TensorFlow C + + library process logging under Windows compilation

1. Preparation

• Windows 10 system, 3.6GHZ CPU, 16G memory
• Visual Studio or 2015
• Download and install Git
• Download and install CMake
• Download Install Swigwin If you do not need Python bindings, you can skip
• Clone TensorFlow
• Switch TensorFlow to the git tag you want to compile
• Modify Tensorflow/contrib/cmake/cmakelists.txt
  

  if(Tensorflow_optimize_for_native_arch) include (Checkcxxcompilerflag) Check_cxx_compiler_flag ("-march=native"compiler_opt_arch_native_supported)if(compiler_opt_arch_native_supported)Set(Cmake_cxx_flags"${cmake_cxx_flags}-march=native") # Add this section belowElse() Check_cxx_compiler_flag ("/ARCH:AVX"compiler_opt_arch_avx_supported)if(compiler_opt_arch_avx_supported)Set(Cmake_cxx_flags"${cmake_cxx_flags}/ARCH:AVX") endif () # End of Add below this section endif () endif ()

2. Steps

2. CMD run vcvarsall.bat AMD64 set specified environment parameters
4. CD Catalog: Tensorflow\contrib\cmake
6. Create a build directory, CD build

7. CMake command

   2. cmake .. -G "Visual Studio 15 2017 Win64" -T host=x64 -DCMAKE_BUILD_TYPE=Release ^
   4. -Dtensorflow_VERBOSE=ON ^
   6. -Dtensorflow_ENABLE_GRPC_SUPPORT=OFF ^
   8. -Dtensorflow_BUILD_PYTHON_BINDINGS=OFF ^
   10. -Dtensorflow_BUILD_CC_EXAMPLE=OFF ^
   12. -Dtensorflow_BUILD_SHARED_LIB=ON ^
   14. -Dtensorflow_WIN_CPU_SIMD_OPTIONS=/arch:AVX
   20. #-DSWIG_EXECUTABLE=swig.exe的路径
   22. #-DPYTHON_EXECUTABLE=python.exe的路径（anaconda3路径/python.exe）
   24. #-DPYTHON_LIBRARIES=python35.lib的路径（anaconda3路径/libs/python35.lib）

   The above command explains:
   -g Specifies the build target version. command line input cmake-g, you can view the optional version list.
   T-Host=64 is one of the many tutorials on the web that I have previously compiled with VS2017 because I did not specify-t host=64, resulting in a 32-bit compiler with fatal error C1060: "The compiler has insufficient heap space" errors.
   Tensorflow_verbose compile-time output specific information, set to on a bit to reduce the compilation speed.
   Tensorflow_enable_grpc_support GRPC Support, if you have a server/client business need, or cross-language GRPC usage requirements, select on.
   Tensorflow_build_python_bindings needs to compile the PYTHON package, select on.
   Tensorflow_build_cc_example compile the test code.
   Tensorflow_build_shared_lib compile the shared library.
   TENSORFLOW_WIN_CPU_SIMD_OPTIONS=/ARCH:AVX2 Select AVX or AVX2, depending on whether your CPU supports the instruction set.
   Tensorflow_enable_gpu whether the GPU is used for compilation.
   The Tensorflow_optimize_for_native_arch can be set to off or not, and will be automatically judged.
   Other optional parameters can be viewed in the tensorflow\tensorflow\contrib\cmake\cmakelists.txt, and for the parameters you are unsure about, keep them by default.

9. Compile
   MSBuild /p:Configuration=Release /p:Platform=x64 /verbosity:detailed tensorflow.vcxproj
这里强调一点，Configuration定义的要与cmake时DCMAKE_BUILD_TYPE定义相同

11. Run Tests

TensorFlow C + + library process logging under Windows compilation