提交 01c00b07 编写于 作者: P peizhilin

fix test issues on windows

test=develop
上级 dc8eca82
...@@ -57,46 +57,43 @@ int main() ...@@ -57,46 +57,43 @@ int main()
return 0; return 0;
}" SSE3_FOUND) }" SSE3_FOUND)
# disable AVX by default on windows # Check AVX
if(NOT WIN32) set(CMAKE_REQUIRED_FLAGS ${AVX_FLAG})
# Check AVX set(AVX_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
set(CMAKE_REQUIRED_FLAGS ${AVX_FLAG}) CHECK_CXX_SOURCE_RUNS("
set(AVX_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) #include <immintrin.h>
CHECK_CXX_SOURCE_RUNS(" int main()
#include <immintrin.h> {
int main() __m256 a = _mm256_set_ps (-1.0f, 2.0f, -3.0f, 4.0f, -1.0f, 2.0f, -3.0f, 4.0f);
{ __m256 b = _mm256_set_ps (1.0f, 2.0f, 3.0f, 4.0f, 1.0f, 2.0f, 3.0f, 4.0f);
__m256 a = _mm256_set_ps (-1.0f, 2.0f, -3.0f, 4.0f, -1.0f, 2.0f, -3.0f, 4.0f); __m256 result = _mm256_add_ps (a, b);
__m256 b = _mm256_set_ps (1.0f, 2.0f, 3.0f, 4.0f, 1.0f, 2.0f, 3.0f, 4.0f); return 0;
__m256 result = _mm256_add_ps (a, b); }" AVX_FOUND)
return 0;
}" AVX_FOUND)
# Check AVX 2 # Check AVX 2
set(CMAKE_REQUIRED_FLAGS ${AVX2_FLAG}) set(CMAKE_REQUIRED_FLAGS ${AVX2_FLAG})
set(AVX2_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) set(AVX2_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS(" CHECK_CXX_SOURCE_RUNS("
#include <immintrin.h> #include <immintrin.h>
int main() int main()
{ {
__m256i a = _mm256_set_epi32 (-1, 2, -3, 4, -1, 2, -3, 4); __m256i a = _mm256_set_epi32 (-1, 2, -3, 4, -1, 2, -3, 4);
__m256i result = _mm256_abs_epi32 (a); __m256i result = _mm256_abs_epi32 (a);
return 0; return 0;
}" AVX2_FOUND) }" AVX2_FOUND)
# Check AVX512F # Check AVX512F
set(CMAKE_REQUIRED_FLAGS ${AVX512F_FLAG}) set(CMAKE_REQUIRED_FLAGS ${AVX512F_FLAG})
set(AVX512F_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE) set(AVX512F_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS(" CHECK_CXX_SOURCE_RUNS("
#include <immintrin.h> #include <immintrin.h>
int main() int main()
{ {
__m512i a = _mm512_set_epi32 (-1, 2, -3, 4, -1, 2, -3, 4, __m512i a = _mm512_set_epi32 (-1, 2, -3, 4, -1, 2, -3, 4,
13, -5, 6, -7, 9, 2, -6, 3); 13, -5, 6, -7, 9, 2, -6, 3);
__m512i result = _mm512_abs_epi32 (a); __m512i result = _mm512_abs_epi32 (a);
return 0; return 0;
}" AVX512F_FOUND) }" AVX512F_FOUND)
endif(NOT WIN32)
set(CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS_RETAINED}) set(CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS_RETAINED})
mark_as_advanced(MMX_FOUND SSE2_FOUND SSE3_FOUND AVX_FOUND AVX2_FOUND AVX512F_FOUND) mark_as_advanced(MMX_FOUND SSE2_FOUND SSE3_FOUND AVX_FOUND AVX2_FOUND AVX512F_FOUND)
...@@ -7,27 +7,17 @@ function(windows_symbolic TARGET) ...@@ -7,27 +7,17 @@ function(windows_symbolic TARGET)
cmake_parse_arguments(windows_symbolic "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) cmake_parse_arguments(windows_symbolic "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
set(final_path ${CMAKE_CURRENT_SOURCE_DIR}/${windows_symbolic_PATH}) set(final_path ${CMAKE_CURRENT_SOURCE_DIR}/${windows_symbolic_PATH})
foreach(src ${windows_symbolic_SRCS}) foreach(src ${windows_symbolic_SRCS})
get_filename_component(src ${src} NAME_WE) get_filename_component(src ${src} NAME_WE)
if (NOT EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${src}.cc OR NOT EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${src}.cu) if (NOT EXISTS ${final_path}/${src}.cc OR NOT EXISTS ${final_path}/${src}.cu)
message(FATAL " ${src}.cc and ${src}.cu must exsits, and ${src}.cu must be symbolic file.") message(FATAL " ${src}.cc and ${src}.cu must exsits, and ${src}.cu must be symbolic file.")
endif() endif()
#only copy the xx.cu to.xx.cu when the content are modified file(GENERATE OUTPUT ${final_path}/.${src}.cu INPUT ${final_path}/${src}.cc)
set(copy_flag 1)
if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/.${src}.cu) add_custom_command(OUTPUT ${final_path}/.${src}.cu
file(READ ${CMAKE_CURRENT_SOURCE_DIR}/${src}.cc SOURCE_STR) COMMAND ${CMAKE_COMMAND} -E copy_if_different "${final_path}/${src}.cc" "${final_path}/.${src}.cu"
file(READ ${CMAKE_CURRENT_SOURCE_DIR}/.${src}.cu TARGET_STR) COMMENT "create hidden file of ${src}.cu")
if (SOURCE_STR STREQUAL TARGET_STR) add_custom_target(${TARGET} ALL DEPENDS .${src}.cu)
set(copy_flag 0)
endif()
endif()
if (copy_flag)
add_custom_command(OUTPUT .${src}.cu
COMMAND ${CMAKE_COMMAND} -E remove ${CMAKE_CURRENT_SOURCE_DIR}/.${src}.cu
COMMAND ${CMAKE_COMMAND} -E copy "${CMAKE_CURRENT_SOURCE_DIR}/${src}.cc" "${CMAKE_CURRENT_SOURCE_DIR}/.${src}.cu"
COMMENT "create hidden file of ${src}.cu")
endif(copy_flag)
add_custom_target(${TARGET} ALL DEPENDS .${src}.cu)
endforeach() endforeach()
endfunction() endfunction()
......
...@@ -215,8 +215,8 @@ class Vector { ...@@ -215,8 +215,8 @@ class Vector {
auto stream = dev_ctx->stream(); auto stream = dev_ctx->stream();
void *src = gpu_->ptr(); void *src = gpu_->ptr();
void *dst = cpu_.data(); void *dst = cpu_.data();
memory::Copy(platform::CPUPlace(), dst, CUDAPlace().get(), src, paddle::memory::Copy(platform::CPUPlace(), dst, CUDAPlace().get(), src,
gpu_->size(), stream); gpu_->size(), stream);
dev_ctx->Wait(); dev_ctx->Wait();
} }
...@@ -261,8 +261,8 @@ class Vector { ...@@ -261,8 +261,8 @@ class Vector {
auto *dev_ctx = static_cast<platform::CUDADeviceContext *>( auto *dev_ctx = static_cast<platform::CUDADeviceContext *>(
platform::DeviceContextPool::Instance().Get(place)); platform::DeviceContextPool::Instance().Get(place));
auto stream = dev_ctx->stream(); auto stream = dev_ctx->stream();
memory::Copy(CUDAPlace().get(), dst, platform::CPUPlace(), src, paddle::memory::Copy(CUDAPlace().get(), dst, platform::CPUPlace(), src,
gpu_->size(), stream); gpu_->size(), stream);
} }
void ImmutableCPU() const { void ImmutableCPU() const {
...@@ -284,7 +284,7 @@ class Vector { ...@@ -284,7 +284,7 @@ class Vector {
bool IsInCPU() const { return flag_ & kDataInCPU; } bool IsInCPU() const { return flag_ & kDataInCPU; }
mutable std::vector<T> cpu_; mutable std::vector<T> cpu_;
mutable memory::AllocationPtr gpu_; mutable paddle::memory::AllocationPtr gpu_;
mutable int flag_; mutable int flag_;
mutable std::mutex mtx_; mutable std::mutex mtx_;
......
...@@ -23,7 +23,8 @@ limitations under the License. */ ...@@ -23,7 +23,8 @@ limitations under the License. */
#include <unordered_map> #include <unordered_map>
#include <unordered_set> #include <unordered_set>
#include "glog/logging.h" // For VLOG() #define GLOG_NO_ABBREVIATED_SEVERITIES // msvc conflict logging with windows.h
#include "glog/logging.h" // For VLOG()
#include "paddle/fluid/framework/attribute.h" #include "paddle/fluid/framework/attribute.h"
#include "paddle/fluid/framework/details/op_registry.h" #include "paddle/fluid/framework/details/op_registry.h"
#include "paddle/fluid/framework/framework.pb.h" #include "paddle/fluid/framework/framework.pb.h"
......
...@@ -3,14 +3,16 @@ set(INFERENCE_DEMO_INSTALL_DIR "${THIRD_PARTY_PATH}/inference_demo" CACHE STRING ...@@ -3,14 +3,16 @@ set(INFERENCE_DEMO_INSTALL_DIR "${THIRD_PARTY_PATH}/inference_demo" CACHE STRING
"A path setting inference demo download directories.") "A path setting inference demo download directories.")
function (inference_download install_dir url filename) function (inference_download install_dir url filename)
message(STATUS "Download inference test stuff from ${url}/${filename}") message(STATUS "Download inference test stuff from ${url}/${filename}")
execute_process(COMMAND bash -c "mkdir -p ${install_dir}") file(DOWNLOAD "${url}/${filename}" "${install_dir}/${filename}")
execute_process(COMMAND bash -c "cd ${install_dir} && wget -q ${url}/${filename}")
message(STATUS "finish downloading ${filename}") message(STATUS "finish downloading ${filename}")
endfunction() endfunction()
function (inference_download_and_uncompress install_dir url filename) function (inference_download_and_uncompress install_dir url filename)
inference_download(${install_dir} ${url} ${filename}) inference_download(${install_dir} ${url} ${filename})
execute_process(COMMAND bash -c "cd ${install_dir} && tar xzf ${filename}") execute_process(
COMMAND ${CMAKE_COMMAND} -E tar xzf ${install_dir}/${filename}
WORKING_DIRECTORY ${install_dir}
)
endfunction() endfunction()
set(WORD2VEC_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/word2vec") set(WORD2VEC_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/word2vec")
......
...@@ -46,7 +46,7 @@ endif() ...@@ -46,7 +46,7 @@ endif()
register_operators(EXCLUDES py_func_op warpctc_op conv_fusion_op DEPS ${OP_HEADER_DEPS} ${OP_PREFETCH_DEPS}) register_operators(EXCLUDES py_func_op warpctc_op conv_fusion_op DEPS ${OP_HEADER_DEPS} ${OP_PREFETCH_DEPS})
# warpctc_op needs cudnn 7 above # warpctc_op needs cudnn 7 above
if (WITH_GPU AND NOT WIN32) if (WITH_GPU)
if (${CUDNN_MAJOR_VERSION} VERSION_LESS 7) if (${CUDNN_MAJOR_VERSION} VERSION_LESS 7)
op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale SRCS warpctc_op.cc warpctc_op.cu.cc) op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale SRCS warpctc_op.cc warpctc_op.cu.cc)
else() else()
......
...@@ -13,6 +13,8 @@ See the License for the specific language governing permissions and ...@@ -13,6 +13,8 @@ See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#pragma once #pragma once
#include <array>
#include "paddle/fluid/framework/eigen.h" #include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/framework/op_registry.h" #include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/operator.h" #include "paddle/fluid/framework/operator.h"
......
...@@ -104,15 +104,19 @@ class HuberLossGradKernel : public framework::OpKernel<T> { ...@@ -104,15 +104,19 @@ class HuberLossGradKernel : public framework::OpKernel<T> {
if (out0) { if (out0) {
out0->mutable_data<T>(context.GetPlace()); out0->mutable_data<T>(context.GetPlace());
auto x_grad = EigenVector<T>::Flatten(*out0); auto x_grad = EigenVector<T>::Flatten(*out0);
// MSVC not treat it well when partial template arguments were specified
x_grad.device(place) = x_grad.device(place) =
out_grad * residual.unaryExpr(HuberLossBackward<T>(delta, -1.0)); out_grad *
residual.unaryExpr(HuberLossBackward<T>(delta, static_cast<T>(-1.0)));
} }
if (out1) { if (out1) {
out1->mutable_data<T>(context.GetPlace()); out1->mutable_data<T>(context.GetPlace());
auto y_grad = EigenVector<T>::Flatten(*out1); auto y_grad = EigenVector<T>::Flatten(*out1);
// MSVC not treat it well when partial template arguments were specified
y_grad.device(place) = y_grad.device(place) =
out_grad * residual.unaryExpr(HuberLossBackward<T>(delta, 1.0)); out_grad *
residual.unaryExpr(HuberLossBackward<T>(delta, static_cast<T>(1.0)));
} }
} }
}; };
......
...@@ -12,6 +12,7 @@ limitations under the License. */ ...@@ -12,6 +12,7 @@ limitations under the License. */
#include <vector> #include <vector>
#define GLOG_NO_ABBREVIATED_SEVERITIES // msvc conflict logging with windows.h
#include "gtest/gtest.h" #include "gtest/gtest.h"
#include "paddle/fluid/framework/lod_tensor.h" #include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/platform/init.h" #include "paddle/fluid/platform/init.h"
......
...@@ -11,6 +11,7 @@ limitations under the License. */ ...@@ -11,6 +11,7 @@ limitations under the License. */
#include "paddle/fluid/platform/float16.h" #include "paddle/fluid/platform/float16.h"
#define GLOG_NO_ABBREVIATED_SEVERITIES // msvc conflict logging with windows.h
#include <glog/logging.h> #include <glog/logging.h>
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include <bitset> #include <bitset>
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册