Commit 0194492
Spencer Bryngelson
fix: add GPU flags to direct srun/flux and regenerate RDMA golden files
The direct MPI execution path was missing GPU binding flags that
the batch templates add outside of mpi_config. For multi-rank GPU
tests (e.g. RDMA MPI), this caused Bus errors on Frontier because
srun didn't bind GPUs to tasks.
Add --gpus-per-task/--gpu-bind to _mpi_cmd for srun, jsrun, and
flux when GPUs are in use, matching what each template does.
Also regenerate RDMA MPI golden files (FA4D8FEF, 1B300F28,
2C9844EF) which were missed when Nt was halved from 50 to 25,
since they require --rdma-mpi to be generated.1 parent 1ef4c77 commit 0194492
7 files changed
Lines changed: 155 additions & 107 deletions
File tree
- tests
- 1B300F28
- 2C9844EF
- FA4D8FEF
- toolchain/mfc/test
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.
0 commit comments