Configuration for a Visual Studio project that depends on tensorflow.dll in C++

tensorflow_cc_windows

Follow instructions in the following link to compile tensorflow.
Build as a shared library -Dtensorflow_BUILD_SHARED_LIB=ON
https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/cmake

Create new empty project in visual studio.
A good example to test as main.cpp would be 
https://github.com/tensorflow/tensorflow/blob/r1.2/tensorflow/examples/label_image/main.cc
The following assumes that Tensorflow was downloaded in C:\tensorflow-r1.2

************
Release
************

Include directories
    C:\tensorflow-r1.2;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\external\eigen_archive;C:\tensorflow-r1.2\third_party\eigen3;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\protobuf\src\protobuf\src%(AdditionalIncludeDirectories)

Preprocessor definitions
    WIN32;_WINDOWS;NDEBUG;_ITERATOR_DEBUG_LEVEL=0;EIGEN_AVOID_STL_ARRAY;NOMINMAX;_WIN32_WINNT=0x0A00;LANG_CXX11;COMPILER_MSVC;OS_WIN;_MBCS;WIN64;WIN32_LEAN_AND_MEAN;NOGDI;PLATFORM_WINDOWS;TENSORFLOW_USE_EIGEN_THREADPOOL;EIGEN_HAS_C99_MATH;%(PreprocessorDefinitions)
	
Additional Library Directories
    C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\Release;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\protobuf\src\protobuf\Release;%(AdditionalLibraryDirectories)
	
Additional Dependencies
    libprotobuf.lib;tensorflow.lib;%(AdditionalDependencies)
    
If the project compiles but it complains at run time about the tensorflow.dll, copy it to the project x64\Release folder.

************
Debug
************
Compile tensorflow static library in debug
MSBuild /p:Configuration=Debug tensorflow_static.vcxproj

Set the custom project to use the tensorflow_static.lib generated in the previous step

Include directories
    C:\tensorflow-r1.2;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\external\eigen_archive;C:\tensorflow-r1.2\third_party\eigen3;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\protobuf\src\protobuf\src%(AdditionalIncludeDirectories)

Preprocessor definitions
    WIN32;_WINDOWS;_ITERATOR_DEBUG_LEVEL=2;EIGEN_AVOID_STL_ARRAY;NOMINMAX;_WIN32_WINNT=0x0A00;LANG_CXX11;COMPILER_MSVC;OS_WIN;_MBCS;WIN64;WIN32_LEAN_AND_MEAN;NOGDI;PLATFORM_WINDOWS;TENSORFLOW_USE_EIGEN_THREADPOOL;EIGEN_HAS_C99_MATH;
    
Additional Library Directories
    C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\fft2d\src\fft2d\Debug;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\farmhash\src\farmhash\Debug;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\gif\src\gif-build\Debug;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\Debug;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\grpc\src\grpc\Debug;C:\tensorflow-r1.2\tensorflow\contrib\cmake\build\protobuf\src\protobuf\Debug;
	
Additional Dependencies
    fft2d.lib;farmhash.lib;giflib.lib;gpr.lib;grpc_unsecure.lib;grpc++_unsecure.lib;tf_protos_cc.lib;libprotobufd.lib;/WHOLEARCHIVE:tensorflow_static.lib;%(AdditionalDependencies)

Is this GPU support? And if so have you build tensorflow with GPU support in Windows with visual studio2015?

Hi,
I have tried to build the library for tensorflow with GPU enabled, however it raise an error about not enough space. It looks like there is an issue with the size of array or any global variable:

You can fin more details in the link below:
https://stackoverflow.com/questions/46315453/error-compiling-static-library-c-for-tensorflow-when-gpu-enabled-in-cmake?noredirect=1#comment79593898_46315453

Could you tell me how could I solve this issue?

I haven't compiled for Windows with GPU support. I'm not sure what is the cause of the problem that you are having, but a few useful tips.

• Compiling dlls in release mode is the better supported than the static libraries, however the debug dll for 1.2 does not work, it should be fixed in 1.3 but I haven't tested yet that version
• Visual studio might fail to pick up the 64bit compiler option, compiling from the command line is recommended
• Building can take a lot of memory, I don't remember the exact number, but I think some people have trouble with machines with 8GB of RAM and less

Building can take a lot of memory, I don't remember the exact number, but I think some people have trouble with machines with 8GB of RAM and less

If we try to use "-Wl,--no-keep-memory" option in order to compile successfully Tensorflow with Cmake on 8GB of RAM and less system?