tanh]
triton_poi_fused_native_layer_norm_tanh_1.run(primals_1, buf0, buf1, primals_2, primals_3, buf2, 256, grid=grid(256), stream=stream0)
del primals_2
del primals_3
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf2, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
buf4 = buf1; del buf1 # reuse
buf5 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_0.run(buf3, buf4, buf5, 64, grid=grid(64), stream=stream0)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3, x_4, x_5], Original ATen: [aten.native_layer_norm, aten.relu, aten.tanh]
triton_poi_fused_native_layer_norm_relu_tanh_2.run(buf3, buf4, buf5, primals_6, primals_7, buf6, 256, grid=grid(256), stream=stream0)
del buf4
del buf5
buf7 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf6, (64, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf7)
buf8 = reinterpret_tensor(buf7, (4, 4, 4, 4), (64, 16, 4, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [add], Original ATen: [aten.add]
triton_poi_fused_add_3.run(buf8, primals_1, primals_9, 256, grid=grid(256), stream=stream0)
del primals_9
return (buf8, primals_1, primals_6, primals_7, buf2, buf3, buf6, primals_8, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import nn
import torch.nn.functional as F
class ResBlock(nn.Module):
def __init__(self, dim, dropout=0):
super(ResBlock, self).__init__()
self.dim = dim
self.dropout = nn.Dropout(dropout)
self.linear1 = nn.Linear(self.dim, self.dim)
self.linear2 = nn.Linear(self.dim, self.dim)
self.layer_norm1 = nn.LayerNorm(self.dim)
self.layer_norm2 = nn.LayerNorm(self.dim)
self.reset_parameters()
def reset_parameters(self):
initScale = 0.1
self.linear1.weight.data.uniform_(-initScale, initScale)
self.linear1.bias.data.zero_()
self.linear2.weight.data.uniform_(-initScale, initScale)
self.linear2.bias.data.zero_()
def forward(self, x):
x_prev = x
x = self.layer_norm1(x)
x = torch.tanh(x)
x = self.linear1(x)
x = self.layer_norm2(x)
x = F.relu(x)
x = torch.tanh(x)
x = self.dropout(x)
x = self.linear2(x)
return x_prev + x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_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-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_tanh_1(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
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
tmp9 = libdevice.tanh(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_relu_tanh_2(in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr3 + x0, xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp4 = tmp2 * tmp3
tmp6 = tmp4 * tmp5
tmp8 = tmp6 + tmp7
tmp9 = tl.full([1], 0, tl.int32)
tmp10 = triton_helpers.maximum(tmp9, tmp8)
tmp11 = libdevice.tanh(tmp10)
tl.store(out_ptr0 + x2, tmp11, 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_ptr0 + x2, xmask)
tmp1 = tl.load(in_out_ptr0 + x2, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((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_1, 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_tanh_1[grid(256)](primals_1,
buf0, buf1, primals_2, primals_3, buf2, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del primals_2
del primals_3
buf3 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(buf2, (64, 4), (
4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf3)
del primals_5
buf4 = buf1
del buf1
buf5 = buf0
del buf0
triton_poi_fused_native_layer_norm_0[grid(64)](buf3, buf4, buf5, 64,
XBLOCK=64, num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_native_layer_norm_relu_tanh_2[grid(256)](buf3,
buf4, buf5, primals_6, primals_7, buf6, 256, XBLOCK=128,
num_warps=4, num_stages=1)
del buf4
del buf5
buf7 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf6, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf7)
buf8 = reinterpret_tensor(buf7, (4, 4, 4, 4), (64, 16, 4, 1), 0)
del buf7
triton_poi_fused_add_3[grid(256)](buf8, primals_1, primals_9, 256,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_9
return (buf8, primals_1, primals_6, primals_7, buf2, buf3, buf6,
primals_8, primals_4)
class ResBlockNew(nn.Module):
def __init__(self, dim, dropout=0):
super(ResBlockNew, self).__init__()
self.dim = dim
self.dropout = nn.Dropout(dropout)
self.linear1 = nn.Linear(self.dim, self.dim)
self.linear2 = nn.Linear(self.dim, self.dim)
self.layer_norm1 = nn.LayerNorm(self.dim)
self.layer_norm2 = nn.LayerNorm(self.dim)
self.reset_parameters()
def reset_parameters(self):
initScale = 0.1
self.linear1.weight.data.uniform_(-initScale, initScale)
self.linear1.bias.data.zero_()
self.linear2.weight.data.uniform_(-initScale, initScale)
self.linear2.bias.data.zero_()
def forward(self, input_0):
primals_4 = self.linear1.weight
primals_2 = self.linear1.bias
primals_8 = self.linear2.weight
primals_3 = self.linear2.bias
primals_5 = self.layer_norm1.weight
primals_6 = self.layer_norm1.bias
primals_7 = self.layer_norm2.weight
primals_9 = self.layer_norm2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
|
JiwanChung/tapm
|
ResBlock
| false
| 8,388
|
[
"MIT"
] | 14
|
ec42b139d1c012daccc55f85e67744488d526476
|
https://github.com/JiwanChung/tapm/tree/ec42b139d1c012daccc55f85e67744488d526476
|
FeatureEncoder
|
# 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/vg/cvgn47a3suf3abdi3yxlfefcismdx34uxftikwkcmdazos4igr4d.py
# Topologically Sorted Source Nodes: [feature_1], Original ATen: [aten.leaky_relu]
# Source node to ATen node mapping:
# feature_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, 0.01), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %view_1, %mul), kwargs = {})
triton_poi_fused_leaky_relu_0 = async_compile.triton('triton_poi_fused_leaky_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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 = 0.01
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = 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: [feature_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
from torch import nn
import torch.nn.functional as F
class FeatureEncoder(nn.Module):
def __init__(self, video_dim, dim):
super(FeatureEncoder, self).__init__()
self.linear = nn.Linear(video_dim, dim)
def forward(self, feature, h=None):
feature = self.linear(feature)
feature = F.leaky_relu(feature)
return feature
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'video_dim': 4, 'dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch 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_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 = 0.01
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3 = 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 FeatureEncoderNew(nn.Module):
def __init__(self, video_dim, dim):
super(FeatureEncoderNew, self).__init__()
self.linear = nn.Linear(video_dim, dim)
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]
|
JiwanChung/tapm
|
FeatureEncoder
| false
| 8,389
|
[
"MIT"
] | 14
|
ec42b139d1c012daccc55f85e67744488d526476
|
https://github.com/JiwanChung/tapm/tree/ec42b139d1c012daccc55f85e67744488d526476
|
net
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/wy/cwy2hamefvfx4yoy3gvok5ugi4f6wharewur3bde4ik4l5cnx3tf.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_1,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
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_tanh_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_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1920
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 30
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vp/cvpxxht56uahspvlsh5nbodmgs4bztkjefvcba436kxwfthuh6pu.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# x_1 => tanh_1
# Graph fragment:
# %tanh_1 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_3,), kwargs = {})
triton_poi_fused_tanh_1 = async_compile.triton('triton_poi_fused_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_tanh_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_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1280
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 20
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (30, 4), (4, 1))
assert_size_stride(primals_2, (30, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (20, 30), (30, 1))
assert_size_stride(primals_5, (20, ), (1, ))
assert_size_stride(primals_6, (4, 20), (20, 1))
assert_size_stride(primals_7, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 30), (30, 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, 30), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 30), (480, 120, 30, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_2, 1920, grid=grid(1920), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 20), (20, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 30), (30, 1), 0), reinterpret_tensor(primals_4, (30, 20), (1, 30), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 20), (320, 80, 20, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.tanh]
triton_poi_fused_tanh_1.run(buf3, primals_5, 1280, grid=grid(1280), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 20), (20, 1), 0), reinterpret_tensor(primals_6, (20, 4), (1, 20), 0), alpha=1, beta=1, out=buf4)
del primals_7
return (reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, buf3, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((30, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((30, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((20, 30), (30, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((20, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 20), (20, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
class net(nn.Module):
def __init__(self, input_dim, output_dim):
super(net, self).__init__()
self.fc1 = nn.Linear(input_dim, 30)
self.fc1.weight.data.normal_(0, 1)
self.fc2 = nn.Linear(30, 20)
self.fc2.weight.data.normal_(0, 1)
self.fc3 = nn.Linear(20, output_dim)
self.fc3.weight.data.normal_(0, 1)
def forward(self, x):
x = F.tanh(self.fc1(x))
x = F.tanh(self.fc2(x))
out = self.fc3(x)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_dim': 4, 'output_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 1920
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 30
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
@triton.jit
def triton_poi_fused_tanh_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 1280
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 20
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (30, 4), (4, 1))
assert_size_stride(primals_2, (30,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (20, 30), (30, 1))
assert_size_stride(primals_5, (20,), (1,))
assert_size_stride(primals_6, (4, 20), (20, 1))
assert_size_stride(primals_7, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 30), (30, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 30), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 30), (480, 120, 30, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(1920)](buf1, primals_2, 1920, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 20), (20, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 30), (30, 1), 0),
reinterpret_tensor(primals_4, (30, 20), (1, 30), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 20), (320, 80, 20, 1), 0)
del buf2
triton_poi_fused_tanh_1[grid(1280)](buf3, primals_5, 1280, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 20),
(20, 1), 0), reinterpret_tensor(primals_6, (20, 4), (1, 20), 0),
alpha=1, beta=1, out=buf4)
del primals_7
return reinterpret_tensor(buf4, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, buf3, primals_6, primals_4
class netNew(nn.Module):
def __init__(self, input_dim, output_dim):
super(netNew, self).__init__()
self.fc1 = nn.Linear(input_dim, 30)
self.fc1.weight.data.normal_(0, 1)
self.fc2 = nn.Linear(30, 20)
self.fc2.weight.data.normal_(0, 1)
self.fc3 = nn.Linear(20, output_dim)
self.fc3.weight.data.normal_(0, 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]
|
Kernels-K/DDPG-pytorch-
|
net
| false
| 8,390
|
[
"MIT"
] | 26
|
9a80a56f52f2232e5bd197521d3d2d388b48c882
|
https://github.com/Kernels-K/DDPG-pytorch-/tree/9a80a56f52f2232e5bd197521d3d2d388b48c882
|
GraphConvolution
|
# 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/mo/cmoh5nwlukrl6peneysoecjki36qsllmnflrtv52q6mnj3gwqwgu.py
# Topologically Sorted Source Nodes: [nodes_1], Original ATen: [aten.leaky_relu]
# Source node to ATen node mapping:
# nodes_1 => gt, mul, where
# Graph fragment:
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%mm, 0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mm, 0.2), kwargs = {})
# %where : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt, %mm, %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=[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_leaky_relu_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_leaky_relu_0(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 = 0.0
tmp2 = tmp0 > tmp1
tmp3 = 0.2
tmp4 = tmp0 * tmp3
tmp5 = tl.where(tmp2, tmp0, tmp4)
tl.store(in_out_ptr0 + (x0), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/hl/chlipq2rvmqikold3rcouxdk6qyogrelwfoj6qpyidl747v63rbu.py
# Topologically Sorted Source Nodes: [nodes_3], Original ATen: [aten.leaky_relu]
# Source node to ATen node mapping:
# nodes_3 => gt_1, mul_1, where_1
# Graph fragment:
# %gt_1 : [num_users=2] = call_function[target=torch.ops.aten.gt.Scalar](args = (%squeeze, 0), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%squeeze, 0.2), kwargs = {})
# %where_1 : [num_users=1] = call_function[target=torch.ops.aten.where.self](args = (%gt_1, %squeeze, %mul_1), kwargs = {})
triton_poi_fused_leaky_relu_1 = async_compile.triton('triton_poi_fused_leaky_relu_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
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_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_leaky_relu_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
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.2
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp7, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = 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, 1), (4, 1, 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, 1), torch.float32)
# Topologically Sorted Source Nodes: [nodes], Original ATen: [aten.mm]
extern_kernels.mm(primals_2, primals_1, out=buf0)
del primals_1
del primals_2
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [nodes_1], Original ATen: [aten.leaky_relu]
stream0 = get_raw_stream(0)
triton_poi_fused_leaky_relu_0.run(buf1, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [nodes_2], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 4), (0, 4, 1), 0), primals_3, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (1, 4, 4), (16, 4, 1))
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [nodes_3], Original ATen: [aten.leaky_relu]
triton_poi_fused_leaky_relu_1.run(buf2, primals_4, buf3, buf4, 16, grid=grid(16), stream=stream0)
del buf2
del primals_4
return (buf4, primals_3, reinterpret_tensor(buf1, (1, 4, 4), (16, 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, 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, 1), (4, 1, 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.nn as nn
class GraphConvolution(nn.Module):
def __init__(self, in_dim, out_dim):
super(GraphConvolution, self).__init__()
self.relu = nn.LeakyReLU(0.2)
self.weight = nn.Conv1d(in_dim, out_dim, 1)
def forward(self, adj, nodes):
nodes = torch.matmul(nodes, adj)
nodes = self.relu(nodes)
nodes = self.weight(nodes)
nodes = self.relu(nodes)
return nodes
def get_inputs():
return [torch.rand([4, 4]), torch.rand([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
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_leaky_relu_0(in_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 = 0.0
tmp2 = tmp0 > tmp1
tmp3 = 0.2
tmp4 = tmp0 * tmp3
tmp5 = tl.where(tmp2, tmp0, tmp4)
tl.store(in_out_ptr0 + x0, tmp5, xmask)
@triton.jit
def triton_poi_fused_leaky_relu_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
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = tmp2 > tmp3
tmp5 = 0.2
tmp6 = tmp2 * tmp5
tmp7 = tl.where(tmp4, tmp2, tmp6)
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp7, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = 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, 1), (4, 1, 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, 1), torch.float32)
extern_kernels.mm(primals_2, primals_1, out=buf0)
del primals_1
del primals_2
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_leaky_relu_0[grid(16)](buf1, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf2 = extern_kernels.convolution(reinterpret_tensor(buf1, (1, 4, 4
), (0, 4, 1), 0), primals_3, stride=(1,), padding=(0,),
dilation=(1,), transposed=False, output_padding=(0,), groups=1,
bias=None)
assert_size_stride(buf2, (1, 4, 4), (16, 4, 1))
buf3 = empty_strided_cuda((4, 4), (4, 1), torch.bool)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_leaky_relu_1[grid(16)](buf2, primals_4, buf3, buf4,
16, XBLOCK=16, num_warps=1, num_stages=1)
del buf2
del primals_4
return buf4, primals_3, reinterpret_tensor(buf1, (1, 4, 4), (16, 4, 1), 0
), buf3
class GraphConvolutionNew(nn.Module):
def __init__(self, in_dim, out_dim):
super(GraphConvolutionNew, self).__init__()
self.relu = nn.LeakyReLU(0.2)
self.weight = nn.Conv1d(in_dim, out_dim, 1)
def forward(self, input_0, input_1):
primals_3 = self.weight.weight
primals_4 = self.weight.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
Kanaricc/TDRG
|
GraphConvolution
| false
| 8,391
|
[
"Apache-2.0"
] | 16
|
91416976c8887877775f516ebee60469449e7e5f
|
https://github.com/Kanaricc/TDRG/tree/91416976c8887877775f516ebee60469449e7e5f
|
ANet
|
# 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/v2/cv24vzifrszeh6ykktpm44yxw6ipfsqq636xos3qcxnfqk5ld5ed.py
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# x_1 => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_1,), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_relu_threshold_backward_0 = async_compile.triton('triton_poi_fused_relu_threshold_backward_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i1', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 1920
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 30
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/4v/c4vkgjo35ehvj3v3z2oxle5g724blvghra57lqsucwpp5wyabanr.py
# Topologically Sorted Source Nodes: [x_3, actions_value], Original ATen: [aten.tanh, aten.mul]
# Source node to ATen node mapping:
# actions_value => mul
# x_3 => tanh
# Graph fragment:
# %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 = (%tanh, 2), kwargs = {})
triton_poi_fused_mul_tanh_1 = async_compile.triton('triton_poi_fused_mul_tanh_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_tanh_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_tanh_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = libdevice.tanh(tmp0)
tmp2 = 2.0
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x0), 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, (30, 4), (4, 1))
assert_size_stride(primals_2, (30, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 30), (30, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 30), (30, 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, 30), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 30), (480, 120, 30, 1), 0); del buf0 # reuse
buf4 = empty_strided_cuda((4, 4, 4, 30), (480, 120, 30, 1), torch.bool)
# Topologically Sorted Source Nodes: [x_1], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf4, 1920, grid=grid(1920), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf1, (64, 30), (30, 1), 0), reinterpret_tensor(primals_4, (30, 4), (1, 30), 0), alpha=1, beta=1, out=buf2)
del primals_5
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_3, actions_value], Original ATen: [aten.tanh, aten.mul]
triton_poi_fused_mul_tanh_1.run(buf2, buf3, 256, grid=grid(256), stream=stream0)
return (buf3, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(buf1, (64, 30), (30, 1), 0), buf2, primals_4, buf4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((30, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((30, ), (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, 30), (30, 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 ANet(nn.Module):
def __init__(self, s_dim, a_dim):
super(ANet, self).__init__()
self.fc1 = nn.Linear(s_dim, 30)
self.fc1.weight.data.normal_(0, 0.1)
self.out = nn.Linear(30, a_dim)
self.out.weight.data.normal_(0, 0.1)
def forward(self, x):
x = self.fc1(x)
x = F.relu(x)
x = self.out(x)
x = F.tanh(x)
actions_value = x * 2
return actions_value
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'s_dim': 4, 'a_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_relu_threshold_backward_0(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 1920
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 30
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_mul_tanh_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = libdevice.tanh(tmp0)
tmp2 = 2.0
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + x0, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (30, 4), (4, 1))
assert_size_stride(primals_2, (30,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 30), (30, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 30), (30, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 30), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 30), (480, 120, 30, 1), 0)
del buf0
buf4 = empty_strided_cuda((4, 4, 4, 30), (480, 120, 30, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0[grid(1920)](buf1,
primals_2, buf4, 1920, 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, 30),
(30, 1), 0), reinterpret_tensor(primals_4, (30, 4), (1, 30), 0),
alpha=1, beta=1, out=buf2)
del primals_5
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_tanh_1[grid(256)](buf2, buf3, 256, XBLOCK=256,
num_warps=4, num_stages=1)
return buf3, reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(buf1, (64, 30), (30, 1), 0
), buf2, primals_4, buf4
class ANetNew(nn.Module):
def __init__(self, s_dim, a_dim):
super(ANetNew, self).__init__()
self.fc1 = nn.Linear(s_dim, 30)
self.fc1.weight.data.normal_(0, 0.1)
self.out = nn.Linear(30, a_dim)
self.out.weight.data.normal_(0, 0.1)
def forward(self, input_0):
primals_1 = self.fc1.weight
primals_2 = self.fc1.bias
primals_4 = self.out.weight
primals_5 = self.out.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
Kernels-K/DDPG-pytorch-
|
ANet
| false
| 8,392
|
[
"MIT"
] | 26
|
9a80a56f52f2232e5bd197521d3d2d388b48c882
|
https://github.com/Kernels-K/DDPG-pytorch-/tree/9a80a56f52f2232e5bd197521d3d2d388b48c882
|
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/tq/ctq3tnkpu5ah3f4fjhbdz4zrfqwhkcvwi7xe6ijdc6ur6cgmqpi5.py
# Topologically Sorted Source Nodes: [mul, numerator, mul_1, sum_2, mul_2, sum_3], Original ATen: [aten.mul, aten.sum]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# mul_2 => mul_2
# numerator => sum_1
# 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_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 = {})
# %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 = {})
triton_per_fused_mul_sum_0 = async_compile.triton('triton_per_fused_mul_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_0', '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_0(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 + ((16*x0) + (64*(r1 // 16)) + (r1 % 16)), xmask, other=0.0)
tmp1 = tl.load(in_ptr1 + ((16*x0) + (64*(r1 // 16)) + (r1 % 16)), 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]
tl.store(out_ptr0 + (x0), tmp6, xmask)
tl.store(out_ptr1 + (x0), tmp11, xmask)
tl.store(out_ptr2 + (x0), tmp16, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ow/cow45uy76wdortu6qm2wtsqlpzlmjty77mvzrpxh2y7u3t4lgfoy.py
# Topologically Sorted Source Nodes: [denominator, clamp, truediv, channelwise_score, channelwise_score_1, score], Original ATen: [aten.add, aten.clamp, aten.div, aten.mul, aten.rsub, aten.sum]
# Source node to ATen node mapping:
# channelwise_score => mul_3
# channelwise_score_1 => sub
# clamp => clamp_min
# denominator => add
# score => sum_4
# truediv => div
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sum_2, %sum_3), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%add, 1e-07), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, %clamp_min), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div, 2), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %mul_3), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%sub,), kwargs = {})
triton_per_fused_add_clamp_div_mul_rsub_sum_1 = async_compile.triton('triton_per_fused_add_clamp_div_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.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_clamp_div_mul_rsub_sum_1', '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_add_clamp_div_mul_rsub_sum_1(in_ptr0, in_ptr1, in_ptr2, 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 = tl.load(in_ptr2 + (r0), None)
tmp3 = tmp1 + tmp2
tmp4 = 1e-07
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp6 = tmp0 / tmp5
tmp7 = 2.0
tmp8 = tmp6 * tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp13 = tl.sum(tmp11, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp13, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, 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)
buf1 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf2 = empty_strided_cuda((4, ), (1, ), torch.float32)
# Topologically Sorted Source Nodes: [mul, numerator, mul_1, sum_2, mul_2, sum_3], Original ATen: [aten.mul, aten.sum]
stream0 = get_raw_stream(0)
triton_per_fused_mul_sum_0.run(arg0_1, arg1_1, buf0, buf1, buf2, 4, 64, grid=grid(4), stream=stream0)
del arg0_1
del arg1_1
buf3 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [denominator, clamp, truediv, channelwise_score, channelwise_score_1, score], Original ATen: [aten.add, aten.clamp, aten.div, aten.mul, aten.rsub, aten.sum]
triton_per_fused_add_clamp_div_mul_rsub_sum_1.run(buf0, buf1, buf2, buf3, 1, 4, grid=grid(1), stream=stream0)
del buf0
del buf1
del buf2
return (buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.utils.data
def flatten_samples(input_):
"""
Flattens a tensor or a variable such that the channel axis is first and the sample axis
is second. The shapes are transformed as follows:
(N, C, H, W) --> (C, N * H * W)
(N, C, D, H, W) --> (C, N * D * H * W)
(N, C) --> (C, N)
The input must be atleast 2d.
"""
num_channels = input_.size(1)
permute_axes = list(range(input_.dim()))
permute_axes[0], permute_axes[1] = permute_axes[1], permute_axes[0]
permuted = input_.permute(*permute_axes).contiguous()
flattened = permuted.view(num_channels, -1)
return flattened
def dice_score(input_, target, invert=False, channelwise=True, eps=1e-07):
if channelwise:
input_ = flatten_samples(input_)
target = flatten_samples(target)
numerator = (input_ * target).sum(-1)
denominator = (input_ * input_).sum(-1) + (target * target).sum(-1)
channelwise_score = 2 * (numerator / denominator.clamp(min=eps))
if invert:
channelwise_score = 1.0 - channelwise_score
score = channelwise_score.sum()
else:
numerator = (input_ * target).sum()
denominator = (input_ * input_).sum() + (target * target).sum()
score = 2.0 * (numerator / denominator.clamp(min=eps))
if invert:
score = 1.0 - score
return score
class DiceLoss(nn.Module):
def __init__(self, channelwise=True, eps=1e-07):
super().__init__()
self.channelwise = channelwise
self.eps = eps
self.init_kwargs = {'channelwise': channelwise, 'eps': self.eps}
def forward(self, input_, target):
return dice_score(input_, target, invert=True, channelwise=self.
channelwise, eps=self.eps)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_per_fused_mul_sum_0(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 + (16 * x0 + 64 * (r1 // 16) + r1 % 16), xmask,
other=0.0)
tmp1 = tl.load(in_ptr1 + (16 * x0 + 64 * (r1 // 16) + r1 % 16), 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]
tl.store(out_ptr0 + x0, tmp6, xmask)
tl.store(out_ptr1 + x0, tmp11, xmask)
tl.store(out_ptr2 + x0, tmp16, xmask)
@triton.jit
def triton_per_fused_add_clamp_div_mul_rsub_sum_1(in_ptr0, in_ptr1, in_ptr2,
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 = tl.load(in_ptr2 + r0, None)
tmp3 = tmp1 + tmp2
tmp4 = 1e-07
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp6 = tmp0 / tmp5
tmp7 = 2.0
tmp8 = tmp6 * tmp7
tmp9 = 1.0
tmp10 = tmp9 - tmp8
tmp11 = tl.broadcast_to(tmp10, [XBLOCK, RBLOCK])
tmp13 = tl.sum(tmp11, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp13, 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,), torch.float32)
buf1 = empty_strided_cuda((4,), (1,), torch.float32)
buf2 = empty_strided_cuda((4,), (1,), torch.float32)
get_raw_stream(0)
triton_per_fused_mul_sum_0[grid(4)](arg0_1, arg1_1, buf0, buf1,
buf2, 4, 64, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
del arg1_1
buf3 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_add_clamp_div_mul_rsub_sum_1[grid(1)](buf0, buf1,
buf2, buf3, 1, 4, XBLOCK=1, num_warps=2, num_stages=1)
del buf0
del buf1
del buf2
return buf3,
def flatten_samples(input_):
"""
Flattens a tensor or a variable such that the channel axis is first and the sample axis
is second. The shapes are transformed as follows:
(N, C, H, W) --> (C, N * H * W)
(N, C, D, H, W) --> (C, N * D * H * W)
(N, C) --> (C, N)
The input must be atleast 2d.
"""
num_channels = input_.size(1)
permute_axes = list(range(input_.dim()))
permute_axes[0], permute_axes[1] = permute_axes[1], permute_axes[0]
permuted = input_.permute(*permute_axes).contiguous()
flattened = permuted.view(num_channels, -1)
return flattened
def dice_score(input_, target, invert=False, channelwise=True, eps=1e-07):
if channelwise:
input_ = flatten_samples(input_)
target = flatten_samples(target)
numerator = (input_ * target).sum(-1)
denominator = (input_ * input_).sum(-1) + (target * target).sum(-1)
channelwise_score = 2 * (numerator / denominator.clamp(min=eps))
if invert:
channelwise_score = 1.0 - channelwise_score
score = channelwise_score.sum()
else:
numerator = (input_ * target).sum()
denominator = (input_ * input_).sum() + (target * target).sum()
score = 2.0 * (numerator / denominator.clamp(min=eps))
if invert:
score = 1.0 - score
return score
class DiceLossNew(nn.Module):
def __init__(self, channelwise=True, eps=1e-07):
super().__init__()
self.channelwise = channelwise
self.eps = eps
self.init_kwargs = {'channelwise': channelwise, 'eps': self.eps}
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
JonasHell/torch-em
|
DiceLoss
| false
| 8,393
|
[
"MIT"
] | 13
|
2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
https://github.com/JonasHell/torch-em/tree/2e008e0cd2f0ea6681581374fce4f9f47b986d55
|
TopKMaxPooling
|
# 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/gp/cgpzobisq4epz675bab55n3i5egvwq4elzdi4mqpguqatsvrh5la.py
# Topologically Sorted Source Nodes: [sort], Original ATen: [aten.sort]
# Source node to ATen node mapping:
# sort => sort
# Graph fragment:
# %sort : [num_users=1] = call_function[target=torch.ops.aten.sort.default](args = (%view, 2, True), kwargs = {})
triton_per_fused_sort_0 = async_compile.triton('triton_per_fused_sort_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[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_sort_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}
)
@triton.jit
def triton_per_fused_sort_0(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 = r1
tmp2 = tmp1.to(tl.int16)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5, tmp6, = triton_helpers.sort_with_index(tmp3, tmp4, None, 1, stable=False, descending=True)
tl.store(out_ptr0 + (r1 + (16*x0)), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/dd/cddjypoux5hrfojvdxfyzz5iys422m4s5ehfksfpk36mcctvj25q.py
# Topologically Sorted Source Nodes: [sum_1, output], Original ATen: [aten.sum, aten.div]
# Source node to ATen node mapping:
# output => div
# sum_1 => sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%slice_1, [2]), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_1, 1), kwargs = {})
triton_poi_fused_div_sum_1 = async_compile.triton('triton_poi_fused_div_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=[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_sum_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_div_sum_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 + (16*x0), xmask, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 16), (64, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [sort], Original ATen: [aten.sort]
stream0 = get_raw_stream(0)
triton_per_fused_sort_0.run(arg0_1, buf0, 16, 16, grid=grid(16), stream=stream0)
del arg0_1
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sum_1, output], Original ATen: [aten.sum, aten.div]
triton_poi_fused_div_sum_1.run(buf0, buf2, 16, grid=grid(16), 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 TopKMaxPooling(nn.Module):
def __init__(self, kmax=1.0):
super(TopKMaxPooling, self).__init__()
self.kmax = kmax
@staticmethod
def get_positive_k(k, n):
if k <= 0:
return 0
elif k < 1:
return round(k * n)
elif k > n:
return int(n)
else:
return int(k)
def forward(self, input):
batch_size = input.size(0)
num_channels = input.size(1)
h = input.size(2)
w = input.size(3)
n = h * w
kmax = self.get_positive_k(self.kmax, n)
sorted, _indices = torch.sort(input.view(batch_size, num_channels,
n), dim=2, descending=True)
region_max = sorted.narrow(2, 0, kmax)
output = region_max.sum(2).div_(kmax)
return output.view(batch_size, num_channels)
def __repr__(self):
return self.__class__.__name__ + ' (kmax=' + str(self.kmax) + ')'
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import 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_sort_0(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 = r1
tmp2 = tmp1.to(tl.int16)
tmp3 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp4 = tl.broadcast_to(tmp2, [XBLOCK, RBLOCK])
tmp5, _tmp6 = triton_helpers.sort_with_index(tmp3, tmp4, None, 1,
stable=False, descending=True)
tl.store(out_ptr0 + (r1 + 16 * x0), tmp5, xmask)
@triton.jit
def triton_poi_fused_div_sum_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 + 16 * x0, xmask, eviction_policy='evict_last')
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 16), (64, 16, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_sort_0[grid(16)](arg0_1, buf0, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
del arg0_1
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_div_sum_1[grid(16)](buf0, buf2, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del buf0
return buf2,
class TopKMaxPoolingNew(nn.Module):
def __init__(self, kmax=1.0):
super(TopKMaxPoolingNew, self).__init__()
self.kmax = kmax
@staticmethod
def get_positive_k(k, n):
if k <= 0:
return 0
elif k < 1:
return round(k * n)
elif k > n:
return int(n)
else:
return int(k)
def __repr__(self):
return self.__class__.__name__ + ' (kmax=' + str(self.kmax) + ')'
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Kanaricc/TDRG
|
TopKMaxPooling
| false
| 8,394
|
[
"Apache-2.0"
] | 16
|
91416976c8887877775f516ebee60469449e7e5f
|
https://github.com/Kanaricc/TDRG/tree/91416976c8887877775f516ebee60469449e7e5f
|
HadamardProduct
|
# 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/pc/cpcfzvtgbrs7uwpmqjpoawjpnkkbj62npkxpjctpbf5sse3erqig.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, %primals_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
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
tl.store(out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4, ), (1, ))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(primals_2, primals_1, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
return (buf0, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class HadamardProduct(nn.Module):
def __init__(self, shape):
super(HadamardProduct, self).__init__()
self.weights = nn.Parameter(torch.rand(shape))
def forward(self, x):
return x * self.weights
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'shape': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_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
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
tl.store(out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2 = args
args.clear()
assert_size_stride(primals_1, (4,), (1,))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(256)](primals_2, primals_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
return buf0, primals_2
class HadamardProductNew(nn.Module):
def __init__(self, shape):
super(HadamardProductNew, self).__init__()
self.weights = nn.Parameter(torch.rand(shape))
def forward(self, input_0):
primals_1 = self.weights
primals_2 = input_0
output = call([primals_1, primals_2])
return output[0]
|
KimUyen/LSTM-BCI-Decoder
|
HadamardProduct
| false
| 8,395
|
[
"MIT"
] | 38
|
c7b4bd108335a4d6c7d99c00c263346026186b0b
|
https://github.com/KimUyen/LSTM-BCI-Decoder/tree/c7b4bd108335a4d6c7d99c00c263346026186b0b
|
ResNetBottleneck
|
# 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/fz/cfzj5bubecclqwixnyw4mgb2x6p65oossc6qcu54ffkv7e56hx4a.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# out => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution,), kwargs = {})
triton_poi_fused_relu_0 = async_compile.triton('triton_poi_fused_relu_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: 'i32'}, 'device': 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_relu_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_relu_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 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(in_out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/xl/cxldz53yhux33th37rzhp6akujnp4o5wil5llpv47kfpc5x5qt2d.py
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# out_3 => add
# out_4 => relu_2
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%convolution_2, %primals_1), kwargs = {})
# %relu_2 : [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_2, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_1 = async_compile.triton('triton_poi_fused_add_relu_threshold_backward_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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_relu_threshold_backward_1(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + (x0), tmp4, xmask)
tl.store(out_ptr0 + (x0), tmp6, xmask)
''', device_str='cuda')
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, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4, 4), (64, 16, 4, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_0.run(buf1, 256, grid=grid(256), stream=stream0)
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(buf1, primals_3, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 4, 4), (64, 16, 4, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [out_1], Original ATen: [aten.relu]
triton_poi_fused_relu_0.run(buf3, 256, grid=grid(256), stream=stream0)
# Topologically Sorted Source Nodes: [out_2], 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, 4, 4), (64, 16, 4, 1))
buf5 = buf4; del buf4 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [out_3, out_4], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_1.run(buf5, primals_1, buf6, 256, grid=grid(256), stream=stream0)
return (buf5, primals_1, primals_2, primals_3, 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((4, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 3, 3), (36, 9, 3, 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)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import nn
import torch.nn.functional as F
class ResNetBottleneck(nn.Module):
def __init__(self, in_channels, out_channels, bottleneck_channels,
stride, downsample=None):
super(ResNetBottleneck, self).__init__()
self.conv1 = nn.Conv2d(in_channels, bottleneck_channels,
kernel_size=1, bias=False)
self.conv2 = nn.Conv2d(bottleneck_channels, bottleneck_channels,
kernel_size=3, stride=stride, padding=1, bias=False)
self.conv3 = nn.Conv2d(bottleneck_channels, out_channels,
kernel_size=1, bias=False)
self.downsample = downsample
def forward(self, x):
residual = x
out = F.relu(self.conv1(x), inplace=True)
out = F.relu(self.conv2(out), inplace=True)
out = self.conv3(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = F.relu(out, inplace=True)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'out_channels': 4, 'bottleneck_channels':
4, 'stride': 1}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch 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_relu_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 = tl.full([1], 0, tl.int32)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tl.store(in_out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_1(in_out_ptr0, in_ptr0,
out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp5 = 0.0
tmp6 = tmp4 <= tmp5
tl.store(in_out_ptr0 + x0, tmp4, xmask)
tl.store(out_ptr0 + x0, tmp6, xmask)
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, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (4, 4, 3, 3), (36, 9, 3, 1))
assert_size_stride(primals_4, (4, 4, 1, 1), (4, 1, 1, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_1, primals_2, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4, 4), (64, 16, 4, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_relu_0[grid(256)](buf1, 256, XBLOCK=128, num_warps
=4, num_stages=1)
buf2 = extern_kernels.convolution(buf1, primals_3, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 4, 4), (64, 16, 4, 1))
buf3 = buf2
del buf2
triton_poi_fused_relu_0[grid(256)](buf3, 256, XBLOCK=128, num_warps
=4, num_stages=1)
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, 4, 4), (64, 16, 4, 1))
buf5 = buf4
del buf4
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_1[grid(256)](buf5,
primals_1, buf6, 256, XBLOCK=256, num_warps=4, num_stages=1)
return buf5, primals_1, primals_2, primals_3, primals_4, buf1, buf3, buf6
class ResNetBottleneckNew(nn.Module):
def __init__(self, in_channels, out_channels, bottleneck_channels,
stride, downsample=None):
super(ResNetBottleneckNew, self).__init__()
self.conv1 = nn.Conv2d(in_channels, bottleneck_channels,
kernel_size=1, bias=False)
self.conv2 = nn.Conv2d(bottleneck_channels, bottleneck_channels,
kernel_size=3, stride=stride, padding=1, bias=False)
self.conv3 = nn.Conv2d(bottleneck_channels, out_channels,
kernel_size=1, bias=False)
self.downsample = downsample
def forward(self, input_0):
primals_2 = self.conv1.weight
primals_3 = self.conv2.weight
primals_4 = self.conv3.weight
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
KH-Kyle/rmp_nav
|
ResNetBottleneck
| false
| 8,396
|
[
"MIT"
] | 30
|
d598fe70664a4cdc0e9b9dd4b52e84aa3de1b551
|
https://github.com/KH-Kyle/rmp_nav/tree/d598fe70664a4cdc0e9b9dd4b52e84aa3de1b551
|
GlobalAttention_text
|
# 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/3o/c3o4fyjxlkj4npsoxsqqqpk3uuygpey2bd23rcgrqskm4bg6g5ni.py
# Topologically Sorted Source Nodes: [targetT], Original ATen: [aten.clone, aten.transpose]
# Source node to ATen node mapping:
# targetT => clone
# Graph fragment:
# %clone : [num_users=2] = call_function[target=torch.ops.aten.clone.default](args = (%permute,), kwargs = {memory_format: torch.contiguous_format})
# %permute_2 : [num_users=1] = call_function[target=torch.ops.aten.permute.default](args = (%clone, [0, 2, 1]), kwargs = {})
triton_poi_fused_clone_transpose_0 = async_compile.triton('triton_poi_fused_clone_transpose_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_clone_transpose_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_transpose_0(in_ptr0, out_ptr0, out_ptr1, 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
x1 = xindex
y0 = yindex
y2 = yindex % 4
y3 = (yindex // 4)
tmp0 = tl.load(in_ptr0 + (x1 + (16*y0)), xmask & ymask)
tl.store(out_ptr0 + (x1 + (16*y0)), tmp0, xmask & ymask)
tl.store(out_ptr1 + (y2 + (4*x1) + (64*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bo/cbobw73zf5udllzw6ypzn2lpl3t5xgyic35frylcttxdzai224c5.py
# Topologically Sorted Source Nodes: [sourceT], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# sourceT => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_2, %primals_3, %primals_4, [1], [0], [1], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 4) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/aq/caqdgjxhqghu4g7d2hcwormbhukc2uvsqhz573l2nir4qeupb3xn.py
# Topologically Sorted Source Nodes: [attn_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_3 => amax, exp, sub, sum_1
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%view_2, [1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_2, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [1], True), kwargs = {})
triton_per_fused__softmax_2 = async_compile.triton('triton_per_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[16, 16],
reduction_hint=ReductionHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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__softmax_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_2(in_ptr0, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex
x0 = xindex % 4
x1 = (xindex // 4)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*r2) + (64*x1)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, float("-inf"))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp5 = tmp0 - tmp4
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = tl.where(xmask, tmp7, 0)
tmp10 = tl.sum(tmp9, 1)[:, None]
tl.store(out_ptr0 + (x3), tmp4, xmask)
tl.store(out_ptr1 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4w/c4w5ze4xkq5wurcwsrniggh5ndyhg6az2kt5ylkmvba5vzbrzicp.py
# Topologically Sorted Source Nodes: [attn_3], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# attn_3 => div, exp, sub
# Graph fragment:
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_2, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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__softmax_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__softmax_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
x3 = xindex
x0 = xindex % 4
x2 = (xindex // 64)
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tl.store(in_out_ptr0 + (x3), tmp5, 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), (16, 4, 1))
assert_size_stride(primals_3, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_4, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [sourceT], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_2, primals_3, stride=(1,), padding=(0,), dilation=(1,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = empty_strided_cuda((4, 16, 4), (64, 1, 16), torch.float32)
buf8 = empty_strided_cuda((4, 4, 16), (64, 1, 4), torch.float32)
# Topologically Sorted Source Nodes: [targetT], Original ATen: [aten.clone, aten.transpose]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_transpose_0.run(primals_1, buf1, buf8, 16, 16, grid=grid(16, 16), stream=stream0)
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [sourceT], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf2, primals_4, 64, grid=grid(64), stream=stream0)
del primals_4
buf3 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [targetT, sourceT, attn], Original ATen: [aten.clone, aten.convolution, aten.bmm]
extern_kernels.bmm(buf1, buf2, out=buf3)
del buf1
buf4 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf5 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
# Topologically Sorted Source Nodes: [attn_3], Original ATen: [aten._softmax]
triton_per_fused__softmax_2.run(buf3, buf4, buf5, 16, 16, grid=grid(16), stream=stream0)
buf6 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [attn_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf6, buf4, buf5, 256, grid=grid(256), stream=stream0)
del buf4
del buf5
buf7 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [text_weighted], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(primals_1, (4, 4, 16), (64, 16, 1), 0), buf6, out=buf7)
return (buf7, primals_2, primals_3, reinterpret_tensor(primals_1, (4, 16, 4), (64, 1, 16), 0), buf6, 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), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 1), (4, 1, 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.nn as nn
import torch.nn.parallel
class GlobalAttention_text(nn.Module):
def __init__(self, idf, cdf):
super(GlobalAttention_text, self).__init__()
self.conv_context = nn.Conv1d(cdf, idf, kernel_size=1, stride=1,
padding=0)
self.sm = nn.Softmax()
self.mask = None
def applyMask(self, mask):
self.mask = mask
def forward(self, input, context):
"""
input: batch x idf x ih x iw (queryL=ihxiw)
context: batch x cdf x sourceL
"""
ih, iw = input.size(2), input.size(3)
queryL = ih * iw
batch_size, sourceL = context.size(0), context.size(2)
target = input.view(batch_size, -1, queryL)
targetT = torch.transpose(target, 1, 2).contiguous()
sourceT = self.conv_context(context)
attn = torch.bmm(targetT, sourceT)
attn = attn.view(batch_size * queryL, sourceL)
if self.mask is not None:
mask = self.mask.repeat(queryL, 1)
attn.data.masked_fill_(mask.data, -float('inf'))
attn = attn.view(batch_size, queryL, sourceL)
attn = torch.nn.Softmax(dim=1)(attn)
text_weighted = torch.bmm(target, attn)
return text_weighted
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'idf': 4, 'cdf': 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
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_transpose_0(in_ptr0, out_ptr0, out_ptr1, 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
x1 = xindex
y0 = yindex
y2 = yindex % 4
y3 = yindex // 4
tmp0 = tl.load(in_ptr0 + (x1 + 16 * y0), xmask & ymask)
tl.store(out_ptr0 + (x1 + 16 * y0), tmp0, xmask & ymask)
tl.store(out_ptr1 + (y2 + 4 * x1 + 64 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 4 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
@triton.jit
def triton_per_fused__softmax_2(in_ptr0, out_ptr0, out_ptr1, xnumel, rnumel,
XBLOCK: tl.constexpr):
xnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r2 = rindex
x0 = xindex % 4
x1 = xindex // 4
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * r2 + 64 * x1), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, float('-inf'))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp5 = tmp0 - tmp4
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = tl.where(xmask, tmp7, 0)
tmp10 = tl.sum(tmp9, 1)[:, None]
tl.store(out_ptr0 + x3, tmp4, xmask)
tl.store(out_ptr1 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused__softmax_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
x3 = xindex
x0 = xindex % 4
x2 = xindex // 64
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp4 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.exp(tmp2)
tmp5 = tmp3 / tmp4
tl.store(in_out_ptr0 + x3, tmp5, 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), (16, 4, 1))
assert_size_stride(primals_3, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_4, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_2, primals_3, stride=(1,),
padding=(0,), dilation=(1,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 4), (16, 4, 1))
buf1 = empty_strided_cuda((4, 16, 4), (64, 1, 16), torch.float32)
buf8 = empty_strided_cuda((4, 4, 16), (64, 1, 4), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_transpose_0[grid(16, 16)](primals_1, buf1,
buf8, 16, 16, XBLOCK=16, YBLOCK=16, num_warps=4, num_stages=1)
buf2 = buf0
del buf0
triton_poi_fused_convolution_1[grid(64)](buf2, primals_4, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
extern_kernels.bmm(buf1, buf2, out=buf3)
del buf1
buf4 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
buf5 = empty_strided_cuda((4, 1, 4), (4, 16, 1), torch.float32)
triton_per_fused__softmax_2[grid(16)](buf3, buf4, buf5, 16, 16,
XBLOCK=1, num_warps=2, num_stages=1)
buf6 = buf3
del buf3
triton_poi_fused__softmax_3[grid(256)](buf6, buf4, buf5, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del buf4
del buf5
buf7 = buf2
del buf2
extern_kernels.bmm(reinterpret_tensor(primals_1, (4, 4, 16), (64,
16, 1), 0), buf6, out=buf7)
return buf7, primals_2, primals_3, reinterpret_tensor(primals_1, (4, 16,
4), (64, 1, 16), 0), buf6, buf8
class GlobalAttention_textNew(nn.Module):
def __init__(self, idf, cdf):
super(GlobalAttention_textNew, self).__init__()
self.conv_context = nn.Conv1d(cdf, idf, kernel_size=1, stride=1,
padding=0)
self.sm = nn.Softmax()
self.mask = None
def applyMask(self, mask):
self.mask = mask
def forward(self, input_0, input_1):
primals_3 = self.conv_context.weight
primals_4 = self.conv_context.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
JoonHong-Kim/T2I_CL
|
GlobalAttention_text
| false
| 8,397
|
[
"MIT"
] | 35
|
c52aa73da903d6e4174eeef2663e5bc1163785b1
|
https://github.com/JoonHong-Kim/T2I_CL/tree/c52aa73da903d6e4174eeef2663e5bc1163785b1
|
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/lg/clgvalhvo7ygyvwhczo52ybtcvk6b7yiltaj5x7jlylvvui4lb63.py
# Topologically Sorted Source Nodes: [hx], Original ATen: [aten.zeros]
# Source node to ATen node mapping:
# hx => full_default
# Graph fragment:
# %full_default : [num_users=3] = call_function[target=torch.ops.aten.full.default](args = ([4, 4], 0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cuda:0, pin_memory: False})
triton_poi_fused_zeros_0 = async_compile.triton('triton_poi_fused_zeros_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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_zeros_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_zeros_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
tmp0 = 0.0
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (12, 4), (4, 1))
assert_size_stride(primals_3, (12, 4), (4, 1))
assert_size_stride(primals_4, (12, ), (1, ))
assert_size_stride(primals_5, (12, ), (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: [hx], Original ATen: [aten.zeros]
stream0 = get_raw_stream(0)
triton_poi_fused_zeros_0.run(buf0, 16, grid=grid(16), stream=stream0)
buf1 = empty_strided_cuda((4, 12), (12, 1), torch.float32)
# Topologically Sorted Source Nodes: [ret], Original ATen: [aten.mm]
extern_kernels.mm(primals_1, reinterpret_tensor(primals_2, (4, 12), (1, 4), 0), out=buf1)
del primals_2
buf2 = empty_strided_cuda((4, 12), (12, 1), torch.float32)
# Topologically Sorted Source Nodes: [ret], Original ATen: [aten.mm]
extern_kernels.mm(buf0, reinterpret_tensor(primals_3, (4, 12), (1, 4), 0), out=buf2)
del primals_3
# Topologically Sorted Source Nodes: [ret], Original ATen: [aten._thnn_fused_gru_cell]
buf3 = torch.ops.aten._thnn_fused_gru_cell.default(buf1, buf2, buf0, primals_4, primals_5)
del buf1
del buf2
del primals_4
del primals_5
buf4 = buf3[0]
buf5 = buf3[1]
del buf3
return (buf4, primals_1, buf0, 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, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((12, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((12, ), (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 GRUCell(nn.Module):
def __init__(self, input_size, hidden_size, init_scale=1.0,
no_weight_init=False):
super(GRUCell, self).__init__()
self.recurrent = nn.GRUCell(input_size, hidden_size)
if not no_weight_init:
for name, param in self.recurrent.named_parameters():
if 'weight' in name:
nn.init.orthogonal_(param, init_scale)
weight_inited = True
elif 'bias' in name:
with torch.no_grad():
param.zero_()
assert weight_inited
def forward(self, x, h=None):
return self.recurrent(x, h)
def get_inputs():
return [torch.rand([4, 4])]
def get_init_inputs():
return [[], {'input_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 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_zeros_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
tmp0 = 0.0
tl.store(out_ptr0 + x0, tmp0, 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, (12, 4), (4, 1))
assert_size_stride(primals_3, (12, 4), (4, 1))
assert_size_stride(primals_4, (12,), (1,))
assert_size_stride(primals_5, (12,), (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_zeros_0[grid(16)](buf0, 16, XBLOCK=16, num_warps=1,
num_stages=1)
buf1 = empty_strided_cuda((4, 12), (12, 1), torch.float32)
extern_kernels.mm(primals_1, reinterpret_tensor(primals_2, (4, 12),
(1, 4), 0), out=buf1)
del primals_2
buf2 = empty_strided_cuda((4, 12), (12, 1), torch.float32)
extern_kernels.mm(buf0, reinterpret_tensor(primals_3, (4, 12), (1,
4), 0), out=buf2)
del primals_3
buf3 = torch.ops.aten._thnn_fused_gru_cell.default(buf1, buf2, buf0,
primals_4, primals_5)
del buf1
del buf2
del primals_4
del primals_5
buf4 = buf3[0]
buf5 = buf3[1]
del buf3
return buf4, primals_1, buf0, buf5
class GRUCellNew(nn.Module):
def __init__(self, input_size, hidden_size, init_scale=1.0,
no_weight_init=False):
super(GRUCellNew, self).__init__()
self.recurrent = nn.GRUCell(input_size, hidden_size)
if not no_weight_init:
for name, param in self.recurrent.named_parameters():
if 'weight' in name:
nn.init.orthogonal_(param, init_scale)
weight_inited = True
elif 'bias' in name:
with torch.no_grad():
param.zero_()
assert weight_inited
def forward(self, input_0):
primals_2 = self.recurrent.weight_ih
primals_3 = self.recurrent.weight_hh
primals_4 = self.recurrent.bias_ih
primals_5 = self.recurrent.bias_hh
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
KH-Kyle/rmp_nav
|
GRUCell
| false
| 8,398
|
[
"MIT"
] | 30
|
d598fe70664a4cdc0e9b9dd4b52e84aa3de1b551
|
https://github.com/KH-Kyle/rmp_nav/tree/d598fe70664a4cdc0e9b9dd4b52e84aa3de1b551
|
Fusion
|
# 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/37/c37l7ezffklnfomgas7mto2m3yaztb5ikq2a6o2uoa4usmb6uodq.py
# Topologically Sorted Source Nodes: [sub, pow_1, neg, add, relu, add_1], Original ATen: [aten.sub, aten.pow, aten.neg, aten.add, aten.relu]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# neg => neg
# pow_1 => pow_1
# relu => relu
# 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 = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%pow_1,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, %arg1_1), kwargs = {})
# %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%add,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg, %relu), kwargs = {})
triton_poi_fused_add_neg_pow_relu_sub_0 = async_compile.triton('triton_poi_fused_add_neg_pow_relu_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_neg_pow_relu_sub_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_add_neg_pow_relu_sub_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp4 = -tmp3
tmp5 = tmp0 + tmp1
tmp6 = tl.full([1], 0, tl.int32)
tmp7 = triton_helpers.maximum(tmp6, tmp5)
tmp8 = tmp4 + tmp7
tl.store(out_ptr0 + (x0), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, 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, neg, add, relu, add_1], Original ATen: [aten.sub, aten.pow, aten.neg, aten.add, aten.relu]
stream0 = get_raw_stream(0)
triton_poi_fused_add_neg_pow_relu_sub_0.run(arg0_1, arg1_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
del arg1_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1, arg1_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data
class Fusion(nn.Module):
""" Crazy multi-modal fusion: negative squared difference minus relu'd sum
"""
def __init__(self):
super().__init__()
def forward(self, x, y):
return -(x - y) ** 2 + F.relu(x + y)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_add_neg_pow_relu_sub_0(in_ptr0, in_ptr1, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp2 = tmp0 - tmp1
tmp3 = tmp2 * tmp2
tmp4 = -tmp3
tmp5 = tmp0 + tmp1
tmp6 = tl.full([1], 0, tl.int32)
tmp7 = triton_helpers.maximum(tmp6, tmp5)
tmp8 = tmp4 + tmp7
tl.store(out_ptr0 + x0, tmp8, 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_add_neg_pow_relu_sub_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 FusionNew(nn.Module):
""" Crazy multi-modal fusion: negative squared difference minus relu'd sum
"""
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]
|
KaihuaTang/VCTree-Visual-Question-Answering
|
Fusion
| false
| 8,399
|
[
"MIT"
] | 31
|
b6b0a8bdb01d45d36de3bded91db42544ad6a593
|
https://github.com/KaihuaTang/VCTree-Visual-Question-Answering/tree/b6b0a8bdb01d45d36de3bded91db42544ad6a593
|
CommandEmbedding
|
# 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/hn/chnlbedg5ibqfgikafvmbktyy5csv2cuovxhtpw4fviqsav6xwtj.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 = ([%view_1, %expand], -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=[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_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 = 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 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 2, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + ((2*x1) + x0), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 4, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + ((-2) + 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')
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, (2, 4), (4, 1))
assert_size_stride(primals_2, (2, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (1, 1, 2), (2, 2, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 2), (2, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 2), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4), (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(buf0, primals_4, buf1, 64, grid=grid(64), stream=stream0)
del buf0
del primals_4
return (buf1, reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((2, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((2, ), (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((1, 1, 2), (2, 2, 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 Tensor
from torch import nn
class CommandEmbedding(nn.Module):
def __init__(self, input_size, output_size):
super().__init__()
self.embedding = nn.Linear(input_size, output_size // 2)
self.encoding = nn.Parameter(torch.rand(1, 1, output_size // 2))
def forward(self, command: 'Tensor') ->Tensor:
return torch.cat([self.embedding(command), self.encoding.expand(
command.size(0), command.size(1), -1)], dim=-1)
def get_inputs():
return [torch.rand([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 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_cat_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 2, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (2 * x1 + x0), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 4, tl.int64)
tmp9 = tl.load(in_ptr1 + (-2 + x0), 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):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (2, 4), (4, 1))
assert_size_stride(primals_2, (2,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (1, 1, 2), (2, 2, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 2), (2, 1), torch.float32)
extern_kernels.addmm(primals_2, reinterpret_tensor(primals_3, (16,
4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 2), (1, 4), 0
), alpha=1, beta=1, out=buf0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(64)](buf0, primals_4, buf1, 64, XBLOCK=
64, num_warps=1, num_stages=1)
del buf0
del primals_4
return buf1, reinterpret_tensor(primals_3, (16, 4), (4, 1), 0)
class CommandEmbeddingNew(nn.Module):
def __init__(self, input_size, output_size):
super().__init__()
self.embedding = nn.Linear(input_size, output_size // 2)
self.encoding = nn.Parameter(torch.rand(1, 1, output_size // 2))
def forward(self, input_0):
primals_4 = self.encoding
primals_1 = self.embedding.weight
primals_2 = self.embedding.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
Kaixhin/GUDRL
|
CommandEmbedding
| false
| 8,400
|
[
"MIT"
] | 26
|
c13fa605a9ffb4c2932390b0b86e476aec62c142
|
https://github.com/Kaixhin/GUDRL/tree/c13fa605a9ffb4c2932390b0b86e476aec62c142
|
BertLayerNormNoVar
|
# 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/s3/cs3abmbgm7lvmghtpxomr2eh7e6hhipu4ylt6bod3zgafqjxmfc2.py
# Topologically Sorted Source Nodes: [u, x, mul, add], Original ATen: [aten.mean, aten.sub, aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# mul => mul
# u => mean
# x => sub
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_1, %mean), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_2, %sub), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %primals_3), kwargs = {})
triton_poi_fused_add_mean_mul_sub_0 = async_compile.triton('triton_poi_fused_add_mean_mul_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_mean_mul_sub_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_mean_mul_sub_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp0 * tmp11
tmp14 = tmp12 + tmp13
tl.store(out_ptr0 + (x2), tmp14, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [u, x, mul, add], Original ATen: [aten.mean, aten.sub, aten.mul, aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_mean_mul_sub_0.run(primals_2, primals_1, primals_3, buf0, 256, grid=grid(256), stream=stream0)
del primals_2
del primals_3
return (buf0, primals_1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class BertLayerNormNoVar(nn.Module):
def __init__(self, hidden_size, eps=1e-12):
super(BertLayerNormNoVar, self).__init__()
self.weight = nn.Parameter(torch.ones(hidden_size))
self.bias = nn.Parameter(torch.zeros(hidden_size))
self.variance_epsilon = eps
def forward(self, x):
u = x.mean(-1, keepdim=True)
x = x - u
return self.weight * x + self.bias
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'hidden_size': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
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_sub_0(in_ptr0, in_ptr1, in_ptr2, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp13 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp9 = 4.0
tmp10 = tmp8 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp0 * tmp11
tmp14 = tmp12 + tmp13
tl.store(out_ptr0 + x2, tmp14, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_mean_mul_sub_0[grid(256)](primals_2, primals_1,
primals_3, buf0, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
del primals_3
return buf0, primals_1
class BertLayerNormNoVarNew(nn.Module):
def __init__(self, hidden_size, eps=1e-12):
super(BertLayerNormNoVarNew, self).__init__()
self.weight = nn.Parameter(torch.ones(hidden_size))
self.bias = nn.Parameter(torch.zeros(hidden_size))
self.variance_epsilon = eps
def forward(self, input_0):
primals_2 = self.weight
primals_3 = self.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
KaidiXu/LiRPA_Verify
|
BertLayerNormNoVar
| false
| 8,401
|
[
"BSD-2-Clause"
] | 14
|
71f5327a8abf136bcfb3e1ec07604628abf8126e
|
https://github.com/KaidiXu/LiRPA_Verify/tree/71f5327a8abf136bcfb3e1ec07604628abf8126e
|
ConvLSTMCell
|
# 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/xc/cxceqiy6okvqxrjbwoo7ygc4plr32q3u54opf7nc2ufvkbmb4dzh.py
# Topologically Sorted Source Nodes: [stacked_inputs], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# stacked_inputs => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_1, %primals_2], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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
x1 = (xindex // 16) % 8
x0 = xindex % 16
x2 = (xindex // 128)
x3 = xindex
tmp0 = x1
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + (16*x1) + (64*x2)), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x0 + (16*((-4) + x1)) + (64*x2)), tmp6 & xmask, other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + (x3), tmp10, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/c2/cc2dm6t5shxvlslm54ib7dbfegegnre2lr6ktlpilmy7q5rgj3he.py
# Topologically Sorted Source Nodes: [xi_1, xf_1, xo_1, xg_1, mul, mul_1, c, tanh_1, h], Original ATen: [aten.sigmoid, aten.tanh, aten.mul, aten.add, aten.sigmoid_backward]
# Source node to ATen node mapping:
# c => add
# h => mul_2
# mul => mul
# mul_1 => mul_1
# tanh_1 => tanh_1
# xf_1 => sigmoid_1
# xg_1 => tanh
# xi_1 => sigmoid
# xo_1 => sigmoid_2
# Graph fragment:
# %sigmoid : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem,), kwargs = {})
# %sigmoid_1 : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem_1,), kwargs = {})
# %sigmoid_2 : [num_users=2] = call_function[target=torch.ops.aten.sigmoid.default](args = (%getitem_2,), kwargs = {})
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%getitem_3,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid_1, %primals_5), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %tanh), kwargs = {})
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
# %tanh_1 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid_2, %tanh_1), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid_1), kwargs = {})
# %mul_14 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid_1, %sub_3), kwargs = {})
triton_poi_fused_add_mul_sigmoid_sigmoid_backward_tanh_1 = async_compile.triton('triton_poi_fused_add_mul_sigmoid_sigmoid_backward_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: '*fp32', 7: '*fp32', 8: '*fp32', 9: 'i32'}, 'device': 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, 8, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_mul_sigmoid_sigmoid_backward_tanh_1', '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_add_mul_sigmoid_sigmoid_backward_tanh_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, out_ptr3, out_ptr4, out_ptr5, xnumel, XBLOCK : tl.constexpr):
xnumel = 144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 36)
x4 = xindex % 36
x1 = (xindex // 9) % 4
x3 = xindex
tmp0 = tl.load(in_ptr0 + (72 + x4 + (144*x2)), xmask)
tmp1 = tl.load(in_ptr1 + (8 + x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (x4 + (144*x2)), xmask)
tmp5 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (108 + x4 + (144*x2)), xmask)
tmp9 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (36 + x4 + (144*x2)), xmask)
tmp13 = tl.load(in_ptr1 + (4 + x1), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr2 + (x3), xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tmp6 = tmp4 + tmp5
tmp7 = tl.sigmoid(tmp6)
tmp10 = tmp8 + tmp9
tmp11 = libdevice.tanh(tmp10)
tmp14 = tmp12 + tmp13
tmp15 = tl.sigmoid(tmp14)
tmp17 = tmp15 * tmp16
tmp18 = tmp7 * tmp11
tmp19 = tmp17 + tmp18
tmp20 = 1.0
tmp21 = tmp20 - tmp15
tmp22 = tmp15 * tmp21
tmp23 = libdevice.tanh(tmp19)
tmp24 = tmp3 * tmp23
tl.store(out_ptr0 + (x3), tmp3, xmask)
tl.store(out_ptr1 + (x3), tmp7, xmask)
tl.store(out_ptr2 + (x3), tmp11, xmask)
tl.store(out_ptr3 + (x3), tmp19, xmask)
tl.store(out_ptr4 + (x3), tmp22, xmask)
tl.store(out_ptr5 + (x3), tmp24, 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, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (16, 8, 4, 4), (128, 16, 4, 1))
assert_size_stride(primals_4, (16, ), (1, ))
assert_size_stride(primals_5, (4, 4, 3, 3), (36, 9, 3, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 8, 4, 4), (128, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [stacked_inputs], 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)
del primals_1
del primals_2
# Topologically Sorted Source Nodes: [gates], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(buf0, primals_3, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 16, 3, 3), (144, 9, 3, 1))
buf3 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf2 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf4 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf6 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
# Topologically Sorted Source Nodes: [xi_1, xf_1, xo_1, xg_1, mul, mul_1, c, tanh_1, h], Original ATen: [aten.sigmoid, aten.tanh, aten.mul, aten.add, aten.sigmoid_backward]
triton_poi_fused_add_mul_sigmoid_sigmoid_backward_tanh_1.run(buf1, primals_4, primals_5, buf3, buf2, buf4, buf5, buf7, buf6, 144, grid=grid(144), stream=stream0)
del buf1
del primals_4
return (buf6, buf5, primals_3, primals_5, buf0, buf2, buf3, buf4, buf5, buf7, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, 8, 4, 4), (128, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, 4, 3, 3), (36, 9, 3, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
from torch.autograd import Variable
class ConvLSTMCell(nn.Module):
def __init__(self, input_channels, hidden_channels, kernel_size, bias=True
):
super(ConvLSTMCell, self).__init__()
assert hidden_channels % 2 == 0
self.input_channels = input_channels
self.hidden_channels = hidden_channels
self.bias = bias
self.kernel_size = kernel_size
self.padding = int((kernel_size - 1) / 2)
self.Gates = nn.Conv2d(self.input_channels + self.hidden_channels,
4 * self.hidden_channels, self.kernel_size, 1, self.padding,
bias=True)
def forward(self, x, h, c):
stacked_inputs = torch.cat((x, h), 1)
gates = self.Gates(stacked_inputs)
xi, xf, xo, xg = gates.chunk(4, 1)
xi = torch.sigmoid(xi)
xf = torch.sigmoid(xf)
xo = torch.sigmoid(xo)
xg = torch.tanh(xg)
c = xf * c + xi * xg
h = xo * torch.tanh(c)
return h, c
def init_hidden(self, batch_size, hidden, shape):
return Variable(torch.zeros(batch_size, hidden, shape[0], shape[1])
), Variable(torch.zeros(batch_size, hidden, shape[0], shape[1]))
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4]), torch.rand(
[4, 4, 3, 3])]
def get_init_inputs():
return [[], {'input_channels': 4, 'hidden_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.triton_helpers import libdevice
import torch.nn as nn
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
@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
x1 = xindex // 16 % 8
x0 = xindex % 16
x2 = xindex // 128
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 16 * x1 + 64 * x2), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (x0 + 16 * (-4 + x1) + 64 * x2), tmp6 & xmask,
other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tl.store(out_ptr0 + x3, tmp10, xmask)
@triton.jit
def triton_poi_fused_add_mul_sigmoid_sigmoid_backward_tanh_1(in_ptr0,
in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, out_ptr3, out_ptr4,
out_ptr5, xnumel, XBLOCK: tl.constexpr):
xnumel = 144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 36
x4 = xindex % 36
x1 = xindex // 9 % 4
x3 = xindex
tmp0 = tl.load(in_ptr0 + (72 + x4 + 144 * x2), xmask)
tmp1 = tl.load(in_ptr1 + (8 + x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (x4 + 144 * x2), xmask)
tmp5 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp8 = tl.load(in_ptr0 + (108 + x4 + 144 * x2), xmask)
tmp9 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp12 = tl.load(in_ptr0 + (36 + x4 + 144 * x2), xmask)
tmp13 = tl.load(in_ptr1 + (4 + x1), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr2 + x3, xmask)
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tmp6 = tmp4 + tmp5
tmp7 = tl.sigmoid(tmp6)
tmp10 = tmp8 + tmp9
tmp11 = libdevice.tanh(tmp10)
tmp14 = tmp12 + tmp13
tmp15 = tl.sigmoid(tmp14)
tmp17 = tmp15 * tmp16
tmp18 = tmp7 * tmp11
tmp19 = tmp17 + tmp18
tmp20 = 1.0
tmp21 = tmp20 - tmp15
tmp22 = tmp15 * tmp21
tmp23 = libdevice.tanh(tmp19)
tmp24 = tmp3 * tmp23
tl.store(out_ptr0 + x3, tmp3, xmask)
tl.store(out_ptr1 + x3, tmp7, xmask)
tl.store(out_ptr2 + x3, tmp11, xmask)
tl.store(out_ptr3 + x3, tmp19, xmask)
tl.store(out_ptr4 + x3, tmp22, xmask)
tl.store(out_ptr5 + x3, tmp24, 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, 4), (64, 16, 4, 1))
assert_size_stride(primals_3, (16, 8, 4, 4), (128, 16, 4, 1))
assert_size_stride(primals_4, (16,), (1,))
assert_size_stride(primals_5, (4, 4, 3, 3), (36, 9, 3, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 8, 4, 4), (128, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(512)](primals_1, primals_2, buf0, 512,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_1
del primals_2
buf1 = extern_kernels.convolution(buf0, primals_3, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 16, 3, 3), (144, 9, 3, 1))
buf3 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf2 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf4 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf7 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
buf6 = empty_strided_cuda((4, 4, 3, 3), (36, 9, 3, 1), torch.float32)
triton_poi_fused_add_mul_sigmoid_sigmoid_backward_tanh_1[grid(144)](
buf1, primals_4, primals_5, buf3, buf2, buf4, buf5, buf7, buf6,
144, XBLOCK=128, num_warps=4, num_stages=1)
del buf1
del primals_4
return buf6, buf5, primals_3, primals_5, buf0, buf2, buf3, buf4, buf5, buf7
class ConvLSTMCellNew(nn.Module):
def __init__(self, input_channels, hidden_channels, kernel_size, bias=True
):
super(ConvLSTMCellNew, self).__init__()
assert hidden_channels % 2 == 0
self.input_channels = input_channels
self.hidden_channels = hidden_channels
self.bias = bias
self.kernel_size = kernel_size
self.padding = int((kernel_size - 1) / 2)
self.Gates = nn.Conv2d(self.input_channels + self.hidden_channels,
4 * self.hidden_channels, self.kernel_size, 1, self.padding,
bias=True)
def init_hidden(self, batch_size, hidden, shape):
return Variable(torch.zeros(batch_size, hidden, shape[0], shape[1])
), Variable(torch.zeros(batch_size, hidden, shape[0], shape[1]))
def forward(self, input_0, input_1, input_2):
primals_3 = self.Gates.weight
primals_4 = self.Gates.bias
primals_1 = input_0
primals_2 = input_1
primals_5 = input_2
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0], output[1]
|
Kwanss/PCLNet
|
ConvLSTMCell
| false
| 8,402
|
[
"MIT"
] | 31
|
d288820975a9daf23eab47c52d7ea6f7dd564725
|
https://github.com/Kwanss/PCLNet/tree/d288820975a9daf23eab47c52d7ea6f7dd564725
|
CAMBlock
|
# 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: [x1], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# x1 => 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/wp/cwpgni2uqhs6rcp5zpreg2yqforqyf2ps3peum3jvr23zklxctyv.py
# Topologically Sorted Source Nodes: [x2], Original ATen: [aten.adaptive_max_pool2d]
# Source node to ATen node mapping:
# x2 => getitem
# Graph fragment:
# %getitem : [num_users=2] = call_function[target=operator.getitem](args = (%adaptive_max_pool2d, 0), kwargs = {})
triton_poi_fused_adaptive_max_pool2d_1 = async_compile.triton('triton_poi_fused_adaptive_max_pool2d_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_1', '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_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 + (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 = 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), tmp30, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/pz/cpzp3m5kdukcylnqwqv7lsxhkeslbvebgcrawt3ydpvq3yefrvim.py
# Topologically Sorted Source Nodes: [x1_1, x1_2], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x1_1 => convolution
# x1_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_2 = async_compile.triton('triton_poi_fused_convolution_relu_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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_relu_2(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (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/ur/curviu3ieiktk2v7as4mymeroi5pvhuiv7vx7xjeyynu2hi7jyfk.py
# Topologically Sorted Source Nodes: [x1_3, x2_3, out, out_1, out_2], Original ATen: [aten.convolution, aten.add, aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# out => add
# out_1 => sigmoid
# out_2 => mul
# x1_3 => convolution_1
# x2_3 => convolution_3
# 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 = {})
# %convolution_3 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu_1, %primals_8, %primals_9, [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, %convolution_3), kwargs = {})
# %sigmoid : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %sigmoid), kwargs = {})
triton_poi_fused_add_convolution_mul_sigmoid_3 = async_compile.triton('triton_poi_fused_add_convolution_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: '*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_convolution_mul_sigmoid_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_add_convolution_mul_sigmoid_3(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
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')
tmp4 = tl.load(in_ptr3 + (x4), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr4 + (x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp6 = tmp4 + tmp5
tmp7 = tmp3 + tmp6
tmp8 = tl.sigmoid(tmp7)
tmp9 = tmp0 * tmp8
tl.store(out_ptr0 + (x3), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/do/cdobl5aazzc5wuk25qh2i4dyoyoe4firam2n24raktdzzjqijwrp.py
# Topologically Sorted Source Nodes: [x1_3, x2_3, out, out_1], Original ATen: [aten.convolution, aten.add, aten.sigmoid, aten.sigmoid_backward]
# Source node to ATen node mapping:
# out => add
# out_1 => sigmoid
# x1_3 => convolution_1
# x2_3 => convolution_3
# 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 = {})
# %convolution_3 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu_1, %primals_8, %primals_9, [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, %convolution_3), kwargs = {})
# %sigmoid : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%add,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %sub), kwargs = {})
triton_poi_fused_add_convolution_sigmoid_sigmoid_backward_4 = async_compile.triton('triton_poi_fused_add_convolution_sigmoid_sigmoid_backward_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_convolution_sigmoid_sigmoid_backward_4', '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_sigmoid_sigmoid_backward_4(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_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 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp7 = tl.sigmoid(tmp6)
tmp8 = 1.0
tmp9 = tmp8 - tmp7
tmp10 = tmp7 * tmp9
tl.store(in_out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (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, ))
assert_size_stride(primals_6, (1, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_7, (1, ), (1, ))
assert_size_stride(primals_8, (4, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_9, (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: [x1], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_per_fused_mean_0.run(buf1, primals_1, 16, 16, grid=grid(16), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x2], Original ATen: [aten.adaptive_max_pool2d]
triton_poi_fused_adaptive_max_pool2d_1.run(primals_1, buf2, 16, grid=grid(16), stream=stream0)
# Topologically Sorted Source Nodes: [x1_1], Original ATen: [aten.convolution]
buf3 = 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(buf3, (4, 1, 1, 1), (1, 1, 1, 1))
buf4 = buf3; del buf3 # reuse
# Topologically Sorted Source Nodes: [x1_1, x1_2], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf4, primals_3, 4, grid=grid(4), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [x1_3], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(buf4, primals_4, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 4, 1, 1), (4, 1, 1, 1))
# Topologically Sorted Source Nodes: [x2_1], Original ATen: [aten.convolution]
buf6 = extern_kernels.convolution(buf2, primals_6, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf6, (4, 1, 1, 1), (1, 1, 1, 1))
buf7 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [x2_1, x2_2], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_2.run(buf7, primals_7, 4, grid=grid(4), stream=stream0)
del primals_7
# Topologically Sorted Source Nodes: [x2_3], Original ATen: [aten.convolution]
buf8 = extern_kernels.convolution(buf7, primals_8, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 4, 1, 1), (4, 1, 1, 1))
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x1_3, x2_3, out, out_1, out_2], Original ATen: [aten.convolution, aten.add, aten.sigmoid, aten.mul]
triton_poi_fused_add_convolution_mul_sigmoid_3.run(primals_1, buf5, primals_5, buf8, primals_9, buf9, 256, grid=grid(256), stream=stream0)
buf10 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [x1_3, x2_3, out, out_1], Original ATen: [aten.convolution, aten.add, aten.sigmoid, aten.sigmoid_backward]
triton_poi_fused_add_convolution_sigmoid_sigmoid_backward_4.run(buf10, primals_5, buf8, primals_9, 16, grid=grid(16), stream=stream0)
del buf8
del primals_5
del primals_9
return (buf9, primals_1, primals_2, primals_4, primals_6, primals_8, buf1, buf2, buf4, buf7, buf10, )
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)
primals_6 = rand_strided((1, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
class CAMBlock(torch.nn.Module):
def __init__(self, inplanes, redr, pool='full'):
super(CAMBlock, self).__init__()
self.planes = inplanes // redr
self.poolingavg = torch.nn.AdaptiveAvgPool2d((1, 1))
self.poolingmax = torch.nn.AdaptiveMaxPool2d((1, 1))
self.avglinear1 = torch.nn.Conv2d(inplanes, self.planes, (1, 1),
padding=0)
self.maxlinear1 = torch.nn.Conv2d(inplanes, self.planes, (1, 1),
padding=0)
self.relu = torch.nn.ReLU(inplace=True)
self.avglinear2 = torch.nn.Conv2d(self.planes, inplanes, (1, 1),
padding=0)
self.maxlinear2 = torch.nn.Conv2d(self.planes, inplanes, (1, 1),
padding=0)
self.sigmod = torch.nn.Sigmoid()
def forward(self, x):
x1 = self.poolingavg(x)
x2 = self.poolingmax(x)
x1 = self.avglinear1(x1)
x1 = self.relu(x1)
x1 = self.avglinear2(x1)
x2 = self.maxlinear1(x2)
x2 = self.relu(x2)
x2 = self.maxlinear2(x2)
out = x1 + x2
out = self.sigmod(out)
out = x * out
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'inplanes': 4, 'redr': 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
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_adaptive_max_pool2d_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 + 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 = 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, tmp30, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_2(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + 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_add_convolution_mul_sigmoid_3(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
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')
tmp4 = tl.load(in_ptr3 + x4, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr4 + x1, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp6 = tmp4 + tmp5
tmp7 = tmp3 + tmp6
tmp8 = tl.sigmoid(tmp7)
tmp9 = tmp0 * tmp8
tl.store(out_ptr0 + x3, tmp9, xmask)
@triton.jit
def triton_poi_fused_add_convolution_sigmoid_sigmoid_backward_4(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_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 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp7 = tl.sigmoid(tmp6)
tmp8 = 1.0
tmp9 = tmp8 - tmp7
tmp10 = tmp7 * tmp9
tl.store(in_out_ptr0 + x2, tmp10, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (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,))
assert_size_stride(primals_6, (1, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_7, (1,), (1,))
assert_size_stride(primals_8, (4, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_9, (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 = empty_strided_cuda((4, 4, 1, 1), (4, 1, 1, 1), torch.float32)
triton_poi_fused_adaptive_max_pool2d_1[grid(16)](primals_1, buf2,
16, XBLOCK=16, num_warps=1, num_stages=1)
buf3 = 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(buf3, (4, 1, 1, 1), (1, 1, 1, 1))
buf4 = buf3
del buf3
triton_poi_fused_convolution_relu_2[grid(4)](buf4, primals_3, 4,
XBLOCK=4, num_warps=1, num_stages=1)
del primals_3
buf5 = extern_kernels.convolution(buf4, primals_4, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 4, 1, 1), (4, 1, 1, 1))
buf6 = extern_kernels.convolution(buf2, primals_6, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf6, (4, 1, 1, 1), (1, 1, 1, 1))
buf7 = buf6
del buf6
triton_poi_fused_convolution_relu_2[grid(4)](buf7, primals_7, 4,
XBLOCK=4, num_warps=1, num_stages=1)
del primals_7
buf8 = extern_kernels.convolution(buf7, primals_8, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 4, 1, 1), (4, 1, 1, 1))
buf9 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_add_convolution_mul_sigmoid_3[grid(256)](primals_1,
buf5, primals_5, buf8, primals_9, buf9, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf10 = buf5
del buf5
triton_poi_fused_add_convolution_sigmoid_sigmoid_backward_4[grid(16)](
buf10, primals_5, buf8, primals_9, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del buf8
del primals_5
del primals_9
return (buf9, primals_1, primals_2, primals_4, primals_6, primals_8,
buf1, buf2, buf4, buf7, buf10)
class CAMBlockNew(torch.nn.Module):
def __init__(self, inplanes, redr, pool='full'):
super(CAMBlockNew, self).__init__()
self.planes = inplanes // redr
self.poolingavg = torch.nn.AdaptiveAvgPool2d((1, 1))
self.poolingmax = torch.nn.AdaptiveMaxPool2d((1, 1))
self.avglinear1 = torch.nn.Conv2d(inplanes, self.planes, (1, 1),
padding=0)
self.maxlinear1 = torch.nn.Conv2d(inplanes, self.planes, (1, 1),
padding=0)
self.relu = torch.nn.ReLU(inplace=True)
self.avglinear2 = torch.nn.Conv2d(self.planes, inplanes, (1, 1),
padding=0)
self.maxlinear2 = torch.nn.Conv2d(self.planes, inplanes, (1, 1),
padding=0)
self.sigmod = torch.nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.avglinear1.weight
primals_3 = self.avglinear1.bias
primals_6 = self.maxlinear1.weight
primals_7 = self.maxlinear1.bias
primals_4 = self.avglinear2.weight
primals_5 = self.avglinear2.bias
primals_8 = self.maxlinear2.weight
primals_9 = self.maxlinear2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9])
return output[0]
|
Knight825/models-pytorch
|
CAMBlock
| false
| 8,403
|
[
"Apache-2.0"
] | 16
|
133559eebb8795d78a32fa44d49408d0c5167ae9
|
https://github.com/Knight825/models-pytorch/tree/133559eebb8795d78a32fa44d49408d0c5167ae9
|
Gram
|
# 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/ue/cuemr6v2riy4j4ikxnha664gorlvrjakljol6wjnoa7bx372cgcp.py
# Topologically Sorted Source Nodes: [gram_1], Original ATen: [aten.div]
# Source node to ATen node mapping:
# gram_1 => div
# Graph fragment:
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mm, 256), 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: 'i32'}, 'device': 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_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_div_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 = 0.00390625
tmp2 = 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, = 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((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [gram], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(arg0_1, (16, 16), (16, 1), 0), reinterpret_tensor(arg0_1, (16, 16), (1, 16), 0), out=buf0)
del arg0_1
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [gram_1], Original ATen: [aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_div_0.run(buf1, 256, grid=grid(256), stream=stream0)
return (buf1, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class Gram(nn.Module):
def __init__(self):
super(Gram, self).__init__()
def forward(self, input):
a, b, c, d = input.size()
feature = input.view(a * b, c * d)
gram = torch.mm(feature, feature.t())
gram /= a * b * c * d
return gram
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
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_div_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 = 0.00390625
tmp2 = tmp0 * tmp1
tl.store(in_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((16, 16), (16, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(arg0_1, (16, 16), (16, 1), 0),
reinterpret_tensor(arg0_1, (16, 16), (1, 16), 0), out=buf0)
del arg0_1
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_div_0[grid(256)](buf1, 256, XBLOCK=128, num_warps=
4, num_stages=1)
return buf1,
class GramNew(nn.Module):
def __init__(self):
super(GramNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
L1aoXingyu/neural-transfer
|
Gram
| false
| 8,404
|
[
"MIT"
] | 45
|
bed445791d823872d9a40ea8927681d8cc99e8df
|
https://github.com/L1aoXingyu/neural-transfer/tree/bed445791d823872d9a40ea8927681d8cc99e8df
|
BiLSTM_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/dh/cdhwwkkwwalf3fekd34n4w7urr47fv5kpqakc24ioadnboe5irlo.py
# Topologically Sorted Source Nodes: [hf], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# hf => clone_2
# Graph fragment:
# %clone_2 : [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, 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 = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + (4*y3)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/iv/civlcl6dr4hxv3h37b66u75jadydavin2nqftyffnz35kihwpnuf.py
# Topologically Sorted Source Nodes: [gates], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates => sigmoid
# Graph fragment:
# %sigmoid : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_4,), 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: '*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_sigmoid_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_sigmoid_1(in_ptr0, in_ptr1, in_ptr2, 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
x3 = (xindex // 4)
x2 = (xindex // 16)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x3)), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x4), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/p2/cp2d4sc2imujujs5jxu7vliwd4n4hrdk3qxz53uhi5aefsvztafv.py
# Topologically Sorted Source Nodes: [cell_, mul, mul_1, cell_f, tanh_1, hidden_f_, eq, hidden_f_2], Original ATen: [aten.tanh, aten.mul, aten.add, aten.eq, aten.where]
# Source node to ATen node mapping:
# cell_ => tanh
# cell_f => add_2
# eq => eq
# hidden_f_ => mul_2
# hidden_f_2 => where
# mul => mul
# mul_1 => mul_1
# tanh_1 => tanh_1
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_4,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_1, %primals_5), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_2, %tanh), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
# %tanh_1 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_2,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_3, %tanh_1), kwargs = {})
# %eq : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%select_5, 0.0), kwargs = {})
# %where : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq, %view_9, %mul_2), kwargs = {})
triton_poi_fused_add_eq_mul_tanh_where_2 = async_compile.triton('triton_poi_fused_add_eq_mul_tanh_where_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, 8, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_eq_mul_tanh_where_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_add_eq_mul_tanh_where_2(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (192 + x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (12 + x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (12 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (4*x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr5 + (32 + x2), xmask)
tmp11 = tl.load(in_ptr5 + (x2), xmask)
tmp12 = tl.load(in_ptr6 + (x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr5 + (16 + x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp7 = 0.0
tmp8 = tmp6 == tmp7
tmp13 = tmp11 * tmp12
tmp15 = tmp14 * tmp5
tmp16 = tmp13 + tmp15
tmp17 = libdevice.tanh(tmp16)
tmp18 = tmp10 * tmp17
tmp19 = tl.where(tmp8, tmp9, tmp18)
tl.store(out_ptr0 + (x2), tmp5, xmask)
tl.store(out_ptr1 + (x2), tmp19, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/qc/cqck52v65u7tj7hlredfj4weo7bnwbaw66siuewilgokmjri5aug.py
# Topologically Sorted Source Nodes: [hb], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# hb => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_3,), 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=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = (yindex // 4)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (64 + 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/qq/cqqwxk6z5fq3pukv6dxohb3t5bbezcl7yryq4nktksqhgmacu43q.py
# Topologically Sorted Source Nodes: [gates_1], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_1 => sigmoid_1
# Graph fragment:
# %sigmoid_1 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_9,), kwargs = {})
triton_poi_fused_sigmoid_4 = async_compile.triton('triton_poi_fused_sigmoid_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_sigmoid_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_sigmoid_4(in_ptr0, in_ptr1, in_ptr2, 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
x3 = (xindex // 4)
x2 = (xindex // 16)
x4 = xindex
tmp0 = tl.load(in_ptr0 + (268 + x0 + (16*x3)), xmask)
tmp1 = tl.load(in_ptr1 + (16 + x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x4), tmp5, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/oc/coc4ueo6l4uzrrtj3acri3icmolona4fwciopr6bx23ya3cvfgzr.py
# Topologically Sorted Source Nodes: [cell__1, mul_3, mul_4, cell_b, tanh_3, hidden_b_, eq_1, hidden_b_2], Original ATen: [aten.tanh, aten.mul, aten.add, aten.eq, aten.where]
# Source node to ATen node mapping:
# cell__1 => tanh_2
# cell_b => add_4
# eq_1 => eq_1
# hidden_b_ => mul_5
# hidden_b_2 => where_1
# mul_3 => mul_3
# mul_4 => mul_4
# tanh_3 => tanh_3
# Graph fragment:
# %tanh_2 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_10,), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_7, %primals_6), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_8, %tanh_2), kwargs = {})
# %add_4 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_3, %mul_4), kwargs = {})
# %tanh_3 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_4,), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_9, %tanh_3), kwargs = {})
# %eq_1 : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%select_11, 0.0), kwargs = {})
# %where_1 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_1, %view_14, %mul_5), kwargs = {})
triton_poi_fused_add_eq_mul_tanh_where_5 = async_compile.triton('triton_poi_fused_add_eq_mul_tanh_where_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: 'i32'}, 'device': 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, 8, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_eq_mul_tanh_where_5', '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_add_eq_mul_tanh_where_5(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (460 + x0 + (16*x1)), xmask)
tmp1 = tl.load(in_ptr1 + (28 + x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (12 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr4 + (x0), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr5 + (32 + x2), xmask)
tmp11 = tl.load(in_ptr5 + (x2), xmask)
tmp12 = tl.load(in_ptr6 + (x0), xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr5 + (16 + x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp7 = 0.0
tmp8 = tmp6 == tmp7
tmp13 = tmp11 * tmp12
tmp15 = tmp14 * tmp5
tmp16 = tmp13 + tmp15
tmp17 = libdevice.tanh(tmp16)
tmp18 = tmp10 * tmp17
tmp19 = tl.where(tmp8, tmp9, tmp18)
tl.store(out_ptr0 + (x2), tmp5, xmask)
tl.store(out_ptr1 + (x2), tmp19, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/si/csibkryuhv6h3nawkdghbrfgrtcaccduopyaueehig7z2sdu55a3.py
# Topologically Sorted Source Nodes: [gates_2], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_2 => sigmoid_2
# Graph fragment:
# %sigmoid_2 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_14,), kwargs = {})
triton_poi_fused_sigmoid_6 = async_compile.triton('triton_poi_fused_sigmoid_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, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_sigmoid_6', '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_sigmoid_6(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (4 + x2 + (16*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gs/cgs3m7o6htt3nb65vukdrpof4dno7i6mchgolpbwwhye5r4hyzyo.py
# Topologically Sorted Source Nodes: [mul, mul_1, cell_f, cell__2, mul_6, mul_7, cell_f_1, tanh_5, hidden_f__1, eq_2, hidden_f_5], Original ATen: [aten.mul, aten.add, aten.tanh, aten.eq, aten.where]
# Source node to ATen node mapping:
# cell__2 => tanh_4
# cell_f => add_2
# cell_f_1 => add_6
# eq_2 => eq_2
# hidden_f_5 => where_2
# hidden_f__1 => mul_8
# mul => mul
# mul_1 => mul_1
# mul_6 => mul_6
# mul_7 => mul_7
# tanh_5 => tanh_5
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_1, %primals_5), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_2, %tanh), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul, %mul_1), kwargs = {})
# %tanh_4 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_16,), kwargs = {})
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_13, %add_2), kwargs = {})
# %mul_7 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_14, %tanh_4), kwargs = {})
# %add_6 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_6, %mul_7), kwargs = {})
# %tanh_5 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_6,), kwargs = {})
# %mul_8 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_15, %tanh_5), kwargs = {})
# %eq_2 : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%select_17, 0.0), kwargs = {})
# %where_2 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_2, %view_19, %mul_8), kwargs = {})
triton_poi_fused_add_eq_mul_tanh_where_7 = async_compile.triton('triton_poi_fused_add_eq_mul_tanh_where_7', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*fp32', 10: '*fp32', 11: '*fp32', 12: 'i32', 13: 'i32'}, 'device': 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, 8, 9, 10, 11), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_eq_mul_tanh_where_7', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 12, '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_eq_mul_tanh_where_7(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, out_ptr0, out_ptr1, out_ptr2, 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 + (196 + x1 + (16*y0)), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + (4*x1)), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + (4*y0)), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + (4*y0)), xmask & ymask)
tmp8 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (16 + x1 + (4*y0)), xmask & ymask)
tmp11 = tl.load(in_ptr6 + (x1 + (4*y0)), xmask & ymask)
tmp15 = tl.load(in_ptr3 + (16 + x1 + (4*y0)), xmask & ymask)
tmp18 = tl.load(in_ptr7 + (1 + (4*y0)), ymask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr8 + (x1 + (4*y0)), xmask & ymask)
tmp22 = tl.load(in_ptr3 + (32 + x1 + (4*y0)), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp9 = tmp7 * tmp8
tmp12 = tmp10 * tmp11
tmp13 = tmp9 + tmp12
tmp14 = tmp6 * tmp13
tmp16 = tmp15 * tmp5
tmp17 = tmp14 + tmp16
tmp19 = 0.0
tmp20 = tmp18 == tmp19
tmp23 = libdevice.tanh(tmp17)
tmp24 = tmp22 * tmp23
tmp25 = tl.where(tmp20, tmp21, tmp24)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + (4*y0)), tmp17, xmask & ymask)
tl.store(out_ptr2 + (x1 + (4*y0)), tmp25, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4v/c4vg75eipp5wv5so4ljmk5fxtnirbmxplvh3yujhdzpi5emniaul.py
# Topologically Sorted Source Nodes: [gates_3], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_3 => sigmoid_3
# Graph fragment:
# %sigmoid_3 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_19,), kwargs = {})
triton_poi_fused_sigmoid_8 = async_compile.triton('triton_poi_fused_sigmoid_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, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_sigmoid_8', '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_sigmoid_8(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (264 + x2 + (16*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (16 + x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/w6/cw645xrs46k5fwfgq6lzkugzdcizudqyqipp3q7mrjuzqjxmcbf4.py
# Topologically Sorted Source Nodes: [gates_4], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_4 => sigmoid_4
# Graph fragment:
# %sigmoid_4 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_24,), kwargs = {})
triton_poi_fused_sigmoid_9 = async_compile.triton('triton_poi_fused_sigmoid_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, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_sigmoid_9', '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_sigmoid_9(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (8 + x2 + (16*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/v7/cv7iyki3pgci6wtqww3aa4lcxry6yaa3dyo5fjerr2rvwnnb6ptp.py
# Topologically Sorted Source Nodes: [mul_3, mul_4, cell_b, cell__3, mul_9, mul_10, cell_b_1, tanh_7, hidden_b__1, eq_3, hidden_b_5, cell__4, mul_12, mul_13, cell_f_2, tanh_9, hidden_f__2, hidden_f_8], Original ATen: [aten.mul, aten.add, aten.tanh, aten.eq, aten.where]
# Source node to ATen node mapping:
# cell__3 => tanh_6
# cell__4 => tanh_8
# cell_b => add_4
# cell_b_1 => add_8
# cell_f_2 => add_10
# eq_3 => eq_3
# hidden_b_5 => where_3
# hidden_b__1 => mul_11
# hidden_f_8 => where_4
# hidden_f__2 => mul_14
# mul_10 => mul_10
# mul_12 => mul_12
# mul_13 => mul_13
# mul_3 => mul_3
# mul_4 => mul_4
# mul_9 => mul_9
# tanh_7 => tanh_7
# tanh_9 => tanh_9
# Graph fragment:
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_7, %primals_6), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_8, %tanh_2), kwargs = {})
# %add_4 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_3, %mul_4), kwargs = {})
# %tanh_6 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_22,), kwargs = {})
# %mul_9 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_19, %add_4), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_20, %tanh_6), kwargs = {})
# %add_8 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_9, %mul_10), kwargs = {})
# %tanh_7 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_8,), kwargs = {})
# %mul_11 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_21, %tanh_7), kwargs = {})
# %eq_3 : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%select_23, 0.0), kwargs = {})
# %where_3 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_3, %view_24, %mul_11), kwargs = {})
# %tanh_8 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_28,), kwargs = {})
# %mul_12 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_25, %add_6), kwargs = {})
# %mul_13 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_26, %tanh_8), kwargs = {})
# %add_10 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_12, %mul_13), kwargs = {})
# %tanh_9 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_10,), kwargs = {})
# %mul_14 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_27, %tanh_9), kwargs = {})
# %where_4 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_3, %view_29, %mul_14), kwargs = {})
triton_poi_fused_add_eq_mul_tanh_where_10 = async_compile.triton('triton_poi_fused_add_eq_mul_tanh_where_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, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*fp32', 10: '*fp32', 11: '*fp32', 12: '*fp32', 13: '*fp32', 14: '*fp32', 15: '*fp32', 16: '*fp32', 17: '*fp32', 18: '*fp32', 19: 'i32', 20: 'i32'}, 'device': 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, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_eq_mul_tanh_where_10', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 20, '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_eq_mul_tanh_where_10(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9, in_ptr10, in_ptr11, in_ptr12, out_ptr0, out_ptr1, out_ptr2, out_ptr3, out_ptr4, out_ptr5, 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 + (456 + x1 + (16*y0)), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (28 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + (4*x1)), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + (4*y0)), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + (4*y0)), xmask & ymask)
tmp8 = tl.load(in_ptr5 + (x1), xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (16 + x1 + (4*y0)), xmask & ymask)
tmp11 = tl.load(in_ptr6 + (x1 + (4*y0)), xmask & ymask)
tmp15 = tl.load(in_ptr3 + (16 + x1 + (4*y0)), xmask & ymask)
tmp18 = tl.load(in_ptr7 + (2 + (4*y0)), ymask, eviction_policy='evict_last')
tmp21 = tl.load(in_ptr8 + (x1 + (4*y0)), xmask & ymask)
tmp22 = tl.load(in_ptr3 + (32 + x1 + (4*y0)), xmask & ymask)
tmp26 = tl.load(in_ptr0 + (200 + x1 + (16*y0)), xmask & ymask)
tmp27 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr9 + (48 + y0 + (4*x1)), xmask & ymask)
tmp32 = tl.load(in_ptr10 + (x1 + (4*y0)), xmask & ymask)
tmp33 = tl.load(in_ptr11 + (x1 + (4*y0)), xmask & ymask)
tmp35 = tl.load(in_ptr10 + (16 + x1 + (4*y0)), xmask & ymask)
tmp38 = tl.load(in_ptr12 + (x1 + (4*y0)), xmask & ymask)
tmp39 = tl.load(in_ptr10 + (32 + x1 + (4*y0)), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp9 = tmp7 * tmp8
tmp12 = tmp10 * tmp11
tmp13 = tmp9 + tmp12
tmp14 = tmp6 * tmp13
tmp16 = tmp15 * tmp5
tmp17 = tmp14 + tmp16
tmp19 = 0.0
tmp20 = tmp18 == tmp19
tmp23 = libdevice.tanh(tmp17)
tmp24 = tmp22 * tmp23
tmp25 = tl.where(tmp20, tmp21, tmp24)
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp31 = libdevice.tanh(tmp30)
tmp34 = tmp32 * tmp33
tmp36 = tmp35 * tmp31
tmp37 = tmp34 + tmp36
tmp40 = libdevice.tanh(tmp37)
tmp41 = tmp39 * tmp40
tmp42 = tl.where(tmp20, tmp38, tmp41)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + (4*y0)), tmp17, xmask & ymask)
tl.store(out_ptr2 + (x1 + (4*y0)), tmp25, xmask & ymask)
tl.store(out_ptr3 + (x1 + (4*y0)), tmp31, xmask & ymask)
tl.store(out_ptr4 + (x1 + (4*y0)), tmp37, xmask & ymask)
tl.store(out_ptr5 + (x1 + (4*y0)), tmp42, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vc/cvcvser4w6g4xoepqoaxb6a4yedoqazwa5tz2xwcktcdvcs7yb3l.py
# Topologically Sorted Source Nodes: [gates_5], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_5 => sigmoid_5
# Graph fragment:
# %sigmoid_5 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_29,), kwargs = {})
triton_poi_fused_sigmoid_11 = async_compile.triton('triton_poi_fused_sigmoid_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, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_sigmoid_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_sigmoid_11(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (260 + x2 + (16*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (16 + x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/q4/cq45bovwn56aajgw53wsabclxvddvdnxs5t3axyhhknd3yw2mxt5.py
# Topologically Sorted Source Nodes: [eq_2, cell__5, mul_15, mul_16, cell_b_2, tanh_11, hidden_b__2, hidden_b_8], Original ATen: [aten.eq, aten.tanh, aten.mul, aten.add, aten.where]
# Source node to ATen node mapping:
# cell__5 => tanh_10
# cell_b_2 => add_12
# eq_2 => eq_2
# hidden_b_8 => where_5
# hidden_b__2 => mul_17
# mul_15 => mul_15
# mul_16 => mul_16
# tanh_11 => tanh_11
# Graph fragment:
# %eq_2 : [num_users=2] = call_function[target=torch.ops.aten.eq.Scalar](args = (%select_17, 0.0), kwargs = {})
# %tanh_10 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_34,), kwargs = {})
# %mul_15 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_31, %add_8), kwargs = {})
# %mul_16 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_32, %tanh_10), kwargs = {})
# %add_12 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_15, %mul_16), kwargs = {})
# %tanh_11 : [num_users=1] = call_function[target=torch.ops.aten.tanh.default](args = (%add_12,), kwargs = {})
# %mul_17 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_33, %tanh_11), kwargs = {})
# %where_5 : [num_users=2] = call_function[target=torch.ops.aten.where.self](args = (%eq_2, %view_34, %mul_17), kwargs = {})
triton_poi_fused_add_eq_mul_tanh_where_12 = async_compile.triton('triton_poi_fused_add_eq_mul_tanh_where_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=[4, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: '*fp32', 10: 'i32', 11: 'i32'}, 'device': 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, 8, 9), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_eq_mul_tanh_where_12', '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_add_eq_mul_tanh_where_12(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, out_ptr1, out_ptr2, 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 + (452 + x1 + (16*y0)), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (28 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + (4*x1)), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + (4*y0)), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + (4*y0)), xmask & ymask)
tmp9 = tl.load(in_ptr3 + (16 + x1 + (4*y0)), xmask & ymask)
tmp12 = tl.load(in_ptr5 + (1 + (4*y0)), ymask, eviction_policy='evict_last')
tmp15 = tl.load(in_ptr6 + (x1 + (4*y0)), xmask & ymask)
tmp16 = tl.load(in_ptr3 + (32 + x1 + (4*y0)), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 * tmp7
tmp10 = tmp9 * tmp5
tmp11 = tmp8 + tmp10
tmp13 = 0.0
tmp14 = tmp12 == tmp13
tmp17 = libdevice.tanh(tmp11)
tmp18 = tmp16 * tmp17
tmp19 = tl.where(tmp14, tmp15, tmp18)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + (4*y0)), tmp11, xmask & ymask)
tl.store(out_ptr2 + (x1 + (4*y0)), tmp19, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/7i/c7ivqwhwvzp6wkjhgfgr3g6sfvvrbjh5i4vz4clc43uafja4m27s.py
# Topologically Sorted Source Nodes: [gates_6], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_6 => sigmoid_6
# Graph fragment:
# %sigmoid_6 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_34,), kwargs = {})
triton_poi_fused_sigmoid_13 = async_compile.triton('triton_poi_fused_sigmoid_13', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_sigmoid_13', '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_sigmoid_13(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (12 + x2 + (16*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4c/c4c7wkd6wz2zp3fulrfggm36urin2l7i56ztaquupqwae2s45g6p.py
# Topologically Sorted Source Nodes: [cell__6, mul_18, mul_19, cell_f_3, tanh_13], Original ATen: [aten.tanh, aten.mul, aten.add]
# Source node to ATen node mapping:
# cell__6 => tanh_12
# cell_f_3 => add_14
# mul_18 => mul_18
# mul_19 => mul_19
# tanh_13 => tanh_13
# Graph fragment:
# %tanh_12 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_40,), kwargs = {})
# %mul_18 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_37, %add_10), kwargs = {})
# %mul_19 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_38, %tanh_12), kwargs = {})
# %add_14 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_18, %mul_19), kwargs = {})
# %tanh_13 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_14,), kwargs = {})
triton_poi_fused_add_mul_tanh_14 = async_compile.triton('triton_poi_fused_add_mul_tanh_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=[4, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32', 8: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_tanh_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_mul_tanh_14(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, out_ptr1, 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 + (204 + x1 + (16*y0)), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + (4*x1)), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + (4*y0)), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + (4*y0)), xmask & ymask)
tmp9 = tl.load(in_ptr3 + (16 + x1 + (4*y0)), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 * tmp7
tmp10 = tmp9 * tmp5
tmp11 = tmp8 + tmp10
tmp12 = libdevice.tanh(tmp11)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + (4*y0)), tmp12, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bq/cbqfvole2ysq4qdzvywtzzpljwuu2bzfbb2w26oskoiu7ue57mxx.py
# Topologically Sorted Source Nodes: [gates_7], Original ATen: [aten.sigmoid]
# Source node to ATen node mapping:
# gates_7 => sigmoid_7
# Graph fragment:
# %sigmoid_7 : [num_users=4] = call_function[target=torch.ops.aten.sigmoid.default](args = (%slice_39,), kwargs = {})
triton_poi_fused_sigmoid_15 = async_compile.triton('triton_poi_fused_sigmoid_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=[16, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_sigmoid_15', '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_sigmoid_15(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = (yindex // 4)
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (256 + x2 + (16*y3)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (16 + x2 + (4*y1)), xmask & ymask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + (4*x2) + (16*y1)), xmask & ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + (4*y3)), tmp5, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/7r/c7r4u7m3aiotucn5yrkylskm5pv5345z73edmfso7e6otlkqng5j.py
# Topologically Sorted Source Nodes: [cell__7, mul_21, mul_22, cell_b_3, tanh_15], Original ATen: [aten.tanh, aten.mul, aten.add]
# Source node to ATen node mapping:
# cell__7 => tanh_14
# cell_b_3 => add_16
# mul_21 => mul_21
# mul_22 => mul_22
# tanh_15 => tanh_15
# Graph fragment:
# %tanh_14 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%select_46,), kwargs = {})
# %mul_21 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_43, %add_12), kwargs = {})
# %mul_22 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%select_44, %tanh_14), kwargs = {})
# %add_16 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_21, %mul_22), kwargs = {})
# %tanh_15 : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%add_16,), kwargs = {})
triton_poi_fused_add_mul_tanh_16 = async_compile.triton('triton_poi_fused_add_mul_tanh_16', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: 'i32', 8: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_tanh_16', '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_tanh_16(in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr0, out_ptr1, 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 + (448 + x1 + (16*y0)), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (28 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + (4*x1)), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + (4*y0)), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + (4*y0)), xmask & ymask)
tmp9 = tl.load(in_ptr3 + (16 + x1 + (4*y0)), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 * tmp7
tmp10 = tmp9 * tmp5
tmp11 = tmp8 + tmp10
tmp12 = libdevice.tanh(tmp11)
tl.store(out_ptr0 + (x1 + (4*y0)), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + (4*y0)), tmp12, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/64/c64yikwy2latso2m5b7qux2wlhh6qn3r352nneytq24qjrfv5nac.py
# Topologically Sorted Source Nodes: [hidden_states_f], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# hidden_states_f => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%view_10, %view_20, %view_30, %view_40],), kwargs = {})
triton_poi_fused_cat_17 = async_compile.triton('triton_poi_fused_cat_17', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_cat_17', '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_cat_17(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 // 16)
x3 = xindex % 16
x1 = (xindex // 4) % 4
x4 = xindex
tmp0 = x2
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 + (x3), 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 + (x3), 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 + (x3), 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 + (3 + (4*x1)), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp20 = 0.0
tmp21 = tmp19 == tmp20
tmp22 = tl.load(in_ptr2 + (x3), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp23 = tl.load(in_ptr4 + (32 + x3), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp24 = tl.load(in_ptr5 + (x3), tmp16 & xmask, eviction_policy='evict_last', other=0.0)
tmp25 = tmp23 * tmp24
tmp26 = tl.where(tmp21, tmp22, tmp25)
tmp27 = tl.full(tmp26.shape, 0.0, tmp26.dtype)
tmp28 = tl.where(tmp16, tmp26, tmp27)
tmp29 = tl.where(tmp14, tmp15, tmp28)
tmp30 = tl.where(tmp9, tmp10, tmp29)
tmp31 = tl.where(tmp4, tmp5, tmp30)
tl.store(out_ptr0 + (x4), tmp31, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/p5/cp5qmaogify3dqrnqntwmg5pzec3q2hdfbe3ctxcixemfoejk5xv.py
# Topologically Sorted Source Nodes: [hidden_states_b], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# hidden_states_b => cat_1
# Graph fragment:
# %cat_1 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%view_45, %view_35, %view_25, %view_15],), kwargs = {})
triton_poi_fused_cat_18 = async_compile.triton('triton_poi_fused_cat_18', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_cat_18', '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_cat_18(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 // 16)
x1 = (xindex // 4) % 4
x3 = xindex % 16
x4 = xindex
tmp0 = x2
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 = 0.0
tmp7 = tmp5 == tmp6
tmp8 = tl.load(in_ptr1 + (x3), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp9 = tl.load(in_ptr2 + (32 + x3), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.load(in_ptr3 + (x3), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = tmp9 * tmp10
tmp12 = tl.where(tmp7, tmp8, tmp11)
tmp13 = tl.full(tmp12.shape, 0.0, tmp12.dtype)
tmp14 = tl.where(tmp4, tmp12, tmp13)
tmp15 = tmp0 >= tmp3
tmp16 = tl.full([1], 2, tl.int64)
tmp17 = tmp0 < tmp16
tmp18 = tmp15 & tmp17
tmp19 = tl.load(in_ptr1 + (x3), tmp18 & xmask, eviction_policy='evict_last', other=0.0)
tmp20 = tmp0 >= tmp16
tmp21 = tl.full([1], 3, tl.int64)
tmp22 = tmp0 < tmp21
tmp23 = tmp20 & tmp22
tmp24 = tl.load(in_ptr4 + (x3), tmp23 & xmask, eviction_policy='evict_last', other=0.0)
tmp25 = tmp0 >= tmp21
tmp26 = tl.full([1], 4, tl.int64)
tmp27 = tmp0 < tmp26
tmp28 = tl.load(in_ptr5 + (x3), tmp25 & xmask, eviction_policy='evict_last', other=0.0)
tmp29 = tl.where(tmp23, tmp24, tmp28)
tmp30 = tl.where(tmp18, tmp19, tmp29)
tmp31 = tl.where(tmp4, tmp14, tmp30)
tl.store(out_ptr0 + (x4), tmp31, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wz/cwzmjlei6ri7xikhyfkqfvjnlkqbkj6muv5ihes2pvkphx62eb4o.py
# Topologically Sorted Source Nodes: [cat_2, hidden_states], Original ATen: [aten.cat, aten.mul]
# Source node to ATen node mapping:
# cat_2 => cat_2
# hidden_states => mul_24
# Graph fragment:
# %cat_2 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%permute_24, %permute_25], -1), kwargs = {})
# %mul_24 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%cat_2, %primals_2), kwargs = {})
triton_poi_fused_cat_mul_19 = async_compile.triton('triton_poi_fused_cat_mul_19', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
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_cat_mul_19', '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_cat_mul_19(in_ptr0, in_ptr1, in_ptr2, 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) % 4
x2 = (xindex // 32)
x3 = (xindex // 8)
x4 = xindex
tmp11 = tl.load(in_ptr2 + (x3), xmask, eviction_policy='evict_last')
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*x2) + (16*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*x2) + (16*x1) + ((-4) + x0)), tmp6 & xmask, eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tmp12 = tmp10 * tmp11
tl.store(out_ptr0 + (x4), tmp12, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/lp/clpa4szqezldbpyk23alqc3bcorbjzy37bykf2funritwlrsnonx.py
# Topologically Sorted Source Nodes: [final_state], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# final_state => cat_3
# Graph fragment:
# %cat_3 : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%select_48, %select_49], -1), kwargs = {})
triton_poi_fused_cat_20 = async_compile.triton('triton_poi_fused_cat_20', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_cat_20', '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_20(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = (xindex // 8)
x2 = xindex
tmp0 = x0
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (48 + (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')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (1, 4), (4, 1))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1, 4), (4, 1))
assert_size_stride(primals_6, (1, 4), (4, 1))
assert_size_stride(primals_7, (8, 4, 4), (16, 4, 1))
assert_size_stride(primals_8, (8, 4, 4), (16, 4, 1))
assert_size_stride(primals_9, (8, 1, 4), (4, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((8, 16, 4), (64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(primals_1, (8, 16, 4), (0, 4, 1), 0), primals_7, out=buf0)
del primals_7
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hf], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(primals_8, buf1, 16, 4, grid=grid(16, 4), stream=stream0)
buf2 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [hf], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 1), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_1.run(buf0, primals_9, buf2, buf3, 48, grid=grid(48), stream=stream0)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cell_, mul, mul_1, cell_f, tanh_1, hidden_f_, eq, hidden_f_2], Original ATen: [aten.tanh, aten.mul, aten.add, aten.eq, aten.where]
triton_poi_fused_add_eq_mul_tanh_where_2.run(buf0, primals_9, buf2, primals_2, primals_3, buf3, primals_5, buf4, buf5, 16, grid=grid(16), stream=stream0)
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hb], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(primals_8, buf6, 16, 4, grid=grid(16, 4), stream=stream0)
del primals_8
buf7 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [hb], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 1), (1, 4), 0), out=buf7)
buf8 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_1], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_4.run(buf0, primals_9, buf7, buf8, 48, grid=grid(48), stream=stream0)
buf9 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf10 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cell__1, mul_3, mul_4, cell_b, tanh_3, hidden_b_, eq_1, hidden_b_2], Original ATen: [aten.tanh, aten.mul, aten.add, aten.eq, aten.where]
triton_poi_fused_add_eq_mul_tanh_where_5.run(buf0, primals_9, buf7, primals_2, primals_4, buf8, primals_6, buf9, buf10, 16, grid=grid(16), stream=stream0)
buf11 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hf_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(buf5, (4, 4), (1, 4), 0), out=buf11)
buf12 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_2], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_6.run(buf0, primals_9, buf11, buf12, 12, 4, grid=grid(12, 4), stream=stream0)
buf13 = reinterpret_tensor(buf7, (4, 4), (4, 1), 0); del buf7 # reuse
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf19 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, mul_1, cell_f, cell__2, mul_6, mul_7, cell_f_1, tanh_5, hidden_f__1, eq_2, hidden_f_5], Original ATen: [aten.mul, aten.add, aten.tanh, aten.eq, aten.where]
triton_poi_fused_add_eq_mul_tanh_where_7.run(buf0, primals_9, buf11, buf12, buf3, primals_5, buf4, primals_2, buf5, buf13, buf14, buf19, 4, 4, grid=grid(4, 4), stream=stream0)
buf15 = buf11; del buf11 # reuse
# Topologically Sorted Source Nodes: [hb_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0), reinterpret_tensor(buf10, (4, 4), (1, 4), 0), out=buf15)
buf16 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_3], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_8.run(buf0, primals_9, buf15, buf16, 12, 4, grid=grid(12, 4), stream=stream0)
buf20 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [hf_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(buf19, (4, 4), (1, 4), 0), out=buf20)
buf21 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_4], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_9.run(buf0, primals_9, buf20, buf21, 12, 4, grid=grid(12, 4), stream=stream0)
buf17 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf18 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf24 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf22 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf23 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf29 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul_3, mul_4, cell_b, cell__3, mul_9, mul_10, cell_b_1, tanh_7, hidden_b__1, eq_3, hidden_b_5, cell__4, mul_12, mul_13, cell_f_2, tanh_9, hidden_f__2, hidden_f_8], Original ATen: [aten.mul, aten.add, aten.tanh, aten.eq, aten.where]
triton_poi_fused_add_eq_mul_tanh_where_10.run(buf0, primals_9, buf15, buf16, buf8, primals_6, buf9, primals_2, buf10, buf20, buf21, buf14, buf19, buf17, buf18, buf24, buf22, buf23, buf29, 4, 4, grid=grid(4, 4), stream=stream0)
buf25 = buf20; del buf20 # reuse
# Topologically Sorted Source Nodes: [hb_2], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0), reinterpret_tensor(buf24, (4, 4), (1, 4), 0), out=buf25)
buf26 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_5], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_11.run(buf0, primals_9, buf25, buf26, 12, 4, grid=grid(12, 4), stream=stream0)
buf27 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf28 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [eq_2, cell__5, mul_15, mul_16, cell_b_2, tanh_11, hidden_b__2, hidden_b_8], Original ATen: [aten.eq, aten.tanh, aten.mul, aten.add, aten.where]
triton_poi_fused_add_eq_mul_tanh_where_12.run(buf0, primals_9, buf25, buf26, buf18, primals_2, buf24, buf27, buf28, buf34, 4, 4, grid=grid(4, 4), stream=stream0)
buf30 = buf25; del buf25 # reuse
# Topologically Sorted Source Nodes: [hf_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0), reinterpret_tensor(buf29, (4, 4), (1, 4), 0), out=buf30)
buf31 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_6], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_13.run(buf0, primals_9, buf30, buf31, 12, 4, grid=grid(12, 4), stream=stream0)
buf32 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf33 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cell__6, mul_18, mul_19, cell_f_3, tanh_13], Original ATen: [aten.tanh, aten.mul, aten.add]
triton_poi_fused_add_mul_tanh_14.run(buf0, primals_9, buf30, buf31, buf23, buf32, buf33, 4, 4, grid=grid(4, 4), stream=stream0)
buf35 = buf30; del buf30 # reuse
# Topologically Sorted Source Nodes: [hb_3], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0), reinterpret_tensor(buf34, (4, 4), (1, 4), 0), out=buf35)
buf36 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [gates_7], Original ATen: [aten.sigmoid]
triton_poi_fused_sigmoid_15.run(buf0, primals_9, buf35, buf36, 12, 4, grid=grid(12, 4), stream=stream0)
buf37 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf38 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [cell__7, mul_21, mul_22, cell_b_3, tanh_15], Original ATen: [aten.tanh, aten.mul, aten.add]
triton_poi_fused_add_mul_tanh_16.run(buf0, primals_9, buf35, buf36, buf28, buf37, buf38, 4, 4, grid=grid(4, 4), stream=stream0)
del buf0
del primals_9
buf39 = reinterpret_tensor(buf35, (4, 4, 4), (16, 4, 1), 0); del buf35 # reuse
# Topologically Sorted Source Nodes: [hidden_states_f], Original ATen: [aten.cat]
triton_poi_fused_cat_17.run(buf5, buf19, buf29, primals_2, buf31, buf33, buf39, 64, grid=grid(64), stream=stream0)
buf40 = reinterpret_tensor(buf15, (4, 4, 4), (16, 4, 1), 0); del buf15 # reuse
# Topologically Sorted Source Nodes: [hidden_states_b], Original ATen: [aten.cat]
triton_poi_fused_cat_18.run(primals_2, buf34, buf36, buf38, buf24, buf10, buf40, 64, grid=grid(64), stream=stream0)
buf41 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
# Topologically Sorted Source Nodes: [cat_2, hidden_states], Original ATen: [aten.cat, aten.mul]
triton_poi_fused_cat_mul_19.run(buf39, buf40, primals_2, buf41, 128, grid=grid(128), stream=stream0)
buf42 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [final_state], Original ATen: [aten.cat]
triton_poi_fused_cat_20.run(buf39, buf40, buf42, 32, grid=grid(32), stream=stream0)
del buf39
del buf40
return (buf41, buf42, primals_1, primals_2, primals_5, primals_6, reinterpret_tensor(buf1, (16, 4), (4, 1), 0), buf3, buf4, reinterpret_tensor(buf6, (16, 4), (4, 1), 0), buf8, buf9, buf12, buf13, buf14, buf16, buf17, buf18, buf21, buf22, buf23, buf26, buf27, buf28, buf31, buf32, buf33, buf36, buf37, buf38, buf34, buf29, buf24, buf19, buf10, buf5, primals_4, 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, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 1), (4, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((8, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((8, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((8, 1, 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])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch as T
import torch.nn as nn
class BiLSTM_Encoder(nn.Module):
def __init__(self, D: 'int', hidden_size: 'int', dropout: 'float'):
super(BiLSTM_Encoder, self).__init__()
self.D = D
self.hidden_size = hidden_size
self.initial_hidden_f = nn.Parameter(T.randn(1, hidden_size))
self.initial_hidden_b = nn.Parameter(T.randn(1, hidden_size))
self.initial_cell_f = nn.Parameter(T.randn(1, hidden_size))
self.initial_cell_b = nn.Parameter(T.randn(1, hidden_size))
self.dropout = nn.Dropout(dropout)
self.weight_ih = nn.Parameter(T.randn(8, D, hidden_size))
self.weight_hh = nn.Parameter(T.randn(8, hidden_size, hidden_size))
self.bias = nn.Parameter(T.zeros(8, 1, hidden_size))
self.init_weights()
def init_weights(self):
for name, param in self.named_parameters():
if 'bias' in name.lower():
nn.init.zeros_(param.data)
elif 'hidden_state' or 'cell' in name.lower():
nn.init.zeros_(param.data)
else:
nn.init.xavier_uniform_(param.data)
def forward(self, x, mask):
N, S, D = x.size()
mask = mask.view(N, S, 1)
hidden_f = self.initial_hidden_f
hidden_b = self.initial_hidden_b
cell_f = self.initial_cell_f
cell_b = self.initial_cell_b
hidden_states_f = []
hidden_states_b = []
self.dropout(self.weight_ih)
weight_hh = self.dropout(self.weight_hh)
x = x.view(1, N * S, D)
x_h = T.matmul(x, self.weight_ih) + self.bias
x_h = x_h.view(8, N, S, self.hidden_size)
for t in range(S):
hidden_f = hidden_f.view(1, -1, self.hidden_size)
xf = x_h[0:4, :, t]
hf = T.matmul(hidden_f, weight_hh[0:4])
preacts = xf + hf
gates = T.sigmoid(preacts[0:3])
f = gates[0]
i = gates[1]
o = gates[2]
cell_ = T.tanh(preacts[3])
cell_f = f * cell_f + i * cell_
hidden_f_ = o * T.tanh(cell_f)
hidden_f = hidden_f.view(-1, self.hidden_size)
hidden_f = T.where(mask[:, t] == 0.0, hidden_f, hidden_f_)
hidden_states_f.append(hidden_f.view(1, N, self.hidden_size))
hidden_b = hidden_b.view(1, -1, self.hidden_size)
xb = x_h[4:, :, S - t - 1]
hb = T.matmul(hidden_b, weight_hh[4:])
preacts = xb + hb
gates = T.sigmoid(preacts[0:3])
f = gates[0]
i = gates[1]
o = gates[2]
cell_ = T.tanh(preacts[3])
cell_b = f * cell_b + i * cell_
hidden_b_ = o * T.tanh(cell_b)
hidden_b = hidden_b.view(-1, self.hidden_size)
hidden_b = T.where(mask[:, S - t - 1] == 0.0, hidden_b, hidden_b_)
hidden_states_b.append(hidden_b.view(1, N, self.hidden_size))
hidden_states_f = T.cat(hidden_states_f, dim=0)
hidden_states_f = T.transpose(hidden_states_f, 0, 1)
hidden_states_b.reverse()
hidden_states_b = T.cat(hidden_states_b, dim=0)
hidden_states_b = T.transpose(hidden_states_b, 0, 1)
hidden_states = T.cat([hidden_states_f, hidden_states_b], dim=-1
) * mask
final_state = T.cat([hidden_states_f[:, -1, :], hidden_states_b[:,
0, :]], dim=-1)
return hidden_states, final_state
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 1])]
def get_init_inputs():
return [[], {'D': 4, 'hidden_size': 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.triton_helpers import libdevice
import torch as T
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, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tl.store(out_ptr0 + (x2 + 4 * y3), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_sigmoid_1(in_ptr0, in_ptr1, in_ptr2, 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
x3 = xindex // 4
x2 = xindex // 16
x4 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x3), xmask)
tmp1 = tl.load(in_ptr1 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp3 = tl.load(in_ptr2 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + x4, tmp5, xmask)
@triton.jit
def triton_poi_fused_add_eq_mul_tanh_where_2(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, out_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (192 + x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (12 + x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (12 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + 4 * x1, xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr5 + (32 + x2), xmask)
tmp11 = tl.load(in_ptr5 + x2, xmask)
tmp12 = tl.load(in_ptr6 + x0, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr5 + (16 + x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp7 = 0.0
tmp8 = tmp6 == tmp7
tmp13 = tmp11 * tmp12
tmp15 = tmp14 * tmp5
tmp16 = tmp13 + tmp15
tmp17 = libdevice.tanh(tmp16)
tmp18 = tmp10 * tmp17
tmp19 = tl.where(tmp8, tmp9, tmp18)
tl.store(out_ptr0 + x2, tmp5, xmask)
tl.store(out_ptr1 + x2, tmp19, xmask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 4
y1 = yindex // 4
y3 = yindex
tmp0 = tl.load(in_ptr0 + (64 + 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_sigmoid_4(in_ptr0, in_ptr1, in_ptr2, 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
x3 = xindex // 4
x2 = xindex // 16
x4 = xindex
tmp0 = tl.load(in_ptr0 + (268 + x0 + 16 * x3), xmask)
tmp1 = tl.load(in_ptr1 + (16 + x0 + 4 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr2 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + x4, tmp5, xmask)
@triton.jit
def triton_poi_fused_add_eq_mul_tanh_where_5(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, out_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (460 + x0 + 16 * x1), xmask)
tmp1 = tl.load(in_ptr1 + (28 + x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (12 + x0), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr3 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr4 + x0, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr5 + (32 + x2), xmask)
tmp11 = tl.load(in_ptr5 + x2, xmask)
tmp12 = tl.load(in_ptr6 + x0, xmask, eviction_policy='evict_last')
tmp14 = tl.load(in_ptr5 + (16 + x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp7 = 0.0
tmp8 = tmp6 == tmp7
tmp13 = tmp11 * tmp12
tmp15 = tmp14 * tmp5
tmp16 = tmp13 + tmp15
tmp17 = libdevice.tanh(tmp16)
tmp18 = tmp10 * tmp17
tmp19 = tl.where(tmp8, tmp9, tmp18)
tl.store(out_ptr0 + x2, tmp5, xmask)
tl.store(out_ptr1 + x2, tmp19, xmask)
@triton.jit
def triton_poi_fused_sigmoid_6(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (4 + x2 + 16 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + 4 * y1), xmask & ymask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_add_eq_mul_tanh_where_7(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, out_ptr0,
out_ptr1, out_ptr2, 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 + (196 + x1 + 16 * y0), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + 4 * x1), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + 4 * y0), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + 4 * y0), xmask & ymask)
tmp8 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (16 + x1 + 4 * y0), xmask & ymask)
tmp11 = tl.load(in_ptr6 + (x1 + 4 * y0), xmask & ymask)
tmp15 = tl.load(in_ptr3 + (16 + x1 + 4 * y0), xmask & ymask)
tmp18 = tl.load(in_ptr7 + (1 + 4 * y0), ymask, eviction_policy='evict_last'
)
tmp21 = tl.load(in_ptr8 + (x1 + 4 * y0), xmask & ymask)
tmp22 = tl.load(in_ptr3 + (32 + x1 + 4 * y0), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp9 = tmp7 * tmp8
tmp12 = tmp10 * tmp11
tmp13 = tmp9 + tmp12
tmp14 = tmp6 * tmp13
tmp16 = tmp15 * tmp5
tmp17 = tmp14 + tmp16
tmp19 = 0.0
tmp20 = tmp18 == tmp19
tmp23 = libdevice.tanh(tmp17)
tmp24 = tmp22 * tmp23
tmp25 = tl.where(tmp20, tmp21, tmp24)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + 4 * y0), tmp17, xmask & ymask)
tl.store(out_ptr2 + (x1 + 4 * y0), tmp25, xmask & ymask)
@triton.jit
def triton_poi_fused_sigmoid_8(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (264 + x2 + 16 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (16 + x2 + 4 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_sigmoid_9(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (8 + x2 + 16 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + 4 * y1), xmask & ymask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_add_eq_mul_tanh_where_10(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, in_ptr8, in_ptr9, in_ptr10,
in_ptr11, in_ptr12, out_ptr0, out_ptr1, out_ptr2, out_ptr3, out_ptr4,
out_ptr5, 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 + (456 + x1 + 16 * y0), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (28 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + 4 * x1), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + 4 * y0), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + 4 * y0), xmask & ymask)
tmp8 = tl.load(in_ptr5 + x1, xmask, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr4 + (16 + x1 + 4 * y0), xmask & ymask)
tmp11 = tl.load(in_ptr6 + (x1 + 4 * y0), xmask & ymask)
tmp15 = tl.load(in_ptr3 + (16 + x1 + 4 * y0), xmask & ymask)
tmp18 = tl.load(in_ptr7 + (2 + 4 * y0), ymask, eviction_policy='evict_last'
)
tmp21 = tl.load(in_ptr8 + (x1 + 4 * y0), xmask & ymask)
tmp22 = tl.load(in_ptr3 + (32 + x1 + 4 * y0), xmask & ymask)
tmp26 = tl.load(in_ptr0 + (200 + x1 + 16 * y0), xmask & ymask)
tmp27 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr9 + (48 + y0 + 4 * x1), xmask & ymask)
tmp32 = tl.load(in_ptr10 + (x1 + 4 * y0), xmask & ymask)
tmp33 = tl.load(in_ptr11 + (x1 + 4 * y0), xmask & ymask)
tmp35 = tl.load(in_ptr10 + (16 + x1 + 4 * y0), xmask & ymask)
tmp38 = tl.load(in_ptr12 + (x1 + 4 * y0), xmask & ymask)
tmp39 = tl.load(in_ptr10 + (32 + x1 + 4 * y0), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp9 = tmp7 * tmp8
tmp12 = tmp10 * tmp11
tmp13 = tmp9 + tmp12
tmp14 = tmp6 * tmp13
tmp16 = tmp15 * tmp5
tmp17 = tmp14 + tmp16
tmp19 = 0.0
tmp20 = tmp18 == tmp19
tmp23 = libdevice.tanh(tmp17)
tmp24 = tmp22 * tmp23
tmp25 = tl.where(tmp20, tmp21, tmp24)
tmp28 = tmp26 + tmp27
tmp30 = tmp28 + tmp29
tmp31 = libdevice.tanh(tmp30)
tmp34 = tmp32 * tmp33
tmp36 = tmp35 * tmp31
tmp37 = tmp34 + tmp36
tmp40 = libdevice.tanh(tmp37)
tmp41 = tmp39 * tmp40
tmp42 = tl.where(tmp20, tmp38, tmp41)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + 4 * y0), tmp17, xmask & ymask)
tl.store(out_ptr2 + (x1 + 4 * y0), tmp25, xmask & ymask)
tl.store(out_ptr3 + (x1 + 4 * y0), tmp31, xmask & ymask)
tl.store(out_ptr4 + (x1 + 4 * y0), tmp37, xmask & ymask)
tl.store(out_ptr5 + (x1 + 4 * y0), tmp42, xmask & ymask)
@triton.jit
def triton_poi_fused_sigmoid_11(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (260 + x2 + 16 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (16 + x2 + 4 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_add_eq_mul_tanh_where_12(in_ptr0, in_ptr1, in_ptr2,
in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, out_ptr1, out_ptr2,
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 + (452 + x1 + 16 * y0), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (28 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + 4 * x1), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + 4 * y0), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + 4 * y0), xmask & ymask)
tmp9 = tl.load(in_ptr3 + (16 + x1 + 4 * y0), xmask & ymask)
tmp12 = tl.load(in_ptr5 + (1 + 4 * y0), ymask, eviction_policy='evict_last'
)
tmp15 = tl.load(in_ptr6 + (x1 + 4 * y0), xmask & ymask)
tmp16 = tl.load(in_ptr3 + (32 + x1 + 4 * y0), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 * tmp7
tmp10 = tmp9 * tmp5
tmp11 = tmp8 + tmp10
tmp13 = 0.0
tmp14 = tmp12 == tmp13
tmp17 = libdevice.tanh(tmp11)
tmp18 = tmp16 * tmp17
tmp19 = tl.where(tmp14, tmp15, tmp18)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + 4 * y0), tmp11, xmask & ymask)
tl.store(out_ptr2 + (x1 + 4 * y0), tmp19, xmask & ymask)
@triton.jit
def triton_poi_fused_sigmoid_13(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (12 + x2 + 16 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + 4 * y1), xmask & ymask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_add_mul_tanh_14(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, out_ptr1, 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 + (204 + x1 + 16 * y0), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (12 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + 4 * x1), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + 4 * y0), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + 4 * y0), xmask & ymask)
tmp9 = tl.load(in_ptr3 + (16 + x1 + 4 * y0), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 * tmp7
tmp10 = tmp9 * tmp5
tmp11 = tmp8 + tmp10
tmp12 = libdevice.tanh(tmp11)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + 4 * y0), tmp12, xmask & ymask)
@triton.jit
def triton_poi_fused_sigmoid_15(in_ptr0, in_ptr1, in_ptr2, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 12
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y1 = yindex // 4
y0 = yindex % 4
tmp0 = tl.load(in_ptr0 + (256 + x2 + 16 * y3), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (16 + x2 + 4 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (y0 + 4 * x2 + 16 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = tl.sigmoid(tmp4)
tl.store(out_ptr0 + (x2 + 4 * y3), tmp5, xmask & ymask)
@triton.jit
def triton_poi_fused_add_mul_tanh_16(in_ptr0, in_ptr1, in_ptr2, in_ptr3,
in_ptr4, out_ptr0, out_ptr1, 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 + (448 + x1 + 16 * y0), xmask & ymask)
tmp1 = tl.load(in_ptr1 + (28 + x1), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr2 + (48 + y0 + 4 * x1), xmask & ymask)
tmp6 = tl.load(in_ptr3 + (x1 + 4 * y0), xmask & ymask)
tmp7 = tl.load(in_ptr4 + (x1 + 4 * y0), xmask & ymask)
tmp9 = tl.load(in_ptr3 + (16 + x1 + 4 * y0), xmask & ymask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp5 = libdevice.tanh(tmp4)
tmp8 = tmp6 * tmp7
tmp10 = tmp9 * tmp5
tmp11 = tmp8 + tmp10
tmp12 = libdevice.tanh(tmp11)
tl.store(out_ptr0 + (x1 + 4 * y0), tmp5, xmask & ymask)
tl.store(out_ptr1 + (x1 + 4 * y0), tmp12, xmask & ymask)
@triton.jit
def triton_poi_fused_cat_17(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 // 16
x3 = xindex % 16
x1 = xindex // 4 % 4
x4 = xindex
tmp0 = x2
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + x3, 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 + x3, 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 + x3, tmp14 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp16 = tmp0 >= tmp12
tl.full([1], 4, tl.int64)
tmp19 = tl.load(in_ptr3 + (3 + 4 * x1), tmp16 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp20 = 0.0
tmp21 = tmp19 == tmp20
tmp22 = tl.load(in_ptr2 + x3, tmp16 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp23 = tl.load(in_ptr4 + (32 + x3), tmp16 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp24 = tl.load(in_ptr5 + x3, tmp16 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp25 = tmp23 * tmp24
tmp26 = tl.where(tmp21, tmp22, tmp25)
tmp27 = tl.full(tmp26.shape, 0.0, tmp26.dtype)
tmp28 = tl.where(tmp16, tmp26, tmp27)
tmp29 = tl.where(tmp14, tmp15, tmp28)
tmp30 = tl.where(tmp9, tmp10, tmp29)
tmp31 = tl.where(tmp4, tmp5, tmp30)
tl.store(out_ptr0 + x4, tmp31, xmask)
@triton.jit
def triton_poi_fused_cat_18(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 // 16
x1 = xindex // 4 % 4
x3 = xindex % 16
x4 = xindex
tmp0 = x2
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 = 0.0
tmp7 = tmp5 == tmp6
tmp8 = tl.load(in_ptr1 + x3, tmp4 & xmask, eviction_policy='evict_last',
other=0.0)
tmp9 = tl.load(in_ptr2 + (32 + x3), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp10 = tl.load(in_ptr3 + x3, tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp11 = tmp9 * tmp10
tmp12 = tl.where(tmp7, tmp8, tmp11)
tmp13 = tl.full(tmp12.shape, 0.0, tmp12.dtype)
tmp14 = tl.where(tmp4, tmp12, tmp13)
tmp15 = tmp0 >= tmp3
tmp16 = tl.full([1], 2, tl.int64)
tmp17 = tmp0 < tmp16
tmp18 = tmp15 & tmp17
tmp19 = tl.load(in_ptr1 + x3, tmp18 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp20 = tmp0 >= tmp16
tmp21 = tl.full([1], 3, tl.int64)
tmp22 = tmp0 < tmp21
tmp23 = tmp20 & tmp22
tmp24 = tl.load(in_ptr4 + x3, tmp23 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp25 = tmp0 >= tmp21
tl.full([1], 4, tl.int64)
tmp28 = tl.load(in_ptr5 + x3, tmp25 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp29 = tl.where(tmp23, tmp24, tmp28)
tmp30 = tl.where(tmp18, tmp19, tmp29)
tmp31 = tl.where(tmp4, tmp14, tmp30)
tl.store(out_ptr0 + x4, tmp31, xmask)
@triton.jit
def triton_poi_fused_cat_mul_19(in_ptr0, in_ptr1, in_ptr2, 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 % 4
x2 = xindex // 32
x3 = xindex // 8
x4 = xindex
tmp11 = tl.load(in_ptr2 + x3, xmask, eviction_policy='evict_last')
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (4 * x2 + 16 * 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 * x2 + 16 * x1 + (-4 + x0)), tmp6 & xmask,
eviction_policy='evict_last', other=0.0)
tmp10 = tl.where(tmp4, tmp5, tmp9)
tmp12 = tmp10 * tmp11
tl.store(out_ptr0 + x4, tmp12, xmask)
@triton.jit
def triton_poi_fused_cat_20(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 8
x1 = xindex // 8
x2 = xindex
tmp0 = x0
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 4, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (48 + 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)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 1), (4, 1, 1))
assert_size_stride(primals_3, (1, 4), (4, 1))
assert_size_stride(primals_4, (1, 4), (4, 1))
assert_size_stride(primals_5, (1, 4), (4, 1))
assert_size_stride(primals_6, (1, 4), (4, 1))
assert_size_stride(primals_7, (8, 4, 4), (16, 4, 1))
assert_size_stride(primals_8, (8, 4, 4), (16, 4, 1))
assert_size_stride(primals_9, (8, 1, 4), (4, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((8, 16, 4), (64, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(primals_1, (8, 16, 4), (0, 4,
1), 0), primals_7, out=buf0)
del primals_7
buf1 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(16, 4)](primals_8, buf1, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf2 = empty_strided_cuda((16, 1), (1, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 1), (1, 4), 0), out=buf2)
buf3 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_1[grid(48)](buf0, primals_9, buf2, buf3,
48, XBLOCK=64, num_warps=1, num_stages=1)
buf4 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf5 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_eq_mul_tanh_where_2[grid(16)](buf0, primals_9,
buf2, primals_2, primals_3, buf3, primals_5, buf4, buf5, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clone_3[grid(16, 4)](primals_8, buf6, 16, 4,
XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
del primals_8
buf7 = buf2
del buf2
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 1), (1, 4), 0), out=buf7)
buf8 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_4[grid(48)](buf0, primals_9, buf7, buf8,
48, XBLOCK=64, num_warps=1, num_stages=1)
buf9 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf10 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_eq_mul_tanh_where_5[grid(16)](buf0, primals_9,
buf7, primals_2, primals_4, buf8, primals_6, buf9, buf10, 16,
XBLOCK=16, num_warps=1, num_stages=1)
buf11 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(buf5, (4, 4), (1, 4), 0), out=buf11)
buf12 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_6[grid(12, 4)](buf0, primals_9, buf11,
buf12, 12, 4, XBLOCK=4, YBLOCK=8, num_warps=1, num_stages=1)
buf13 = reinterpret_tensor(buf7, (4, 4), (4, 1), 0)
del buf7
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf19 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_eq_mul_tanh_where_7[grid(4, 4)](buf0,
primals_9, buf11, buf12, buf3, primals_5, buf4, primals_2, buf5,
buf13, buf14, buf19, 4, 4, XBLOCK=4, YBLOCK=4, num_warps=1,
num_stages=1)
buf15 = buf11
del buf11
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0),
reinterpret_tensor(buf10, (4, 4), (1, 4), 0), out=buf15)
buf16 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_8[grid(12, 4)](buf0, primals_9, buf15,
buf16, 12, 4, XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf20 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(buf19, (4, 4), (1, 4), 0), out=buf20)
buf21 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_9[grid(12, 4)](buf0, primals_9, buf20,
buf21, 12, 4, XBLOCK=4, YBLOCK=8, num_warps=1, num_stages=1)
buf17 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf18 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf24 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf22 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf23 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf29 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_eq_mul_tanh_where_10[grid(4, 4)](buf0,
primals_9, buf15, buf16, buf8, primals_6, buf9, primals_2,
buf10, buf20, buf21, buf14, buf19, buf17, buf18, buf24, buf22,
buf23, buf29, 4, 4, XBLOCK=4, YBLOCK=4, num_warps=1, num_stages=1)
buf25 = buf20
del buf20
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0),
reinterpret_tensor(buf24, (4, 4), (1, 4), 0), out=buf25)
buf26 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_11[grid(12, 4)](buf0, primals_9, buf25,
buf26, 12, 4, XBLOCK=4, YBLOCK=16, num_warps=1, num_stages=1)
buf27 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf28 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_eq_mul_tanh_where_12[grid(4, 4)](buf0,
primals_9, buf25, buf26, buf18, primals_2, buf24, buf27, buf28,
buf34, 4, 4, XBLOCK=4, YBLOCK=4, num_warps=1, num_stages=1)
buf30 = buf25
del buf25
extern_kernels.mm(reinterpret_tensor(buf1, (16, 4), (4, 1), 0),
reinterpret_tensor(buf29, (4, 4), (1, 4), 0), out=buf30)
buf31 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_13[grid(12, 4)](buf0, primals_9, buf30,
buf31, 12, 4, XBLOCK=4, YBLOCK=8, num_warps=1, num_stages=1)
buf32 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf33 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_mul_tanh_14[grid(4, 4)](buf0, primals_9, buf30,
buf31, buf23, buf32, buf33, 4, 4, XBLOCK=4, YBLOCK=4, num_warps
=1, num_stages=1)
buf35 = buf30
del buf30
extern_kernels.mm(reinterpret_tensor(buf6, (16, 4), (4, 1), 0),
reinterpret_tensor(buf34, (4, 4), (1, 4), 0), out=buf35)
buf36 = empty_strided_cuda((3, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sigmoid_15[grid(12, 4)](buf0, primals_9, buf35,
buf36, 12, 4, XBLOCK=4, YBLOCK=8, num_warps=1, num_stages=1)
buf37 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
buf38 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_mul_tanh_16[grid(4, 4)](buf0, primals_9, buf35,
buf36, buf28, buf37, buf38, 4, 4, XBLOCK=4, YBLOCK=4, num_warps
=1, num_stages=1)
del buf0
del primals_9
buf39 = reinterpret_tensor(buf35, (4, 4, 4), (16, 4, 1), 0)
del buf35
triton_poi_fused_cat_17[grid(64)](buf5, buf19, buf29, primals_2,
buf31, buf33, buf39, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf40 = reinterpret_tensor(buf15, (4, 4, 4), (16, 4, 1), 0)
del buf15
triton_poi_fused_cat_18[grid(64)](primals_2, buf34, buf36, buf38,
buf24, buf10, buf40, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf41 = empty_strided_cuda((4, 4, 8), (32, 8, 1), torch.float32)
triton_poi_fused_cat_mul_19[grid(128)](buf39, buf40, primals_2,
buf41, 128, XBLOCK=128, num_warps=4, num_stages=1)
buf42 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
triton_poi_fused_cat_20[grid(32)](buf39, buf40, buf42, 32, XBLOCK=
32, num_warps=1, num_stages=1)
del buf39
del buf40
return (buf41, buf42, primals_1, primals_2, primals_5, primals_6,
reinterpret_tensor(buf1, (16, 4), (4, 1), 0), buf3, buf4,
reinterpret_tensor(buf6, (16, 4), (4, 1), 0), buf8, buf9, buf12,
buf13, buf14, buf16, buf17, buf18, buf21, buf22, buf23, buf26,
buf27, buf28, buf31, buf32, buf33, buf36, buf37, buf38, buf34,
buf29, buf24, buf19, buf10, buf5, primals_4, primals_3)
class BiLSTM_EncoderNew(nn.Module):
def __init__(self, D: 'int', hidden_size: 'int', dropout: 'float'):
super(BiLSTM_EncoderNew, self).__init__()
self.D = D
self.hidden_size = hidden_size
self.initial_hidden_f = nn.Parameter(T.randn(1, hidden_size))
self.initial_hidden_b = nn.Parameter(T.randn(1, hidden_size))
self.initial_cell_f = nn.Parameter(T.randn(1, hidden_size))
self.initial_cell_b = nn.Parameter(T.randn(1, hidden_size))
self.dropout = nn.Dropout(dropout)
self.weight_ih = nn.Parameter(T.randn(8, D, hidden_size))
self.weight_hh = nn.Parameter(T.randn(8, hidden_size, hidden_size))
self.bias = nn.Parameter(T.zeros(8, 1, hidden_size))
self.init_weights()
def init_weights(self):
for name, param in self.named_parameters():
if 'bias' in name.lower():
nn.init.zeros_(param.data)
elif 'hidden_state' or 'cell' in name.lower():
nn.init.zeros_(param.data)
else:
nn.init.xavier_uniform_(param.data)
def forward(self, input_0, input_1):
primals_3 = self.initial_hidden_f
primals_4 = self.initial_hidden_b
primals_5 = self.initial_cell_f
primals_6 = self.initial_cell_b
primals_7 = self.weight_ih
primals_8 = self.weight_hh
primals_9 = self.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])
return output[0], output[1]
|
JRC1995/BERT-Disaster-Classification-Capsule-Routing
|
BiLSTM_Encoder
| false
| 8,405
|
[
"MIT"
] | 16
|
520d2b37af309c95f09bcda321915cffae803086
|
https://github.com/JRC1995/BERT-Disaster-Classification-Capsule-Routing/tree/520d2b37af309c95f09bcda321915cffae803086
|
MultiHeadQKVAttention
|
# 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/m2/cm2m4qsjyjz2a6kyhcrh5cr3nu35vr4fchqqwxvyt5do5bxsewau.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_3,), 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=[4, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + (16*y0)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/om/com5a2jdtzsd2gxozlfy64n7thpakjehaajnnx7weffss4gxj2sy.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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
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/5m/c5mma4y56ura3imiphserxkqyervoqe3bptp4i4swvp3yenvzn36.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/o5/co52ni3kaq2u2zi4ueyjtns22aw3kcfcubdifioq6mckemevvn2u.py
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_3 => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), 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=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (4, 4), (4, 1))
assert_size_stride(primals_11, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf0)
del primals_4
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf1)
del primals_6
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf2)
del primals_8
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, primals_5, buf3, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_5
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf1, primals_7, buf4, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_7
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf2, primals_9, buf8, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_9
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [o], Original ATen: [aten.bmm]
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf9, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [linear_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf11)
del primals_11
return (reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (16, 4), (4, 1), 0), buf7, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), primals_10, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import math
import torch
import numpy as np
import torch.nn.functional as F
import torch.nn as nn
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
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_k': 4, 'd_v': 4, 'n_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 math
import numpy as np
import torch.nn.functional as F
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, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x1), xmask & ymask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + 16 * y0), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x1), xmask & ymask, eviction_policy
='evict_last')
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (4, 4), (4, 1))
assert_size_stride(primals_11, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf0)
del primals_4
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf1)
del primals_6
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf2)
del primals_8
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(4, 16)](buf0, primals_5, buf3, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_5
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf0
triton_poi_fused_clone_0[grid(4, 16)](buf1, primals_7, buf4, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_7
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf7 = buf5
del buf5
triton_poi_fused__softmax_2[grid(256)](buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_clone_0[grid(4, 16)](buf2, primals_9, buf8, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_9
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 4, 1), (4, 1,
0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_3[grid(16, 4)](buf9, buf10, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.addmm(primals_11, reinterpret_tensor(buf10, (16, 4),
(4, 1), 0), reinterpret_tensor(primals_10, (4, 4), (1, 4), 0),
alpha=1, beta=1, out=buf11)
del primals_11
return reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0
), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_3, (16, 4), (4, 1), 0
), buf7, reinterpret_tensor(buf10, (16, 4), (4, 1), 0
), primals_10, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 1), 0)
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttentionNew(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, input_0, input_1, input_2):
primals_4 = self.q_projector.weight
primals_5 = self.q_projector.bias
primals_6 = self.k_projector.weight
primals_7 = self.k_projector.bias
primals_8 = self.v_projector.weight
primals_9 = self.v_projector.bias
primals_10 = self.o_projector.weight
primals_11 = self.o_projector.bias
primals_1 = input_0
primals_2 = input_1
primals_3 = input_2
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
|
KohavTal/SCAE_Project
|
MultiHeadQKVAttention
| false
| 8,406
|
[
"Apache-2.0"
] | 40
|
bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
https://github.com/KohavTal/SCAE_Project/tree/bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
MAB
|
# 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/m2/cm2m4qsjyjz2a6kyhcrh5cr3nu35vr4fchqqwxvyt5do5bxsewau.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_3,), 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=[4, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + (16*y0)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/om/com5a2jdtzsd2gxozlfy64n7thpakjehaajnnx7weffss4gxj2sy.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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
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/5m/c5mma4y56ura3imiphserxkqyervoqe3bptp4i4swvp3yenvzn36.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/o5/co52ni3kaq2u2zi4ueyjtns22aw3kcfcubdifioq6mckemevvn2u.py
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_3 => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), 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=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/l3/cl35un3kcsf7pvplhghgirnuph6p7ili57xhcg3be5ggxi7rvwhl.py
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
# Source node to ATen node mapping:
# h_1 => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_21, %primals_1), 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=[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_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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ip/cipnm4mskrxmuoiysdead6jtudm6fsnxgrtfs44ii4ovxepoidk3.py
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
# Source node to ATen node mapping:
# h_2 => add_1
# relu => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_23,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %relu), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_5 = async_compile.triton('triton_poi_fused_add_relu_threshold_backward_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 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_add_relu_threshold_backward_5', '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_relu_threshold_backward_5(in_ptr0, in_ptr1, in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + (x2), tmp6, xmask)
tl.store(out_ptr1 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, ), (1, ))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf0)
del primals_3
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf1)
del primals_5
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, primals_4, buf3, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_4
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf1, primals_6, buf4, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_6
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf2, primals_8, buf8, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [o], Original ATen: [aten.bmm]
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf9, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf11)
buf12 = reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0); del buf11 # reuse
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
triton_poi_fused_add_4.run(buf12, primals_10, primals_1, 64, grid=grid(64), stream=stream0)
del primals_10
buf13 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf12, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_5.run(buf12, buf13, primals_12, buf14, buf15, 64, grid=grid(64), stream=stream0)
del buf13
del primals_12
return (buf14, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0), buf7, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(buf12, (16, 4), (4, 1), 0), buf15, primals_11, primals_9, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((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)
primals_7 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import math
import torch
import numpy as np
import torch.nn.functional as F
import torch.nn as nn
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'d': 4, 'n_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 math
import numpy as np
import torch.nn.functional as F
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, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x1), xmask & ymask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + 16 * y0), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x1), xmask & ymask, eviction_policy
='evict_last')
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_4(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_5(in_ptr0, in_ptr1,
in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + x2, tmp6, xmask)
tl.store(out_ptr1 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4,), (1,))
assert_size_stride(primals_5, (4, 4), (4, 1))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf0)
del primals_3
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf1)
del primals_5
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_2, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf2)
del primals_7
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(4, 16)](buf0, primals_4, buf3, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_4
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf0
triton_poi_fused_clone_0[grid(4, 16)](buf1, primals_6, buf4, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_6
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf7 = buf5
del buf5
triton_poi_fused__softmax_2[grid(256)](buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_clone_0[grid(4, 16)](buf2, primals_8, buf8, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_8
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 4, 1), (4, 1,
0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_3[grid(16, 4)](buf9, buf10, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.mm(reinterpret_tensor(buf10, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf11)
buf12 = reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0)
del buf11
triton_poi_fused_add_4[grid(64)](buf12, primals_10, primals_1, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_10
buf13 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf12, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_5[grid(64)](buf12,
buf13, primals_12, buf14, buf15, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf13
del primals_12
return buf14, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_2, (16, 4), (4, 1), 0
), buf7, reinterpret_tensor(buf10, (16, 4), (4, 1), 0
), reinterpret_tensor(buf12, (16, 4), (4, 1), 0
), buf15, primals_11, primals_9, reinterpret_tensor(buf8, (16, 1, 4
), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 1), 0)
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MABNew(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, input_0, input_1):
primals_3 = self.mqkv.q_projector.weight
primals_4 = self.mqkv.q_projector.bias
primals_5 = self.mqkv.k_projector.weight
primals_6 = self.mqkv.k_projector.bias
primals_7 = self.mqkv.v_projector.weight
primals_8 = self.mqkv.v_projector.bias
primals_9 = self.mqkv.o_projector.weight
primals_10 = self.mqkv.o_projector.bias
primals_11 = self.fc.weight
primals_12 = self.fc.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, primals_11, primals_12])
return output[0]
|
KohavTal/SCAE_Project
|
MAB
| false
| 8,407
|
[
"Apache-2.0"
] | 40
|
bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
https://github.com/KohavTal/SCAE_Project/tree/bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
ConditionalLayerNorm
|
# 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/hy/chyz7kuep75o42kybftuykj5bewzkpchoywodbhchelaxc4urm7f.py
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# linear => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%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=[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_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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = (xindex // 16)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x2)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x3), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ys/cys6padbalp6boucferrgazhxn665ce6lbap46k2ptohqotp45y4.py
# Topologically Sorted Source Nodes: [mean, std, gamma, beta, sub, mul, add_2, truediv, x], Original ATen: [aten.mean, aten.std, aten.add, aten.sub, aten.mul, aten.div]
# Source node to ATen node mapping:
# add_2 => add_2
# beta => add_1
# gamma => add
# mean => mean
# mul => mul
# std => sqrt, var
# sub => sub
# truediv => div
# x => add_3
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%primals_1, [-1], True), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%primals_1, [-1]), kwargs = {correction: 1.0, keepdim: True})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %primals_4), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %primals_6), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%primals_1, %mean), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add, %sub), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt, 1e-06), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul, %add_2), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div, %add_1), kwargs = {})
triton_poi_fused_add_div_mean_mul_std_sub_1 = async_compile.triton('triton_poi_fused_add_div_mean_mul_std_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=[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_add_div_mean_mul_std_sub_1', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], '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_add_div_mean_mul_std_sub_1(in_out_ptr0, in_out_ptr1, 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
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 + (x2), xmask)
tmp4 = tl.load(in_ptr1 + (4*x1), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp32 = tl.load(in_out_ptr1 + (x2), xmask)
tmp33 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = 4.0
tmp12 = tmp10 / tmp11
tmp13 = tmp3 - tmp12
tmp14 = tmp2 * tmp13
tmp15 = tmp4 - tmp12
tmp16 = tmp15 * tmp15
tmp17 = tmp5 - tmp12
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp12
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = tmp9 - tmp12
tmp24 = tmp23 * tmp23
tmp25 = tmp22 + tmp24
tmp26 = 3.0
tmp27 = tmp25 / tmp26
tmp28 = libdevice.sqrt(tmp27)
tmp29 = 1e-06
tmp30 = tmp28 + tmp29
tmp31 = tmp14 / tmp30
tmp34 = tmp32 + tmp33
tmp35 = tmp31 + tmp34
tl.store(in_out_ptr1 + (x2), tmp35, 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, 4), (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((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(primals_2, buf0, 64, grid=grid(64), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.mm]
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf2)
del primals_5
buf3 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0); del buf1 # reuse
buf4 = reinterpret_tensor(buf2, (4, 4, 4), (16, 4, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [mean, std, gamma, beta, sub, mul, add_2, truediv, x], Original ATen: [aten.mean, aten.std, aten.add, aten.sub, aten.mul, aten.div]
triton_poi_fused_add_div_mean_mul_std_sub_1.run(buf3, buf4, primals_4, primals_1, primals_6, 64, grid=grid(64), stream=stream0)
del buf3
del primals_4
del primals_6
return (buf4, primals_1, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (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, 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 torch
from sklearn.metrics import *
from torch import nn
class ConditionalLayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-06):
super(ConditionalLayerNorm, self).__init__()
self.eps = eps
self.gamma_dense = nn.Linear(hidden_size, hidden_size, bias=False)
self.beta_dense = nn.Linear(hidden_size, hidden_size, bias=False)
self.gamma = nn.Parameter(torch.ones(hidden_size))
self.beta = nn.Parameter(torch.zeros(hidden_size))
nn.init.zeros_(self.gamma_dense.weight)
nn.init.zeros_(self.beta_dense.weight)
def forward(self, x, condition):
"""
:param x: [b, t, e]
:param condition: [b, e]
:return:
"""
mean = x.mean(-1, keepdim=True)
std = x.std(-1, keepdim=True)
condition = condition.unsqueeze(1).expand_as(x)
gamma = self.gamma_dense(condition) + self.gamma
beta = self.beta_dense(condition) + self.beta
x = gamma * (x - mean) / (std + self.eps) + beta
return x
def get_inputs():
return [torch.rand([4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'hidden_size': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from sklearn.metrics import *
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_clone_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x2 = xindex // 16
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x2), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + x3, tmp0, xmask)
@triton.jit
def triton_poi_fused_add_div_mean_mul_std_sub_1(in_out_ptr0, in_out_ptr1,
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
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 + x2, xmask)
tmp4 = tl.load(in_ptr1 + 4 * x1, xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr1 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr1 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp32 = tl.load(in_out_ptr1 + x2, xmask)
tmp33 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp6 = tmp4 + tmp5
tmp8 = tmp6 + tmp7
tmp10 = tmp8 + tmp9
tmp11 = 4.0
tmp12 = tmp10 / tmp11
tmp13 = tmp3 - tmp12
tmp14 = tmp2 * tmp13
tmp15 = tmp4 - tmp12
tmp16 = tmp15 * tmp15
tmp17 = tmp5 - tmp12
tmp18 = tmp17 * tmp17
tmp19 = tmp16 + tmp18
tmp20 = tmp7 - tmp12
tmp21 = tmp20 * tmp20
tmp22 = tmp19 + tmp21
tmp23 = tmp9 - tmp12
tmp24 = tmp23 * tmp23
tmp25 = tmp22 + tmp24
tmp26 = 3.0
tmp27 = tmp25 / tmp26
tmp28 = libdevice.sqrt(tmp27)
tmp29 = 1e-06
tmp30 = tmp28 + tmp29
tmp31 = tmp14 / tmp30
tmp34 = tmp32 + tmp33
tmp35 = tmp31 + tmp34
tl.store(in_out_ptr1 + x2, tmp35, 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, 4), (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((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(64)](primals_2, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
del primals_3
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf2)
del primals_5
buf3 = reinterpret_tensor(buf1, (4, 4, 4), (16, 4, 1), 0)
del buf1
buf4 = reinterpret_tensor(buf2, (4, 4, 4), (16, 4, 1), 0)
del buf2
triton_poi_fused_add_div_mean_mul_std_sub_1[grid(64)](buf3, buf4,
primals_4, primals_1, primals_6, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf3
del primals_4
del primals_6
return buf4, primals_1, reinterpret_tensor(buf0, (16, 4), (4, 1), 0)
class ConditionalLayerNormNew(nn.Module):
def __init__(self, hidden_size, eps=1e-06):
super(ConditionalLayerNormNew, self).__init__()
self.eps = eps
self.gamma_dense = nn.Linear(hidden_size, hidden_size, bias=False)
self.beta_dense = nn.Linear(hidden_size, hidden_size, bias=False)
self.gamma = nn.Parameter(torch.ones(hidden_size))
self.beta = nn.Parameter(torch.zeros(hidden_size))
nn.init.zeros_(self.gamma_dense.weight)
nn.init.zeros_(self.beta_dense.weight)
def forward(self, input_0, input_1):
primals_4 = self.gamma
primals_6 = self.beta
primals_2 = self.gamma_dense.weight
primals_3 = self.beta_dense.weight
primals_1 = input_0
primals_5 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6])
return output[0]
|
JiaweiSheng/CasEE
|
ConditionalLayerNorm
| false
| 8,408
|
[
"MIT"
] | 44
|
af69432baf34d150f4721a4b4119002555758601
|
https://github.com/JiaweiSheng/CasEE/tree/af69432baf34d150f4721a4b4119002555758601
|
VisTransformerDecoderLayer
|
# 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_3, 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_7,), 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/r3/cr3yj6dvepflkmsphii5jijgumcvcjwxzi7xsjhbn5x6msa6avwl.py
# Topologically Sorted Source Nodes: [tgt, tgt_1], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# tgt => add
# tgt_1 => var_mean
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %squeeze), kwargs = {})
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add, [1]), kwargs = {correction: 0, keepdim: True})
triton_poi_fused_add_native_layer_norm_4 = async_compile.triton('triton_poi_fused_add_native_layer_norm_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=[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_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_add_native_layer_norm_4(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/bk/cbk6ogr4fxxafxq737m3dmvcf777c7g6ws4gbqnhtzkyg4f7mlbe.py
# Topologically Sorted Source Nodes: [tgt, tgt_1], Original ATen: [aten.add, aten.native_layer_norm]
# Source node to ATen node mapping:
# tgt => add
# tgt_1 => add_1, add_2, mul_1, mul_2, rsqrt, sub_1
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_1, %squeeze), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_6, 1e-05), kwargs = {})
# %rsqrt : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_1,), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add, %getitem_7), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_1, %rsqrt), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_1, %primals_6), kwargs = {})
# %add_2 : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_7), kwargs = {})
triton_poi_fused_add_native_layer_norm_5 = async_compile.triton('triton_poi_fused_add_native_layer_norm_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*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_5', '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_5(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/6t/c6tnsmtuu3suynz2ifqymnfdxlfvcse7jkckh7hudvirutnyhqo6.py
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# multi_head_attention_forward_1 => mean_1
# Graph fragment:
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view_23, [1]), kwargs = {})
triton_poi_fused_mean_6 = async_compile.triton('triton_poi_fused_mean_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_mean_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_mean_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
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + (x0), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/c4/cc4dqkyamsucxmjrjcvvg3ydlbeeie3lo3r4whs3y5wqsmmem3gh.py
# Topologically Sorted Source Nodes: [tgt_2], Original ATen: [aten.add]
# Source node to ATen node mapping:
# tgt_2 => add_3
# Graph fragment:
# %add_3 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %squeeze_2), kwargs = {})
triton_poi_fused_add_7 = async_compile.triton('triton_poi_fused_add_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_7', '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_7(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_ptr0 + (x2), xmask)
tmp1 = tl.load(in_out_ptr0 + (x2), xmask)
tmp2 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/x7/cx77v65phn2ltvve3kt3h4jamwjhkohhviebke5cut77oqaa7hv5.py
# Topologically Sorted Source Nodes: [tgt_3], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# tgt_3 => add_4, rsqrt_1, var_mean_1
# Graph fragment:
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_3, [1]), kwargs = {correction: 0, keepdim: True})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_14, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_4,), kwargs = {})
triton_poi_fused_native_layer_norm_8 = async_compile.triton('triton_poi_fused_native_layer_norm_8', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_8', '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_8(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/bp/cbpgtlftqe6wfbyra37xz5scae7dmze5b4us4dl5654nlqku5m6n.py
# Topologically Sorted Source Nodes: [tgt_3], Original ATen: [aten.native_layer_norm]
# Source node to ATen node mapping:
# tgt_3 => add_4, add_5, mul_4, mul_5, rsqrt_1, sub_3, var_mean_1
# Graph fragment:
# %var_mean_1 : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%add_3, [1]), kwargs = {correction: 0, keepdim: True})
# %add_4 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_14, 1e-05), kwargs = {})
# %rsqrt_1 : [num_users=1] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add_4,), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_3, %getitem_15), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub_3, %rsqrt_1), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul_4, %primals_13), kwargs = {})
# %add_5 : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_5, %primals_14), kwargs = {})
triton_poi_fused_native_layer_norm_9 = async_compile.triton('triton_poi_fused_native_layer_norm_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: '*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_9', '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_9(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/k4/ck4ejdtxok3522cwutqt2ogd6tsq7bhgk5dt7wocvnwzbskweyfm.py
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu]
# Source node to ATen node mapping:
# relu => relu
# Graph fragment:
# %add_tensor_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mm_default_1, %primals_16), kwargs = {})
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%add_tensor_1,), kwargs = {})
triton_poi_fused_relu_10 = async_compile.triton('triton_poi_fused_relu_10', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192],
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_10', '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_10(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 8192
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 2048
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19, primals_20 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (12, 4), (4, 1))
assert_size_stride(primals_3, (12, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, ), (1, ))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (12, 4), (4, 1))
assert_size_stride(primals_10, (12, ), (1, ))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4, ), (1, ))
assert_size_stride(primals_13, (4, ), (1, ))
assert_size_stride(primals_14, (4, ), (1, ))
assert_size_stride(primals_15, (2048, 4), (4, 1))
assert_size_stride(primals_16, (2048, ), (1, ))
assert_size_stride(primals_17, (4, 2048), (2048, 1))
assert_size_stride(primals_18, (4, ), (1, ))
assert_size_stride(primals_19, (4, ), (1, ))
assert_size_stride(primals_20, (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: [], Original ATen: []
extern_kernels.mm(primals_1, 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: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_3, (4, ), (1, ), 4), primals_1, reinterpret_tensor(primals_2, (4, 4), (1, 4), 16), alpha=1, beta=1, out=buf1)
buf2 = 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_3, (4, ), (1, ), 8), primals_1, reinterpret_tensor(primals_2, (4, 4), (1, 4), 32), alpha=1, beta=1, out=buf2)
del primals_2
buf3 = reinterpret_tensor(buf0, (4, 4, 1), (1, 4, 16), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(buf3, primals_3, 16, grid=grid(16), stream=stream0)
del primals_3
buf4 = 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(buf3, reinterpret_tensor(buf1, (4, 1, 4), (1, 1, 4), 0), out=buf4)
buf5 = 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(buf4, buf5, 64, grid=grid(64), stream=stream0)
buf6 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf5, buf6, 64, grid=grid(64), stream=stream0)
buf7 = 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(buf6, reinterpret_tensor(buf2, (4, 4, 1), (1, 4, 1), 0), out=buf7)
buf8 = 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(buf7, buf8, 4, 4, grid=grid(4, 4), stream=stream0)
buf9 = reinterpret_tensor(buf7, (4, 4), (4, 1), 0); del buf7 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(buf8, (4, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf9)
del primals_5
buf10 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf11 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
# Topologically Sorted Source Nodes: [tgt, tgt_1], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_4.run(primals_1, buf9, buf10, buf11, 4, grid=grid(4), stream=stream0)
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [tgt, tgt_1], Original ATen: [aten.add, aten.native_layer_norm]
triton_poi_fused_add_native_layer_norm_5.run(primals_1, buf9, buf10, buf11, primals_6, primals_7, buf12, 16, grid=grid(16), stream=stream0)
del primals_7
buf13 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf12, reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_10, (4, ), (1, ), 4), primals_8, reinterpret_tensor(primals_9, (4, 4), (1, 4), 16), alpha=1, beta=1, out=buf14)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.addmm]
extern_kernels.addmm(reinterpret_tensor(primals_10, (4, ), (1, ), 8), primals_8, reinterpret_tensor(primals_9, (4, 4), (1, 4), 32), alpha=1, beta=1, out=buf15)
buf16 = reinterpret_tensor(buf13, (4, 4, 1), (1, 4, 16), 0); del buf13 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.mul]
triton_poi_fused_mul_0.run(buf16, primals_10, 16, grid=grid(16), stream=stream0)
del primals_10
buf17 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.bmm]
extern_kernels.bmm(buf16, reinterpret_tensor(buf14, (4, 1, 4), (1, 1, 4), 0), out=buf17)
buf18 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf17, buf18, 64, grid=grid(64), stream=stream0)
buf19 = buf17; del buf17 # reuse
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf18, buf19, 64, grid=grid(64), stream=stream0)
del buf18
buf20 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.bmm]
extern_kernels.bmm(buf19, reinterpret_tensor(buf15, (4, 4, 1), (1, 4, 1), 0), out=buf20)
buf21 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf20, buf21, 4, 4, grid=grid(4, 4), stream=stream0)
buf22 = reinterpret_tensor(buf20, (4, 4), (4, 1), 0); del buf20 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf21, (4, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf22)
buf23 = empty_strided_cuda((1, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [multi_head_attention_forward_1], Original ATen: [aten.mean]
triton_poi_fused_mean_6.run(buf19, buf23, 16, grid=grid(16), stream=stream0)
buf24 = buf22; del buf22 # reuse
# Topologically Sorted Source Nodes: [tgt_2], Original ATen: [aten.add]
triton_poi_fused_add_7.run(buf24, buf12, primals_12, 16, grid=grid(16), stream=stream0)
del primals_12
buf25 = buf11; del buf11 # reuse
buf26 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [tgt_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_8.run(buf24, buf25, buf26, 4, grid=grid(4), stream=stream0)
buf27 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [tgt_3], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_9.run(buf24, buf25, buf26, primals_13, primals_14, buf27, 16, grid=grid(16), stream=stream0)
del primals_14
buf28 = empty_strided_cuda((4, 2048), (2048, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf27, reinterpret_tensor(primals_15, (4, 2048), (1, 4), 0), out=buf28)
buf29 = buf28; del buf28 # reuse
# Topologically Sorted Source Nodes: [relu], Original ATen: [aten.relu]
triton_poi_fused_relu_10.run(buf29, primals_16, 8192, grid=grid(8192), stream=stream0)
del primals_16
buf30 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(buf29, reinterpret_tensor(primals_17, (2048, 4), (1, 2048), 0), out=buf30)
buf31 = buf30; del buf30 # reuse
# Topologically Sorted Source Nodes: [tgt_4], Original ATen: [aten.add]
triton_poi_fused_add_7.run(buf31, buf27, primals_18, 16, grid=grid(16), stream=stream0)
del primals_18
buf32 = buf26; del buf26 # reuse
buf33 = buf25; del buf25 # reuse
# Topologically Sorted Source Nodes: [tgt_5], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_8.run(buf31, buf32, buf33, 4, grid=grid(4), stream=stream0)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [tgt_5], Original ATen: [aten.native_layer_norm]
triton_poi_fused_native_layer_norm_9.run(buf31, buf32, buf33, primals_19, primals_20, buf34, 16, grid=grid(16), stream=stream0)
del buf32
del buf33
del primals_20
return (buf34, reinterpret_tensor(buf23, (4, 4), (4, 1), 0), primals_1, primals_6, primals_13, primals_19, buf6, reinterpret_tensor(buf8, (4, 4), (4, 1), 0), buf9, buf12, primals_8, buf19, reinterpret_tensor(buf21, (4, 4), (4, 1), 0), buf24, buf27, buf29, buf31, primals_17, primals_15, primals_11, reinterpret_tensor(buf15, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf16, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf14, (4, 4, 1), (1, 4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (4, 1), 0), primals_4, reinterpret_tensor(buf2, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf3, (4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf1, (4, 4, 1), (1, 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((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((12, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((12, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((12, ), (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, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_14 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_15 = rand_strided((2048, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((2048, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((4, 2048), (2048, 1), device='cuda:0', dtype=torch.float32)
primals_18 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_19 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_20 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19, primals_20])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import Tensor
from typing import Tuple
from typing import Optional
import torch.nn as nn
class VisTransformerDecoderLayer(nn.TransformerDecoderLayer):
def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1,
activation='relu', layer_norm_eps=1e-05, batch_first=False, device=
None, dtype=None) ->None:
super(VisTransformerDecoderLayer, self).__init__(d_model, nhead,
dim_feedforward=dim_feedforward, dropout=dropout, activation=
activation, layer_norm_eps=layer_norm_eps, batch_first=
batch_first, device=device, dtype=dtype)
def forward(self, tgt: 'Tensor', memory: 'Tensor', tgt_mask:
'Optional[Tensor]'=None, memory_mask: 'Optional[Tensor]'=None,
tgt_key_padding_mask: 'Optional[Tensor]'=None,
memory_key_padding_mask: 'Optional[Tensor]'=None) ->Tuple[Tensor,
Tensor]:
"""Pass the inputs (and mask) through the decoder layer.
Args:
tgt: the sequence to the decoder layer (required).
memory: the sequence from the last layer of the encoder (required).
tgt_mask: the mask for the tgt sequence (optional).
memory_mask: the mask for the memory sequence (optional).
tgt_key_padding_mask: the mask for the tgt keys per batch (optional).
memory_key_padding_mask: the mask for the memory keys per batch (optional).
Shape:
see the docs in Transformer class.
"""
tgt2 = self.self_attn(tgt, tgt, tgt, attn_mask=tgt_mask,
key_padding_mask=tgt_key_padding_mask)[0]
tgt = tgt + self.dropout1(tgt2)
tgt = self.norm1(tgt)
tgt2, attn_output_weights = self.multihead_attn(tgt, memory, memory,
attn_mask=memory_mask, key_padding_mask=memory_key_padding_mask)
tgt = tgt + self.dropout2(tgt2)
tgt = self.norm2(tgt)
tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt))))
tgt = tgt + self.dropout3(tgt2)
tgt = self.norm3(tgt)
return tgt, attn_output_weights
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'d_model': 4, 'nhead': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_mul_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_native_layer_norm_4(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_5(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_mean_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
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + x0, tmp8, xmask)
@triton.jit
def triton_poi_fused_add_7(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_ptr0 + x2, xmask)
tmp1 = tl.load(in_out_ptr0 + x2, xmask)
tmp2 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tmp0 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_native_layer_norm_8(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_9(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_relu_10(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 2048
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13, primals_14, primals_15, primals_16, primals_17,
primals_18, primals_19, primals_20) = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (12, 4), (4, 1))
assert_size_stride(primals_3, (12,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4,), (1,))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (12, 4), (4, 1))
assert_size_stride(primals_10, (12,), (1,))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (4,), (1,))
assert_size_stride(primals_13, (4,), (1,))
assert_size_stride(primals_14, (4,), (1,))
assert_size_stride(primals_15, (2048, 4), (4, 1))
assert_size_stride(primals_16, (2048,), (1,))
assert_size_stride(primals_17, (4, 2048), (2048, 1))
assert_size_stride(primals_18, (4,), (1,))
assert_size_stride(primals_19, (4,), (1,))
assert_size_stride(primals_20, (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_1, reinterpret_tensor(primals_2, (4, 4),
(1, 4), 0), out=buf0)
buf1 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_3, (4,), (1,), 4),
primals_1, reinterpret_tensor(primals_2, (4, 4), (1, 4), 16),
alpha=1, beta=1, out=buf1)
buf2 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_3, (4,), (1,), 8),
primals_1, reinterpret_tensor(primals_2, (4, 4), (1, 4), 32),
alpha=1, beta=1, out=buf2)
del primals_2
buf3 = reinterpret_tensor(buf0, (4, 4, 1), (1, 4, 16), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_mul_0[grid(16)](buf3, primals_3, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_3
buf4 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(buf3, reinterpret_tensor(buf1, (4, 1, 4), (1, 1,
4), 0), out=buf4)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(64)](buf4, buf5, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf6 = buf4
del buf4
triton_poi_fused__softmax_2[grid(64)](buf5, buf6, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf7 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(buf6, reinterpret_tensor(buf2, (4, 4, 1), (1, 4,
1), 0), out=buf7)
buf8 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
triton_poi_fused_clone_3[grid(4, 4)](buf7, buf8, 4, 4, XBLOCK=4,
YBLOCK=4, num_warps=1, num_stages=1)
buf9 = reinterpret_tensor(buf7, (4, 4), (4, 1), 0)
del buf7
extern_kernels.addmm(primals_5, reinterpret_tensor(buf8, (4, 4), (4,
1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha
=1, beta=1, out=buf9)
del primals_5
buf10 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
buf11 = empty_strided_cuda((4, 1), (1, 4), torch.float32)
triton_poi_fused_add_native_layer_norm_4[grid(4)](primals_1, buf9,
buf10, buf11, 4, XBLOCK=4, num_warps=1, num_stages=1)
buf12 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_add_native_layer_norm_5[grid(16)](primals_1, buf9,
buf10, buf11, primals_6, primals_7, buf12, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_7
buf13 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf12, reinterpret_tensor(primals_9, (4, 4), (1,
4), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_10, (4,), (1,), 4),
primals_8, reinterpret_tensor(primals_9, (4, 4), (1, 4), 16),
alpha=1, beta=1, out=buf14)
buf15 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.addmm(reinterpret_tensor(primals_10, (4,), (1,), 8),
primals_8, reinterpret_tensor(primals_9, (4, 4), (1, 4), 32),
alpha=1, beta=1, out=buf15)
buf16 = reinterpret_tensor(buf13, (4, 4, 1), (1, 4, 16), 0)
del buf13
triton_poi_fused_mul_0[grid(16)](buf16, primals_10, 16, XBLOCK=16,
num_warps=1, num_stages=1)
del primals_10
buf17 = buf5
del buf5
extern_kernels.bmm(buf16, reinterpret_tensor(buf14, (4, 1, 4), (1,
1, 4), 0), out=buf17)
buf18 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(64)](buf17, buf18, 64, XBLOCK=64,
num_warps=1, num_stages=1)
buf19 = buf17
del buf17
triton_poi_fused__softmax_2[grid(64)](buf18, buf19, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf18
buf20 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(buf19, reinterpret_tensor(buf15, (4, 4, 1), (1,
4, 1), 0), out=buf20)
buf21 = empty_strided_cuda((4, 4, 1), (4, 1, 1), torch.float32)
triton_poi_fused_clone_3[grid(4, 4)](buf20, buf21, 4, 4, XBLOCK=4,
YBLOCK=4, num_warps=1, num_stages=1)
buf22 = reinterpret_tensor(buf20, (4, 4), (4, 1), 0)
del buf20
extern_kernels.mm(reinterpret_tensor(buf21, (4, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf22)
buf23 = empty_strided_cuda((1, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_mean_6[grid(16)](buf19, buf23, 16, XBLOCK=16,
num_warps=1, num_stages=1)
buf24 = buf22
del buf22
triton_poi_fused_add_7[grid(16)](buf24, buf12, primals_12, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_12
buf25 = buf11
del buf11
buf26 = buf10
del buf10
triton_poi_fused_native_layer_norm_8[grid(4)](buf24, buf25, buf26,
4, XBLOCK=4, num_warps=1, num_stages=1)
buf27 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_native_layer_norm_9[grid(16)](buf24, buf25, buf26,
primals_13, primals_14, buf27, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del primals_14
buf28 = empty_strided_cuda((4, 2048), (2048, 1), torch.float32)
extern_kernels.mm(buf27, reinterpret_tensor(primals_15, (4, 2048),
(1, 4), 0), out=buf28)
buf29 = buf28
del buf28
triton_poi_fused_relu_10[grid(8192)](buf29, primals_16, 8192,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_16
buf30 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
extern_kernels.mm(buf29, reinterpret_tensor(primals_17, (2048, 4),
(1, 2048), 0), out=buf30)
buf31 = buf30
del buf30
triton_poi_fused_add_7[grid(16)](buf31, buf27, primals_18, 16,
XBLOCK=16, num_warps=1, num_stages=1)
del primals_18
buf32 = buf26
del buf26
buf33 = buf25
del buf25
triton_poi_fused_native_layer_norm_8[grid(4)](buf31, buf32, buf33,
4, XBLOCK=4, num_warps=1, num_stages=1)
buf34 = empty_strided_cuda((4, 4), (4, 1), torch.float32)
triton_poi_fused_native_layer_norm_9[grid(16)](buf31, buf32, buf33,
primals_19, primals_20, buf34, 16, XBLOCK=16, num_warps=1,
num_stages=1)
del buf32
del buf33
del primals_20
return (buf34, reinterpret_tensor(buf23, (4, 4), (4, 1), 0), primals_1,
primals_6, primals_13, primals_19, buf6, reinterpret_tensor(buf8, (
4, 4), (4, 1), 0), buf9, buf12, primals_8, buf19,
reinterpret_tensor(buf21, (4, 4), (4, 1), 0), buf24, buf27, buf29,
buf31, primals_17, primals_15, primals_11, reinterpret_tensor(buf15,
(4, 1, 4), (1, 1, 4), 0), reinterpret_tensor(buf16, (4, 1, 4), (1,
1, 4), 0), reinterpret_tensor(buf14, (4, 4, 1), (1, 4, 1), 0),
reinterpret_tensor(primals_9, (4, 4), (4, 1), 0), primals_4,
reinterpret_tensor(buf2, (4, 1, 4), (1, 1, 4), 0),
reinterpret_tensor(buf3, (4, 1, 4), (1, 1, 4), 0),
reinterpret_tensor(buf1, (4, 4, 1), (1, 4, 1), 0))
class VisTransformerDecoderLayerNew(nn.TransformerDecoderLayer):
def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1,
activation='relu', layer_norm_eps=1e-05, batch_first=False, device=
None, dtype=None) ->None:
super(VisTransformerDecoderLayerNew, self).__init__(d_model, nhead,
dim_feedforward=dim_feedforward, dropout=dropout, activation=
activation, layer_norm_eps=layer_norm_eps, batch_first=
batch_first, device=device, dtype=dtype)
def forward(self, input_0, input_1):
primals_2 = self.self_attn.in_proj_weight
primals_3 = self.self_attn.in_proj_bias
primals_1 = self.self_attn.out_proj.weight
primals_5 = self.self_attn.out_proj.bias
primals_9 = self.multihead_attn.in_proj_weight
primals_10 = self.multihead_attn.in_proj_bias
primals_4 = self.multihead_attn.out_proj.weight
primals_6 = self.multihead_attn.out_proj.bias
primals_15 = self.linear1.weight
primals_16 = self.linear1.bias
primals_17 = self.linear2.weight
primals_7 = self.linear2.bias
primals_12 = self.norm1.weight
primals_13 = self.norm1.bias
primals_14 = self.norm2.weight
primals_18 = self.norm2.bias
primals_19 = self.norm3.weight
primals_20 = self.norm3.bias
primals_8 = input_0
primals_11 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13, primals_14,
primals_15, primals_16, primals_17, primals_18, primals_19,
primals_20])
return output[0], output[1]
|
Kamino666/Video-Captioning-Transformer
|
VisTransformerDecoderLayer
| false
| 8,409
|
[
"Apache-2.0"
] | 14
|
06e6c95d9bf11d61f5825be3c640e489521f9934
|
https://github.com/Kamino666/Video-Captioning-Transformer/tree/06e6c95d9bf11d61f5825be3c640e489521f9934
|
SAB
|
# 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/m2/cm2m4qsjyjz2a6kyhcrh5cr3nu35vr4fchqqwxvyt5do5bxsewau.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_3,), 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=[4, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + (16*y0)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/om/com5a2jdtzsd2gxozlfy64n7thpakjehaajnnx7weffss4gxj2sy.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_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
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/5m/c5mma4y56ura3imiphserxkqyervoqe3bptp4i4swvp3yenvzn36.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/o5/co52ni3kaq2u2zi4ueyjtns22aw3kcfcubdifioq6mckemevvn2u.py
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_3 => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), 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=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/l3/cl35un3kcsf7pvplhghgirnuph6p7ili57xhcg3be5ggxi7rvwhl.py
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
# Source node to ATen node mapping:
# h_1 => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_21, %primals_1), 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=[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_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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ip/cipnm4mskrxmuoiysdead6jtudm6fsnxgrtfs44ii4ovxepoidk3.py
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
# Source node to ATen node mapping:
# h_2 => add_1
# relu => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_23,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %relu), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_5 = async_compile.triton('triton_poi_fused_add_relu_threshold_backward_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 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_add_relu_threshold_backward_5', '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_relu_threshold_backward_5(in_ptr0, in_ptr1, in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + (x2), tmp6, xmask)
tl.store(out_ptr1 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, ), (1, ))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4, ), (1, ))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4, ), (1, ))
assert_size_stride(primals_10, (4, 4), (4, 1))
assert_size_stride(primals_11, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, primals_3, buf3, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_3
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf1, primals_5, buf4, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_1.run(buf5, buf6, 256, grid=grid(256), stream=stream0)
buf7 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
triton_poi_fused_clone_0.run(buf2, primals_7, buf8, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_7
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [o], Original ATen: [aten.bmm]
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 4, 1), (4, 1, 0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
triton_poi_fused_clone_3.run(buf9, buf10, 16, 4, grid=grid(16, 4), stream=stream0)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0); del buf9 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf11)
buf12 = reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0); del buf11 # reuse
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
triton_poi_fused_add_4.run(buf12, primals_9, primals_1, 64, grid=grid(64), stream=stream0)
del primals_9
buf13 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf12, (16, 4), (4, 1), 0), reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_5.run(buf12, buf13, primals_11, buf14, buf15, 64, grid=grid(64), stream=stream0)
del buf13
del primals_11
return (buf14, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), buf7, reinterpret_tensor(buf10, (16, 4), (4, 1), 0), reinterpret_tensor(buf12, (16, 4), (4, 1), 0), buf15, primals_10, primals_8, reinterpret_tensor(buf8, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, 4), (4, 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)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import math
import torch
import numpy as np
import torch.nn.functional as F
import torch.nn as nn
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
class SAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.mab = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
def forward(self, x, presence=None):
return self.mab(x, x, presence)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'d': 4, 'n_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 math
import numpy as np
import torch.nn.functional as F
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, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x1), xmask & ymask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + 16 * y0), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x1), xmask & ymask, eviction_policy
='evict_last')
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_4(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_5(in_ptr0, in_ptr1,
in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + x2, tmp6, xmask)
tl.store(out_ptr1 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4,), (1,))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
assert_size_stride(primals_6, (4, 4), (4, 1))
assert_size_stride(primals_7, (4,), (1,))
assert_size_stride(primals_8, (4, 4), (4, 1))
assert_size_stride(primals_9, (4,), (1,))
assert_size_stride(primals_10, (4, 4), (4, 1))
assert_size_stride(primals_11, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_2, (4, 4), (1, 4), 0), out=buf0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), out=buf1)
del primals_4
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_6, (4, 4), (1, 4), 0), out=buf2)
del primals_6
buf3 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(4, 16)](buf0, primals_3, buf3, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_3
buf4 = reinterpret_tensor(buf0, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf0
triton_poi_fused_clone_0[grid(4, 16)](buf1, primals_5, buf4, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf4, (16, 1, 4), (4, 0, 1), 0), out=buf5)
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_1[grid(256)](buf5, buf6, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf7 = buf5
del buf5
triton_poi_fused__softmax_2[grid(256)](buf6, buf7, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf6
buf8 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_clone_0[grid(4, 16)](buf2, primals_7, buf8, 4, 16,
XBLOCK=8, YBLOCK=4, num_warps=1, num_stages=1)
del primals_7
buf9 = reinterpret_tensor(buf2, (16, 4, 1), (4, 1, 1), 0)
del buf2
extern_kernels.bmm(buf7, reinterpret_tensor(buf8, (16, 4, 1), (4, 1,
0), 0), out=buf9)
buf10 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_3[grid(16, 4)](buf9, buf10, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf11 = reinterpret_tensor(buf9, (16, 4), (4, 1), 0)
del buf9
extern_kernels.mm(reinterpret_tensor(buf10, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_8, (4, 4), (1, 4), 0), out=buf11)
buf12 = reinterpret_tensor(buf11, (4, 4, 4), (16, 4, 1), 0)
del buf11
triton_poi_fused_add_4[grid(64)](buf12, primals_9, primals_1, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_9
buf13 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf12, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_10, (4, 4), (1, 4), 0), out=buf13)
buf14 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_5[grid(64)](buf12,
buf13, primals_11, buf14, buf15, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf13
del primals_11
return buf14, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), buf7, reinterpret_tensor(buf10, (16, 4), (4, 1), 0
), reinterpret_tensor(buf12, (16, 4), (4, 1), 0
), buf15, primals_10, primals_8, reinterpret_tensor(buf8, (16, 1, 4
), (4, 1, 1), 0), reinterpret_tensor(buf3, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 1), 0)
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
class SABNew(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.mab = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
def forward(self, input_0):
primals_2 = self.mab.mqkv.q_projector.weight
primals_3 = self.mab.mqkv.q_projector.bias
primals_4 = self.mab.mqkv.k_projector.weight
primals_5 = self.mab.mqkv.k_projector.bias
primals_6 = self.mab.mqkv.v_projector.weight
primals_7 = self.mab.mqkv.v_projector.bias
primals_8 = self.mab.mqkv.o_projector.weight
primals_9 = self.mab.mqkv.o_projector.bias
primals_10 = self.mab.fc.weight
primals_11 = self.mab.fc.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
|
KohavTal/SCAE_Project
|
SAB
| false
| 8,410
|
[
"Apache-2.0"
] | 40
|
bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
https://github.com/KohavTal/SCAE_Project/tree/bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
AvgPoolShortCut
|
# 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/4b/c4brz7ereswzcaqtbznmyf4sucbm3djdkkcc2nnv63dvnoccs6do.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=1] = call_function[target=torch.ops.aten.avg_pool2d.default](args = (%arg0_1, [1, 1], [1, 1]), kwargs = {})
triton_poi_fused_avg_pool2d_0 = async_compile.triton('triton_poi_fused_avg_pool2d_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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': 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_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.avg_pool2d]
stream0 = get_raw_stream(0)
triton_poi_fused_avg_pool2d_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
from torch import nn
from torch.nn import functional as F
class AvgPoolShortCut(nn.Module):
def __init__(self, stride, out_c, in_c):
super(AvgPoolShortCut, self).__init__()
self.stride = stride
self.out_c = out_c
self.in_c = in_c
def forward(self, x):
if x.shape[2] % 2 != 0:
x = F.avg_pool2d(x, 1, self.stride)
else:
x = F.avg_pool2d(x, self.stride, self.stride)
pad = torch.zeros(x.shape[0], self.out_c - self.in_c, x.shape[2], x
.shape[3], device=x.device)
x = torch.cat((x, pad), dim=1)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'stride': 1, 'out_c': 4, 'in_c': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch 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_avg_pool2d_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 1.0
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_avg_pool2d_0[grid(256)](arg0_1, buf0, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class AvgPoolShortCutNew(nn.Module):
def __init__(self, stride, out_c, in_c):
super(AvgPoolShortCutNew, self).__init__()
self.stride = stride
self.out_c = out_c
self.in_c = in_c
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Karthik-Ragunath/DDU
|
AvgPoolShortCut
| false
| 8,411
|
[
"MIT"
] | 43
|
b9daae9304bdeb222857884ef8cb3b6b3d004d33
|
https://github.com/Karthik-Ragunath/DDU/tree/b9daae9304bdeb222857884ef8cb3b6b3d004d33
|
CNet
|
# 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/qk/cqkiz525yi2cjplmd3dj5polsvupseswfkemi6lbx374vnvnzsy7.py
# Topologically Sorted Source Nodes: [add, net], Original ATen: [aten.add, aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# add => add
# net => relu
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_1, %view_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=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 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_add_relu_threshold_backward_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_add_relu_threshold_backward_0(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 1920
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 30
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 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp7 = tl.full([1], 0, tl.int32)
tmp8 = triton_helpers.maximum(tmp7, tmp6)
tmp9 = 0.0
tmp10 = tmp8 <= tmp9
tl.store(in_out_ptr0 + (x2), tmp8, xmask)
tl.store(out_ptr0 + (x2), tmp10, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8 = args
args.clear()
assert_size_stride(primals_1, (30, 4), (4, 1))
assert_size_stride(primals_2, (30, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (30, 4), (4, 1))
assert_size_stride(primals_5, (30, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_7, (1, 30), (30, 1))
assert_size_stride(primals_8, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 30), (30, 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, 30), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((64, 30), (30, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 30), (1, 4), 0), out=buf1)
del primals_4
buf2 = reinterpret_tensor(buf0, (4, 4, 4, 30), (480, 120, 30, 1), 0); del buf0 # reuse
buf5 = empty_strided_cuda((4, 4, 4, 30), (480, 120, 30, 1), torch.bool)
# Topologically Sorted Source Nodes: [add, net], 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_2, buf1, primals_5, buf5, 1920, grid=grid(1920), stream=stream0)
del buf1
del primals_2
del primals_5
buf4 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [actions_value], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, reinterpret_tensor(buf2, (64, 30), (30, 1), 0), reinterpret_tensor(primals_7, (30, 1), (1, 30), 0), alpha=1, beta=1, out=buf4)
del primals_8
return (reinterpret_tensor(buf4, (4, 4, 4, 1), (16, 4, 1, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_6, (64, 4), (4, 1), 0), reinterpret_tensor(buf2, (64, 30), (30, 1), 0), primals_7, 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((30, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((30, ), (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((30, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((30, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, 30), (30, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
class CNet(nn.Module):
def __init__(self, s_dim, a_dim):
super(CNet, self).__init__()
self.fcs = nn.Linear(s_dim, 30)
self.fcs.weight.data.normal_(0, 0.1)
self.fca = nn.Linear(a_dim, 30)
self.fca.weight.data.normal_(0, 0.1)
self.out = nn.Linear(30, 1)
self.out.weight.data.normal_(0, 0.1)
def forward(self, s, a):
x = self.fcs(s)
y = self.fca(a)
net = F.relu(x + y)
actions_value = self.out(net)
return actions_value
class net(nn.Module):
def __init__(self, input_dim, output_dim):
super(net, self).__init__()
self.fc1 = nn.Linear(input_dim, 30)
self.fc1.weight.data.normal_(0, 1)
self.fc2 = nn.Linear(30, 20)
self.fc2.weight.data.normal_(0, 1)
self.fc3 = nn.Linear(20, output_dim)
self.fc3.weight.data.normal_(0, 1)
def forward(self, x):
x = F.tanh(self.fc1(x))
x = F.tanh(self.fc2(x))
out = self.fc3(x)
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'s_dim': 4, 'a_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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,
in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 1920
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 30
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 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp5 = tmp3 + tmp4
tmp6 = tmp2 + tmp5
tmp7 = tl.full([1], 0, tl.int32)
tmp8 = triton_helpers.maximum(tmp7, tmp6)
tmp9 = 0.0
tmp10 = tmp8 <= tmp9
tl.store(in_out_ptr0 + x2, tmp8, xmask)
tl.store(out_ptr0 + x2, tmp10, 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, (30, 4), (4, 1))
assert_size_stride(primals_2, (30,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (30, 4), (4, 1))
assert_size_stride(primals_5, (30,), (1,))
assert_size_stride(primals_6, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_7, (1, 30), (30, 1))
assert_size_stride(primals_8, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 30), (30, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 30), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((64, 30), (30, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_6, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_4, (4, 30), (1, 4), 0), out=buf1)
del primals_4
buf2 = reinterpret_tensor(buf0, (4, 4, 4, 30), (480, 120, 30, 1), 0)
del buf0
buf5 = empty_strided_cuda((4, 4, 4, 30), (480, 120, 30, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_add_relu_threshold_backward_0[grid(1920)](buf2,
primals_2, buf1, primals_5, buf5, 1920, XBLOCK=128, num_warps=4,
num_stages=1)
del buf1
del primals_2
del primals_5
buf4 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_8, reinterpret_tensor(buf2, (64, 30),
(30, 1), 0), reinterpret_tensor(primals_7, (30, 1), (1, 30), 0),
alpha=1, beta=1, out=buf4)
del primals_8
return reinterpret_tensor(buf4, (4, 4, 4, 1), (16, 4, 1, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), reinterpret_tensor(primals_6, (64, 4), (4, 1), 0
), reinterpret_tensor(buf2, (64, 30), (30, 1), 0), primals_7, buf5
class CNetNew(nn.Module):
def __init__(self, s_dim, a_dim):
super(CNetNew, self).__init__()
self.fcs = nn.Linear(s_dim, 30)
self.fcs.weight.data.normal_(0, 0.1)
self.fca = nn.Linear(a_dim, 30)
self.fca.weight.data.normal_(0, 0.1)
self.out = nn.Linear(30, 1)
self.out.weight.data.normal_(0, 0.1)
def forward(self, input_0, input_1):
primals_1 = self.fcs.weight
primals_2 = self.fcs.bias
primals_4 = self.fca.weight
primals_5 = self.fca.bias
primals_7 = self.out.weight
primals_8 = self.out.bias
primals_3 = input_0
primals_6 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0]
class net(nn.Module):
def __init__(self, input_dim, output_dim):
super(net, self).__init__()
self.fc1 = nn.Linear(input_dim, 30)
self.fc1.weight.data.normal_(0, 1)
self.fc2 = nn.Linear(30, 20)
self.fc2.weight.data.normal_(0, 1)
self.fc3 = nn.Linear(20, output_dim)
self.fc3.weight.data.normal_(0, 1)
def forward(self, x):
x = F.tanh(self.fc1(x))
x = F.tanh(self.fc2(x))
out = self.fc3(x)
return out
|
Kernels-K/DDPG-pytorch-
|
CNet
| false
| 8,412
|
[
"MIT"
] | 26
|
9a80a56f52f2232e5bd197521d3d2d388b48c882
|
https://github.com/Kernels-K/DDPG-pytorch-/tree/9a80a56f52f2232e5bd197521d3d2d388b48c882
|
HSwish
|
# 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/rw/crwdk3j4ojtagun7zhqwyf5bqe7salukzcept7ykzhjfr5nro3cm.py
# Topologically Sorted Source Nodes: [add, hardtanh, mul, truediv], Original ATen: [aten.add, aten.hardtanh, aten.mul, aten.div]
# Source node to ATen node mapping:
# add => add
# hardtanh => clamp_max, clamp_min
# mul => mul
# truediv => div
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 3.0), kwargs = {})
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%add, 0.0), kwargs = {})
# %clamp_max : [num_users=1] = call_function[target=torch.ops.aten.clamp_max.default](args = (%clamp_min, 6.0), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %clamp_max), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul, 6.0), kwargs = {})
triton_poi_fused_add_div_hardtanh_mul_0 = async_compile.triton('triton_poi_fused_add_div_hardtanh_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_hardtanh_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_hardtanh_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 = 3.0
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = 6.0
tmp6 = triton_helpers.minimum(tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = 0.16666666666666666
tmp9 = tmp7 * tmp8
tl.store(out_ptr0 + (x0), tmp9, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [add, hardtanh, mul, truediv], Original ATen: [aten.add, aten.hardtanh, aten.mul, aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_hardtanh_mul_0.run(arg0_1, buf0, 256, grid=grid(256), stream=stream0)
del arg0_1
return (buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([arg0_1])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn
class HSwish(nn.Module):
"""
H-Swish activation function from 'Searching for MobileNetV3,' https://arxiv.org/abs/1905.02244.
Parameters:
----------
inplace : bool
Whether to use inplace version of the module.
"""
def __init__(self, inplace=True):
super(HSwish, self).__init__()
self.inplace = inplace
self.relu = nn.ReLU6(inplace=self.inplace)
def forward(self, x):
return x * self.relu(x + 3.0) / 6.0
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
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
@triton.jit
def triton_poi_fused_add_div_hardtanh_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 = 3.0
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = 6.0
tmp6 = triton_helpers.minimum(tmp4, tmp5)
tmp7 = tmp0 * tmp6
tmp8 = 0.16666666666666666
tmp9 = tmp7 * tmp8
tl.store(out_ptr0 + x0, tmp9, 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_hardtanh_mul_0[grid(256)](arg0_1, buf0,
256, XBLOCK=256, num_warps=4, num_stages=1)
del arg0_1
return buf0,
class HSwishNew(nn.Module):
"""
H-Swish activation function from 'Searching for MobileNetV3,' https://arxiv.org/abs/1905.02244.
Parameters:
----------
inplace : bool
Whether to use inplace version of the module.
"""
def __init__(self, inplace=True):
super(HSwishNew, self).__init__()
self.inplace = inplace
self.relu = nn.ReLU6(inplace=self.inplace)
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Kthyeon/micronet_neurips_challenge
|
HSwish
| false
| 8,413
|
[
"MIT"
] | 19
|
9f71fb752e8fbd5abca07be530f7fb19e164125c
|
https://github.com/Kthyeon/micronet_neurips_challenge/tree/9f71fb752e8fbd5abca07be530f7fb19e164125c
|
SAMblock
|
# 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/7r/c7r2mp7slj6ma5k33x7azsnkn4fzyuvu5pejkuomkbcqmlt3dtua.py
# Topologically Sorted Source Nodes: [sp], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# sp => cat
# Graph fragment:
# %cat : [num_users=2] = call_function[target=torch.ops.aten.cat.default](args = ([%getitem, %mean], 1), kwargs = {})
triton_poi_fused_cat_0 = async_compile.triton('triton_poi_fused_cat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp9 = triton_helpers.maximum(tmp7, tmp8)
tmp10 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp4 & xmask, eviction_policy='evict_last', other=0.0)
tmp11 = triton_helpers.maximum(tmp9, tmp10)
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tmp15 = tl.full([1], 2, tl.int64)
tmp16 = tmp0 < tmp15
tmp17 = tl.load(in_ptr0 + (x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp18 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp19 = tmp17 + tmp18
tmp20 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp21 = tmp19 + tmp20
tmp22 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp23 = tmp21 + tmp22
tmp24 = 4.0
tmp25 = tmp23 / tmp24
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp14, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp13, tmp27)
tl.store(out_ptr0 + (x3), tmp28, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/id/cidq7yz75eih55wcintoo4y4wimds2ct66mggi4tcj5vxeakq36c.py
# Topologically Sorted Source Nodes: [sp_1], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# sp_1 => convolution
# Graph fragment:
# %convolution : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%cat, %primals_2, %primals_3, [1, 1], [3, 3], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 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
tl.store(in_out_ptr0 + (x0), tmp3, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/rw/crw5uk2cftormrsyby7jmeo5c4fmy2udygypvlxxysucldgzv7ep.py
# Topologically Sorted Source Nodes: [sp_2, x], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# sp_2 => sigmoid
# x => mul
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution,), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %primals_1), kwargs = {})
triton_poi_fused_mul_sigmoid_2 = async_compile.triton('triton_poi_fused_mul_sigmoid_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_mul_sigmoid_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x2 = (xindex // 64)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (x3), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x3), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 7, 7), (98, 49, 7, 1))
assert_size_stride(primals_3, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sp], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, buf0, 128, grid=grid(128), stream=stream0)
# Topologically Sorted Source Nodes: [sp_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: [sp_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf2, primals_3, 64, grid=grid(64), stream=stream0)
del primals_3
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sp_2, x], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_2.run(buf2, primals_1, buf3, 256, grid=grid(256), stream=stream0)
return (buf3, primals_1, primals_2, buf0, buf2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 2, 7, 7), (98, 49, 7, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
class SAMblock(torch.nn.Module):
def __init__(self, size=7, model='full', outplanes=None):
super(SAMblock, self).__init__()
self.outplanes = outplanes
if self.outplanes is None:
self.outplanes = 1
self.model = model
self.conv1 = torch.nn.Conv2d(2, self.outplanes, (size, size),
stride=1, padding=size // 2)
if self.model != 'full':
self.conv1 = torch.nn.Conv2d(1, self.outplanes, (size, size),
stride=1, padding=size // 2)
self.sigmod = torch.nn.Sigmoid()
def forward(self, x):
if self.model == 'mean':
meanpool = torch.mean(x, 1, True)
sp = meanpool
if self.model == 'max':
maxpool, _ = torch.max(x, 1, True)
sp = maxpool
if self.model == 'full':
maxpool, _ = torch.max(x, 1, True)
meanpool = torch.mean(x, 1, True)
sp = torch.cat([maxpool, meanpool], dim=1)
sp = self.conv1(sp)
sp = self.sigmod(sp)
x = sp * 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
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 128
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 16 % 2
x0 = xindex % 16
x2 = xindex // 32
x3 = xindex
tmp0 = x1
tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp4 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp6 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp9 = triton_helpers.maximum(tmp7, tmp8)
tmp10 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp4 & xmask,
eviction_policy='evict_last', other=0.0)
tmp11 = triton_helpers.maximum(tmp9, tmp10)
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp17 = tl.load(in_ptr0 + (x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp18 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp19 = tmp17 + tmp18
tmp20 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp21 = tmp19 + tmp20
tmp22 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), tmp14 & xmask,
eviction_policy='evict_last', other=0.0)
tmp23 = tmp21 + tmp22
tmp24 = 4.0
tmp25 = tmp23 / tmp24
tmp26 = tl.full(tmp25.shape, 0.0, tmp25.dtype)
tmp27 = tl.where(tmp14, tmp25, tmp26)
tmp28 = tl.where(tmp4, tmp13, tmp27)
tl.store(out_ptr0 + x3, tmp28, xmask)
@triton.jit
def triton_poi_fused_convolution_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
tl.store(in_out_ptr0 + x0, tmp3, xmask)
@triton.jit
def triton_poi_fused_mul_sigmoid_2(in_ptr0, in_ptr1, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x2 = xindex // 64
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr1 + x3, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + x3, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 2, 7, 7), (98, 49, 7, 1))
assert_size_stride(primals_3, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 2, 4, 4), (32, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(128)](primals_1, buf0, 128, XBLOCK=128,
num_warps=4, num_stages=1)
buf1 = extern_kernels.convolution(buf0, primals_2, stride=(1, 1),
padding=(3, 3), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 1, 4, 4), (16, 16, 4, 1))
buf2 = buf1
del buf1
triton_poi_fused_convolution_1[grid(64)](buf2, primals_3, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_3
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_poi_fused_mul_sigmoid_2[grid(256)](buf2, primals_1, buf3,
256, XBLOCK=256, num_warps=4, num_stages=1)
return buf3, primals_1, primals_2, buf0, buf2
class SAMblockNew(torch.nn.Module):
def __init__(self, size=7, model='full', outplanes=None):
super(SAMblockNew, self).__init__()
self.outplanes = outplanes
if self.outplanes is None:
self.outplanes = 1
self.model = model
self.conv1 = torch.nn.Conv2d(2, self.outplanes, (size, size),
stride=1, padding=size // 2)
if self.model != 'full':
self.conv1 = torch.nn.Conv2d(1, self.outplanes, (size, size),
stride=1, padding=size // 2)
self.sigmod = torch.nn.Sigmoid()
def forward(self, input_0):
primals_2 = self.conv1.weight
primals_3 = self.conv1.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
Knight825/models-pytorch
|
SAMblock
| false
| 8,414
|
[
"Apache-2.0"
] | 16
|
133559eebb8795d78a32fa44d49408d0c5167ae9
|
https://github.com/Knight825/models-pytorch/tree/133559eebb8795d78a32fa44d49408d0c5167ae9
|
CrossAttention
|
# 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/qy/cqygb2qwws4alrj2ckver6xxou33xclrdrez4nmivbvxxa23xdtf.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024, 4096], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 1024
xnumel = 4096
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = (yindex // 256)
tmp0 = tl.load(in_ptr0 + (x2 + (4096*y3)), None, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (256*x2) + (1048576*y1)), tmp0, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/u7/cu7a662azidbq3tmhjleikux7irx7jvowvfwsbmc4iwezhqjgc3c.py
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_4, %primals_2, %primals_3, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[128, 4096], 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, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_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_convolution_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 128
xnumel = 4096
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y0 = yindex % 32
y1 = (yindex // 32)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (32*x2) + (131072*y1)), ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4096*y3)), tmp2, ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gw/cgw47fm7lqkgooogmtmaw2mqt4hnfuh7negcehsiawlcdu4iaphj.py
# Topologically Sorted Source Nodes: [mask_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# mask_1 => amax, div, exp, sub, sum_1
# 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 = {})
# %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_red_fused__softmax_2 = async_compile.triton('triton_red_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.reduction(
size_hints=[16384, 4096],
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_red_fused__softmax_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 3, '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_red_fused__softmax_2(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr, RBLOCK : tl.constexpr):
xnumel = 16384
rnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
rbase = tl.arange(0, RBLOCK)[None, :]
x0 = xindex
_tmp2 = tl.full([XBLOCK, RBLOCK], float("-inf"), tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r1 = rindex
tmp0 = tl.load(in_ptr0 + (r1 + (4096*x0)), rmask, eviction_policy='evict_last', other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = triton_helpers.maximum(_tmp2, tmp1)
_tmp2 = tl.where(rmask, tmp3, _tmp2)
tmp2 = triton_helpers.max2(_tmp2, 1)[:, None]
_tmp8 = tl.full([XBLOCK, RBLOCK], 0, tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r1 = rindex
tmp4 = tl.load(in_ptr0 + (r1 + (4096*x0)), rmask, eviction_policy='evict_last', other=0.0)
tmp5 = tmp4 - tmp2
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = _tmp8 + tmp7
_tmp8 = tl.where(rmask, tmp9, _tmp8)
tmp8 = tl.sum(_tmp8, 1)[:, None]
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r1 = rindex
tmp10 = tl.load(in_ptr0 + (r1 + (4096*x0)), rmask, eviction_policy='evict_first', other=0.0)
tmp11 = tmp10 - tmp2
tmp12 = tl_math.exp(tmp11)
tmp13 = tmp12 / tmp8
tl.store(out_ptr2 + (r1 + (4096*x0)), tmp13, rmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/uc/cucqjmgbyhlerewb4vgxo6itmnpcw4qkazlfq7ordh6mg42vaq3y.py
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d_2 => convolution_2
# Graph fragment:
# %convolution_2 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_1, %primals_7, %primals_8, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_3 = async_compile.triton('triton_poi_fused_convolution_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=[1024, 4096], 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, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_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_3(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 1024
xnumel = 4096
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y0 = yindex % 256
y1 = (yindex // 256)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (256*x2) + (1048576*y1)), None, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + (4096*y3)), tmp2, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8 = args
args.clear()
assert_size_stride(primals_1, (4, 256, 64, 64), (1048576, 4096, 64, 1))
assert_size_stride(primals_2, (32, 256, 1, 1), (256, 1, 1, 1))
assert_size_stride(primals_3, (32, ), (1, ))
assert_size_stride(primals_4, (4, 256, 64, 64), (1048576, 4096, 64, 1))
assert_size_stride(primals_5, (32, 256, 1, 1), (256, 1, 1, 1))
assert_size_stride(primals_6, (32, ), (1, ))
assert_size_stride(primals_7, (256, 256, 1, 1), (256, 1, 1, 1))
assert_size_stride(primals_8, (256, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 256, 64, 64), (1048576, 1, 16384, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(primals_1, buf0, 1024, 4096, grid=grid(1024, 4096), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 256, 64, 64), (1048576, 1, 16384, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_0.run(primals_4, buf1, 1024, 4096, grid=grid(1024, 4096), stream=stream0)
del primals_4
# Topologically Sorted Source Nodes: [conv2d], 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, 32, 64, 64), (131072, 1, 2048, 32))
buf3 = empty_strided_cuda((4, 32, 64, 64), (131072, 4096, 64, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv2d], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf2, primals_3, buf3, 128, 4096, grid=grid(128, 4096), stream=stream0)
del primals_3
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
buf4 = extern_kernels.convolution(buf1, primals_5, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 32, 64, 64), (131072, 1, 2048, 32))
buf5 = reinterpret_tensor(buf2, (4, 32, 64, 64), (131072, 4096, 64, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [conv2d_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf4, primals_6, buf5, 128, 4096, grid=grid(128, 4096), stream=stream0)
del buf4
del primals_6
buf6 = empty_strided_cuda((4, 4096, 4096), (16777216, 4096, 1), torch.float32)
# Topologically Sorted Source Nodes: [mask], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 4096, 32), (131072, 1, 4096), 0), reinterpret_tensor(buf5, (4, 32, 4096), (131072, 4096, 1), 0), out=buf6)
buf9 = empty_strided_cuda((4, 4096, 4096), (16777216, 4096, 1), torch.float32)
# Topologically Sorted Source Nodes: [mask_1], Original ATen: [aten._softmax]
triton_red_fused__softmax_2.run(buf6, buf9, 16384, 4096, grid=grid(16384), stream=stream0)
del buf6
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
buf10 = extern_kernels.convolution(buf0, primals_7, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 256, 64, 64), (1048576, 1, 16384, 256))
buf11 = empty_strided_cuda((4, 256, 64, 64), (1048576, 4096, 64, 1), torch.float32)
# Topologically Sorted Source Nodes: [conv2d_2], Original ATen: [aten.convolution]
triton_poi_fused_convolution_3.run(buf10, primals_8, buf11, 1024, 4096, grid=grid(1024, 4096), stream=stream0)
del primals_8
buf12 = reinterpret_tensor(buf10, (4, 256, 4096), (1048576, 4096, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [feat], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf11, (4, 256, 4096), (1048576, 4096, 1), 0), reinterpret_tensor(buf9, (4, 4096, 4096), (16777216, 1, 4096), 0), out=buf12)
return (reinterpret_tensor(buf12, (4, 256, 64, 64), (1048576, 4096, 64, 1), 0), buf0, primals_2, buf1, primals_5, primals_7, buf9, reinterpret_tensor(buf11, (4, 4096, 256), (1048576, 1, 4096), 0), reinterpret_tensor(buf3, (4, 32, 4096), (131072, 4096, 1), 0), reinterpret_tensor(buf5, (4, 4096, 32), (131072, 1, 4096), 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, 256, 64, 64), (1048576, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((32, 256, 1, 1), (256, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 256, 64, 64), (1048576, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((32, 256, 1, 1), (256, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((32, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((256, 256, 1, 1), (256, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class CrossAttention(nn.Module):
def __init__(self, in_channel=256, ratio=8):
super(CrossAttention, self).__init__()
self.conv_query = nn.Conv2d(in_channel, in_channel // ratio,
kernel_size=1)
self.conv_key = nn.Conv2d(in_channel, in_channel // ratio,
kernel_size=1)
self.conv_value = nn.Conv2d(in_channel, in_channel, kernel_size=1)
def forward(self, rgb, depth):
bz, c, h, w = rgb.shape
depth_q = self.conv_query(depth).view(bz, -1, h * w).permute(0, 2, 1)
depth_k = self.conv_key(depth).view(bz, -1, h * w)
mask = torch.bmm(depth_q, depth_k)
mask = torch.softmax(mask, dim=-1)
rgb_v = self.conv_value(rgb).view(bz, c, -1)
feat = torch.bmm(rgb_v, mask.permute(0, 2, 1))
feat = feat.view(bz, c, h, w)
return feat
def get_inputs():
return [torch.rand([4, 256, 64, 64]), torch.rand([4, 256, 64, 64])]
def get_init_inputs():
return [[], {}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import 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_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = yindex // 256
tmp0 = tl.load(in_ptr0 + (x2 + 4096 * y3), None, eviction_policy=
'evict_last')
tl.store(out_ptr0 + (y0 + 256 * x2 + 1048576 * y1), tmp0, None)
@triton.jit
def triton_poi_fused_convolution_1(in_ptr0, in_ptr1, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 128
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y0 = yindex % 32
y1 = yindex // 32
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 32 * x2 + 131072 * y1), ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4096 * y3), tmp2, ymask)
@triton.jit
def triton_red_fused__softmax_2(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK:
tl.constexpr, RBLOCK: tl.constexpr):
rnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
rbase = tl.arange(0, RBLOCK)[None, :]
x0 = xindex
_tmp2 = tl.full([XBLOCK, RBLOCK], float('-inf'), tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r1 = rindex
tmp0 = tl.load(in_ptr0 + (r1 + 4096 * x0), rmask, eviction_policy=
'evict_last', other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = triton_helpers.maximum(_tmp2, tmp1)
_tmp2 = tl.where(rmask, tmp3, _tmp2)
tmp2 = triton_helpers.max2(_tmp2, 1)[:, None]
_tmp8 = tl.full([XBLOCK, RBLOCK], 0, tl.float32)
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r1 = rindex
tmp4 = tl.load(in_ptr0 + (r1 + 4096 * x0), rmask, eviction_policy=
'evict_last', other=0.0)
tmp5 = tmp4 - tmp2
tmp6 = tl_math.exp(tmp5)
tmp7 = tl.broadcast_to(tmp6, [XBLOCK, RBLOCK])
tmp9 = _tmp8 + tmp7
_tmp8 = tl.where(rmask, tmp9, _tmp8)
tmp8 = tl.sum(_tmp8, 1)[:, None]
for roffset in range(0, rnumel, RBLOCK):
rindex = roffset + rbase
rmask = rindex < rnumel
r1 = rindex
tmp10 = tl.load(in_ptr0 + (r1 + 4096 * x0), rmask, eviction_policy=
'evict_first', other=0.0)
tmp11 = tmp10 - tmp2
tmp12 = tl_math.exp(tmp11)
tmp13 = tmp12 / tmp8
tl.store(out_ptr2 + (r1 + 4096 * x0), tmp13, rmask)
@triton.jit
def triton_poi_fused_convolution_3(in_ptr0, in_ptr1, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y0 = yindex % 256
y1 = yindex // 256
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 256 * x2 + 1048576 * y1), None,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x2 + 4096 * y3), tmp2, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8) = args
args.clear()
assert_size_stride(primals_1, (4, 256, 64, 64), (1048576, 4096, 64, 1))
assert_size_stride(primals_2, (32, 256, 1, 1), (256, 1, 1, 1))
assert_size_stride(primals_3, (32,), (1,))
assert_size_stride(primals_4, (4, 256, 64, 64), (1048576, 4096, 64, 1))
assert_size_stride(primals_5, (32, 256, 1, 1), (256, 1, 1, 1))
assert_size_stride(primals_6, (32,), (1,))
assert_size_stride(primals_7, (256, 256, 1, 1), (256, 1, 1, 1))
assert_size_stride(primals_8, (256,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 256, 64, 64), (1048576, 1, 16384, 256
), torch.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(1024, 4096)](primals_1, buf0, 1024, 4096,
XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 256, 64, 64), (1048576, 1, 16384, 256
), torch.float32)
triton_poi_fused_0[grid(1024, 4096)](primals_4, buf1, 1024, 4096,
XBLOCK=32, YBLOCK=32, num_warps=4, num_stages=1)
del primals_4
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, 32, 64, 64), (131072, 1, 2048, 32))
buf3 = empty_strided_cuda((4, 32, 64, 64), (131072, 4096, 64, 1),
torch.float32)
triton_poi_fused_convolution_1[grid(128, 4096)](buf2, primals_3,
buf3, 128, 4096, XBLOCK=64, YBLOCK=64, num_warps=8, num_stages=1)
del primals_3
buf4 = extern_kernels.convolution(buf1, primals_5, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf4, (4, 32, 64, 64), (131072, 1, 2048, 32))
buf5 = reinterpret_tensor(buf2, (4, 32, 64, 64), (131072, 4096, 64,
1), 0)
del buf2
triton_poi_fused_convolution_1[grid(128, 4096)](buf4, primals_6,
buf5, 128, 4096, XBLOCK=64, YBLOCK=64, num_warps=8, num_stages=1)
del buf4
del primals_6
buf6 = empty_strided_cuda((4, 4096, 4096), (16777216, 4096, 1),
torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf3, (4, 4096, 32), (131072,
1, 4096), 0), reinterpret_tensor(buf5, (4, 32, 4096), (131072,
4096, 1), 0), out=buf6)
buf9 = empty_strided_cuda((4, 4096, 4096), (16777216, 4096, 1),
torch.float32)
triton_red_fused__softmax_2[grid(16384)](buf6, buf9, 16384, 4096,
XBLOCK=1, RBLOCK=2048, num_warps=16, num_stages=1)
del buf6
buf10 = extern_kernels.convolution(buf0, primals_7, stride=(1, 1),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 256, 64, 64), (1048576, 1, 16384, 256))
buf11 = empty_strided_cuda((4, 256, 64, 64), (1048576, 4096, 64, 1),
torch.float32)
triton_poi_fused_convolution_3[grid(1024, 4096)](buf10, primals_8,
buf11, 1024, 4096, XBLOCK=16, YBLOCK=256, num_warps=8, num_stages=1
)
del primals_8
buf12 = reinterpret_tensor(buf10, (4, 256, 4096), (1048576, 4096, 1), 0
)
del buf10
extern_kernels.bmm(reinterpret_tensor(buf11, (4, 256, 4096), (
1048576, 4096, 1), 0), reinterpret_tensor(buf9, (4, 4096, 4096),
(16777216, 1, 4096), 0), out=buf12)
return reinterpret_tensor(buf12, (4, 256, 64, 64), (1048576, 4096, 64,
1), 0
), buf0, primals_2, buf1, primals_5, primals_7, buf9, reinterpret_tensor(
buf11, (4, 4096, 256), (1048576, 1, 4096), 0), reinterpret_tensor(buf3,
(4, 32, 4096), (131072, 4096, 1), 0), reinterpret_tensor(buf5, (4,
4096, 32), (131072, 1, 4096), 0)
class CrossAttentionNew(nn.Module):
def __init__(self, in_channel=256, ratio=8):
super(CrossAttentionNew, self).__init__()
self.conv_query = nn.Conv2d(in_channel, in_channel // ratio,
kernel_size=1)
self.conv_key = nn.Conv2d(in_channel, in_channel // ratio,
kernel_size=1)
self.conv_value = nn.Conv2d(in_channel, in_channel, kernel_size=1)
def forward(self, input_0, input_1):
primals_2 = self.conv_query.weight
primals_3 = self.conv_query.bias
primals_5 = self.conv_key.weight
primals_6 = self.conv_key.bias
primals_7 = self.conv_value.weight
primals_8 = self.conv_value.bias
primals_1 = input_0
primals_4 = input_1
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8])
return output[0]
|
JosephChenHub/DPANet
|
CrossAttention
| false
| 8,415
|
[
"MIT"
] | 19
|
68cf40a405d8c8c6506884079cd0a206d6d58e63
|
https://github.com/JosephChenHub/DPANet/tree/68cf40a405d8c8c6506884079cd0a206d6d58e63
|
ISAB
|
# 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/clieqwiqohdxw3rro7yubltcvmtjwclpy4jznpytei347h6na5r4.py
# Topologically Sorted Source Nodes: [repeat], Original ATen: [aten.repeat]
# Source node to ATen node mapping:
# repeat => repeat
# Graph fragment:
# %repeat : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_2, [4, 1, 1]), kwargs = {})
triton_poi_fused_repeat_0 = async_compile.triton('triton_poi_fused_repeat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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_repeat_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_repeat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
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/ls/clscue37cenpwdtzqux56gb6cc4oukj57bzuosuaav5mip3j73tk.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_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=[4, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + (16*y0)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kz/ckzqylporms4fvgcrqg44ypprwpanp6hf222rji24wskr3b44aga.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4d/c4dndrlfjcamjfnn3ng5agjc3ahefdgw6jcsnn6hm4ljwpbfbe7h.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wg/cwgjlocuqhpaklaousvfvznecyut7vf6pbamkv5aenho7h5t5xko.py
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_3 => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mu/cmu55cc2rrop6dujmkmg24kdnnkehbytpnejw6dzpsfvbjwmug5x.py
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
# Source node to ATen node mapping:
# h_1 => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_21, %repeat), kwargs = {})
triton_poi_fused_add_5 = async_compile.triton('triton_poi_fused_add_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_5', '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_5(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/6j/c6jhrcgdwigdocqbjr3po3tf6sp6t6udjnhf23dicfmtlnqloh6u.py
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
# Source node to ATen node mapping:
# h_2 => add_1
# relu => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_23,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %relu), kwargs = {})
# %le_1 : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_6 = async_compile.triton('triton_poi_fused_add_relu_threshold_backward_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 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_add_relu_threshold_backward_6', '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_relu_threshold_backward_6(in_ptr0, in_ptr1, in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + (x2), tmp6, xmask)
tl.store(out_ptr1 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19, primals_20, primals_21, primals_22 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4), (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, ))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, 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, 4), (4, 1))
assert_size_stride(primals_16, (4, ), (1, ))
assert_size_stride(primals_17, (4, 4), (4, 1))
assert_size_stride(primals_18, (4, ), (1, ))
assert_size_stride(primals_19, (4, 4), (4, 1))
assert_size_stride(primals_20, (4, ), (1, ))
assert_size_stride(primals_21, (4, 4), (4, 1))
assert_size_stride(primals_22, (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: [repeat], Original ATen: [aten.repeat]
stream0 = get_raw_stream(0)
triton_poi_fused_repeat_0.run(primals_2, buf0, 64, grid=grid(64), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((16, 4), (4, 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, 4), (1, 4), 0), out=buf2)
del primals_5
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf3)
del primals_7
buf4 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf1, primals_4, buf4, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_4
buf5 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf2, primals_6, buf5, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_6
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf5, (16, 1, 4), (4, 0, 1), 0), out=buf6)
buf7 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
buf8 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf7, buf8, 256, grid=grid(256), stream=stream0)
buf9 = reinterpret_tensor(buf2, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf3, primals_8, buf9, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_8
buf10 = reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [o], Original ATen: [aten.bmm]
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (16, 4, 1), (4, 1, 0), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf10, buf11, 16, 4, grid=grid(16, 4), stream=stream0)
buf12 = reinterpret_tensor(buf10, (16, 4), (4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf11, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf12)
buf13 = reinterpret_tensor(buf12, (4, 4, 4), (16, 4, 1), 0); del buf12 # reuse
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
triton_poi_fused_add_5.run(buf13, primals_10, buf0, 64, grid=grid(64), stream=stream0)
del primals_10
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf14)
buf15 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf15)
del primals_13
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf32 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_6.run(buf13, buf14, primals_12, buf16, buf32, 64, grid=grid(64), stream=stream0)
del primals_12
buf17 = buf14; del buf14 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf16, (16, 4), (4, 1), 0), reinterpret_tensor(primals_15, (4, 4), (1, 4), 0), out=buf17)
buf18 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf16, (16, 4), (4, 1), 0), reinterpret_tensor(primals_17, (4, 4), (1, 4), 0), out=buf18)
buf19 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_4], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf15, primals_14, buf19, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_14
buf20 = reinterpret_tensor(buf15, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf15 # reuse
# Topologically Sorted Source Nodes: [contiguous_5], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf17, primals_16, buf20, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_16
buf21 = buf7; del buf7 # reuse
# Topologically Sorted Source Nodes: [routing_2], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf19, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf20, (16, 1, 4), (4, 0, 1), 0), out=buf21)
buf22 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf21, buf22, 256, grid=grid(256), stream=stream0)
buf23 = buf21; del buf21 # reuse
# Topologically Sorted Source Nodes: [routing_3], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf22, buf23, 256, grid=grid(256), stream=stream0)
del buf22
buf24 = reinterpret_tensor(buf17, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf17 # reuse
# Topologically Sorted Source Nodes: [contiguous_6], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf18, primals_18, buf24, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_18
buf25 = reinterpret_tensor(buf18, (16, 4, 1), (4, 1, 1), 0); del buf18 # reuse
# Topologically Sorted Source Nodes: [o_2], Original ATen: [aten.bmm]
extern_kernels.bmm(buf23, reinterpret_tensor(buf24, (16, 4, 1), (4, 1, 0), 0), out=buf25)
buf26 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_7], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf25, buf26, 16, 4, grid=grid(16, 4), stream=stream0)
buf27 = reinterpret_tensor(buf25, (16, 4), (4, 1), 0); del buf25 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf26, (16, 4), (4, 1), 0), reinterpret_tensor(primals_19, (4, 4), (1, 4), 0), out=buf27)
buf28 = reinterpret_tensor(buf27, (4, 4, 4), (16, 4, 1), 0); del buf27 # reuse
# Topologically Sorted Source Nodes: [h_4], Original ATen: [aten.add]
triton_poi_fused_add_5.run(buf28, primals_20, primals_1, 64, grid=grid(64), stream=stream0)
del primals_20
buf29 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf28, (16, 4), (4, 1), 0), reinterpret_tensor(primals_21, (4, 4), (1, 4), 0), out=buf29)
buf30 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf31 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu_1, h_5], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_6.run(buf28, buf29, primals_22, buf30, buf31, 64, grid=grid(64), stream=stream0)
del buf29
del primals_22
return (buf30, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), buf8, reinterpret_tensor(buf11, (16, 4), (4, 1), 0), reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(buf16, (16, 4), (4, 1), 0), buf23, reinterpret_tensor(buf26, (16, 4), (4, 1), 0), reinterpret_tensor(buf28, (16, 4), (4, 1), 0), buf31, primals_21, primals_19, reinterpret_tensor(buf24, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf19, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf20, (16, 4, 1), (4, 1, 1), 0), primals_17, primals_15, buf32, primals_11, primals_9, reinterpret_tensor(buf9, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 1), 0), 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, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 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)
primals_7 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_16 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_17 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_18 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_19 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_20 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_21 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_22 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12, primals_13, primals_14, primals_15, primals_16, primals_17, primals_18, primals_19, primals_20, primals_21, primals_22])
return print_performance(fn, 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.functional as F
import torch.nn as nn
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
class ISAB(nn.Module):
def __init__(self, d, n_heads, n_inducing_points, layer_norm=False):
super().__init__()
self.mab0 = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
self.mab1 = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
self.I = nn.Parameter(torch.zeros(1, n_inducing_points, d),
requires_grad=True)
with torch.no_grad():
nn.init.xavier_uniform_(self.I)
def forward(self, x, presence=None):
batch_size = x.shape[0]
h = self.mab0(self.I.repeat(batch_size, 1, 1), x, presence)
return self.mab1(x, h)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'d': 4, 'n_heads': 4, 'n_inducing_points': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import math
import numpy as np
import torch.nn.functional as F
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_repeat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
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_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x1), xmask & ymask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + 16 * y0), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x1), xmask & ymask, eviction_policy
='evict_last')
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_5(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_6(in_ptr0, in_ptr1,
in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + x2, tmp6, xmask)
tl.store(out_ptr1 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12,
primals_13, primals_14, primals_15, primals_16, primals_17,
primals_18, primals_19, primals_20, primals_21, primals_22) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4), (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,))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4, 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, 4), (4, 1))
assert_size_stride(primals_16, (4,), (1,))
assert_size_stride(primals_17, (4, 4), (4, 1))
assert_size_stride(primals_18, (4,), (1,))
assert_size_stride(primals_19, (4, 4), (4, 1))
assert_size_stride(primals_20, (4,), (1,))
assert_size_stride(primals_21, (4, 4), (4, 1))
assert_size_stride(primals_22, (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_repeat_0[grid(64)](primals_2, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf2)
del primals_5
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf3)
del primals_7
buf4 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_1[grid(4, 16)](buf1, primals_4, buf4, 4, 16,
XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_4
buf5 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_clone_1[grid(4, 16)](buf2, primals_6, buf5, 4, 16,
XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_6
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf5, (16, 1, 4), (4, 0, 1), 0), out=buf6)
buf7 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(256)](buf6, buf7, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf8 = buf6
del buf6
triton_poi_fused__softmax_3[grid(256)](buf7, buf8, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf9 = reinterpret_tensor(buf2, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf2
triton_poi_fused_clone_1[grid(4, 16)](buf3, primals_8, buf9, 4, 16,
XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_8
buf10 = reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 1), 0)
del buf3
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (16, 4, 1), (4, 1,
0), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_4[grid(16, 4)](buf10, buf11, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf12 = reinterpret_tensor(buf10, (16, 4), (4, 1), 0)
del buf10
extern_kernels.mm(reinterpret_tensor(buf11, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf12)
buf13 = reinterpret_tensor(buf12, (4, 4, 4), (16, 4, 1), 0)
del buf12
triton_poi_fused_add_5[grid(64)](buf13, primals_10, buf0, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_10
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf14)
buf15 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_13, (4, 4), (1, 4), 0), out=buf15)
del primals_13
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf32 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_6[grid(64)](buf13,
buf14, primals_12, buf16, buf32, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del primals_12
buf17 = buf14
del buf14
extern_kernels.mm(reinterpret_tensor(buf16, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_15, (4, 4), (1, 4), 0), out=buf17)
buf18 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf16, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_17, (4, 4), (1, 4), 0), out=buf18)
buf19 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_1[grid(4, 16)](buf15, primals_14, buf19, 4,
16, XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_14
buf20 = reinterpret_tensor(buf15, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf15
triton_poi_fused_clone_1[grid(4, 16)](buf17, primals_16, buf20, 4,
16, XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_16
buf21 = buf7
del buf7
extern_kernels.bmm(reinterpret_tensor(buf19, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf20, (16, 1, 4), (4, 0, 1), 0), out=buf21)
buf22 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(256)](buf21, buf22, 256, XBLOCK=
128, num_warps=4, num_stages=1)
buf23 = buf21
del buf21
triton_poi_fused__softmax_3[grid(256)](buf22, buf23, 256, XBLOCK=
256, num_warps=4, num_stages=1)
del buf22
buf24 = reinterpret_tensor(buf17, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf17
triton_poi_fused_clone_1[grid(4, 16)](buf18, primals_18, buf24, 4,
16, XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_18
buf25 = reinterpret_tensor(buf18, (16, 4, 1), (4, 1, 1), 0)
del buf18
extern_kernels.bmm(buf23, reinterpret_tensor(buf24, (16, 4, 1), (4,
1, 0), 0), out=buf25)
buf26 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_4[grid(16, 4)](buf25, buf26, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf27 = reinterpret_tensor(buf25, (16, 4), (4, 1), 0)
del buf25
extern_kernels.mm(reinterpret_tensor(buf26, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_19, (4, 4), (1, 4), 0), out=buf27)
buf28 = reinterpret_tensor(buf27, (4, 4, 4), (16, 4, 1), 0)
del buf27
triton_poi_fused_add_5[grid(64)](buf28, primals_20, primals_1, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_20
buf29 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf28, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_21, (4, 4), (1, 4), 0), out=buf29)
buf30 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf31 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_6[grid(64)](buf28,
buf29, primals_22, buf30, buf31, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf29
del primals_22
return (buf30, reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), buf8,
reinterpret_tensor(buf11, (16, 4), (4, 1), 0), reinterpret_tensor(
buf13, (16, 4), (4, 1), 0), reinterpret_tensor(buf16, (16, 4), (4,
1), 0), buf23, reinterpret_tensor(buf26, (16, 4), (4, 1), 0),
reinterpret_tensor(buf28, (16, 4), (4, 1), 0), buf31, primals_21,
primals_19, reinterpret_tensor(buf24, (16, 1, 4), (4, 1, 1), 0),
reinterpret_tensor(buf19, (16, 1, 4), (4, 1, 1), 0),
reinterpret_tensor(buf20, (16, 4, 1), (4, 1, 1), 0), primals_17,
primals_15, buf32, primals_11, primals_9, reinterpret_tensor(buf9,
(16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 1, 4), (4,
1, 1), 0), reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 1), 0),
primals_3)
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
class ISABNew(nn.Module):
def __init__(self, d, n_heads, n_inducing_points, layer_norm=False):
super().__init__()
self.mab0 = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
self.mab1 = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
self.I = nn.Parameter(torch.zeros(1, n_inducing_points, d),
requires_grad=True)
with torch.no_grad():
nn.init.xavier_uniform_(self.I)
def forward(self, input_0):
primals_2 = self.I
primals_3 = self.mab0.mqkv.q_projector.weight
primals_4 = self.mab0.mqkv.q_projector.bias
primals_5 = self.mab0.mqkv.k_projector.weight
primals_6 = self.mab0.mqkv.k_projector.bias
primals_7 = self.mab0.mqkv.v_projector.weight
primals_8 = self.mab0.mqkv.v_projector.bias
primals_9 = self.mab0.mqkv.o_projector.weight
primals_10 = self.mab0.mqkv.o_projector.bias
primals_11 = self.mab0.fc.weight
primals_12 = self.mab0.fc.bias
primals_13 = self.mab1.mqkv.q_projector.weight
primals_14 = self.mab1.mqkv.q_projector.bias
primals_15 = self.mab1.mqkv.k_projector.weight
primals_16 = self.mab1.mqkv.k_projector.bias
primals_17 = self.mab1.mqkv.v_projector.weight
primals_18 = self.mab1.mqkv.v_projector.bias
primals_19 = self.mab1.mqkv.o_projector.weight
primals_20 = self.mab1.mqkv.o_projector.bias
primals_21 = self.mab1.fc.weight
primals_22 = self.mab1.fc.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12, primals_13, primals_14,
primals_15, primals_16, primals_17, primals_18, primals_19,
primals_20, primals_21, primals_22])
return output[0]
|
KohavTal/SCAE_Project
|
ISAB
| false
| 8,416
|
[
"Apache-2.0"
] | 40
|
bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
https://github.com/KohavTal/SCAE_Project/tree/bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
PositionWiseFeedForwardNetworks
|
# 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/sm/csmenfx32ekzp3jy4hjjyfdm37wlwf77smzilpa5revmsd7esfwf.py
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.relu, aten.threshold_backward]
# Source node to ATen node mapping:
# outputs => relu
# Graph fragment:
# %relu : [num_users=1] = 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 = (%view_6, 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
x4 = xindex
x0 = xindex % 4
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 + (x4), tmp6, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ox/coxzkq4kueijjjccjhdtvzhfquura4nj5lgusydc343b4la3telx.py
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.view]
# Source node to ATen node mapping:
# linear_1 => view_7
# Graph fragment:
# %view_7 : [num_users=2] = call_function[target=torch.ops.aten.reshape.default](args = (%view_6, [64, 4]), kwargs = {})
triton_poi_fused_view_1 = async_compile.triton('triton_poi_fused_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=[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_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_view_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 % 4
x1 = (xindex // 4)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x1) + (16*((x1 % 4) // 4)) + (64*(((4*((x1 // 4) % 4)) + (x1 % 4)) // 16))), xmask)
tl.store(out_ptr0 + (x2), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (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
buf4 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [outputs], Original ATen: [aten.relu, aten.threshold_backward]
stream0 = get_raw_stream(0)
triton_poi_fused_relu_threshold_backward_0.run(buf1, primals_2, buf4, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.view]
triton_poi_fused_view_1.run(buf1, buf2, 256, grid=grid(256), stream=stream0)
buf3 = reinterpret_tensor(buf1, (64, 4), (4, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, buf2, reinterpret_tensor(primals_4, (4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
return (reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf2, primals_4, buf4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((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
from torch.nn import functional as F
def Linear(in_features, out_features, bias=True):
m = nn.Linear(in_features, out_features, bias)
nn.init.xavier_uniform_(m.weight)
if bias:
nn.init.constant_(m.bias, 0.0)
return m
class PositionWiseFeedForwardNetworks(nn.Module):
def __init__(self, input_size, output_size, d_ff):
super().__init__()
self.W_1 = Linear(input_size, d_ff, bias=True)
self.W_2 = Linear(d_ff, output_size, bias=True)
nn.init.constant_(self.W_1.bias, 0.0)
nn.init.constant_(self.W_2.bias, 0.0)
def forward(self, input):
outputs = F.relu(self.W_1(input), inplace=True)
return self.W_2(outputs)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'input_size': 4, 'output_size': 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 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
x4 = xindex
x0 = xindex % 4
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 + x4, tmp6, xmask)
@triton.jit
def triton_poi_fused_view_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 % 4
x1 = xindex // 4
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x1 + 16 * (x1 % 4 // 4) + 64 * ((4 *
(x1 // 4 % 4) + x1 % 4) // 16)), xmask)
tl.store(out_ptr0 + x2, tmp0, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (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
buf4 = 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, buf4, 256, XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
triton_poi_fused_view_1[grid(256)](buf1, buf2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
buf3 = reinterpret_tensor(buf1, (64, 4), (4, 1), 0)
del buf1
extern_kernels.addmm(primals_5, buf2, reinterpret_tensor(primals_4,
(4, 4), (1, 4), 0), alpha=1, beta=1, out=buf3)
del primals_5
return reinterpret_tensor(buf3, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf2, primals_4, buf4
def Linear(in_features, out_features, bias=True):
m = nn.Linear(in_features, out_features, bias)
nn.init.xavier_uniform_(m.weight)
if bias:
nn.init.constant_(m.bias, 0.0)
return m
class PositionWiseFeedForwardNetworksNew(nn.Module):
def __init__(self, input_size, output_size, d_ff):
super().__init__()
self.W_1 = Linear(input_size, d_ff, bias=True)
self.W_2 = Linear(d_ff, output_size, bias=True)
nn.init.constant_(self.W_1.bias, 0.0)
nn.init.constant_(self.W_2.bias, 0.0)
def forward(self, input_0):
primals_1 = self.W_1.weight
primals_2 = self.W_1.bias
primals_4 = self.W_2.weight
primals_5 = self.W_2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
L-Zhe/FasySeq
|
PositionWiseFeedForwardNetworks
| false
| 8,417
|
[
"Apache-2.0"
] | 34
|
2cd2abd290666b1e118d8ad11c973b58ca4f0573
|
https://github.com/L-Zhe/FasySeq/tree/2cd2abd290666b1e118d8ad11c973b58ca4f0573
|
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/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.py
# Topologically Sorted Source Nodes: [se], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# se => 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: [se_1, se_2], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# se_1 => convolution
# se_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/x7/cx7evvvm7te22h7xf3yh7pnjatqie5vy54vyorfffrtctztd4wn5.py
# Topologically Sorted Source Nodes: [se_3], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# se_3 => 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/qj/cqjyghtzlqqgfsi5spfanbmvqxah56ghatqjdqw3zpkokadehk6j.py
# Topologically Sorted Source Nodes: [se_4, x], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# se_4 => sigmoid
# x => mul
# 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 = (%sigmoid, %primals_1), 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
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + (x2), xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + (x2), tmp3, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, 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: [se], 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: [se_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: [se_1, se_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: [se_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 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [se_3], 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: [se_4, x], Original ATen: [aten.sigmoid, aten.mul]
triton_poi_fused_mul_sigmoid_3.run(buf5, primals_1, buf6, 256, grid=grid(256), stream=stream0)
return (buf6, primals_1, primals_2, primals_4, buf1, 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((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
class SEBlock(torch.nn.Module):
def __init__(self, inplanes, redr, poolflag='avg'):
super(SEBlock, self).__init__()
if poolflag == 'max':
self.pool = torch.nn.AdaptiveMaxPool2d((1, 1))
if poolflag == 'avg':
self.pool = torch.nn.AdaptiveAvgPool2d((1, 1))
self.replanes = inplanes // redr
self.linear1 = torch.nn.Conv2d(inplanes, self.replanes, (1, 1),
padding=0)
self.relu = torch.nn.ReLU(inplace=True)
self.linear2 = torch.nn.Conv2d(self.replanes, inplanes, (1, 1),
padding=0)
self.sigmod = torch.nn.Sigmoid()
def forward(self, x):
se = self.pool(x)
se = self.linear1(se)
se = self.relu(se)
se = self.linear2(se)
se = self.sigmod(se)
x = se * x
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'inplanes': 4, 'redr': 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
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_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
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr1 + x2, xmask)
tmp1 = tl.sigmoid(tmp0)
tmp3 = tmp1 * tmp2
tl.store(out_ptr0 + x2, tmp3, xmask)
def call(args):
primals_1, primals_2, primals_3, 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 = 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)](buf5, primals_1, buf6,
256, XBLOCK=256, num_warps=4, num_stages=1)
return buf6, primals_1, primals_2, primals_4, buf1, buf3, buf5
class SEBlockNew(torch.nn.Module):
def __init__(self, inplanes, redr, poolflag='avg'):
super(SEBlockNew, self).__init__()
if poolflag == 'max':
self.pool = torch.nn.AdaptiveMaxPool2d((1, 1))
if poolflag == 'avg':
self.pool = torch.nn.AdaptiveAvgPool2d((1, 1))
self.replanes = inplanes // redr
self.linear1 = torch.nn.Conv2d(inplanes, self.replanes, (1, 1),
padding=0)
self.relu = torch.nn.ReLU(inplace=True)
self.linear2 = torch.nn.Conv2d(self.replanes, inplanes, (1, 1),
padding=0)
self.sigmod = torch.nn.Sigmoid()
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]
|
Knight825/models-pytorch
|
SEBlock
| false
| 8,418
|
[
"Apache-2.0"
] | 16
|
133559eebb8795d78a32fa44d49408d0c5167ae9
|
https://github.com/Knight825/models-pytorch/tree/133559eebb8795d78a32fa44d49408d0c5167ae9
|
FourierEmbedding
|
# 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/u5/cu5k3whvhlntwfwcahwmvbgdom6k5v4wiblsbncuovm3zy3ejx4w.py
# Topologically Sorted Source Nodes: [z], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# z => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%unsqueeze, %unsqueeze_1], 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=[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_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 = 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 = 2.0
tmp7 = tmp5 * tmp6
tmp8 = 0.3333333333333333
tmp9 = tmp7 * tmp8
tmp10 = 1.0
tmp11 = tmp9 - tmp10
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tmp15 = tl.full([1], 2, tl.int64)
tmp16 = tmp0 < tmp15
tmp17 = tl.load(in_ptr1 + (x1), tmp14 & xmask, eviction_policy='evict_last', other=0.0)
tmp18 = tmp17 * tmp6
tmp19 = tmp18 * tmp8
tmp20 = tmp19 - tmp10
tmp21 = tl.full(tmp20.shape, 0.0, tmp20.dtype)
tmp22 = tl.where(tmp14, tmp20, tmp21)
tmp23 = tl.where(tmp4, tmp13, tmp22)
tl.store(out_ptr0 + (x2), tmp23, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/66/c66eyweomhv45cfxhrdx7hpucqf4jqo4fas5ztzz5f6urmroruhe.py
# Topologically Sorted Source Nodes: [sin], Original ATen: [aten.sin]
# Source node to ATen node mapping:
# sin => sin
# Graph fragment:
# %sin : [num_users=1] = call_function[target=torch.ops.aten.sin.default](args = (%view_1,), kwargs = {})
triton_poi_fused_sin_1 = async_compile.triton('triton_poi_fused_sin_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_sin_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_sin_1(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl_math.sin(tmp0)
tl.store(out_ptr0 + (x0), tmp1, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 2), (2, 1))
assert_size_stride(primals_4, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
# Topologically Sorted Source Nodes: [z], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(primals_1, primals_2, buf0, 32, grid=grid(32), stream=stream0)
del primals_1
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_4, reinterpret_tensor(buf0, (16, 2), (2, 1), 0), reinterpret_tensor(primals_3, (2, 4), (1, 2), 0), alpha=1, beta=1, out=buf1)
del primals_3
del primals_4
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [sin], Original ATen: [aten.sin]
triton_poi_fused_sin_1.run(buf1, buf2, 64, grid=grid(64), stream=stream0)
return (buf2, reinterpret_tensor(buf0, (16, 2), (2, 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, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 2), (2, 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
from torch import nn
class FourierEmbedding(nn.Module):
def __init__(self, features, height, width, **kwargs):
super().__init__(**kwargs)
self.projector = nn.Linear(2, features)
self._height = height
self._width = width
def forward(self, y, x):
x_norm = 2 * x / (self._width - 1) - 1
y_norm = 2 * y / (self._height - 1) - 1
z = torch.cat((x_norm.unsqueeze(2), y_norm.unsqueeze(2)), dim=2)
return torch.sin(self.projector(z))
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'features': 4, 'height': 4, 'width': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import math as tl_math
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_cat_0(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 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 = 2.0
tmp7 = tmp5 * tmp6
tmp8 = 0.3333333333333333
tmp9 = tmp7 * tmp8
tmp10 = 1.0
tmp11 = tmp9 - tmp10
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tl.full([1], 2, tl.int64)
tmp17 = tl.load(in_ptr1 + x1, tmp14 & xmask, eviction_policy=
'evict_last', other=0.0)
tmp18 = tmp17 * tmp6
tmp19 = tmp18 * tmp8
tmp20 = tmp19 - tmp10
tmp21 = tl.full(tmp20.shape, 0.0, tmp20.dtype)
tmp22 = tl.where(tmp14, tmp20, tmp21)
tmp23 = tl.where(tmp4, tmp13, tmp22)
tl.store(out_ptr0 + x2, tmp23, xmask)
@triton.jit
def triton_poi_fused_sin_1(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl_math.sin(tmp0)
tl.store(out_ptr0 + x0, tmp1, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, 4), (4, 1))
assert_size_stride(primals_3, (4, 2), (2, 1))
assert_size_stride(primals_4, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 2), (8, 2, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(32)](primals_1, primals_2, buf0, 32,
XBLOCK=32, num_warps=1, num_stages=1)
del primals_1
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_4, reinterpret_tensor(buf0, (16, 2), (
2, 1), 0), reinterpret_tensor(primals_3, (2, 4), (1, 2), 0),
alpha=1, beta=1, out=buf1)
del primals_3
del primals_4
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_sin_1[grid(64)](buf1, buf2, 64, XBLOCK=64,
num_warps=1, num_stages=1)
return buf2, reinterpret_tensor(buf0, (16, 2), (2, 1), 0), buf1
class FourierEmbeddingNew(nn.Module):
def __init__(self, features, height, width, **kwargs):
super().__init__(**kwargs)
self.projector = nn.Linear(2, features)
self._height = height
self._width = width
def forward(self, input_0, input_1):
primals_3 = self.projector.weight
primals_4 = self.projector.bias
primals_1 = input_0
primals_2 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
LS4GAN/uvcgan
|
FourierEmbedding
| false
| 8,419
|
[
"BSD-2-Clause"
] | 20
|
376439ae2a9be684ff279ddf634fe137aadc5df5
|
https://github.com/LS4GAN/uvcgan/tree/376439ae2a9be684ff279ddf634fe137aadc5df5
|
Critic
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/pv/cpvq5akoxii6zri2lssiqurrqsl5nuea43ylpa6xzb433ugqbchi.py
# Topologically Sorted Source Nodes: [z], Original ATen: [aten.tanh]
# Source node to ATen node mapping:
# z => tanh
# Graph fragment:
# %tanh : [num_users=2] = call_function[target=torch.ops.aten.tanh.default](args = (%view_1,), kwargs = {})
triton_poi_fused_tanh_0 = async_compile.triton('triton_poi_fused_tanh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_tanh_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_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 4096
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + (x2), tmp3, None)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7 = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 64), (64, 1))
assert_size_stride(primals_5, (64, ), (1, ))
assert_size_stride(primals_6, (1, 64), (64, 1))
assert_size_stride(primals_7, (1, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [z], Original ATen: [aten.tanh]
stream0 = get_raw_stream(0)
triton_poi_fused_tanh_0.run(buf1, primals_2, 4096, grid=grid(4096), stream=stream0)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0), reinterpret_tensor(primals_4, (64, 64), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [z_1], Original ATen: [aten.tanh]
triton_poi_fused_tanh_0.run(buf3, primals_5, 4096, grid=grid(4096), stream=stream0)
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
# Topologically Sorted Source Nodes: [v], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64), (64, 1), 0), reinterpret_tensor(primals_6, (64, 1), (1, 64), 0), alpha=1, beta=1, out=buf5)
del primals_7
return (reinterpret_tensor(buf5, (4, 4, 4, 1), (16, 4, 1, 1), 0), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), buf1, buf3, primals_6, primals_4, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((64, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((64, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((1, 64), (64, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((1, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class Critic(nn.Module):
def __init__(self, state_dim, hidden_dim=64):
super(Critic, self).__init__()
self.l1 = nn.Linear(state_dim, hidden_dim)
self.l2 = nn.Linear(hidden_dim, hidden_dim)
self.l3 = nn.Linear(hidden_dim, 1)
def forward(self, state):
z = torch.tanh(self.l1(state))
z = torch.tanh(self.l2(z))
v = self.l3(z)
return v
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'state_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_tanh_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr
):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 64
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = libdevice.tanh(tmp2)
tl.store(in_out_ptr0 + x2, tmp3, None)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7) = args
args.clear()
assert_size_stride(primals_1, (64, 4), (4, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (64, 64), (64, 1))
assert_size_stride(primals_5, (64,), (1,))
assert_size_stride(primals_6, (1, 64), (64, 1))
assert_size_stride(primals_7, (1,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 64), (1, 4), 0), out=buf0)
del primals_1
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_tanh_0[grid(4096)](buf1, primals_2, 4096, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_2
buf2 = empty_strided_cuda((64, 64), (64, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf1, (64, 64), (64, 1), 0),
reinterpret_tensor(primals_4, (64, 64), (1, 64), 0), out=buf2)
buf3 = reinterpret_tensor(buf2, (4, 4, 4, 64), (1024, 256, 64, 1), 0)
del buf2
triton_poi_fused_tanh_0[grid(4096)](buf3, primals_5, 4096, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_5
buf5 = empty_strided_cuda((64, 1), (1, 1), torch.float32)
extern_kernels.addmm(primals_7, reinterpret_tensor(buf3, (64, 64),
(64, 1), 0), reinterpret_tensor(primals_6, (64, 1), (1, 64), 0),
alpha=1, beta=1, out=buf5)
del primals_7
return reinterpret_tensor(buf5, (4, 4, 4, 1), (16, 4, 1, 1), 0
), reinterpret_tensor(primals_3, (64, 4), (4, 1), 0
), buf1, buf3, primals_6, primals_4
class CriticNew(nn.Module):
def __init__(self, state_dim, hidden_dim=64):
super(CriticNew, self).__init__()
self.l1 = nn.Linear(state_dim, hidden_dim)
self.l2 = nn.Linear(hidden_dim, hidden_dim)
self.l3 = nn.Linear(hidden_dim, 1)
def forward(self, input_0):
primals_1 = self.l1.weight
primals_2 = self.l1.bias
primals_4 = self.l2.weight
primals_5 = self.l2.bias
primals_6 = self.l3.weight
primals_7 = self.l3.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7])
return output[0]
|
LQNew/LWDRL
|
Critic
| false
| 8,420
|
[
"MIT"
] | 11
|
0e4fab077a0cfbd27590b840557f4fda033c74ff
|
https://github.com/LQNew/LWDRL/tree/0e4fab077a0cfbd27590b840557f4fda033c74ff
|
PMA
|
# 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/clieqwiqohdxw3rro7yubltcvmtjwclpy4jznpytei347h6na5r4.py
# Topologically Sorted Source Nodes: [repeat], Original ATen: [aten.repeat]
# Source node to ATen node mapping:
# repeat => repeat
# Graph fragment:
# %repeat : [num_users=2] = call_function[target=torch.ops.aten.repeat.default](args = (%primals_2, [4, 1, 1]), kwargs = {})
triton_poi_fused_repeat_0 = async_compile.triton('triton_poi_fused_repeat_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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_repeat_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_repeat_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
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/ls/clscue37cenpwdtzqux56gb6cc4oukj57bzuosuaav5mip3j73tk.py
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous => clone
# Graph fragment:
# %clone : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_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=[4, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (4*x1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + (16*y0)), tmp2, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kz/ckzqylporms4fvgcrqg44ypprwpanp6hf222rji24wskr3b44aga.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => amax, exp, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%bmm, [-1], True), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%bmm, %amax), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%sub,), kwargs = {})
triton_poi_fused__softmax_2 = async_compile.triton('triton_poi_fused__softmax_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + (x2), tmp9, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/4d/c4dndrlfjcamjfnn3ng5agjc3ahefdgw6jcsnn6hm4ljwpbfbe7h.py
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
# Source node to ATen node mapping:
# routing_1 => div_1, sum_1
# Graph fragment:
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%exp, [-1], True), kwargs = {})
# %div_1 : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%exp, %sum_1), kwargs = {})
triton_poi_fused__softmax_3 = async_compile.triton('triton_poi_fused__softmax_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (4*x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + (4*x1)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + (4*x1)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + (x2), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wg/cwgjlocuqhpaklaousvfvznecyut7vf6pbamkv5aenho7h5t5xko.py
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# contiguous_3 => clone_3
# Graph fragment:
# %clone_3 : [num_users=1] = call_function[target=torch.ops.aten.clone.default](args = (%permute_7,), kwargs = {memory_format: torch.contiguous_format})
triton_poi_fused_clone_4 = async_compile.triton('triton_poi_fused_clone_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16, 4], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (16*x1)), xmask & ymask, eviction_policy='evict_last')
tl.store(out_ptr0 + (x1 + (4*y0)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mu/cmu55cc2rrop6dujmkmg24kdnnkehbytpnejw6dzpsfvbjwmug5x.py
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
# Source node to ATen node mapping:
# h_1 => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_21, %repeat), kwargs = {})
triton_poi_fused_add_5 = async_compile.triton('triton_poi_fused_add_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_5', '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_5(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + (x2), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/6j/c6jhrcgdwigdocqbjr3po3tf6sp6t6udjnhf23dicfmtlnqloh6u.py
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
# Source node to ATen node mapping:
# h_2 => add_1
# relu => relu
# Graph fragment:
# %relu : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%view_23,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %relu), kwargs = {})
# %le : [num_users=1] = call_function[target=torch.ops.aten.le.Scalar](args = (%relu, 0), kwargs = {})
triton_poi_fused_add_relu_threshold_backward_6 = async_compile.triton('triton_poi_fused_add_relu_threshold_backward_6', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*i1', 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_add_relu_threshold_backward_6', '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_relu_threshold_backward_6(in_ptr0, in_ptr1, in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x2), xmask)
tmp2 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + (x2), tmp6, xmask)
tl.store(out_ptr1 + (x2), tmp8, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4), (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, ))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4, ), (1, ))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4, ), (1, ))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (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: [repeat], Original ATen: [aten.repeat]
stream0 = get_raw_stream(0)
triton_poi_fused_repeat_0.run(primals_2, buf0, 64, grid=grid(64), stream=stream0)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((16, 4), (4, 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, 4), (1, 4), 0), out=buf2)
del primals_5
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf3)
del primals_7
buf4 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf1, primals_4, buf4, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_4
buf5 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf1 # reuse
# Topologically Sorted Source Nodes: [contiguous_1], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf2, primals_6, buf5, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_6
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 0), 0), reinterpret_tensor(buf5, (16, 1, 4), (4, 0, 1), 0), out=buf6)
buf7 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_2.run(buf6, buf7, 256, grid=grid(256), stream=stream0)
buf8 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [routing_1], Original ATen: [aten._softmax]
triton_poi_fused__softmax_3.run(buf7, buf8, 256, grid=grid(256), stream=stream0)
del buf7
buf9 = reinterpret_tensor(buf2, (4, 4, 4, 1), (16, 4, 1, 1), 0); del buf2 # reuse
# Topologically Sorted Source Nodes: [contiguous_2], Original ATen: [aten.clone]
triton_poi_fused_clone_1.run(buf3, primals_8, buf9, 4, 16, grid=grid(4, 16), stream=stream0)
del primals_8
buf10 = reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 1), 0); del buf3 # reuse
# Topologically Sorted Source Nodes: [o], Original ATen: [aten.bmm]
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (16, 4, 1), (4, 1, 0), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous_3], Original ATen: [aten.clone]
triton_poi_fused_clone_4.run(buf10, buf11, 16, 4, grid=grid(16, 4), stream=stream0)
buf12 = reinterpret_tensor(buf10, (16, 4), (4, 1), 0); del buf10 # reuse
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf11, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf12)
buf13 = reinterpret_tensor(buf12, (4, 4, 4), (16, 4, 1), 0); del buf12 # reuse
# Topologically Sorted Source Nodes: [h_1], Original ATen: [aten.add]
triton_poi_fused_add_5.run(buf13, primals_10, buf0, 64, grid=grid(64), stream=stream0)
del primals_10
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0), reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf14)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [relu, h_2], Original ATen: [aten.relu, aten.add, aten.threshold_backward]
triton_poi_fused_add_relu_threshold_backward_6.run(buf13, buf14, primals_12, buf15, buf16, 64, grid=grid(64), stream=stream0)
del buf14
del primals_12
return (buf15, reinterpret_tensor(buf0, (16, 4), (4, 1), 0), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), buf8, reinterpret_tensor(buf11, (16, 4), (4, 1), 0), reinterpret_tensor(buf13, (16, 4), (4, 1), 0), buf16, primals_11, primals_9, reinterpret_tensor(buf9, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 1, 1), 0), reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 1), 0), 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, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 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)
primals_7 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_12 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11, primals_12])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import math
import torch
import numpy as np
import torch.nn.functional as F
import torch.nn as nn
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
class PMA(nn.Module):
def __init__(self, d, n_heads, n_seeds, layer_norm=False):
super().__init__()
self.mab = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
self.S = nn.Parameter(torch.zeros(1, n_seeds, d), requires_grad=True)
with torch.no_grad():
nn.init.xavier_uniform_(self.S)
def forward(self, x, presence=None):
batch_size = x.shape[0]
return self.mab(self.S.repeat(batch_size, 1, 1), x, presence)
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'d': 4, 'n_heads': 4, 'n_seeds': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import math as tl_math
import math
import numpy as np
import torch.nn.functional as F
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_repeat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
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_clone_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel,
YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 4
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 4 * x1), xmask & ymask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x1 + 16 * y0), tmp2, xmask & ymask)
@triton.jit
def triton_poi_fused__softmax_2(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tmp9 = tl_math.exp(tmp8)
tl.store(out_ptr0 + x2, tmp9, xmask)
@triton.jit
def triton_poi_fused__softmax_3(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr
):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x1 = xindex // 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + 4 * x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (1 + 4 * x1), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (2 + 4 * x1), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (3 + 4 * x1), xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp5 = tmp3 + tmp4
tmp7 = tmp5 + tmp6
tmp8 = tmp0 / tmp7
tl.store(out_ptr0 + x2, tmp8, xmask)
@triton.jit
def triton_poi_fused_clone_4(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
ynumel = 16
xnumel = 4
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x1 = xindex
y0 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 16 * x1), xmask & ymask, eviction_policy
='evict_last')
tl.store(out_ptr0 + (x1 + 4 * y0), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_add_5(in_out_ptr0, in_ptr0, in_ptr1, xnumel, XBLOCK:
tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tl.store(in_out_ptr0 + x2, tmp4, xmask)
@triton.jit
def triton_poi_fused_add_relu_threshold_backward_6(in_ptr0, in_ptr1,
in_ptr2, 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 % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x2, xmask)
tmp2 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp3 = tmp1 + tmp2
tmp4 = tl.full([1], 0, tl.int32)
tmp5 = triton_helpers.maximum(tmp4, tmp3)
tmp6 = tmp0 + tmp5
tmp7 = 0.0
tmp8 = tmp5 <= tmp7
tl.store(out_ptr0 + x2, tmp6, xmask)
tl.store(out_ptr1 + x2, tmp8, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11, primals_12
) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (1, 4, 4), (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,))
assert_size_stride(primals_7, (4, 4), (4, 1))
assert_size_stride(primals_8, (4,), (1,))
assert_size_stride(primals_9, (4, 4), (4, 1))
assert_size_stride(primals_10, (4,), (1,))
assert_size_stride(primals_11, (4, 4), (4, 1))
assert_size_stride(primals_12, (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_repeat_0[grid(64)](primals_2, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_2
buf1 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf0, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_3, (4, 4), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_5, (4, 4), (1, 4), 0), out=buf2)
del primals_5
buf3 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_1, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_7, (4, 4), (1, 4), 0), out=buf3)
del primals_7
buf4 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_1[grid(4, 16)](buf1, primals_4, buf4, 4, 16,
XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_4
buf5 = reinterpret_tensor(buf1, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf1
triton_poi_fused_clone_1[grid(4, 16)](buf2, primals_6, buf5, 4, 16,
XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_6
buf6 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf4, (16, 4, 1), (4, 1, 0),
0), reinterpret_tensor(buf5, (16, 1, 4), (4, 0, 1), 0), out=buf6)
buf7 = empty_strided_cuda((16, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused__softmax_2[grid(256)](buf6, buf7, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf8 = buf6
del buf6
triton_poi_fused__softmax_3[grid(256)](buf7, buf8, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del buf7
buf9 = reinterpret_tensor(buf2, (4, 4, 4, 1), (16, 4, 1, 1), 0)
del buf2
triton_poi_fused_clone_1[grid(4, 16)](buf3, primals_8, buf9, 4, 16,
XBLOCK=16, YBLOCK=4, num_warps=1, num_stages=1)
del primals_8
buf10 = reinterpret_tensor(buf3, (16, 4, 1), (4, 1, 1), 0)
del buf3
extern_kernels.bmm(buf8, reinterpret_tensor(buf9, (16, 4, 1), (4, 1,
0), 0), out=buf10)
buf11 = empty_strided_cuda((4, 4, 4, 1), (16, 4, 1, 1), torch.float32)
triton_poi_fused_clone_4[grid(16, 4)](buf10, buf11, 16, 4, XBLOCK=4,
YBLOCK=16, num_warps=1, num_stages=1)
buf12 = reinterpret_tensor(buf10, (16, 4), (4, 1), 0)
del buf10
extern_kernels.mm(reinterpret_tensor(buf11, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_9, (4, 4), (1, 4), 0), out=buf12)
buf13 = reinterpret_tensor(buf12, (4, 4, 4), (16, 4, 1), 0)
del buf12
triton_poi_fused_add_5[grid(64)](buf13, primals_10, buf0, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del primals_10
buf14 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(buf13, (16, 4), (4, 1), 0),
reinterpret_tensor(primals_11, (4, 4), (1, 4), 0), out=buf14)
buf15 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
buf16 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.bool)
triton_poi_fused_add_relu_threshold_backward_6[grid(64)](buf13,
buf14, primals_12, buf15, buf16, 64, XBLOCK=64, num_warps=1,
num_stages=1)
del buf14
del primals_12
return buf15, reinterpret_tensor(buf0, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), buf8, reinterpret_tensor(buf11, (16, 4), (4, 1), 0
), reinterpret_tensor(buf13, (16, 4), (4, 1), 0
), buf16, primals_11, primals_9, reinterpret_tensor(buf9, (16, 1, 4
), (4, 1, 1), 0), reinterpret_tensor(buf4, (16, 1, 4), (4, 1, 1), 0
), reinterpret_tensor(buf5, (16, 4, 1), (4, 1, 1), 0), primals_3
def qkv_attention(queries, keys, values, presence=None):
"""
Transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
d_k = queries.shape[-1]
routing = torch.matmul(queries, keys.transpose(1, 2))
if presence is not None:
routing -= (1.0 - presence.unsqueeze(-2)) * 1e+32
routing = F.softmax(routing / np.sqrt(d_k), -1)
return torch.matmul(routing, values)
class MultiHeadQKVAttention(nn.Module):
"""Multi-head version of Transformer-like attention."""
def __init__(self, d_k, d_v, n_heads):
super().__init__()
self.d_k = d_k
self.d_v = d_v
self.n_heads = n_heads
d_k_p = int(math.ceil(d_k / n_heads)) * n_heads
d_v_p = int(math.ceil(d_v / n_heads)) * n_heads
self.q_projector = nn.Linear(d_k, d_k_p)
self.k_projector = nn.Linear(d_k, d_k_p)
self.v_projector = nn.Linear(d_v, d_v_p)
self.o_projector = nn.Linear(d_v_p, d_v)
def forward(self, queries, keys, values, presence=None):
"""
Multi-head transformer-like self-attention.
Args:
queries: Tensor of shape [B, N, d_k].
keys: Tensor of shape [B, M, d_k].
values: : Tensor of shape [B, M, d_v].
presence: None or tensor of shape [B, M].
Returns:
Tensor of shape [B, N, d_v]
"""
assert queries.shape[2] == keys.shape[2]
assert keys.shape[1] == values.shape[1]
if presence is not None:
assert values.shape[:2] == presence.shape
B, N, _d_k = queries.shape
M, _d_v = values.shape[1:]
H = self.n_heads
q_p = self.q_projector(queries)
k_p = self.k_projector(keys)
v_p = self.v_projector(values)
del queries, keys, values
q = q_p.view(B, N, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, N, -1)
k = k_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
v = v_p.view(B, M, H, -1).permute(2, 0, 1, 3).contiguous().view(H *
B, M, -1)
if presence is not None:
presence = presence.repeat(self.n_heads, 1)
o = qkv_attention(q, k, v, presence)
o = o.view(H, B, N, -1).permute(1, 2, 0, 3).contiguous().view(B, N, -1)
return self.o_projector(o)
class MAB(nn.Module):
def __init__(self, d, n_heads, layer_norm=False):
super().__init__()
self.layer_norm = layer_norm
self.mqkv = MultiHeadQKVAttention(d_k=d, d_v=d, n_heads=n_heads)
if layer_norm:
self.ln0 = nn.LayerNorm(d)
self.ln1 = nn.LayerNorm(d)
self.fc = nn.Linear(d, d)
def forward(self, queries, keys, presence=None):
h = self.mqkv(queries, keys, keys, presence)
h = h + queries
if presence is not None:
assert presence.shape[1] == queries.shape[1] == keys.shape[1]
h = h * presence.unsqueeze(-1)
if self.layer_norm:
h = self.ln0(h)
h = h + F.relu(self.fc(h))
if self.layer_norm:
h = self.ln1(h)
return h
class PMANew(nn.Module):
def __init__(self, d, n_heads, n_seeds, layer_norm=False):
super().__init__()
self.mab = MAB(d=d, n_heads=n_heads, layer_norm=layer_norm)
self.S = nn.Parameter(torch.zeros(1, n_seeds, d), requires_grad=True)
with torch.no_grad():
nn.init.xavier_uniform_(self.S)
def forward(self, input_0):
primals_2 = self.S
primals_3 = self.mab.mqkv.q_projector.weight
primals_4 = self.mab.mqkv.q_projector.bias
primals_5 = self.mab.mqkv.k_projector.weight
primals_6 = self.mab.mqkv.k_projector.bias
primals_7 = self.mab.mqkv.v_projector.weight
primals_8 = self.mab.mqkv.v_projector.bias
primals_9 = self.mab.mqkv.o_projector.weight
primals_10 = self.mab.mqkv.o_projector.bias
primals_11 = self.mab.fc.weight
primals_12 = self.mab.fc.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11, primals_12])
return output[0]
|
KohavTal/SCAE_Project
|
PMA
| false
| 8,421
|
[
"Apache-2.0"
] | 40
|
bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
https://github.com/KohavTal/SCAE_Project/tree/bc6d1c3697fcb9327dd96e9657c3299b47cf355e
|
MeanMap
|
# 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/vu/cvuk26doyygvttg2zp55bryzzvez7ugdrqavksjocjhjxdz7aa6d.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 = (%arg0_1, [1]), kwargs = {})
triton_poi_fused_mean_0 = async_compile.triton('triton_poi_fused_mean_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_mean_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_mean_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 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 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + (x2), 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), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.mean]
stream0 = get_raw_stream(0)
triton_poi_fused_mean_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.autograd
class MeanMap(nn.Module):
"""
Compute vanilla mean on a 4D tensor. This acts as a standard PyTorch layer.
The Mean is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) Recommend a tensor with only positive values. (After a ReLU)
Any real value will work.
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self):
super(MeanMap, self).__init__()
def forward(self, x):
assert torch.is_tensor(x), 'input must be a Torch Tensor'
assert len(x.size()) > 2, 'input must have at least three dims'
x = torch.mean(x, dim=1)
return 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
import torch.autograd
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_mean_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 4.0
tmp8 = tmp6 / tmp7
tl.store(out_ptr0 + x2, 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), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mean_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class MeanMapNew(nn.Module):
"""
Compute vanilla mean on a 4D tensor. This acts as a standard PyTorch layer.
The Mean is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) Recommend a tensor with only positive values. (After a ReLU)
Any real value will work.
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self):
super(MeanMapNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
MeanMap
| false
| 8,422
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
SMOEScaleMap
|
# 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/qk/cqkolwpncvzra6vqtv4vy7f7amxmwyh2t7od3iftyjt7eyliy2ci.py
# Topologically Sorted Source Nodes: [x, m, log2, log2_1, mean_1, k, th], Original ATen: [aten.add, aten.mean, aten.log2, aten.sub, aten.mul]
# Source node to ATen node mapping:
# k => sub
# log2 => log2
# log2_1 => log2_1
# m => mean
# mean_1 => mean_1
# th => mul
# x => add
# Graph fragment:
# %add : [num_users=2] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=2] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log2 : [num_users=1] = call_function[target=torch.ops.aten.log2.default](args = (%mean,), kwargs = {})
# %log2_1 : [num_users=1] = call_function[target=torch.ops.aten.log2.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log2_1, [1]), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log2, %mean_1), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, %mean), kwargs = {})
triton_poi_fused_add_log2_mean_mul_sub_0 = async_compile.triton('triton_poi_fused_add_log2_mean_mul_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_log2_mean_mul_sub_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_log2_mean_mul_sub_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp1 = 1e-07
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = tmp2 + tmp4
tmp7 = tmp6 + tmp1
tmp8 = tmp5 + tmp7
tmp10 = tmp9 + tmp1
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = libdevice.log2(tmp13)
tmp15 = libdevice.log2(tmp2)
tmp16 = libdevice.log2(tmp4)
tmp17 = tmp15 + tmp16
tmp18 = libdevice.log2(tmp7)
tmp19 = tmp17 + tmp18
tmp20 = libdevice.log2(tmp10)
tmp21 = tmp19 + tmp20
tmp22 = tmp21 / tmp12
tmp23 = tmp14 - tmp22
tmp24 = tmp23 * tmp13
tl.store(out_ptr0 + (x2), tmp24, 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: [x, m, log2, log2_1, mean_1, k, th], Original ATen: [aten.add, aten.mean, aten.log2, aten.sub, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_add_log2_mean_mul_sub_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.autograd
class SMOEScaleMap(nn.Module):
"""
Compute SMOE Scale on a 4D tensor. This acts as a standard PyTorch layer.
SMOE Scale is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) A tensor with only positive values. (After a ReLU)
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self, run_relu=False):
super(SMOEScaleMap, self).__init__()
"""
SMOE Scale must take in values > 0. Optionally, we can run a ReLU to do that.
"""
if run_relu:
self.relu = nn.ReLU(inplace=False)
else:
self.relu = None
def forward(self, x):
assert torch.is_tensor(x), 'input must be a Torch Tensor'
assert len(x.size()) > 2, 'input must have at least three dims'
"""
If we do not have a convenient ReLU to pluck from, we can do it here
"""
if self.relu is not None:
x = self.relu(x)
"""
avoid log(0)
"""
x = x + 1e-07
"""
This is one form. We can also use the log only form.
"""
m = torch.mean(x, dim=1)
k = torch.log2(m) - torch.mean(torch.log2(x), dim=1)
th = k * m
return th
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
import torch.autograd
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_log2_mean_mul_sub_0(in_ptr0, out_ptr0, xnumel,
XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp1 = 1e-07
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = tmp2 + tmp4
tmp7 = tmp6 + tmp1
tmp8 = tmp5 + tmp7
tmp10 = tmp9 + tmp1
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = libdevice.log2(tmp13)
tmp15 = libdevice.log2(tmp2)
tmp16 = libdevice.log2(tmp4)
tmp17 = tmp15 + tmp16
tmp18 = libdevice.log2(tmp7)
tmp19 = tmp17 + tmp18
tmp20 = libdevice.log2(tmp10)
tmp21 = tmp19 + tmp20
tmp22 = tmp21 / tmp12
tmp23 = tmp14 - tmp22
tmp24 = tmp23 * tmp13
tl.store(out_ptr0 + x2, tmp24, 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_log2_mean_mul_sub_0[grid(64)](arg0_1, buf0, 64,
XBLOCK=64, num_warps=1, num_stages=1)
del arg0_1
return buf0,
class SMOEScaleMapNew(nn.Module):
"""
Compute SMOE Scale on a 4D tensor. This acts as a standard PyTorch layer.
SMOE Scale is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) A tensor with only positive values. (After a ReLU)
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self, run_relu=False):
super(SMOEScaleMapNew, self).__init__()
"""
SMOE Scale must take in values > 0. Optionally, we can run a ReLU to do that.
"""
if run_relu:
self.relu = nn.ReLU(inplace=False)
else:
self.relu = None
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
SMOEScaleMap
| false
| 8,423
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
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/ol/coljqhqn2ngky5ed74qawwhwl656ahahxt5nc7rr3rlliikrfrsw.py
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# weight => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, 0.1767766952966369), 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 = 0.1767766952966369
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: [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, %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: [weight], 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: [conv2d], 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: [conv2d], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf2, primals_2, 16, grid=grid(16), stream=stream0)
del primals_2
return (buf2, buf0, 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 torch
import torch.nn as nn
from math import sqrt
def equal_lr(module, name='weight'):
EqualLR.apply(module, name)
return module
class EqualLR:
def __init__(self, name):
self.name = name
def compute_weight(self, module):
weight = getattr(module, self.name + '_orig')
fan_in = weight.data.size(1) * weight.data[0][0].numel()
return weight * sqrt(2 / fan_in)
@staticmethod
def apply(module, name):
fn = EqualLR(name)
weight = getattr(module, name)
del module._parameters[name]
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
module.register_forward_pre_hook(fn)
return fn
def __call__(self, module, input):
weight = self.compute_weight(module)
setattr(module, self.name, weight)
class EqualConv2d(nn.Module):
def __init__(self, *args, **kwargs):
super().__init__()
conv = nn.Conv2d(*args, **kwargs)
conv.weight.data.normal_()
conv.bias.data.zero_()
self.conv = equal_lr(conv)
def forward(self, input):
return self.conv(input)
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
import torch.nn as nn
from math import sqrt
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 = 0.1767766952966369
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=256,
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, buf0, primals_3, buf0
def equal_lr(module, name='weight'):
EqualLR.apply(module, name)
return module
class EqualLR:
def __init__(self, name):
self.name = name
def compute_weight(self, module):
weight = getattr(module, self.name + '_orig')
fan_in = weight.data.size(1) * weight.data[0][0].numel()
return weight * sqrt(2 / fan_in)
@staticmethod
def apply(module, name):
fn = EqualLR(name)
weight = getattr(module, name)
del module._parameters[name]
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
module.register_forward_pre_hook(fn)
return fn
def __call__(self, module, input):
weight = self.compute_weight(module)
setattr(module, self.name, weight)
class EqualConv2dNew(nn.Module):
def __init__(self, *args, **kwargs):
super().__init__()
conv = nn.Conv2d(*args, **kwargs)
conv.weight.data.normal_()
conv.bias.data.zero_()
self.conv = equal_lr(conv)
def forward(self, input_0):
primals_2 = self.conv.bias
primals_1 = self.conv.weight_orig
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
KwonGihyun/DiagonalGAN
|
EqualConv2d
| false
| 8,424
|
[
"MIT"
] | 13
|
9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
https://github.com/KwonGihyun/DiagonalGAN/tree/9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
SE
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/yg/cygooswl5gkxugqq2ejgag2vtcqhtumn2j3notsgzty3xoxbrq4v.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.mean]
# Source node to ATen node mapping:
# out => 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/bq/cbqs25ilz2wluqikclvslctdlvul4rcbcdk3m2pcvrxffj6hdrw5.py
# Topologically Sorted Source Nodes: [conv2d, out_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# conv2d => convolution
# out_1 => 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=[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: [conv2d_1], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv2d_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: [out_2, out_3], Original ATen: [aten.sigmoid, aten.mul]
# Source node to ATen node mapping:
# out_2 => sigmoid
# out_3 => mul
# 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, 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: [out], 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: [conv2d], 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, 4, 1, 1), (4, 1, 1, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [conv2d, out_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: [conv2d_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: [conv2d_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: [out_2, out_3], 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, buf1, 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
import torch.nn as nn
import torch.nn.functional as F
class SE(nn.Module):
"""Squeeze-and-Excitation block."""
def __init__(self, in_planes, se_planes):
super(SE, self).__init__()
self.se1 = nn.Conv2d(in_planes, se_planes, kernel_size=1, bias=True)
self.se2 = nn.Conv2d(se_planes, in_planes, kernel_size=1, bias=True)
def forward(self, x):
out = F.adaptive_avg_pool2d(x, (1, 1))
out = F.relu(self.se1(out))
out = self.se2(out).sigmoid()
out = x * out
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_planes': 4, 'se_planes': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_mean_0(in_out_ptr0, in_ptr0, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 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, 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, 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, buf1, buf3, buf5
class SENew(nn.Module):
"""Squeeze-and-Excitation block."""
def __init__(self, in_planes, se_planes):
super(SENew, self).__init__()
self.se1 = nn.Conv2d(in_planes, se_planes, kernel_size=1, bias=True)
self.se2 = nn.Conv2d(se_planes, in_planes, kernel_size=1, bias=True)
def forward(self, input_0):
primals_2 = self.se1.weight
primals_3 = self.se1.bias
primals_4 = self.se2.weight
primals_5 = self.se2.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
LIJUNYI95/SuperAdam
|
SE
| false
| 8,425
|
[
"MIT"
] | 14
|
00fc8a4d90bd037ccb9b871fbc64482818457b93
|
https://github.com/LIJUNYI95/SuperAdam/tree/00fc8a4d90bd037ccb9b871fbc64482818457b93
|
StdMap
|
# 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/ym/cymrn47gljfk23csjcqwzae3btwxkqjopyc6asz7i6pkmx2y77pp.py
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.std]
# Source node to ATen node mapping:
# x => sqrt, var
# Graph fragment:
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%arg0_1, [1]), kwargs = {correction: 1.0})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%var,), kwargs = {})
triton_poi_fused_std_0 = async_compile.triton('triton_poi_fused_std_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_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_std_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_std_0(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp2 = 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 = 3.0
tmp21 = tmp19 / tmp20
tmp22 = libdevice.sqrt(tmp21)
tl.store(out_ptr0 + (x2), tmp22, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.std]
stream0 = get_raw_stream(0)
triton_poi_fused_std_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.autograd
class StdMap(nn.Module):
"""
Compute vanilla standard deviation on a 4D tensor. This acts as a standard PyTorch layer.
Standard Deviation is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) Recommend a tensor with only positive values. (After a ReLU)
Any real value will work.
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self):
super(StdMap, self).__init__()
def forward(self, x):
assert torch.is_tensor(x), 'input must be a Torch Tensor'
assert len(x.size()) > 2, 'input must have at least three dims'
x = torch.std(x, dim=1)
return 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 torch.nn as nn
import torch.autograd
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_std_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp2 = 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 = 3.0
tmp21 = tmp19 / tmp20
tmp22 = libdevice.sqrt(tmp21)
tl.store(out_ptr0 + x2, tmp22, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_std_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class StdMapNew(nn.Module):
"""
Compute vanilla standard deviation on a 4D tensor. This acts as a standard PyTorch layer.
Standard Deviation is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) Recommend a tensor with only positive values. (After a ReLU)
Any real value will work.
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self):
super(StdMapNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
StdMap
| false
| 8,426
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
RangeNorm2D
|
# 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/y6/cy6wrdk7w3v6irqdpgj5h7c5osop4flxcgmudzw6yf7soycxgc4a.py
# Topologically Sorted Source Nodes: [max_1, min_1, nval, range_1, gt, eps_mat, range_2, x_1], Original ATen: [aten.max, aten.min, aten.sub, aten.gt, aten.add, aten.where, aten.div]
# Source node to ATen node mapping:
# eps_mat => full_default
# gt => gt
# max_1 => max_1
# min_1 => min_1
# nval => sub
# range_1 => sub_1
# range_2 => where
# x_1 => div
# Graph fragment:
# %max_1 : [num_users=1] = call_function[target=torch.ops.aten.max.dim](args = (%view, 1), kwargs = {})
# %min_1 : [num_users=1] = call_function[target=torch.ops.aten.min.dim](args = (%view, 1), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view, %view_2), kwargs = {})
# %sub_1 : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view_1, %view_2), kwargs = {})
# %gt : [num_users=1] = call_function[target=torch.ops.aten.gt.Scalar](args = (%sub_1, 1e-09), kwargs = {})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([4, 1], 9.999999717180685e-10), 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 = (%gt, %sub_1, %full_default), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %where), kwargs = {})
triton_per_fused_add_div_gt_max_min_sub_where_0 = async_compile.triton('triton_per_fused_add_div_gt_max_min_sub_where_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 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, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_gt_max_min_sub_where_0', '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_add_div_gt_max_min_sub_where_0(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, float("-inf"))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp6 = tl.where(xmask, tmp1, float("inf"))
tmp7 = triton_helpers.min2(tmp6, 1)[:, None]
tmp8 = tmp0 - tmp7
tmp9 = tmp4 - tmp7
tmp10 = 1e-09
tmp11 = tmp9 > tmp10
tmp12 = 9.999999717180685e-10
tmp13 = tl.where(tmp11, tmp9, tmp12)
tmp14 = tmp8 / tmp13
tl.store(out_ptr2 + (r1 + (16*x0)), tmp14, 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)
buf4 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [max_1, min_1, nval, range_1, gt, eps_mat, range_2, x_1], Original ATen: [aten.max, aten.min, aten.sub, aten.gt, aten.add, aten.where, aten.div]
stream0 = get_raw_stream(0)
triton_per_fused_add_div_gt_max_min_sub_where_0.run(arg0_1, buf4, 4, 16, grid=grid(4), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf4, (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
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.autograd
class RangeNorm2D(nn.Module):
"""
This will normalize a saliency map to range from 0 to 1 via linear range function.
Input and output will be a 3D tensor of size [batch size x height x width].
Input can be any real valued number (supported by hardware)
Output will range from 0 to 1
Parameters:
full_norm: This forces the values to range completely from 0 to 1.
"""
def __init__(self, full_norm=True, eps=1e-09):
super(RangeNorm2D, self).__init__()
self.full_norm = full_norm
self.eps = eps
def forward(self, x):
"""
Input:
x: A Torch Tensor image with shape [batch size x height x width] e.g. [64,7,7]
All values should be real positive (i.e. >= 0).
Return:
x: x Normalized by dividing by either the min value or the range between max and min.
Each max/min is computed for each batch item.
"""
assert torch.is_tensor(x), 'Input must be a Torch Tensor'
assert len(x.size()
) == 3, 'Input should be sizes [batch size x height x width]'
s0 = x.size()[0]
s1 = x.size()[1]
s2 = x.size()[2]
x = x.reshape(s0, s1 * s2)
xmax = x.max(dim=1)[0].reshape(s0, 1)
if self.full_norm:
xmin = x.min(dim=1)[0].reshape(s0, 1)
nval = x - xmin
range = xmax - xmin
else:
nval = x
range = xmax
"""
prevent divide by zero by setting zero to a small number
Simply adding eps does not work will in this case. So we use torch.where to set a minimum value.
"""
eps_mat = torch.zeros_like(range) + self.eps
range = torch.where(range > self.eps, range, eps_mat)
x = nval / range
x = x.reshape(s0, s1, s2)
return x
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 import triton_helpers
import torch.nn as nn
import torch.autograd
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_add_div_gt_max_min_sub_where_0(in_ptr0, out_ptr2,
xnumel, rnumel, XBLOCK: tl.constexpr):
xnumel = 4
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, float('-inf'))
tmp4 = triton_helpers.max2(tmp3, 1)[:, None]
tmp6 = tl.where(xmask, tmp1, float('inf'))
tmp7 = triton_helpers.min2(tmp6, 1)[:, None]
tmp8 = tmp0 - tmp7
tmp9 = tmp4 - tmp7
tmp10 = 1e-09
tmp11 = tmp9 > tmp10
tmp12 = 9.999999717180685e-10
tmp13 = tl.where(tmp11, tmp9, tmp12)
tmp14 = tmp8 / tmp13
tl.store(out_ptr2 + (r1 + 16 * x0), tmp14, 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)
buf4 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_add_div_gt_max_min_sub_where_0[grid(4)](arg0_1,
buf4, 4, 16, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
return reinterpret_tensor(buf4, (4, 4, 4), (16, 4, 1), 0),
class RangeNorm2DNew(nn.Module):
"""
This will normalize a saliency map to range from 0 to 1 via linear range function.
Input and output will be a 3D tensor of size [batch size x height x width].
Input can be any real valued number (supported by hardware)
Output will range from 0 to 1
Parameters:
full_norm: This forces the values to range completely from 0 to 1.
"""
def __init__(self, full_norm=True, eps=1e-09):
super(RangeNorm2DNew, self).__init__()
self.full_norm = full_norm
self.eps = eps
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
RangeNorm2D
| false
| 8,427
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
MaxMap
|
# 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/be/cbe62bamauzqfphojicwtl7ho6myyvnylv4btfha4nkoa3jpcrke.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 % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tl.store(out_ptr0 + (x2), tmp6, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [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
import torch.nn as nn
import torch.autograd
class MaxMap(nn.Module):
"""
Compute vanilla mean on a 4D tensor. This acts as a standard PyTorch layer.
The Max is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) Recommend a tensor with only positive values. (After a ReLU)
Any real value will work.
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self):
super(MaxMap, self).__init__()
def forward(self, x):
assert torch.is_tensor(x), 'input must be a Torch Tensor'
assert len(x.size()) > 2, 'input must have at least three dims'
x = torch.max(x, dim=1)[0]
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
import torch.autograd
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 % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp1 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp5 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp2 = triton_helpers.maximum(tmp0, tmp1)
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp6 = triton_helpers.maximum(tmp4, tmp5)
tl.store(out_ptr0 + x2, tmp6, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_max_0[grid(64)](arg0_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
return buf0,
class MaxMapNew(nn.Module):
"""
Compute vanilla mean on a 4D tensor. This acts as a standard PyTorch layer.
The Max is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) Recommend a tensor with only positive values. (After a ReLU)
Any real value will work.
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self):
super(MaxMapNew, self).__init__()
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
MaxMap
| false
| 8,428
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
InfoNCE_loss_vectorized
|
# 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/cpoe4jf5rd2jtxre5frkxszfgta5tmna5miyocthivabzvjnckhw.py
# Topologically Sorted Source Nodes: [similarities], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
# Source node to ATen node mapping:
# similarities => 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 = (%select, 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 = (%select, %clamp_min), kwargs = {})
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_4, 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 = (%select_4, %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=[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_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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (12 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (x3), xmask)
tmp17 = tl.load(in_ptr1 + (x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (4 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (8 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (12 + x0 + (16*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/x4/cx4ikvss35qp72zdkepsqffz766walj3bsti724i3txsoupnbzik.py
# Topologically Sorted Source Nodes: [similarities, similarities_1, exp_tensor, sum_1], Original ATen: [aten.sum, aten.div, aten.exp]
# Source node to ATen node mapping:
# exp_tensor => exp
# similarities => sum_3
# similarities_1 => div_2
# sum_1 => sum_4
# Graph fragment:
# %sum_3 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%mul, [1]), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sum_3, 4), kwargs = {})
# %exp : [num_users=2] = call_function[target=torch.ops.aten.exp.default](args = (%div_2,), kwargs = {})
# %sum_4 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%exp,), kwargs = {})
triton_per_fused_div_exp_sum_1 = async_compile.triton('triton_per_fused_div_exp_sum_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[1, 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_div_exp_sum_1', 'mutated_arg_names': [], '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_div_exp_sum_1(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 % 4
r1 = (rindex // 4)
tmp0 = tl.load(in_ptr0 + (r0 + (16*r1)), None)
tmp1 = tl.load(in_ptr0 + (4 + r0 + (16*r1)), None)
tmp3 = tl.load(in_ptr0 + (8 + r0 + (16*r1)), None)
tmp5 = tl.load(in_ptr0 + (12 + r0 + (16*r1)), None)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tmp9 = tl_math.exp(tmp8)
tmp10 = tl.broadcast_to(tmp9, [XBLOCK, RBLOCK])
tmp12 = tl.sum(tmp10, 1)[:, None]
tl.store(out_ptr0 + (tl.full([XBLOCK, 1], 0, tl.int32)), tmp12, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/re/crefqcphnx7mocofjtm62ovu6wh7k3msevqr2iapwbareettmfmc.py
# Topologically Sorted Source Nodes: [similarities_2], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
# Source node to ATen node mapping:
# similarities_2 => clamp_min_2, clamp_min_3, div_4, div_5, mul_1, pow_5, pow_6, pow_7, pow_8, sum_5, sum_6
# Graph fragment:
# %pow_5 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_1, 2), kwargs = {})
# %sum_5 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_5, [1], True), kwargs = {})
# %pow_6 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_5, 0.5), kwargs = {})
# %clamp_min_2 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_6, 1e-08), kwargs = {})
# %div_5 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_1, %clamp_min_2), kwargs = {})
# %pow_7 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_5, 2), kwargs = {})
# %sum_6 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_7, [1], True), kwargs = {})
# %pow_8 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_6, 0.5), kwargs = {})
# %clamp_min_3 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_8, 1e-08), kwargs = {})
# %div_4 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_5, %clamp_min_3), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_5, %div_4), kwargs = {})
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_2 = async_compile.triton('triton_poi_fused_clamp_min_div_linalg_vector_norm_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=[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_clamp_min_div_linalg_vector_norm_mul_2', '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_2(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (64 + x3), xmask)
tmp1 = tl.load(in_ptr0 + (64 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (68 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (72 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (76 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (64 + x3), xmask)
tmp17 = tl.load(in_ptr1 + (64 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (68 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (72 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (76 + x0 + (16*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/eg/cegv7ez4r3nhfbyjr5a7iemqrcp577x7uxyjeedmiwgjcevua72k.py
# Topologically Sorted Source Nodes: [similarities_4], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
# Source node to ATen node mapping:
# similarities_4 => clamp_min_4, clamp_min_5, div_8, div_9, mul_2, pow_10, pow_11, pow_12, pow_9, sum_10, sum_9
# Graph fragment:
# %pow_9 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_2, 2), kwargs = {})
# %sum_9 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_9, [1], True), kwargs = {})
# %pow_10 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_9, 0.5), kwargs = {})
# %clamp_min_4 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_10, 1e-08), kwargs = {})
# %div_9 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_2, %clamp_min_4), kwargs = {})
# %pow_11 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_6, 2), kwargs = {})
# %sum_10 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_11, [1], True), kwargs = {})
# %pow_12 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_10, 0.5), kwargs = {})
# %clamp_min_5 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_12, 1e-08), kwargs = {})
# %div_8 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_6, %clamp_min_5), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_9, %div_8), kwargs = {})
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_3 = async_compile.triton('triton_poi_fused_clamp_min_div_linalg_vector_norm_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: 'i32'}, 'device': 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_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_clamp_min_div_linalg_vector_norm_mul_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
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (128 + x3), xmask)
tmp1 = tl.load(in_ptr0 + (128 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (132 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (136 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (140 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (128 + x3), xmask)
tmp17 = tl.load(in_ptr1 + (128 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (132 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (136 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (140 + x0 + (16*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/eg/cegftiqockibyeggghlmjar4rbd4k4shclu3hnfs4etbkoqng5eq.py
# Topologically Sorted Source Nodes: [similarities_6], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
# Source node to ATen node mapping:
# similarities_6 => clamp_min_6, clamp_min_7, div_12, div_13, mul_3, pow_13, pow_14, pow_15, pow_16, sum_13, sum_14
# Graph fragment:
# %pow_13 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_3, 2), kwargs = {})
# %sum_13 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_13, [1], True), kwargs = {})
# %pow_14 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_13, 0.5), kwargs = {})
# %clamp_min_6 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_14, 1e-08), kwargs = {})
# %div_13 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_3, %clamp_min_6), kwargs = {})
# %pow_15 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%select_7, 2), kwargs = {})
# %sum_14 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_15, [1], True), kwargs = {})
# %pow_16 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sum_14, 0.5), kwargs = {})
# %clamp_min_7 : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%pow_16, 1e-08), kwargs = {})
# %div_12 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_7, %clamp_min_7), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%div_13, %div_12), kwargs = {})
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_4 = async_compile.triton('triton_poi_fused_clamp_min_div_linalg_vector_norm_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: 'i32'}, 'device': 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_4', '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_4(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
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (192 + x3), xmask)
tmp1 = tl.load(in_ptr0 + (192 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (196 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (200 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp9 = tl.load(in_ptr0 + (204 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp16 = tl.load(in_ptr1 + (192 + x3), xmask)
tmp17 = tl.load(in_ptr1 + (192 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp19 = tl.load(in_ptr1 + (196 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp22 = tl.load(in_ptr1 + (200 + x0 + (16*x2)), xmask, eviction_policy='evict_last')
tmp25 = tl.load(in_ptr1 + (204 + x0 + (16*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/dx/cdxpinddr5taszdtmg7xcwnaano7mpoidpebaefrqoq7vjbxtdw7.py
# Topologically Sorted Source Nodes: [loss, log, loss_1, total_loss, loss_2, log_1, loss_3, total_loss_1, loss_4, log_2, loss_5, total_loss_2, loss_6, log_3, loss_7, total_loss_3, total_loss_final], Original ATen: [aten.div, aten.log, aten.neg, aten.add]
# Source node to ATen node mapping:
# log => log
# log_1 => log_1
# log_2 => log_2
# log_3 => log_3
# loss => div_3
# loss_1 => neg
# loss_2 => div_7
# loss_3 => neg_1
# loss_4 => div_11
# loss_5 => neg_2
# loss_6 => div_15
# loss_7 => neg_3
# total_loss => add
# total_loss_1 => add_1
# total_loss_2 => add_2
# total_loss_3 => add_3
# total_loss_final => div_16
# Graph fragment:
# %div_3 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_8, %sum_4), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%div_3,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%log,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%neg, 0), kwargs = {})
# %div_7 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_9, %sum_8), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%div_7,), kwargs = {})
# %neg_1 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%log_1,), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %neg_1), kwargs = {})
# %div_11 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_10, %sum_12), kwargs = {})
# %log_2 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%div_11,), kwargs = {})
# %neg_2 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%log_2,), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %neg_2), kwargs = {})
# %div_15 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%select_11, %sum_16), kwargs = {})
# %log_3 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%div_15,), kwargs = {})
# %neg_3 : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%log_3,), kwargs = {})
# %add_3 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_2, %neg_3), kwargs = {})
# %div_16 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_3, 4), kwargs = {})
triton_poi_fused_add_div_log_neg_5 = async_compile.triton('triton_poi_fused_add_div_log_neg_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],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: '*fp32', 7: '*fp32', 8: '*fp32', 9: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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, 8), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_add_div_log_neg_5', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 20, '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_log_neg_5(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, xnumel, XBLOCK : tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp10 = tl.load(in_ptr1 + (0))
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp17 = tl.load(in_ptr2 + (x0), xmask)
tmp18 = tl.load(in_ptr2 + (4 + x0), xmask)
tmp20 = tl.load(in_ptr2 + (8 + x0), xmask)
tmp22 = tl.load(in_ptr2 + (12 + x0), xmask)
tmp26 = tl.load(in_ptr3 + (0))
tmp27 = tl.broadcast_to(tmp26, [XBLOCK])
tmp32 = tl.load(in_ptr4 + (x0), xmask)
tmp33 = tl.load(in_ptr4 + (4 + x0), xmask)
tmp35 = tl.load(in_ptr4 + (8 + x0), xmask)
tmp37 = tl.load(in_ptr4 + (12 + x0), xmask)
tmp41 = tl.load(in_ptr5 + (0))
tmp42 = tl.broadcast_to(tmp41, [XBLOCK])
tmp47 = tl.load(in_ptr6 + (x0), xmask)
tmp48 = tl.load(in_ptr6 + (4 + x0), xmask)
tmp50 = tl.load(in_ptr6 + (8 + x0), xmask)
tmp52 = tl.load(in_ptr6 + (12 + x0), xmask)
tmp56 = tl.load(in_ptr7 + (0))
tmp57 = tl.broadcast_to(tmp56, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tmp9 = tl_math.exp(tmp8)
tmp12 = tmp9 / tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = -tmp13
tmp15 = 0.0
tmp16 = tmp14 + tmp15
tmp19 = tmp17 + tmp18
tmp21 = tmp19 + tmp20
tmp23 = tmp21 + tmp22
tmp24 = tmp23 * tmp7
tmp25 = tl_math.exp(tmp24)
tmp28 = tmp25 / tmp27
tmp29 = tl_math.log(tmp28)
tmp30 = -tmp29
tmp31 = tmp16 + tmp30
tmp34 = tmp32 + tmp33
tmp36 = tmp34 + tmp35
tmp38 = tmp36 + tmp37
tmp39 = tmp38 * tmp7
tmp40 = tl_math.exp(tmp39)
tmp43 = tmp40 / tmp42
tmp44 = tl_math.log(tmp43)
tmp45 = -tmp44
tmp46 = tmp31 + tmp45
tmp49 = tmp47 + tmp48
tmp51 = tmp49 + tmp50
tmp53 = tmp51 + tmp52
tmp54 = tmp53 * tmp7
tmp55 = tl_math.exp(tmp54)
tmp58 = tmp55 / tmp57
tmp59 = tl_math.log(tmp58)
tmp60 = -tmp59
tmp61 = tmp46 + tmp60
tmp62 = tmp61 * tmp7
tl.store(in_out_ptr0 + (x0), tmp62, 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), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [similarities], 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(arg0_1, arg1_1, buf0, 64, grid=grid(64), stream=stream0)
buf1 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [similarities, similarities_1, exp_tensor, sum_1], Original ATen: [aten.sum, aten.div, aten.exp]
triton_per_fused_div_exp_sum_1.run(buf0, buf1, 1, 16, grid=grid(1), stream=stream0)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [similarities_2], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_2.run(arg0_1, arg1_1, buf2, 64, grid=grid(64), stream=stream0)
buf3 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [similarities_2, similarities_3, exp_tensor_1, sum_2], Original ATen: [aten.sum, aten.div, aten.exp]
triton_per_fused_div_exp_sum_1.run(buf2, buf3, 1, 16, grid=grid(1), stream=stream0)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [similarities_4], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_3.run(arg0_1, arg1_1, buf5, 64, grid=grid(64), stream=stream0)
buf6 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [similarities_4, similarities_5, exp_tensor_2, sum_3], Original ATen: [aten.sum, aten.div, aten.exp]
triton_per_fused_div_exp_sum_1.run(buf5, buf6, 1, 16, grid=grid(1), stream=stream0)
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [similarities_6], Original ATen: [aten.linalg_vector_norm, aten.clamp_min, aten.div, aten.mul]
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_4.run(arg0_1, arg1_1, buf7, 64, grid=grid(64), stream=stream0)
del arg0_1
del arg1_1
buf8 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [similarities_6, similarities_7, exp_tensor_3, sum_4], Original ATen: [aten.sum, aten.div, aten.exp]
triton_per_fused_div_exp_sum_1.run(buf7, buf8, 1, 16, grid=grid(1), stream=stream0)
buf4 = empty_strided_cuda((4, ), (1, ), torch.float32)
buf9 = buf4; del buf4 # reuse
buf10 = buf9; del buf9 # reuse
# Topologically Sorted Source Nodes: [loss, log, loss_1, total_loss, loss_2, log_1, loss_3, total_loss_1, loss_4, log_2, loss_5, total_loss_2, loss_6, log_3, loss_7, total_loss_3, total_loss_final], Original ATen: [aten.div, aten.log, aten.neg, aten.add]
triton_poi_fused_add_div_log_neg_5.run(buf10, buf0, buf1, buf2, buf3, buf5, buf6, buf7, buf8, 4, grid=grid(4), stream=stream0)
del buf0
del buf1
del buf2
del buf3
del buf5
del buf6
del buf7
del buf8
return (buf10, )
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 InfoNCE_loss_vectorized(nn.Module):
"""
SimCLR loss: https://github.com/google-research/simclr // https://github.com/sthalles/SimCLR
"""
def __init__(self, temperature):
super(InfoNCE_loss_vectorized, self).__init__()
self.temperature = temperature
self.cos = nn.CosineSimilarity(dim=1)
def forward(self, graph_out, sent_out):
total_loss = 0
for g, s in zip(graph_out, sent_out):
similarities = self.cos(g, s)
similarities = similarities / self.temperature
exp_tensor = torch.exp(similarities)
loss = exp_tensor[0] / torch.sum(exp_tensor)
loss = -torch.log(loss)
total_loss = total_loss + loss
total_loss_final = total_loss / len(graph_out)
return total_loss_final
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'temperature': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (12 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + x3, xmask)
tmp17 = tl.load(in_ptr1 + (x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr1 + (4 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + (8 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr1 + (12 + x0 + 16 * 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_per_fused_div_exp_sum_1(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 % 4
r1 = rindex // 4
tmp0 = tl.load(in_ptr0 + (r0 + 16 * r1), None)
tmp1 = tl.load(in_ptr0 + (4 + r0 + 16 * r1), None)
tmp3 = tl.load(in_ptr0 + (8 + r0 + 16 * r1), None)
tmp5 = tl.load(in_ptr0 + (12 + r0 + 16 * r1), None)
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tmp9 = tl_math.exp(tmp8)
tmp10 = tl.broadcast_to(tmp9, [XBLOCK, RBLOCK])
tmp12 = tl.sum(tmp10, 1)[:, None]
tl.store(out_ptr0 + tl.full([XBLOCK, 1], 0, tl.int32), tmp12, None)
@triton.jit
def triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_2(in_ptr0,
in_ptr1, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 4
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + (64 + x3), xmask)
tmp1 = tl.load(in_ptr0 + (64 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (68 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (72 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (76 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + (64 + x3), xmask)
tmp17 = tl.load(in_ptr1 + (64 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr1 + (68 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + (72 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr1 + (76 + x0 + 16 * 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_clamp_min_div_linalg_vector_norm_mul_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
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + (128 + x3), xmask)
tmp1 = tl.load(in_ptr0 + (128 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (132 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (136 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (140 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + (128 + x3), xmask)
tmp17 = tl.load(in_ptr1 + (128 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr1 + (132 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + (136 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr1 + (140 + x0 + 16 * 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_clamp_min_div_linalg_vector_norm_mul_4(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
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + (192 + x3), xmask)
tmp1 = tl.load(in_ptr0 + (192 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = tl.load(in_ptr0 + (196 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (200 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp9 = tl.load(in_ptr0 + (204 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp16 = tl.load(in_ptr1 + (192 + x3), xmask)
tmp17 = tl.load(in_ptr1 + (192 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp19 = tl.load(in_ptr1 + (196 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp22 = tl.load(in_ptr1 + (200 + x0 + 16 * x2), xmask, eviction_policy=
'evict_last')
tmp25 = tl.load(in_ptr1 + (204 + x0 + 16 * 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_add_div_log_neg_5(in_out_ptr0, in_ptr0, in_ptr1,
in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, in_ptr7, xnumel, XBLOCK:
tl.constexpr):
xnumel = 4
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + (4 + x0), xmask)
tmp3 = tl.load(in_ptr0 + (8 + x0), xmask)
tmp5 = tl.load(in_ptr0 + (12 + x0), xmask)
tmp10 = tl.load(in_ptr1 + 0)
tmp11 = tl.broadcast_to(tmp10, [XBLOCK])
tmp17 = tl.load(in_ptr2 + x0, xmask)
tmp18 = tl.load(in_ptr2 + (4 + x0), xmask)
tmp20 = tl.load(in_ptr2 + (8 + x0), xmask)
tmp22 = tl.load(in_ptr2 + (12 + x0), xmask)
tmp26 = tl.load(in_ptr3 + 0)
tmp27 = tl.broadcast_to(tmp26, [XBLOCK])
tmp32 = tl.load(in_ptr4 + x0, xmask)
tmp33 = tl.load(in_ptr4 + (4 + x0), xmask)
tmp35 = tl.load(in_ptr4 + (8 + x0), xmask)
tmp37 = tl.load(in_ptr4 + (12 + x0), xmask)
tmp41 = tl.load(in_ptr5 + 0)
tmp42 = tl.broadcast_to(tmp41, [XBLOCK])
tmp47 = tl.load(in_ptr6 + x0, xmask)
tmp48 = tl.load(in_ptr6 + (4 + x0), xmask)
tmp50 = tl.load(in_ptr6 + (8 + x0), xmask)
tmp52 = tl.load(in_ptr6 + (12 + x0), xmask)
tmp56 = tl.load(in_ptr7 + 0)
tmp57 = tl.broadcast_to(tmp56, [XBLOCK])
tmp2 = tmp0 + tmp1
tmp4 = tmp2 + tmp3
tmp6 = tmp4 + tmp5
tmp7 = 0.25
tmp8 = tmp6 * tmp7
tmp9 = tl_math.exp(tmp8)
tmp12 = tmp9 / tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = -tmp13
tmp15 = 0.0
tmp16 = tmp14 + tmp15
tmp19 = tmp17 + tmp18
tmp21 = tmp19 + tmp20
tmp23 = tmp21 + tmp22
tmp24 = tmp23 * tmp7
tmp25 = tl_math.exp(tmp24)
tmp28 = tmp25 / tmp27
tmp29 = tl_math.log(tmp28)
tmp30 = -tmp29
tmp31 = tmp16 + tmp30
tmp34 = tmp32 + tmp33
tmp36 = tmp34 + tmp35
tmp38 = tmp36 + tmp37
tmp39 = tmp38 * tmp7
tmp40 = tl_math.exp(tmp39)
tmp43 = tmp40 / tmp42
tmp44 = tl_math.log(tmp43)
tmp45 = -tmp44
tmp46 = tmp31 + tmp45
tmp49 = tmp47 + tmp48
tmp51 = tmp49 + tmp50
tmp53 = tmp51 + tmp52
tmp54 = tmp53 * tmp7
tmp55 = tl_math.exp(tmp54)
tmp58 = tmp55 / tmp57
tmp59 = tl_math.log(tmp58)
tmp60 = -tmp59
tmp61 = tmp46 + tmp60
tmp62 = tmp61 * tmp7
tl.store(in_out_ptr0 + x0, tmp62, 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), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_0[grid(64)](
arg0_1, arg1_1, buf0, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf1 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_div_exp_sum_1[grid(1)](buf0, buf1, 1, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_2[grid(64)](
arg0_1, arg1_1, buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf3 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_div_exp_sum_1[grid(1)](buf2, buf3, 1, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf5 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_3[grid(64)](
arg0_1, arg1_1, buf5, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf6 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_div_exp_sum_1[grid(1)](buf5, buf6, 1, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf7 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_clamp_min_div_linalg_vector_norm_mul_4[grid(64)](
arg0_1, arg1_1, buf7, 64, XBLOCK=64, num_warps=1, num_stages=1)
del arg0_1
del arg1_1
buf8 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_div_exp_sum_1[grid(1)](buf7, buf8, 1, 16, XBLOCK=1,
num_warps=2, num_stages=1)
buf4 = empty_strided_cuda((4,), (1,), torch.float32)
buf9 = buf4
del buf4
buf10 = buf9
del buf9
triton_poi_fused_add_div_log_neg_5[grid(4)](buf10, buf0, buf1, buf2,
buf3, buf5, buf6, buf7, buf8, 4, XBLOCK=4, num_warps=1,
num_stages=1)
del buf0
del buf1
del buf2
del buf3
del buf5
del buf6
del buf7
del buf8
return buf10,
class InfoNCE_loss_vectorizedNew(nn.Module):
"""
SimCLR loss: https://github.com/google-research/simclr // https://github.com/sthalles/SimCLR
"""
def __init__(self, temperature):
super(InfoNCE_loss_vectorizedNew, self).__init__()
self.temperature = temperature
self.cos = nn.CosineSimilarity(dim=1)
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
LIIR-KULeuven/CLDR_CLNER_models
|
InfoNCE_loss_vectorized
| false
| 8,429
|
[
"MIT"
] | 12
|
5fe47a988b88a36d0ccf4484aff5ab70c59f39d6
|
https://github.com/LIIR-KULeuven/CLDR_CLNER_models/tree/5fe47a988b88a36d0ccf4484aff5ab70c59f39d6
|
ClassificationModel
|
# 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/6z/c6zb2twnjioimzrk4rie3z6om2pycozdr6corfhcdm5q6f7h4oqz.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[1024, 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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_0', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 1024
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 4
y1 = (yindex // 4)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (4*x2) + (36*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/62/c62myuf63oepurbxeoz2olrr5dwoipydpve4ekfnunp6gerrikor.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_1 = async_compile.triton('triton_poi_fused_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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_1(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
x2 = xindex
y3 = yindex
y0 = yindex % 4
y1 = (yindex // 4)
tmp0 = tl.load(in_ptr0 + (x2 + (16*y3)), xmask & ymask)
tl.store(out_ptr0 + (y0 + (4*x2) + (64*y1)), tmp0, xmask & ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/sf/csfbo2wg4qzrcculqiz2kklr2dwz3q6kpztuuavooookip4gpwcb.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_2 = async_compile.triton('triton_poi_fused_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[65536, 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), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_2', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_2(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 65536
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = (yindex // 256)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (256*x2) + (2304*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/66/c66wyyjpsnatq6od5cev6tenkovm6iijhpkvolh22f7cpiccrbj4.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_3 = async_compile.triton('triton_poi_fused_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 184320
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = (yindex // 256)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (256*x2) + (2304*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gc/cgctjlgnbq4wfwdwoevcpfb7vf3b2chjk7tvdxs3kk4uy3k2jnp7.py
# Topologically Sorted Source Nodes: [out, out_1], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# out => convolution
# out_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_4 = async_compile.triton('triton_poi_fused_convolution_relu_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[16384],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_4', '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_4(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 16384
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
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)
tl.store(in_out_ptr0 + (x2), tmp4, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/mx/cmxmwjlunpgkjjeidhzy2o4cgx2nth5kjthef775uacebbx34z2t.py
# Topologically Sorted Source Nodes: [out_8, out_9], Original ATen: [aten.convolution, aten.sigmoid, aten.sigmoid_backward]
# Source node to ATen node mapping:
# out_8 => convolution_4
# out_9 => sigmoid
# Graph fragment:
# %convolution_4 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%relu_3, %primals_10, %primals_11, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %sigmoid : [num_users=3] = call_function[target=torch.ops.aten.sigmoid.default](args = (%convolution_4,), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1, %sigmoid), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %sub), kwargs = {})
triton_poi_fused_convolution_sigmoid_sigmoid_backward_5 = async_compile.triton('triton_poi_fused_convolution_sigmoid_sigmoid_backward_5', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[4096, 16], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_convolution_sigmoid_sigmoid_backward_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_convolution_sigmoid_sigmoid_backward_5(in_ptr0, in_ptr1, out_ptr0, out_ptr1, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 2880
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 720
y1 = (yindex // 720)
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + (720*x2) + (11520*y1)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tmp4 = 1.0
tmp5 = tmp4 - tmp3
tmp6 = tmp3 * tmp5
tl.store(out_ptr0 + (x2 + (16*y3)), tmp3, xmask & ymask)
tl.store(out_ptr1 + (y0 + (720*x2) + (11520*y1)), tmp6, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (256, 4, 3, 3), (36, 9, 3, 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, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_5, (256, ), (1, ))
assert_size_stride(primals_6, (256, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_7, (256, ), (1, ))
assert_size_stride(primals_8, (256, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_9, (256, ), (1, ))
assert_size_stride(primals_10, (720, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_11, (720, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((256, 4, 3, 3), (36, 1, 12, 4), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(primals_1, buf0, 1024, 9, grid=grid(1024, 9), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 1, 16, 4), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_1.run(primals_3, buf1, 16, 16, grid=grid(16, 16), stream=stream0)
del primals_3
buf2 = empty_strided_cuda((256, 256, 3, 3), (2304, 1, 768, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(primals_4, buf2, 65536, 9, grid=grid(65536, 9), stream=stream0)
del primals_4
buf3 = empty_strided_cuda((256, 256, 3, 3), (2304, 1, 768, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(primals_6, buf3, 65536, 9, grid=grid(65536, 9), stream=stream0)
del primals_6
buf4 = empty_strided_cuda((256, 256, 3, 3), (2304, 1, 768, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_2.run(primals_8, buf4, 65536, 9, grid=grid(65536, 9), stream=stream0)
del primals_8
buf5 = empty_strided_cuda((720, 256, 3, 3), (2304, 1, 768, 256), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
triton_poi_fused_3.run(primals_10, buf5, 184320, 9, grid=grid(184320, 9), stream=stream0)
del primals_10
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf6 = extern_kernels.convolution(buf1, buf0, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf6, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf7 = buf6; del buf6 # reuse
# Topologically Sorted Source Nodes: [out, out_1], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_4.run(buf7, primals_2, 16384, grid=grid(16384), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [out_2], Original ATen: [aten.convolution]
buf8 = extern_kernels.convolution(buf7, buf2, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf9 = buf8; del buf8 # reuse
# Topologically Sorted Source Nodes: [out_2, out_3], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_4.run(buf9, primals_5, 16384, grid=grid(16384), stream=stream0)
del primals_5
# Topologically Sorted Source Nodes: [out_4], Original ATen: [aten.convolution]
buf10 = extern_kernels.convolution(buf9, buf3, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf11 = buf10; del buf10 # reuse
# Topologically Sorted Source Nodes: [out_4, out_5], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_4.run(buf11, primals_7, 16384, grid=grid(16384), stream=stream0)
del primals_7
# Topologically Sorted Source Nodes: [out_6], Original ATen: [aten.convolution]
buf12 = extern_kernels.convolution(buf11, buf4, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf13 = buf12; del buf12 # reuse
# Topologically Sorted Source Nodes: [out_6, out_7], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_4.run(buf13, primals_9, 16384, grid=grid(16384), stream=stream0)
del primals_9
# Topologically Sorted Source Nodes: [out_8], Original ATen: [aten.convolution]
buf14 = extern_kernels.convolution(buf13, buf5, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf14, (4, 720, 4, 4), (11520, 1, 2880, 720))
buf15 = empty_strided_cuda((4, 720, 4, 4), (11520, 16, 4, 1), torch.float32)
buf16 = empty_strided_cuda((4, 720, 4, 4), (11520, 1, 2880, 720), torch.float32)
# Topologically Sorted Source Nodes: [out_8, out_9], Original ATen: [aten.convolution, aten.sigmoid, aten.sigmoid_backward]
triton_poi_fused_convolution_sigmoid_sigmoid_backward_5.run(buf14, primals_11, buf15, buf16, 2880, 16, grid=grid(2880, 16), stream=stream0)
del buf14
del primals_11
return (reinterpret_tensor(buf15, (4, 4, 4, 9, 80), (11520, 4, 1, 1280, 16), 0), buf0, buf1, buf2, buf3, buf4, buf5, buf7, buf9, buf11, buf13, buf16, )
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, 3, 3), (36, 9, 3, 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, 3, 3), (2304, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((256, 256, 3, 3), (2304, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((256, 256, 3, 3), (2304, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((256, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((720, 256, 3, 3), (2304, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((720, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class ClassificationModel(nn.Module):
def __init__(self, num_features_in, num_anchors=9, num_classes=80,
prior=0.01, feature_size=256):
super(ClassificationModel, self).__init__()
self.num_classes = num_classes
self.num_anchors = num_anchors
self.conv1 = nn.Conv2d(num_features_in, feature_size, kernel_size=3,
padding=1)
self.act1 = nn.ReLU()
self.conv2 = nn.Conv2d(feature_size, feature_size, kernel_size=3,
padding=1)
self.act2 = nn.ReLU()
self.conv3 = nn.Conv2d(feature_size, feature_size, kernel_size=3,
padding=1)
self.act3 = nn.ReLU()
self.conv4 = nn.Conv2d(feature_size, feature_size, kernel_size=3,
padding=1)
self.act4 = nn.ReLU()
self.output = nn.Conv2d(feature_size, num_anchors * num_classes,
kernel_size=3, padding=1)
self.output_act = nn.Sigmoid()
def forward(self, x):
out = self.conv1(x)
out = self.act1(out)
out = self.conv2(out)
out = self.act2(out)
out = self.conv3(out)
out = self.act3(out)
out = self.conv4(out)
out = self.act4(out)
out = self.output(out)
out = self.output_act(out)
out1 = out.permute(0, 2, 3, 1)
batch_size, width, height, _channels = out1.shape
out2 = out1.view(batch_size, width, height, self.num_anchors, self.
num_classes)
return out2
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'num_features_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
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_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 4
y1 = yindex // 4
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 4 * x2 + 36 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_1(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
x2 = xindex
y3 = yindex
y0 = yindex % 4
y1 = yindex // 4
tmp0 = tl.load(in_ptr0 + (x2 + 16 * y3), xmask & ymask)
tl.store(out_ptr0 + (y0 + 4 * x2 + 64 * y1), tmp0, xmask & ymask)
@triton.jit
def triton_poi_fused_2(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)
) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = yindex // 256
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 256 * x2 + 2304 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_3(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = (tl.program_id(1) + tl.program_id(2) * tl.num_programs(1)
) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 256
y1 = yindex // 256
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 256 * x2 + 2304 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_4(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
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)
tl.store(in_out_ptr0 + x2, tmp4, None)
@triton.jit
def triton_poi_fused_convolution_sigmoid_sigmoid_backward_5(in_ptr0,
in_ptr1, out_ptr0, out_ptr1, ynumel, xnumel, YBLOCK: tl.constexpr,
XBLOCK: tl.constexpr):
ynumel = 2880
xnumel = 16
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y0 = yindex % 720
y1 = yindex // 720
y3 = yindex
tmp0 = tl.load(in_ptr0 + (y0 + 720 * x2 + 11520 * y1), xmask & ymask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.sigmoid(tmp2)
tmp4 = 1.0
tmp5 = tmp4 - tmp3
tmp6 = tmp3 * tmp5
tl.store(out_ptr0 + (x2 + 16 * y3), tmp3, xmask & ymask)
tl.store(out_ptr1 + (y0 + 720 * x2 + 11520 * y1), tmp6, xmask & ymask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (256, 4, 3, 3), (36, 9, 3, 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, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_5, (256,), (1,))
assert_size_stride(primals_6, (256, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_7, (256,), (1,))
assert_size_stride(primals_8, (256, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_9, (256,), (1,))
assert_size_stride(primals_10, (720, 256, 3, 3), (2304, 9, 3, 1))
assert_size_stride(primals_11, (720,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((256, 4, 3, 3), (36, 1, 12, 4), torch.float32
)
get_raw_stream(0)
triton_poi_fused_0[grid(1024, 9)](primals_1, buf0, 1024, 9, XBLOCK=
16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 1, 16, 4), torch.float32)
triton_poi_fused_1[grid(16, 16)](primals_3, buf1, 16, 16, XBLOCK=16,
YBLOCK=16, num_warps=4, num_stages=1)
del primals_3
buf2 = empty_strided_cuda((256, 256, 3, 3), (2304, 1, 768, 256),
torch.float32)
triton_poi_fused_2[grid(65536, 9)](primals_4, buf2, 65536, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_4
buf3 = empty_strided_cuda((256, 256, 3, 3), (2304, 1, 768, 256),
torch.float32)
triton_poi_fused_2[grid(65536, 9)](primals_6, buf3, 65536, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_6
buf4 = empty_strided_cuda((256, 256, 3, 3), (2304, 1, 768, 256),
torch.float32)
triton_poi_fused_2[grid(65536, 9)](primals_8, buf4, 65536, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_8
buf5 = empty_strided_cuda((720, 256, 3, 3), (2304, 1, 768, 256),
torch.float32)
triton_poi_fused_3[grid(184320, 9)](primals_10, buf5, 184320, 9,
XBLOCK=16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_10
buf6 = extern_kernels.convolution(buf1, buf0, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf6, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf7 = buf6
del buf6
triton_poi_fused_convolution_relu_4[grid(16384)](buf7, primals_2,
16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf8 = extern_kernels.convolution(buf7, buf2, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf8, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf9 = buf8
del buf8
triton_poi_fused_convolution_relu_4[grid(16384)](buf9, primals_5,
16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf10 = extern_kernels.convolution(buf9, buf3, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf10, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf11 = buf10
del buf10
triton_poi_fused_convolution_relu_4[grid(16384)](buf11, primals_7,
16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_7
buf12 = extern_kernels.convolution(buf11, buf4, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf12, (4, 256, 4, 4), (4096, 1, 1024, 256))
buf13 = buf12
del buf12
triton_poi_fused_convolution_relu_4[grid(16384)](buf13, primals_9,
16384, XBLOCK=128, num_warps=4, num_stages=1)
del primals_9
buf14 = extern_kernels.convolution(buf13, buf5, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf14, (4, 720, 4, 4), (11520, 1, 2880, 720))
buf15 = empty_strided_cuda((4, 720, 4, 4), (11520, 16, 4, 1), torch
.float32)
buf16 = empty_strided_cuda((4, 720, 4, 4), (11520, 1, 2880, 720),
torch.float32)
triton_poi_fused_convolution_sigmoid_sigmoid_backward_5[grid(2880, 16)
](buf14, primals_11, buf15, buf16, 2880, 16, XBLOCK=16, YBLOCK=
32, num_warps=4, num_stages=1)
del buf14
del primals_11
return reinterpret_tensor(buf15, (4, 4, 4, 9, 80), (11520, 4, 1, 1280,
16), 0
), buf0, buf1, buf2, buf3, buf4, buf5, buf7, buf9, buf11, buf13, buf16
class ClassificationModelNew(nn.Module):
def __init__(self, num_features_in, num_anchors=9, num_classes=80,
prior=0.01, feature_size=256):
super(ClassificationModelNew, self).__init__()
self.num_classes = num_classes
self.num_anchors = num_anchors
self.conv1 = nn.Conv2d(num_features_in, feature_size, kernel_size=3,
padding=1)
self.act1 = nn.ReLU()
self.conv2 = nn.Conv2d(feature_size, feature_size, kernel_size=3,
padding=1)
self.act2 = nn.ReLU()
self.conv3 = nn.Conv2d(feature_size, feature_size, kernel_size=3,
padding=1)
self.act3 = nn.ReLU()
self.conv4 = nn.Conv2d(feature_size, feature_size, kernel_size=3,
padding=1)
self.act4 = nn.ReLU()
self.output = nn.Conv2d(feature_size, num_anchors * num_classes,
kernel_size=3, padding=1)
self.output_act = nn.Sigmoid()
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_6 = self.conv3.weight
primals_7 = self.conv3.bias
primals_8 = self.conv4.weight
primals_9 = self.conv4.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]
|
LLYXC/OXNet
|
ClassificationModel
| false
| 8,430
|
[
"Apache-2.0"
] | 13
|
4fb67a8c42b9158a8e563c4b68a157e4dedd9c66
|
https://github.com/LLYXC/OXNet/tree/4fb67a8c42b9158a8e563c4b68a157e4dedd9c66
|
TwoLayerNet
|
# 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/47/c472nztfmbxxfmhai7yhlryycsnpwjfi355joszqptwlh7nsb7pe.py
# Topologically Sorted Source Nodes: [h_relu], Original ATen: [aten.clamp, aten.ge]
# Source node to ATen node mapping:
# h_relu => clamp_min
# Graph fragment:
# %clamp_min : [num_users=1] = call_function[target=torch.ops.aten.clamp_min.default](args = (%view_1, 0), kwargs = {})
# %ge : [num_users=1] = call_function[target=torch.ops.aten.ge.Scalar](args = (%view_1, 0), kwargs = {})
triton_poi_fused_clamp_ge_0 = async_compile.triton('triton_poi_fused_clamp_ge_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*i1', 4: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_clamp_ge_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_ge_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = tmp2 >= tmp3
tl.store(out_ptr0 + (x2), tmp4, xmask)
tl.store(out_ptr1 + (x2), tmp5, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0), reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
# Topologically Sorted Source Nodes: [h_relu], Original ATen: [aten.clamp, aten.ge]
stream0 = get_raw_stream(0)
triton_poi_fused_clamp_ge_0.run(buf0, primals_2, buf1, buf3, 256, grid=grid(256), stream=stream0)
del primals_2
buf2 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [y_pred], 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
class TwoLayerNet(torch.nn.Module):
def __init__(self, D_in, H, D_out):
super(TwoLayerNet, self).__init__()
self.linear1 = torch.nn.Linear(D_in, H)
self.linear2 = torch.nn.Linear(H, D_out)
def forward(self, x):
h_relu = self.linear1(x).clamp(min=0)
y_pred = self.linear2(h_relu)
return y_pred
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'D_in': 4, 'H': 4, 'D_out': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_clamp_ge_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = 0.0
tmp4 = triton_helpers.maximum(tmp2, tmp3)
tmp5 = tmp2 >= tmp3
tl.store(out_ptr0 + x2, tmp4, xmask)
tl.store(out_ptr1 + x2, tmp5, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4), (4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_4, (4, 4), (4, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4), (4, 1), torch.float32)
extern_kernels.mm(reinterpret_tensor(primals_3, (64, 4), (4, 1), 0),
reinterpret_tensor(primals_1, (4, 4), (1, 4), 0), out=buf0)
del primals_1
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf3 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.bool)
get_raw_stream(0)
triton_poi_fused_clamp_ge_0[grid(256)](buf0, primals_2, buf1, buf3,
256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = buf0
del buf0
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 TwoLayerNetNew(torch.nn.Module):
def __init__(self, D_in, H, D_out):
super(TwoLayerNetNew, self).__init__()
self.linear1 = torch.nn.Linear(D_in, H)
self.linear2 = torch.nn.Linear(H, D_out)
def forward(self, input_0):
primals_1 = self.linear1.weight
primals_2 = self.linear1.bias
primals_4 = self.linear2.weight
primals_5 = self.linear2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
KentonMurray/ProxGradPytorch
|
TwoLayerNet
| false
| 8,431
|
[
"MIT"
] | 27
|
c534a49142ac9ec149ca67de24bb0487fde1607b
|
https://github.com/KentonMurray/ProxGradPytorch/tree/c534a49142ac9ec149ca67de24bb0487fde1607b
|
DiagonalwiseRefactorization
|
# 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/dv/cdvmujdx4wzp3xf3q2rdnt72xso7clbkadprbhlirztwjsalq3rx.py
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# weight => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, %primals_2), 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):
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, 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: [weight], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(primals_1, primals_2, buf0, 256, grid=grid(256), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf1 = extern_kernels.convolution(primals_3, buf0, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 3, 3), (36, 9, 3, 1))
return (buf1, primals_2, primals_3, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((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, 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.parallel
import torch.optim
import torch
import torch.nn as nn
import torch.utils.data
import torch.utils.data.distributed
def get_mask(in_channels, channels, ks):
in_channels = int(in_channels)
channels = int(channels)
if len(ks) == 1:
mask = np.zeros((int(in_channels), int(channels), int(ks[0])))
elif len(ks) == 2:
mask = np.zeros((int(in_channels), int(channels), int(ks[0]), int(
ks[1])))
elif len(ks) == 3:
mask = np.zeros((int(in_channels), int(channels), int(ks[0]), int(
ks[1]), int(ks[2])))
else:
raise Error('not implement yet')
for _ in range(in_channels):
mask[_, _ % channels, :, :] = 1.0
return mask
class DiagonalwiseRefactorization(nn.Module):
def __init__(self, in_channels, ks, stride=1, groups=1):
super(DiagonalwiseRefactorization, self).__init__()
channels = in_channels // groups
self.in_channels = in_channels
self.groups = groups
self.stride = stride
p = (np.array(ks) - 1) // 2
self.p = p.tolist()
self.mask = nn.Parameter(torch.Tensor(get_mask(in_channels,
channels, ks=ks)), requires_grad=False)
self.weight = nn.Parameter(torch.Tensor(in_channels, channels, *ks),
requires_grad=True)
torch.nn.init.xavier_uniform_(self.weight.data)
self.weight.data.mul_(self.mask.data)
if len(ks) == 1:
self.conv = nn.functional.conv1d
elif len(ks) == 2:
self.conv = nn.functional.conv2d
elif len(ks) == 3:
self.conv = nn.functional.conv3d
else:
raise Error(
'The kernal size in DiagonalwiseRefactorization is wrong!')
def forward(self, x):
weight = torch.mul(self.weight, self.mask)
x = self.conv(x, weight, bias=None, stride=self.stride, padding=
self.p, groups=self.groups)
return x
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'ks': [4, 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 numpy as np
import torch.nn.parallel
import torch.optim
import torch
import torch.nn as nn
import torch.utils.data
import torch.utils.data.distributed
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@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):
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, 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, primals_2, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
buf1 = extern_kernels.convolution(primals_3, buf0, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf1, (4, 4, 3, 3), (36, 9, 3, 1))
return buf1, primals_2, primals_3, buf0
def get_mask(in_channels, channels, ks):
in_channels = int(in_channels)
channels = int(channels)
if len(ks) == 1:
mask = np.zeros((int(in_channels), int(channels), int(ks[0])))
elif len(ks) == 2:
mask = np.zeros((int(in_channels), int(channels), int(ks[0]), int(
ks[1])))
elif len(ks) == 3:
mask = np.zeros((int(in_channels), int(channels), int(ks[0]), int(
ks[1]), int(ks[2])))
else:
raise Error('not implement yet')
for _ in range(in_channels):
mask[_, _ % channels, :, :] = 1.0
return mask
class DiagonalwiseRefactorizationNew(nn.Module):
def __init__(self, in_channels, ks, stride=1, groups=1):
super(DiagonalwiseRefactorizationNew, self).__init__()
channels = in_channels // groups
self.in_channels = in_channels
self.groups = groups
self.stride = stride
p = (np.array(ks) - 1) // 2
self.p = p.tolist()
self.mask = nn.Parameter(torch.Tensor(get_mask(in_channels,
channels, ks=ks)), requires_grad=False)
self.weight = nn.Parameter(torch.Tensor(in_channels, channels, *ks),
requires_grad=True)
torch.nn.init.xavier_uniform_(self.weight.data)
self.weight.data.mul_(self.mask.data)
if len(ks) == 1:
self.conv = nn.functional.conv1d
elif len(ks) == 2:
self.conv = nn.functional.conv2d
elif len(ks) == 3:
self.conv = nn.functional.conv3d
else:
raise Error(
'The kernal size in DiagonalwiseRefactorization is wrong!')
def forward(self, input_0):
primals_1 = self.mask
primals_2 = self.weight
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
LaputaDream/region-based-non-local-network
|
DiagonalwiseRefactorization
| false
| 8,432
|
[
"MIT"
] | 18
|
98e5fb3d8010e8c5360ac3066fdc06c37106d7dc
|
https://github.com/LaputaDream/region-based-non-local-network/tree/98e5fb3d8010e8c5360ac3066fdc06c37106d7dc
|
GroupLinear
|
# 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/ij/cij4ltov5vxwiaide426vddz4rlp6sfsgsldnnxoa2ympzgmlefc.py
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.add]
# Source node to ATen node mapping:
# output => add
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%view_3, %unsqueeze), 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': ['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_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x2), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
assert_size_stride(primals_3, (1, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4, 1), (4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [matmul], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(primals_3, (64, 4, 4), (0, 4, 1), 0), reinterpret_tensor(primals_1, (64, 4, 1), (4, 1, 1), 0), out=buf0)
del primals_3
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 1, 4, 1), (64, 16, 4, 4, 1, 1), 0); del buf0 # reuse
# Topologically Sorted Source Nodes: [output], Original ATen: [aten.add]
stream0 = get_raw_stream(0)
triton_poi_fused_add_0.run(buf1, primals_2, 256, grid=grid(256), stream=stream0)
del primals_2
return (reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0), reinterpret_tensor(primals_1, (64, 1, 4), (4, 1, 1), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((1, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((1, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.utils.data
class GroupLinear(nn.Module):
def __init__(self, in_features, out_features, groups, bias=True):
super(GroupLinear, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.groups = groups
self.weight = nn.Parameter(torch.Tensor(groups, out_features //
groups, in_features // groups))
if bias:
self.bias = nn.Parameter(torch.Tensor(groups, out_features //
groups))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
for weight in self.weight:
nn.init.kaiming_uniform_(weight, nonlinearity='linear')
if self.bias is not None:
nn.init.constant_(self.bias, 0)
def forward(self, input):
"""
Args:
input (Tensor): shape (*, in_features)
"""
batch_size = input.shape[:-1]
if self.bias is not None:
output = self.weight @ input.reshape(*batch_size, self.groups,
self.in_features // self.groups, 1) + self.bias[..., None]
else:
output = self.weight @ input.reshape(*batch_size, self.groups,
self.in_features // self.groups, 1)
return output.reshape(*batch_size, self.out_features)
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'in_features': 4, 'out_features': 4, 'groups': 1}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import torch.nn as nn
import torch.utils.data
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_add_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 4
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x2, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (1, 4), (4, 1))
assert_size_stride(primals_3, (1, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((64, 4, 1), (4, 1, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(primals_3, (64, 4, 4), (0, 4,
1), 0), reinterpret_tensor(primals_1, (64, 4, 1), (4, 1, 1), 0),
out=buf0)
del primals_3
buf1 = reinterpret_tensor(buf0, (4, 4, 4, 1, 4, 1), (64, 16, 4, 4,
1, 1), 0)
del buf0
get_raw_stream(0)
triton_poi_fused_add_0[grid(256)](buf1, primals_2, 256, XBLOCK=256,
num_warps=4, num_stages=1)
del primals_2
return reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 16, 4, 1), 0
), reinterpret_tensor(primals_1, (64, 1, 4), (4, 1, 1), 0)
class GroupLinearNew(nn.Module):
def __init__(self, in_features, out_features, groups, bias=True):
super(GroupLinearNew, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.groups = groups
self.weight = nn.Parameter(torch.Tensor(groups, out_features //
groups, in_features // groups))
if bias:
self.bias = nn.Parameter(torch.Tensor(groups, out_features //
groups))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
for weight in self.weight:
nn.init.kaiming_uniform_(weight, nonlinearity='linear')
if self.bias is not None:
nn.init.constant_(self.bias, 0)
def forward(self, input_0):
primals_3 = self.weight
primals_2 = self.bias
primals_1 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
Lakonik/EPro-PnP
|
GroupLinear
| false
| 8,433
|
[
"Apache-2.0"
] | 19
|
931df847190ce10eddd1dc3e3168ce1a2f295ffa
|
https://github.com/Lakonik/EPro-PnP/tree/931df847190ce10eddd1dc3e3168ce1a2f295ffa
|
GammaScaleMap
|
# 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/fk/cfkf6enhrdxuolzpcw3wzgmkvdxbf2yeamw6dswhcd7tiorzwjps.py
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, sub_2, s3, pow_1, mul, add_1, rt, nm, dn, truediv, k], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.rsub, aten.pow, aten.mul, aten.sqrt, aten.div]
# Source node to ATen node mapping:
# add_1 => add_1
# dn => mul_1
# k => add_3
# log => log
# log_1 => log_1
# mean => mean
# mean_1 => mean_1
# mul => mul
# nm => add_2
# pow_1 => pow_1
# rt => sqrt
# s => sub
# s3 => sub_1
# sub_2 => sub_2
# truediv => div
# x => add
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%mean,), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log_1, [1]), kwargs = {})
# %sub : [num_users=14] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %mean_1), kwargs = {})
# %sub_2 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (3.0, %sub), kwargs = {})
# %sub_1 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub, 3.0), kwargs = {})
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_1, 2), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, 24.0), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%pow_1, %mul), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%add_1,), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_2, %sqrt), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sub, 12.0), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_2, %mul_1), kwargs = {})
# %add_3 : [num_users=6] = call_function[target=torch.ops.aten.add.Tensor](args = (%div, 1e-07), kwargs = {})
triton_poi_fused_add_div_log_mean_mul_pow_rsub_sqrt_sub_0 = async_compile.triton('triton_poi_fused_add_div_log_mean_mul_pow_rsub_sqrt_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.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_div_log_mean_mul_pow_rsub_sqrt_sub_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_add_div_log_mean_mul_pow_rsub_sqrt_sub_0(in_out_ptr0, in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = (xindex // 16)
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (64*x1)), xmask)
tmp3 = tl.load(in_ptr0 + (16 + x0 + (64*x1)), xmask)
tmp6 = tl.load(in_ptr0 + (32 + x0 + (64*x1)), xmask)
tmp9 = tl.load(in_ptr0 + (48 + x0 + (64*x1)), xmask)
tmp1 = 1e-07
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = tmp2 + tmp4
tmp7 = tmp6 + tmp1
tmp8 = tmp5 + tmp7
tmp10 = tmp9 + tmp1
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tl_math.log(tmp13)
tmp15 = tl_math.log(tmp2)
tmp16 = tl_math.log(tmp4)
tmp17 = tmp15 + tmp16
tmp18 = tl_math.log(tmp7)
tmp19 = tmp17 + tmp18
tmp20 = tl_math.log(tmp10)
tmp21 = tmp19 + tmp20
tmp22 = tmp21 / tmp12
tmp23 = tmp14 - tmp22
tmp24 = 3.0
tmp25 = tmp23 - tmp24
tmp26 = tmp25 * tmp25
tmp27 = 24.0
tmp28 = tmp23 * tmp27
tmp29 = tmp26 + tmp28
tmp30 = libdevice.sqrt(tmp29)
tmp31 = tmp24 - tmp23
tmp32 = tmp31 + tmp30
tmp33 = 12.0
tmp34 = tmp23 * tmp33
tmp35 = tmp32 / tmp34
tmp36 = tmp35 + tmp1
tl.store(in_out_ptr0 + (x2), tmp35, xmask)
tl.store(out_ptr0 + (x2), tmp36, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cp/ccpzc7bnwy5ex77gkagui54sgmgom2lohbjngjpoztrk3ogip5hv.py
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k, log_2, sub_3, nm_1, reciprocal, z, zz, mul_2, reciprocal_1, a, truediv_1, b, mul_3, truediv_2, c, mul_4, truediv_3, d, e, pow_3, reciprocal_2, e_1, dn_1, truediv_5, k2], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
# Source node to ATen node mapping:
# a => sub_5
# b => sub_6
# c => add_5
# d => add_6
# dn_1 => sub_7
# e => div_4
# e_1 => add_7
# k => add_3
# k2 => sub_8
# log => log
# log_1 => log_1
# log_2 => log_2
# mean => mean
# mean_1 => mean_1
# mul_2 => mul_2
# mul_3 => mul_3
# mul_4 => mul_4
# nm_1 => sub_4
# pow_3 => pow_3
# reciprocal => reciprocal
# reciprocal_1 => reciprocal_1
# reciprocal_2 => reciprocal_2
# s => sub
# sub_3 => sub_3
# truediv_1 => div_1
# truediv_2 => div_2
# truediv_3 => div_3
# truediv_5 => div_5
# x => add
# z => add_4
# zz => pow_2
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%mean,), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log_1, [1]), kwargs = {})
# %sub : [num_users=14] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %mean_1), kwargs = {})
# %add_3 : [num_users=6] = call_function[target=torch.ops.aten.add.Tensor](args = (%div, 1e-07), kwargs = {})
# %log_2 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add_3,), kwargs = {})
# %sub_3 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log_2, %digamma), kwargs = {})
# %sub_4 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_3, %sub), kwargs = {})
# %reciprocal : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%add_3,), kwargs = {})
# %add_4 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_3, 1.0), kwargs = {})
# %pow_2 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_4, 2), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_2, 7.0), kwargs = {})
# %reciprocal_1 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_2,), kwargs = {})
# %sub_5 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_1), kwargs = {})
# %div_1 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_5, %pow_2), kwargs = {})
# %sub_6 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_1), kwargs = {})
# %mul_3 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_4, 3.0), kwargs = {})
# %div_2 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_6, %mul_3), kwargs = {})
# %add_5 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_2, 1.0), kwargs = {})
# %mul_4 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_4, 2.0), kwargs = {})
# %div_3 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_5, %mul_4), kwargs = {})
# %add_6 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_3, 1.0), kwargs = {})
# %div_4 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_6, %add_4), kwargs = {})
# %pow_3 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_3, 2.0), kwargs = {})
# %reciprocal_2 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_3,), kwargs = {})
# %add_7 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_4, %reciprocal_2), kwargs = {})
# %sub_7 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal, %add_7), kwargs = {})
# %div_5 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_4, %sub_7), kwargs = {})
# %sub_8 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_3, %div_5), kwargs = {})
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_1 = async_compile.triton('triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_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=[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_add_div_log_mean_mul_pow_reciprocal_rsub_sub_1', 'mutated_arg_names': ['in_out_ptr0'], '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_div_log_mean_mul_pow_reciprocal_rsub_sub_1(in_out_ptr0, 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
x0 = xindex % 16
x1 = (xindex // 16)
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp4 = tl.load(in_ptr0 + (x2), xmask)
tmp6 = tl.load(in_ptr1 + (x0 + (64*x1)), xmask)
tmp8 = tl.load(in_ptr1 + (16 + x0 + (64*x1)), xmask)
tmp11 = tl.load(in_ptr1 + (32 + x0 + (64*x1)), xmask)
tmp14 = tl.load(in_ptr1 + (48 + x0 + (64*x1)), xmask)
tmp30 = tl.load(in_ptr2 + (x2), xmask)
tmp1 = 1e-07
tmp2 = tmp0 + tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp3 - tmp4
tmp7 = tmp6 + tmp1
tmp9 = tmp8 + tmp1
tmp10 = tmp7 + tmp9
tmp12 = tmp11 + tmp1
tmp13 = tmp10 + tmp12
tmp15 = tmp14 + tmp1
tmp16 = tmp13 + tmp15
tmp17 = 4.0
tmp18 = tmp16 / tmp17
tmp19 = tl_math.log(tmp18)
tmp20 = tl_math.log(tmp7)
tmp21 = tl_math.log(tmp9)
tmp22 = tmp20 + tmp21
tmp23 = tl_math.log(tmp12)
tmp24 = tmp22 + tmp23
tmp25 = tl_math.log(tmp15)
tmp26 = tmp24 + tmp25
tmp27 = tmp26 / tmp17
tmp28 = tmp19 - tmp27
tmp29 = tmp5 - tmp28
tmp31 = tl.full([1], 1, tl.int32)
tmp32 = tmp31 / tmp30
tmp33 = 1.0
tmp34 = tmp30 + tmp33
tmp35 = tmp34 * tmp34
tmp36 = 7.0
tmp37 = tmp35 * tmp36
tmp38 = tmp31 / tmp37
tmp39 = 0.2
tmp40 = tmp39 - tmp38
tmp41 = tmp40 / tmp35
tmp42 = tmp33 - tmp41
tmp43 = 3.0
tmp44 = tmp34 * tmp43
tmp45 = tmp42 / tmp44
tmp46 = tmp45 + tmp33
tmp47 = 2.0
tmp48 = tmp34 * tmp47
tmp49 = tmp46 / tmp48
tmp50 = tmp49 + tmp33
tmp51 = tmp50 / tmp34
tmp52 = tmp30 * tmp30
tmp53 = tmp31 / tmp52
tmp54 = tmp51 + tmp53
tmp55 = tmp32 - tmp54
tmp56 = tmp29 / tmp55
tmp57 = tmp30 - tmp56
tl.store(in_out_ptr0 + (x2), tmp56, xmask)
tl.store(out_ptr0 + (x2), tmp57, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ac/cacnwf5nwa2gxch4csd5vbrr53xvpaxk2jshkrqlyka4taohph4n.py
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2, log_3, sub_9, nm_2, reciprocal_3, z_1, zz_1, mul_5, reciprocal_4, a_1, truediv_6, b_1, mul_6, truediv_7, c_1, mul_7, truediv_8, d_1, e_2, pow_5, reciprocal_5, e_3, dn_2, truediv_10, k2_1], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
# Source node to ATen node mapping:
# a_1 => sub_11
# b_1 => sub_12
# c_1 => add_9
# d_1 => add_10
# dn_2 => sub_13
# e_2 => div_9
# e_3 => add_11
# k2 => sub_8
# k2_1 => sub_14
# log => log
# log_1 => log_1
# log_3 => log_3
# mean => mean
# mean_1 => mean_1
# mul_5 => mul_5
# mul_6 => mul_6
# mul_7 => mul_7
# nm_2 => sub_10
# pow_5 => pow_5
# reciprocal_3 => reciprocal_3
# reciprocal_4 => reciprocal_4
# reciprocal_5 => reciprocal_5
# s => sub
# sub_9 => sub_9
# truediv_10 => div_10
# truediv_6 => div_6
# truediv_7 => div_7
# truediv_8 => div_8
# x => add
# z_1 => add_8
# zz_1 => pow_4
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%mean,), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log_1, [1]), kwargs = {})
# %sub : [num_users=14] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %mean_1), kwargs = {})
# %sub_8 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%add_3, %div_5), kwargs = {})
# %log_3 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sub_8,), kwargs = {})
# %sub_9 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log_3, %digamma_1), kwargs = {})
# %sub_10 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_9, %sub), kwargs = {})
# %reciprocal_3 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_8,), kwargs = {})
# %add_8 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_8, 1.0), kwargs = {})
# %pow_4 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_8, 2), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_4, 7.0), kwargs = {})
# %reciprocal_4 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_5,), kwargs = {})
# %sub_11 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_4), kwargs = {})
# %div_6 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_11, %pow_4), kwargs = {})
# %sub_12 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_6), kwargs = {})
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_8, 3.0), kwargs = {})
# %div_7 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_12, %mul_6), kwargs = {})
# %add_9 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_7, 1.0), kwargs = {})
# %mul_7 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_8, 2.0), kwargs = {})
# %div_8 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_9, %mul_7), kwargs = {})
# %add_10 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_8, 1.0), kwargs = {})
# %div_9 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_10, %add_8), kwargs = {})
# %pow_5 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_8, 2.0), kwargs = {})
# %reciprocal_5 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_5,), kwargs = {})
# %add_11 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_9, %reciprocal_5), kwargs = {})
# %sub_13 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal_3, %add_11), kwargs = {})
# %div_10 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_10, %sub_13), kwargs = {})
# %sub_14 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_8, %div_10), kwargs = {})
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2 = async_compile.triton('triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2', 'mutated_arg_names': ['in_out_ptr0'], '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_div_log_mean_mul_pow_reciprocal_rsub_sub_2(in_out_ptr0, 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
x2 = xindex
x0 = xindex % 16
x1 = (xindex // 16)
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x2), xmask)
tmp4 = tl.load(in_ptr1 + (x2), xmask)
tmp6 = tl.load(in_ptr2 + (x0 + (64*x1)), xmask)
tmp9 = tl.load(in_ptr2 + (16 + x0 + (64*x1)), xmask)
tmp12 = tl.load(in_ptr2 + (32 + x0 + (64*x1)), xmask)
tmp15 = tl.load(in_ptr2 + (48 + x0 + (64*x1)), xmask)
tmp31 = tl.load(in_ptr3 + (x2), xmask)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp3 - tmp4
tmp7 = 1e-07
tmp8 = tmp6 + tmp7
tmp10 = tmp9 + tmp7
tmp11 = tmp8 + tmp10
tmp13 = tmp12 + tmp7
tmp14 = tmp11 + tmp13
tmp16 = tmp15 + tmp7
tmp17 = tmp14 + tmp16
tmp18 = 4.0
tmp19 = tmp17 / tmp18
tmp20 = tl_math.log(tmp19)
tmp21 = tl_math.log(tmp8)
tmp22 = tl_math.log(tmp10)
tmp23 = tmp21 + tmp22
tmp24 = tl_math.log(tmp13)
tmp25 = tmp23 + tmp24
tmp26 = tl_math.log(tmp16)
tmp27 = tmp25 + tmp26
tmp28 = tmp27 / tmp18
tmp29 = tmp20 - tmp28
tmp30 = tmp5 - tmp29
tmp32 = tl.full([1], 1, tl.int32)
tmp33 = tmp32 / tmp31
tmp34 = 1.0
tmp35 = tmp31 + tmp34
tmp36 = tmp35 * tmp35
tmp37 = 7.0
tmp38 = tmp36 * tmp37
tmp39 = tmp32 / tmp38
tmp40 = 0.2
tmp41 = tmp40 - tmp39
tmp42 = tmp41 / tmp36
tmp43 = tmp34 - tmp42
tmp44 = 3.0
tmp45 = tmp35 * tmp44
tmp46 = tmp43 / tmp45
tmp47 = tmp46 + tmp34
tmp48 = 2.0
tmp49 = tmp35 * tmp48
tmp50 = tmp47 / tmp49
tmp51 = tmp50 + tmp34
tmp52 = tmp51 / tmp35
tmp53 = tmp31 * tmp31
tmp54 = tmp32 / tmp53
tmp55 = tmp52 + tmp54
tmp56 = tmp33 - tmp55
tmp57 = tmp30 / tmp56
tmp58 = tmp31 - tmp57
tl.store(in_out_ptr0 + (x2), tmp30, xmask)
tl.store(out_ptr0 + (x2), tmp58, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/23/c23dnslt4c2jdgtblpo22kyagp5vbtmqmmwxvvvgu7zpxpjojrbb.py
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, reciprocal_3, z_1, zz_1, mul_5, reciprocal_4, a_1, truediv_6, b_1, mul_6, truediv_7, c_1, mul_7, truediv_8, d_1, e_2, pow_5, reciprocal_5, e_3, dn_2, truediv_10, k2_1, log_4, sub_15, nm_3, reciprocal_6, z_2, zz_2, mul_8, reciprocal_7, a_2, truediv_11, b_2, mul_9, truediv_12, c_2, mul_10, truediv_13, d_2, e_4, pow_7, reciprocal_8, e_5, dn_3, truediv_15, k2_2], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
# Source node to ATen node mapping:
# a_1 => sub_11
# a_2 => sub_17
# b_1 => sub_12
# b_2 => sub_18
# c_1 => add_9
# c_2 => add_13
# d_1 => add_10
# d_2 => add_14
# dn_2 => sub_13
# dn_3 => sub_19
# e_2 => div_9
# e_3 => add_11
# e_4 => div_14
# e_5 => add_15
# k2_1 => sub_14
# k2_2 => sub_20
# log => log
# log_1 => log_1
# log_4 => log_4
# mean => mean
# mean_1 => mean_1
# mul_10 => mul_10
# mul_5 => mul_5
# mul_6 => mul_6
# mul_7 => mul_7
# mul_8 => mul_8
# mul_9 => mul_9
# nm_3 => sub_16
# pow_5 => pow_5
# pow_7 => pow_7
# reciprocal_3 => reciprocal_3
# reciprocal_4 => reciprocal_4
# reciprocal_5 => reciprocal_5
# reciprocal_6 => reciprocal_6
# reciprocal_7 => reciprocal_7
# reciprocal_8 => reciprocal_8
# s => sub
# sub_15 => sub_15
# truediv_10 => div_10
# truediv_11 => div_11
# truediv_12 => div_12
# truediv_13 => div_13
# truediv_15 => div_15
# truediv_6 => div_6
# truediv_7 => div_7
# truediv_8 => div_8
# x => add
# z_1 => add_8
# z_2 => add_12
# zz_1 => pow_4
# zz_2 => pow_6
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%mean,), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log_1, [1]), kwargs = {})
# %sub : [num_users=14] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %mean_1), kwargs = {})
# %reciprocal_3 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_8,), kwargs = {})
# %add_8 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_8, 1.0), kwargs = {})
# %pow_4 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_8, 2), kwargs = {})
# %mul_5 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_4, 7.0), kwargs = {})
# %reciprocal_4 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_5,), kwargs = {})
# %sub_11 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_4), kwargs = {})
# %div_6 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_11, %pow_4), kwargs = {})
# %sub_12 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_6), kwargs = {})
# %mul_6 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_8, 3.0), kwargs = {})
# %div_7 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_12, %mul_6), kwargs = {})
# %add_9 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_7, 1.0), kwargs = {})
# %mul_7 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_8, 2.0), kwargs = {})
# %div_8 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_9, %mul_7), kwargs = {})
# %add_10 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_8, 1.0), kwargs = {})
# %div_9 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_10, %add_8), kwargs = {})
# %pow_5 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_8, 2.0), kwargs = {})
# %reciprocal_5 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_5,), kwargs = {})
# %add_11 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_9, %reciprocal_5), kwargs = {})
# %sub_13 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal_3, %add_11), kwargs = {})
# %div_10 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_10, %sub_13), kwargs = {})
# %sub_14 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_8, %div_10), kwargs = {})
# %log_4 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sub_14,), kwargs = {})
# %sub_15 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log_4, %digamma_2), kwargs = {})
# %sub_16 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_15, %sub), kwargs = {})
# %reciprocal_6 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_14,), kwargs = {})
# %add_12 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_14, 1.0), kwargs = {})
# %pow_6 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_12, 2), kwargs = {})
# %mul_8 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_6, 7.0), kwargs = {})
# %reciprocal_7 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_8,), kwargs = {})
# %sub_17 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_7), kwargs = {})
# %div_11 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_17, %pow_6), kwargs = {})
# %sub_18 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_11), kwargs = {})
# %mul_9 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_12, 3.0), kwargs = {})
# %div_12 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_18, %mul_9), kwargs = {})
# %add_13 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_12, 1.0), kwargs = {})
# %mul_10 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_12, 2.0), kwargs = {})
# %div_13 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_13, %mul_10), kwargs = {})
# %add_14 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_13, 1.0), kwargs = {})
# %div_14 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_14, %add_12), kwargs = {})
# %pow_7 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_14, 2.0), kwargs = {})
# %reciprocal_8 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_7,), kwargs = {})
# %add_15 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_14, %reciprocal_8), kwargs = {})
# %sub_19 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal_6, %add_15), kwargs = {})
# %div_15 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_16, %sub_19), kwargs = {})
# %sub_20 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_14, %div_15), kwargs = {})
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_3 = async_compile.triton('triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_add_div_log_mean_mul_pow_reciprocal_rsub_sub_3', 'mutated_arg_names': ['in_out_ptr0'], '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_div_log_mean_mul_pow_reciprocal_rsub_sub_3(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
x1 = xindex % 16
x2 = (xindex // 16)
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp30 = tl.load(in_out_ptr0 + (x0), xmask)
tmp32 = tl.load(in_ptr2 + (x1 + (64*x2)), xmask)
tmp35 = tl.load(in_ptr2 + (16 + x1 + (64*x2)), xmask)
tmp38 = tl.load(in_ptr2 + (32 + x1 + (64*x2)), xmask)
tmp41 = tl.load(in_ptr2 + (48 + x1 + (64*x2)), xmask)
tmp57 = tl.load(in_ptr3 + (x0), xmask)
tmp2 = tl.full([1], 1, tl.int32)
tmp3 = tmp2 / tmp0
tmp4 = 1.0
tmp5 = tmp0 + tmp4
tmp6 = tmp5 * tmp5
tmp7 = 7.0
tmp8 = tmp6 * tmp7
tmp9 = tmp2 / tmp8
tmp10 = 0.2
tmp11 = tmp10 - tmp9
tmp12 = tmp11 / tmp6
tmp13 = tmp4 - tmp12
tmp14 = 3.0
tmp15 = tmp5 * tmp14
tmp16 = tmp13 / tmp15
tmp17 = tmp16 + tmp4
tmp18 = 2.0
tmp19 = tmp5 * tmp18
tmp20 = tmp17 / tmp19
tmp21 = tmp20 + tmp4
tmp22 = tmp21 / tmp5
tmp23 = tmp0 * tmp0
tmp24 = tmp2 / tmp23
tmp25 = tmp22 + tmp24
tmp26 = tmp3 - tmp25
tmp27 = tmp1 / tmp26
tmp28 = tmp0 - tmp27
tmp29 = tl_math.log(tmp28)
tmp31 = tmp29 - tmp30
tmp33 = 1e-07
tmp34 = tmp32 + tmp33
tmp36 = tmp35 + tmp33
tmp37 = tmp34 + tmp36
tmp39 = tmp38 + tmp33
tmp40 = tmp37 + tmp39
tmp42 = tmp41 + tmp33
tmp43 = tmp40 + tmp42
tmp44 = 4.0
tmp45 = tmp43 / tmp44
tmp46 = tl_math.log(tmp45)
tmp47 = tl_math.log(tmp34)
tmp48 = tl_math.log(tmp36)
tmp49 = tmp47 + tmp48
tmp50 = tl_math.log(tmp39)
tmp51 = tmp49 + tmp50
tmp52 = tl_math.log(tmp42)
tmp53 = tmp51 + tmp52
tmp54 = tmp53 / tmp44
tmp55 = tmp46 - tmp54
tmp56 = tmp31 - tmp55
tmp58 = tmp2 / tmp57
tmp59 = tmp57 + tmp4
tmp60 = tmp59 * tmp59
tmp61 = tmp60 * tmp7
tmp62 = tmp2 / tmp61
tmp63 = tmp10 - tmp62
tmp64 = tmp63 / tmp60
tmp65 = tmp4 - tmp64
tmp66 = tmp59 * tmp14
tmp67 = tmp65 / tmp66
tmp68 = tmp67 + tmp4
tmp69 = tmp59 * tmp18
tmp70 = tmp68 / tmp69
tmp71 = tmp70 + tmp4
tmp72 = tmp71 / tmp59
tmp73 = tmp57 * tmp57
tmp74 = tmp2 / tmp73
tmp75 = tmp72 + tmp74
tmp76 = tmp58 - tmp75
tmp77 = tmp56 / tmp76
tmp78 = tmp57 - tmp77
tl.store(in_out_ptr0 + (x0), tmp77, xmask)
tl.store(out_ptr0 + (x0), tmp78, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/5r/c5ravvqqldt4ehwikip7iu66obt35uy2ozmkhn7zqdu2h6ttu5as.py
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, reciprocal_9, z_3, zz_3, mul_11, reciprocal_10, a_3, truediv_16, b_3, mul_12, truediv_17, c_3, mul_13, truediv_18, d_3, e_6, pow_9, reciprocal_11, e_7, dn_4, truediv_20, k2_3, log_6, sub_27, nm_5, reciprocal_12, z_4, zz_4, mul_14, reciprocal_13, a_4, truediv_21, b_4, mul_15, truediv_22, c_4, mul_16, truediv_23, d_4, e_8, pow_11, reciprocal_14, e_9, dn_5, truediv_25, k2_4], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
# Source node to ATen node mapping:
# a_3 => sub_23
# a_4 => sub_29
# b_3 => sub_24
# b_4 => sub_30
# c_3 => add_17
# c_4 => add_21
# d_3 => add_18
# d_4 => add_22
# dn_4 => sub_25
# dn_5 => sub_31
# e_6 => div_19
# e_7 => add_19
# e_8 => div_24
# e_9 => add_23
# k2_3 => sub_26
# k2_4 => sub_32
# log => log
# log_1 => log_1
# log_6 => log_6
# mean => mean
# mean_1 => mean_1
# mul_11 => mul_11
# mul_12 => mul_12
# mul_13 => mul_13
# mul_14 => mul_14
# mul_15 => mul_15
# mul_16 => mul_16
# nm_5 => sub_28
# pow_11 => pow_11
# pow_9 => pow_9
# reciprocal_10 => reciprocal_10
# reciprocal_11 => reciprocal_11
# reciprocal_12 => reciprocal_12
# reciprocal_13 => reciprocal_13
# reciprocal_14 => reciprocal_14
# reciprocal_9 => reciprocal_9
# s => sub
# sub_27 => sub_27
# truediv_16 => div_16
# truediv_17 => div_17
# truediv_18 => div_18
# truediv_20 => div_20
# truediv_21 => div_21
# truediv_22 => div_22
# truediv_23 => div_23
# truediv_25 => div_25
# x => add
# z_3 => add_16
# z_4 => add_20
# zz_3 => pow_8
# zz_4 => pow_10
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%mean,), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log_1, [1]), kwargs = {})
# %sub : [num_users=14] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %mean_1), kwargs = {})
# %reciprocal_9 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_20,), kwargs = {})
# %add_16 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_20, 1.0), kwargs = {})
# %pow_8 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_16, 2), kwargs = {})
# %mul_11 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_8, 7.0), kwargs = {})
# %reciprocal_10 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_11,), kwargs = {})
# %sub_23 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_10), kwargs = {})
# %div_16 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_23, %pow_8), kwargs = {})
# %sub_24 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_16), kwargs = {})
# %mul_12 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_16, 3.0), kwargs = {})
# %div_17 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_24, %mul_12), kwargs = {})
# %add_17 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_17, 1.0), kwargs = {})
# %mul_13 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_16, 2.0), kwargs = {})
# %div_18 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_17, %mul_13), kwargs = {})
# %add_18 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_18, 1.0), kwargs = {})
# %div_19 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_18, %add_16), kwargs = {})
# %pow_9 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_20, 2.0), kwargs = {})
# %reciprocal_11 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_9,), kwargs = {})
# %add_19 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_19, %reciprocal_11), kwargs = {})
# %sub_25 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal_9, %add_19), kwargs = {})
# %div_20 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_22, %sub_25), kwargs = {})
# %sub_26 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_20, %div_20), kwargs = {})
# %log_6 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sub_26,), kwargs = {})
# %sub_27 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log_6, %digamma_4), kwargs = {})
# %sub_28 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_27, %sub), kwargs = {})
# %reciprocal_12 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_26,), kwargs = {})
# %add_20 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_26, 1.0), kwargs = {})
# %pow_10 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_20, 2), kwargs = {})
# %mul_14 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_10, 7.0), kwargs = {})
# %reciprocal_13 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_14,), kwargs = {})
# %sub_29 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_13), kwargs = {})
# %div_21 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_29, %pow_10), kwargs = {})
# %sub_30 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_21), kwargs = {})
# %mul_15 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_20, 3.0), kwargs = {})
# %div_22 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_30, %mul_15), kwargs = {})
# %add_21 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_22, 1.0), kwargs = {})
# %mul_16 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_20, 2.0), kwargs = {})
# %div_23 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_21, %mul_16), kwargs = {})
# %add_22 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_23, 1.0), kwargs = {})
# %div_24 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_22, %add_20), kwargs = {})
# %pow_11 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_26, 2.0), kwargs = {})
# %reciprocal_14 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_11,), kwargs = {})
# %add_23 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_24, %reciprocal_14), kwargs = {})
# %sub_31 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal_12, %add_23), kwargs = {})
# %div_25 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_28, %sub_31), kwargs = {})
# %sub_32 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_26, %div_25), kwargs = {})
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4 = async_compile.triton('triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32'}, 'device': 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_div_log_mean_mul_pow_reciprocal_rsub_sub_4', 'mutated_arg_names': ['in_out_ptr0'], '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_div_log_mean_mul_pow_reciprocal_rsub_sub_4(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
x1 = xindex % 16
x2 = (xindex // 16)
tmp0 = tl.load(in_out_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr0 + (x0), xmask)
tmp30 = tl.load(in_ptr1 + (x0), xmask)
tmp32 = tl.load(in_ptr2 + (x1 + (64*x2)), xmask)
tmp35 = tl.load(in_ptr2 + (16 + x1 + (64*x2)), xmask)
tmp38 = tl.load(in_ptr2 + (32 + x1 + (64*x2)), xmask)
tmp41 = tl.load(in_ptr2 + (48 + x1 + (64*x2)), xmask)
tmp57 = tl.load(in_ptr3 + (x0), xmask)
tmp2 = tl.full([1], 1, tl.int32)
tmp3 = tmp2 / tmp0
tmp4 = 1.0
tmp5 = tmp0 + tmp4
tmp6 = tmp5 * tmp5
tmp7 = 7.0
tmp8 = tmp6 * tmp7
tmp9 = tmp2 / tmp8
tmp10 = 0.2
tmp11 = tmp10 - tmp9
tmp12 = tmp11 / tmp6
tmp13 = tmp4 - tmp12
tmp14 = 3.0
tmp15 = tmp5 * tmp14
tmp16 = tmp13 / tmp15
tmp17 = tmp16 + tmp4
tmp18 = 2.0
tmp19 = tmp5 * tmp18
tmp20 = tmp17 / tmp19
tmp21 = tmp20 + tmp4
tmp22 = tmp21 / tmp5
tmp23 = tmp0 * tmp0
tmp24 = tmp2 / tmp23
tmp25 = tmp22 + tmp24
tmp26 = tmp3 - tmp25
tmp27 = tmp1 / tmp26
tmp28 = tmp0 - tmp27
tmp29 = tl_math.log(tmp28)
tmp31 = tmp29 - tmp30
tmp33 = 1e-07
tmp34 = tmp32 + tmp33
tmp36 = tmp35 + tmp33
tmp37 = tmp34 + tmp36
tmp39 = tmp38 + tmp33
tmp40 = tmp37 + tmp39
tmp42 = tmp41 + tmp33
tmp43 = tmp40 + tmp42
tmp44 = 4.0
tmp45 = tmp43 / tmp44
tmp46 = tl_math.log(tmp45)
tmp47 = tl_math.log(tmp34)
tmp48 = tl_math.log(tmp36)
tmp49 = tmp47 + tmp48
tmp50 = tl_math.log(tmp39)
tmp51 = tmp49 + tmp50
tmp52 = tl_math.log(tmp42)
tmp53 = tmp51 + tmp52
tmp54 = tmp53 / tmp44
tmp55 = tmp46 - tmp54
tmp56 = tmp31 - tmp55
tmp58 = tmp2 / tmp57
tmp59 = tmp57 + tmp4
tmp60 = tmp59 * tmp59
tmp61 = tmp60 * tmp7
tmp62 = tmp2 / tmp61
tmp63 = tmp10 - tmp62
tmp64 = tmp63 / tmp60
tmp65 = tmp4 - tmp64
tmp66 = tmp59 * tmp14
tmp67 = tmp65 / tmp66
tmp68 = tmp67 + tmp4
tmp69 = tmp59 * tmp18
tmp70 = tmp68 / tmp69
tmp71 = tmp70 + tmp4
tmp72 = tmp71 / tmp59
tmp73 = tmp57 * tmp57
tmp74 = tmp2 / tmp73
tmp75 = tmp72 + tmp74
tmp76 = tmp58 - tmp75
tmp77 = tmp56 / tmp76
tmp78 = tmp57 - tmp77
tl.store(in_out_ptr0 + (x0), tmp77, xmask)
tl.store(out_ptr0 + (x0), tmp78, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ar/carxneddoiwjpi7pqfjcyon5zit4yz2i5tujfyoqfyzxpc77xxtr.py
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2_8, log_11, sub_57, nm_10, reciprocal_27, z_9, zz_9, mul_29, reciprocal_28, a_9, truediv_46, b_9, mul_30, truediv_47, c_9, mul_31, truediv_48, d_9, e_18, pow_21, reciprocal_29, e_19, dn_10, truediv_50, k2_9, reciprocal_30, mean_2, th], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
# Source node to ATen node mapping:
# a_9 => sub_59
# b_9 => sub_60
# c_9 => add_41
# d_9 => add_42
# dn_10 => sub_61
# e_18 => div_49
# e_19 => add_43
# k2_8 => sub_56
# k2_9 => sub_62
# log => log
# log_1 => log_1
# log_11 => log_11
# mean => mean
# mean_1 => mean_1
# mean_2 => mean_2
# mul_29 => mul_29
# mul_30 => mul_30
# mul_31 => mul_31
# nm_10 => sub_58
# pow_21 => pow_21
# reciprocal_27 => reciprocal_27
# reciprocal_28 => reciprocal_28
# reciprocal_29 => reciprocal_29
# reciprocal_30 => reciprocal_30
# s => sub
# sub_57 => sub_57
# th => mul_32
# truediv_46 => div_46
# truediv_47 => div_47
# truediv_48 => div_48
# truediv_50 => div_50
# x => add
# z_9 => add_40
# zz_9 => pow_20
# Graph fragment:
# %add : [num_users=3] = call_function[target=torch.ops.aten.add.Tensor](args = (%arg0_1, 1e-07), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %log : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%mean,), kwargs = {})
# %log_1 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%add,), kwargs = {})
# %mean_1 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%log_1, [1]), kwargs = {})
# %sub : [num_users=14] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log, %mean_1), kwargs = {})
# %sub_56 : [num_users=6] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_50, %div_45), kwargs = {})
# %log_11 : [num_users=1] = call_function[target=torch.ops.aten.log.default](args = (%sub_56,), kwargs = {})
# %sub_57 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%log_11, %digamma_9), kwargs = {})
# %sub_58 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_57, %sub), kwargs = {})
# %reciprocal_27 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_56,), kwargs = {})
# %add_40 : [num_users=4] = call_function[target=torch.ops.aten.add.Tensor](args = (%sub_56, 1.0), kwargs = {})
# %pow_20 : [num_users=2] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%add_40, 2), kwargs = {})
# %mul_29 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%pow_20, 7.0), kwargs = {})
# %reciprocal_28 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%mul_29,), kwargs = {})
# %sub_59 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (0.2, %reciprocal_28), kwargs = {})
# %div_46 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_59, %pow_20), kwargs = {})
# %sub_60 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (1.0, %div_46), kwargs = {})
# %mul_30 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_40, 3.0), kwargs = {})
# %div_47 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_60, %mul_30), kwargs = {})
# %add_41 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_47, 1.0), kwargs = {})
# %mul_31 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%add_40, 2.0), kwargs = {})
# %div_48 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_41, %mul_31), kwargs = {})
# %add_42 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_48, 1.0), kwargs = {})
# %div_49 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_42, %add_40), kwargs = {})
# %pow_21 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%sub_56, 2.0), kwargs = {})
# %reciprocal_29 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%pow_21,), kwargs = {})
# %add_43 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%div_49, %reciprocal_29), kwargs = {})
# %sub_61 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%reciprocal_27, %add_43), kwargs = {})
# %div_50 : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub_58, %sub_61), kwargs = {})
# %sub_62 : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%sub_56, %div_50), kwargs = {})
# %reciprocal_30 : [num_users=1] = call_function[target=torch.ops.aten.reciprocal.default](args = (%sub_62,), kwargs = {})
# %mean_2 : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%add, [1]), kwargs = {})
# %mul_32 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%reciprocal_30, %mean_2), kwargs = {})
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_5 = async_compile.triton('triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_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: '*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_add_div_log_mean_mul_pow_reciprocal_rsub_sub_5', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], '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_div_log_mean_mul_pow_reciprocal_rsub_sub_5(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
x1 = (xindex // 16)
tmp0 = tl.load(in_out_ptr0 + (x2), xmask)
tmp1 = tl.load(in_ptr0 + (x2), xmask)
tmp4 = tl.load(in_ptr1 + (x2), xmask)
tmp6 = tl.load(in_ptr2 + (x0 + (64*x1)), xmask)
tmp9 = tl.load(in_ptr2 + (16 + x0 + (64*x1)), xmask)
tmp12 = tl.load(in_ptr2 + (32 + x0 + (64*x1)), xmask)
tmp15 = tl.load(in_ptr2 + (48 + x0 + (64*x1)), xmask)
tmp31 = tl.load(in_out_ptr1 + (x2), xmask)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp3 - tmp4
tmp7 = 1e-07
tmp8 = tmp6 + tmp7
tmp10 = tmp9 + tmp7
tmp11 = tmp8 + tmp10
tmp13 = tmp12 + tmp7
tmp14 = tmp11 + tmp13
tmp16 = tmp15 + tmp7
tmp17 = tmp14 + tmp16
tmp18 = 4.0
tmp19 = tmp17 / tmp18
tmp20 = tl_math.log(tmp19)
tmp21 = tl_math.log(tmp8)
tmp22 = tl_math.log(tmp10)
tmp23 = tmp21 + tmp22
tmp24 = tl_math.log(tmp13)
tmp25 = tmp23 + tmp24
tmp26 = tl_math.log(tmp16)
tmp27 = tmp25 + tmp26
tmp28 = tmp27 / tmp18
tmp29 = tmp20 - tmp28
tmp30 = tmp5 - tmp29
tmp32 = tl.full([1], 1, tl.int32)
tmp33 = tmp32 / tmp31
tmp34 = 1.0
tmp35 = tmp31 + tmp34
tmp36 = tmp35 * tmp35
tmp37 = 7.0
tmp38 = tmp36 * tmp37
tmp39 = tmp32 / tmp38
tmp40 = 0.2
tmp41 = tmp40 - tmp39
tmp42 = tmp41 / tmp36
tmp43 = tmp34 - tmp42
tmp44 = 3.0
tmp45 = tmp35 * tmp44
tmp46 = tmp43 / tmp45
tmp47 = tmp46 + tmp34
tmp48 = 2.0
tmp49 = tmp35 * tmp48
tmp50 = tmp47 / tmp49
tmp51 = tmp50 + tmp34
tmp52 = tmp51 / tmp35
tmp53 = tmp31 * tmp31
tmp54 = tmp32 / tmp53
tmp55 = tmp52 + tmp54
tmp56 = tmp33 - tmp55
tmp57 = tmp30 / tmp56
tmp58 = tmp31 - tmp57
tmp59 = tmp32 / tmp58
tmp60 = tmp59 * tmp19
tl.store(in_out_ptr1 + (x2), tmp60, 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)
buf1 = buf0; del buf0 # reuse
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, sub_2, s3, pow_1, mul, add_1, rt, nm, dn, truediv, k], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.rsub, aten.pow, aten.mul, aten.sqrt, aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_log_mean_mul_pow_rsub_sqrt_sub_0.run(buf1, arg0_1, buf2, 64, grid=grid(64), stream=stream0)
# Topologically Sorted Source Nodes: [k, digamma], Original ATen: [aten.add, aten.digamma]
buf3 = torch.ops.aten.digamma.default(buf2)
buf4 = buf3
del buf3
buf5 = buf1; del buf1 # reuse
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k, log_2, sub_3, nm_1, reciprocal, z, zz, mul_2, reciprocal_1, a, truediv_1, b, mul_3, truediv_2, c, mul_4, truediv_3, d, e, pow_3, reciprocal_2, e_1, dn_1, truediv_5, k2], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_1.run(buf5, buf4, arg0_1, buf2, buf6, 64, grid=grid(64), stream=stream0)
# Topologically Sorted Source Nodes: [k2, digamma_1], Original ATen: [aten.sub, aten.digamma]
buf7 = torch.ops.aten.digamma.default(buf6)
buf8 = buf7
del buf7
buf9 = buf2; del buf2 # reuse
buf10 = buf4; del buf4 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2, log_3, sub_9, nm_2, reciprocal_3, z_1, zz_1, mul_5, reciprocal_4, a_1, truediv_6, b_1, mul_6, truediv_7, c_1, mul_7, truediv_8, d_1, e_2, pow_5, reciprocal_5, e_3, dn_2, truediv_10, k2_1], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2.run(buf9, buf5, buf8, arg0_1, buf6, buf10, 64, grid=grid(64), stream=stream0)
del buf5
# Topologically Sorted Source Nodes: [reciprocal_3, z_1, zz_1, mul_5, reciprocal_4, a_1, truediv_6, b_1, mul_6, truediv_7, c_1, mul_7, truediv_8, d_1, e_2, pow_5, reciprocal_5, e_3, dn_2, truediv_10, k2_1, digamma_2], Original ATen: [aten.reciprocal, aten.add, aten.pow, aten.mul, aten.rsub, aten.div, aten.sub, aten.digamma]
buf11 = torch.ops.aten.digamma.default(buf10)
buf12 = buf11
del buf11
buf13 = buf12; del buf12 # reuse
buf14 = buf13; del buf13 # reuse
buf15 = buf8; del buf8 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, reciprocal_3, z_1, zz_1, mul_5, reciprocal_4, a_1, truediv_6, b_1, mul_6, truediv_7, c_1, mul_7, truediv_8, d_1, e_2, pow_5, reciprocal_5, e_3, dn_2, truediv_10, k2_1, log_4, sub_15, nm_3, reciprocal_6, z_2, zz_2, mul_8, reciprocal_7, a_2, truediv_11, b_2, mul_9, truediv_12, c_2, mul_10, truediv_13, d_2, e_4, pow_7, reciprocal_8, e_5, dn_3, truediv_15, k2_2], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_3.run(buf14, buf6, buf9, arg0_1, buf10, buf15, 64, grid=grid(64), stream=stream0)
del buf6
# Topologically Sorted Source Nodes: [k2_2, digamma_3], Original ATen: [aten.sub, aten.digamma]
buf16 = torch.ops.aten.digamma.default(buf15)
buf17 = buf16
del buf16
buf18 = buf10; del buf10 # reuse
buf19 = buf9; del buf9 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2_2, log_5, sub_21, nm_4, reciprocal_9, z_3, zz_3, mul_11, reciprocal_10, a_3, truediv_16, b_3, mul_12, truediv_17, c_3, mul_13, truediv_18, d_3, e_6, pow_9, reciprocal_11, e_7, dn_4, truediv_20, k2_3], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2.run(buf18, buf14, buf17, arg0_1, buf15, buf19, 64, grid=grid(64), stream=stream0)
del buf14
# Topologically Sorted Source Nodes: [reciprocal_9, z_3, zz_3, mul_11, reciprocal_10, a_3, truediv_16, b_3, mul_12, truediv_17, c_3, mul_13, truediv_18, d_3, e_6, pow_9, reciprocal_11, e_7, dn_4, truediv_20, k2_3, digamma_4], Original ATen: [aten.reciprocal, aten.add, aten.pow, aten.mul, aten.rsub, aten.div, aten.sub, aten.digamma]
buf20 = torch.ops.aten.digamma.default(buf19)
buf21 = buf20
del buf20
buf22 = buf15; del buf15 # reuse
buf23 = buf22; del buf22 # reuse
buf24 = buf17; del buf17 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, reciprocal_9, z_3, zz_3, mul_11, reciprocal_10, a_3, truediv_16, b_3, mul_12, truediv_17, c_3, mul_13, truediv_18, d_3, e_6, pow_9, reciprocal_11, e_7, dn_4, truediv_20, k2_3, log_6, sub_27, nm_5, reciprocal_12, z_4, zz_4, mul_14, reciprocal_13, a_4, truediv_21, b_4, mul_15, truediv_22, c_4, mul_16, truediv_23, d_4, e_8, pow_11, reciprocal_14, e_9, dn_5, truediv_25, k2_4], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4.run(buf23, buf18, buf21, arg0_1, buf19, buf24, 64, grid=grid(64), stream=stream0)
del buf18
# Topologically Sorted Source Nodes: [k2_4, digamma_5], Original ATen: [aten.sub, aten.digamma]
buf25 = torch.ops.aten.digamma.default(buf24)
buf26 = buf25
del buf25
buf27 = buf19; del buf19 # reuse
buf28 = buf21; del buf21 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2_4, log_7, sub_33, nm_6, reciprocal_15, z_5, zz_5, mul_17, reciprocal_16, a_5, truediv_26, b_5, mul_18, truediv_27, c_5, mul_19, truediv_28, d_5, e_10, pow_13, reciprocal_17, e_11, dn_6, truediv_30, k2_5], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2.run(buf27, buf23, buf26, arg0_1, buf24, buf28, 64, grid=grid(64), stream=stream0)
del buf23
# Topologically Sorted Source Nodes: [reciprocal_15, z_5, zz_5, mul_17, reciprocal_16, a_5, truediv_26, b_5, mul_18, truediv_27, c_5, mul_19, truediv_28, d_5, e_10, pow_13, reciprocal_17, e_11, dn_6, truediv_30, k2_5, digamma_6], Original ATen: [aten.reciprocal, aten.add, aten.pow, aten.mul, aten.rsub, aten.div, aten.sub, aten.digamma]
buf29 = torch.ops.aten.digamma.default(buf28)
buf30 = buf29
del buf29
buf31 = buf24; del buf24 # reuse
buf32 = buf31; del buf31 # reuse
buf33 = buf26; del buf26 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, reciprocal_15, z_5, zz_5, mul_17, reciprocal_16, a_5, truediv_26, b_5, mul_18, truediv_27, c_5, mul_19, truediv_28, d_5, e_10, pow_13, reciprocal_17, e_11, dn_6, truediv_30, k2_5, log_8, sub_39, nm_7, reciprocal_18, z_6, zz_6, mul_20, reciprocal_19, a_6, truediv_31, b_6, mul_21, truediv_32, c_6, mul_22, truediv_33, d_6, e_12, pow_15, reciprocal_20, e_13, dn_7, truediv_35, k2_6], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4.run(buf32, buf27, buf30, arg0_1, buf28, buf33, 64, grid=grid(64), stream=stream0)
del buf27
# Topologically Sorted Source Nodes: [k2_6, digamma_7], Original ATen: [aten.sub, aten.digamma]
buf34 = torch.ops.aten.digamma.default(buf33)
buf35 = buf34
del buf34
buf36 = buf28; del buf28 # reuse
buf37 = buf30; del buf30 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2_6, log_9, sub_45, nm_8, reciprocal_21, z_7, zz_7, mul_23, reciprocal_22, a_7, truediv_36, b_7, mul_24, truediv_37, c_7, mul_25, truediv_38, d_7, e_14, pow_17, reciprocal_23, e_15, dn_8, truediv_40, k2_7], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2.run(buf36, buf32, buf35, arg0_1, buf33, buf37, 64, grid=grid(64), stream=stream0)
del buf32
# Topologically Sorted Source Nodes: [reciprocal_21, z_7, zz_7, mul_23, reciprocal_22, a_7, truediv_36, b_7, mul_24, truediv_37, c_7, mul_25, truediv_38, d_7, e_14, pow_17, reciprocal_23, e_15, dn_8, truediv_40, k2_7, digamma_8], Original ATen: [aten.reciprocal, aten.add, aten.pow, aten.mul, aten.rsub, aten.div, aten.sub, aten.digamma]
buf38 = torch.ops.aten.digamma.default(buf37)
buf39 = buf38
del buf38
buf40 = buf33; del buf33 # reuse
buf41 = buf40; del buf40 # reuse
buf42 = buf35; del buf35 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, reciprocal_21, z_7, zz_7, mul_23, reciprocal_22, a_7, truediv_36, b_7, mul_24, truediv_37, c_7, mul_25, truediv_38, d_7, e_14, pow_17, reciprocal_23, e_15, dn_8, truediv_40, k2_7, log_10, sub_51, nm_9, reciprocal_24, z_8, zz_8, mul_26, reciprocal_25, a_8, truediv_41, b_8, mul_27, truediv_42, c_8, mul_28, truediv_43, d_8, e_16, pow_19, reciprocal_26, e_17, dn_9, truediv_45, k2_8], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4.run(buf41, buf36, buf39, arg0_1, buf37, buf42, 64, grid=grid(64), stream=stream0)
del buf36
del buf39
# Topologically Sorted Source Nodes: [k2_8, digamma_9], Original ATen: [aten.sub, aten.digamma]
buf43 = torch.ops.aten.digamma.default(buf42)
buf44 = buf43
del buf43
buf45 = buf37; del buf37 # reuse
buf46 = buf42; del buf42 # reuse
# Topologically Sorted Source Nodes: [x, mean, log, log_1, mean_1, s, k2_8, log_11, sub_57, nm_10, reciprocal_27, z_9, zz_9, mul_29, reciprocal_28, a_9, truediv_46, b_9, mul_30, truediv_47, c_9, mul_31, truediv_48, d_9, e_18, pow_21, reciprocal_29, e_19, dn_10, truediv_50, k2_9, reciprocal_30, mean_2, th], Original ATen: [aten.add, aten.mean, aten.log, aten.sub, aten.reciprocal, aten.pow, aten.mul, aten.rsub, aten.div]
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_5.run(buf45, buf46, buf41, buf44, arg0_1, 64, grid=grid(64), stream=stream0)
del arg0_1
del buf41
del buf44
del buf45
return (buf46, )
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.autograd
class GammaScaleMap(nn.Module):
"""
Compute Gamma Scale on a 4D tensor (The hard way). This acts as a standard PyTorch layer.
Gamma Scale is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) A tensor with only positive values. (After a ReLU)
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self, run_relu=False):
super(GammaScaleMap, self).__init__()
"""
SMOE Scale must take in values > 0. Optionally, we can run a ReLU to do that.
"""
if run_relu:
self.relu = nn.ReLU(inplace=False)
else:
self.relu = None
def _trigamma(self, x):
"""
We need this line since recursion is not good for x < 1.0
Note that we take + torch.reciprocal(x.pow(2)) at the end because:
trigamma(z) = trigamma(z + 1) + 1/z^2
"""
z = x + 1.0
zz = z.pow(2)
a = 0.2 - torch.reciprocal(7.0 * zz)
b = 1.0 - a / zz
c = 1.0 + b / (3.0 * z)
d = 1.0 + c / (2.0 * z)
e = d / z
e = e + torch.reciprocal(x.pow(2.0))
return e
def _k_update(self, k, s):
nm = torch.log(k) - torch.digamma(k) - s
dn = torch.reciprocal(k) - self._trigamma(k)
k2 = k - nm / dn
return k2
def _compute_k_est(self, x, i=10, dim=1):
"""
Calculate s
"""
s = torch.log(torch.mean(x, dim=dim)) - torch.mean(torch.log(x),
dim=dim)
"""
Get estimate of k to within 1.5%
NOTE: K gets smaller as log variance s increases
"""
s3 = s - 3.0
rt = torch.sqrt(s3.pow(2) + 24.0 * s)
nm = 3.0 - s + rt
dn = 12.0 * s
k = nm / dn + 1e-07
"""
Do i Newton-Raphson steps to get closer than 1.5%
For i=5 gets us within 4 or 5 decimal places
"""
for _ in range(i):
k = self._k_update(k, s)
return k
def forward(self, x):
assert torch.is_tensor(x), 'input must be a Torch Tensor'
assert len(x.size()) > 2, 'input must have at least three dims'
"""
If we do not have a convenient ReLU to pluck from, we can do it here
"""
if self.relu is not None:
x = self.relu(x)
"""
avoid log(0)
"""
x = x + 1e-07
k = self._compute_k_est(x)
th = torch.reciprocal(k) * torch.mean(x, dim=1)
return th
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
import torch.nn as nn
import torch.autograd
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_log_mean_mul_pow_rsub_sqrt_sub_0(in_out_ptr0,
in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 16
x1 = xindex // 16
x2 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 64 * x1), xmask)
tmp3 = tl.load(in_ptr0 + (16 + x0 + 64 * x1), xmask)
tmp6 = tl.load(in_ptr0 + (32 + x0 + 64 * x1), xmask)
tmp9 = tl.load(in_ptr0 + (48 + x0 + 64 * x1), xmask)
tmp1 = 1e-07
tmp2 = tmp0 + tmp1
tmp4 = tmp3 + tmp1
tmp5 = tmp2 + tmp4
tmp7 = tmp6 + tmp1
tmp8 = tmp5 + tmp7
tmp10 = tmp9 + tmp1
tmp11 = tmp8 + tmp10
tmp12 = 4.0
tmp13 = tmp11 / tmp12
tmp14 = tl_math.log(tmp13)
tmp15 = tl_math.log(tmp2)
tmp16 = tl_math.log(tmp4)
tmp17 = tmp15 + tmp16
tmp18 = tl_math.log(tmp7)
tmp19 = tmp17 + tmp18
tmp20 = tl_math.log(tmp10)
tmp21 = tmp19 + tmp20
tmp22 = tmp21 / tmp12
tmp23 = tmp14 - tmp22
tmp24 = 3.0
tmp25 = tmp23 - tmp24
tmp26 = tmp25 * tmp25
tmp27 = 24.0
tmp28 = tmp23 * tmp27
tmp29 = tmp26 + tmp28
tmp30 = libdevice.sqrt(tmp29)
tmp31 = tmp24 - tmp23
tmp32 = tmp31 + tmp30
tmp33 = 12.0
tmp34 = tmp23 * tmp33
tmp35 = tmp32 / tmp34
tmp36 = tmp35 + tmp1
tl.store(in_out_ptr0 + x2, tmp35, xmask)
tl.store(out_ptr0 + x2, tmp36, xmask)
@triton.jit
def triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_1(in_out_ptr0
, 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
x0 = xindex % 16
x1 = xindex // 16
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp4 = tl.load(in_ptr0 + x2, xmask)
tmp6 = tl.load(in_ptr1 + (x0 + 64 * x1), xmask)
tmp8 = tl.load(in_ptr1 + (16 + x0 + 64 * x1), xmask)
tmp11 = tl.load(in_ptr1 + (32 + x0 + 64 * x1), xmask)
tmp14 = tl.load(in_ptr1 + (48 + x0 + 64 * x1), xmask)
tmp30 = tl.load(in_ptr2 + x2, xmask)
tmp1 = 1e-07
tmp2 = tmp0 + tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp3 - tmp4
tmp7 = tmp6 + tmp1
tmp9 = tmp8 + tmp1
tmp10 = tmp7 + tmp9
tmp12 = tmp11 + tmp1
tmp13 = tmp10 + tmp12
tmp15 = tmp14 + tmp1
tmp16 = tmp13 + tmp15
tmp17 = 4.0
tmp18 = tmp16 / tmp17
tmp19 = tl_math.log(tmp18)
tmp20 = tl_math.log(tmp7)
tmp21 = tl_math.log(tmp9)
tmp22 = tmp20 + tmp21
tmp23 = tl_math.log(tmp12)
tmp24 = tmp22 + tmp23
tmp25 = tl_math.log(tmp15)
tmp26 = tmp24 + tmp25
tmp27 = tmp26 / tmp17
tmp28 = tmp19 - tmp27
tmp29 = tmp5 - tmp28
tmp31 = tl.full([1], 1, tl.int32)
tmp32 = tmp31 / tmp30
tmp33 = 1.0
tmp34 = tmp30 + tmp33
tmp35 = tmp34 * tmp34
tmp36 = 7.0
tmp37 = tmp35 * tmp36
tmp38 = tmp31 / tmp37
tmp39 = 0.2
tmp40 = tmp39 - tmp38
tmp41 = tmp40 / tmp35
tmp42 = tmp33 - tmp41
tmp43 = 3.0
tmp44 = tmp34 * tmp43
tmp45 = tmp42 / tmp44
tmp46 = tmp45 + tmp33
tmp47 = 2.0
tmp48 = tmp34 * tmp47
tmp49 = tmp46 / tmp48
tmp50 = tmp49 + tmp33
tmp51 = tmp50 / tmp34
tmp52 = tmp30 * tmp30
tmp53 = tmp31 / tmp52
tmp54 = tmp51 + tmp53
tmp55 = tmp32 - tmp54
tmp56 = tmp29 / tmp55
tmp57 = tmp30 - tmp56
tl.store(in_out_ptr0 + x2, tmp56, xmask)
tl.store(out_ptr0 + x2, tmp57, xmask)
@triton.jit
def triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2(in_out_ptr0
, 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
x2 = xindex
x0 = xindex % 16
x1 = xindex // 16
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x2, xmask)
tmp4 = tl.load(in_ptr1 + x2, xmask)
tmp6 = tl.load(in_ptr2 + (x0 + 64 * x1), xmask)
tmp9 = tl.load(in_ptr2 + (16 + x0 + 64 * x1), xmask)
tmp12 = tl.load(in_ptr2 + (32 + x0 + 64 * x1), xmask)
tmp15 = tl.load(in_ptr2 + (48 + x0 + 64 * x1), xmask)
tmp31 = tl.load(in_ptr3 + x2, xmask)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp3 - tmp4
tmp7 = 1e-07
tmp8 = tmp6 + tmp7
tmp10 = tmp9 + tmp7
tmp11 = tmp8 + tmp10
tmp13 = tmp12 + tmp7
tmp14 = tmp11 + tmp13
tmp16 = tmp15 + tmp7
tmp17 = tmp14 + tmp16
tmp18 = 4.0
tmp19 = tmp17 / tmp18
tmp20 = tl_math.log(tmp19)
tmp21 = tl_math.log(tmp8)
tmp22 = tl_math.log(tmp10)
tmp23 = tmp21 + tmp22
tmp24 = tl_math.log(tmp13)
tmp25 = tmp23 + tmp24
tmp26 = tl_math.log(tmp16)
tmp27 = tmp25 + tmp26
tmp28 = tmp27 / tmp18
tmp29 = tmp20 - tmp28
tmp30 = tmp5 - tmp29
tmp32 = tl.full([1], 1, tl.int32)
tmp33 = tmp32 / tmp31
tmp34 = 1.0
tmp35 = tmp31 + tmp34
tmp36 = tmp35 * tmp35
tmp37 = 7.0
tmp38 = tmp36 * tmp37
tmp39 = tmp32 / tmp38
tmp40 = 0.2
tmp41 = tmp40 - tmp39
tmp42 = tmp41 / tmp36
tmp43 = tmp34 - tmp42
tmp44 = 3.0
tmp45 = tmp35 * tmp44
tmp46 = tmp43 / tmp45
tmp47 = tmp46 + tmp34
tmp48 = 2.0
tmp49 = tmp35 * tmp48
tmp50 = tmp47 / tmp49
tmp51 = tmp50 + tmp34
tmp52 = tmp51 / tmp35
tmp53 = tmp31 * tmp31
tmp54 = tmp32 / tmp53
tmp55 = tmp52 + tmp54
tmp56 = tmp33 - tmp55
tmp57 = tmp30 / tmp56
tmp58 = tmp31 - tmp57
tl.store(in_out_ptr0 + x2, tmp30, xmask)
tl.store(out_ptr0 + x2, tmp58, xmask)
@triton.jit
def triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_3(in_out_ptr0
, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
x1 = xindex % 16
x2 = xindex // 16
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr1 + x0, xmask)
tmp30 = tl.load(in_out_ptr0 + x0, xmask)
tmp32 = tl.load(in_ptr2 + (x1 + 64 * x2), xmask)
tmp35 = tl.load(in_ptr2 + (16 + x1 + 64 * x2), xmask)
tmp38 = tl.load(in_ptr2 + (32 + x1 + 64 * x2), xmask)
tmp41 = tl.load(in_ptr2 + (48 + x1 + 64 * x2), xmask)
tmp57 = tl.load(in_ptr3 + x0, xmask)
tmp2 = tl.full([1], 1, tl.int32)
tmp3 = tmp2 / tmp0
tmp4 = 1.0
tmp5 = tmp0 + tmp4
tmp6 = tmp5 * tmp5
tmp7 = 7.0
tmp8 = tmp6 * tmp7
tmp9 = tmp2 / tmp8
tmp10 = 0.2
tmp11 = tmp10 - tmp9
tmp12 = tmp11 / tmp6
tmp13 = tmp4 - tmp12
tmp14 = 3.0
tmp15 = tmp5 * tmp14
tmp16 = tmp13 / tmp15
tmp17 = tmp16 + tmp4
tmp18 = 2.0
tmp19 = tmp5 * tmp18
tmp20 = tmp17 / tmp19
tmp21 = tmp20 + tmp4
tmp22 = tmp21 / tmp5
tmp23 = tmp0 * tmp0
tmp24 = tmp2 / tmp23
tmp25 = tmp22 + tmp24
tmp26 = tmp3 - tmp25
tmp27 = tmp1 / tmp26
tmp28 = tmp0 - tmp27
tmp29 = tl_math.log(tmp28)
tmp31 = tmp29 - tmp30
tmp33 = 1e-07
tmp34 = tmp32 + tmp33
tmp36 = tmp35 + tmp33
tmp37 = tmp34 + tmp36
tmp39 = tmp38 + tmp33
tmp40 = tmp37 + tmp39
tmp42 = tmp41 + tmp33
tmp43 = tmp40 + tmp42
tmp44 = 4.0
tmp45 = tmp43 / tmp44
tmp46 = tl_math.log(tmp45)
tmp47 = tl_math.log(tmp34)
tmp48 = tl_math.log(tmp36)
tmp49 = tmp47 + tmp48
tmp50 = tl_math.log(tmp39)
tmp51 = tmp49 + tmp50
tmp52 = tl_math.log(tmp42)
tmp53 = tmp51 + tmp52
tmp54 = tmp53 / tmp44
tmp55 = tmp46 - tmp54
tmp56 = tmp31 - tmp55
tmp58 = tmp2 / tmp57
tmp59 = tmp57 + tmp4
tmp60 = tmp59 * tmp59
tmp61 = tmp60 * tmp7
tmp62 = tmp2 / tmp61
tmp63 = tmp10 - tmp62
tmp64 = tmp63 / tmp60
tmp65 = tmp4 - tmp64
tmp66 = tmp59 * tmp14
tmp67 = tmp65 / tmp66
tmp68 = tmp67 + tmp4
tmp69 = tmp59 * tmp18
tmp70 = tmp68 / tmp69
tmp71 = tmp70 + tmp4
tmp72 = tmp71 / tmp59
tmp73 = tmp57 * tmp57
tmp74 = tmp2 / tmp73
tmp75 = tmp72 + tmp74
tmp76 = tmp58 - tmp75
tmp77 = tmp56 / tmp76
tmp78 = tmp57 - tmp77
tl.store(in_out_ptr0 + x0, tmp77, xmask)
tl.store(out_ptr0 + x0, tmp78, xmask)
@triton.jit
def triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4(in_out_ptr0
, in_ptr0, in_ptr1, in_ptr2, in_ptr3, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
x1 = xindex % 16
x2 = xindex // 16
tmp0 = tl.load(in_out_ptr0 + x0, xmask)
tmp1 = tl.load(in_ptr0 + x0, xmask)
tmp30 = tl.load(in_ptr1 + x0, xmask)
tmp32 = tl.load(in_ptr2 + (x1 + 64 * x2), xmask)
tmp35 = tl.load(in_ptr2 + (16 + x1 + 64 * x2), xmask)
tmp38 = tl.load(in_ptr2 + (32 + x1 + 64 * x2), xmask)
tmp41 = tl.load(in_ptr2 + (48 + x1 + 64 * x2), xmask)
tmp57 = tl.load(in_ptr3 + x0, xmask)
tmp2 = tl.full([1], 1, tl.int32)
tmp3 = tmp2 / tmp0
tmp4 = 1.0
tmp5 = tmp0 + tmp4
tmp6 = tmp5 * tmp5
tmp7 = 7.0
tmp8 = tmp6 * tmp7
tmp9 = tmp2 / tmp8
tmp10 = 0.2
tmp11 = tmp10 - tmp9
tmp12 = tmp11 / tmp6
tmp13 = tmp4 - tmp12
tmp14 = 3.0
tmp15 = tmp5 * tmp14
tmp16 = tmp13 / tmp15
tmp17 = tmp16 + tmp4
tmp18 = 2.0
tmp19 = tmp5 * tmp18
tmp20 = tmp17 / tmp19
tmp21 = tmp20 + tmp4
tmp22 = tmp21 / tmp5
tmp23 = tmp0 * tmp0
tmp24 = tmp2 / tmp23
tmp25 = tmp22 + tmp24
tmp26 = tmp3 - tmp25
tmp27 = tmp1 / tmp26
tmp28 = tmp0 - tmp27
tmp29 = tl_math.log(tmp28)
tmp31 = tmp29 - tmp30
tmp33 = 1e-07
tmp34 = tmp32 + tmp33
tmp36 = tmp35 + tmp33
tmp37 = tmp34 + tmp36
tmp39 = tmp38 + tmp33
tmp40 = tmp37 + tmp39
tmp42 = tmp41 + tmp33
tmp43 = tmp40 + tmp42
tmp44 = 4.0
tmp45 = tmp43 / tmp44
tmp46 = tl_math.log(tmp45)
tmp47 = tl_math.log(tmp34)
tmp48 = tl_math.log(tmp36)
tmp49 = tmp47 + tmp48
tmp50 = tl_math.log(tmp39)
tmp51 = tmp49 + tmp50
tmp52 = tl_math.log(tmp42)
tmp53 = tmp51 + tmp52
tmp54 = tmp53 / tmp44
tmp55 = tmp46 - tmp54
tmp56 = tmp31 - tmp55
tmp58 = tmp2 / tmp57
tmp59 = tmp57 + tmp4
tmp60 = tmp59 * tmp59
tmp61 = tmp60 * tmp7
tmp62 = tmp2 / tmp61
tmp63 = tmp10 - tmp62
tmp64 = tmp63 / tmp60
tmp65 = tmp4 - tmp64
tmp66 = tmp59 * tmp14
tmp67 = tmp65 / tmp66
tmp68 = tmp67 + tmp4
tmp69 = tmp59 * tmp18
tmp70 = tmp68 / tmp69
tmp71 = tmp70 + tmp4
tmp72 = tmp71 / tmp59
tmp73 = tmp57 * tmp57
tmp74 = tmp2 / tmp73
tmp75 = tmp72 + tmp74
tmp76 = tmp58 - tmp75
tmp77 = tmp56 / tmp76
tmp78 = tmp57 - tmp77
tl.store(in_out_ptr0 + x0, tmp77, xmask)
tl.store(out_ptr0 + x0, tmp78, xmask)
@triton.jit
def triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_5(in_out_ptr0
, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
x1 = xindex // 16
tmp0 = tl.load(in_out_ptr0 + x2, xmask)
tmp1 = tl.load(in_ptr0 + x2, xmask)
tmp4 = tl.load(in_ptr1 + x2, xmask)
tmp6 = tl.load(in_ptr2 + (x0 + 64 * x1), xmask)
tmp9 = tl.load(in_ptr2 + (16 + x0 + 64 * x1), xmask)
tmp12 = tl.load(in_ptr2 + (32 + x0 + 64 * x1), xmask)
tmp15 = tl.load(in_ptr2 + (48 + x0 + 64 * x1), xmask)
tmp31 = tl.load(in_out_ptr1 + x2, xmask)
tmp2 = tmp0 - tmp1
tmp3 = tl_math.log(tmp2)
tmp5 = tmp3 - tmp4
tmp7 = 1e-07
tmp8 = tmp6 + tmp7
tmp10 = tmp9 + tmp7
tmp11 = tmp8 + tmp10
tmp13 = tmp12 + tmp7
tmp14 = tmp11 + tmp13
tmp16 = tmp15 + tmp7
tmp17 = tmp14 + tmp16
tmp18 = 4.0
tmp19 = tmp17 / tmp18
tmp20 = tl_math.log(tmp19)
tmp21 = tl_math.log(tmp8)
tmp22 = tl_math.log(tmp10)
tmp23 = tmp21 + tmp22
tmp24 = tl_math.log(tmp13)
tmp25 = tmp23 + tmp24
tmp26 = tl_math.log(tmp16)
tmp27 = tmp25 + tmp26
tmp28 = tmp27 / tmp18
tmp29 = tmp20 - tmp28
tmp30 = tmp5 - tmp29
tmp32 = tl.full([1], 1, tl.int32)
tmp33 = tmp32 / tmp31
tmp34 = 1.0
tmp35 = tmp31 + tmp34
tmp36 = tmp35 * tmp35
tmp37 = 7.0
tmp38 = tmp36 * tmp37
tmp39 = tmp32 / tmp38
tmp40 = 0.2
tmp41 = tmp40 - tmp39
tmp42 = tmp41 / tmp36
tmp43 = tmp34 - tmp42
tmp44 = 3.0
tmp45 = tmp35 * tmp44
tmp46 = tmp43 / tmp45
tmp47 = tmp46 + tmp34
tmp48 = 2.0
tmp49 = tmp35 * tmp48
tmp50 = tmp47 / tmp49
tmp51 = tmp50 + tmp34
tmp52 = tmp51 / tmp35
tmp53 = tmp31 * tmp31
tmp54 = tmp32 / tmp53
tmp55 = tmp52 + tmp54
tmp56 = tmp33 - tmp55
tmp57 = tmp30 / tmp56
tmp58 = tmp31 - tmp57
tmp59 = tmp32 / tmp58
tmp60 = tmp59 * tmp19
tl.store(in_out_ptr1 + x2, tmp60, 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)
buf1 = buf0
del buf0
buf2 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_div_log_mean_mul_pow_rsub_sqrt_sub_0[grid(64)](
buf1, arg0_1, buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
buf3 = torch.ops.aten.digamma.default(buf2)
buf4 = buf3
del buf3
buf5 = buf1
del buf1
buf6 = empty_strided_cuda((4, 4, 4), (16, 4, 1), torch.float32)
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_1[grid
(64)](buf5, buf4, arg0_1, buf2, buf6, 64, XBLOCK=64, num_warps=
1, num_stages=1)
buf7 = torch.ops.aten.digamma.default(buf6)
buf8 = buf7
del buf7
buf9 = buf2
del buf2
buf10 = buf4
del buf4
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2[grid
(64)](buf9, buf5, buf8, arg0_1, buf6, buf10, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf5
buf11 = torch.ops.aten.digamma.default(buf10)
buf12 = buf11
del buf11
buf13 = buf12
del buf12
buf14 = buf13
del buf13
buf15 = buf8
del buf8
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_3[grid
(64)](buf14, buf6, buf9, arg0_1, buf10, buf15, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf6
buf16 = torch.ops.aten.digamma.default(buf15)
buf17 = buf16
del buf16
buf18 = buf10
del buf10
buf19 = buf9
del buf9
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2[grid
(64)](buf18, buf14, buf17, arg0_1, buf15, buf19, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf14
buf20 = torch.ops.aten.digamma.default(buf19)
buf21 = buf20
del buf20
buf22 = buf15
del buf15
buf23 = buf22
del buf22
buf24 = buf17
del buf17
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4[grid
(64)](buf23, buf18, buf21, arg0_1, buf19, buf24, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf18
buf25 = torch.ops.aten.digamma.default(buf24)
buf26 = buf25
del buf25
buf27 = buf19
del buf19
buf28 = buf21
del buf21
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2[grid
(64)](buf27, buf23, buf26, arg0_1, buf24, buf28, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf23
buf29 = torch.ops.aten.digamma.default(buf28)
buf30 = buf29
del buf29
buf31 = buf24
del buf24
buf32 = buf31
del buf31
buf33 = buf26
del buf26
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4[grid
(64)](buf32, buf27, buf30, arg0_1, buf28, buf33, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf27
buf34 = torch.ops.aten.digamma.default(buf33)
buf35 = buf34
del buf34
buf36 = buf28
del buf28
buf37 = buf30
del buf30
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_2[grid
(64)](buf36, buf32, buf35, arg0_1, buf33, buf37, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf32
buf38 = torch.ops.aten.digamma.default(buf37)
buf39 = buf38
del buf38
buf40 = buf33
del buf33
buf41 = buf40
del buf40
buf42 = buf35
del buf35
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_4[grid
(64)](buf41, buf36, buf39, arg0_1, buf37, buf42, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del buf36
del buf39
buf43 = torch.ops.aten.digamma.default(buf42)
buf44 = buf43
del buf43
buf45 = buf37
del buf37
buf46 = buf42
del buf42
triton_poi_fused_add_div_log_mean_mul_pow_reciprocal_rsub_sub_5[grid
(64)](buf45, buf46, buf41, buf44, arg0_1, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del arg0_1
del buf41
del buf44
del buf45
return buf46,
class GammaScaleMapNew(nn.Module):
"""
Compute Gamma Scale on a 4D tensor (The hard way). This acts as a standard PyTorch layer.
Gamma Scale is computed independantly for each batch item at each location x,y
Input should be:
(1) A tensor of size [batch x channels x height x width]
(2) A tensor with only positive values. (After a ReLU)
Output is a 3D tensor of size [batch x height x width]
"""
def __init__(self, run_relu=False):
super(GammaScaleMapNew, self).__init__()
"""
SMOE Scale must take in values > 0. Optionally, we can run a ReLU to do that.
"""
if run_relu:
self.relu = nn.ReLU(inplace=False)
else:
self.relu = None
def _trigamma(self, x):
"""
We need this line since recursion is not good for x < 1.0
Note that we take + torch.reciprocal(x.pow(2)) at the end because:
trigamma(z) = trigamma(z + 1) + 1/z^2
"""
z = x + 1.0
zz = z.pow(2)
a = 0.2 - torch.reciprocal(7.0 * zz)
b = 1.0 - a / zz
c = 1.0 + b / (3.0 * z)
d = 1.0 + c / (2.0 * z)
e = d / z
e = e + torch.reciprocal(x.pow(2.0))
return e
def _k_update(self, k, s):
nm = torch.log(k) - torch.digamma(k) - s
dn = torch.reciprocal(k) - self._trigamma(k)
k2 = k - nm / dn
return k2
def _compute_k_est(self, x, i=10, dim=1):
"""
Calculate s
"""
s = torch.log(torch.mean(x, dim=dim)) - torch.mean(torch.log(x),
dim=dim)
"""
Get estimate of k to within 1.5%
NOTE: K gets smaller as log variance s increases
"""
s3 = s - 3.0
rt = torch.sqrt(s3.pow(2) + 24.0 * s)
nm = 3.0 - s + rt
dn = 12.0 * s
k = nm / dn + 1e-07
"""
Do i Newton-Raphson steps to get closer than 1.5%
For i=5 gets us within 4 or 5 decimal places
"""
for _ in range(i):
k = self._k_update(k, s)
return k
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
GammaScaleMap
| false
| 8,434
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
L2Norm
|
# 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/5x/c5xvxdio3xjphaookra3l26tpxdd6tqkav2zx33jmjfl7ffs6k6i.py
# Topologically Sorted Source Nodes: [pow_1, sum_1, sqrt, norm, x, out], Original ATen: [aten.pow, aten.sum, aten.sqrt, aten.add, aten.div, aten.mul]
# Source node to ATen node mapping:
# norm => add
# out => mul
# pow_1 => pow_1
# sqrt => sqrt
# sum_1 => sum_1
# x => div
# Graph fragment:
# %pow_1 : [num_users=1] = call_function[target=torch.ops.aten.pow.Tensor_Scalar](args = (%primals_1, 2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.dim_IntList](args = (%pow_1, [1], True), kwargs = {})
# %sqrt : [num_users=1] = call_function[target=torch.ops.aten.sqrt.default](args = (%sum_1,), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%sqrt, 1e-10), kwargs = {})
# %div : [num_users=3] = call_function[target=torch.ops.aten.div.Tensor](args = (%primals_1, %add), kwargs = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%expand, %div), kwargs = {})
# %copy_ : [num_users=0] = call_function[target=torch.ops.aten.copy_.default](args = (%primals_1, %div), kwargs = {})
triton_poi_fused_add_div_mul_pow_sqrt_sum_0 = async_compile.triton('triton_poi_fused_add_div_mul_pow_sqrt_sum_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*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_add_div_mul_pow_sqrt_sum_0', 'mutated_arg_names': ['in_ptr0', 'out_ptr2'], '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_div_mul_pow_sqrt_sum_0(in_ptr0, in_ptr1, out_ptr0, out_ptr1, out_ptr2, xnumel, XBLOCK : tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
x1 = (xindex // 16) % 4
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 + (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-10
tmp14 = tmp12 + tmp13
tmp15 = tmp0 / tmp14
tmp17 = tmp16 * tmp15
tl.store(out_ptr0 + (x3), tmp15, xmask)
tl.store(out_ptr1 + (x3), tmp17, xmask)
tl.store(out_ptr2 + (x3), 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, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [pow_1, sum_1, sqrt, norm, x, out], Original ATen: [aten.pow, aten.sum, aten.sqrt, aten.add, aten.div, aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_mul_pow_sqrt_sum_0.run(primals_1, primals_2, buf0, buf1, primals_1, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_2
return (buf1, buf0, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
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)
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 math import sqrt as sqrt
from itertools import product as product
import torch.nn as nn
import torch.nn.init as init
class L2Norm(nn.Module):
def __init__(self, n_channels, scale):
super(L2Norm, self).__init__()
self.n_channels = n_channels
self.gamma = scale or None
self.eps = 1e-10
self.weight = nn.Parameter(torch.Tensor(self.n_channels))
self.reset_parameters()
def reset_parameters(self):
init.constant(self.weight, self.gamma)
def forward(self, x):
norm = x.pow(2).sum(dim=1, keepdim=True).sqrt() + self.eps
x /= norm
out = self.weight.unsqueeze(0).unsqueeze(2).unsqueeze(3).expand_as(x
) * x
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'n_channels': 4, 'scale': 1.0}]
|
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 math import sqrt as sqrt
from itertools import product as product
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
@triton.jit
def triton_poi_fused_add_div_mul_pow_sqrt_sum_0(in_ptr0, in_ptr1, out_ptr0,
out_ptr1, out_ptr2, xnumel, XBLOCK: tl.constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
x1 = xindex // 16 % 4
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 + 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-10
tmp14 = tmp12 + tmp13
tmp15 = tmp0 / tmp14
tmp17 = tmp16 * tmp15
tl.store(out_ptr0 + x3, tmp15, xmask)
tl.store(out_ptr1 + x3, tmp17, xmask)
tl.store(out_ptr2 + x3, tmp15, 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, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
buf1 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_add_div_mul_pow_sqrt_sum_0[grid(256)](primals_1,
primals_2, buf0, buf1, primals_1, 256, XBLOCK=256, num_warps=4,
num_stages=1)
del primals_1
del primals_2
return buf1, buf0
class L2NormNew(nn.Module):
def __init__(self, n_channels, scale):
super(L2NormNew, self).__init__()
self.n_channels = n_channels
self.gamma = scale or None
self.eps = 1e-10
self.weight = nn.Parameter(torch.Tensor(self.n_channels))
self.reset_parameters()
def reset_parameters(self):
init.constant(self.weight, self.gamma)
def forward(self, input_0):
primals_2 = self.weight
primals_1 = input_0
output = call([primals_1, primals_2])
return output[0]
|
Kalana304/realtime-action-detection
|
L2Norm
| false
| 8,435
|
[
"MIT"
] | 26
|
a40178c749d60c135290c40a8ac658bac253f0d4
|
https://github.com/Kalana304/realtime-action-detection/tree/a40178c749d60c135290c40a8ac658bac253f0d4
|
AdaptiveInstanceNorm
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/62/c62ipoefw56tbmppuqtdlczqscl2iztyp62s2k32xsqyn2fpedeh.py
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# weight => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, 0.7071067811865476), 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=[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_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 = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.7071067811865476
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/wl/cwlt4vfun3nfpvifiuqyv24qaha4ukioermi2tj5iwtnycfpzlra.py
# Topologically Sorted Source Nodes: [out, mul_1, out_1], Original ATen: [aten._native_batch_norm_legit, aten.mul, aten.add]
# Source node to ATen node mapping:
# mul_1 => mul_2
# out => add, rsqrt, var_mean
# out_1 => add_1
# Graph fragment:
# %var_mean : [num_users=2] = call_function[target=torch.ops.aten.var_mean.correction](args = (%view, [0, 2, 3]), kwargs = {correction: 0, keepdim: True})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%getitem_2, 1e-05), kwargs = {})
# %rsqrt : [num_users=2] = call_function[target=torch.ops.aten.rsqrt.default](args = (%add,), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%getitem, %view_1), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %getitem_1), kwargs = {})
triton_per_fused__native_batch_norm_legit_add_mul_1 = async_compile.triton('triton_per_fused__native_batch_norm_legit_add_mul_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[16, 16],
reduction_hint=ReductionHint.INNER,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: '*fp32', 5: '*fp32', 6: 'i32', 7: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_per_fused__native_batch_norm_legit_add_mul_1', 'mutated_arg_names': ['in_out_ptr0'], 'no_x_dim': False, 'num_load': 5, 'num_reduction': 4, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False}
)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_mul_1(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 16
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
x2 = xindex % 4
x3 = (xindex // 4)
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp22 = tl.load(in_ptr1 + (x2 + (8*x3)), xmask, eviction_policy='evict_last')
tmp23 = tl.load(in_ptr2 + (x2), xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (4 + x2 + (8*x3)), xmask, eviction_policy='evict_last')
tmp29 = tl.load(in_ptr2 + (4 + x2), xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = 16.0
tmp18 = tmp16 / tmp17
tmp19 = 1e-05
tmp20 = tmp18 + tmp19
tmp21 = libdevice.rsqrt(tmp20)
tmp24 = tmp22 + tmp23
tmp25 = tmp0 - tmp10
tmp26 = tmp25 * tmp21
tmp27 = tmp24 * tmp26
tmp30 = tmp28 + tmp29
tmp31 = tmp27 + tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + (x0), tmp21, xmask)
tl.store(out_ptr1 + (r1 + (16*x0)), tmp31, xmask)
tl.store(out_ptr0 + (x0), tmp10, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (8, 4), (4, 1))
assert_size_stride(primals_2, (8, ), (1, ))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((8, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(primals_1, buf0, 32, grid=grid(32), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
# Topologically Sorted Source Nodes: [], Original ATen: []
extern_kernels.mm(primals_3, reinterpret_tensor(buf0, (4, 8), (1, 4), 0), out=buf1)
buf2 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 1, 1), torch.float32)
buf3 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 16, 16), torch.float32)
buf5 = reinterpret_tensor(buf3, (1, 16, 1, 1), (16, 1, 1, 1), 0); del buf3 # reuse
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out, mul_1, out_1], Original ATen: [aten._native_batch_norm_legit, aten.mul, aten.add]
triton_per_fused__native_batch_norm_legit_add_mul_1.run(buf5, primals_4, buf1, primals_2, buf2, buf6, 16, 16, grid=grid(16), stream=stream0)
del buf1
del primals_2
return (buf6, buf0, primals_3, primals_4, buf2, buf5, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((8, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((8, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
from math import sqrt
def equal_lr(module, name='weight'):
EqualLR.apply(module, name)
return module
class EqualLR:
def __init__(self, name):
self.name = name
def compute_weight(self, module):
weight = getattr(module, self.name + '_orig')
fan_in = weight.data.size(1) * weight.data[0][0].numel()
return weight * sqrt(2 / fan_in)
@staticmethod
def apply(module, name):
fn = EqualLR(name)
weight = getattr(module, name)
del module._parameters[name]
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
module.register_forward_pre_hook(fn)
return fn
def __call__(self, module, input):
weight = self.compute_weight(module)
setattr(module, self.name, weight)
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim):
super().__init__()
linear = nn.Linear(in_dim, out_dim)
linear.weight.data.normal_()
linear.bias.data.zero_()
self.linear = equal_lr(linear)
def forward(self, input):
return self.linear(input)
class AdaptiveInstanceNorm(nn.Module):
def __init__(self, in_channel, style_dim):
super().__init__()
self.norm = nn.InstanceNorm2d(in_channel)
self.style = EqualLinear(style_dim, in_channel * 2)
self.style.linear.bias.data[:in_channel] = 1
self.style.linear.bias.data[in_channel:] = 0
def forward(self, input, style):
style = self.style(style).unsqueeze(2).unsqueeze(3)
gamma, beta = style.chunk(chunks=2, dim=1)
out = self.norm(input)
out = gamma * out + beta
return out
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'in_channel': 4, 'style_dim': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
import torch.nn as nn
from math import sqrt
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 = 32
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.7071067811865476
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_per_fused__native_batch_norm_legit_add_mul_1(in_out_ptr0,
in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, xnumel, rnumel, XBLOCK:
tl.constexpr):
xnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
x2 = xindex % 4
x3 = xindex // 4
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp22 = tl.load(in_ptr1 + (x2 + 8 * x3), xmask, eviction_policy=
'evict_last')
tmp23 = tl.load(in_ptr2 + x2, xmask, eviction_policy='evict_last')
tmp28 = tl.load(in_ptr1 + (4 + x2 + 8 * x3), xmask, eviction_policy=
'evict_last')
tmp29 = tl.load(in_ptr2 + (4 + x2), xmask, eviction_policy='evict_last')
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tl.where(xmask, tmp1, 0)
tmp4 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp6 = tl.where(xmask, tmp4, 0)
tmp7 = tl.sum(tmp6, 1)[:, None]
tmp8 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp9 = tmp8.to(tl.float32)
tmp10 = tmp7 / tmp9
tmp11 = tmp1 - tmp10
tmp12 = tmp11 * tmp11
tmp13 = tl.broadcast_to(tmp12, [XBLOCK, RBLOCK])
tmp15 = tl.where(xmask, tmp13, 0)
tmp16 = tl.sum(tmp15, 1)[:, None]
tmp17 = 16.0
tmp18 = tmp16 / tmp17
tmp19 = 1e-05
tmp20 = tmp18 + tmp19
tmp21 = libdevice.rsqrt(tmp20)
tmp24 = tmp22 + tmp23
tmp25 = tmp0 - tmp10
tmp26 = tmp25 * tmp21
tmp27 = tmp24 * tmp26
tmp30 = tmp28 + tmp29
tmp31 = tmp27 + tmp30
tl.debug_barrier()
tl.store(in_out_ptr0 + x0, tmp21, xmask)
tl.store(out_ptr1 + (r1 + 16 * x0), tmp31, xmask)
tl.store(out_ptr0 + x0, tmp10, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (8, 4), (4, 1))
assert_size_stride(primals_2, (8,), (1,))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((8, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(32)](primals_1, buf0, 32, XBLOCK=32,
num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 8), (8, 1), torch.float32)
extern_kernels.mm(primals_3, reinterpret_tensor(buf0, (4, 8), (1, 4
), 0), out=buf1)
buf2 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 1, 1), torch.float32)
buf3 = empty_strided_cuda((1, 16, 1, 1), (16, 1, 16, 16), torch.float32
)
buf5 = reinterpret_tensor(buf3, (1, 16, 1, 1), (16, 1, 1, 1), 0)
del buf3
buf6 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
triton_per_fused__native_batch_norm_legit_add_mul_1[grid(16)](buf5,
primals_4, buf1, primals_2, buf2, buf6, 16, 16, XBLOCK=8,
num_warps=2, num_stages=1)
del buf1
del primals_2
return buf6, buf0, primals_3, primals_4, buf2, buf5
def equal_lr(module, name='weight'):
EqualLR.apply(module, name)
return module
class EqualLR:
def __init__(self, name):
self.name = name
def compute_weight(self, module):
weight = getattr(module, self.name + '_orig')
fan_in = weight.data.size(1) * weight.data[0][0].numel()
return weight * sqrt(2 / fan_in)
@staticmethod
def apply(module, name):
fn = EqualLR(name)
weight = getattr(module, name)
del module._parameters[name]
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
module.register_forward_pre_hook(fn)
return fn
def __call__(self, module, input):
weight = self.compute_weight(module)
setattr(module, self.name, weight)
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim):
super().__init__()
linear = nn.Linear(in_dim, out_dim)
linear.weight.data.normal_()
linear.bias.data.zero_()
self.linear = equal_lr(linear)
def forward(self, input):
return self.linear(input)
class AdaptiveInstanceNormNew(nn.Module):
def __init__(self, in_channel, style_dim):
super().__init__()
self.norm = nn.InstanceNorm2d(in_channel)
self.style = EqualLinear(style_dim, in_channel * 2)
self.style.linear.bias.data[:in_channel] = 1
self.style.linear.bias.data[in_channel:] = 0
def forward(self, input_0, input_1):
primals_2 = self.style.linear.bias
primals_1 = self.style.linear.weight_orig
primals_4 = input_0
primals_3 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0]
|
KwonGihyun/DiagonalGAN
|
AdaptiveInstanceNorm
| false
| 8,437
|
[
"MIT"
] | 13
|
9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
https://github.com/KwonGihyun/DiagonalGAN/tree/9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
AdaptiveAttention
|
# 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/xf/cxf3byjr37ma2znuig476zvopiwegchp5jobm22ko4v5rlho3xac.py
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
# Source node to ATen node mapping:
# weight => mul
# Graph fragment:
# %mul : [num_users=2] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_1, 0.7071067811865476), kwargs = {})
triton_poi_fused_mul_0 = async_compile.triton('triton_poi_fused_mul_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[64],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': 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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = 0.7071067811865476
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/cy/ccyophyfjixfdl6nkpfgyxswhhotlkhyklpmnbx7p57hcmtgj7mp.py
# Topologically Sorted Source Nodes: [h_2, mul_1, mul_2, add], Original ATen: [aten.sigmoid, aten.mul, aten.add]
# Source node to ATen node mapping:
# add => add
# h_2 => sigmoid
# mul_1 => mul_1
# mul_2 => mul_2
# Graph fragment:
# %sigmoid : [num_users=1] = call_function[target=torch.ops.aten.sigmoid.default](args = (%view,), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%sigmoid, %primals_5), kwargs = {})
# %mul_2 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%primals_4, %mul_1), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%mul_2, %primals_5), kwargs = {})
triton_poi_fused_add_mul_sigmoid_1 = async_compile.triton('triton_poi_fused_add_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=[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_sigmoid_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_mul_sigmoid_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + (0))
tmp1 = tl.broadcast_to(tmp0, [XBLOCK])
tmp2 = tl.load(in_ptr1 + (x2), xmask)
tmp4 = tl.load(in_ptr2 + (x0), xmask, eviction_policy='evict_last')
tmp3 = tl.sigmoid(tmp2)
tmp5 = tmp3 * tmp4
tmp6 = tmp1 * tmp5
tmp7 = tmp6 + tmp4
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, (16, 4), (4, 1))
assert_size_stride(primals_2, (16, ), (1, ))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [weight], Original ATen: [aten.mul]
stream0 = get_raw_stream(0)
triton_poi_fused_mul_0.run(primals_1, buf0, 64, grid=grid(64), stream=stream0)
del primals_1
buf1 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [h], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_2, primals_3, reinterpret_tensor(buf0, (4, 16), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_2
buf2 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [h_2, mul_1, mul_2, add], Original ATen: [aten.sigmoid, aten.mul, aten.add]
triton_poi_fused_add_mul_sigmoid_1.run(primals_4, buf1, primals_5, buf2, 64, grid=grid(64), stream=stream0)
return (buf2, buf0, primals_3, primals_4, primals_5, 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((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((1, 1, 1, 1), (1, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_5 = 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])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
from math import sqrt
def equal_lr(module, name='weight'):
EqualLR.apply(module, name)
return module
class EqualLR:
def __init__(self, name):
self.name = name
def compute_weight(self, module):
weight = getattr(module, self.name + '_orig')
fan_in = weight.data.size(1) * weight.data[0][0].numel()
return weight * sqrt(2 / fan_in)
@staticmethod
def apply(module, name):
fn = EqualLR(name)
weight = getattr(module, name)
del module._parameters[name]
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
module.register_forward_pre_hook(fn)
return fn
def __call__(self, module, input):
weight = self.compute_weight(module)
setattr(module, self.name, weight)
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim):
super().__init__()
linear = nn.Linear(in_dim, out_dim)
linear.weight.data.normal_()
linear.bias.data.zero_()
self.linear = equal_lr(linear)
def forward(self, input):
return self.linear(input)
class AdaptiveAttention(nn.Module):
def __init__(self, img_dim, style_dim):
super().__init__()
self.img_dim = img_dim
self.fc = EqualLinear(style_dim, img_dim ** 2)
self.gamma = nn.Parameter(torch.ones(1, 1, 1, 1))
def forward(self, x, p):
h = self.fc(p)
h = h.view(h.size(0), 1, self.img_dim, self.img_dim)
h = F.sigmoid(h)
return self.gamma * (h * x) + x
def get_inputs():
return [torch.rand([4, 4]), torch.rand([4, 4])]
def get_init_inputs():
return [[], {'img_dim': 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
import torch.nn as nn
from math import sqrt
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 = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + x0, xmask)
tmp1 = 0.7071067811865476
tmp2 = tmp0 * tmp1
tl.store(out_ptr0 + x0, tmp2, xmask)
@triton.jit
def triton_poi_fused_add_mul_sigmoid_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0,
xnumel, XBLOCK: tl.constexpr):
xnumel = 64
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex
x0 = xindex % 16
tmp0 = tl.load(in_ptr0 + 0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK])
tmp2 = tl.load(in_ptr1 + x2, xmask)
tmp4 = tl.load(in_ptr2 + x0, xmask, eviction_policy='evict_last')
tmp3 = tl.sigmoid(tmp2)
tmp5 = tmp3 * tmp4
tmp6 = tmp1 * tmp5
tmp7 = tmp6 + tmp4
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, (16, 4), (4, 1))
assert_size_stride(primals_2, (16,), (1,))
assert_size_stride(primals_3, (4, 4), (4, 1))
assert_size_stride(primals_4, (1, 1, 1, 1), (1, 1, 1, 1))
assert_size_stride(primals_5, (4, 4), (4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_mul_0[grid(64)](primals_1, buf0, 64, XBLOCK=64,
num_warps=1, num_stages=1)
del primals_1
buf1 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
extern_kernels.addmm(primals_2, primals_3, reinterpret_tensor(buf0,
(4, 16), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_2
buf2 = empty_strided_cuda((4, 1, 4, 4), (16, 16, 4, 1), torch.float32)
triton_poi_fused_add_mul_sigmoid_1[grid(64)](primals_4, buf1,
primals_5, buf2, 64, XBLOCK=64, num_warps=1, num_stages=1)
return buf2, buf0, primals_3, primals_4, primals_5, buf1
def equal_lr(module, name='weight'):
EqualLR.apply(module, name)
return module
class EqualLR:
def __init__(self, name):
self.name = name
def compute_weight(self, module):
weight = getattr(module, self.name + '_orig')
fan_in = weight.data.size(1) * weight.data[0][0].numel()
return weight * sqrt(2 / fan_in)
@staticmethod
def apply(module, name):
fn = EqualLR(name)
weight = getattr(module, name)
del module._parameters[name]
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
module.register_forward_pre_hook(fn)
return fn
def __call__(self, module, input):
weight = self.compute_weight(module)
setattr(module, self.name, weight)
class EqualLinear(nn.Module):
def __init__(self, in_dim, out_dim):
super().__init__()
linear = nn.Linear(in_dim, out_dim)
linear.weight.data.normal_()
linear.bias.data.zero_()
self.linear = equal_lr(linear)
def forward(self, input):
return self.linear(input)
class AdaptiveAttentionNew(nn.Module):
def __init__(self, img_dim, style_dim):
super().__init__()
self.img_dim = img_dim
self.fc = EqualLinear(style_dim, img_dim ** 2)
self.gamma = nn.Parameter(torch.ones(1, 1, 1, 1))
def forward(self, input_0, input_1):
primals_4 = self.gamma
primals_2 = self.fc.linear.bias
primals_1 = self.fc.linear.weight_orig
primals_3 = input_0
primals_5 = input_1
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
KwonGihyun/DiagonalGAN
|
AdaptiveAttention
| false
| 8,438
|
[
"MIT"
] | 13
|
9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
https://github.com/KwonGihyun/DiagonalGAN/tree/9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
ConvTemporalGraphical
|
# 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/gv/cgvn7pohkb2ct3ysv7pb6blj4kyhihkyh4bagoxrigwcx3bg37gg.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.clone]
# Source node to ATen node mapping:
# x_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_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: '*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 = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = (xindex // 4) % 4
x4 = (xindex // 256)
x5 = (xindex // 16) % 16
x3 = (xindex // 64) % 4
x6 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (4*x5) + (64*x1) + (256*x4)), xmask)
tmp1 = tl.load(in_ptr1 + (x3 + (4*x1)), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x6), 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), (16, 4, 1))
assert_size_stride(primals_2, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (16, ), (1, ))
assert_size_stride(primals_4, (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_4, primals_2, stride=(1, 1), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = empty_strided_cuda((4, 4, 4, 4, 4, 1), (256, 64, 16, 4, 1, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf0, primals_3, buf1, 1024, grid=grid(1024), stream=stream0)
del buf0
del primals_3
buf2 = empty_strided_cuda((1, 64, 4), (256, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.bmm]
extern_kernels.bmm(reinterpret_tensor(buf1, (1, 64, 16), (0, 16, 1), 0), reinterpret_tensor(primals_1, (1, 16, 4), (64, 4, 1), 0), out=buf2)
del buf1
return (reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0), primals_2, primals_4, reinterpret_tensor(primals_1, (1, 4, 16), (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((16, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
class ConvTemporalGraphical(nn.Module):
"""The basic module for applying a graph convolution.
Args:
in_channels (int): Number of channels in the input sequence data
out_channels (int): Number of channels produced by the convolution
kernel_size (int): Size of the graph convolving kernel
t_kernel_size (int): Size of the temporal convolving kernel
t_stride (int, optional): Stride of the temporal convolution. Default: 1
t_padding (int, optional): Temporal zero-padding added to both sides of
the input. Default: 0
t_dilation (int, optional): Spacing between temporal kernel elements.
Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output.
Default: ``True``
Shape:
- Input[0]: Input graph sequence in :math:`(N, in_channels, T_{in}, V)` format
- Input[1]: Input graph adjacency matrix in :math:`(K, V, V)` format
- Output[0]: Outpu graph sequence in :math:`(N, out_channels, T_{out}, V)` format
- Output[1]: Graph adjacency matrix for output data in :math:`(K, V, V)` format
where
:math:`N` is a batch size,
:math:`K` is the spatial kernel size, as :math:`K == kernel_size[1]`,
:math:`T_{in}/T_{out}` is a length of input/output sequence,
:math:`V` is the number of graph nodes.
"""
def __init__(self, in_channels, out_channels, kernel_size,
t_kernel_size=1, t_stride=1, t_padding=0, t_dilation=1, bias=True):
super().__init__()
self.kernel_size = kernel_size
self.conv = nn.Conv2d(in_channels, out_channels * kernel_size,
kernel_size=(t_kernel_size, 1), padding=(t_padding, 0), stride=
(t_stride, 1), dilation=(t_dilation, 1), bias=bias)
def forward(self, x, A):
assert A.size(0) == self.kernel_size
x = self.conv(x)
n, kc, t, v = x.size()
x = x.view(n, self.kernel_size, kc // self.kernel_size, t, v)
x = torch.einsum('nkctv,kvw->nctw', (x, A))
return x.contiguous(), A
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([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
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 = 1024
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex % 4
x1 = xindex // 4 % 4
x4 = xindex // 256
x5 = xindex // 16 % 16
x3 = xindex // 64 % 4
x6 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 4 * x5 + 64 * x1 + 256 * x4), xmask)
tmp1 = tl.load(in_ptr1 + (x3 + 4 * x1), xmask, eviction_policy='evict_last'
)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + x6, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (16, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_3, (16,), (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 = extern_kernels.convolution(primals_4, primals_2, stride=(1,
1), padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf0, (4, 16, 4, 4), (256, 16, 4, 1))
buf1 = empty_strided_cuda((4, 4, 4, 4, 4, 1), (256, 64, 16, 4, 1, 1
), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(1024)](buf0, primals_3, buf1, 1024,
XBLOCK=128, num_warps=4, num_stages=1)
del buf0
del primals_3
buf2 = empty_strided_cuda((1, 64, 4), (256, 4, 1), torch.float32)
extern_kernels.bmm(reinterpret_tensor(buf1, (1, 64, 16), (0, 16, 1),
0), reinterpret_tensor(primals_1, (1, 16, 4), (64, 4, 1), 0),
out=buf2)
del buf1
return reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 16, 4, 1), 0
), primals_2, primals_4, reinterpret_tensor(primals_1, (1, 4, 16),
(64, 1, 4), 0)
class ConvTemporalGraphicalNew(nn.Module):
"""The basic module for applying a graph convolution.
Args:
in_channels (int): Number of channels in the input sequence data
out_channels (int): Number of channels produced by the convolution
kernel_size (int): Size of the graph convolving kernel
t_kernel_size (int): Size of the temporal convolving kernel
t_stride (int, optional): Stride of the temporal convolution. Default: 1
t_padding (int, optional): Temporal zero-padding added to both sides of
the input. Default: 0
t_dilation (int, optional): Spacing between temporal kernel elements.
Default: 1
bias (bool, optional): If ``True``, adds a learnable bias to the output.
Default: ``True``
Shape:
- Input[0]: Input graph sequence in :math:`(N, in_channels, T_{in}, V)` format
- Input[1]: Input graph adjacency matrix in :math:`(K, V, V)` format
- Output[0]: Outpu graph sequence in :math:`(N, out_channels, T_{out}, V)` format
- Output[1]: Graph adjacency matrix for output data in :math:`(K, V, V)` format
where
:math:`N` is a batch size,
:math:`K` is the spatial kernel size, as :math:`K == kernel_size[1]`,
:math:`T_{in}/T_{out}` is a length of input/output sequence,
:math:`V` is the number of graph nodes.
"""
def __init__(self, in_channels, out_channels, kernel_size,
t_kernel_size=1, t_stride=1, t_padding=0, t_dilation=1, bias=True):
super().__init__()
self.kernel_size = kernel_size
self.conv = nn.Conv2d(in_channels, out_channels * kernel_size,
kernel_size=(t_kernel_size, 1), padding=(t_padding, 0), stride=
(t_stride, 1), dilation=(t_dilation, 1), bias=bias)
def forward(self, input_0, input_1):
primals_2 = self.conv.weight
primals_3 = self.conv.bias
primals_4 = input_0
primals_1 = input_1
output = call([primals_1, primals_2, primals_3, primals_4])
return output[0], output[1]
|
Levigty/AimCLR
|
ConvTemporalGraphical
| false
| 8,439
|
[
"MIT"
] | 25
|
6cd73767f17748792508647355fa324fa63e235d
|
https://github.com/Levigty/AimCLR/tree/6cd73767f17748792508647355fa324fa63e235d
|
DropBlockT_1d
|
# 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/nc/cncq7jlbf7req5ht7vs4lajl6zdv5jinugrmeved2y7rieiv63cg.py
# Topologically Sorted Source Nodes: [setitem], Original ATen: [aten.lift_fresh, aten.index_put]
# Source node to ATen node mapping:
# setitem => full_default, index_put
# Graph fragment:
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 2.0), 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 = (%arg1_1, [%ge], %full_default), kwargs = {})
triton_poi_fused_index_put_lift_fresh_0 = async_compile.triton('triton_poi_fused_index_put_lift_fresh_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_0', 'mutated_arg_names': ['in_ptr0', '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_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 = 0.9
tmp2 = tmp0 >= tmp1
tmp3 = 2.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/kn/ckndrvngdbo3rrepy6nu5g7xshvxtbdxbgemeoicukp54reaos4y.py
# Topologically Sorted Source Nodes: [setitem_1], Original ATen: [aten.lift_fresh, aten.index_put]
# Source node to ATen node mapping:
# setitem_1 => full_default_1, index_put_1
# Graph fragment:
# %full_default_1 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 1.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %index_put_1 : [num_users=2] = call_function[target=torch.ops.aten.index_put_.default](args = (%index_put, [%lt], %full_default_1), kwargs = {})
triton_poi_fused_index_put_lift_fresh_1 = async_compile.triton('triton_poi_fused_index_put_lift_fresh_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_index_put_lift_fresh_1', 'mutated_arg_names': ['in_ptr0', '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_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 = 0.9
tmp2 = tmp0 < tmp1
tmp3 = 1.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tl.store(out_ptr0 + (x0), tmp4, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ho/chom2bhiwf5qu3yh2stu4bihs3qshknsqtsoqy6xmdfjqaf4a3yc.py
# Topologically Sorted Source Nodes: [setitem_2, sum_1], Original ATen: [aten.lift_fresh, aten.index_put, aten.sum]
# Source node to ATen node mapping:
# setitem_2 => full_default_2, index_put_2
# sum_1 => sum_1
# Graph fragment:
# %full_default_2 : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 0.0), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %index_put_2 : [num_users=3] = call_function[target=torch.ops.aten.index_put_.default](args = (%index_put_1, [%eq], %full_default_2), kwargs = {})
# %sum_1 : [num_users=1] = call_function[target=torch.ops.aten.sum.default](args = (%index_put_2,), kwargs = {})
triton_per_fused_index_put_lift_fresh_sum_2 = async_compile.triton('triton_per_fused_index_put_lift_fresh_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, 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_index_put_lift_fresh_sum_2', 'mutated_arg_names': ['in_ptr0', 'out_ptr0'], 'no_x_dim': True, '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_index_put_lift_fresh_sum_2(in_ptr0, out_ptr0, out_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 = 2.0
tmp2 = tmp0 == tmp1
tmp3 = 0.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tmp5 = tl.broadcast_to(tmp4, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tl.store(out_ptr0 + (tl.broadcast_to(r0, [RBLOCK])), tmp4, None)
tl.store(out_ptr1 + (tl.full([1], 0, tl.int32)), tmp7, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/az/caznebbkxbo5xyds3w5qgcukajrizzf3vwnjnwn6lxjvefgg4d3u.py
# Topologically Sorted Source Nodes: [mul, mul_1, truediv], Original ATen: [aten.mul, aten.div]
# Source node to ATen node mapping:
# mul => mul
# mul_1 => mul_1
# truediv => div
# Graph fragment:
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view, %index_put_2), kwargs = {})
# %mul_1 : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%mul, 256), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%mul_1, %sum_1), kwargs = {})
triton_poi_fused_div_mul_3 = async_compile.triton('triton_poi_fused_div_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, 4], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*fp32', 3: '*fp32', 4: 'i32', 5: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_div_mul_3(in_ptr0, in_ptr1, in_ptr2, 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 + ((4*x2) + (16*(y0 // 4)) + (64*y1) + (y0 % 4)), xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + (4*y3)), xmask & ymask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr2 + (0))
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, YBLOCK])
tmp2 = tmp0 * tmp1
tmp3 = 256.0
tmp4 = tmp2 * tmp3
tmp7 = tmp4 / tmp6
tl.store(out_ptr0 + (x2 + (4*y3)), tmp7, xmask & ymask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, arg1_1 = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(arg1_1, (4, 16, 4), (64, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [setitem], Original ATen: [aten.lift_fresh, aten.index_put]
stream0 = get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_0.run(arg1_1, arg1_1, 256, grid=grid(256), stream=stream0)
# Topologically Sorted Source Nodes: [setitem_1], Original ATen: [aten.lift_fresh, aten.index_put]
triton_poi_fused_index_put_lift_fresh_1.run(arg1_1, arg1_1, 256, grid=grid(256), stream=stream0)
buf6 = empty_strided_cuda((), (), torch.float32)
# Topologically Sorted Source Nodes: [setitem_2, sum_1], Original ATen: [aten.lift_fresh, aten.index_put, aten.sum]
triton_per_fused_index_put_lift_fresh_sum_2.run(arg1_1, arg1_1, buf6, 1, 256, grid=grid(1), stream=stream0)
buf7 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [mul, mul_1, truediv], Original ATen: [aten.mul, aten.div]
triton_poi_fused_div_mul_3.run(arg0_1, arg1_1, buf6, buf7, 64, 4, grid=grid(64, 4), stream=stream0)
del arg0_1
del arg1_1
del buf6
return (reinterpret_tensor(buf7, (4, 4, 4, 4), (64, 16, 1, 4), 0), )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
arg0_1 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
arg1_1 = rand_strided((4, 16, 4), (64, 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 DropBlockT_1d(nn.Module):
def __init__(self, keep_prob=0.9):
super(DropBlockT_1d, self).__init__()
self.keep_prob = keep_prob
def forward(self, input, mask):
n, c, t, v = input.size()
input1 = input.permute(0, 1, 3, 2).contiguous().view(n, c * v, t)
mask[mask >= self.keep_prob] = 2.0
mask[mask < self.keep_prob] = 1.0
mask[mask == 2.0] = 0.0
return (input1 * mask * mask.numel() / mask.sum()).view(n, c, v, t
).permute(0, 1, 3, 2)
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 16, 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
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_index_put_lift_fresh_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 = 0.9
tmp2 = tmp0 >= tmp1
tmp3 = 2.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_poi_fused_index_put_lift_fresh_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 = 0.9
tmp2 = tmp0 < tmp1
tmp3 = 1.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tl.store(out_ptr0 + x0, tmp4, xmask)
@triton.jit
def triton_per_fused_index_put_lift_fresh_sum_2(in_ptr0, out_ptr0, out_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 = 2.0
tmp2 = tmp0 == tmp1
tmp3 = 0.0
tmp4 = tl.where(tmp2, tmp3, tmp0)
tmp5 = tl.broadcast_to(tmp4, [RBLOCK])
tmp7 = triton_helpers.promote_to_tensor(tl.sum(tmp5, 0))
tl.store(out_ptr0 + tl.broadcast_to(r0, [RBLOCK]), tmp4, None)
tl.store(out_ptr1 + tl.full([1], 0, tl.int32), tmp7, None)
@triton.jit
def triton_poi_fused_div_mul_3(in_ptr0, in_ptr1, in_ptr2, 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 + (4 * x2 + 16 * (y0 // 4) + 64 * y1 + y0 % 4),
xmask & ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (x2 + 4 * y3), xmask & ymask, eviction_policy=
'evict_last')
tmp5 = tl.load(in_ptr2 + 0)
tmp6 = tl.broadcast_to(tmp5, [XBLOCK, YBLOCK])
tmp2 = tmp0 * tmp1
tmp3 = 256.0
tmp4 = tmp2 * tmp3
tmp7 = tmp4 / tmp6
tl.store(out_ptr0 + (x2 + 4 * y3), tmp7, xmask & ymask)
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, 16, 4), (64, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
get_raw_stream(0)
triton_poi_fused_index_put_lift_fresh_0[grid(256)](arg1_1, arg1_1,
256, XBLOCK=128, num_warps=4, num_stages=1)
triton_poi_fused_index_put_lift_fresh_1[grid(256)](arg1_1, arg1_1,
256, XBLOCK=256, num_warps=4, num_stages=1)
buf6 = empty_strided_cuda((), (), torch.float32)
triton_per_fused_index_put_lift_fresh_sum_2[grid(1)](arg1_1, arg1_1,
buf6, 1, 256, num_warps=2, num_stages=1)
buf7 = empty_strided_cuda((4, 16, 4), (64, 4, 1), torch.float32)
triton_poi_fused_div_mul_3[grid(64, 4)](arg0_1, arg1_1, buf6, buf7,
64, 4, XBLOCK=4, YBLOCK=64, num_warps=4, num_stages=1)
del arg0_1
del arg1_1
del buf6
return reinterpret_tensor(buf7, (4, 4, 4, 4), (64, 16, 1, 4), 0),
class DropBlockT_1dNew(nn.Module):
def __init__(self, keep_prob=0.9):
super(DropBlockT_1dNew, self).__init__()
self.keep_prob = keep_prob
def forward(self, input_0, input_1):
arg0_1 = input_0
arg1_1 = input_1
output = call([arg0_1, arg1_1])
return output[0]
|
Levigty/AimCLR
|
DropBlockT_1d
| false
| 8,440
|
[
"MIT"
] | 25
|
6cd73767f17748792508647355fa324fa63e235d
|
https://github.com/Levigty/AimCLR/tree/6cd73767f17748792508647355fa324fa63e235d
|
GaussNorm2D
|
# 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/7p/c7pozh4btr2mbqngpjlc7jjcl3rvmhvmosru4rguy6eryt5zjot2.py
# Topologically Sorted Source Nodes: [m, sub, s, sqrt, mul, truediv, erf, add, x_1], Original ATen: [aten.mean, aten.sub, aten.std, aten.sqrt, aten.mul, aten.div, aten.erf, aten.add]
# Source node to ATen node mapping:
# add => add
# erf => erf
# m => mean
# mul => mul
# s => var
# sqrt => full_default
# sub => sub
# truediv => div
# x_1 => mul_1
# Graph fragment:
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.dim](args = (%view, [1]), kwargs = {})
# %sub : [num_users=1] = call_function[target=torch.ops.aten.sub.Tensor](args = (%view, %view_1), kwargs = {})
# %var : [num_users=1] = call_function[target=torch.ops.aten.var.correction](args = (%view, [1]), kwargs = {correction: 1.0})
# %full_default : [num_users=1] = call_function[target=torch.ops.aten.full.default](args = ([], 1.4142135381698608), kwargs = {dtype: torch.float32, layout: torch.strided, device: cpu, pin_memory: False})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%view_2, %full_default), kwargs = {})
# %div : [num_users=1] = call_function[target=torch.ops.aten.div.Tensor](args = (%sub, %mul), 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 = (%add, 0.5), kwargs = {})
triton_per_fused_add_div_erf_mean_mul_sqrt_std_sub_0 = async_compile.triton('triton_per_fused_add_div_erf_mean_mul_sqrt_std_sub_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.persistent_reduction(
size_hints=[4, 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, 3), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_div_erf_mean_mul_sqrt_std_sub_0', 'mutated_arg_names': [], '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_add_div_erf_mean_mul_sqrt_std_sub_0(in_ptr0, out_ptr2, xnumel, rnumel, XBLOCK : tl.constexpr):
xnumel = 4
rnumel = 16
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
roffset = 0
rmask = tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + (16*x0)), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp6 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp8 = tl.where(xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tmp10 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp11 = tmp10.to(tl.float32)
tmp12 = tmp9 / tmp11
tmp13 = tmp1 - tmp12
tmp14 = tmp13 * tmp13
tmp15 = tl.broadcast_to(tmp14, [XBLOCK, RBLOCK])
tmp17 = tl.where(xmask, tmp15, 0)
tmp18 = tl.sum(tmp17, 1)[:, None]
tmp19 = 16.0
tmp20 = tmp4 / tmp19
tmp21 = tmp0 - tmp20
tmp22 = 15.0
tmp23 = tmp18 / tmp22
tmp24 = libdevice.sqrt(tmp23)
tmp25 = 1.4142135381698608
tmp26 = tmp24 * tmp25
tmp27 = tmp21 / tmp26
tmp28 = libdevice.erf(tmp27)
tmp29 = 1.0
tmp30 = tmp28 + tmp29
tmp31 = 0.5
tmp32 = tmp30 * tmp31
tl.store(out_ptr2 + (r1 + (16*x0)), tmp32, 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)
buf4 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [m, sub, s, sqrt, mul, truediv, erf, add, x_1], Original ATen: [aten.mean, aten.sub, aten.std, aten.sqrt, aten.mul, aten.div, aten.erf, aten.add]
stream0 = get_raw_stream(0)
triton_per_fused_add_div_erf_mean_mul_sqrt_std_sub_0.run(arg0_1, buf4, 4, 16, grid=grid(4), stream=stream0)
del arg0_1
return (reinterpret_tensor(buf4, (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
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.autograd
class GaussNorm2D(nn.Module):
"""
This will normalize a saliency map to range from 0 to 1 via normal cumulative distribution function.
Input and output will be a 3D tensor of size [batch size x height x width].
Input can be any real valued number (supported by hardware)
Output will range from 0 to 1
Notes:
(1) GammaNorm2D will produce slightly better results
The sum ROAR/KAR will improve from 1.44 to 1.45 for FastCAM using GradCAM.
(2) This method is a bit less expensive than GammaNorm2D.
"""
def __init__(self, const_mean=None, const_std=None):
super(GaussNorm2D, self).__init__()
assert isinstance(const_mean, float) or const_mean is None
assert isinstance(const_std, float) or const_std is None
self.const_mean = const_mean
self.const_std = const_std
def forward(self, x):
"""
Input:
x: A Torch Tensor image with shape [batch size x height x width] e.g. [64,7,7]
Return:
x: x Normalized by computing mean and std over each individual batch item and squashed with a
Normal/Gaussian CDF.
"""
assert torch.is_tensor(x), 'Input must be a Torch Tensor'
assert len(x.size()
) == 3, 'Input should be sizes [batch size x height x width]'
s0 = x.size()[0]
s1 = x.size()[1]
s2 = x.size()[2]
x = x.reshape(s0, s1 * s2)
"""
Compute Mean
"""
if self.const_mean is None:
m = x.mean(dim=1)
m = m.reshape(m.size()[0], 1)
else:
m = self.const_mean
"""
Compute Standard Deviation
"""
if self.const_std is None:
s = x.std(dim=1)
s = s.reshape(s.size()[0], 1)
else:
s = self.const_std
"""
The normal cumulative distribution function is used to squash the values from within the range of 0 to 1
"""
x = 0.5 * (1.0 + torch.erf((x - m) / (s * torch.sqrt(torch.tensor(
2.0)))))
x = x.reshape(s0, s1, s2)
return x
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 libdevice
import torch.nn as nn
import torch.autograd
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_per_fused_add_div_erf_mean_mul_sqrt_std_sub_0(in_ptr0, out_ptr2,
xnumel, rnumel, XBLOCK: tl.constexpr):
xnumel = 4
RBLOCK: tl.constexpr = 16
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
rindex = tl.arange(0, RBLOCK)[None, :]
tl.full([XBLOCK, RBLOCK], True, tl.int1)
r1 = rindex
x0 = xindex
tmp0 = tl.load(in_ptr0 + (r1 + 16 * x0), xmask, other=0.0)
tmp1 = tl.broadcast_to(tmp0, [XBLOCK, RBLOCK])
tmp3 = tl.where(xmask, tmp1, 0)
tmp4 = tl.sum(tmp3, 1)[:, None]
tmp6 = tl.broadcast_to(tmp1, [XBLOCK, RBLOCK])
tmp8 = tl.where(xmask, tmp6, 0)
tmp9 = tl.sum(tmp8, 1)[:, None]
tmp10 = tl.full([XBLOCK, 1], 16, tl.int32)
tmp11 = tmp10.to(tl.float32)
tmp12 = tmp9 / tmp11
tmp13 = tmp1 - tmp12
tmp14 = tmp13 * tmp13
tmp15 = tl.broadcast_to(tmp14, [XBLOCK, RBLOCK])
tmp17 = tl.where(xmask, tmp15, 0)
tmp18 = tl.sum(tmp17, 1)[:, None]
tmp19 = 16.0
tmp20 = tmp4 / tmp19
tmp21 = tmp0 - tmp20
tmp22 = 15.0
tmp23 = tmp18 / tmp22
tmp24 = libdevice.sqrt(tmp23)
tmp25 = 1.4142135381698608
tmp26 = tmp24 * tmp25
tmp27 = tmp21 / tmp26
tmp28 = libdevice.erf(tmp27)
tmp29 = 1.0
tmp30 = tmp28 + tmp29
tmp31 = 0.5
tmp32 = tmp30 * tmp31
tl.store(out_ptr2 + (r1 + 16 * x0), tmp32, 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)
buf4 = empty_strided_cuda((4, 16), (16, 1), torch.float32)
get_raw_stream(0)
triton_per_fused_add_div_erf_mean_mul_sqrt_std_sub_0[grid(4)](arg0_1,
buf4, 4, 16, XBLOCK=1, num_warps=2, num_stages=1)
del arg0_1
return reinterpret_tensor(buf4, (4, 4, 4), (16, 4, 1), 0),
class GaussNorm2DNew(nn.Module):
"""
This will normalize a saliency map to range from 0 to 1 via normal cumulative distribution function.
Input and output will be a 3D tensor of size [batch size x height x width].
Input can be any real valued number (supported by hardware)
Output will range from 0 to 1
Notes:
(1) GammaNorm2D will produce slightly better results
The sum ROAR/KAR will improve from 1.44 to 1.45 for FastCAM using GradCAM.
(2) This method is a bit less expensive than GammaNorm2D.
"""
def __init__(self, const_mean=None, const_std=None):
super(GaussNorm2DNew, self).__init__()
assert isinstance(const_mean, float) or const_mean is None
assert isinstance(const_std, float) or const_std is None
self.const_mean = const_mean
self.const_std = const_std
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
LLNL/fastcam
|
GaussNorm2D
| false
| 8,441
|
[
"BSD-3-Clause"
] | 25
|
99cefe37528014247319468cf05f54fef259d3bf
|
https://github.com/LLNL/fastcam/tree/99cefe37528014247319468cf05f54fef259d3bf
|
FirstNet
|
# 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/te/ctef7mzwyjkfyg3f43zkxcxrqjh56sog4eojwdtrxx5ecrksbhsb.py
# Unsorted Source Nodes: [], Original ATen: []
# Source node to ATen node mapping:
triton_poi_fused_0 = async_compile.triton('triton_poi_fused_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[8192, 16], tile_hint=TileHint.SQUARE,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32', 3: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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_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_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 8192
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 64
y1 = (yindex // 64)
tmp0 = tl.load(in_ptr0 + (x2 + (9*y3)), xmask, eviction_policy='evict_last')
tl.store(out_ptr0 + (y0 + (64*x2) + (576*y1)), tmp0, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/bs/cbsrrow5un3n6xpmn6m4epddnq2wdszchl6ffyn6dnlk232oasjh.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], [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_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=[256, 4096], 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, 4), equal_to_1=())]},
inductor_meta={'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_relu_1', 'mutated_arg_names': [], 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': 'B098E03CDA7B8ADC90DAFFDF24A2956451D1B13F297756A5DCC209498AA53705', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': False},
min_elem_per_thread=0
)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_ptr0, in_ptr1, out_ptr0, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 256
xnumel = 4096
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y3 = yindex
y0 = yindex % 64
y1 = (yindex // 64)
tmp0 = tl.load(in_ptr0 + (x2 + (4096*y3)), ymask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr1 + (y0), ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1, 1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(out_ptr0 + (y0 + (64*x2) + (262144*y1)), tmp4, ymask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ma/cmainzeqjobzoawmdmzvkfd3cg6hup2rnekj33lv2wb62vtrjpqv.py
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x_2 => getitem, getitem_1
# Graph fragment:
# %getitem : [num_users=2] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 0), kwargs = {})
# %getitem_1 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_2 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_2', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[262144],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i8', 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_max_pool2d_with_indices_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_max_pool2d_with_indices_2(in_ptr0, out_ptr0, out_ptr1, xnumel, XBLOCK : tl.constexpr):
xnumel = 262144
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 64
x1 = (xindex // 64) % 32
x2 = (xindex // 2048)
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + (128*x1) + (8192*x2)), None)
tmp1 = tl.load(in_ptr0 + (64 + x0 + (128*x1) + (8192*x2)), None)
tmp3 = tl.load(in_ptr0 + (4096 + x0 + (128*x1) + (8192*x2)), None)
tmp5 = tl.load(in_ptr0 + (4160 + x0 + (128*x1) + (8192*x2)), None)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (x3), tmp6, None)
tl.store(out_ptr1 + (x3), tmp16, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/gk/cgkac2t4jvdqung4oouqcv6mynjak323ilfwj3p4qgvaortqf3lz.py
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.convolution, aten.relu]
# Source node to ATen node mapping:
# x_3 => convolution_1
# x_4 => relu_1
# Graph fragment:
# %convolution_1 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%getitem, %primals_4, %primals_5, [1, 1], [1, 1], [1, 1], False, [0, 0], 1), kwargs = {})
# %relu_1 : [num_users=2] = call_function[target=torch.ops.aten.relu.default](args = (%convolution_1,), kwargs = {})
triton_poi_fused_convolution_relu_3 = async_compile.triton('triton_poi_fused_convolution_relu_3', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[524288],
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_3', '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_3(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 524288
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + (x2), None)
tmp1 = tl.load(in_ptr0 + (x0), None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + (x2), tmp4, None)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/vg/cvg565pmgpsaclzuyyhufywleedihcvdk5cyodbpujd7uuaxcuxs.py
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.max_pool2d_with_indices]
# Source node to ATen node mapping:
# x_5 => getitem_2, getitem_3
# Graph fragment:
# %getitem_2 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets_1, 0), kwargs = {})
# %getitem_3 : [num_users=1] = call_function[target=operator.getitem](args = (%_low_memory_max_pool2d_with_offsets_1, 1), kwargs = {})
triton_poi_fused_max_pool2d_with_indices_4 = async_compile.triton('triton_poi_fused_max_pool2d_with_indices_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=[1024, 128], tile_hint=TileHint.DEFAULT,
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: '*i8', 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_poi_fused_max_pool2d_with_indices_4', '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_pool2d_with_indices_4(in_ptr0, out_ptr0, out_ptr1, ynumel, xnumel, YBLOCK : tl.constexpr, XBLOCK : tl.constexpr):
ynumel = 1024
xnumel = 128
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x3 = xindex
y0 = yindex % 16
y4 = (yindex // 16)
y2 = (yindex // 256)
y5 = yindex % 256
y6 = yindex
tmp0 = tl.load(in_ptr0 + (x3 + (256*y0) + (8192*y4)), xmask, eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (128 + x3 + (256*y0) + (8192*y4)), xmask, eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (4096 + x3 + (256*y0) + (8192*y4)), xmask, eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (4224 + x3 + (256*y0) + (8192*y4)), xmask, eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1, 1], 1, tl.int8)
tmp9 = tl.full([1, 1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1, 1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1, 1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (y5 + (256*x3) + (32768*y2)), tmp6, xmask)
tl.store(out_ptr1 + (x3 + (128*y6)), tmp16, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (64, 1, 3, 3), (9, 9, 3, 1))
assert_size_stride(primals_2, (64, ), (1, ))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
assert_size_stride(primals_4, (128, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_5, (128, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((128, 64, 3, 3), (576, 1, 192, 64), torch.float32)
# Unsorted Source Nodes: [], Original ATen: []
stream0 = get_raw_stream(0)
triton_poi_fused_0.run(primals_4, buf0, 8192, 9, grid=grid(8192, 9), stream=stream0)
del primals_4
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.convolution]
buf1 = 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(buf1, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf2 = empty_strided_cuda((4, 64, 64, 64), (262144, 1, 4096, 64), torch.float32)
# Topologically Sorted Source Nodes: [x, x_1], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_1.run(buf1, primals_2, buf2, 256, 4096, grid=grid(256, 4096), stream=stream0)
del buf1
del primals_2
buf3 = empty_strided_cuda((4, 64, 32, 32), (65536, 1, 2048, 64), torch.float32)
buf4 = empty_strided_cuda((4, 64, 32, 32), (65536, 1, 2048, 64), torch.int8)
# Topologically Sorted Source Nodes: [x_2], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_2.run(buf2, buf3, buf4, 262144, grid=grid(262144), stream=stream0)
# Topologically Sorted Source Nodes: [x_3], Original ATen: [aten.convolution]
buf5 = extern_kernels.convolution(buf3, buf0, stride=(1, 1), padding=(1, 1), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 128, 32, 32), (131072, 1, 4096, 128))
buf6 = buf5; del buf5 # reuse
# Topologically Sorted Source Nodes: [x_3, x_4], Original ATen: [aten.convolution, aten.relu]
triton_poi_fused_convolution_relu_3.run(buf6, primals_5, 524288, grid=grid(524288), stream=stream0)
del primals_5
buf7 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1), torch.float32)
buf8 = empty_strided_cuda((4, 128, 16, 16), (32768, 1, 2048, 128), torch.int8)
# Topologically Sorted Source Nodes: [x_5], Original ATen: [aten.max_pool2d_with_indices]
triton_poi_fused_max_pool2d_with_indices_4.run(buf6, buf7, buf8, 1024, 128, grid=grid(1024, 128), stream=stream0)
return (buf7, primals_1, primals_3, buf0, buf2, buf3, buf4, buf6, 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((64, 1, 3, 3), (9, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((64, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 1, 64, 64), (4096, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((128, 64, 3, 3), (576, 9, 3, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((128, ), (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 FirstNet(nn.Module):
def __init__(self):
super(FirstNet, self).__init__()
self.conv1 = nn.Conv2d(in_channels=1, out_channels=64, kernel_size=
3, padding=1, stride=1)
self.conv2 = nn.Conv2d(64, 128, 3, padding=1)
def forward(self, x):
x = self.conv1(x)
x = F.relu(x)
x = F.max_pool2d(x, 2, 2)
x = self.conv2(x)
x = F.relu(x)
x = F.max_pool2d(x, 2, 2)
return x
def get_inputs():
return [torch.rand([4, 1, 64, 64])]
def get_init_inputs():
return [[], {}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime import triton_helpers
import torch.nn as nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_0(in_ptr0, out_ptr0, ynumel, xnumel, YBLOCK: tl.
constexpr, XBLOCK: tl.constexpr):
xnumel = 9
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x2 = xindex
y3 = yindex
y0 = yindex % 64
y1 = yindex // 64
tmp0 = tl.load(in_ptr0 + (x2 + 9 * y3), xmask, eviction_policy='evict_last'
)
tl.store(out_ptr0 + (y0 + 64 * x2 + 576 * y1), tmp0, xmask)
@triton.jit
def triton_poi_fused_convolution_relu_1(in_ptr0, in_ptr1, out_ptr0, ynumel,
xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
ynumel = 256
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
ymask = yindex < ynumel
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
x2 = xindex
y3 = yindex
y0 = yindex % 64
y1 = yindex // 64
tmp0 = tl.load(in_ptr0 + (x2 + 4096 * y3), ymask, eviction_policy=
'evict_last')
tmp1 = tl.load(in_ptr1 + y0, ymask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1, 1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(out_ptr0 + (y0 + 64 * x2 + 262144 * y1), tmp4, ymask)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_2(in_ptr0, out_ptr0, out_ptr1,
xnumel, XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x0 = xindex % 64
x1 = xindex // 64 % 32
x2 = xindex // 2048
x3 = xindex
tmp0 = tl.load(in_ptr0 + (x0 + 128 * x1 + 8192 * x2), None)
tmp1 = tl.load(in_ptr0 + (64 + x0 + 128 * x1 + 8192 * x2), None)
tmp3 = tl.load(in_ptr0 + (4096 + x0 + 128 * x1 + 8192 * x2), None)
tmp5 = tl.load(in_ptr0 + (4160 + x0 + 128 * x1 + 8192 * x2), None)
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1], 1, tl.int8)
tmp9 = tl.full([1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + x3, tmp6, None)
tl.store(out_ptr1 + x3, tmp16, None)
@triton.jit
def triton_poi_fused_convolution_relu_3(in_out_ptr0, in_ptr0, xnumel,
XBLOCK: tl.constexpr):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
tl.full([XBLOCK], True, tl.int1)
x2 = xindex
x0 = xindex % 128
tmp0 = tl.load(in_out_ptr0 + x2, None)
tmp1 = tl.load(in_ptr0 + x0, None, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tmp3 = tl.full([1], 0, tl.int32)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tl.store(in_out_ptr0 + x2, tmp4, None)
@triton.jit
def triton_poi_fused_max_pool2d_with_indices_4(in_ptr0, out_ptr0, out_ptr1,
ynumel, xnumel, YBLOCK: tl.constexpr, XBLOCK: tl.constexpr):
xnumel = 128
yoffset = tl.program_id(1) * YBLOCK
yindex = yoffset + tl.arange(0, YBLOCK)[None, :]
tl.full([XBLOCK, YBLOCK], True, tl.int1)
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
xmask = xindex < xnumel
x3 = xindex
y0 = yindex % 16
y4 = yindex // 16
y2 = yindex // 256
y5 = yindex % 256
y6 = yindex
tmp0 = tl.load(in_ptr0 + (x3 + 256 * y0 + 8192 * y4), xmask,
eviction_policy='evict_last')
tmp1 = tl.load(in_ptr0 + (128 + x3 + 256 * y0 + 8192 * y4), xmask,
eviction_policy='evict_last')
tmp3 = tl.load(in_ptr0 + (4096 + x3 + 256 * y0 + 8192 * y4), xmask,
eviction_policy='evict_last')
tmp5 = tl.load(in_ptr0 + (4224 + x3 + 256 * y0 + 8192 * y4), xmask,
eviction_policy='evict_last')
tmp2 = triton_helpers.maximum(tmp1, tmp0)
tmp4 = triton_helpers.maximum(tmp3, tmp2)
tmp6 = triton_helpers.maximum(tmp5, tmp4)
tmp7 = tmp1 > tmp0
tmp8 = tl.full([1, 1], 1, tl.int8)
tmp9 = tl.full([1, 1], 0, tl.int8)
tmp10 = tl.where(tmp7, tmp8, tmp9)
tmp11 = tmp3 > tmp2
tmp12 = tl.full([1, 1], 2, tl.int8)
tmp13 = tl.where(tmp11, tmp12, tmp10)
tmp14 = tmp5 > tmp4
tmp15 = tl.full([1, 1], 3, tl.int8)
tmp16 = tl.where(tmp14, tmp15, tmp13)
tl.store(out_ptr0 + (y5 + 256 * x3 + 32768 * y2), tmp6, xmask)
tl.store(out_ptr1 + (x3 + 128 * y6), tmp16, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (64, 1, 3, 3), (9, 9, 3, 1))
assert_size_stride(primals_2, (64,), (1,))
assert_size_stride(primals_3, (4, 1, 64, 64), (4096, 4096, 64, 1))
assert_size_stride(primals_4, (128, 64, 3, 3), (576, 9, 3, 1))
assert_size_stride(primals_5, (128,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((128, 64, 3, 3), (576, 1, 192, 64), torch
.float32)
get_raw_stream(0)
triton_poi_fused_0[grid(8192, 9)](primals_4, buf0, 8192, 9, XBLOCK=
16, YBLOCK=64, num_warps=4, num_stages=1)
del primals_4
buf1 = 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(buf1, (4, 64, 64, 64), (262144, 4096, 64, 1))
buf2 = empty_strided_cuda((4, 64, 64, 64), (262144, 1, 4096, 64),
torch.float32)
triton_poi_fused_convolution_relu_1[grid(256, 4096)](buf1,
primals_2, buf2, 256, 4096, XBLOCK=32, YBLOCK=32, num_warps=4,
num_stages=1)
del buf1
del primals_2
buf3 = empty_strided_cuda((4, 64, 32, 32), (65536, 1, 2048, 64),
torch.float32)
buf4 = empty_strided_cuda((4, 64, 32, 32), (65536, 1, 2048, 64),
torch.int8)
triton_poi_fused_max_pool2d_with_indices_2[grid(262144)](buf2, buf3,
buf4, 262144, XBLOCK=512, num_warps=8, num_stages=1)
buf5 = extern_kernels.convolution(buf3, buf0, stride=(1, 1),
padding=(1, 1), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf5, (4, 128, 32, 32), (131072, 1, 4096, 128))
buf6 = buf5
del buf5
triton_poi_fused_convolution_relu_3[grid(524288)](buf6, primals_5,
524288, XBLOCK=1024, num_warps=4, num_stages=1)
del primals_5
buf7 = empty_strided_cuda((4, 128, 16, 16), (32768, 256, 16, 1),
torch.float32)
buf8 = empty_strided_cuda((4, 128, 16, 16), (32768, 1, 2048, 128),
torch.int8)
triton_poi_fused_max_pool2d_with_indices_4[grid(1024, 128)](buf6,
buf7, buf8, 1024, 128, XBLOCK=32, YBLOCK=32, num_warps=4,
num_stages=1)
return buf7, primals_1, primals_3, buf0, buf2, buf3, buf4, buf6, buf8
class FirstNetNew(nn.Module):
def __init__(self):
super(FirstNetNew, self).__init__()
self.conv1 = nn.Conv2d(in_channels=1, out_channels=64, kernel_size=
3, padding=1, stride=1)
self.conv2 = nn.Conv2d(64, 128, 3, padding=1)
def forward(self, input_0):
primals_1 = self.conv1.weight
primals_2 = self.conv1.bias
primals_4 = self.conv2.weight
primals_5 = self.conv2.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
Koukyosyumei/AIJack
|
FirstNet
| false
| 8,442
|
[
"MIT"
] | 24
|
9545d3828907b54965ede85e0e12cb32eef54294
|
https://github.com/Koukyosyumei/AIJack/tree/9545d3828907b54965ede85e0e12cb32eef54294
|
FusedDownsample
|
# 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/qf/cqfpjcbno7orqx35hioypcgzp7qglgr7wqdebmvpel54tvilauub.py
# Topologically Sorted Source Nodes: [add, add_1, add_2, weight_1], Original ATen: [aten.add, aten.div]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# weight_1 => div
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%slice_4, %slice_8), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %slice_12), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %slice_16), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_2, 4), kwargs = {})
triton_poi_fused_add_div_0 = async_compile.triton('triton_poi_fused_add_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=[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_add_div_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_add_div_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 5) % 5
x0 = xindex % 5
x2 = (xindex // 25)
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 = x0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (x0 + (4*x1) + (16*x2)), tmp10 & xmask, other=0.0)
tmp12 = 0.1767766952966369
tmp13 = tmp11 * tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp10, tmp13, tmp14)
tmp16 = (-1) + x1
tmp17 = tmp16 >= tmp1
tmp18 = tmp16 < tmp3
tmp19 = tmp17 & tmp18
tmp20 = tmp19 & tmp6
tmp21 = tmp20 & tmp7
tmp22 = tl.load(in_ptr0 + ((-4) + x0 + (4*x1) + (16*x2)), tmp21 & xmask, other=0.0)
tmp23 = tmp22 * tmp12
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp21, tmp23, tmp24)
tmp26 = tmp15 + tmp25
tmp27 = (-1) + x0
tmp28 = tmp27 >= tmp1
tmp29 = tmp27 < tmp3
tmp30 = tmp8 & tmp28
tmp31 = tmp30 & tmp29
tmp32 = tl.load(in_ptr0 + ((-1) + x0 + (4*x1) + (16*x2)), tmp31 & xmask, other=0.0)
tmp33 = tmp32 * tmp12
tmp34 = tl.full(tmp33.shape, 0.0, tmp33.dtype)
tmp35 = tl.where(tmp31, tmp33, tmp34)
tmp36 = tmp26 + tmp35
tmp37 = tmp19 & tmp28
tmp38 = tmp37 & tmp29
tmp39 = tl.load(in_ptr0 + ((-5) + x0 + (4*x1) + (16*x2)), tmp38 & xmask, other=0.0)
tmp40 = tmp39 * tmp12
tmp41 = tl.full(tmp40.shape, 0.0, tmp40.dtype)
tmp42 = tl.where(tmp38, tmp40, tmp41)
tmp43 = tmp36 + tmp42
tmp44 = 0.25
tmp45 = tmp43 * tmp44
tl.store(in_out_ptr0 + (x4), tmp45, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/l2/cl2i74pzzl4wk7ch45c33ead7jflgqqtjdxk4goxparpm7agz76c.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, %div, %primals_2, [2, 2], [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=[16384],
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 = 14400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 900) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 64, 64), (16384, 4096, 64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 5, 5), (100, 25, 5, 1), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [add, add_1, add_2, weight_1], Original ATen: [aten.add, aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_0.run(buf1, primals_1, 400, grid=grid(400), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(primals_3, buf1, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=False, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 30, 30), (3600, 900, 30, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf3, primals_2, 14400, grid=grid(14400), stream=stream0)
del primals_2
return (buf3, primals_3, 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, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 64, 64), (16384, 4096, 64, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
from math import sqrt
class FusedDownsample(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, padding=0):
super().__init__()
weight = torch.randn(out_channel, in_channel, kernel_size, kernel_size)
bias = torch.zeros(out_channel)
fan_in = in_channel * kernel_size * kernel_size
self.multiplier = sqrt(2 / fan_in)
self.weight = nn.Parameter(weight)
self.bias = nn.Parameter(bias)
self.pad = padding
def forward(self, input):
weight = F.pad(self.weight * self.multiplier, [1, 1, 1, 1])
weight = (weight[:, :, 1:, 1:] + weight[:, :, :-1, 1:] + weight[:,
:, 1:, :-1] + weight[:, :, :-1, :-1]) / 4
out = F.conv2d(input, weight, self.bias, stride=2, padding=self.pad)
return out
def get_inputs():
return [torch.rand([4, 4, 64, 64])]
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 torch.nn as nn
from math import sqrt
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_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 5 % 5
x0 = xindex % 5
x2 = xindex // 25
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 = x0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (x0 + 4 * x1 + 16 * x2), tmp10 & xmask, other=0.0
)
tmp12 = 0.1767766952966369
tmp13 = tmp11 * tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp10, tmp13, tmp14)
tmp16 = -1 + x1
tmp17 = tmp16 >= tmp1
tmp18 = tmp16 < tmp3
tmp19 = tmp17 & tmp18
tmp20 = tmp19 & tmp6
tmp21 = tmp20 & tmp7
tmp22 = tl.load(in_ptr0 + (-4 + x0 + 4 * x1 + 16 * x2), tmp21 & xmask,
other=0.0)
tmp23 = tmp22 * tmp12
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp21, tmp23, tmp24)
tmp26 = tmp15 + tmp25
tmp27 = -1 + x0
tmp28 = tmp27 >= tmp1
tmp29 = tmp27 < tmp3
tmp30 = tmp8 & tmp28
tmp31 = tmp30 & tmp29
tmp32 = tl.load(in_ptr0 + (-1 + x0 + 4 * x1 + 16 * x2), tmp31 & xmask,
other=0.0)
tmp33 = tmp32 * tmp12
tmp34 = tl.full(tmp33.shape, 0.0, tmp33.dtype)
tmp35 = tl.where(tmp31, tmp33, tmp34)
tmp36 = tmp26 + tmp35
tmp37 = tmp19 & tmp28
tmp38 = tmp37 & tmp29
tmp39 = tl.load(in_ptr0 + (-5 + x0 + 4 * x1 + 16 * x2), tmp38 & xmask,
other=0.0)
tmp40 = tmp39 * tmp12
tmp41 = tl.full(tmp40.shape, 0.0, tmp40.dtype)
tmp42 = tl.where(tmp38, tmp40, tmp41)
tmp43 = tmp36 + tmp42
tmp44 = 0.25
tmp45 = tmp43 * tmp44
tl.store(in_out_ptr0 + x4, tmp45, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 14400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 900 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 64, 64), (16384, 4096, 64, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 5, 5), (100, 25, 5, 1), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_add_div_0[grid(400)](buf1, primals_1, 400, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_1
buf2 = extern_kernels.convolution(primals_3, buf1, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=False,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 30, 30), (3600, 900, 30, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_1[grid(14400)](buf3, primals_2, 14400,
XBLOCK=256, num_warps=4, num_stages=1)
del primals_2
return buf3, primals_3, buf1
class FusedDownsampleNew(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, padding=0):
super().__init__()
weight = torch.randn(out_channel, in_channel, kernel_size, kernel_size)
bias = torch.zeros(out_channel)
fan_in = in_channel * kernel_size * kernel_size
self.multiplier = sqrt(2 / fan_in)
self.weight = nn.Parameter(weight)
self.bias = nn.Parameter(bias)
self.pad = 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]
|
KwonGihyun/DiagonalGAN
|
FusedDownsample
| false
| 8,443
|
[
"MIT"
] | 13
|
9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
https://github.com/KwonGihyun/DiagonalGAN/tree/9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
AttentionCrossEntropy
|
# 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/u2/cu2beycg2t2ghizs6f4qom7bxbxmajhdaakuyq6y2korxywhp6ba.py
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
# Source node to ATen node mapping:
# log_softmax => amax, sub
# Graph fragment:
# %amax : [num_users=1] = call_function[target=torch.ops.aten.amax.default](args = (%arg1_1, [1], True), kwargs = {})
# %sub : [num_users=2] = call_function[target=torch.ops.aten.sub.Tensor](args = (%arg1_1, %amax), kwargs = {})
triton_poi_fused__log_softmax_0 = async_compile.triton('triton_poi_fused__log_softmax_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[256],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = (xindex // 64)
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + (64*x2)), xmask, eviction_policy='evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + (x3), tmp8, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/ln/clnlj6wy43fv2frbtrh4bvz2vyislzyzsoluffjff3th4msvqs47.py
# Topologically Sorted Source Nodes: [log_softmax, cross_loss, mean, loss], Original ATen: [aten._log_softmax, aten.mul, aten.mean, aten.neg]
# Source node to ATen node mapping:
# cross_loss => mul
# log_softmax => exp, log, sub_1, sum_1
# loss => neg
# mean => mean
# 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 = {})
# %mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%arg0_1, %sub_1), kwargs = {})
# %mean : [num_users=1] = call_function[target=torch.ops.aten.mean.default](args = (%mul,), kwargs = {})
# %neg : [num_users=1] = call_function[target=torch.ops.aten.neg.default](args = (%mean,), kwargs = {})
triton_per_fused__log_softmax_mean_mul_neg_1 = async_compile.triton('triton_per_fused__log_softmax_mean_mul_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.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__log_softmax_mean_mul_neg_1', 'mutated_arg_names': ['in_out_ptr0'], '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__log_softmax_mean_mul_neg_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)
r3 = rindex
r0 = rindex % 16
r2 = (rindex // 64)
tmp0 = tl.load(in_ptr0 + (r3), None)
tmp1 = tl.load(in_ptr1 + (r3), None)
tmp2 = tl.load(in_ptr1 + (r0 + (64*r2)), None, eviction_policy='evict_last')
tmp4 = tl.load(in_ptr1 + (16 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp7 = tl.load(in_ptr1 + (32 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp10 = tl.load(in_ptr1 + (48 + r0 + (64*r2)), None, eviction_policy='evict_last')
tmp3 = tl_math.exp(tmp2)
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp3 + tmp5
tmp8 = tl_math.exp(tmp7)
tmp9 = tmp6 + tmp8
tmp11 = tl_math.exp(tmp10)
tmp12 = tmp9 + tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = tmp1 - tmp13
tmp15 = tmp0 * tmp14
tmp16 = tl.broadcast_to(tmp15, [RBLOCK])
tmp18 = triton_helpers.promote_to_tensor(tl.sum(tmp16, 0))
tmp19 = 256.0
tmp20 = tmp18 / tmp19
tmp21 = -tmp20
tl.debug_barrier()
tl.store(in_out_ptr0 + (tl.full([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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [log_softmax], Original ATen: [aten._log_softmax]
stream0 = get_raw_stream(0)
triton_poi_fused__log_softmax_0.run(arg1_1, buf0, 256, grid=grid(256), stream=stream0)
del arg1_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1; del buf1 # reuse
# Topologically Sorted Source Nodes: [log_softmax, cross_loss, mean, loss], Original ATen: [aten._log_softmax, aten.mul, aten.mean, aten.neg]
triton_per_fused__log_softmax_mean_mul_neg_1.run(buf2, arg0_1, buf0, 1, 256, grid=grid(1), stream=stream0)
del arg0_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)
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 AttentionCrossEntropy(nn.Module):
def __init__(self):
super(AttentionCrossEntropy, self).__init__()
def forward(self, input, target):
cross_loss = torch.mul(target.float(), F.log_softmax(input, dim=1))
loss = torch.neg(torch.mean(cross_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_poi_fused__log_softmax_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 256
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x0 = xindex % 16
x2 = xindex // 64
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + (x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp2 = tl.load(in_ptr0 + (16 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp4 = tl.load(in_ptr0 + (32 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp6 = tl.load(in_ptr0 + (48 + x0 + 64 * x2), xmask, eviction_policy=
'evict_last')
tmp3 = triton_helpers.maximum(tmp1, tmp2)
tmp5 = triton_helpers.maximum(tmp3, tmp4)
tmp7 = triton_helpers.maximum(tmp5, tmp6)
tmp8 = tmp0 - tmp7
tl.store(out_ptr0 + x3, tmp8, xmask)
@triton.jit
def triton_per_fused__log_softmax_mean_mul_neg_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)
r3 = rindex
r0 = rindex % 16
r2 = rindex // 64
tmp0 = tl.load(in_ptr0 + r3, None)
tmp1 = tl.load(in_ptr1 + r3, None)
tmp2 = tl.load(in_ptr1 + (r0 + 64 * r2), None, eviction_policy='evict_last'
)
tmp4 = tl.load(in_ptr1 + (16 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp7 = tl.load(in_ptr1 + (32 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp10 = tl.load(in_ptr1 + (48 + r0 + 64 * r2), None, eviction_policy=
'evict_last')
tmp3 = tl_math.exp(tmp2)
tmp5 = tl_math.exp(tmp4)
tmp6 = tmp3 + tmp5
tmp8 = tl_math.exp(tmp7)
tmp9 = tmp6 + tmp8
tmp11 = tl_math.exp(tmp10)
tmp12 = tmp9 + tmp11
tmp13 = tl_math.log(tmp12)
tmp14 = tmp1 - tmp13
tmp15 = tmp0 * tmp14
tmp16 = tl.broadcast_to(tmp15, [RBLOCK])
tmp18 = triton_helpers.promote_to_tensor(tl.sum(tmp16, 0))
tmp19 = 256.0
tmp20 = tmp18 / tmp19
tmp21 = -tmp20
tl.debug_barrier()
tl.store(in_out_ptr0 + tl.full([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((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused__log_softmax_0[grid(256)](arg1_1, buf0, 256,
XBLOCK=128, num_warps=4, num_stages=1)
del arg1_1
buf1 = empty_strided_cuda((), (), torch.float32)
buf2 = buf1
del buf1
triton_per_fused__log_softmax_mean_mul_neg_1[grid(1)](buf2, arg0_1,
buf0, 1, 256, num_warps=2, num_stages=1)
del arg0_1
del buf0
return buf2,
class AttentionCrossEntropyNew(nn.Module):
def __init__(self):
super(AttentionCrossEntropyNew, 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]
|
LindgeW/sentiment-analysis-based-on-attention
|
AttentionCrossEntropy
| false
| 8,444
|
[
"Apache-2.0"
] | 13
|
82ea37c8ef84eec56082d60001b1179b4c12f416
|
https://github.com/LindgeW/sentiment-analysis-based-on-attention/tree/82ea37c8ef84eec56082d60001b1179b4c12f416
|
CausalConv1d
|
# 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/na/cnak6v36bwpgvwoplicu6iqan6mtuzm37prkdglarz33d6ftf5ev.py
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv1d => convolution
# Graph fragment:
# %convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1], [2], [2], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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 = 96
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 6) % 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, 2), (8, 2, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,), padding=(2,), dilation=(2,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 6), (24, 6, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_2, 96, grid=grid(96), stream=stream0)
del primals_2
return (reinterpret_tensor(buf1, (4, 4, 4), (24, 6, 1), 0), primals_1, primals_3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 2), (8, 2, 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)
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 CausalConv1d(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=2, dilation=2):
super(CausalConv1d, self).__init__()
self.padding = dilation
self.causal_conv = nn.Conv1d(in_channels, out_channels, kernel_size,
padding=self.padding, dilation=dilation)
def forward(self, minibatch):
return self.causal_conv(minibatch)[:, :, :-self.padding]
def get_inputs():
return [torch.rand([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 import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 96
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 6 % 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, 2), (8, 2, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,),
padding=(2,), dilation=(2,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 6), (24, 6, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(96)](buf1, primals_2, 96,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
return reinterpret_tensor(buf1, (4, 4, 4), (24, 6, 1), 0
), primals_1, primals_3
class CausalConv1dNew(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=2, dilation=2):
super(CausalConv1dNew, self).__init__()
self.padding = dilation
self.causal_conv = nn.Conv1d(in_channels, out_channels, kernel_size,
padding=self.padding, dilation=dilation)
def forward(self, input_0):
primals_1 = self.causal_conv.weight
primals_2 = self.causal_conv.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3])
return output[0]
|
LittleGuoKe/Entity-Concept-enhanced-Few-shot-Relation-Extraction
|
CausalConv1d
| false
| 8,445
|
[
"MIT"
] | 19
|
b41386bdc70a3b84731bdbf700ff1ba4eda6675d
|
https://github.com/LittleGuoKe/Entity-Concept-enhanced-Few-shot-Relation-Extraction/tree/b41386bdc70a3b84731bdbf700ff1ba4eda6675d
|
MultiHeadAttention
|
# AOT ID: ['0_forward']
from ctypes import c_void_p, c_long, c_int
import torch
import math
import random
import os
import tempfile
from math import inf, nan
from torch._inductor.hooks import run_intermediate_hooks
from torch._inductor.utils import maybe_profile
from torch._inductor.codegen.memory_planning import _align as align
from torch import device, empty_strided
from torch._inductor.async_compile import AsyncCompile
from torch._inductor.select_algorithm import extern_kernels
from torch._inductor.codegen.multi_kernel import MultiKernelCall
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid, split_scan_grid, grid_combo_kernels, start_graph, end_graph
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
aten = torch.ops.aten
inductor_ops = torch.ops.inductor
_quantized = torch.ops._quantized
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cpu = torch._C._dynamo.guards._empty_strided_cpu
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
empty_strided_xpu = torch._C._dynamo.guards._empty_strided_xpu
reinterpret_tensor = torch._C._dynamo.guards._reinterpret_tensor
alloc_from_pool = torch.ops.inductor._alloc_from_pool
async_compile = AsyncCompile()
# kernel path: runs/run_shard_1/inductor_cache/gh/cghdn4pihhs4yx6f6a4ig3pjgfy5zh6dghclq5xeandkuwdvuk5c.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_7,), 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: 'i32'}, 'device': 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 = 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)
tl.store(out_ptr0 + (x0), tmp0, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (16, 4), (4, 1))
assert_size_stride(primals_3, (16, ), (1, ))
assert_size_stride(primals_4, (16, 4), (4, 1))
assert_size_stride(primals_5, (16, ), (1, ))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (16, 4), (4, 1))
assert_size_stride(primals_8, (16, ), (1, ))
assert_size_stride(primals_9, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_10, (4, 16), (16, 1))
assert_size_stride(primals_11, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 16), (1, 4), 0), alpha=1, beta=1, out=buf0)
del primals_2
del primals_3
buf1 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_1], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 16), (1, 4), 0), alpha=1, beta=1, out=buf1)
del primals_4
del primals_5
buf2 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_2], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_9, (16, 4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 16), (1, 4), 0), alpha=1, beta=1, out=buf2)
del primals_7
del primals_8
# Topologically Sorted Source Nodes: [], Original ATen: []
buf3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 4, 16, 1), 0), reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 4, 16, 1), 0), reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 4, 16, 1), 0), None, True, scale=0.5)
buf4 = buf3[0]
buf5 = buf3[1]
buf6 = buf3[2]
buf7 = buf3[3]
del buf3
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [contiguous], Original ATen: [aten.clone]
stream0 = get_raw_stream(0)
triton_poi_fused_clone_0.run(buf4, buf8, 256, grid=grid(256), stream=stream0)
buf9 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
# Topologically Sorted Source Nodes: [linear_3], Original ATen: [aten.addmm]
extern_kernels.addmm(primals_11, reinterpret_tensor(buf8, (16, 16), (16, 1), 0), reinterpret_tensor(primals_10, (16, 4), (1, 16), 0), alpha=1, beta=1, out=buf9)
del primals_11
return (reinterpret_tensor(buf9, (4, 4, 4), (16, 4, 1), 0), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0), reinterpret_tensor(primals_6, (16, 4), (4, 1), 0), reinterpret_tensor(primals_9, (16, 4), (4, 1), 0), reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 4, 16, 1), 0), reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 4, 16, 1), 0), reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 4, 16, 1), 0), buf4, buf5, buf6, buf7, reinterpret_tensor(buf8, (16, 16), (16, 1), 0), primals_10, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_5 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_6 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_7 = rand_strided((16, 4), (4, 1), device='cuda:0', dtype=torch.float32)
primals_8 = rand_strided((16, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_9 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_10 = rand_strided((4, 16), (16, 1), device='cuda:0', dtype=torch.float32)
primals_11 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3, primals_4, primals_5, primals_6, primals_7, primals_8, primals_9, primals_10, primals_11])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import math
import torch
from torch import nn
from torch.nn import functional as F
from numpy import inf
from math import inf
def Linear(in_features, out_features, bias=True):
m = nn.Linear(in_features, out_features, bias)
nn.init.xavier_uniform_(m.weight)
if bias:
nn.init.constant_(m.bias, 0.0)
return m
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, d_qk, d_v, num_head):
super().__init__()
self.d_model = d_model
self.num_head = num_head
self.d_qk = d_qk
self.d_v = d_v
self.W_Q = Linear(d_model, num_head * d_qk)
self.W_K = Linear(d_model, num_head * d_qk)
self.W_V = Linear(d_model, num_head * d_v)
self.W_out = Linear(d_v * num_head, d_model)
def ScaledDotProductAttention(self, query, keys, values, mask=None):
score = torch.matmul(query, keys.transpose(-1, -2)) / math.sqrt(self
.d_model)
if mask is not None:
score.masked_fill_(mask.unsqueeze(1), -inf)
weight = F.softmax(score, dim=-1)
return torch.matmul(weight, values)
def forward(self, Q, K, V, mask=None):
batch_size = Q.size(0)
query = self.W_Q(Q).view(batch_size, Q.size(1), self.num_head, self
.d_qk)
keys = self.W_K(K).view(batch_size, K.size(1), self.num_head, self.d_qk
)
values = self.W_V(V).view(batch_size, V.size(1), self.num_head,
self.d_v)
query.transpose_(1, 2)
keys.transpose_(1, 2)
values.transpose_(1, 2)
outputs = self.ScaledDotProductAttention(query, keys, values, mask)
del query, keys, values
outputs = outputs.transpose(1, 2).contiguous().view(batch_size, -1,
self.d_v * self.num_head)
return self.W_out(outputs)
def cal_one_vector(self, vector, memory, memory_new, i):
batch_size = vector.size(0)
query = self.W_Q(vector).view(batch_size, vector.size(1), self.
num_head, self.d_qk)
key = self.W_K(vector).view(batch_size, vector.size(1), self.
num_head, self.d_qk)
value = self.W_V(vector).view(batch_size, vector.size(1), self.
num_head, self.d_v)
query.transpose_(1, 2)
key.transpose_(1, 2)
value.transpose_(1, 2)
outputs = torch.cat((key.unsqueeze(-1), value.unsqueeze(-1)), dim=-1)
del key, value
if memory is not None:
if memory_new is None:
memory_new = torch.cat((memory[:, i, ...], outputs), dim=2
).unsqueeze(1)
else:
_m = torch.cat((memory[:, i, ...], outputs), dim=2)
memory_new = torch.cat((memory_new, _m.unsqueeze(1)), dim=1)
elif memory_new is None:
memory_new = outputs.unsqueeze(1)
else:
memory_new = torch.cat((memory_new, outputs.unsqueeze(1)), dim=1)
outputs = self.ScaledDotProductAttention(query, memory_new[:, i,
..., 0], memory_new[:, i, ..., 1])
outputs = outputs.transpose(1, 2).contiguous().view(batch_size, -1,
self.d_v * self.num_head)
return self.W_out(outputs), memory_new
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_qk': 4, 'd_v': 4, 'num_head': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
import math
from torch import nn
from torch.nn import functional as F
from numpy import inf
from math import inf
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 = 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)
tl.store(out_ptr0 + x0, tmp0, xmask)
def call(args):
(primals_1, primals_2, primals_3, primals_4, primals_5, primals_6,
primals_7, primals_8, primals_9, primals_10, primals_11) = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_2, (16, 4), (4, 1))
assert_size_stride(primals_3, (16,), (1,))
assert_size_stride(primals_4, (16, 4), (4, 1))
assert_size_stride(primals_5, (16,), (1,))
assert_size_stride(primals_6, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_7, (16, 4), (4, 1))
assert_size_stride(primals_8, (16,), (1,))
assert_size_stride(primals_9, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_10, (4, 16), (16, 1))
assert_size_stride(primals_11, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.addmm(primals_3, reinterpret_tensor(primals_1, (16,
4), (4, 1), 0), reinterpret_tensor(primals_2, (4, 16), (1, 4),
0), alpha=1, beta=1, out=buf0)
del primals_2
del primals_3
buf1 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.addmm(primals_5, reinterpret_tensor(primals_6, (16,
4), (4, 1), 0), reinterpret_tensor(primals_4, (4, 16), (1, 4),
0), alpha=1, beta=1, out=buf1)
del primals_4
del primals_5
buf2 = empty_strided_cuda((16, 16), (16, 1), torch.float32)
extern_kernels.addmm(primals_8, reinterpret_tensor(primals_9, (16,
4), (4, 1), 0), reinterpret_tensor(primals_7, (4, 16), (1, 4),
0), alpha=1, beta=1, out=buf2)
del primals_7
del primals_8
buf3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(
reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 4, 16, 1), 0),
reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 4, 16, 1), 0),
reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 4, 16, 1), 0), None,
True, scale=0.5)
buf4 = buf3[0]
buf5 = buf3[1]
buf6 = buf3[2]
buf7 = buf3[3]
del buf3
buf8 = empty_strided_cuda((4, 4, 4, 4), (64, 16, 4, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_clone_0[grid(256)](buf4, buf8, 256, XBLOCK=128,
num_warps=4, num_stages=1)
buf9 = empty_strided_cuda((16, 4), (4, 1), torch.float32)
extern_kernels.addmm(primals_11, reinterpret_tensor(buf8, (16, 16),
(16, 1), 0), reinterpret_tensor(primals_10, (16, 4), (1, 16), 0
), alpha=1, beta=1, out=buf9)
del primals_11
return reinterpret_tensor(buf9, (4, 4, 4), (16, 4, 1), 0
), reinterpret_tensor(primals_1, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_6, (16, 4), (4, 1), 0
), reinterpret_tensor(primals_9, (16, 4), (4, 1), 0
), reinterpret_tensor(buf0, (4, 4, 4, 4), (64, 4, 16, 1), 0
), reinterpret_tensor(buf1, (4, 4, 4, 4), (64, 4, 16, 1), 0
), reinterpret_tensor(buf2, (4, 4, 4, 4), (64, 4, 16, 1), 0
), buf4, buf5, buf6, buf7, reinterpret_tensor(buf8, (16, 16), (16,
1), 0), primals_10
def Linear(in_features, out_features, bias=True):
m = nn.Linear(in_features, out_features, bias)
nn.init.xavier_uniform_(m.weight)
if bias:
nn.init.constant_(m.bias, 0.0)
return m
class MultiHeadAttentionNew(nn.Module):
def __init__(self, d_model, d_qk, d_v, num_head):
super().__init__()
self.d_model = d_model
self.num_head = num_head
self.d_qk = d_qk
self.d_v = d_v
self.W_Q = Linear(d_model, num_head * d_qk)
self.W_K = Linear(d_model, num_head * d_qk)
self.W_V = Linear(d_model, num_head * d_v)
self.W_out = Linear(d_v * num_head, d_model)
def ScaledDotProductAttention(self, query, keys, values, mask=None):
score = torch.matmul(query, keys.transpose(-1, -2)) / math.sqrt(self
.d_model)
if mask is not None:
score.masked_fill_(mask.unsqueeze(1), -inf)
weight = F.softmax(score, dim=-1)
return torch.matmul(weight, values)
def cal_one_vector(self, vector, memory, memory_new, i):
batch_size = vector.size(0)
query = self.W_Q(vector).view(batch_size, vector.size(1), self.
num_head, self.d_qk)
key = self.W_K(vector).view(batch_size, vector.size(1), self.
num_head, self.d_qk)
value = self.W_V(vector).view(batch_size, vector.size(1), self.
num_head, self.d_v)
query.transpose_(1, 2)
key.transpose_(1, 2)
value.transpose_(1, 2)
outputs = torch.cat((key.unsqueeze(-1), value.unsqueeze(-1)), dim=-1)
del key, value
if memory is not None:
if memory_new is None:
memory_new = torch.cat((memory[:, i, ...], outputs), dim=2
).unsqueeze(1)
else:
_m = torch.cat((memory[:, i, ...], outputs), dim=2)
memory_new = torch.cat((memory_new, _m.unsqueeze(1)), dim=1)
elif memory_new is None:
memory_new = outputs.unsqueeze(1)
else:
memory_new = torch.cat((memory_new, outputs.unsqueeze(1)), dim=1)
outputs = self.ScaledDotProductAttention(query, memory_new[:, i,
..., 0], memory_new[:, i, ..., 1])
outputs = outputs.transpose(1, 2).contiguous().view(batch_size, -1,
self.d_v * self.num_head)
return self.W_out(outputs), memory_new
def forward(self, input_0, input_1, input_2):
primals_2 = self.W_Q.weight
primals_3 = self.W_Q.bias
primals_4 = self.W_K.weight
primals_5 = self.W_K.bias
primals_7 = self.W_V.weight
primals_8 = self.W_V.bias
primals_10 = self.W_out.weight
primals_11 = self.W_out.bias
primals_1 = input_0
primals_6 = input_1
primals_9 = input_2
output = call([primals_1, primals_2, primals_3, primals_4,
primals_5, primals_6, primals_7, primals_8, primals_9,
primals_10, primals_11])
return output[0]
|
L-Zhe/FasySeq
|
MultiHeadAttention
| false
| 8,446
|
[
"Apache-2.0"
] | 34
|
2cd2abd290666b1e118d8ad11c973b58ca4f0573
|
https://github.com/L-Zhe/FasySeq/tree/2cd2abd290666b1e118d8ad11c973b58ca4f0573
|
DenseBlock
|
# 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/na/cnak6v36bwpgvwoplicu6iqan6mtuzm37prkdglarz33d6ftf5ev.py
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
# Source node to ATen node mapping:
# conv1d => convolution
# Graph fragment:
# %convolution : [num_users=2] = call_function[target=torch.ops.aten.convolution.default](args = (%primals_3, %primals_1, %primals_2, [1], [2], [2], False, [0], 1), kwargs = {})
triton_poi_fused_convolution_0 = async_compile.triton('triton_poi_fused_convolution_0', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[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_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 = 96
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 6) % 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/25/c254wu6xcta5qkwqoh6zxsylo4hl72c5ne4lzeafhps64ivdeyed.py
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.cat]
# Source node to ATen node mapping:
# out => cat
# Graph fragment:
# %cat : [num_users=1] = call_function[target=torch.ops.aten.cat.default](args = ([%primals_3, %mul], 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: '*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_cat_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_cat_1(in_ptr0, in_ptr1, in_ptr2, 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 // 4) % 8
x0 = xindex % 4
x2 = (xindex // 32)
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 + (4*x1) + (16*x2)), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tmp7 = tl.full([1], 8, tl.int64)
tmp8 = tmp0 < tmp7
tmp9 = tl.load(in_ptr1 + (x0 + (6*((-4) + x1)) + (24*x2)), tmp6 & xmask, other=0.0)
tmp10 = libdevice.tanh(tmp9)
tmp11 = tl.load(in_ptr2 + (x0 + (6*((-4) + x1)) + (24*x2)), tmp6 & xmask, other=0.0)
tmp12 = tl.sigmoid(tmp11)
tmp13 = tmp10 * tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp6, tmp13, tmp14)
tmp16 = tl.where(tmp4, tmp5, tmp15)
tl.store(out_ptr0 + (x3), tmp16, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 2), (8, 2, 1))
assert_size_stride(primals_2, (4, ), (1, ))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4, 2), (8, 2, 1))
assert_size_stride(primals_5, (4, ), (1, ))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,), padding=(2,), dilation=(2,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 6), (24, 6, 1))
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
stream0 = get_raw_stream(0)
triton_poi_fused_convolution_0.run(buf1, primals_2, 96, grid=grid(96), stream=stream0)
del primals_2
# Topologically Sorted Source Nodes: [conv1d_1], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(primals_3, primals_4, stride=(1,), padding=(2,), dilation=(2,), transposed=False, output_padding=(0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 6), (24, 6, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [conv1d_1], Original ATen: [aten.convolution]
triton_poi_fused_convolution_0.run(buf3, primals_5, 96, grid=grid(96), stream=stream0)
del primals_5
buf4 = empty_strided_cuda((4, 8, 4), (32, 4, 1), torch.float32)
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.cat]
triton_poi_fused_cat_1.run(primals_3, buf1, buf3, buf4, 128, grid=grid(128), stream=stream0)
return (buf4, primals_1, primals_3, primals_4, buf1, buf3, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((4, 4, 2), (8, 2, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4), (16, 4, 1), device='cuda:0', dtype=torch.float32)
primals_4 = rand_strided((4, 4, 2), (8, 2, 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
from torch.nn import functional as F
class CausalConv1d(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=2, dilation=2):
super(CausalConv1d, self).__init__()
self.padding = dilation
self.causal_conv = nn.Conv1d(in_channels, out_channels, kernel_size,
padding=self.padding, dilation=dilation)
def forward(self, minibatch):
return self.causal_conv(minibatch)[:, :, :-self.padding]
class DenseBlock(nn.Module):
def __init__(self, in_channels, filters, dilation=2):
super(DenseBlock, self).__init__()
self.causal_conv1 = CausalConv1d(in_channels, filters, dilation=
dilation)
self.causal_conv2 = CausalConv1d(in_channels, filters, dilation=
dilation)
def forward(self, minibatch):
tanh = F.tanh(self.causal_conv1(minibatch))
sig = F.sigmoid(self.causal_conv2(minibatch))
out = torch.cat([minibatch, tanh * sig], dim=1)
return out
def get_inputs():
return [torch.rand([4, 4, 4])]
def get_init_inputs():
return [[], {'in_channels': 4, 'filters': 4}]
|
import torch
from torch._inductor.select_algorithm import extern_kernels
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
from torch._inductor.runtime.triton_helpers import libdevice
from torch import nn
assert_size_stride = torch._C._dynamo.guards.assert_size_stride
empty_strided_cuda = torch._C._dynamo.guards._empty_strided_cuda
@triton.jit
def triton_poi_fused_convolution_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 96
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 6 % 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_cat_1(in_ptr0, in_ptr1, in_ptr2, 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 // 4 % 8
x0 = xindex % 4
x2 = xindex // 32
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 + 4 * x1 + 16 * x2), tmp4 & xmask, other=0.0)
tmp6 = tmp0 >= tmp3
tl.full([1], 8, tl.int64)
tmp9 = tl.load(in_ptr1 + (x0 + 6 * (-4 + x1) + 24 * x2), tmp6 & xmask,
other=0.0)
tmp10 = libdevice.tanh(tmp9)
tmp11 = tl.load(in_ptr2 + (x0 + 6 * (-4 + x1) + 24 * x2), tmp6 & xmask,
other=0.0)
tmp12 = tl.sigmoid(tmp11)
tmp13 = tmp10 * tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp6, tmp13, tmp14)
tmp16 = tl.where(tmp4, tmp5, tmp15)
tl.store(out_ptr0 + x3, tmp16, xmask)
def call(args):
primals_1, primals_2, primals_3, primals_4, primals_5 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 2), (8, 2, 1))
assert_size_stride(primals_2, (4,), (1,))
assert_size_stride(primals_3, (4, 4, 4), (16, 4, 1))
assert_size_stride(primals_4, (4, 4, 2), (8, 2, 1))
assert_size_stride(primals_5, (4,), (1,))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = extern_kernels.convolution(primals_3, primals_1, stride=(1,),
padding=(2,), dilation=(2,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf0, (4, 4, 6), (24, 6, 1))
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_convolution_0[grid(96)](buf1, primals_2, 96,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
buf2 = extern_kernels.convolution(primals_3, primals_4, stride=(1,),
padding=(2,), dilation=(2,), transposed=False, output_padding=(
0,), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 6), (24, 6, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_0[grid(96)](buf3, primals_5, 96,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_5
buf4 = empty_strided_cuda((4, 8, 4), (32, 4, 1), torch.float32)
triton_poi_fused_cat_1[grid(128)](primals_3, buf1, buf3, buf4, 128,
XBLOCK=128, num_warps=4, num_stages=1)
return buf4, primals_1, primals_3, primals_4, buf1, buf3
class CausalConv1d(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=2, dilation=2):
super(CausalConv1d, self).__init__()
self.padding = dilation
self.causal_conv = nn.Conv1d(in_channels, out_channels, kernel_size,
padding=self.padding, dilation=dilation)
def forward(self, minibatch):
return self.causal_conv(minibatch)[:, :, :-self.padding]
class DenseBlockNew(nn.Module):
def __init__(self, in_channels, filters, dilation=2):
super(DenseBlockNew, self).__init__()
self.causal_conv1 = CausalConv1d(in_channels, filters, dilation=
dilation)
self.causal_conv2 = CausalConv1d(in_channels, filters, dilation=
dilation)
def forward(self, input_0):
primals_1 = self.causal_conv1.causal_conv.weight
primals_2 = self.causal_conv1.causal_conv.bias
primals_4 = self.causal_conv2.causal_conv.weight
primals_5 = self.causal_conv2.causal_conv.bias
primals_3 = input_0
output = call([primals_1, primals_2, primals_3, primals_4, primals_5])
return output[0]
|
LittleGuoKe/Entity-Concept-enhanced-Few-shot-Relation-Extraction
|
DenseBlock
| false
| 8,447
|
[
"MIT"
] | 19
|
b41386bdc70a3b84731bdbf700ff1ba4eda6675d
|
https://github.com/LittleGuoKe/Entity-Concept-enhanced-Few-shot-Relation-Extraction/tree/b41386bdc70a3b84731bdbf700ff1ba4eda6675d
|
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/yu/cyuvi6hfyqjmxhq5wefkgmloticjjwb4aqsy2b3ssfrdlttmumth.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_1, %primals_2), kwargs = {})
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%primals_3, %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
x1 = (xindex // 16) % 4
tmp0 = tl.load(in_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr1 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + (x3), xmask)
tmp3 = tmp1 * tmp2
tmp4 = tmp0 + tmp3
tl.store(out_ptr0 + (x3), tmp4, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 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, add], 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, 256, grid=grid(256), stream=stream0)
del primals_1
del primals_3
return (buf0, primals_2, )
def benchmark_compiled_module(times=10, repeat=10):
from torch._dynamo.testing import rand_strided
from torch._inductor.utils import print_performance
primals_1 = rand_strided((1, 4, 1, 1), (4, 1, 1, 1), device='cuda:0', dtype=torch.float32)
primals_2 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
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 NoiseInjection(nn.Module):
def __init__(self, channel):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1, channel, 1, 1))
def forward(self, image, noise):
return image + self.weight * noise
def get_inputs():
return [torch.rand([4, 4, 4, 4]), torch.rand([4, 4, 4, 4])]
def get_init_inputs():
return [[], {'channel': 4}]
|
import torch
import triton
import triton.language as tl
from torch._inductor.runtime.triton_heuristics import grid
from torch._C import _cuda_getCurrentRawStream as get_raw_stream
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
x3 = xindex
x1 = xindex // 16 % 4
tmp0 = tl.load(in_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr1 + x1, xmask, eviction_policy='evict_last')
tmp2 = tl.load(in_ptr2 + x3, xmask)
tmp3 = tmp1 * tmp2
tmp4 = tmp0 + tmp3
tl.store(out_ptr0 + x3, tmp4, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (1, 4, 1, 1), (4, 1, 1, 1))
assert_size_stride(primals_2, (4, 4, 4, 4), (64, 16, 4, 1))
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_add_mul_0[grid(256)](primals_3, primals_1,
primals_2, buf0, 256, XBLOCK=128, num_warps=4, num_stages=1)
del primals_1
del primals_3
return buf0, primals_2
class NoiseInjectionNew(nn.Module):
def __init__(self, channel):
super().__init__()
self.weight = nn.Parameter(torch.zeros(1, channel, 1, 1))
def forward(self, input_0, input_1):
primals_1 = self.weight
primals_2 = input_0
primals_3 = input_1
output = call([primals_1, primals_2, primals_3])
return output[0]
|
KwonGihyun/DiagonalGAN
|
NoiseInjection
| false
| 8,448
|
[
"MIT"
] | 13
|
9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
https://github.com/KwonGihyun/DiagonalGAN/tree/9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
SymmetricPad2d
|
# 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/g7/cg7ad5vsmxcpmslazz5kudd2typtohhdmh3lhfs7jnkgyafk34sv.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 = ([%constant_pad_nd, %constant_pad_nd_1, %constant_pad_nd_2, %constant_pad_nd_3], 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': 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, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = (xindex // 25) % 4
x1 = (xindex // 5) % 5
x0 = xindex % 5
x3 = (xindex // 100)
x6 = xindex
tmp0 = x2
tmp1 = tl.full([1], 0, tl.int64)
tmp2 = tmp0 >= tmp1
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = (-1) + x1
tmp6 = tmp5 >= tmp1
tmp7 = (-1) + x0
tmp8 = tmp7 >= tmp1
tmp9 = tmp6 & tmp8
tmp10 = tmp9 & tmp4
tmp11 = tl.load(in_ptr0 + ((-5) + x0 + (4*x1) + (64*x3)), tmp10 & xmask, eviction_policy='evict_last', other=0.0)
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tmp15 = tl.full([1], 2, tl.int64)
tmp16 = tmp0 < tmp15
tmp17 = tmp14 & tmp16
tmp18 = x1
tmp19 = tl.full([1], 4, tl.int64)
tmp20 = tmp18 < tmp19
tmp21 = tmp20 & tmp8
tmp22 = tmp21 & tmp17
tmp23 = tl.load(in_ptr0 + (15 + x0 + (4*x1) + (64*x3)), tmp22 & xmask, eviction_policy='evict_last', other=0.0)
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp17, tmp23, tmp24)
tmp26 = tmp0 >= tmp15
tmp27 = tl.full([1], 3, tl.int64)
tmp28 = tmp0 < tmp27
tmp29 = tmp26 & tmp28
tmp30 = x0
tmp31 = tmp30 < tmp19
tmp32 = tmp6 & tmp31
tmp33 = tmp32 & tmp29
tmp34 = tl.load(in_ptr0 + (28 + x0 + (4*x1) + (64*x3)), tmp33 & xmask, eviction_policy='evict_last', other=0.0)
tmp35 = tl.full(tmp34.shape, 0.0, tmp34.dtype)
tmp36 = tl.where(tmp29, tmp34, tmp35)
tmp37 = tmp0 >= tmp27
tmp38 = tmp0 < tmp19
tmp39 = tmp20 & tmp31
tmp40 = tmp39 & tmp37
tmp41 = tl.load(in_ptr0 + (48 + x0 + (4*x1) + (64*x3)), tmp40 & xmask, eviction_policy='evict_last', other=0.0)
tmp42 = tl.full(tmp41.shape, 0.0, tmp41.dtype)
tmp43 = tl.where(tmp37, tmp41, tmp42)
tmp44 = tl.where(tmp29, tmp36, tmp43)
tmp45 = tl.where(tmp17, tmp25, tmp44)
tmp46 = tl.where(tmp4, tmp13, tmp45)
tl.store(out_ptr0 + (x6), tmp46, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 5, 5), (100, 25, 5, 1), torch.float32)
# Topologically Sorted Source Nodes: [x], Original ATen: [aten.cat]
stream0 = get_raw_stream(0)
triton_poi_fused_cat_0.run(arg0_1, buf0, 400, grid=grid(400), 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 SymmetricPad2d(nn.Module):
"""symmetric 0-pad to splited tensors and concat"""
def __init__(self, pad=1):
super(SymmetricPad2d, self).__init__()
self.padding1 = nn.ZeroPad2d((pad, 0, pad, 0))
self.padding2 = nn.ZeroPad2d((pad, 0, 0, pad))
self.padding3 = nn.ZeroPad2d((0, pad, pad, 0))
self.padding4 = nn.ZeroPad2d((0, pad, 0, pad))
def forward(self, x):
sub = x.shape[1] // 4
x1, x2, x3, x4 = x[:, :sub], x[:, sub:2 * sub], x[:, 2 * sub:3 * sub
], x[:, 3 * sub:]
x1, x2, x3, x4 = self.padding1(x1), self.padding2(x2), self.padding3(x3
), self.padding4(x4)
x = torch.cat((x1, x2, x3, x4), dim=1)
return 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_cat_0(in_ptr0, out_ptr0, xnumel, XBLOCK: tl.constexpr):
xnumel = 400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x2 = xindex // 25 % 4
x1 = xindex // 5 % 5
x0 = xindex % 5
x3 = xindex // 100
x6 = xindex
tmp0 = x2
tmp1 = tl.full([1], 0, tl.int64)
tmp3 = tl.full([1], 1, tl.int64)
tmp4 = tmp0 < tmp3
tmp5 = -1 + x1
tmp6 = tmp5 >= tmp1
tmp7 = -1 + x0
tmp8 = tmp7 >= tmp1
tmp9 = tmp6 & tmp8
tmp10 = tmp9 & tmp4
tmp11 = tl.load(in_ptr0 + (-5 + x0 + 4 * x1 + 64 * x3), tmp10 & xmask,
eviction_policy='evict_last', other=0.0)
tmp12 = tl.full(tmp11.shape, 0.0, tmp11.dtype)
tmp13 = tl.where(tmp4, tmp11, tmp12)
tmp14 = tmp0 >= tmp3
tmp15 = tl.full([1], 2, tl.int64)
tmp16 = tmp0 < tmp15
tmp17 = tmp14 & tmp16
tmp18 = x1
tmp19 = tl.full([1], 4, tl.int64)
tmp20 = tmp18 < tmp19
tmp21 = tmp20 & tmp8
tmp22 = tmp21 & tmp17
tmp23 = tl.load(in_ptr0 + (15 + x0 + 4 * x1 + 64 * x3), tmp22 & xmask,
eviction_policy='evict_last', other=0.0)
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp17, tmp23, tmp24)
tmp26 = tmp0 >= tmp15
tmp27 = tl.full([1], 3, tl.int64)
tmp28 = tmp0 < tmp27
tmp29 = tmp26 & tmp28
tmp30 = x0
tmp31 = tmp30 < tmp19
tmp32 = tmp6 & tmp31
tmp33 = tmp32 & tmp29
tmp34 = tl.load(in_ptr0 + (28 + x0 + 4 * x1 + 64 * x3), tmp33 & xmask,
eviction_policy='evict_last', other=0.0)
tmp35 = tl.full(tmp34.shape, 0.0, tmp34.dtype)
tmp36 = tl.where(tmp29, tmp34, tmp35)
tmp37 = tmp0 >= tmp27
tmp39 = tmp20 & tmp31
tmp40 = tmp39 & tmp37
tmp41 = tl.load(in_ptr0 + (48 + x0 + 4 * x1 + 64 * x3), tmp40 & xmask,
eviction_policy='evict_last', other=0.0)
tmp42 = tl.full(tmp41.shape, 0.0, tmp41.dtype)
tmp43 = tl.where(tmp37, tmp41, tmp42)
tmp44 = tl.where(tmp29, tmp36, tmp43)
tmp45 = tl.where(tmp17, tmp25, tmp44)
tmp46 = tl.where(tmp4, tmp13, tmp45)
tl.store(out_ptr0 + x6, tmp46, xmask)
def call(args):
arg0_1, = args
args.clear()
assert_size_stride(arg0_1, (4, 4, 4, 4), (64, 16, 4, 1))
with torch.cuda._DeviceGuard(0):
torch.cuda.set_device(0)
buf0 = empty_strided_cuda((4, 4, 5, 5), (100, 25, 5, 1), torch.float32)
get_raw_stream(0)
triton_poi_fused_cat_0[grid(400)](arg0_1, buf0, 400, XBLOCK=128,
num_warps=4, num_stages=1)
del arg0_1
return buf0,
class SymmetricPad2dNew(nn.Module):
"""symmetric 0-pad to splited tensors and concat"""
def __init__(self, pad=1):
super(SymmetricPad2dNew, self).__init__()
self.padding1 = nn.ZeroPad2d((pad, 0, pad, 0))
self.padding2 = nn.ZeroPad2d((pad, 0, 0, pad))
self.padding3 = nn.ZeroPad2d((0, pad, pad, 0))
self.padding4 = nn.ZeroPad2d((0, pad, 0, pad))
def forward(self, input_0):
arg0_1 = input_0
output = call([arg0_1])
return output[0]
|
Lee-Gihun/Micronet_GSJ
|
SymmetricPad2d
| false
| 8,449
|
[
"MIT"
] | 12
|
72289bb66507b6c3b4d14f2e5916dec718a1b198
|
https://github.com/Lee-Gihun/Micronet_GSJ/tree/72289bb66507b6c3b4d14f2e5916dec718a1b198
|
FusedUpsample
|
# 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/qf/cqfpjcbno7orqx35hioypcgzp7qglgr7wqdebmvpel54tvilauub.py
# Topologically Sorted Source Nodes: [add, add_1, add_2, weight_1], Original ATen: [aten.add, aten.div]
# Source node to ATen node mapping:
# add => add
# add_1 => add_1
# add_2 => add_2
# weight_1 => div
# Graph fragment:
# %add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%slice_4, %slice_8), kwargs = {})
# %add_1 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add, %slice_12), kwargs = {})
# %add_2 : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%add_1, %slice_16), kwargs = {})
# %div : [num_users=2] = call_function[target=torch.ops.aten.div.Tensor](args = (%add_2, 4), kwargs = {})
triton_poi_fused_add_div_0 = async_compile.triton('triton_poi_fused_add_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=[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_add_div_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_add_div_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = (xindex // 5) % 5
x0 = xindex % 5
x2 = (xindex // 25)
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 = x0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (x0 + (4*x1) + (16*x2)), tmp10 & xmask, other=0.0)
tmp12 = 0.1767766952966369
tmp13 = tmp11 * tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp10, tmp13, tmp14)
tmp16 = (-1) + x1
tmp17 = tmp16 >= tmp1
tmp18 = tmp16 < tmp3
tmp19 = tmp17 & tmp18
tmp20 = tmp19 & tmp6
tmp21 = tmp20 & tmp7
tmp22 = tl.load(in_ptr0 + ((-4) + x0 + (4*x1) + (16*x2)), tmp21 & xmask, other=0.0)
tmp23 = tmp22 * tmp12
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp21, tmp23, tmp24)
tmp26 = tmp15 + tmp25
tmp27 = (-1) + x0
tmp28 = tmp27 >= tmp1
tmp29 = tmp27 < tmp3
tmp30 = tmp8 & tmp28
tmp31 = tmp30 & tmp29
tmp32 = tl.load(in_ptr0 + ((-1) + x0 + (4*x1) + (16*x2)), tmp31 & xmask, other=0.0)
tmp33 = tmp32 * tmp12
tmp34 = tl.full(tmp33.shape, 0.0, tmp33.dtype)
tmp35 = tl.where(tmp31, tmp33, tmp34)
tmp36 = tmp26 + tmp35
tmp37 = tmp19 & tmp28
tmp38 = tmp37 & tmp29
tmp39 = tl.load(in_ptr0 + ((-5) + x0 + (4*x1) + (16*x2)), tmp38 & xmask, other=0.0)
tmp40 = tmp39 * tmp12
tmp41 = tl.full(tmp40.shape, 0.0, tmp40.dtype)
tmp42 = tl.where(tmp38, tmp40, tmp41)
tmp43 = tmp36 + tmp42
tmp44 = 0.25
tmp45 = tmp43 * tmp44
tl.store(in_out_ptr0 + (x4), tmp45, xmask)
''', device_str='cuda')
# kernel path: runs/run_shard_1/inductor_cache/5a/c5azowqg6ohglmafftsxyhwic76fsypgkun7hqb4dv4k54eda6nu.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, %div, %primals_2, [2, 2], [0, 0], [1, 1], True, [0, 0], 1), kwargs = {})
triton_poi_fused_convolution_1 = async_compile.triton('triton_poi_fused_convolution_1', '''
import triton
import triton.language as tl
from triton.compiler.compiler import AttrsDescriptor
from torch._inductor.runtime import triton_helpers, triton_heuristics
from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, instance_descriptor, DeviceProperties
@triton_heuristics.pointwise(
size_hints=[2048],
filename=__file__,
triton_meta={'signature': {0: '*fp32', 1: '*fp32', 2: 'i32'}, 'device': DeviceProperties(type='cuda', index=0, cc=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 = 1936
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = (xindex // 121) % 4
tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + (x3), tmp2, xmask)
''', device_str='cuda')
async_compile.wait(globals())
del async_compile
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4, ), (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, 5, 5), (100, 25, 5, 1), torch.float32)
buf1 = buf0; del buf0 # reuse
# Topologically Sorted Source Nodes: [add, add_1, add_2, weight_1], Original ATen: [aten.add, aten.div]
stream0 = get_raw_stream(0)
triton_poi_fused_add_div_0.run(buf1, primals_1, 400, grid=grid(400), stream=stream0)
del primals_1
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
buf2 = extern_kernels.convolution(primals_3, buf1, stride=(2, 2), padding=(0, 0), dilation=(1, 1), transposed=True, output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 11, 11), (484, 121, 11, 1))
buf3 = buf2; del buf2 # reuse
# Topologically Sorted Source Nodes: [out], Original ATen: [aten.convolution]
triton_poi_fused_convolution_1.run(buf3, primals_2, 1936, grid=grid(1936), stream=stream0)
del primals_2
return (buf3, primals_3, 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, ), (1, ), device='cuda:0', dtype=torch.float32)
primals_3 = rand_strided((4, 4, 4, 4), (64, 16, 4, 1), device='cuda:0', dtype=torch.float32)
fn = lambda: call([primals_1, primals_2, primals_3])
return print_performance(fn, times=times, repeat=repeat)
if __name__ == "__main__":
from torch._inductor.wrapper_benchmark import compiled_module_main
compiled_module_main('None', benchmark_compiled_module)
|
import torch
import torch.nn as nn
import torch.nn.functional as F
from math import sqrt
class FusedUpsample(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, padding=0):
super().__init__()
weight = torch.randn(in_channel, out_channel, kernel_size, kernel_size)
bias = torch.zeros(out_channel)
fan_in = in_channel * kernel_size * kernel_size
self.multiplier = sqrt(2 / fan_in)
self.weight = nn.Parameter(weight)
self.bias = nn.Parameter(bias)
self.pad = padding
def forward(self, input):
weight = F.pad(self.weight * self.multiplier, [1, 1, 1, 1])
weight = (weight[:, :, 1:, 1:] + weight[:, :, :-1, 1:] + weight[:,
:, 1:, :-1] + weight[:, :, :-1, :-1]) / 4
out = F.conv_transpose2d(input, weight, self.bias, stride=2,
padding=self.pad)
return out
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 torch.nn as nn
from math import sqrt
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_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl.
constexpr):
xnumel = 400
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x1 = xindex // 5 % 5
x0 = xindex % 5
x2 = xindex // 25
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 = x0
tmp6 = tmp5 >= tmp1
tmp7 = tmp5 < tmp3
tmp8 = tmp2 & tmp4
tmp9 = tmp8 & tmp6
tmp10 = tmp9 & tmp7
tmp11 = tl.load(in_ptr0 + (x0 + 4 * x1 + 16 * x2), tmp10 & xmask, other=0.0
)
tmp12 = 0.1767766952966369
tmp13 = tmp11 * tmp12
tmp14 = tl.full(tmp13.shape, 0.0, tmp13.dtype)
tmp15 = tl.where(tmp10, tmp13, tmp14)
tmp16 = -1 + x1
tmp17 = tmp16 >= tmp1
tmp18 = tmp16 < tmp3
tmp19 = tmp17 & tmp18
tmp20 = tmp19 & tmp6
tmp21 = tmp20 & tmp7
tmp22 = tl.load(in_ptr0 + (-4 + x0 + 4 * x1 + 16 * x2), tmp21 & xmask,
other=0.0)
tmp23 = tmp22 * tmp12
tmp24 = tl.full(tmp23.shape, 0.0, tmp23.dtype)
tmp25 = tl.where(tmp21, tmp23, tmp24)
tmp26 = tmp15 + tmp25
tmp27 = -1 + x0
tmp28 = tmp27 >= tmp1
tmp29 = tmp27 < tmp3
tmp30 = tmp8 & tmp28
tmp31 = tmp30 & tmp29
tmp32 = tl.load(in_ptr0 + (-1 + x0 + 4 * x1 + 16 * x2), tmp31 & xmask,
other=0.0)
tmp33 = tmp32 * tmp12
tmp34 = tl.full(tmp33.shape, 0.0, tmp33.dtype)
tmp35 = tl.where(tmp31, tmp33, tmp34)
tmp36 = tmp26 + tmp35
tmp37 = tmp19 & tmp28
tmp38 = tmp37 & tmp29
tmp39 = tl.load(in_ptr0 + (-5 + x0 + 4 * x1 + 16 * x2), tmp38 & xmask,
other=0.0)
tmp40 = tmp39 * tmp12
tmp41 = tl.full(tmp40.shape, 0.0, tmp40.dtype)
tmp42 = tl.where(tmp38, tmp40, tmp41)
tmp43 = tmp36 + tmp42
tmp44 = 0.25
tmp45 = tmp43 * tmp44
tl.store(in_out_ptr0 + x4, tmp45, xmask)
@triton.jit
def triton_poi_fused_convolution_1(in_out_ptr0, in_ptr0, xnumel, XBLOCK: tl
.constexpr):
xnumel = 1936
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x3 = xindex
x1 = xindex // 121 % 4
tmp0 = tl.load(in_out_ptr0 + x3, xmask)
tmp1 = tl.load(in_ptr0 + x1, xmask, eviction_policy='evict_last')
tmp2 = tmp0 + tmp1
tl.store(in_out_ptr0 + x3, tmp2, xmask)
def call(args):
primals_1, primals_2, primals_3 = args
args.clear()
assert_size_stride(primals_1, (4, 4, 4, 4), (64, 16, 4, 1))
assert_size_stride(primals_2, (4,), (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, 5, 5), (100, 25, 5, 1), torch.float32)
buf1 = buf0
del buf0
get_raw_stream(0)
triton_poi_fused_add_div_0[grid(400)](buf1, primals_1, 400, XBLOCK=
128, num_warps=4, num_stages=1)
del primals_1
buf2 = extern_kernels.convolution(primals_3, buf1, stride=(2, 2),
padding=(0, 0), dilation=(1, 1), transposed=True,
output_padding=(0, 0), groups=1, bias=None)
assert_size_stride(buf2, (4, 4, 11, 11), (484, 121, 11, 1))
buf3 = buf2
del buf2
triton_poi_fused_convolution_1[grid(1936)](buf3, primals_2, 1936,
XBLOCK=128, num_warps=4, num_stages=1)
del primals_2
return buf3, primals_3, buf1
class FusedUpsampleNew(nn.Module):
def __init__(self, in_channel, out_channel, kernel_size, padding=0):
super().__init__()
weight = torch.randn(in_channel, out_channel, kernel_size, kernel_size)
bias = torch.zeros(out_channel)
fan_in = in_channel * kernel_size * kernel_size
self.multiplier = sqrt(2 / fan_in)
self.weight = nn.Parameter(weight)
self.bias = nn.Parameter(bias)
self.pad = 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]
|
KwonGihyun/DiagonalGAN
|
FusedUpsample
| false
| 8,450
|
[
"MIT"
] | 13
|
9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
https://github.com/KwonGihyun/DiagonalGAN/tree/9e401c00e741d700f85df2c715ee11c1e66e1d1c
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.