Closed donnaaboise closed 6 years ago
Original comment by Donna Calhoun (Bitbucket: donnaaboise, GitHub: donnaaboise):
That would be great. But then if we "officially" support OpenMP, it would be good to test it with MPI. I tried, but found that the performance was quite slow, which was surprising. I am not really sure how MPI and OpenMP interact.
Would your student be able to do a few ForestClaw runs to possibly test the OpenMP/MPI combination? That would be great!
On the other hand, I found that at least on the desktop, MPI was faster than OpenMP, so I am not sure what is really gained by using OpenMP, unless we use it at the grid level, i.e. over i,j loops. Currently, I am only using it to parallelize the code at the patch level.
Original comment by Carsten Burstedde (Bitbucket: cburstedde, GitHub: cburstedde):
Ok I'll have my grad student program something.
I think it is easy enough to include openMP in the configure options - no need to make the build system more complicated.
Originally reported by: Donna Calhoun (Bitbucket: donnaaboise, GitHub: donnaaboise)
It is easy enough to add
-fopenmp
to the configure options, but we could also think about adding a check in the build system (although most compilers these days probably support openmp ...)