Do concurrent adventures: all the roads lead to GPUs

Advocacy
1

I spent some time two weeks ago writing some code to test out some portability of CPU and GPU code. I made some cool graphs and learned a lot. I put it together into a blog post: Performance Portability of Do Concurrent in Fortran :: Sueños Guajiros: HPC and random stuff

This is a “first draft” on a potential post for the Fortran lang website for GPU programming using native Fortran + a little of OpenMP.

If you find typos forgive me, I wrote this in Markdown with no help of a spell check and I am not a native speaker of the English language.

10 Likes
2

This is great, @jorgeg. Thank you. If I can think of suggestions, I’ll just raise these directly through gh issue on the repo.

3

@jorgeg by any chance have you experimented with the combo GNU + nvptx offloading for those tests?

4

Not really yet! I think I do have a version of GNU that I compiled with offloading support…let me check later today

Wait, can I do do concurrent on GPUs with gfortran? I thought it was only openmp

5

Not sure, I tried a few tests with openmp target with gnu 12 or 13 but haven’t dared to try with DC yet.

1 Like