I'm working on an osx machine, which I have recently learned leads the g++ command back to clang. I have a custom op I want to use in tensorflow, and want to use the openmp capabilities of a certain library to gain performance. Unfortunately clang doesn't support openmp it seems, so I installed gcc through homebrew and compiled my library using that. The compilation went fine, however when loading the .so file using tf.load_op_library, I get "segmentation fault: 11"
This is without the -fopenmp option either.
The code worked fine using clang, but as soon as I switch to gcc 4.9, 5, or 6, I get this segfault.
Has anyone else had this issue? Should I start an issue on the tensorflow github page?