experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Split scan ops GPU code into multiple files

Browse Source 
This file was a bottleneck during compilation, often taking many minutes to compile. In local testing this change reduces the wall-clock build time for the scan ops GPU kernels from 107s to 96s.

PiperOrigin-RevId: 228304727
This commit is contained in:
James Keeling 2019-01-08 03:28:39 -08:00 committed by TensorFlower Gardener
parent 0ef4b19044
commit 3436665db2
5 changed files with 107 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
tensorflow/core/kernels/BUILD
View File

@ -3279,7 +3279,15 @@ tf_kernel_library(
tf_kernel_library(
name = "scan_ops",
prefix = "scan_ops",
srcs = ["scan_ops.cc"],
hdrs = ["scan_ops.h"],
gpu_srcs = [
"scan_ops.h",
"scan_ops_gpu.h",
"scan_ops_gpu_double.cu.cc",
"scan_ops_gpu_float.cu.cc",
"scan_ops_gpu_half.cu.cc",
],
deps = MATH_DEPS + if_cuda(["@cub_archive//:cub"]),
)

16
tensorflow/core/kernels/scan_ops_gpu.cu.cc → tensorflow/core/kernels/scan_ops_gpu.h
View File

@ -13,6 +13,9 @@ See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#ifndef TENSORFLOW_CORE_KERNELS_SCAN_OPS_GPU_H_
#define TENSORFLOW_CORE_KERNELS_SCAN_OPS_GPU_H_
#if GOOGLE_CUDA
#define EIGEN_USE_GPU
@ -290,17 +293,8 @@ struct Scan<GPUDevice, Eigen::internal::ProdReducer<T>, T> {
};
} // namespace functor
#define DEFINE(REDUCER, T) template struct functor::Scan<GPUDevice, REDUCER, T>;
#define DEFINE_FOR_ALL_REDUCERS(T) \
DEFINE(Eigen::internal::SumReducer<T>, T); \
DEFINE(Eigen::internal::ProdReducer<T>, T);
TF_CALL_GPU_NUMBER_TYPES(DEFINE_FOR_ALL_REDUCERS);
#undef DEFINE_FOR_ALL_REDUCERS
#undef DEFINE
} // end namespace tensorflow
#endif // GOOGLE_CUDA
#endif // TENSORFLOW_CORE_KERNELS_SCAN_OPS_GPU_H_

31
tensorflow/core/kernels/scan_ops_gpu_double.cu.cc Normal file
View File

@ -0,0 +1,31 @@
/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#if GOOGLE_CUDA
#define EIGEN_USE_GPU
#include "tensorflow/core/kernels/scan_ops.h"
#include "tensorflow/core/kernels/scan_ops_gpu.h"
namespace tensorflow {
using Eigen::GpuDevice;
template struct functor::Scan<GpuDevice, Eigen::internal::SumReducer<double>,
double>;
template struct functor::Scan<GpuDevice, Eigen::internal::ProdReducer<double>,
double>;
} // namespace tensorflow
#endif // GOOGLE_CUDA

31
tensorflow/core/kernels/scan_ops_gpu_float.cu.cc Normal file
View File

@ -0,0 +1,31 @@
/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#if GOOGLE_CUDA
#define EIGEN_USE_GPU
#include "tensorflow/core/kernels/scan_ops.h"
#include "tensorflow/core/kernels/scan_ops_gpu.h"
namespace tensorflow {
using Eigen::GpuDevice;
template struct functor::Scan<GpuDevice, Eigen::internal::SumReducer<float>,
float>;
template struct functor::Scan<GpuDevice, Eigen::internal::ProdReducer<float>,
float>;
} // namespace tensorflow
#endif // GOOGLE_CUDA

31
tensorflow/core/kernels/scan_ops_gpu_half.cu.cc Normal file
View File

@ -0,0 +1,31 @@
/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#if GOOGLE_CUDA
#define EIGEN_USE_GPU
#include "tensorflow/core/kernels/scan_ops.h"
#include "tensorflow/core/kernels/scan_ops_gpu.h"
namespace tensorflow {
using Eigen::GpuDevice;
template struct functor::Scan<
GpuDevice, Eigen::internal::SumReducer<Eigen::half>, Eigen::half>;
template struct functor::Scan<
GpuDevice, Eigen::internal::ProdReducer<Eigen::half>, Eigen::half>;
} // namespace tensorflow
#endif // GOOGLE_CUDA