Skip to content

Commit 02e9b0c

Browse files
author
Spencer Bryngelson
committed
fix: set MPICH_GPU_SUPPORT_ENABLED for GPU test dispatch
The Frontier template sets MPICH_GPU_SUPPORT_ENABLED=1 for GPU runs, but this was outside mpi_config so the direct test runner never set it. Without it, GPU-aware MPI (RDMA) cannot handle device pointers, causing Bus errors on multi-rank GPU tests.
1 parent 2a6b783 commit 02e9b0c

File tree

1 file changed

+1
-0
lines changed

1 file changed

+1
-0
lines changed

toolchain/mfc/test/case.py

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -309,6 +309,7 @@ def run(self, targets: List[Union[str, MFCTarget]], gpus: Set[int]) -> subproces
309309
gpu_ids = ",".join(str(g) for g in gpus)
310310
env["CUDA_VISIBLE_DEVICES"] = gpu_ids
311311
env["HIP_VISIBLE_DEVICES"] = gpu_ids
312+
env["MPICH_GPU_SUPPORT_ENABLED"] = "1"
312313

313314
# Resolve binary paths using the original (unmodified) params for slug
314315
slug_case = self.to_input_file()

0 commit comments

Comments
 (0)