diff --git a/paddle/platform/CMakeLists.txt b/paddle/platform/CMakeLists.txt index e93592cc4cbacca9ee0202a24b3c813cc23c2202..358d14f4555e1d046c8e7b91e23d54fb504926e5 100644 --- a/paddle/platform/CMakeLists.txt +++ b/paddle/platform/CMakeLists.txt @@ -5,4 +5,11 @@ nv_test(cuda_test SRCS cuda_test.cu) cc_library(place SRCS place.cc) cc_test(place_test SRCS place_test.cc DEPS place glog gflags) -nv_test(device_context_test SRCS device_context_test.cc DEPS dynload_cuda dynamic_loader eigen3 place) +IF(WITH_GPU) + set(GPU_CTX_DEPS dynload_cuda dynamic_loader) +ELSE() + set(GPU_CTX_DEPS) +ENDIF() + +cc_library(device_context SRCS device_context.cc DEPS place eigen3 ${GPU_CTX_DEPS}) +nv_test(device_context_test SRCS device_context_test.cc DEPS device_context glog gflags) diff --git a/paddle/platform/cuda_device_context.h b/paddle/platform/cuda_device_context.h deleted file mode 100644 index 8a9d15e8a8410c7f268ad32d798c408bfbe949bc..0000000000000000000000000000000000000000 --- a/paddle/platform/cuda_device_context.h +++ /dev/null @@ -1,159 +0,0 @@ -/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - -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. */ - -#pragma once - -#include "paddle/framework/enforce.h" -#include "paddle/platform/cuda.h" -#include "paddle/platform/dynload/cublas.h" -#include "paddle/platform/dynload/cudnn.h" -#include "paddle/platform/dynload/curand.h" -#define EIGEN_USE_GPU -#include "paddle/platform/device_context.h" -#include "paddle/platform/place.h" -#include "unsupported/Eigen/CXX11/Tensor" - -namespace paddle { -namespace platform { - -class GPUPlaceGuard { - public: - explicit GPUPlaceGuard(GPUPlace new_place) : previous_(GetCurrentDeviceId()) { - if (previous_ != new_place) { - paddle::platform::SetDeviceId(new_place.device); - } - } - - ~GPUPlaceGuard() { paddle::platform::SetDeviceId(previous_.device); } - - private: - GPUPlace previous_; -}; - -class CUDADeviceContext : public DeviceContext { - public: - explicit CUDADeviceContext(const GPUPlace gpu_place) : gpu_place_(gpu_place) { - GPUPlaceGuard guard(gpu_place_); - paddle::platform::throw_on_error(cudaStreamCreate(&stream_), - "cudaStreamCreate failed"); - eigen_stream_ = new Eigen::CudaStreamDevice(&stream_); - eigen_device_ = new Eigen::GpuDevice(eigen_stream_); - } - - Place GetPlace() const override { - Place retv = GPUPlace(); - return retv; - } - - void Wait() { - paddle::platform::throw_on_error(cudaStreamSynchronize(stream_), - "cudaStreamSynchronize failed"); - } - - cudaStream_t stream() { return stream_; } - - Eigen::GpuDevice eigen_device() { return *eigen_device_; } - - cublasHandle_t cublas_handle() { - if (!blas_handle_) { - GPUPlaceGuard guard(gpu_place_); - PADDLE_ENFORCE(paddle::platform::dynload::cublasCreate(&blas_handle_) == - CUBLAS_STATUS_SUCCESS, - "cublasCreate failed"); - PADDLE_ENFORCE(paddle::platform::dynload::cublasSetStream( - blas_handle_, stream_) == CUBLAS_STATUS_SUCCESS, - "cublasSetStream failed"); - } - return blas_handle_; - } - - cudnnHandle_t cudnn_handle() { - if (!dnn_handle_) { - GPUPlaceGuard guard(gpu_place_); - PADDLE_ENFORCE(paddle::platform::dynload::cudnnCreate(&dnn_handle_) == - CUDNN_STATUS_SUCCESS, - "cudnnCreate failed"); - PADDLE_ENFORCE(paddle::platform::dynload::cudnnSetStream( - dnn_handle_, stream_) == CUDNN_STATUS_SUCCESS, - "cudnnSetStream failed"); - } - return dnn_handle_; - } - - curandGenerator_t curand_generator() { - if (!rand_generator_) { - GPUPlaceGuard guard(gpu_place_); - PADDLE_ENFORCE(paddle::platform::dynload::curandCreateGenerator( - &rand_generator_, CURAND_RNG_PSEUDO_DEFAULT) == - CURAND_STATUS_SUCCESS, - "curandCreateGenerator failed"); - PADDLE_ENFORCE( - paddle::platform::dynload::curandSetPseudoRandomGeneratorSeed( - rand_generator_, random_seed_) == CURAND_STATUS_SUCCESS, - "curandSetPseudoRandomGeneratorSeed failed"); - PADDLE_ENFORCE(paddle::platform::dynload::curandSetStream( - rand_generator_, stream_) == CURAND_STATUS_SUCCESS, - "curandSetStream failed"); - } - return rand_generator_; - } - - ~CUDADeviceContext() { - Wait(); - if (blas_handle_) { - PADDLE_ENFORCE(paddle::platform::dynload::cublasDestroy(blas_handle_) == - CUBLAS_STATUS_SUCCESS, - "cublasDestroy failed"); - } - - if (dnn_handle_) { - PADDLE_ENFORCE(paddle::platform::dynload::cudnnDestroy(dnn_handle_) == - CUDNN_STATUS_SUCCESS, - "cudnnDestroy failed"); - } - - if (rand_generator_) { - PADDLE_ENFORCE(paddle::platform::dynload::curandDestroyGenerator( - rand_generator_) == CURAND_STATUS_SUCCESS, - "curandDestroyGenerator failed"); - } - - delete eigen_stream_; - delete eigen_device_; - - paddle::platform::throw_on_error(cudaStreamDestroy(stream_), - "cudaStreamDestroy failed"); - } - - private: - GPUPlace gpu_place_; - cudaStream_t stream_; - - Eigen::CudaStreamDevice* eigen_stream_; - Eigen::GpuDevice* eigen_device_; - - cublasHandle_t blas_handle_{nullptr}; - - cudnnHandle_t dnn_handle_{nullptr}; - - int random_seed_; - curandGenerator_t rand_generator_{nullptr}; -}; - -template <> -Eigen::GpuDevice DeviceContext::get_eigen_device() { - return dynamic_cast(this)->eigen_device(); -} -} // namespace platform -} // namespace paddle diff --git a/paddle/platform/device_context.cc b/paddle/platform/device_context.cc new file mode 100644 index 0000000000000000000000000000000000000000..8d800ec49987df36daf8a5c1a961bea49786ad46 --- /dev/null +++ b/paddle/platform/device_context.cc @@ -0,0 +1,24 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. +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. */ + +#include "paddle/platform/device_context.h" + +namespace paddle { +namespace platform { +namespace dynload { +namespace dummy { +// Make DeviceContext A library. +int DUMMY_VAR_FOR_DEV_CTX = 0; + +} // namespace dummy +} // namespace dynload +} // namespace platform +} // namespace paddle \ No newline at end of file diff --git a/paddle/platform/device_context.h b/paddle/platform/device_context.h index d2f7cf62166c287e987d28c6ad1a191dd0ae9a84..5b4b5e29994b9a5371b5ed55b12ef89748d0658b 100644 --- a/paddle/platform/device_context.h +++ b/paddle/platform/device_context.h @@ -1,11 +1,8 @@ /* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - 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. @@ -13,9 +10,17 @@ See the License for the specific language governing permissions and limitations under the License. */ #pragma once + #include "paddle/framework/enforce.h" -#include "paddle/platform/place.h" -#include "unsupported/Eigen/CXX11/Tensor" +#ifndef PADDLE_ONLY_CPU +#include "paddle/platform/cuda.h" +#include "paddle/platform/dynload/cublas.h" +#include "paddle/platform/dynload/cudnn.h" +#include "paddle/platform/dynload/curand.h" +#define EIGEN_USE_GPU +#endif +#include +#include namespace paddle { namespace platform { @@ -23,11 +28,10 @@ namespace platform { class DeviceContext { public: virtual ~DeviceContext() {} + virtual Place GetPlace() const = 0; template inline DeviceType get_eigen_device(); - - virtual Place GetPlace() const = 0; }; class CPUDeviceContext : public DeviceContext { @@ -52,5 +56,138 @@ template <> Eigen::DefaultDevice DeviceContext::get_eigen_device() { return dynamic_cast(this)->eigen_device(); } + +#ifndef PADDLE_ONLY_CPU + +class GPUPlaceGuard { + public: + explicit GPUPlaceGuard(GPUPlace new_place) : previous_(GetCurrentDeviceId()) { + if (previous_ != new_place) { + paddle::platform::SetDeviceId(new_place.device); + } + } + + ~GPUPlaceGuard() { paddle::platform::SetDeviceId(previous_.device); } + + private: + GPUPlace previous_; +}; + +class CUDADeviceContext : public DeviceContext { + public: + explicit CUDADeviceContext(const GPUPlace gpu_place) : gpu_place_(gpu_place) { + GPUPlaceGuard guard(gpu_place_); + paddle::platform::throw_on_error(cudaStreamCreate(&stream_), + "cudaStreamCreate failed"); + eigen_stream_ = new Eigen::CudaStreamDevice(&stream_); + eigen_device_ = new Eigen::GpuDevice(eigen_stream_); + } + + Place GetPlace() const override { + Place retv = GPUPlace(); + return retv; + } + + void Wait() { + paddle::platform::throw_on_error(cudaStreamSynchronize(stream_), + "cudaStreamSynchronize failed"); + } + + cudaStream_t stream() { return stream_; } + + Eigen::GpuDevice eigen_device() { return *eigen_device_; } + + cublasHandle_t cublas_handle() { + if (!blas_handle_) { + GPUPlaceGuard guard(gpu_place_); + PADDLE_ENFORCE(paddle::platform::dynload::cublasCreate(&blas_handle_) == + CUBLAS_STATUS_SUCCESS, + "cublasCreate failed"); + PADDLE_ENFORCE(paddle::platform::dynload::cublasSetStream( + blas_handle_, stream_) == CUBLAS_STATUS_SUCCESS, + "cublasSetStream failed"); + } + return blas_handle_; + } + + cudnnHandle_t cudnn_handle() { + if (!dnn_handle_) { + GPUPlaceGuard guard(gpu_place_); + PADDLE_ENFORCE(paddle::platform::dynload::cudnnCreate(&dnn_handle_) == + CUDNN_STATUS_SUCCESS, + "cudnnCreate failed"); + PADDLE_ENFORCE(paddle::platform::dynload::cudnnSetStream( + dnn_handle_, stream_) == CUDNN_STATUS_SUCCESS, + "cudnnSetStream failed"); + } + return dnn_handle_; + } + + curandGenerator_t curand_generator() { + if (!rand_generator_) { + GPUPlaceGuard guard(gpu_place_); + PADDLE_ENFORCE(paddle::platform::dynload::curandCreateGenerator( + &rand_generator_, CURAND_RNG_PSEUDO_DEFAULT) == + CURAND_STATUS_SUCCESS, + "curandCreateGenerator failed"); + PADDLE_ENFORCE( + paddle::platform::dynload::curandSetPseudoRandomGeneratorSeed( + rand_generator_, random_seed_) == CURAND_STATUS_SUCCESS, + "curandSetPseudoRandomGeneratorSeed failed"); + PADDLE_ENFORCE(paddle::platform::dynload::curandSetStream( + rand_generator_, stream_) == CURAND_STATUS_SUCCESS, + "curandSetStream failed"); + } + return rand_generator_; + } + + ~CUDADeviceContext() { + Wait(); + if (blas_handle_) { + PADDLE_ENFORCE(paddle::platform::dynload::cublasDestroy(blas_handle_) == + CUBLAS_STATUS_SUCCESS, + "cublasDestroy failed"); + } + + if (dnn_handle_) { + PADDLE_ENFORCE(paddle::platform::dynload::cudnnDestroy(dnn_handle_) == + CUDNN_STATUS_SUCCESS, + "cudnnDestroy failed"); + } + + if (rand_generator_) { + PADDLE_ENFORCE(paddle::platform::dynload::curandDestroyGenerator( + rand_generator_) == CURAND_STATUS_SUCCESS, + "curandDestroyGenerator failed"); + } + + delete eigen_stream_; + delete eigen_device_; + + paddle::platform::throw_on_error(cudaStreamDestroy(stream_), + "cudaStreamDestroy failed"); + } + + private: + GPUPlace gpu_place_; + cudaStream_t stream_; + + Eigen::CudaStreamDevice* eigen_stream_; + Eigen::GpuDevice* eigen_device_; + + cublasHandle_t blas_handle_{nullptr}; + + cudnnHandle_t dnn_handle_{nullptr}; + + int random_seed_; + curandGenerator_t rand_generator_{nullptr}; +}; + +template <> +Eigen::GpuDevice DeviceContext::get_eigen_device() { + return dynamic_cast(this)->eigen_device(); +} +#endif + } // namespace platform } // namespace paddle diff --git a/paddle/platform/device_context_test.cc b/paddle/platform/device_context_test.cc index abaaaececf5b0d623be90d7c9e67b5b6da05f803..913e3c0aa96e150cd4f6192d2e6f6fc2227b4311 100644 --- a/paddle/platform/device_context_test.cc +++ b/paddle/platform/device_context_test.cc @@ -12,8 +12,8 @@ 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. */ +#include "paddle/platform/device_context.h" #include "gtest/gtest.h" -#include "paddle/platform/cuda_device_context.h" using DEVICE_GPU = Eigen::GpuDevice; TEST(Device, Init) {