Fix ninja compilation bug and warning on windows (#32987)

* fix ninja compilation bug on windows * polish windows ci * polish windows ci

Fix ninja compilation bug and warning on windows (#32987)
* fix ninja compilation bug on windows * polish windows ci * polish windows ci
accf284b · Zhou Wei · GitHub · f91e0f45 · accf284b · accf284b
10 changed file
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -20,6 +20,13 @@ set(PADDLE_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR})

 include(system)

+# Note(zhouwei): Ninja Generator will set CMAKE_BUILD_TYPE to Debug
+if(NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE "Release" CACHE STRING
+      "Choose the type of build, options are: Debug Release RelWithDebInfo MinSizeRel"
+      FORCE)
+endif()
+
 project(paddle CXX C)

 # enable language CUDA
@@ -213,12 +220,6 @@ if(NOT PY_VERSION)
 endif()
 set(PYBIND11_PYTHON_VERSION ${PY_VERSION})

-# CMAKE_BUILD_TYPE
-if(NOT CMAKE_BUILD_TYPE)
-    set(CMAKE_BUILD_TYPE "Release" CACHE STRING
-      "Choose the type of build, options are: Debug Release RelWithDebInfo MinSizeRel"
-      FORCE)
-endif()

 # the type of sanitizer, options are: Address, Leak, Memory, Thread, Undefined. Default: OFF
 if(SANITIZER_TYPE AND NOT "${SANITIZER_TYPE}" MATCHES "^(Address|Leak|Memory|Thread|Undefined)$")

--- a/cmake/cuda.cmake
+++ b/cmake/cuda.cmake
@@ -205,23 +205,16 @@ set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-extended-lambda")
 if(WIN32)
  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler \"/wd4244 /wd4267 /wd4819 \"")
  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler /bigobj")
-  if(CMAKE_BUILD_TYPE  STREQUAL "Debug")
-    # match the cl's _ITERATOR_DEBUG_LEVEL
-    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler \"-g -G -D_DEBUG\"")
-    if(MSVC_STATIC_CRT)
-      set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler /MTd")
-    else()
-      set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler /MDd")
-    endif()
-  elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
-    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler \"-DNDEBUG\"")
-    if(MSVC_STATIC_CRT)
-      set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler /MT")
-    else()
-      set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler /MD")
-    endif()
-  else()
-    message(FATAL "Windows only support Release or Debug build now. Please set visual studio build type to Release/Debug, x64 build.")
+  if(MSVC_STATIC_CRT)
+    set(CMAKE_CUDA_FLAGS_DEBUG   "${CMAKE_CUDA_FLAGS_DEBUG} -Xcompiler /MTd")
+    set(CMAKE_CUDA_FLAGS_RELEASE  "${CMAKE_CUDA_FLAGS_RELEASE} -Xcompiler /MT")
+    foreach(flag_var
+        CMAKE_CUDA_FLAGS CMAKE_CUDA_FLAGS_DEBUG CMAKE_CUDA_FLAGS_RELEASE
+        CMAKE_CUDA_FLAGS_MINSIZEREL CMAKE_CUDA_FLAGS_RELWITHDEBINFO)
+        if(${flag_var} MATCHES "-MD")
+            string(REGEX REPLACE "-MD" "-MT" ${flag_var} "${${flag_var}}")
+        endif()
+    endforeach(flag_var)
  endif()
 endif()


--- a/cmake/external/mkldnn.cmake
+++ b/cmake/external/mkldnn.cmake
@@ -60,8 +60,8 @@ ExternalProject_Add(
    DEPENDS             ${MKLDNN_DEPENDS}
    PREFIX              ${MKLDNN_PREFIX_DIR}
    SOURCE_DIR          ${MKLDNN_SOURCE_DIR}
-    BUILD_ALWAYS        1
-    # UPDATE_COMMAND      ""
+    UPDATE_COMMAND      ""
+    #BUILD_ALWAYS        1
    CMAKE_ARGS          -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
                        -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
                        -DCMAKE_CXX_FLAGS_RELEASE=${CMAKE_CXX_FLAGS_RELEASE}

--- a/cmake/external/warpctc.cmake
+++ b/cmake/external/warpctc.cmake
@@ -100,9 +100,9 @@ else()
        "${WARPCTC_DOWNLOAD_CMD}"
        PREFIX          ${WARPCTC_PREFIX_DIR}
        SOURCE_DIR      ${WARPCTC_SOURCE_DIR}
-        #UPDATE_COMMAND  ""
+        UPDATE_COMMAND  ""
        PATCH_COMMAND   ""
-        BUILD_ALWAYS    1
+        #BUILD_ALWAYS    1
        CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
                        -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
                        -DCMAKE_C_FLAGS=${WARPCTC_C_FLAGS}

--- a/cmake/flags.cmake
+++ b/cmake/flags.cmake
@@ -28,7 +28,12 @@ function(CheckCompilerCXX14Flag)
 endfunction()

 CheckCompilerCXX14Flag()
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
+if(NOT WIN32)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
+else()
+    set(CMAKE_CXX_STANDARD 14)
+endif()
+
 # safe_set_flag
 #
 # Set a compile flag only if compiler is support

--- a/cmake/generic.cmake
+++ b/cmake/generic.cmake
@@ -92,7 +92,7 @@ include_directories(${CMAKE_CURRENT_BINARY_DIR})
 # including io directory for inference lib paddle_api.h
 include_directories("${PADDLE_SOURCE_DIR}/paddle/fluid/framework/io")

-if(NOT APPLE)
+if(NOT APPLE AND NOT WIN32)
  find_package(Threads REQUIRED)
  link_libraries(${CMAKE_THREAD_LIBS_INIT})
  if(WITH_PSLIB OR WITH_DISTRIBUTE)
@@ -100,7 +100,7 @@ if(NOT APPLE)
  else()
    set(CMAKE_CXX_LINK_EXECUTABLE "${CMAKE_CXX_LINK_EXECUTABLE} -pthread -ldl -lrt")
  endif()
-endif(NOT APPLE)
+endif()

 set_property(GLOBAL PROPERTY FLUID_MODULES "")
 # find all fluid modules is used for paddle fluid static library

--- a/cmake/init.cmake
+++ b/cmake/init.cmake
@@ -17,16 +17,30 @@ if(NOT WIN32)
    set(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG")
    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
    set(CMAKE_CXX_FLAGS_MINSIZEREL "-Os -DNDEBUG")
+    
+    if(WITH_GPU)
+        set(CMAKE_CUDA_FLAGS_DEBUG "-g")
+        set(CMAKE_CUDA_FLAGS_RELEASE "-O3 -DNDEBUG")
+        set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
+        set(CMAKE_CUDA_FLAGS_MINSIZEREL "-O1 -DNDEBUG")
+    endif()
 else()
-    set(CMAKE_C_FLAGS_DEBUG "/Zi /DEBUG")
-    set(CMAKE_C_FLAGS_RELEASE "/O2 /DNDEBUG")
-    set(CMAKE_C_FLAGS_RELWITHDEBINFO "/O2 /DNDEBUG")
-    set(CMAKE_C_FLAGS_MINSIZEREL "/Os /DNDEBUG")
+    set(CMAKE_C_FLAGS_DEBUG "/MDd /Zi /Ob0 /Od /RTC1")
+    set(CMAKE_C_FLAGS_RELEASE "/MD /O2 /Ob2 /DNDEBUG")
+    set(CMAKE_C_FLAGS_RELWITHDEBINFO "/MD /Zi /O2 /Ob1 /DNDEBUG")
+    set(CMAKE_C_FLAGS_MINSIZEREL "/MD /O1 /Ob1 /DNDEBUG")

-    set(CMAKE_CXX_FLAGS_DEBUG "/Zi /DEBUG")
-    set(CMAKE_CXX_FLAGS_RELEASE "/O2 /DNDEBUG")
-    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "/O2 /DNDEBUG")
-    set(CMAKE_CXX_FLAGS_MINSIZEREL "/Os /DNDEBUG")
+    set(CMAKE_CXX_FLAGS_DEBUG "/MDd /Zi /Ob0 /Od /RTC1")
+    set(CMAKE_CXX_FLAGS_RELEASE "/MD /O2 /Ob2 /DNDEBUG")
+    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "/MD /Zi /O2 /Ob1 /DNDEBUG")
+    set(CMAKE_CXX_FLAGS_MINSIZEREL "/MD /O1 /Ob1 /DNDEBUG")
+
+    if(WITH_GPU)
+        set(CMAKE_CUDA_FLAGS_DEBUG "-Xcompiler=\"-MDd -Zi -Ob0 -Od /RTC1\"")
+        set(CMAKE_CUDA_FLAGS_RELEASE "-Xcompiler=\"-MD -O2 -Ob2\" -DNDEBUG")
+        set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "-Xcompiler=\"-MD -Zi -O2 -Ob1\" -DNDEBUG")
+        set(CMAKE_CUDA_FLAGS_MINSIZEREL "-Xcompiler=\"-MD -O1 -Ob1\" -DNDEBUG")
+    endif()

    # It can specify CUDA compile flag manualy,
    # its use is to remvoe /Zi to reduce GPU static library size. But it's dangerous
@@ -34,10 +48,3 @@ else()
    # Now, it's only used in VS2015 + CUDA:[10.0, 10.2]
    set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
 endif()
-
-if(WITH_GPU)
-    set(CMAKE_CUDA_FLAGS_DEBUG "-g")
-    set(CMAKE_CUDA_FLAGS_RELEASE "-O3 -DNDEBUG")
-    set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "-O2 -g -DNDEBUG")
-    set(CMAKE_CUDA_FLAGS_MINSIZEREL "-O1 -DNDEBUG")
-endif()
--- a/paddle/scripts/paddle_build.bat
+++ b/paddle/scripts/paddle_build.bat
@@ -18,7 +18,7 @@ rem       Paddle CI Task On Windows Platform
 rem =================================================

 @ECHO ON
-setlocal
+setlocal enabledelayedexpansion

 rem -------clean up environment-----------
 set work_dir=%cd%
@@ -63,7 +63,7 @@ if not defined WITH_PYTHON set WITH_PYTHON=ON
 if not defined ON_INFER set ON_INFER=ON
 if not defined WITH_INFERENCE_API_TEST set WITH_INFERENCE_API_TEST=ON
 if not defined WITH_STATIC_LIB set WITH_STATIC_LIB=ON
-if not defined WITH_TPCACHE set WITH_TPCACHE=ON
+if not defined WITH_TPCACHE set WITH_TPCACHE=OFF
 if not defined WITH_CLCACHE set WITH_CLCACHE=OFF
 if not defined WITH_CACHE set WITH_CACHE=OFF
 if not defined WITH_UNITY_BUILD set WITH_UNITY_BUILD=OFF
@@ -236,6 +236,8 @@ call "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary
 set DISTUTILS_USE_SDK=1
 rem Windows 10 Kit bin dir
 set PATH=C:\Program Files (x86)\Windows Kits\10\bin\10.0.17763.0\x64;%PATH%
+rem Use 64-bit ToolSet to compile
+set PreferredToolArchitecture=x64

 for /F %%# in ('wmic os get localdatetime^|findstr 20') do set start=%%#
 set start=%start:~4,10%
@@ -263,12 +265,12 @@ rem ------initialize the python environment------
 @ECHO ON
 set PYTHON_EXECUTABLE=%PYTHON_ROOT%\python.exe
 set PATH=%PYTHON_ROOT%;%PYTHON_ROOT%\Scripts;%PATH%
-if %WITH_PYTHON% == "ON" (
+if "%WITH_PYTHON%" == "ON" (
    where python
    where pip
    pip install wheel --user
    pip install -r %work_dir%\python\requirements.txt --user
-    if %ERRORLEVEL% NEQ 0 (
+    if !ERRORLEVEL! NEQ 0 (
        echo pip install requirements.txt failed!
        exit /b 7
    )
@@ -329,14 +331,14 @@ if "%WITH_GPU%"=="ON" (
 )

 :cmake_impl
-echo cmake .. -G %GENERATOR% -T host=x64 -DCMAKE_BUILD_TYPE=Release -DWITH_AVX=%WITH_AVX% -DWITH_GPU=%WITH_GPU% -DWITH_MKL=%WITH_MKL% ^
+echo cmake .. -G %GENERATOR% -DCMAKE_BUILD_TYPE=Release -DWITH_AVX=%WITH_AVX% -DWITH_GPU=%WITH_GPU% -DWITH_MKL=%WITH_MKL% ^
 -DWITH_TESTING=%WITH_TESTING% -DWITH_PYTHON=%WITH_PYTHON% -DPYTHON_EXECUTABLE=%PYTHON_EXECUTABLE% -DON_INFER=%ON_INFER% ^
 -DWITH_INFERENCE_API_TEST=%WITH_INFERENCE_API_TEST% -DTHIRD_PARTY_PATH=%THIRD_PARTY_PATH% ^
 -DINFERENCE_DEMO_INSTALL_DIR=%INFERENCE_DEMO_INSTALL_DIR% -DWITH_STATIC_LIB=%WITH_STATIC_LIB% ^
 -DWITH_TENSORRT=%WITH_TENSORRT% -DTENSORRT_ROOT="%TENSORRT_ROOT%" -DMSVC_STATIC_CRT=%MSVC_STATIC_CRT% ^
 -DWITH_UNITY_BUILD=%WITH_UNITY_BUILD% -DCUDA_ARCH_NAME=%CUDA_ARCH_NAME%

-cmake .. -G %GENERATOR% -DCMAKE_BUILD_TYPE=Release -T host=x64 -DWITH_AVX=%WITH_AVX% -DWITH_GPU=%WITH_GPU% -DWITH_MKL=%WITH_MKL% ^
+cmake .. -G %GENERATOR% -DCMAKE_BUILD_TYPE=Release -DWITH_AVX=%WITH_AVX% -DWITH_GPU=%WITH_GPU% -DWITH_MKL=%WITH_MKL% ^
 -DWITH_TESTING=%WITH_TESTING% -DWITH_PYTHON=%WITH_PYTHON% -DPYTHON_EXECUTABLE=%PYTHON_EXECUTABLE% -DON_INFER=%ON_INFER% ^
 -DWITH_INFERENCE_API_TEST=%WITH_INFERENCE_API_TEST% -DTHIRD_PARTY_PATH=%THIRD_PARTY_PATH% ^
 -DINFERENCE_DEMO_INSTALL_DIR=%INFERENCE_DEMO_INSTALL_DIR% -DWITH_STATIC_LIB=%WITH_STATIC_LIB% ^
@@ -366,7 +368,7 @@ echo Build third_party the %build_times% time:
 if %GENERATOR% == "Ninja" (
    ninja third_party
 ) else (
-    MSBuild /m /p:PreferredToolArchitecture=x64 /p:Configuration=Release /verbosity:quiet third_party.vcxproj
+    MSBuild /m /p:PreferredToolArchitecture=x64 /p:Configuration=Release /verbosity:%LOG_LEVEL% third_party.vcxproj
 )
 if %ERRORLEVEL% NEQ 0 (
    set /a build_times=%build_times%+1  
@@ -412,10 +414,10 @@ if "%WITH_TESTING%"=="ON" (

 echo Build Paddle the %build_times% time:
 if %GENERATOR% == "Ninja" (
-    ninja -j %PARALLEL_PROJECT_COUNT%
+    ninja all
 ) else (
    if "%WITH_CLCACHE%"=="OFF" (
-        MSBuild /m:%PARALLEL_PROJECT_COUNT% /p:PreferredToolArchitecture=x64 /p:TrackFileAccess=false /p:Configuration=Release /verbosity:%LOG_LEVEL% ALL_BUILD.vcxproj
+        MSBuild /m:%PARALLEL_PROJECT_COUNT% /p:PreferredToolArchitecture=x64 /p:Configuration=Release /verbosity:%LOG_LEVEL% ALL_BUILD.vcxproj
    ) else (
        MSBuild /m:%PARALLEL_PROJECT_COUNT% /p:PreferredToolArchitecture=x64 /p:TrackFileAccess=false /p:CLToolExe=clcache.exe /p:CLToolPath=%PYTHON_ROOT%\Scripts /p:Configuration=Release /verbosity:%LOG_LEVEL% ALL_BUILD.vcxproj
    )
@@ -644,7 +646,7 @@ echo     git fetch upstream $BRANCH # develop is not fetched>>  check_change_of_
 echo fi>>  check_change_of_unittest.sh
 echo git checkout -b origin_pr >>  check_change_of_unittest.sh
 echo git checkout -f $BRANCH >>  check_change_of_unittest.sh
-echo cmake .. -G %GENERATOR% -T host=x64 -DWITH_AVX=%WITH_AVX% -DWITH_GPU=%WITH_GPU% -DWITH_MKL=%WITH_MKL% ^
+echo cmake .. -G %GENERATOR% -DWITH_AVX=%WITH_AVX% -DWITH_GPU=%WITH_GPU% -DWITH_MKL=%WITH_MKL% ^
 -DWITH_TESTING=%WITH_TESTING% -DWITH_PYTHON=%WITH_PYTHON% -DPYTHON_EXECUTABLE=%PYTHON_EXECUTABLE% -DON_INFER=%ON_INFER% ^
 -DWITH_INFERENCE_API_TEST=%WITH_INFERENCE_API_TEST% -DTHIRD_PARTY_PATH=%THIRD_PARTY_PATH% ^
 -DINFERENCE_DEMO_INSTALL_DIR=%INFERENCE_DEMO_INSTALL_DIR% -DWITH_STATIC_LIB=%WITH_STATIC_LIB% ^

--- a/tools/parallel_UT_rule.py
+++ b/tools/parallel_UT_rule.py
@@ -650,7 +650,6 @@ TETRAD_PARALLEL_JOB = [
    'test_collective_wait',
    'test_collective_split_row_linear',
    'test_collective_split_embedding',
-    'test_custom_attrs_jit',
    'float16_gpu_test',
    'test_leaky_relu_grad_grad_functor',
    'test_complex_simplenet',

--- a/tools/windows/build_compile_environment.bat
+++ b/tools/windows/build_compile_environment.bat
@@ -132,7 +132,7 @@ goto :eof
 :vs
 echo ">>>>>>>> step [4/7]: Visual Studio 2017 "
 cmd /C "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvars64.bat"  > nul 2> nul || call :install_visual_studio
-goto :cuda10
+goto :cuda

 :install_visual_studio
 echo There is not Visual Studio in this PC, will install VS2017.
@@ -153,7 +153,7 @@ goto :eof
 :: ===== end step 4: Visual Studio 2017 =====

 :: ===== start step 5: CUDA 11 =====
-:cuda10
+:cuda
 echo ">>>>>>>> step [5/7]: CUDA 11.2"
 cmd /C nvcc --version 2> nul | findstr /C:"11.2" > nul 2> nul || call :install_cuda
 goto java-jre