Revert "fix windows for optimization of elementwise_add Op (#31068)"

This reverts commit 364cfa2686.

CMakeLists.txt

@@ -12,8 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License
 
-cmake_minimum_required(VERSION 3.15)
-cmake_policy(VERSION 3.10)
+cmake_minimum_required(VERSION 3.10)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
 set(PADDLE_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR})
 set(PADDLE_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR})
@@ -39,6 +38,11 @@ endif()
 if (WITH_GPU  AND WITH_ASCEND)
     message(FATAL_ERROR "Error when compile GPU and ASCEND at the same time")
 endif()
+# cmake 3.12, 3.13, 3.14 will append gcc link options to nvcc, and nvcc doesn't recognize them.
+if(WITH_GPU AND (${CMAKE_VERSION} VERSION_GREATER_EQUAL 3.12) AND (${CMAKE_VERSION} VERSION_LESS 3.15))
+    message(FATAL_ERROR "cmake ${CMAKE_VERSION} is not supported when WITH_GPU=ON because of bug https://cmake.org/pipermail/cmake/2018-September/068195.html. "
+       "You can use cmake 3.16 (recommended), 3.10, 3.11, 3.15 or 3.17. Please refer to the install document: https://cmake.org/install/")
+endif()
 
 if(WITH_GPU AND NOT APPLE)
     enable_language(CUDA)
@@ -57,7 +61,6 @@ if(WITH_MUSL)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-error=deprecated-declarations -Wno-deprecated-declarations -Wno-error=pessimizing-move -Wno-error=deprecated-copy")
 endif()
 
-
 if(WIN32)
     option(MSVC_STATIC_CRT "use static C Runtime library by default" ON)
 
@@ -69,13 +72,6 @@ if(WIN32)
     set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} /bigobj")
     set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS_RELEASE} /bigobj")
 
-    if("${CMAKE_GENERATOR}" STREQUAL "Ninja")
-        set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /Zc:inline")
-        set(CMAKE_C_FLAGS_RELEASE  "${CMAKE_C_FLAGS_RELEASE} /Zc:inline")
-        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} /Zc:inline")
-        set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS_RELEASE} /Zc:inline")
-    endif()
-
     if (MSVC_STATIC_CRT)
         message(STATUS "Use static C runtime time, refer to https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-library-features?view=vs-2019")
         set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /MTd")
@@ -92,7 +88,7 @@ if(WIN32)
             endif()
         endforeach(flag_var)
     endif()
-    
+
     # NOTE(Avin0323): Less parallel count result in faster compilation.
     math(EXPR PROCESS_MAX "${CPU_CORES} * 2 / 3")
     # windows build turn off warnings, use parallel compiling.
@@ -120,10 +116,6 @@ if(WIN32)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
 
-    foreach(flag_var CMAKE_SHARED_LINKER_FLAGS CMAKE_STATIC_LINKER_FLAGS CMAKE_EXE_LINKER_FLAGS CMAKE_LINKER_FLAGS)
-        set(${flag_var} "${${flag_var}} /ignore:4049 /ignore:4217 /ignore:4006 /ignore:4221")
-    endforeach(flag_var)
-
     if (WITH_WIN_DUMP_DBG)
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /Zi")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Zi")

README.md

@@ -36,7 +36,7 @@ pip install paddlepaddle-gpu
 ```
 More infomation about installation, please view [Quick Install](https://www.paddlepaddle.org.cn/install/quick)
 
-Now our developers can acquire Tesla V100 online computing resources for free. If you create a program by AI Studio, you will obtain 10 hours to train models online per day. [Click here to start](https://aistudio.baidu.com/aistudio/index).
+Now our developers can acquire Tesla V100 online computing resources for free. If you create a program by AI Studio, you will obtain 12 hours to train models online per day. If you can insist on that for five consecutive days, then you will receive an extra 48 hours. [Click here to start](https://ai.baidu.com/support/news?action=detail&id=981).
 
 ## FOUR LEADING TECHNOLOGIES
 

README_cn.md

@@ -32,7 +32,7 @@ pip install paddlepaddle-gpu
 ```
 更多安装信息详见官网 [安装说明](https://www.paddlepaddle.org.cn/install/quick)
 
-PaddlePaddle用户可领取**免费Tesla V100在线算力资源**，训练模型更高效。**每日登陆即送10小时**，[前往使用免费算力](https://aistudio.baidu.com/aistudio/index)。
+PaddlePaddle用户可领取**免费Tesla V100在线算力资源**，训练模型更高效。**每日登陆即送12小时**，**连续五天运行再加送48小时**，[前往使用免费算力](https://ai.baidu.com/support/news?action=detail&id=981)。
 
 ## 四大领先技术
 

cmake/cuda.cmake

@@ -74,7 +74,7 @@ endfunction()
 #   select_nvcc_arch_flags(out_variable)
 function(select_nvcc_arch_flags out_variable)
   # List of arch names
-  set(archs_names "Kepler" "Maxwell" "Pascal" "Volta" "Turing" "Ampere" "All" "Manual")
+  set(archs_names "Kepler" "Maxwell" "Pascal" "Volta" "Turing" "All" "Manual")
   set(archs_name_default "Auto")
   list(APPEND archs_names "Auto")
 
@@ -91,7 +91,7 @@ function(select_nvcc_arch_flags out_variable)
 
   if(${CUDA_ARCH_NAME} STREQUAL "Manual")
     set(CUDA_ARCH_BIN ${paddle_known_gpu_archs} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
-    set(CUDA_ARCH_PTX ""                        CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
+    set(CUDA_ARCH_PTX "50"                     CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
     mark_as_advanced(CUDA_ARCH_BIN CUDA_ARCH_PTX)
   else()
     unset(CUDA_ARCH_BIN CACHE)
@@ -108,8 +108,6 @@ function(select_nvcc_arch_flags out_variable)
     set(cuda_arch_bin "70")
   elseif(${CUDA_ARCH_NAME} STREQUAL "Turing")
     set(cuda_arch_bin "75")
-  elseif(${CUDA_ARCH_NAME} STREQUAL "Ampere")
-    set(cuda_arch_bin "80")
   elseif(${CUDA_ARCH_NAME} STREQUAL "All")
     set(cuda_arch_bin ${paddle_known_gpu_archs})
   elseif(${CUDA_ARCH_NAME} STREQUAL "Auto")
@@ -177,22 +175,14 @@ elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 10.0) # CUDA 9.x
   set(paddle_known_gpu_archs ${paddle_known_gpu_archs9})
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0) # CUDA 10.x
   set(paddle_known_gpu_archs ${paddle_known_gpu_archs10})
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 11.2) # CUDA 11.0/11.1
+elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0) # CUDA 11.x
   set(paddle_known_gpu_archs ${paddle_known_gpu_archs11})
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0) # CUDA 11.2+
-  set(paddle_known_gpu_archs "${paddle_known_gpu_archs11} 86")
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 endif()
 
 if (NOT ${CMAKE_CUDA_COMPILER_VERSION} LESS 10.0)
@@ -208,11 +198,14 @@ select_nvcc_arch_flags(NVCC_FLAGS_EXTRA)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} ${NVCC_FLAGS_EXTRA}")
 message(STATUS "NVCC_FLAGS_EXTRA: ${NVCC_FLAGS_EXTRA}")
 
-# Set C++14 support
+# Set C++11 support
 set(CUDA_PROPAGATE_HOST_FLAGS OFF)
 # Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
 # So, don't set these flags here.
-set(CMAKE_CUDA_STANDARD 14)
+if (NOT WIN32) # windows msvc2015 support c++11 natively.
+    # -std=c++11 -fPIC not recoginize by msvc, -Xcompiler will be added by cmake.
+  set(CMAKE_CUDA_STANDARD 11)
+endif(NOT WIN32)
 
 # (Note) For windows, if delete /W[1-4], /W1 will be added defaultly and conflic with -w
 # So replace /W[1-4] with /W0

cmake/cudnn.cmake

@@ -94,7 +94,7 @@ macro(find_cudnn_version cudnn_header_file)
                 "${CUDNN_MAJOR_VERSION} * 1000 +
                  ${CUDNN_MINOR_VERSION} * 100 + ${CUDNN_PATCHLEVEL_VERSION}")
             message(STATUS "Current cuDNN header is ${cudnn_header_file} "
-              "Current cuDNN version is v${CUDNN_MAJOR_VERSION}.${CUDNN_MINOR_VERSION}.${CUDNN_PATCHLEVEL_VERSION}. ")
+              "Current cuDNN version is v${CUDNN_MAJOR_VERSION}.${CUDNN_MINOR_VERSION}. ")
         endif()
     endif()
 endmacro()

cmake/external/cryptopp.cmake

@@ -22,13 +22,6 @@ SET(CRYPTOPP_TAG        CRYPTOPP_8_2_0)
 
 IF(WIN32)
   SET(CRYPTOPP_LIBRARIES "${CRYPTOPP_INSTALL_DIR}/lib/cryptopp-static.lib" CACHE FILEPATH "cryptopp library." FORCE)
-  # There is a compilation parameter 'FI\"winapifamily.h\"' can't be used correctly
-  # with Ninja on Windows. The only difference between the patch file and original
-  # file is that the compilation parameters are changed to 'FIwinapifamily.h'. This
-  # patch command can be removed when upgrading to a higher version.
-  if("${CMAKE_GENERATOR}" STREQUAL "Ninja")
-    set(CRYPTOPP_PATCH_COMMAND ${CMAKE_COMMAND} -E copy_if_different "${PADDLE_SOURCE_DIR}/patches/cryptopp/CMakeLists.txt" "<SOURCE_DIR>/")
-  endif()
 ELSE(WIN32)
   SET(CRYPTOPP_LIBRARIES "${CRYPTOPP_INSTALL_DIR}/lib/libcryptopp.a" CACHE FILEPATH "cryptopp library." FORCE)
 ENDIF(WIN32)
@@ -60,13 +53,11 @@ ExternalProject_Add(
     "${CRYPTOPP_DOWNLOAD_CMD}"
     PREFIX          ${CRYPTOPP_PREFIX_DIR}
     SOURCE_DIR      ${CRYPTOPP_SOURCE_DIR}
-    UPDATE_COMMAND  ""
     PATCH_COMMAND
     COMMAND ${CMAKE_COMMAND} -E remove_directory "<SOURCE_DIR>/cmake/"
     COMMAND git clone ${GIT_URL}/noloader/cryptopp-cmake "<SOURCE_DIR>/cmake"
     COMMAND cd "<SOURCE_DIR>/cmake" && git checkout tags/${CRYPTOPP_TAG} -b ${CRYPTOPP_TAG}
     COMMAND ${CMAKE_COMMAND} -E copy_directory "<SOURCE_DIR>/cmake/" "<SOURCE_DIR>/"
-    COMMAND ${CRYPTOPP_PATCH_COMMAND}
     INSTALL_DIR     ${CRYPTOPP_INSTALL_DIR}
     CMAKE_ARGS ${CRYPTOPP_CMAKE_ARGS}
     CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${CRYPTOPP_INSTALL_DIR}

cmake/external/mkldnn.cmake

@@ -20,7 +20,7 @@ SET(MKLDNN_SOURCE_DIR     ${THIRD_PARTY_PATH}/mkldnn/src/extern_mkldnn)
 SET(MKLDNN_INSTALL_DIR    ${THIRD_PARTY_PATH}/install/mkldnn)
 SET(MKLDNN_INC_DIR        "${MKLDNN_INSTALL_DIR}/include" CACHE PATH "mkldnn include directory." FORCE)
 SET(MKLDNN_REPOSITORY     ${GIT_URL}/oneapi-src/oneDNN.git)
-SET(MKLDNN_TAG            72efa005effb49595933e033cc732f215ef0445a)
+SET(MKLDNN_TAG            3d53cd3f17ce7ca365c980f0e1e50359751ca038)
 
 # Introduce variables:
 # * CMAKE_INSTALL_LIBDIR

cmake/external/protobuf.cmake

@@ -250,8 +250,5 @@ IF(NOT PROTOBUF_FOUND)
 
     SET(PROTOBUF_PROTOC_EXECUTABLE ${extern_protobuf_PROTOC_EXECUTABLE}
         CACHE FILEPATH "protobuf executable." FORCE)
-    # `EXTERN_PROTOBUF_DEPEND` used in cmake function `proto_library` to ensure
-    # `protoc.exe` existed before calling it.
-    set(EXTERN_PROTOBUF_DEPEND extern_protobuf)
     PROMPT_PROTOBUF_LIB(extern_protobuf)
 ENDIF(NOT PROTOBUF_FOUND)

cmake/external/warpctc.cmake

@@ -14,15 +14,11 @@
 
 INCLUDE(ExternalProject)
 
-IF(WITH_ROCM)
-    add_definitions(-DWARPCTC_WITH_HIP)
-ENDIF()
-
 SET(WARPCTC_PREFIX_DIR  ${THIRD_PARTY_PATH}/warpctc)
 SET(WARPCTC_SOURCE_DIR  ${THIRD_PARTY_PATH}/warpctc/src/extern_warpctc)
 SET(WARPCTC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/warpctc)
 set(WARPCTC_REPOSITORY  ${GIT_URL}/baidu-research/warp-ctc.git)
-set(WARPCTC_TAG         c690fc5755abbdbdc98ef78d51ec10a6748a8cd1)
+set(WARPCTC_TAG         95a461eddeabd51099ef059dcfada1117eb1bfb8)
 
 SET(WARPCTC_INCLUDE_DIR "${WARPCTC_INSTALL_DIR}/include"
     CACHE PATH "Warp-ctc Directory" FORCE)
@@ -53,15 +49,14 @@ ExternalProject_Add(
     BUILD_ALWAYS    1
     CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
                     -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-                    -DCMAKE_C_FLAGS=$<FILTER:${CMAKE_C_FLAGS},EXCLUDE,/Zc:inline>
-                    -DCMAKE_C_FLAGS_DEBUG=$<FILTER:${CMAKE_C_FLAGS_DEBUG},EXCLUDE,/Zc:inline>
-                    -DCMAKE_C_FLAGS_RELEASE=$<FILTER:${CMAKE_C_FLAGS_RELEASE},EXCLUDE,/Zc:inline>
-                    -DCMAKE_CXX_FLAGS=$<FILTER:${CMAKE_CXX_FLAGS},EXCLUDE,/Zc:inline>
-                    -DCMAKE_CXX_FLAGS_RELEASE=$<FILTER:${CMAKE_CXX_FLAGS_RELEASE},EXCLUDE,/Zc:inline>
-                    -DCMAKE_CXX_FLAGS_DEBUG=$<FILTER:${CMAKE_CXX_FLAGS_DEBUG},EXCLUDE,/Zc:inline>
+                    -DCMAKE_C_FLAGS=${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=${CMAKE_CXX_FLAGS_DEBUG}
                     -DCMAKE_INSTALL_PREFIX=${WARPCTC_INSTALL_DIR}
                     -DWITH_GPU=${WITH_GPU}
-                    -DWITH_ROCM=${WITH_ROCM}
                     -DWITH_OMP=${USE_OMP}
                     -DWITH_TORCH=OFF
                     -DCMAKE_DISABLE_FIND_PACKAGE_Torch=ON

cmake/external/xpu.cmake

@@ -4,8 +4,6 @@ endif()
 
 INCLUDE(ExternalProject)
 SET(XPU_PROJECT                 "extern_xpu")
-SET(XPU_API_LIB_NAME            "libxpuapi.so")
-SET(XPU_RT_LIB_NAME             "libxpurt.so")
 
 if(NOT XPU_SDK_ROOT)
   if (WITH_AARCH64)
@@ -13,7 +11,7 @@ if(NOT XPU_SDK_ROOT)
   elseif(WITH_SUNWAY)
       SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/sunway/xpu_2021_01_13.tar.gz" CACHE STRING "" FORCE)
   else()
-      SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2021_03_30.tar.gz" CACHE STRING "" FORCE)
+      SET(XPU_URL    "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/xpu_2021_02_19.tar.gz" CACHE STRING "" FORCE)
   endif()
 
   SET(XPU_SOURCE_DIR              "${THIRD_PARTY_PATH}/xpu")
@@ -22,6 +20,8 @@ if(NOT XPU_SDK_ROOT)
   SET(XPU_API_INC_DIR             "${THIRD_PARTY_PATH}/install/xpu/include")
   SET(XPU_LIB_DIR                 "${THIRD_PARTY_PATH}/install/xpu/lib")
 
+  SET(XPU_API_LIB_NAME            "libxpuapi.so")
+  SET(XPU_RT_LIB_NAME             "libxpurt.so")
   SET(XPU_API_LIB                 "${XPU_LIB_DIR}/${XPU_API_LIB_NAME}")
   SET(XPU_RT_LIB                  "${XPU_LIB_DIR}/${XPU_RT_LIB_NAME}")
 
@@ -49,7 +49,6 @@ else()
   SET(XPU_API_INC_DIR   "${XPU_SDK_ROOT}/XTDK/include/")
   SET(XPU_API_LIB "${XPU_SDK_ROOT}/XTDK/shlib/libxpuapi.so")
   SET(XPU_RT_LIB "${XPU_SDK_ROOT}/XTDK/runtime/shlib/libxpurt.so")
-  SET(XPU_LIB_DIR "${XPU_SDK_ROOT}/XTDK/shlib/")
 endif()
 
 INCLUDE_DIRECTORIES(${XPU_API_INC_DIR})

cmake/flags.cmake

@@ -4,10 +4,10 @@ include(CheckCCompilerFlag)
 include(CheckCXXSymbolExists)
 include(CheckTypeSize)
 
-function(CheckCompilerCXX14Flag)
+function(CheckCompilerCXX11Flag)
     if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-        if(${CMAKE_CXX_COMPILER_VERSION} VERSION_LESS 5.4)
-            message(FATAL_ERROR "Unsupported GCC version. GCC >= 5.4 required.")
+        if(${CMAKE_CXX_COMPILER_VERSION} VERSION_LESS 4.8)
+            message(FATAL_ERROR "Unsupported GCC version. GCC >= 4.8 required.")
         elseif(${CMAKE_CXX_COMPILER_VERSION} VERSION_GREATER 8.2)
             message(WARNING "Found GCC ${CMAKE_CXX_COMPILER_VERSION} which is too high, recommended to use GCC 8.2")
         endif()
@@ -20,15 +20,23 @@ function(CheckCompilerCXX14Flag)
                 message(FATAL_ERROR "Unsupported AppleClang version. AppleClang >= 5.1 required.")
             endif()
         else()
-            if (${CMAKE_CXX_COMPILER_VERSION} VERSION_LESS 3.4)
-                message(FATAL_ERROR "Unsupported Clang version. Clang >= 3.4 required.")
+            if (${CMAKE_CXX_COMPILER_VERSION} VERSION_LESS 3.3)
+                message(FATAL_ERROR "Unsupported Clang version. Clang >= 3.3 required.")
             endif()
         endif()
     endif()
 endfunction()
 
-CheckCompilerCXX14Flag()
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
+CheckCompilerCXX11Flag()
+if (WITH_GPU)
+    if (${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.0)
+       set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
+    else()
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+    endif()
+else()
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+endif()
 # safe_set_flag
 #
 # Set a compile flag only if compiler is support

cmake/generic.cmake

@@ -260,8 +260,8 @@ function(merge_static_libs TARGET_NAME)
     # msvc will put libarary in directory of "/Release/xxxlib" by default
     #       COMMAND cmake -E remove "${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_BUILD_TYPE}/${TARGET_NAME}.lib"
     add_custom_command(TARGET ${TARGET_NAME} POST_BUILD
-      COMMAND cmake -E make_directory $<TARGET_FILE_DIR:${TARGET_NAME}>
-      COMMAND lib /OUT:$<TARGET_FILE:${TARGET_NAME}> ${libfiles}
+      COMMAND cmake -E make_directory "${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_BUILD_TYPE}"
+      COMMAND lib /OUT:${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_BUILD_TYPE}/lib${TARGET_NAME}.lib ${libfiles}
       )
   endif(WIN32)
 endfunction(merge_static_libs)
@@ -492,7 +492,7 @@ function(nv_library TARGET_NAME)
         message(FATAL "Please specify source file or library in nv_library.")
       endif()
     endif(nv_library_SRCS)
-    if((CUDA_VERSION GREATER 9.2) AND (CUDA_VERSION LESS 11.0) AND (MSVC_VERSION LESS 1910))
+    if (WIN32 AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
       set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
     endif()
   endif()
@@ -510,7 +510,7 @@ function(nv_binary TARGET_NAME)
       add_dependencies(${TARGET_NAME} ${nv_binary_DEPS})
       common_link(${TARGET_NAME})
     endif()
-    if((CUDA_VERSION GREATER 9.2) AND (CUDA_VERSION LESS 11.0) AND (MSVC_VERSION LESS 1910))
+    if (WIN32 AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
       set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
     endif()
   endif()
@@ -537,7 +537,7 @@ function(nv_test TARGET_NAME)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cpu_deterministic=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_init_allocated_mem=true)
     set_property(TEST ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
-    if((CUDA_VERSION GREATER 9.2) AND (CUDA_VERSION LESS 11.0) AND (MSVC_VERSION LESS 1910))
+    if (WIN32 AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
       set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
     endif()
   endif()
@@ -753,8 +753,7 @@ function(paddle_protobuf_generate_cpp SRCS HDRS)
       COMMAND ${PROTOBUF_PROTOC_EXECUTABLE}
       -I${CMAKE_CURRENT_SOURCE_DIR}
       --cpp_out "${CMAKE_CURRENT_BINARY_DIR}" ${ABS_FIL}
-      # Set `EXTERN_PROTOBUF_DEPEND` only if need to compile `protoc.exe`.
-      DEPENDS ${ABS_FIL} ${EXTERN_PROTOBUF_DEPEND}
+      DEPENDS ${ABS_FIL} protoc
       COMMENT "Running C++ protocol buffer compiler on ${FIL}"
       VERBATIM )
   endforeach()
@@ -795,8 +794,7 @@ function(py_test TARGET_NAME)
     if(WITH_COVERAGE AND NOT (WITH_INCREMENTAL_COVERAGE AND "$ENV{PADDLE_GIT_DIFF_PY_FILE}" STREQUAL ""))
       add_test(NAME ${TARGET_NAME}
               COMMAND ${CMAKE_COMMAND} -E env FLAGS_init_allocated_mem=true FLAGS_cudnn_deterministic=true
-              FLAGS_cpu_deterministic=true
-              PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
+              FLAGS_cpu_deterministic=true ${py_test_ENVS}
               COVERAGE_FILE=${PADDLE_BINARY_DIR}/python-coverage.data
               ${PYTHON_EXECUTABLE} -m coverage run --branch -p ${py_test_SRCS} ${py_test_ARGS}
               WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})

cmake/hip.cmake

@@ -45,7 +45,6 @@ set(THRUST_DEVICE_SYSTEM THRUST_DEVICE_SYSTEM_HIP)
 # define HIP_CXX_FLAGS
 list(APPEND HIP_CXX_FLAGS -fPIC)
 list(APPEND HIP_CXX_FLAGS -D__HIP_PLATFORM_HCC__=1)
-# Note(qili93): HIP has compile conflicts of float16.h as platform::float16 overload std::is_floating_point and std::is_integer
 list(APPEND HIP_CXX_FLAGS -D__HIP_NO_HALF_CONVERSIONS__=1)
 list(APPEND HIP_CXX_FLAGS -Wno-macro-redefined)
 list(APPEND HIP_CXX_FLAGS -Wno-inconsistent-missing-override)

cmake/inference_lib.cmake

@@ -164,11 +164,11 @@ copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_INSTALL_DIR})
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
 if(WIN32)
     if(WITH_STATIC_LIB)
-        set(paddle_inference_lib $<TARGET_FILE_DIR:paddle_inference>/libpaddle_inference.lib
-                             $<TARGET_FILE_DIR:paddle_inference>/paddle_inference.*)
+        set(paddle_inference_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/libpaddle_inference.lib
+                             ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_inference.*)
     else()
-        set(paddle_inference_lib $<TARGET_FILE_DIR:paddle_inference_shared>/paddle_inference.dll
-                             $<TARGET_FILE_DIR:paddle_inference_shared>/paddle_inference.lib)
+        set(paddle_inference_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_inference.dll
+                             ${PADDLE_BINARY_DIR}/paddle/fluid/inference/${CMAKE_BUILD_TYPE}/paddle_inference.lib)
     endif()
     copy(inference_lib_dist
             SRCS  ${src_dir}/inference/api/paddle_*.h ${paddle_inference_lib}
@@ -189,19 +189,6 @@ copy(inference_lib_dist
         DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/crypto/)
 include_directories(${CMAKE_BINARY_DIR}/../paddle/fluid/framework/io)
 
-copy(inference_lib_dist
-        SRCS  ${PADDLE_SOURCE_DIR}/paddle/fluid/extension/include/*
-        DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/)
-copy(inference_lib_dist
-        SRCS  ${PADDLE_SOURCE_DIR}/paddle/fluid/platform/complex64.h
-        DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/)
-copy(inference_lib_dist
-        SRCS  ${PADDLE_SOURCE_DIR}/paddle/fluid/platform/complex128.h
-        DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/)
-copy(inference_lib_dist
-        SRCS  ${PADDLE_SOURCE_DIR}/paddle/fluid/platform/float16.h
-        DSTS  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/)
-
 # CAPI inference library for only inference
 set(PADDLE_INFERENCE_C_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_inference_c_install_dir" CACHE STRING
 "A path setting CAPI paddle inference shared")
@@ -209,7 +196,7 @@ copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_C_INSTALL_DIR})
 
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
 if(WIN32)
-  set(paddle_inference_c_lib $<TARGET_FILE_DIR:paddle_inference_c>/paddle_inference_c.*)
+  set(paddle_inference_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/${CMAKE_BUILD_TYPE}/paddle_inference_c.*)
 else(WIN32)
   set(paddle_inference_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_inference_c.*)
 endif(WIN32)

cmake/init.cmake

@@ -18,10 +18,6 @@ if(NOT WIN32)
     set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
     set(CMAKE_CXX_FLAGS_MINSIZEREL "-Os -DNDEBUG")
 else()
-    # It has not been used now, it can specify CUDA compile flag manualy,
-    # its use is to remvoe /Zi to reduce GPU static library size. But it's dangerous
-    # because CUDA will update by nvidia, then error will occur.
-    # Now, it's used in CUDA:[10.0, 10.2]
     set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
 endif()
 

cmake/paddle_win.props

@@ -15,7 +15,7 @@
             <Warning>InheritFromHost</Warning>
 
             <BaseCommandLineTemplate>-ccbin "%(VCBinDir)" -x cu [GenerateRelocatableDeviceCode] [Include] [RequiredIncludes] [InterleaveSourceInPTX] [GPUDebugInfo] [GenerateLineInfo] [Keep] [KeepDir] [MaxRegCount] [PtxAsOptionV] [TargetMachinePlatform] [NvccCompilation] [CudaRuntime] [AdditionalOptions]</BaseCommandLineTemplate>
-            <BuildCommandLineTemplate>--use-local-env $(CudaClVersion)</BuildCommandLineTemplate>
+            <BuildCommandLineTemplate>--use-local-env --cl-version $(CudaClVersion)</BuildCommandLineTemplate>
             <BuildDynamicCommandLineTemplate>[CodeGeneration]</BuildDynamicCommandLineTemplate>
             <CleanCommandLineTemplate>-clean</CleanCommandLineTemplate>
             <!-- <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] $(CudaForceSynchronousPdbWrites) /Zi [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate> -->

cmake/third_party.cmake

@@ -222,7 +222,7 @@ if(WITH_MKLDNN)
 endif()
 
 include(external/protobuf)  	# find first, then download, build, install protobuf
-if(TARGET extern_protobuf)
+if(NOT PROTOBUF_FOUND OR WIN32)
     list(APPEND third_party_deps extern_protobuf)
 endif()
 
@@ -311,14 +311,11 @@ if(WITH_DGC)
 endif()
 
 if (WITH_LITE)
-    message(STATUS "Compile Paddle with Lite Engine.")
     include(external/lite)
 endif (WITH_LITE)
 
 if (WITH_CRYPTO)
     include(external/cryptopp)   # download, build, install cryptopp
-    list(APPEND third_party_deps extern_cryptopp)
-    add_definitions(-DPADDLE_WITH_CRYPTO)
 endif (WITH_CRYPTO)
 
 add_custom_target(third_party ALL DEPENDS ${third_party_deps})

go/README_cn.md

@@ -50,7 +50,6 @@ output_data := value.Interface().([][]float32)
 
 运行
 ```bash
-go mod init github.com/paddlepaddle
 export LD_LIBRARY_PATH=`pwd`/paddle_c/paddle/lib:$LD_LIBRARY_PATH
 go run ./demo/mobilenet.go
 ```

go/demo/mobilenet.go

@@ -13,7 +13,7 @@
 // limitations under the License.
 package main
 
-import "github.com/paddlepaddle/paddle"
+import "../paddle"
 import "strings"
 import "io/ioutil"
 import "strconv"

go/paddle/common.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
 // #include <stdbool.h>
 // #include <paddle_c_api.h>
 import "C"

go/paddle/config.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
 // #include <stdbool.h>
 // #include <stdlib.h>
 // #include <paddle_c_api.h>

go/paddle/predictor.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
 // #include <stdbool.h>
 // #include "paddle_c_api.h"
 import "C"
@@ -88,7 +88,7 @@ func (predictor *Predictor) GetInputNames() []string {
 }
 
 func (predictor *Predictor) GetOutputNames() []string {
-	names := make([]string, predictor.GetOutputNum())
+	names := make([]string, predictor.GetInputNum())
 	for i := 0; i < len(names); i++ {
 		names[i] = predictor.GetOutputName(i)
 	}

go/paddle/tensor.go

@@ -15,7 +15,7 @@
 package paddle
 
 // #cgo CFLAGS: -I${SRCDIR}/../paddle_c/paddle/include
-// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_inference_c
+// #cgo LDFLAGS: -L${SRCDIR}/../paddle_c/paddle/lib -lpaddle_fluid_c
 // #include <stdbool.h>
 // #include <stdlib.h>
 // #include <string.h>
@@ -209,7 +209,7 @@ func DecodeTensor(r *bytes.Reader, shape []int32, t reflect.Type, ptr reflect.Va
 		value := reflect.Indirect(ptr)
 		value.Set(reflect.MakeSlice(t, int(shape[0]), int(shape[0])))
 		if len(shape) == 1 && value.Len() > 0 {
-			switch value.Index(0).Kind() {
+			switch value.Index(1).Kind() {
 			case reflect.Uint8, reflect.Int32, reflect.Int64, reflect.Float32:
 				binary.Read(r, Endian(), value.Interface())
 				return

paddle/extension.h

@@ -15,4 +15,4 @@ limitations under the License. */
 #pragma once
 
 // All paddle apis in C++ frontend
-#include "paddle/fluid/extension/include/ext_all.h"
+#include "paddle/fluid/extension/include/all.h"

paddle/fluid/distributed/fleet.cc

@@ -501,7 +501,7 @@ void FleetWrapper::ShrinkDenseTable(int table_id, Scope* scope,
     if (name.find("batch_sum") != std::string::npos) {
       Variable* var = scope->FindVar(name);
       CHECK(var != nullptr) << "var[" << name << "] not found";
-      VLOG(3) << "prepare shrink dense batch_sum";
+      VLOG(0) << "prepare shrink dense batch_sum";
       LoDTensor* tensor = var->GetMutable<LoDTensor>();
       float* g = tensor->data<float>();