Skip to content

Commit 53e1e9d

Browse files
committed
ompi/mpi_init: fix barrier
Relax CPU usage pressure from the application processes when doing modex and barrier in ompi_mpi_init. We see significant latencies in SLURM/pmix plugin barrier progress because app processes are aggressively call opal_progress pushing away daemon process doing collective progress. (cherry-ported from 0861884) Signed-off-by: Artem Polyakov <[email protected]>
1 parent 2959423 commit 53e1e9d

File tree

1 file changed

+2
-2
lines changed

1 file changed

+2
-2
lines changed

ompi/runtime/ompi_mpi_init.c

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -639,7 +639,7 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
639639
if (NULL != opal_pmix.fence_nb) {
640640
opal_pmix.fence_nb(NULL, opal_pmix_collect_all_data,
641641
fence_release, (void*)&active);
642-
OMPI_WAIT_FOR_COMPLETION(active);
642+
OMPI_LAZY_WAIT_FOR_COMPLETION(active);
643643
} else {
644644
opal_pmix.fence(NULL, opal_pmix_collect_all_data);
645645
}
@@ -809,7 +809,7 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
809809
if (NULL != opal_pmix.fence_nb) {
810810
opal_pmix.fence_nb(NULL, opal_pmix_collect_all_data,
811811
fence_release, (void*)&active);
812-
OMPI_WAIT_FOR_COMPLETION(active);
812+
OMPI_LAZY_WAIT_FOR_COMPLETION(active);
813813
} else {
814814
opal_pmix.fence(NULL, opal_pmix_collect_all_data);
815815
}

0 commit comments

Comments
 (0)