lomp | Little OpenMP Library | GPU library

Solved in the comments by Craig Estey. So here's my tutorial for compiling openMP on macOS.

1. install llvm with brew install llvm
2. install libomp with brew install libomp

Now, it clang and clang++ are calling llvm compilation, OpenMP should compile. But that's where I ran into the issue of missing ld linked.

1. use find /usr/local -xdev -name '*libomp*' to find where libomp is installed, probably at /usr/local/opt/libomp

Now when you compile/in your makefile, use clang instead of cc/gcc and add LDFLAGS += -L /libomp/lib and CFLAGS += -I/usr/local/opt/libomp/include to your make file. Compiling should now work.

When using brew to target the M1 natively, the brew installed clang is not in /usr/bin. The brew instructions show that the preferred location for it is /opt/homebrew. The clang in /usr/bin is a link to the Apple command line tools one; you can see that by using -v.

It seems you're using different clang (judging from the paths like /usr/local/opt/libomp/lib). Likely the one provided by conda. In order to perform build with LTO you need to use the compatible toolchain (e.g. build everything with Apple-provided toolchain or with conda-provided one).

You're best off using pgc++ to link. Compiling main.cpp with g++ is fine, but the PGI compiler at link time will implicitly include some initialization routines that are needed for OpenACC and CUDA interoperability. Without this initialization, you'll see runtime errors like this one.

Finally i find the solution...

1. In your venv:

pip install Cython setuptools

1. Find where is your gcc compiler... if you have already install do as follows, in other case find and install one...

$ mdfind gcc | grep gcc

1. Put the whole gcc directory in your project's venv. (path: /My_Project/venv/gcc)

2. and setup.py should be the following... (path: /My_Project/package/setup.py)