polish windows ci (#32964)

cmake/external/mkldnn.cmake

@@ -101,8 +101,11 @@ ADD_DEPENDENCIES(mkldnn ${MKLDNN_PROJECT})
 # it can be directly contained in wheel or capi
 if(WIN32)
     SET(MKLDNN_SHARED_LIB ${MKLDNN_INSTALL_DIR}/bin/mkldnn.dll)
+
+    file(TO_NATIVE_PATH ${MKLDNN_INSTALL_DIR} NATIVE_MKLDNN_INSTALL_DIR)
+    file(TO_NATIVE_PATH ${MKLDNN_SHARED_LIB} NATIVE_MKLDNN_SHARED_LIB)
     ADD_CUSTOM_COMMAND(TARGET ${MKLDNN_PROJECT} POST_BUILD
-        COMMAND ${CMAKE_COMMAND} -E copy ${MKLDNN_INSTALL_DIR}/bin/dnnl.dll ${MKLDNN_SHARED_LIB})
+        COMMAND (copy ${NATIVE_MKLDNN_INSTALL_DIR}\\bin\\dnnl.dll ${NATIVE_MKLDNN_SHARED_LIB} /Y))
     add_custom_command(TARGET ${MKLDNN_PROJECT} POST_BUILD VERBATIM
         COMMAND dumpbin /exports ${MKLDNN_INSTALL_DIR}/bin/mkldnn.dll > ${MKLDNN_INSTALL_DIR}/bin/exports.txt)
     add_custom_command(TARGET ${MKLDNN_PROJECT} POST_BUILD VERBATIM

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

@@ -72,8 +72,12 @@ if(WITH_GPU)
   if(NOT WIN32)
     set(CUDA_LIB "/usr/local/cuda/lib64/" CACHE STRING "CUDA Library")
   else()
-    if(CUDA_LIB STREQUAL "")
-      set(CUDA_LIB "C:\\Program\ Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v8.0\\lib\\x64")
+    if(NOT DEFINED CUDA_LIB)
+      if(DEFINED ENV{CUDA_PATH})
+        set(CUDA_LIB "$ENV{CUDA_PATH}\\lib\\x64")
+      else()
+        set(CUDA_LIB "C:\\Program\ Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.2\\lib\\x64")
+      endif()
     endif()
   endif(NOT WIN32)
 endif()

paddle/fluid/inference/api/demo_ci/run.sh

@@ -85,7 +85,7 @@ for WITH_STATIC_LIB in ON OFF; do
   if [ $(echo `uname` | grep "Win") != "" ]; then
     # TODO(wilber, T8T9): Do we still need to support windows gpu static library
     if [ $TEST_GPU_CPU == ON ] && [ $WITH_STATIC_LIB == ON ]; then
-      return 0
+      continue
     fi
     # -----simple_on_word2vec on windows-----
     cmake .. -G "Visual Studio 15 2017" -A x64 -T host=x64 -DPADDLE_LIB=${inference_install_dir} \

paddle/fluid/pybind/CMakeLists.txt

@@ -142,7 +142,7 @@ if(WITH_PYTHON)
     "${op_function_generator_path}\\op_function_generator.exe ${tmp_impl_file}\n"
     "if %ERRORLEVEL% NEQ 0 (\n"
     "    set /a build_times=%build_times%+1\n"
-    "    if %build_times% GEQ 3 (\n"
+    "    if %build_times% GEQ 10 (\n"
     "        exit /b 1\n"
     "    ) else (\n"
     "        goto :retry\n"

paddle/scripts/paddle_build.bat

@@ -161,6 +161,7 @@ set WITH_MKL=ON
 set WITH_GPU=ON
 set WITH_AVX=ON
 set MSVC_STATIC_CRT=OFF
+set ON_INFER=ON
 
 call :cmake || goto cmake_error
 call :build || goto build_error
@@ -177,12 +178,13 @@ set WITH_GPU=OFF
 set WITH_AVX=OFF
 set MSVC_STATIC_CRT=ON
 set retry_times=1
+set ON_INFER=OFF
 
 call :cmake || goto cmake_error
 call :build || goto build_error
 call :test_whl_pacakage || goto test_whl_pacakage_error
 call :test_unit || goto test_unit_error
-call :test_inference || goto test_inference_error
+:: call :test_inference || goto test_inference_error
 :: call :check_change_of_unittest || goto check_change_of_unittest_error
 goto:success
 
@@ -191,7 +193,7 @@ rem ------Build windows avx whl package------
 set WITH_AVX=ON
 set ON_INFER=OFF
 set CUDA_ARCH_NAME=All
-set retry_times=4
+set retry_times=3
 
 call :cmake || goto cmake_error
 call :build || goto build_error
@@ -203,7 +205,7 @@ rem ------Build windows no-avx whl package------
 set WITH_AVX=OFF
 set ON_INFER=OFF
 set CUDA_ARCH_NAME=All
-set retry_times=4
+set retry_times=3
 
 call :cmake || goto cmake_error
 call :build || goto build_error
@@ -265,7 +267,7 @@ if "%WITH_GPU%"=="ON" (
 )
 
 rem ------initialize the python environment------
-@ECHO ON
+@ECHO OFF
 set PYTHON_EXECUTABLE=%PYTHON_ROOT%\python.exe
 set PATH=%PYTHON_ROOT%;%PYTHON_ROOT%\Scripts;%PATH%
 if "%WITH_PYTHON%" == "ON" (
@@ -420,7 +422,7 @@ if %GENERATOR% == "Ninja" (
     ninja all
 ) else (
     if "%WITH_CLCACHE%"=="OFF" (
-        MSBuild /m:%PARALLEL_PROJECT_COUNT% /p:PreferredToolArchitecture=x64 /p:Configuration=Release /verbosity:%LOG_LEVEL% ALL_BUILD.vcxproj
+        MSBuild /m:%PARALLEL_PROJECT_COUNT% /p:PreferredToolArchitecture=x64 /p:TrackFileAccess=false /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
     )
@@ -648,12 +650,12 @@ 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% -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% ^
+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% -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% >>  check_change_of_unittest.sh
+-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% >>  check_change_of_unittest.sh
 echo cat ^<^<EOF>>  check_change_of_unittest.sh
 echo     ============================================       >>  check_change_of_unittest.sh
 echo     Generate unit tests.spec of develop.               >>  check_change_of_unittest.sh

tools/check_added_ut.sh

@@ -66,7 +66,7 @@ rm -rf prec_build
 if [[ "$SYSTEM" == "Linux" ]] || [[ "$SYSTEM" == "Darwin" ]];then
     rm $PADDLE_ROOT/br-ut $PADDLE_ROOT/pr-ut $PADDLE_ROOT/paddle/scripts/paddle_build_pre.sh
 elif [[ "$SYSTEM" == "Windows_NT" ]];then
-    rm $PADDLE_ROOT/br-ut $PADDLE_ROOT/pr-ut $PADDLE_ROOT/get_added_ut.sh
+    rm $PADDLE_ROOT/br-ut $PADDLE_ROOT/pr-ut $PADDLE_ROOT/win_cmake.sh
 fi
 git checkout -f $CURBRANCH
 echo $CURBRANCH

tools/parallel_UT_rule.py

@@ -159,7 +159,6 @@ CPU_PARALLEL_JOB = [
     'test_matmul_transpose_reshape_fuse_pass',
     'test_matmul_mkldnn_op',
     'test_matmul_bf16_mkldnn_op',
-    'test_math_op_patch',
     'test_match_matrix_tensor_op',
     'test_lookup_table_dequant_op',
     'test_logging_utils',
@@ -175,7 +174,6 @@ CPU_PARALLEL_JOB = [
     'test_layer_norm_mkldnn_op',
     'test_layer_norm_bf16_mkldnn_op',
     'test_layer',
-    'test_lambv2_op',
     'test_is_test_pass',
     'test_ir_skip_layernorm_pass',
     'test_ir_graph',
@@ -657,6 +655,8 @@ TETRAD_PARALLEL_JOB = [
 # It run 2 job each time, If it failed due to Insufficient GPU memory or CUBLAS_STATUS_ALLOC_FAILED,
 # just remove it from this list.
 TWO_PARALLEL_JOB = [
+    'test_lambv2_op',
+    'test_math_op_patch',
     'test_tensor_to_numpy',
     'zero_copy_tensor_test',
     'sequence_pooling_test',