Merge pull request #14931 from wopeizl/windows/mkl

add mkl support for windows

CMakeLists.txt

@@ -126,16 +126,12 @@ if(ANDROID OR IOS)
     add_definitions(-DPADDLE_MOBILE_INFERENCE)
 endif()
 
-if (APPLE OR WIN32)
+if (APPLE)
     set(WITH_MKL OFF CACHE STRING
-        "Disable MKL for building on mac and windows" FORCE)
+        "Disable MKL for building on mac" FORCE)
 endif()
 
 if (WIN32)
-    set(WITH_DSO OFF CACHE STRING
-            "Disable DSO when compiling for Windows" FORCE)
-    set(WITH_MKL OFF CACHE STRING
-            "Disable MKL when compiling for Windows" FORCE)
     set(WITH_DISTRIBUTE OFF CACHE STRING
             "Disable DISTRIBUTE when compiling for Windows" FORCE)
     set(WITH_C_API OFF CACHE STRING

cmake/external/mkldnn.cmake

@@ -23,15 +23,14 @@ SET(MKLDNN_SOURCES_DIR    ${THIRD_PARTY_PATH}/mkldnn)
 SET(MKLDNN_INSTALL_DIR    ${THIRD_PARTY_PATH}/install/mkldnn)
 SET(MKLDNN_INC_DIR        "${MKLDNN_INSTALL_DIR}/include" CACHE PATH "mkldnn include directory." FORCE)
 
-IF(WIN32 OR APPLE)
+IF(APPLE)
     MESSAGE(WARNING
-        "Windows or Mac is not supported with MKLDNN in Paddle yet."
+        "Mac is not supported with MKLDNN in Paddle yet."
         "Force WITH_MKLDNN=OFF")
-    SET(WITH_MKLDNN OFF CACHE STRING "Disable MKLDNN in Windows and MacOS" FORCE)
+    SET(WITH_MKLDNN OFF CACHE STRING "Disable MKLDNN in MacOS" FORCE)
     return()
 ENDIF()
 
-SET(MKLDNN_LIB "${MKLDNN_INSTALL_DIR}/lib/libmkldnn.so" CACHE FILEPATH "mkldnn library." FORCE)
 MESSAGE(STATUS "Set ${MKLDNN_INSTALL_DIR}/lib to runtime path")
 SET(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)
 SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${MKLDNN_INSTALL_DIR}/lib")
@@ -44,10 +43,14 @@ IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
 ELSE()
     MESSAGE(FATAL_ERROR "Should enable MKLML when build MKLDNN")
 ENDIF()
-SET(MKLDNN_FLAG "-Wno-error=strict-overflow -Wno-error=unused-result -Wno-error=array-bounds")
-SET(MKLDNN_FLAG "${MKLDNN_FLAG} -Wno-unused-result -Wno-unused-value")
-SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} ${MKLDNN_FLAG}")
-SET(MKLDNN_CXXFLAG "${CMAKE_CXX_FLAGS} ${MKLDNN_FLAG}")
+
+IF(NOT WIN32)
+    SET(MKLDNN_FLAG "-Wno-error=strict-overflow -Wno-error=unused-result -Wno-error=array-bounds")
+    SET(MKLDNN_FLAG "${MKLDNN_FLAG} -Wno-unused-result -Wno-unused-value")
+    SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} ${MKLDNN_FLAG}")
+    SET(MKLDNN_CXXFLAG "${CMAKE_CXX_FLAGS} ${MKLDNN_FLAG}")
+ENDIF(NOT WIN32)
+
 ExternalProject_Add(
     ${MKLDNN_PROJECT}
     ${EXTERNAL_PROJECT_LOG_ARGS}
@@ -58,8 +61,15 @@ ExternalProject_Add(
     UPDATE_COMMAND      ""
     CMAKE_ARGS          -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
     CMAKE_ARGS          -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
+    CMAKE_ARGS          -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
+    CMAKE_ARGS          -DCMAKE_CXX_FLAGS_RELEASE=${CMAKE_CXX_FLAGS_RELEASE}
+    CMAKE_ARGS          -DCMAKE_CXX_FLAGS_DEBUG=${CMAKE_CXX_FLAGS_DEBUG}
+    CMAKE_ARGS          -DCMAKE_C_FLAGS=${CMAKE_C_FLAGS}
+    CMAKE_ARGS          -DCMAKE_C_FLAGS_DEBUG=${CMAKE_C_FLAGS_DEBUG}
+    CMAKE_ARGS          -DCMAKE_C_FLAGS_RELEASE=${CMAKE_C_FLAGS_RELEASE}
     CMAKE_ARGS          -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR}
     CMAKE_ARGS          -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}
+    CMAKE_ARGS          -DCMAKE_POSITION_INDEPENDENT_CODE=ON
     CMAKE_ARGS          -DMKLROOT=${MKLML_ROOT}
     CMAKE_ARGS          -DCMAKE_C_FLAGS=${MKLDNN_CFLAG}
     CMAKE_ARGS          -DCMAKE_CXX_FLAGS=${MKLDNN_CXXFLAG}
@@ -67,6 +77,11 @@ ExternalProject_Add(
     CMAKE_CACHE_ARGS    -DCMAKE_INSTALL_PREFIX:PATH=${MKLDNN_INSTALL_DIR}
                         -DMKLROOT:PATH=${MKLML_ROOT}
 )
+if(WIN32)
+    SET(MKLDNN_LIB "${MKLDNN_INSTALL_DIR}/lib/mkldnn.lib" CACHE FILEPATH "mkldnn library." FORCE)
+else(WIN32)
+    SET(MKLDNN_LIB "${MKLDNN_INSTALL_DIR}/lib/libmkldnn.so" CACHE FILEPATH "mkldnn library." FORCE)
+endif(WIN32)
 
 ADD_LIBRARY(shared_mkldnn SHARED IMPORTED GLOBAL)
 SET_PROPERTY(TARGET shared_mkldnn PROPERTY IMPORTED_LOCATION ${MKLDNN_LIB})
@@ -85,10 +100,14 @@ ADD_DEPENDENCIES(mkldnn ${MKLDNN_PROJECT})
 
 # copy the real so.0 lib to install dir
 # it can be directly contained in wheel or capi
-SET(MKLDNN_SHARED_LIB ${MKLDNN_INSTALL_DIR}/libmkldnn.so.0)
-ADD_CUSTOM_COMMAND(OUTPUT ${MKLDNN_SHARED_LIB}
-    COMMAND cp ${MKLDNN_LIB} ${MKLDNN_SHARED_LIB}
-    DEPENDS mkldnn)
+if(WIN32)
+    SET(MKLDNN_SHARED_LIB ${MKLDNN_INSTALL_DIR}/lib/mkldnn.dll)
+else(WIN32)
+    SET(MKLDNN_SHARED_LIB ${MKLDNN_INSTALL_DIR}/libmkldnn.so.0)
+    ADD_CUSTOM_COMMAND(OUTPUT ${MKLDNN_SHARED_LIB}
+            COMMAND ${CMAKE_COMMAND} -E copy ${MKLDNN_LIB} ${MKLDNN_SHARED_LIB}
+            DEPENDS mkldnn)
+endif(WIN32)
 ADD_CUSTOM_TARGET(mkldnn_shared_lib ALL DEPENDS ${MKLDNN_SHARED_LIB})
 
 IF(WITH_C_API)

cmake/external/mklml.cmake

@@ -16,56 +16,67 @@ IF(NOT ${WITH_MKLML})
   return()
 ENDIF(NOT ${WITH_MKLML})
 
-IF(WIN32 OR APPLE)
+IF(APPLE)
     MESSAGE(WARNING
-        "Windows or Mac is not supported with MKLML in Paddle yet."
+        "Mac is not supported with MKLML in Paddle yet."
         "Force WITH_MKLML=OFF")
     SET(WITH_MKLML OFF CACHE STRING "Disable MKLML package in Windows and MacOS" FORCE)
     return()
 ENDIF()
 
 INCLUDE(ExternalProject)
-
-SET(MKLML_PROJECT       "extern_mklml")
-IF((NOT DEFINED MKLML_VER) OR (NOT DEFINED MKLML_URL))
-  MESSAGE(STATUS "use pre defined download url")
-  SET(MKLML_VER "mklml_lnx_2019.0.20180710" CACHE STRING "" FORCE)
-  SET(MKLML_URL "http://paddlepaddledeps.cdn.bcebos.com/${MKLML_VER}.tgz" CACHE STRING "" FORCE)
-ENDIF()
-MESSAGE(STATUS "MKLML_VER: ${MKLML_VER}, MKLML_URL: ${MKLML_URL}")
-SET(MKLML_SOURCE_DIR    "${THIRD_PARTY_PATH}/mklml")
-SET(MKLML_DOWNLOAD_DIR  "${MKLML_SOURCE_DIR}/src/${MKLML_PROJECT}")
 SET(MKLML_DST_DIR       "mklml")
 SET(MKLML_INSTALL_ROOT  "${THIRD_PARTY_PATH}/install")
 SET(MKLML_INSTALL_DIR   ${MKLML_INSTALL_ROOT}/${MKLML_DST_DIR})
 SET(MKLML_ROOT          ${MKLML_INSTALL_DIR})
 SET(MKLML_INC_DIR       ${MKLML_ROOT}/include)
 SET(MKLML_LIB_DIR       ${MKLML_ROOT}/lib)
-SET(MKLML_LIB           ${MKLML_LIB_DIR}/libmklml_intel.so)
-SET(MKLML_IOMP_LIB      ${MKLML_LIB_DIR}/libiomp5.so)
+if(WIN32)
+    SET(MKLML_LIB                 ${MKLML_LIB_DIR}/mklml.lib)
+    SET(MKLML_IOMP_LIB            ${MKLML_LIB_DIR}/libiomp5md.lib)
+    SET(MKLML_SHARED_LIB          ${MKLML_LIB_DIR}/mklml.dll)
+    SET(MKLML_SHARED_IOMP_LIB     ${MKLML_LIB_DIR}/libiomp5md.dll)
+else()
+    SET(MKLML_LIB                 ${MKLML_LIB_DIR}/libmklml_intel.so)
+    SET(MKLML_IOMP_LIB            ${MKLML_LIB_DIR}/libiomp5.so)
+    SET(MKLML_SHARED_LIB          ${MKLML_LIB_DIR}/libmklml_intel.so)
+    SET(MKLML_SHARED_IOMP_LIB     ${MKLML_LIB_DIR}/libiomp5.so)
+endif()
 SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${MKLML_ROOT}/lib")
 
-INCLUDE_DIRECTORIES(${MKLML_INC_DIR})
+IF((NOT DEFINED MKLML_VER) OR (NOT DEFINED MKLML_URL))
+    MESSAGE(STATUS "use pre defined download url")
+    if(WIN32)
+        SET(MKLML_VER "mklml_win_2019.0.20180710" CACHE STRING "" FORCE)
+        SET(MKLML_URL "https://paddlepaddledeps.cdn.bcebos.com/${MKLML_VER}.zip" CACHE STRING "" FORCE)
+    else()
+        SET(MKLML_VER "mklml_lnx_2019.0.20180710" CACHE STRING "" FORCE)
+        SET(MKLML_URL "http://paddlepaddledeps.cdn.bcebos.com/${MKLML_VER}.tgz" CACHE STRING "" FORCE)
+    ENDIF()
+endif()
 
-FILE(WRITE ${MKLML_DOWNLOAD_DIR}/CMakeLists.txt
-  "PROJECT(MKLML)\n"
-  "cmake_minimum_required(VERSION 3.0)\n"
-  "install(DIRECTORY ${MKLML_VER}/include ${MKLML_VER}/lib \n"
-  "        DESTINATION ${MKLML_DST_DIR})\n")
+SET(MKLML_PROJECT       "extern_mklml")
+MESSAGE(STATUS "MKLML_VER: ${MKLML_VER}, MKLML_URL: ${MKLML_URL}")
+SET(MKLML_SOURCE_DIR    "${THIRD_PARTY_PATH}/mklml")
+SET(MKLML_DOWNLOAD_DIR  "${MKLML_SOURCE_DIR}/src/${MKLML_PROJECT}")
 
 ExternalProject_Add(
     ${MKLML_PROJECT}
     ${EXTERNAL_PROJECT_LOG_ARGS}
-    PREFIX                ${MKLML_SOURCE_DIR}
+    PREFIX                 ${MKLML_SOURCE_DIR}
+    URL                    ${MKLML_URL}
     DOWNLOAD_DIR          ${MKLML_DOWNLOAD_DIR}
-    DOWNLOAD_COMMAND      wget --no-check-certificate ${MKLML_URL} -c -q -O ${MKLML_VER}.tgz 
-                          && tar zxf ${MKLML_VER}.tgz
     DOWNLOAD_NO_PROGRESS  1
-    UPDATE_COMMAND        ""
-    CMAKE_ARGS            -DCMAKE_INSTALL_PREFIX=${MKLML_INSTALL_ROOT}
-    CMAKE_CACHE_ARGS      -DCMAKE_INSTALL_PREFIX:PATH=${MKLML_INSTALL_ROOT}
+    CONFIGURE_COMMAND     ""
+    BUILD_COMMAND         ""
+    UPDATE_COMMAND ""
+    INSTALL_COMMAND
+        ${CMAKE_COMMAND} -E copy_directory ${MKLML_DOWNLOAD_DIR}/include ${MKLML_INC_DIR} &&
+        ${CMAKE_COMMAND} -E copy_directory ${MKLML_DOWNLOAD_DIR}/lib ${MKLML_LIB_DIR}
 )
 
+INCLUDE_DIRECTORIES(${MKLML_INC_DIR})
+
 ADD_LIBRARY(mklml SHARED IMPORTED GLOBAL)
 SET_PROPERTY(TARGET mklml PROPERTY IMPORTED_LOCATION ${MKLML_LIB})
 ADD_DEPENDENCIES(mklml ${MKLML_PROJECT})

cmake/generic.cmake

@@ -267,7 +267,11 @@ function(cc_library TARGET_NAME)
           list(APPEND cc_library_DEPS dynload_mklml)
         endif()
         add_dependencies(${TARGET_NAME} mklml)
-        target_link_libraries(${TARGET_NAME} "-L${MKLML_LIB_DIR} -liomp5 -Wl,--as-needed")
+        if(WIN32)
+          target_link_libraries(${TARGET_NAME} ${MKLML_IOMP_LIB})
+        else(WIN32)
+          target_link_libraries(${TARGET_NAME} "-L${MKLML_LIB_DIR} -liomp5 -Wl,--as-needed")
+        endif(WIN32)
       endif()
       # remove link to python, see notes at:
       # https://github.com/pybind/pybind11/blob/master/docs/compiling.rst#building-manually

cmake/inference_lib.cmake

@@ -115,20 +115,20 @@ if (NOT PROTOBUF_FOUND OR WIN32)
             )
 endif ()
 
-if (NOT CBLAS_FOUND)
-    set(dst_dir "${FLUID_INSTALL_DIR}/third_party/install/openblas")
-    copy(openblas_lib
-            SRCS ${CBLAS_INSTALL_DIR}/lib ${CBLAS_INSTALL_DIR}/include
-            DSTS ${dst_dir} ${dst_dir}
-            DEPS extern_openblas
-            )
-elseif (WITH_MKLML)
+if (WITH_MKLML)
     set(dst_dir "${FLUID_INSTALL_DIR}/third_party/install/mklml")
     copy(mklml_lib
             SRCS ${MKLML_LIB} ${MKLML_IOMP_LIB} ${MKLML_INC_DIR}
             DSTS ${dst_dir}/lib ${dst_dir}/lib ${dst_dir}
             DEPS mklml
             )
+elseif (NOT CBLAS_FOUND OR WIN32)
+    set(dst_dir "${FLUID_INSTALL_DIR}/third_party/install/openblas")
+    copy(openblas_lib
+            SRCS ${CBLAS_INSTALL_DIR}/lib ${CBLAS_INSTALL_DIR}/include
+            DSTS ${dst_dir} ${dst_dir}
+            DEPS extern_openblas
+            )
 endif ()
 
 if (WITH_MKLDNN)

paddle/fluid/inference/api/demo_ci/CMakeLists.txt

@@ -89,12 +89,21 @@ endif()
 
 if(WITH_MKL)
   include_directories("${PADDLE_LIB}/third_party/install/mklml/include")
-  set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel${CMAKE_SHARED_LIBRARY_SUFFIX}
-               ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5${CMAKE_SHARED_LIBRARY_SUFFIX})
+  if(NOT WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel${CMAKE_SHARED_LIBRARY_SUFFIX}
+                 ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5${CMAKE_SHARED_LIBRARY_SUFFIX})
+  else(WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml${CMAKE_SHARED_LIBRARY_SUFFIX}
+            ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5md${CMAKE_SHARED_LIBRARY_SUFFIX})
+  endif(WIN32)
   set(MKLDNN_PATH "${PADDLE_LIB}/third_party/install/mkldnn")
   if(EXISTS ${MKLDNN_PATH})
     include_directories("${MKLDNN_PATH}/include")
-    set(MKLDNN_LIB ${MKLDNN_PATH}/lib/libmkldnn.so.0)
+    if(WIN32)
+      set(MKLDNN_LIB ${MKLDNN_PATH}/lib/mkldnn.lib)
+    else(WIN32)
+      set(MKLDNN_LIB ${MKLDNN_PATH}/lib/libmkldnn.so.0)
+    endif(WIN32)
   endif()
 else()
   set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas${CMAKE_STATIC_LIBRARY_SUFFIX})

paddle/fluid/operators/elementwise/elementwise_mul_mkldnn_op.cc

@@ -16,11 +16,14 @@ limitations under the License. */
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
 
+#include "paddle/fluid/operators/jit/kernels.h"
+#include "paddle/fluid/platform/cpu_info.h"
 #include "paddle/fluid/platform/mkldnn_helper.h"
 
-#include "paddle/fluid/operators/jit/kernels.h"
+#ifdef PADDLE_WITH_XBYAK
 #include "xbyak/xbyak.h"
 #include "xbyak/xbyak_util.h"
+#endif
 
 namespace paddle {
 namespace operators {
@@ -81,8 +84,7 @@ class ElementwiseMulMKLDNNKernel : public framework::OpKernel<T> {
     UpdateDataFormat(ctx, const_cast<Tensor*>(x), "x_data_format");
     UpdateDataFormat(ctx, const_cast<Tensor*>(y), "y_data_format");
 
-    Xbyak::util::Cpu cpu;
-    const bool is_avx512_enabled = cpu.has(Xbyak::util::Cpu::tAVX512F);
+    const bool is_avx512_enabled = platform::MayIUse(platform::avx512f);
     const bool are_dims_divisable = !(x_int_dims[1] % 16);
     const bool is_x_format_correct = x->format() == memory::format::nChw16c;
     const bool is_y_format_correct = y->format() == memory::format::nc;

paddle/fluid/operators/math/detail/lstm_cpu_kernel.h

@@ -17,6 +17,12 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/detail/activation_functions.h"
 #include "paddle/fluid/operators/math/lstm_compute.h"
 
+#if defined(_WIN32)
+#if defined(__AVX2__) || defined(__AVX__)
+inline __m256 operator+=(__m256 a, __m256 b) { return _mm256_add_ps(a, b); }
+#endif
+#endif
+
 namespace paddle {
 namespace operators {
 namespace math {

paddle/fluid/platform/cpu_info.cc

@@ -22,7 +22,6 @@ limitations under the License. */
 #ifdef __APPLE__
 #include <sys/sysctl.h>
 #include <sys/types.h>
-
 #elif defined(_WIN32)
 #define NOMINMAX  // msvc max/min macro conflict with std::min/max
 #include <windows.h>

paddle/fluid/platform/dynload/dynamic_loader.cc

@@ -227,6 +227,8 @@ void* GetTensorRtDsoHandle() {
 void* GetMKLMLDsoHandle() {
 #if defined(__APPLE__) || defined(__OSX__)
   return GetDsoHandleFromSearchPath(FLAGS_mklml_dir, "libmklml_intel.dylib");
+#elif defined(_WIN32)
+  return GetDsoHandleFromSearchPath(FLAGS_mklml_dir, "mklml.dll");
 #else
   return GetDsoHandleFromSearchPath(FLAGS_mklml_dir, "libmklml_intel.so");
 #endif

paddle/fluid/train/demo/CMakeLists.txt

@@ -35,16 +35,26 @@ add_executable(demo_trainer demo_trainer.cc)
 
 if(WITH_MKLDNN)
   include_directories("${PADDLE_LIB}/third_party/install/mkldnn/include")
-  set(MKLDNN_LIB ${PADDLE_LIB}/third_party/install/mkldnn/lib/libmkldnn.so.0)
-endif()
+  if(WIN32)
+    set(MKLDNN_LIB ${PADDLE_LIB}/third_party/install/mkldnn/lib/mkldnn.lib)
+  else(WIN32)
+    set(MKLDNN_LIB ${PADDLE_LIB}/third_party/install/mkldnn/lib/libmkldnn.so.0)
+  endif(WIN32)
+endif(WITH_MKLDNN)
 
 if(WITH_MKL)
   include_directories("${PADDLE_LIB}/third_party/install/mklml/include")
-  set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel.so)
+  if(WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/mklml.lib)
+  else(WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel.so)
+  endif(WIN32)
 else()
   if(APPLE)
     set(MATH_LIB cblas)
-  else(APPLE)
+  elseif(WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas.lib)
+  else()
     set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas.a)
   endif(APPLE)
 endif()

python/CMakeLists.txt

@@ -48,18 +48,13 @@ configure_file(${CMAKE_CURRENT_SOURCE_DIR}/setup.py.in
 IF(WIN32)
     # Python would use the .pyd by default under Windows series platform
     set(FLUID_DST_DIR ${PADDLE_BINARY_DIR}/python/paddle/fluid/)
-    get_filename_component(openblas_refpath ${CBLAS_LIBRARIES} DIRECTORY)
     set(FLUID_CORE ${FLUID_DST_DIR}/core.pyd)
-    add_custom_command(OUTPUT ${FLUID_CORE}
-            COMMAND cmake -E copy $<TARGET_FILE:paddle_pybind> ${FLUID_CORE}
-            COMMAND cmake -E copy ${openblas_refpath}/openblas.dll ${FLUID_DST_DIR}
-            DEPENDS paddle_pybind)
 ELSE()
     set(FLUID_CORE ${PADDLE_BINARY_DIR}/python/paddle/fluid/core.so)
-    add_custom_command(OUTPUT ${FLUID_CORE}
-            COMMAND cmake -E copy $<TARGET_FILE:paddle_pybind> ${FLUID_CORE}
-            DEPENDS paddle_pybind)
 ENDIF()
+add_custom_command(OUTPUT ${FLUID_CORE}
+        COMMAND cmake -E copy $<TARGET_FILE:paddle_pybind> ${FLUID_CORE}
+        DEPENDS paddle_pybind)
 add_custom_target(copy_paddle_pybind ALL DEPENDS ${FLUID_CORE})
 
 IF(WIN32)

python/setup.py.in

@@ -140,8 +140,6 @@ if '${WITH_FLUID_ONLY}'== 'OFF':
                    '${PADDLE_BINARY_DIR}/paddle/scripts/paddle']
 
 package_data={'paddle.fluid': ['core' + (ext_name if os.name != 'nt' else '.pyd')]}
-if os.name == 'nt':
-    package_data['paddle.fluid'] += ['openblas' + ext_name]
 
 if '${WITH_FLUID_ONLY}'== 'OFF':
     package_data['paddle.v2.master']=['libpaddle_master' + ext_name]
@@ -166,11 +164,17 @@ package_data['paddle.libs']=[('libwarpctc' if os.name != 'nt' else 'warpctc') +
 shutil.copy('${WARPCTC_LIBRARIES}', libs_path)
 
 if '${WITH_MKL}' == 'ON':
-    shutil.copy('${MKLML_LIB}', libs_path)
-    shutil.copy('${MKLML_IOMP_LIB}', libs_path)
-    package_data['paddle.libs']+=['libmklml_intel' + ext_name,'libiomp5' + ext_name]
+    shutil.copy('${MKLML_SHARED_LIB}', libs_path)
+    shutil.copy('${MKLML_SHARED_IOMP_LIB}', libs_path)
+    package_data['paddle.libs']+=[('libmklml_intel' if os.name != 'nt' else 'mklml') + ext_name, ('libiomp5' if os.name != 'nt' else 'libiomp5md') + ext_name]
+else:
+    if os.name == 'nt':
+        # copy the openblas.dll
+        shutil.copy(os.path.dirname('${CBLAS_LIBRARIES}') + '/openblas' + ext_name, libs_path)
+        package_data['paddle.libs'] += ['openblas' + ext_name]
+
 if '${WITH_MKLDNN}' == 'ON':
-    if '${CMAKE_BUILD_TYPE}' == 'Release':
+    if '${CMAKE_BUILD_TYPE}' == 'Release' and os.name != 'nt':
         # only change rpath in Release mode.
         # TODO(typhoonzero): use install_name_tool to patch mkl libs once
         # we can support mkl on mac.
@@ -181,7 +185,7 @@ if '${WITH_MKLDNN}' == 'ON':
         command = "patchelf --set-rpath '$ORIGIN/' ${MKLDNN_SHARED_LIB}"
         if os.system(command) != 0:
             raise Exception("patch libmkldnn.so failed, command: %s" % command)
-    package_data['paddle.libs']+=['libmkldnn.so.0']
+    package_data['paddle.libs']+=['libmkldnn.so.0' if os.name != 'nt' else ('mkldnn' + ext_name)]
     shutil.copy('${MKLDNN_SHARED_LIB}', libs_path)
 if '${WITH_NGRAPH}' == 'ON':
     # only change rpath in Release mode,