windows lib size crop from 5.4G to 3.9G (#27477)

cmake/generic.cmake

@@ -446,6 +446,9 @@ function(nv_library TARGET_NAME)
         message(FATAL "Please specify source file or library in nv_library.")
       endif()
     endif(nv_library_SRCS)
+    if (WIN32)
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
+    endif(WIN32)
   endif()
 endfunction(nv_library)
 
@@ -461,6 +464,9 @@ function(nv_binary TARGET_NAME)
       add_dependencies(${TARGET_NAME} ${nv_binary_DEPS})
       common_link(${TARGET_NAME})
     endif()
+    if (WIN32)
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
+    endif(WIN32)
   endif()
 endfunction(nv_binary)
 
@@ -482,6 +488,9 @@ 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 ${TARGET_NAME} PROPERTY ENVIRONMENT FLAGS_cudnn_deterministic=true)
+    if (WIN32)
+      set_target_properties(${TARGET_NAME} PROPERTIES VS_USER_PROPS ${WIN_PROPS})
+    endif(WIN32)
   endif()
 endfunction(nv_test)
 

cmake/inference_lib.cmake

@@ -19,9 +19,8 @@ set(PADDLE_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_install_dir" CACHE STRING
 set(PADDLE_INFERENCE_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_inference_install_dir" CACHE STRING
   "A path setting paddle inference shared and static libraries")
   
-# TODO(zhaolong)
-# At present, the size of static lib in Windows exceeds the system limit,
-# so the generation of static lib is temporarily turned off.
+# At present, the size of static lib in Windows is very large,
+# so we need to crop the library size.
 if(WIN32)
     #todo: remove the option 
     option(WITH_STATIC_LIB "Compile demo with static/shared library, default use dynamic."   OFF)
@@ -196,7 +195,11 @@ set(PADDLE_INFERENCE_C_INSTALL_DIR "${CMAKE_BINARY_DIR}/paddle_inference_c_insta
 copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_C_INSTALL_DIR})
 
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
-set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_fluid_c.*)
+if(WIN32)
+  set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/${CMAKE_BUILD_TYPE}/paddle_fluid_c.*)
+else(WIN32)
+  set(paddle_fluid_c_lib ${PADDLE_BINARY_DIR}/paddle/fluid/inference/capi/libpaddle_fluid_c.*)
+endif(WIN32)
 
 copy(inference_lib_dist
       SRCS  ${src_dir}/inference/capi/paddle_c_api.h  ${paddle_fluid_c_lib}

cmake/init.cmake

@@ -26,4 +26,7 @@ if(WITH_GPU)
     set(CMAKE_CUDA_FLAGS_MINSIZEREL "-O1 -DNDEBUG")
 endif()
 
+if(WIN32)
+    set(WIN_PROPS ${CMAKE_SOURCE_DIR}/cmake/paddle_win.props)
+endif()
 

cmake/paddle_win.props

+<?xml version="1.0" encoding="utf-8"?>
+<Project xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+    <ItemDefinitionGroup>
+        <CudaCompile>
+            <!-- Project schema: Host properties -->
+            <UseHostDefines>true</UseHostDefines>
+            <Emulation>false</Emulation>
+            <HostDebugInfo Condition="'$(Configuration)' == 'Debug'">true</HostDebugInfo>
+            <HostDebugInfo Condition="'$(Configuration)' != 'Debug'">false</HostDebugInfo>
+            <FastMath>false</FastMath>
+            <Optimization>InheritFromHost</Optimization>
+            <Runtime>InheritFromHost</Runtime>
+            <RuntimeChecks>InheritFromHost</RuntimeChecks>
+            <TypeInfo>InheritFromHost</TypeInfo>
+            <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</BuildCommandLineTemplate>
+            <BuildDynamicCommandLineTemplate>[CodeGeneration]</BuildDynamicCommandLineTemplate>
+            <CleanCommandLineTemplate>-clean</CleanCommandLineTemplate>
+            <!-- <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] [ProgramDataBaseFileName] $(CudaForceSynchronousPdbWrites) /Zi [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate> -->
+            <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] [ProgramDataBaseFileName] $(CudaForceSynchronousPdbWrites) [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate>
+
+            <DriverApiCommandLineTemplate>%(BaseCommandLineTemplate) [CompileOut] "%(FullPath)"</DriverApiCommandLineTemplate>
+            <RuntimeApiCommandLineTemplate>%(BaseCommandLineTemplate) [HostDebugInfo] [Emulation] [FastMath] [Defines] %(HostCommandLineTemplate) [CompileOut] "%(FullPath)"</RuntimeApiCommandLineTemplate>
+
+            <CommandLineTemplate>
+# (Approximate command-line.  Settings inherited from host are not visible below.)
+# (Please see the output window after a build for the full command-line)
+
+# Driver API (NVCC Compilation Type is .cubin, .gpu, or .ptx)
+set CUDAFE_FLAGS=--sdk_dir "$(WindowsSdkDir)"
+"$(CudaToolkitNvccPath)" %(BuildCommandLineTemplate) %(DriverApiCommandLineTemplate)
+
+# Runtime API (NVCC Compilation Type is hybrid object or .c file)
+set CUDAFE_FLAGS=--sdk_dir "$(WindowsSdkDir)"
+"$(CudaToolkitNvccPath)" %(BuildCommandLineTemplate) %(RuntimeApiCommandLineTemplate)
+            </CommandLineTemplate>
+            <ExecutionDescription>Compiling CUDA source file %(Identity)...</ExecutionDescription>
+            <ExclusionDescription>Skipping CUDA source file %(Identity) (excluded from build).</ExclusionDescription>
+
+            <!-- Miscellaneous -->
+            <PropsCacheOutputFile>%(Filename)%(Extension).cache</PropsCacheOutputFile>
+            <PropsCacheOutputPath>$(IntDir)%(PropsCacheOutputFile)</PropsCacheOutputPath>
+
+            <CudaCompileCoreProject>$(MSBuildProjectFullPath)</CudaCompileCoreProject>
+        </CudaCompile>
+
+        <CudaLink>
+            <PerformDeviceLink>true</PerformDeviceLink>
+            <LinkOut>$(IntDir)$(TargetName).device-link.obj</LinkOut>
+
+            <AdditionalLibraryDirectories></AdditionalLibraryDirectories>
+            <UseHostLibraryDirectories>true</UseHostLibraryDirectories>
+            <AdditionalDependencies></AdditionalDependencies>
+            <UseHostLibraryDependencies>true</UseHostLibraryDependencies>
+
+            <GPUDebugInfo>InheritFromProject</GPUDebugInfo>
+            <Optimization>InheritFromProject</Optimization>
+
+            <!-- Implicitly inherited from the project via @(CudaCompile) -->
+            <CodeGeneration></CodeGeneration>
+            <RuntimeChecks></RuntimeChecks>
+            <Runtime></Runtime>
+            <TargetMachinePlatform></TargetMachinePlatform>
+            <TypeInfo></TypeInfo>
+            <Warning></Warning>
+
+            <Inputs></Inputs>
+
+            <!-- <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] /Zi [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate> -->
+            <HostCommandLineTemplate>-Xcompiler &quot;/EHsc [Warning] /nologo [Optimization] [RuntimeChecks] [Runtime] [TypeInfo]&quot;</HostCommandLineTemplate>
+            <LinkCommandLineTemplate>"$(CudaToolkitNvccPath)" -dlink [LinkOut] %(HostCommandLineTemplate) [AdditionalLibraryDirectories] [AdditionalDependencies] [AdditionalOptions] [CodeGeneration] [GPUDebugInfo] [TargetMachinePlatform] [Inputs]</LinkCommandLineTemplate>
+            <CommandLineTemplate>
+# (Approximate command-line.  Settings inherited from host are not visible below.)
+# (Please see the output window after a build for the full command-line)
+
+%(LinkCommandLineTemplate)
+            </CommandLineTemplate>
+        </CudaLink>
+
+        <Link>
+            <AdditionalLibraryDirectories>%(AdditionalLibraryDirectories);$(CudaToolkitLibDir)</AdditionalLibraryDirectories>
+        </Link>
+
+        <ClCompile>
+            <AdditionalIncludeDirectories>%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir)</AdditionalIncludeDirectories>
+        </ClCompile>
+    </ItemDefinitionGroup>
+</Project>
+

paddle/fluid/inference/CMakeLists.txt

@@ -44,14 +44,9 @@ add_subdirectory(api)
 set(STATIC_INFERENCE_API paddle_inference_api analysis_predictor
      zero_copy_tensor reset_tensor_array 
         analysis_config paddle_pass_builder activation_functions ${mkldnn_quantizer_cfg})
-# TODO(xingzhaolong, jiweibo): remove this and create_static_lib(paddle_fluid) on windows GPU
-if(WIN32 AND WITH_GPU)
-  cc_library(paddle_fluid DEPS ${fluid_modules} ${STATIC_INFERENCE_API}) 
-else()
-  create_static_lib(paddle_fluid ${fluid_modules} ${STATIC_INFERENCE_API}) 
-endif()
+create_static_lib(paddle_fluid ${fluid_modules} ${STATIC_INFERENCE_API})
 
-if(NOT APPLE AND NOT WIN32)
+if(NOT APPLE)
   # TODO(liuyiqu: Temporarily disable the link flag because it is not support on Mac.
   set(LINK_FLAGS "-Wl,--retain-symbols-file ${CMAKE_CURRENT_SOURCE_DIR}/paddle_fluid.sym")
   set_target_properties(paddle_fluid PROPERTIES LINK_FLAGS "${LINK_FLAGS}")

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

@@ -68,11 +68,6 @@ rm -rf *
 
 for WITH_STATIC_LIB in ON OFF; do
   if [ $(echo `uname` | grep "Win") != "" ]; then
-    # TODO(xingzhaolong, jiweibo): remove this if windows GPU library is ready.
-    if [ $TEST_GPU_CPU == ON] && [ $WITH_STATIC_LIB ==ON ]; then
-      return 0
-    fi
-    
     # -----simple_on_word2vec on windows-----
     cmake .. -G "Visual Studio 14 2015" -A x64 -DPADDLE_LIB=${inference_install_dir} \
       -DWITH_MKL=$TURN_ON_MKL \