conda-forge / gromacs-feedstock

A conda-smithy repository for gromacs.
BSD 3-Clause "New" or "Revised" License
0 stars 9 forks source link

Enable the CUDA-MPI build #36

Closed xiki-tempula closed 9 months ago

xiki-tempula commented 10 months ago

Checklist

https://github.com/conda-forge/gromacs-feedstock/issues/35

conda-forge-webservices[bot] commented 10 months ago

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

xiki-tempula commented 10 months ago

@conda-forge-admin, please rerender

github-actions[bot] commented 10 months ago

Hi! This is the friendly automated conda-forge-webservice.

I tried to rerender for you, but it looks like there was nothing to do.

This message was generated by GitHub actions workflow run https://github.com/conda-forge/gromacs-feedstock/actions/runs/7034764408.

mabraham commented 10 months ago

This would work as you intend only if the openmpi package has suitable cuda awareness already built in. Does it?

If not then automated or manual tests would pass because GROMACS will fall back onto the CPU if the openmpi linked at runtime does not assert that it has cuda awareness. That check can be bypassed with env var GMX_FORCE_GPU_AWARE_MPI but that would only work if the testing machine actually has compatible GPUs. I'd be surprised if they did, so someone will need to test this build manually.

xiki-tempula commented 10 months ago

This MPI seems to be cuda aware?

-- Performing Test HAVE_MPI_EXT
-- Performing Test HAVE_MPI_EXT - Success
-- Performing Test MPI_SUPPORTS_CUDA_AWARE_DETECTION
-- Performing Test MPI_SUPPORTS_CUDA_AWARE_DETECTION - Success
-- Checking for MPI_SUPPORTS_CUDA_AWARE_DETECTION - yes

https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=838665&view=logs&j=4f922444-fdfe-5dcf-b824-02f86439ef14

xiki-tempula commented 10 months ago

I built this package locally and it seems to be working perfectly fine with GMX_ENABLE_DIRECT_GPU_COMM=True and MPI cuda replex.

mabraham commented 9 months ago

I assume there is no need to bump the build number as this MR is just adding a new build