Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into dev_sequence_padding_op

Author: JiayiFeng

.gitignore

@@ -5,6 +5,7 @@ python/paddle/v2/fluid/tests/book/image_classification_resnet.inference.model/
 python/paddle/v2/fluid/tests/book/image_classification_vgg.inference.model/
 python/paddle/v2/fluid/tests/book/label_semantic_roles.inference.model/
 *.DS_Store
+*.vs
 build/
 build_doc/
 *.user
@@ -15,6 +16,7 @@ build_doc/
 .cproject
 .pydevproject
 .settings/
+CMakeSettings.json
 Makefile
 .test_env/
 third_party/

CMakeLists.txt

@@ -24,6 +24,9 @@ message(STATUS "CXX compiler: ${CMAKE_CXX_COMPILER}, version: "
         "${CMAKE_CXX_COMPILER_ID} ${CMAKE_CXX_COMPILER_VERSION}")
 message(STATUS "C compiler: ${CMAKE_C_COMPILER}, version: "
         "${CMAKE_C_COMPILER_ID} ${CMAKE_C_COMPILER_VERSION}")
+if(WIN32)
+    set(CMAKE_STATIC_LIBRARY_PREFIX lib)
+endif(WIN32)
 
 if(NOT CMAKE_CROSSCOMPILING)
     find_package(CUDA QUIET)
@@ -138,12 +141,6 @@ else()
     set(THIRD_PARTY_BUILD_TYPE Release)
 endif()
 
-if(WITH_MKL)
-  option(MKL_SPLIT_GEMM "PaddlePaddle MKL gemm would split to small ones" OFF)
-  if (MKL_SPLIT_GEMM)
-    add_definitions(-DPADDLE_MKL_SPLIT_GEMM)
-  endif()
-endif()
 set(WITH_MKLML ${WITH_MKL})
 if (NOT DEFINED WITH_MKLDNN)
     if (WITH_MKL AND AVX2_FOUND)
@@ -171,14 +168,21 @@ include(external/python)    # download, build, install python
 include(external/openblas)  # download, build, install openblas
 include(external/mkldnn)    # download, build, install mkldnn
 include(external/swig)      # download, build, install swig
-include(external/warpctc)   # download, build, install warpctc
 include(external/boost)     # download boost
 include(external/any)       # download libn::any
 include(external/eigen)     # download eigen3
 include(external/pybind11)  # download pybind11
 include(external/cares)
 include(external/cub)
 
+if (NOT WIN32)
+# there is no official support of snappystream, warpctc, nccl, cupti in windows
+include(external/snappy)    # download snappy
+include(external/snappystream) # download snappystream
+include(external/warpctc)   # download, build, install warpctc
+include(cupti)
+endif (NOT WIN32)
+
 if(WITH_DISTRIBUTE)
     if(WITH_GRPC)
         include(external/grpc)
@@ -200,26 +204,25 @@ if(WITH_BRPC_RDMA)
     endif()
 endif()
 
-include(external/snappy)    # download snappy
-include(external/snappystream)
+
 include(external/threadpool)
+include(flags)              # set paddle compile flags
+include(cudnn)              # set cudnn libraries, must before configure
+include(configure)          # add paddle env configuration
 
-set(WITH_ANAKIN OFF CACHE STRING "Disable Anakin first, will add it later." FORCE)
 if(WITH_GPU)
     include(cuda)
     include(tensorrt)
     include(external/anakin)
+elseif()
+    set(WITH_ANAKIN OFF CACHE STRING "Anakin is used in GPU only now." FORCE)
 endif()
 
-include(cudnn)              # set cudnn libraries, must before configure
-include(cupti)
-include(configure)          # add paddle env configuration
 include(generic)            # simplify cmake module
 include(package)            # set paddle packages
 include(ccache)             # set ccache for compilation
 include(util)               # set unittest and link libs
 include(rdma)               # set rdma libraries
-include(flags)              # set paddle compile flags
 include(version)            # set PADDLE_VERSION
 include(coveralls)          # set code coverage
 include(inference_lib)      # add paddle fluid inference libraries

cmake/configure.cmake

@@ -50,13 +50,22 @@ if(NOT WITH_PROFILER)
 endif(NOT WITH_PROFILER)
 
 if(NOT CMAKE_CROSSCOMPILING)
-    if(WITH_AVX AND AVX_FOUND)
+    if(WITH_AVX AND AVX512F_FOUND)
+        set(SIMD_FLAG ${AVX512F_FLAG})
+    elseif(WITH_AVX AND AVX2_FOUND)
+        set(SIMD_FLAG ${AVX2_FLAG})
+    elseif(WITH_AVX AND AVX_FOUND)
         set(SIMD_FLAG ${AVX_FLAG})
     elseif(SSE3_FOUND)
         set(SIMD_FLAG ${SSE3_FLAG})
     endif()
 endif()
 
+if(WIN32)
+  # windows stupid compile option for all targets.
+  add_definitions(-D_XKEYCHECK_H)
+endif(WIN32)
+
 if(NOT WITH_GOLANG)
     add_definitions(-DPADDLE_WITHOUT_GOLANG)
 endif(NOT WITH_GOLANG)
@@ -99,12 +108,21 @@ if(WITH_GPU)
     endif()
     if(WITH_ANAKIN)
         if(${CUDA_VERSION_MAJOR} VERSION_LESS 8)
-            message(FATAL_ERROR "Anakin needs CUDA >= 8.0 to compile")
+            message(WARNING "Anakin needs CUDA >= 8.0 to compile. Force WITH_ANAKIN=OFF")
+            set(WITH_ANAKIN OFF CACHE STRING "Anakin is valid only when CUDA >= 8.0." FORCE)
         endif()
         if(${CUDNN_MAJOR_VERSION} VERSION_LESS 7)
-            message(FATAL_ERROR "Anakin needs CUDNN >= 7.0 to compile")
+            message(WARNING "Anakin needs CUDNN >= 7.0 to compile. Force WITH_ANAKIN=OFF")
+            set(WITH_ANAKIN OFF CACHE STRING "Anakin is valid only when CUDNN >= 7.0." FORCE)
         endif()
     endif()
+    if(WITH_ANAKIN)
+        # NOTICE(minqiyang): the end slash is important because $CUDNN_INCLUDE_DIR
+        # is a softlink to real cudnn.h directory
+        set(ENV{CUDNN_INCLUDE_DIR} "${CUDNN_INCLUDE_DIR}/")
+        get_filename_component(CUDNN_LIBRARY_DIR ${CUDNN_LIBRARY} DIRECTORY)
+        set(ENV{CUDNN_LIBRARY} ${CUDNN_LIBRARY_DIR})
+    endif()
 elseif(WITH_AMD_GPU)
     add_definitions(-DPADDLE_WITH_HIP)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D__HIP_PLATFORM_HCC__")

cmake/cudnn.cmake

@@ -25,8 +25,25 @@ list(APPEND CUDNN_CHECK_LIBRARY_DIRS
     $ENV{CUDNN_ROOT}
     $ENV{CUDNN_ROOT}/lib64
     $ENV{CUDNN_ROOT}/lib
-    /usr/lib)
-find_library(CUDNN_LIBRARY NAMES libcudnn.so libcudnn.dylib # libcudnn_static.a
+    /usr/lib
+	${CUDA_TOOLKIT_ROOT_DIR}
+	${CUDA_TOOLKIT_ROOT_DIR}/lib/x64
+	)
+set(CUDNN_LIB_NAME "")
+if (LINUX)
+set(CUDNN_LIB_NAME "libcudnn.so")
+endif(LINUX)
+
+if(WIN32)
+# only support cudnn7
+set(CUDNN_LIB_NAME "cudnn.lib" "cudnn64_7.dll")
+endif(WIN32)
+
+if(Apple)
+set(CUDNN_LIB_NAME "libcudnn.dylib" "libcudnn.so")
+endif(Apple)
+
+find_library(CUDNN_LIBRARY NAMES ${CUDNN_LIB_NAME} # libcudnn_static.a
     PATHS ${CUDNN_CHECK_LIBRARY_DIRS} ${CUDNN_INCLUDE_DIR} ${__libpath_hist}
           NO_DEFAULT_PATH
     DOC "Path to cuDNN library.")

cmake/external/anakin.cmake

@@ -2,6 +2,11 @@ if (NOT WITH_ANAKIN)
   return()
 endif()
 
+option(ANAKIN_ENABLE_OP_TIMER      "Get more detailed information with Anakin op time"        OFF)
+if(ANAKIN_ENABLE_OP_TIMER)
+  add_definitions(-DPADDLE_ANAKIN_ENABLE_OP_TIMER)
+endif()
+
 INCLUDE(ExternalProject)
 set(ANAKIN_SOURCE_DIR  ${THIRD_PARTY_PATH}/anakin)
 # the anakin install dir is only default one now
@@ -11,33 +16,45 @@ set(ANAKIN_LIBRARY     ${ANAKIN_INSTALL_DIR})
 set(ANAKIN_SHARED_LIB  ${ANAKIN_LIBRARY}/libanakin.so)
 set(ANAKIN_SABER_LIB   ${ANAKIN_LIBRARY}/libanakin_saber_common.so)
 
-# TODO(luotao): ANAKIN_MODLE_URL will move to demo ci later.
-set(ANAKIN_MODLE_URL "http://paddle-inference-dist.bj.bcebos.com/mobilenet_v2.anakin.bin")
+# TODO(luotao): ANAKIN_MODLE_URL etc will move to demo ci later.
+set(INFERENCE_URL "http://paddle-inference-dist.bj.bcebos.com")
+set(ANAKIN_MODLE_URL "${INFERENCE_URL}/mobilenet_v2.anakin.bin")
+set(ANAKIN_RNN_MODLE_URL "${INFERENCE_URL}/anakin_test%2Fditu_rnn.anakin2.model.bin")
+set(ANAKIN_RNN_DATA_URL "${INFERENCE_URL}/anakin_test%2Fditu_rnn_data.txt")
 execute_process(COMMAND bash -c "mkdir -p ${ANAKIN_SOURCE_DIR}")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_MODLE_URL}")
+execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_MODLE_URL} -N")
+execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_MODLE_URL} -N")
+execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_DATA_URL} -N")
 
 include_directories(${ANAKIN_INCLUDE})
 include_directories(${ANAKIN_INCLUDE}/saber/)
+include_directories(${ANAKIN_INCLUDE}/saber/core/)
+include_directories(${ANAKIN_INCLUDE}/saber/funcs/impl/x86/)
+include_directories(${ANAKIN_INCLUDE}/saber/funcs/impl/cuda/base/cuda_c/)
 
-set(ANAKIN_COMPILE_EXTRA_FLAGS 
+set(ANAKIN_COMPILE_EXTRA_FLAGS
     -Wno-error=unused-but-set-variable -Wno-unused-but-set-variable
-    -Wno-error=unused-variable -Wno-unused-variable 
+    -Wno-error=unused-variable -Wno-unused-variable
     -Wno-error=format-extra-args -Wno-format-extra-args
     -Wno-error=comment -Wno-comment 
     -Wno-error=format -Wno-format 
+    -Wno-error=maybe-uninitialized -Wno-maybe-uninitialized
     -Wno-error=switch -Wno-switch
-    -Wno-error=return-type -Wno-return-type 
+    -Wno-error=return-type -Wno-return-type
     -Wno-error=non-virtual-dtor -Wno-non-virtual-dtor
+    -Wno-error=ignored-qualifiers
+    -Wno-ignored-qualifiers
     -Wno-sign-compare
-    -Wno-reorder 
+    -Wno-reorder
     -Wno-error=cpp)
 
 ExternalProject_Add(
     extern_anakin
     ${EXTERNAL_PROJECT_LOG_ARGS}
-    # TODO(luotao): use PaddlePaddle/Anakin later
+    DEPENDS             ${MKLML_PROJECT}
+    # Anakin codes error on Intel(R) Xeon(R) Gold 5117 CPU, temporary do not compile avx512 related code.
     GIT_REPOSITORY      "https://github.com/luotao1/Anakin"
-    GIT_TAG             "3957ae9263eaa0b1986758dac60a88852afb09be"
+    GIT_TAG             "211d1fc5d813d70c0c14072f9083cf25f40940ea"
     PREFIX              ${ANAKIN_SOURCE_DIR}
     UPDATE_COMMAND      ""
     CMAKE_ARGS          -DUSE_GPU_PLACE=YES
@@ -46,6 +63,8 @@ ExternalProject_Add(
                         -DPROTOBUF_ROOT=${THIRD_PARTY_PATH}/install/protobuf
                         -DMKLML_ROOT=${THIRD_PARTY_PATH}/install/mklml
                         -DCUDNN_ROOT=${CUDNN_ROOT}
+                        -DCUDNN_INCLUDE_DIR=${CUDNN_INCLUDE_DIR}
+                        -DENABLE_OP_TIMER=${ANAKIN_ENABLE_OP_TIMER}
                         ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS    -DCMAKE_INSTALL_PREFIX:PATH=${ANAKIN_INSTALL_DIR}
 )

cmake/external/boost.cmake

@@ -28,29 +28,36 @@ 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()
-MESSAGE(STATUS "BOOST_TAR: ${BOOST_TAR}, BOOST_URL: ${BOOST_URL}")
+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)
+
 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)
 
 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
+    && tar zxf ${BOOST_TAR}.tar.gz
     DOWNLOAD_NO_PROGRESS  1
     PREFIX                ${BOOST_SOURCES_DIR}
     CONFIGURE_COMMAND     ""
     BUILD_COMMAND         ""
     INSTALL_COMMAND       ""
     UPDATE_COMMAND        ""
 )
+endif(NOT WIN32)
 
-if (${CMAKE_VERSION} VERSION_LESS "3.3.0")
+if (${CMAKE_VERSION} VERSION_LESS "3.3.0" OR NOT WIN32)
     set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/boost_dummy.c)
     file(WRITE ${dummyfile} "const char *dummy = \"${dummyfile}\";")
     add_library(boost STATIC ${dummyfile})

cmake/external/gflags.cmake

@@ -18,7 +18,7 @@ SET(GFLAGS_SOURCES_DIR ${THIRD_PARTY_PATH}/gflags)
 SET(GFLAGS_INSTALL_DIR ${THIRD_PARTY_PATH}/install/gflags)
 SET(GFLAGS_INCLUDE_DIR "${GFLAGS_INSTALL_DIR}/include" CACHE PATH "gflags include directory." FORCE)
 IF(WIN32)
-  set(GFLAGS_LIBRARIES "${GFLAGS_INSTALL_DIR}/lib/gflags.lib" CACHE FILEPATH "GFLAGS_LIBRARIES" FORCE)
+  set(GFLAGS_LIBRARIES "${GFLAGS_INSTALL_DIR}/lib/libgflags.lib" CACHE FILEPATH "GFLAGS_LIBRARIES" FORCE)
 ELSE(WIN32)
   set(GFLAGS_LIBRARIES "${GFLAGS_INSTALL_DIR}/lib/libgflags.a" CACHE FILEPATH "GFLAGS_LIBRARIES" FORCE)
 ENDIF(WIN32)
@@ -45,7 +45,13 @@ ExternalProject_Add(
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
 )
-
+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
+  )
+  ENDIF()
+ENDIF(WIN32)
 ADD_LIBRARY(gflags STATIC IMPORTED GLOBAL)
 SET_PROPERTY(TARGET gflags PROPERTY IMPORTED_LOCATION ${GFLAGS_LIBRARIES})
 ADD_DEPENDENCIES(gflags extern_gflags)
@@ -60,3 +66,4 @@ IF(WITH_C_API)
     INSTALL(FILES ${GFLAGS_LIBRARIES} DESTINATION third_party/gflags/lib)
   ENDIF()
 ENDIF()
+

cmake/external/glog.cmake

@@ -60,6 +60,13 @@ ExternalProject_Add(
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
 )
+IF(WIN32)
+  IF(NOT EXISTS "${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
+  )
+  ENDIF()
+ENDIF(WIN32)
 
 ADD_LIBRARY(glog STATIC IMPORTED GLOBAL)
 SET_PROPERTY(TARGET glog PROPERTY IMPORTED_LOCATION ${GLOG_LIBRARIES})

cmake/external/mkldnn.cmake

@@ -54,7 +54,7 @@ ExternalProject_Add(
     ${EXTERNAL_PROJECT_LOG_ARGS}
     DEPENDS             ${MKLDNN_DEPENDS}
     GIT_REPOSITORY      "https://github.com/01org/mkl-dnn.git"
-    GIT_TAG             "a29d8487a63afca3d5b8c5bbdbb473cf8ccc6e51"
+    GIT_TAG             "64e03a1939e0d526aa8e9f2e3f7dc0ad8d372944"
     PREFIX              ${MKLDNN_SOURCES_DIR}
     UPDATE_COMMAND      ""
     CMAKE_ARGS          -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}