A (not yet enough) simple example combining cffi, numpy, and openMP.

Running (release mode)

$ OMP_NUM_THREADS=2 python3 -O test.py use_openmp
Building for performance with OpenMP...
Test 1 OK!
Test 2 OK! -- 5120x5120
numpy time -> 0.1006176 s/call
C time -> 0.0816459 s/call
C extension is 18.8552499761% faster

Running (debug mode)

python3 test.py

From the original documentation

Running On Mac OS X with OpenMP:

--without-multilib is mandatory for building with OpenMP support https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60670

brew reinstall gcc --without-multilib
USE_OPENMP=true CC=gcc-5 python3 -O test_add_array_f.py
USE_OPENMP=true CC=gcc-5 python3 -O test_add_array_i.py