Build tensorflow with gpu support on OSX Sierra

tensorflow-gpu-sierra

disable sip in recovery mode
  start recovery mode with Cmd + R on startup
  start terminal
  csrutil disable
  
download and install cuda 8
download cuDNN v6.0 for cuda 8 (register first)

install xcode 7.2
  current xcode version is not supported
  https://srikanthpagadala.github.io/notes/2016/11/07/enable-gpu-support-for-tensorflow-on-macos
    • download Xcode 7.2.1 from the apple developer website
    • create a new directory /Applications/XCode7.2.1/
    • copy the entire XCode.App inside /Applications/XCode7.2.1
    • run sudo xcode-select -s /Applications/XCode7.2.1/Xcode.app/

move cuda libs
  cd /Users/tobi/Downloads
  sudo mv -v cuda/lib/libcudnn* /usr/local/cuda/lib
  sudo mv -v cuda/include/cudnn.h /usr/local/cuda/include

clone git
  git clone --recurse-submodules https://github.com/tensorflow/tensorflow
  git checkout remotes/origin/r1.1

add to .bash_profile
  export CUDA_HOME=/usr/local/cuda
  export DYLD_LIBRARY_PATH=/usr/local/cuda/lib:/usr/local/cuda/extras/CUPTI/lib    
  export LD_LIBRARY_PATH=$DYLD_LIBRARY_PATH
  export PATH=$CUDA_HOME/bi:$DYLD_LIBRARY_PATH:$PATH

activate venv
  source /Users/tobi/python3/bin/activate

uninstall old tensorflow version
  pip uninstall tensorflow
  pip uninstall tensorflow-gpu

avoid missing rpath error
https://github.com/tensorflow/tensorflow/issues/4187
  cd /Users/tobi/tensorflow/
  touch -t YYYYMMDDhhmm.SS genrule-setup.sh

configure
  check your gpu prerequisites:
  http://stackoverflow.com/questions/33651810/the-minimum-required-cuda-capability-is-3-5
  https://developer.nvidia.com/cuda-gpus?nocache=1492985895

  TF_UNOFFICIAL_SETTING=1 ./configure

build
  bazel build --config=cuda --config=opt --action_env PATH --action_env LD_LIBRARY_PATH --action_env DYLD_LIBRARY_PATH //tensorflow/tools/pip_package:build_pip_package

Thank you for posting this! Have you attempted to build r1.2, despite no official support?

Thanks for this, minor typo in environmental variables

export PATH=$CUDA_HOME/bin:$DYLD_LIBRARY_PATH:$PATH

Hi,

I am exactly doing the steps but I am failing on the last step of the .configure, I am getting two errors:

1. error loading package 'tensorflow/stream_executor': Encountered error while reading extension file 'cuda/build_defs.bzl': no such package '@local_config_cuda//cuda': Traceback (most recent call last):
2. Auto-Configuration Error: Cannot find line containing '#define CUDNN_MAJOR' in /usr/local/cuda/include/cudnn.h

How can I solve this, any suggests?