/hello_tf_c_api

Neural Network TensorFlow C API.

Primary LanguageC++MIT LicenseMIT

Example TensorFlow C API

Example TensorFlow C API logo

Branch Linux/OSX Windows License Codacy
master Build Status Build status License Codacy Badge

Example how to run TensorFlow lib C API on Windows, Linux and macOS(Darwin).

Build example

Windows

git clone --depth 1 https://github.com/Neargye/hello_tf_c_api
cd hello_tf_c_api
mkdir build
cd build
cmake -G "Visual Studio 15 2017" -A x64 ..
cmake --build . --config Debug

Linux and macOS(Darwin)

git clone --depth 1 https://github.com/Neargye/hello_tf_c_api
cd hello_tf_c_api
mkdir build
cd build
cmake -G "Unix Makefiles" ..
cmake --build .

Remarks

  • After the build, you can find the TensorFlow lib in the folder hello_tf_c_api/tensorflow/lib, and header in hello_tf_c_api/tensorflow/include.
  • The tensorflow in the repository is compiled in x64 mode. Make sure that project target 64-bit platforms.
  • Make sure that the tensorflow lib is in Output Directory or either in the directory contained by the %PATH% environment variable.

Get tensorflow lib

For x64 CPU, you can download the tensorflow.so, tensorflow.dll and tensorflow.lib from https://github.com/Neargye/tensorflow/releases.

Or build lib which version you need from the sources, with CPU or GPU support.

Create .lib file from .dll for windows

Open the Visual Studio Command Prompt, you find its shortcut in "Start"->"Programs"->"Microsoft Visual Studio"->"Tools". Now run the dumpbin command to get a list of all exported functions of your dll:

dumpbin /exports yourpath/tensorflow.dll

This will print quite a bit of text to the console. However we are only interested in the functions:

    ordinal hint RVA      name

          1    0 028D4AB8 ?DEVICE_CPU@tensorflow@@3QEBDEB
          2    1 028D4AC0 ?DEVICE_GPU@tensorflow@@3QEBDEB
          3    2 028D4AC8 ?DEVICE_SYCL@tensorflow@@3QEBDEB
          4    3 028E1380 ?kDatasetGraphKey@GraphDatasetBase@tensorflow@@2QBDB
          5    4 028E1390 ?kDatasetGraphOutputNodeKey@GraphDatasetBase@tensorflow@@2QBDB
          6    5 03242488 ?tracing_engine_@Tracing@port@tensorflow@@0U?$atomic@PEAVEngine@Tracing@port@tensorflow@@@std@@A
          7    6 001996C0 TFE_ContextAddFunction
          8    7 00199710 TFE_ContextAddFunctionDef
          9    8 001997D0 TFE_ContextAsyncClearError
         10    9 001997E0 TFE_ContextAsyncWait
         11    A 00199830 TFE_ContextClearCaches
...

Now copy all those function names (only the names!) and paste them into a new textfile. Name the nextfile tensorflow.def and put the line “EXPORTS” at its top. My tensorflow.def file looks like this:

EXPORTS
?DEVICE_CPU@tensorflow@@3QEBDEB
?DEVICE_GPU@tensorflow@@3QEBDEB
?DEVICE_SYCL@tensorflow@@3QEBDEB
?kDatasetGraphKey@GraphDatasetBase@tensorflow@@2QBDB
?kDatasetGraphOutputNodeKey@GraphDatasetBase@tensorflow@@2QBDB
?tracing_engine_@Tracing@port@tensorflow@@0U?$atomic@PEAVEngine@Tracing@port@tensorflow@@@std@@A
TFE_ContextAddFunction
TFE_ContextAddFunctionDef
TFE_ContextAsyncClearError
TFE_ContextAsyncWait
TFE_ContextClearCaches
...

Now from that definition file, we can finally create the .lib file. We use the “lib” tool for this, so run this command in your Visual Studio Command Prompt:

lib /def:yourpath/tensorflow.def /OUT:yourpath/tensorflow.lib /MACHINE:X64

/MACHINE:X64 - fow x64 build, and /MACHINE:X86 for x32 build.

Link tensorflow lib

CMakeLists.txt

link_directories(yourpath/to/tensorflow) # path to tensorflow lib
... # other
target_link_libraries(<target> <PRIVATE|PUBLIC|INTERFACE> tensorflow)

Visual Studio

"Project"->"Properties"->Configuration Properties"->"Linker"->"Additional Dependencies" and add path to your tensorflow.lib as a next line.

Make sure that the tensorflow.dll is in Output Directory (by default, this is Debug\Release under your project's folder) or either in the directory contained by the %PATH% environment variable.

Licensed under the MIT License