parallelization questions
Dear Anthony,
There are several levels of parallelism in gmshfem and who-does-what is not so clear to me. Solving is parallel through the blas (I'm using openblas) used by mumps and setting export OPENBLAS_NUM_THREADS=X seems to work. Assembly relies on openmp. When I set e.g. OPENBLAS_NUM_THREADS=8 and export OMP_NUM_THREADS=8, my program gets really slow. In fact, 8 instances of the code seem to be running... Do you know what I am doing wrong?
Kind regards, Guillaume