OSX Sierra Tensorflow build error: ld: file not found: #rpath/CUDA.framework/Versions/A/CUDA - macos

Question

I have followed the instruction in:
https://gist.github.com/notilas/a30e29ce514970e821a34153c1e78b3f
But cannot complete it.
OSX: Sierra
Tensorflow version 1.1.0 (Google says v1.2 does not support OSX CUDA)
CUDA Tool kit : 8.0
CUDNN : 6.0
Xcode : 7.2.1
Anaconda : 4.2 (Python version 3.5)
Error Log:
ERROR: /Users/so041e/ml/tensorflow/tensorflow/python/BUILD:2534:1:
Linking of rule '//tensorflow/python:_pywrap_tensorflow_internal.so'
failed: link_dynamic_library.sh failed: error executing command
external/bazel_tools/tools/cpp/link_dynamic_library.sh no ignored
ignored ignored
external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc
-shared -o ... (remaining 455 argument(s) skipped): com.google.devtools.build.lib.shell.BadExitStatusException: Process
exited with status 1.
clang: warning: argument unused during compilation: '-pthread'
ld: file not found: #rpath/CUDA.framework/Versions/A/CUDA for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
Target //tensorflow/tools/pip_package:build_pip_package failed to build
Use --verbose_failures to see the command lines of failed build steps.
[.bash_profile]
export PATH="/Users/so041e/anaconda/bin:$PATH"
export CUDA_HOME=/usr/local/cuda
export HOME=/Users/so041e
export PATH="$CUDA_HOME/bin:$PATH"
export DYLD_LIBRARY_PATH="/usr/local/cuda/lib:/Developer/NVIDIA/CUDA8.0/lib":$DYLD_LIBRARY_PATH
export LD_LIBRARY_PATH=$DYLD_LIBRARY_PATH
export PATH=$DYLD_LIBRARY_PATH:$PATH
export PATH="//anaconda/bin:$PATH"
Moved CUDNN lib and include to /user/local/cuda
sudo mv -v cuda/lib/libcudnn* /usr/local/cuda/lib
sudo mv -v cuda/include/cudnn.h /usr/local/cuda/include
Didn't use "vent" Just used single python 3.5 at this moment.
Tried both, but no difference.
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
bazel build --config=cuda //tensorflow/tools/pip_package:build_pip_package

Answer 1

This might be a bit late, but I had this exact same problem and I managed to fix it.
First, #rpath/CUDA.framework/Versions/A/CUDA is a dynamic library install name for libcuda.dylib, which is found in /usr/local/cuda/lib. So do
otool -l /usr/local/cuda/lib/libcuda.dylib
Check where you see #rpath/CUDA.framework/Versions/A/CUDA; on my system it was in the command LC_REEXPORT_DYLIB. From here, it seems dyld doesn't resolve the #rpath for the LC_REEXPORT_DYLIB command, only LC_LOAD*_DYLIB commands. Meaning, it looks for the literal path "#rpath/CUDA.framework/Versions/A/CUDA". So you're going to have to change that by doing
sudo install_name_tool -change #rpath/CUDA.framework/Versions/A/CUDA \
/Library/Frameworks/CUDA.framework/Versions/A/CUDA \
/usr/local/cuda/lib/libcuda.dylib
This should resolve your problem.
Now why your system (and mine) has this install name for libcuda.dylib? I have absolutely no clue.