Merge pull request #14320 from wopeizl/windows/online

Windows/online
panyx0718-patch-1
dzhwinter 6 years ago committed by GitHub
commit d3aed98d86
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

@ -26,6 +26,11 @@ message(STATUS "C compiler: ${CMAKE_C_COMPILER}, version: "
"${CMAKE_C_COMPILER_ID} ${CMAKE_C_COMPILER_VERSION}")
if(WIN32)
set(CMAKE_STATIC_LIBRARY_PREFIX lib)
add_definitions("/DGOOGLE_GLOG_DLL_DECL=")
set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} /bigobj /MTd")
set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /bigobj /MT")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /bigobj /MTd")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /bigobj /MT")
endif(WIN32)
if(NOT CMAKE_CROSSCOMPILING)

@ -157,6 +157,9 @@ list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY})
if(NOT WITH_DSO)
# TODO(panyx0718): CUPTI only allows DSO?
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUPTI_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY})
if(WIN32)
set_property(GLOBAL PROPERTY CUDA_MODULES ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY})
endif(WIN32)
endif(NOT WITH_DSO)
# setting nvcc arch flags
@ -196,10 +199,12 @@ elseif(CMAKE_BUILD_TYPE STREQUAL "MinSizeRel")
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELEASE})
endif()
else(NOT WIN32)
if(CMAKE_BUILD_TYPE STREQUAL "Release")
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
list(APPEND CUDA_NVCC_FLAGS "-g -G")
elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
list(APPEND CUDA_NVCC_FLAGS "-O3 -DNDEBUG")
else()
message(FATAL "Windows only support Release build now. Please set visual studio build type to Release, x64 build.")
message(FATAL "Windows only support Release or Debug build now. Please set visual studio build type to Release/Debug, x64 build.")
endif()
endif(NOT WIN32)

@ -2,7 +2,12 @@ if(NOT WITH_GPU)
return()
endif()
set(CUDNN_ROOT "/usr" CACHE PATH "CUDNN ROOT")
if(WIN32)
set(CUDNN_ROOT ${CUDA_TOOLKIT_ROOT_DIR})
else(WIN32)
set(CUDNN_ROOT "/usr" CACHE PATH "CUDNN ROOT")
endif(WIN32)
find_path(CUDNN_INCLUDE_DIR cudnn.h
PATHS ${CUDNN_ROOT} ${CUDNN_ROOT}/include
$ENV{CUDNN_ROOT} $ENV{CUDNN_ROOT}/include ${CUDA_TOOLKIT_INCLUDE}

@ -28,34 +28,28 @@ if((NOT DEFINED BOOST_TAR) OR (NOT DEFINED BOOST_URL))
set(BOOST_TAR "boost_1_41_0" CACHE STRING "" FORCE)
set(BOOST_URL "http://paddlepaddledeps.cdn.bcebos.com/${BOOST_TAR}.tar.gz" CACHE STRING "" FORCE)
endif()
IF (WIN32)
MESSAGE(WARNING, "In windows, boost can not be downloaded automaticlly, please build it manually and put it at " ${THIRD_PARTY_PATH}install/boost)
else()
MESSAGE(STATUS "BOOST_TAR: ${BOOST_TAR}, BOOST_URL: ${BOOST_URL}")
ENDIF(WIN32)
MESSAGE(STATUS "BOOST_TAR: ${BOOST_TAR}, BOOST_URL: ${BOOST_URL}")
set(BOOST_SOURCES_DIR ${THIRD_PARTY_PATH}/boost)
set(BOOST_DOWNLOAD_DIR "${BOOST_SOURCES_DIR}/src/${BOOST_PROJECT}")
set(BOOST_INCLUDE_DIR "${BOOST_DOWNLOAD_DIR}/${BOOST_TAR}" CACHE PATH "boost include directory." FORCE)
set_directory_properties(PROPERTIES CLEAN_NO_CUSTOM 1)
set(BOOST_INCLUDE_DIR "${BOOST_DOWNLOAD_DIR}" CACHE PATH "boost include directory." FORCE)
set_directory_properties(PROPERTIES CLEAN_NO_CUSTOM 1)
include_directories(${BOOST_INCLUDE_DIR})
if (NOT WIN32)
ExternalProject_Add(
${BOOST_PROJECT}
${EXTERNAL_PROJECT_LOG_ARGS}
DOWNLOAD_DIR ${BOOST_DOWNLOAD_DIR}
DOWNLOAD_COMMAND wget --no-check-certificate ${BOOST_URL} -c -q -O ${BOOST_TAR}.tar.gz
&& tar zxf ${BOOST_TAR}.tar.gz
URL ${BOOST_URL}
DOWNLOAD_NO_PROGRESS 1
PREFIX ${BOOST_SOURCES_DIR}
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
UPDATE_COMMAND ""
)
endif(NOT WIN32)
)
if (${CMAKE_VERSION} VERSION_LESS "3.3.0" OR NOT WIN32)
set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/boost_dummy.c)

@ -35,7 +35,12 @@ ExternalProject_Add(
CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-DCMAKE_CXX_FLAGS_RELEASE=${CMAKE_CXX_FLAGS_RELEASE}
-DCMAKE_CXX_FLAGS_DEBUG=${CMAKE_CXX_FLAGS_DEBUG}
-DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-DCMAKE_C_FLAGS_DEBUG=${CMAKE_C_FLAGS_DEBUG}
-DCMAKE_C_FLAGS_RELEASE=${CMAKE_C_FLAGS_RELEASE}
-DBUILD_STATIC_LIBS=ON
-DCMAKE_INSTALL_PREFIX=${GFLAGS_INSTALL_DIR}
-DCMAKE_POSITION_INDEPENDENT_CODE=ON
-DBUILD_TESTING=OFF
@ -48,8 +53,8 @@ ExternalProject_Add(
IF(WIN32)
IF(NOT EXISTS "${GFLAGS_INSTALL_DIR}/lib/libgflags.lib")
add_custom_command(TARGET extern_gflags POST_BUILD
COMMAND cmake -E rename ${GFLAGS_INSTALL_DIR}/lib/gflags_static.lib ${GFLAGS_INSTALL_DIR}/lib/libgflags.lib
)
COMMAND cmake -E copy ${GFLAGS_INSTALL_DIR}/lib/gflags_static.lib ${GFLAGS_INSTALL_DIR}/lib/libgflags.lib
)
ENDIF()
ENDIF(WIN32)
ADD_LIBRARY(gflags STATIC IMPORTED GLOBAL)

@ -46,7 +46,11 @@ ExternalProject_Add(
CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
-DCMAKE_CXX_FLAGS_RELEASE=${CMAKE_CXX_FLAGS_RELEASE}
-DCMAKE_CXX_FLAGS_DEBUG=${CMAKE_CXX_FLAGS_DEBUG}
-DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
-DCMAKE_C_FLAGS_DEBUG=${CMAKE_C_FLAGS_DEBUG}
-DCMAKE_C_FLAGS_RELEASE=${CMAKE_C_FLAGS_RELEASE}
-DCMAKE_INSTALL_PREFIX=${GLOG_INSTALL_DIR}
-DCMAKE_INSTALL_LIBDIR=${GLOG_INSTALL_DIR}/lib
-DCMAKE_POSITION_INDEPENDENT_CODE=ON
@ -63,7 +67,7 @@ ExternalProject_Add(
IF(WIN32)
IF(NOT EXISTS "${GLOG_INSTALL_DIR}/lib/libglog.lib")
add_custom_command(TARGET extern_glog POST_BUILD
COMMAND cmake -E rename ${GLOG_INSTALL_DIR}/lib/glog.lib ${GLOG_INSTALL_DIR}/lib/libglog.lib
COMMAND cmake -E copy ${GLOG_INSTALL_DIR}/lib/glog.lib ${GLOG_INSTALL_DIR}/lib/libglog.lib
)
ENDIF()
ENDIF(WIN32)

@ -17,12 +17,8 @@ IF(USE_EIGEN_FOR_BLAS)
ENDIF(USE_EIGEN_FOR_BLAS)
INCLUDE(cblas)
# IF(WIN32 AND NOT ${CBLAS_FOUND})
IF(NOT ${CBLAS_FOUND})
INCLUDE(ExternalProject)
SET(CBLAS_SOURCES_DIR ${THIRD_PARTY_PATH}/openblas)
@ -34,6 +30,7 @@ IF(NOT ${CBLAS_FOUND})
CACHE FILEPATH "openblas library." FORCE)
ADD_DEFINITIONS(-DPADDLE_USE_OPENBLAS)
IF (WIN32)
SET(CBLAS_FOUND true)
MESSAGE(WARNING, "In windows, openblas only support msvc build, please build it manually and put it at " ${CBLAS_INSTALL_DIR})

@ -140,7 +140,6 @@ endmacro()
set(PROTOBUF_ROOT "" CACHE PATH "Folder contains protobuf")
IF (WIN32)
SET(PROTOBUF_ROOT ${THIRD_PARTY_PATH}/install/protobuf)
MESSAGE(WARNING, "In windows, protobuf only support msvc build, please build it manually and put it at " ${PROTOBUF_ROOT})
ENDIF(WIN32)
if (NOT "${PROTOBUF_ROOT}" STREQUAL "")
@ -188,13 +187,20 @@ FUNCTION(build_protobuf TARGET_NAME BUILD_FOR_HOST)
SET(OPTIONAL_ARGS
"-DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}"
"-DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}"
"-DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}"
"-DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}"
"-DCMAKE_C_FLAGS_DEBUG=${CMAKE_C_FLAGS_DEBUG}"
"-DCMAKE_C_FLAGS_RELEASE=${CMAKE_C_FLAGS_RELEASE}"
"-DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}"
"-DCMAKE_CXX_FLAGS_RELEASE=${CMAKE_CXX_FLAGS_RELEASE}"
"-DCMAKE_CXX_FLAGS_DEBUG=${CMAKE_CXX_FLAGS_DEBUG}"
"-Dprotobuf_WITH_ZLIB=ON"
"-DZLIB_ROOT:FILEPATH=${ZLIB_ROOT}"
${EXTERNAL_OPTIONAL_ARGS})
SET(OPTIONAL_CACHE_ARGS "-DZLIB_ROOT:STRING=${ZLIB_ROOT}")
ENDIF()
IF(WIN32)
SET(OPTIONAL_ARGS ${OPTIONAL_ARGS} "-DCMAKE_GENERATOR_PLATFORM=x64")
ENDIF()
SET(PROTOBUF_REPO "https://github.com/google/protobuf.git")
SET(PROTOBUF_TAG "9f75c5aa851cd877fb0d93ccc31b8567a6706546")

@ -21,6 +21,48 @@ INCLUDE(python_module)
FIND_PACKAGE(PythonInterp ${PY_VERSION})
FIND_PACKAGE(PythonLibs ${PY_VERSION})
if(WIN32)
execute_process(COMMAND "${PYTHON_EXECUTABLE}" "-c"
"from distutils import sysconfig as s;import sys;import struct;
print(sys.prefix);
print(s.get_config_var('LDVERSION') or s.get_config_var('VERSION'));
"
RESULT_VARIABLE _PYTHON_SUCCESS
OUTPUT_VARIABLE _PYTHON_VALUES
ERROR_VARIABLE _PYTHON_ERROR_VALUE)
if(NOT _PYTHON_SUCCESS MATCHES 0)
set(PYTHONLIBS_FOUND FALSE)
return()
endif()
# Convert the process output into a list
string(REGEX REPLACE ";" "\\\\;" _PYTHON_VALUES ${_PYTHON_VALUES})
string(REGEX REPLACE "\n" ";" _PYTHON_VALUES ${_PYTHON_VALUES})
list(GET _PYTHON_VALUES 0 PYTHON_PREFIX)
list(GET _PYTHON_VALUES 1 PYTHON_LIBRARY_SUFFIX)
# Make sure all directory separators are '/'
string(REGEX REPLACE "\\\\" "/" PYTHON_PREFIX ${PYTHON_PREFIX})
set(PYTHON_LIBRARY
"${PYTHON_PREFIX}/libs/Python${PYTHON_LIBRARY_SUFFIX}.lib")
# when run in a venv, PYTHON_PREFIX points to it. But the libraries remain in the
# original python installation. They may be found relative to PYTHON_INCLUDE_DIR.
if(NOT EXISTS "${PYTHON_LIBRARY}")
get_filename_component(_PYTHON_ROOT ${PYTHON_INCLUDE_DIR} DIRECTORY)
set(PYTHON_LIBRARY
"${_PYTHON_ROOT}/libs/Python${PYTHON_LIBRARY_SUFFIX}.lib")
endif()
# raise an error if the python libs are still not found.
if(NOT EXISTS "${PYTHON_LIBRARY}")
message(FATAL_ERROR "Python libraries not found")
endif()
SET(PYTHON_LIBRARIES "${PYTHON_LIBRARY}")
endif(WIN32)
# Fixme: Maybe find a static library. Get SHARED/STATIC by FIND_PACKAGE.
ADD_LIBRARY(python SHARED IMPORTED GLOBAL)
SET_PROPERTY(TARGET python PROPERTY IMPORTED_LOCATION ${PYTHON_LIBRARIES})

@ -14,23 +14,52 @@ ELSE()
ENDIF(APPLE)
ENDIF()
ExternalProject_Add(
extern_xxhash
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/Cyan4973/xxHash"
GIT_TAG "v0.6.5"
PREFIX ${XXHASH_SOURCE_DIR}
DOWNLOAD_NAME "xxhash"
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_IN_SOURCE 1
PATCH_COMMAND
BUILD_COMMAND ${BUILD_CMD}
INSTALL_COMMAND export PREFIX=${XXHASH_INSTALL_DIR}/ && make install
TEST_COMMAND ""
)
if(WIN32)
ExternalProject_Add(
extern_xxhash
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/Cyan4973/xxHash"
GIT_TAG "v0.6.5"
PREFIX ${XXHASH_SOURCE_DIR}
DOWNLOAD_NAME "xxhash"
UPDATE_COMMAND ""
BUILD_IN_SOURCE 1
PATCH_COMMAND
CONFIGURE_COMMAND
${CMAKE_COMMAND} ${XXHASH_SOURCE_DIR}/src/extern_xxhash/cmake_unofficial
-DCMAKE_INSTALL_PREFIX:PATH=${XXHASH_INSTALL_DIR}
-DCMAKE_BUILD_TYPE:STRING=${CMAKE_BUILD_TYPE}
-DCMAKE_VERBOSE_MAKEFILE:BOOL=OFF
-DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
-DBUILD_XXHSUM=OFF
-DCMAKE_GENERATOR_PLATFORM=x64
-DBUILD_SHARED_LIBS=OFF
${OPTIONAL_CACHE_ARGS}
TEST_COMMAND ""
)
else()
ExternalProject_Add(
extern_xxhash
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/Cyan4973/xxHash"
GIT_TAG "v0.6.5"
PREFIX ${XXHASH_SOURCE_DIR}
DOWNLOAD_NAME "xxhash"
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_IN_SOURCE 1
PATCH_COMMAND
BUILD_COMMAND ${BUILD_CMD}
INSTALL_COMMAND export PREFIX=${XXHASH_INSTALL_DIR}/ && make install
TEST_COMMAND ""
)
endif()
set(XXHASH_LIBRARIES "${XXHASH_INSTALL_DIR}/lib/libxxhash.a")
if (WIN32)
set(XXHASH_LIBRARIES "${XXHASH_INSTALL_DIR}/lib/xxhash.lib")
else()
set(XXHASH_LIBRARIES "${XXHASH_INSTALL_DIR}/lib/libxxhash.a")
endif ()
INCLUDE_DIRECTORIES(${XXHASH_INCLUDE_DIR})
add_library(xxhash STATIC IMPORTED GLOBAL)

@ -266,7 +266,11 @@ function(cc_library TARGET_NAME)
if("${cc_library_DEPS};" MATCHES "python;")
list(REMOVE_ITEM cc_library_DEPS python)
add_dependencies(${TARGET_NAME} python)
target_link_libraries(${TARGET_NAME} "-Wl,-undefined,dynamic_lookup")
if(WIN32)
target_link_libraries(${TARGET_NAME} ${PYTHON_LIBRARIES})
else()
target_link_libraries(${TARGET_NAME} "-Wl,-undefined,dynamic_lookup")
endif(WIN32)
endif()
target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
add_dependencies(${TARGET_NAME} ${cc_library_DEPS})
@ -288,6 +292,45 @@ function(cc_library TARGET_NAME)
endif(cc_library_SRCS)
endfunction(cc_library)
# The link operation under windows may exceeds the maximum characters limit, simply break the link command
# into multiple link opeartion can fix that, say
# original:
# lib /out:target.lib a.lib b.lib c.lib d.lib
# after:
# 1. lib /out:dummy_lib_1.lib a.lib b.lib
# 2. lib /out:dummy_lib_2.lib c.lib d.lib
# 1. lib /out:target.lib dummy_lib_1.lib dummy_lib_2.lib
function(sep_library TARGET_NAME)
set(options STATIC static SHARED shared)
set(oneValueArgs "")
set(multiValueArgs SRCS DEPS)
cmake_parse_arguments(sep_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
set(dummy_index 1)
set(dummy_offset 1)
# the dummy target would be consisted of limit size libraries
set(dummy_limit 50)
list(LENGTH sep_library_DEPS sep_all_len)
foreach(v ${sep_library_DEPS})
list(APPEND dummy_list ${v})
list(LENGTH dummy_list listlen )
if ((${listlen} GREATER ${dummy_limit}) OR (${dummy_offset} EQUAL ${sep_all_len}))
message("create dummy library ${TARGET_NAME}_dummy_lib_${dummy_index} for ${TARGET_NAME}")
cc_library(${TARGET_NAME}_dummy_lib_${dummy_index} STATIC DEPS ${dummy_list})
foreach(i ${dummy_list})
list(REMOVE_AT dummy_list 0)
endforeach()
list(APPEND ${TARGET_NAME}_dummy_list ${TARGET_NAME}_dummy_lib_${dummy_index})
MATH(EXPR dummy_index "${dummy_index}+1")
endif()
MATH(EXPR dummy_offset "${dummy_offset}+1")
endforeach()
if(${sep_library_SHARED})
cc_library(${TARGET_NAME} SHARED SRCS ${sep_library_SRCS} DEPS ${${TARGET_NAME}_dummy_list})
else(${sep_library_SHARED})
cc_library(${TARGET_NAME} STATIC SRCS ${sep_library_SRCS} DEPS ${${TARGET_NAME}_dummy_list})
endif(${sep_library_SHARED})
endfunction(sep_library)
function(cc_binary TARGET_NAME)
set(options "")
set(oneValueArgs "")

File diff suppressed because it is too large Load Diff

@ -1 +1 @@
../../../CONTRIBUTING.md
../../../CONTRIBUTING.md

@ -4,11 +4,12 @@ add_subdirectory(framework)
add_subdirectory(operators)
add_subdirectory(string)
if (NOT WIN32)
add_subdirectory(pybind)
if (NOT WIN32)
add_subdirectory(recordio)
endif(NOT WIN32)
# NOTE: please add subdirectory inference at last.
add_subdirectory(inference)
add_subdirectory(train)

@ -137,23 +137,31 @@ cc_test(version_test SRCS version_test.cc DEPS version)
cc_library(proto_desc SRCS var_desc.cc op_desc.cc block_desc.cc program_desc.cc DEPS shape_inference op_info operator glog version)
cc_library(ngraph_bridge SRCS ngraph_bridge.cc DEPS operator framework_proto)
if(NOT WIN32)
cc_library(ngraph_operator SRCS ngraph_operator.cc DEPS ngraph_bridge operator op_info device_context tensor scope glog
shape_inference data_transform lod_tensor profiler)
endif(NOT WIN32)
cc_library(op_registry SRCS op_registry.cc DEPS op_proto_maker op_info operator glog proto_desc)
nv_test(op_registry_test SRCS op_registry_test.cc DEPS op_registry)
if (NOT WIN32)
py_proto_compile(framework_py_proto SRCS framework.proto)
# Generate an empty __init__.py to make framework_py_proto as a valid python module.
add_custom_target(framework_py_proto_init ALL COMMAND ${CMAKE_COMMAND} -E touch __init__.py)
add_dependencies(framework_py_proto framework_py_proto_init)
add_custom_command(TARGET framework_py_proto POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${PADDLE_BINARY_DIR}/python/paddle/fluid/proto
COMMAND cp *.py ${PADDLE_BINARY_DIR}/python/paddle/fluid/proto/
COMMENT "Copy generated python proto into directory paddle/fluid/proto."
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
if (NOT WIN32)
add_custom_command(TARGET framework_py_proto POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${PADDLE_BINARY_DIR}/python/paddle/fluid/proto
COMMAND cp *.py ${PADDLE_BINARY_DIR}/python/paddle/fluid/proto/
COMMENT "Copy generated python proto into directory paddle/fluid/proto."
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
else(NOT WIN32)
string(REPLACE "/" "\\" proto_dstpath "${PADDLE_BINARY_DIR}/python/paddle/fluid/proto/")
add_custom_command(TARGET framework_py_proto POST_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory ${PADDLE_BINARY_DIR}/python/paddle/fluid/proto
COMMAND copy /Y *.py ${proto_dstpath}
COMMENT "Copy generated python proto into directory paddle/fluid/proto."
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
endif(NOT WIN32)
cc_library(lod_rank_table SRCS lod_rank_table.cc DEPS lod_tensor)
@ -167,7 +175,11 @@ if(WITH_DISTRIBUTE)
set(DISTRIBUTE_COMPILE_FLAGS "-Wno-non-virtual-dtor -Wno-error=non-virtual-dtor -Wno-error=delete-non-virtual-dtor")
set_source_files_properties(executor.cc PROPERTIES COMPILE_FLAGS ${DISTRIBUTE_COMPILE_FLAGS})
else()
cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method graph_to_program_pass ngraph_operator)
if(NOT WIN32)
cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method graph_to_program_pass ngraph_operator)
else(NOT WIN32)
cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method graph_to_program_pass)
endif(NOT WIN32)
cc_test(test_naive_executor SRCS naive_executor_test.cc DEPS naive_executor elementwise_add_op)
endif()

@ -29,7 +29,7 @@ template <typename T>
class GarbageCollector {
public:
GarbageCollector(const platform::Place &place, size_t max_memory_size)
: max_memory_size_(std::max(max_memory_size, static_cast<size_t>(1))) {
: max_memory_size_((std::max)(max_memory_size, static_cast<size_t>(1))) {
garbages_.reset(new std::deque<T *>());
dev_ctx_ = platform::DeviceContextPool::Instance().Get(place);
}

@ -211,12 +211,12 @@ void PrepareLSTMWeight(const LoDTensor& W_forget_w0,
VLOG(30) << "LSTMWeight resized to " << out->dims();
float* out_data = out->mutable_data<float>(platform::CPUPlace());
std::array<const float*, 4> tensors(
{{W_forget_w0.data<float>(), W_input_w0.data<float>(),
W_output_w0.data<float>(), W_cell_w0.data<float>()}});
std::array<const float*, 4> tensors1(
{{W_forget_w1.data<float>(), W_input_w1.data<float>(),
W_output_w1.data<float>(), W_cell_w1.data<float>()}});
std::array<const float*, 4> tensors{
W_forget_w0.data<float>(), W_input_w0.data<float>(),
W_output_w0.data<float>(), W_cell_w0.data<float>()};
std::array<const float*, 4> tensors1{
W_forget_w1.data<float>(), W_input_w1.data<float>(),
W_output_w1.data<float>(), W_cell_w1.data<float>()};
for (int row = 0; row < D; row++) {
for (int col = 0; col < 4; col++) {
@ -238,9 +238,9 @@ void PrepareLSTMWeight(const LoDTensor& W_forget_w0,
void PrepareLSTMBias(const LoDTensor& B_forget, const LoDTensor& B_input,
const LoDTensor& B_output, const LoDTensor& B_cell,
LoDTensor* out) {
std::array<const float*, 4> tensors(
{{B_forget.data<float>(), B_input.data<float>(), B_output.data<float>(),
B_cell.data<float>()}});
std::array<const float*, 4> tensors{
B_forget.data<float>(), B_input.data<float>(), B_output.data<float>(),
B_cell.data<float>()};
PADDLE_ENFORCE_EQ(B_forget.dims().size(), 1);
int D = B_forget.dims()[0];

@ -17,7 +17,12 @@ limitations under the License. */
namespace paddle {
namespace framework {
namespace ir {
// msvc15 don't support constexpr in correct way.
#if !defined(_WIN32)
constexpr char Node::kControlDepVarName[];
#else
const char Node::kControlDepVarName[] = "__control_var";
#endif
std::unique_ptr<Node> CreateNodeForTest(const std::string& name,
Node::Type type) {

@ -55,7 +55,11 @@ class Node {
}
enum class Type { kOperation, kVariable };
#if !defined(_WIN32) // msvc not support constexpr correctly.
static constexpr char kControlDepVarName[] = "__control_var";
#else
static const char kControlDepVarName[];
#endif
Type NodeType() const { return type_; }

@ -197,26 +197,26 @@ struct PassRegistrar : public Registrar {
msg)
// Register a new pass that can be applied on the IR.
#define REGISTER_PASS(pass_type, pass_class) \
STATIC_ASSERT_PASS_GLOBAL_NAMESPACE( \
__reg_pass__##pass_type, \
"REGISTER_PASS must be called in global namespace"); \
static ::paddle::framework::ir::PassRegistrar<pass_class> \
__pass_registrar_##pass_type##__(#pass_type); \
int TouchPassRegistrar_##pass_type() { \
__pass_registrar_##pass_type##__.Touch(); \
return 0; \
} \
static ::paddle::framework::ir::PassRegistrar<pass_class> \
&__pass_tmp_registrar_##pass_type##__ __attribute__((unused)) = \
#define REGISTER_PASS(pass_type, pass_class) \
STATIC_ASSERT_PASS_GLOBAL_NAMESPACE( \
__reg_pass__##pass_type, \
"REGISTER_PASS must be called in global namespace"); \
static ::paddle::framework::ir::PassRegistrar<pass_class> \
__pass_registrar_##pass_type##__(#pass_type); \
int TouchPassRegistrar_##pass_type() { \
__pass_registrar_##pass_type##__.Touch(); \
return 0; \
} \
static ::paddle::framework::ir::PassRegistrar<pass_class> \
&__pass_tmp_registrar_##pass_type##__ UNUSED = \
__pass_registrar_##pass_type##__
#define USE_PASS(pass_type) \
STATIC_ASSERT_PASS_GLOBAL_NAMESPACE( \
__use_pass_itself_##pass_type, \
"USE_PASS must be called in global namespace"); \
extern int TouchPassRegistrar_##pass_type(); \
static int use_pass_itself_##pass_type##_ __attribute__((unused)) = \
#define USE_PASS(pass_type) \
STATIC_ASSERT_PASS_GLOBAL_NAMESPACE( \
__use_pass_itself_##pass_type, \
"USE_PASS must be called in global namespace"); \
extern int TouchPassRegistrar_##pass_type(); \
static int use_pass_itself_##pass_type##_ UNUSED = \
TouchPassRegistrar_##pass_type()
} // namespace ir

@ -150,14 +150,17 @@ void OperatorBase::Run(const Scope& scope, const platform::Place& place) {
#endif
}
// The profile has a process-wide mutex, results in serious performance issue
// in concurrency scenerio. Here use an `if` to fix this issue.
// Please not remove the `if`, ask @Superjomn if there are any concern.
// The profile has a process-wide mutex, results in serious performance issue
// in concurrency scenerio. Here use an `if` to fix this issue.
// Please not remove the `if`, ask @Superjomn if there are any concern.
#ifndef _WIN32
if (platform::IsProfileEnabled()) {
platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
platform::RecordEvent record_event(Type(), pool.Get(place));
RunImpl(scope, place);
} else {
} else
#endif
{
RunImpl(scope, place);
}
VLOG(30) << place << " " << DebugStringEx(&scope);

@ -16,9 +16,21 @@ cc_library(paddle_fluid_api
DEPS ${FLUID_CORE_MODULES} ${GLOB_OP_LIB})
get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
get_property(cuda_modules GLOBAL PROPERTY CUDA_MODULES)
get_property(fluid_third_partys GLOBAL PROPERTY FLUID_THRID_PARTYS)
if (WIN32)
list(APPEND fluid_third_partys gflags glog protobuf cblas)
endif(WIN32)
# paddle_fluid_origin exclude inference api interface
cc_library(paddle_fluid_origin DEPS ${fluid_modules} paddle_fluid_api)
if(WIN32)
sep_library(paddle_fluid_origin DEPS ${fluid_modules} paddle_fluid_api)
if(WITH_GPU AND NOT WITH_DSO)
target_link_libraries(paddle_fluid_origin ${cuda_modules})
endif(WITH_GPU AND NOT WITH_DSO)
else(WIN32)
cc_library(paddle_fluid_origin DEPS ${fluid_modules} paddle_fluid_api)
endif(WIN32)
add_subdirectory(api)
@ -28,8 +40,16 @@ set(SHARED_INFERENCE_SRCS
${CMAKE_CURRENT_SOURCE_DIR}/api/analysis_predictor.cc
${CMAKE_CURRENT_SOURCE_DIR}/api/details/zero_copy_tensor.cc)
# Create static library
cc_library(paddle_fluid DEPS ${fluid_modules} ${STATIC_INFERENCE_APIS} zero_copy_tensor reset_tensor_array analysis_config paddle_pass_builder)
if(WIN32)
sep_library(paddle_fluid DEPS ${fluid_modules} ${STATIC_INFERENCE_APIS} zero_copy_tensor reset_tensor_array
analysis_config paddle_pass_builder)
if(WITH_GPU AND NOT WITH_DSO)
target_link_libraries(paddle_fluid ${cuda_modules})
endif(WITH_GPU AND NOT WITH_DSO)
else(WIN32)
cc_library(paddle_fluid DEPS ${fluid_modules} ${STATIC_INFERENCE_APIS} zero_copy_tensor reset_tensor_array
analysis_config paddle_pass_builder)
endif(WIN32)
if(NOT APPLE)
# TODO(liuyiqu: Temporarily disable the link flag because it is not support on Mac.
@ -38,11 +58,20 @@ if(NOT APPLE)
endif()
# Create shared library
cc_library(paddle_fluid_shared SHARED SRCS ${SHARED_INFERENCE_SRCS}
DEPS ${fluid_modules} paddle_fluid_api reset_tensor_array analysis_config paddle_pass_builder)
if(WIN32)
sep_library(paddle_fluid_shared SHARED SRCS ${SHARED_INFERENCE_SRCS}
DEPS ${fluid_modules} paddle_fluid_api reset_tensor_array analysis_config paddle_pass_builder)
target_link_libraries(paddle_fluid_shared shlwapi)
if(WITH_GPU AND NOT WITH_DSO)
target_link_libraries(paddle_fluid_origin ${cuda_modules})
endif(WITH_GPU AND NOT WITH_DSO)
else(WIN32)
cc_library(paddle_fluid_shared SHARED SRCS ${SHARED_INFERENCE_SRCS}
DEPS ${fluid_modules} paddle_fluid_api reset_tensor_array analysis_config paddle_pass_builder)
endif()
set_target_properties(paddle_fluid_shared PROPERTIES OUTPUT_NAME paddle_fluid)
if(NOT APPLE)
if(NOT APPLE AND NOT WIN32)
# TODO(liuyiqun): Temporarily disable the link flag because it is not support on Mac.
set(LINK_FLAGS "-Wl,--version-script ${CMAKE_CURRENT_SOURCE_DIR}/paddle_fluid.map")
set_target_properties(paddle_fluid_shared PROPERTIES LINK_FLAGS "${LINK_FLAGS}")

@ -26,6 +26,7 @@ limitations under the License. */
#include "paddle/fluid/framework/scope.h"
#include "paddle/fluid/framework/variable.h"
#include "paddle/fluid/platform/enforce.h"
#include "paddle/fluid/platform/port.h"
namespace paddle {
namespace inference {
@ -124,20 +125,6 @@ T &GetFromScope(const framework::Scope &scope, const std::string &name) {
return *var->GetMutable<T>();
}
static void ExecShellCommand(const std::string &cmd, std::string *message) {
char buffer[128];
std::shared_ptr<FILE> pipe(popen(cmd.c_str(), "r"), pclose);
if (!pipe) {
LOG(ERROR) << "error running command: " << cmd;
return;
}
while (!feof(pipe.get())) {
if (fgets(buffer, 128, pipe.get()) != nullptr) {
*message += buffer;
}
}
}
static framework::proto::ProgramDesc LoadProgramDesc(
const std::string &model_path) {
std::ifstream fin(model_path, std::ios::in | std::ios::binary);

@ -15,9 +15,14 @@
#pragma once
#include <glog/logging.h>
#if !defined(_WIN32)
#include <sys/time.h>
#else
#endif
#include <algorithm>
#include <chrono> // NOLINT
#include <iterator>
#include <numeric>
#include <sstream>
#include <string>

@ -94,7 +94,8 @@ function(op_library TARGET)
# remove windows unsupported op, because windows has no nccl, no warpctc such ops.
foreach(windows_unsupport_op "nccl_op" "gen_nccl_id_op" "warpctc_op" "hierarchical_sigmoid_op"
"crf_decoding_op" "select_op" "lstmp_op" "gru_op" "fusion_gru_op" "lstm_op" "fusion_lstm_op" "cumsum_op"
"fusion_seqconv_eltadd_relu_op" "channel_send_op" "channel_create_op" "channel_close_op" "channel_recv_op")
"fusion_seqconv_eltadd_relu_op" "channel_send_op" "channel_create_op" "channel_close_op" "channel_recv_op"
"fusion_seqexpand_concat_fc_op" "attention_lstm_op" "fused_embedding_fc_lstm_op" "fc_op")
if ("${TARGET}" STREQUAL "${windows_unsupport_op}")
return()
endif()
@ -308,8 +309,10 @@ op_library(flatten_op DEPS reshape_op)
op_library(sequence_pad_op DEPS sequence_padding)
op_library(unstack_op DEPS stack_op)
op_library(fake_quantize_op DEPS memory)
if (NOT WIN32)
op_library(crf_decoding_op DEPS jit_kernel)
op_library(fusion_lstm_op DEPS jit_kernel)
endif(NOT WIN32)
if (WITH_GPU)
op_library(conv_op DEPS vol2col depthwise_conv im2col)
op_library(layer_norm_op DEPS cub)
@ -325,8 +328,8 @@ op_library(save_op DEPS lod_tensor)
op_library(load_op DEPS lod_tensor)
op_library(save_combine_op DEPS lod_tensor)
op_library(load_combine_op DEPS lod_tensor)
op_library(tensor_array_to_tensor_op DEPS concat_op)
op_library(concat_op DEPS concat_and_split)
op_library(tensor_array_to_tensor_op DEPS concat_op)
list(REMOVE_ITEM GENERAL_OPS ${DEPS_OPS})

Some files were not shown because too many files have changed in this diff Show More

Loading…
Cancel
Save