"Please set CLANG_HOME"

[inferrna]

cocl on branch origin/working-on-incremental-function-building tensorflow-cl on branch tensorflow-cl

INFO: From Compiling tensorflow/core/kernels/gather_functor_gpu.cu.cc:

Please set CLANG_HOME

cocl built with CLANG_HOME=/usr/lib/llvm-3.8 and installed into /usr/local/

[hughperkins]

Yes, I need to integrate the latest cuda-on-cl with tensorflow. I havent done that yet.

[hughperkins]

This might be addressed in https://github.com/hughperkins/cuda-on-cl/commit/d6ef4ee6c105fa50efffcb88cc2c0309c21c774f ? Eg, now, after doing sudo make install from the cuda-on-cl build directory, I can do:

 ~$ mkdir /tmp/foo
 ~$ cd /tmp/foo
 /tmp/foo$ env | grep CLANG
 /tmp/foo$ cp ~/git/cuda-on-cl/test/cocl/cuda_sample.cu .
 /tmp/foo$ cocl cuda_sample.cu 
 /tmp/foo$ ./cuda_sample 

In this latest version cocl wraps cocl_wrapped, and defines CLANG_HOME:

$ cat /usr/local/bin/cocl
export CLANG_HOME=/usr/lib/llvm-3.8_
/usr/local/bin/cocl_wrapped "$@"

(my clang_home is in a weird place, cos I moved it, just for fun :-P )

[hughperkins]

(Hmmm, still broken from tensorflow-cl though. let me fix that, and then update you)

[hughperkins]

(works slightly better with https://github.com/hughperkins/tensorflow-cl/commit/ef92f337c904d98035d7ce3d847a4159a3ffdd04 , but still doesnt quite build yet, I think)

[hughperkins]

I think it builds to completon again now in: 9ae5906

[hughperkins]

I think this is fixed now, right?