From 459b52158f118286fb4b2988e4778ee531b7ad87 Mon Sep 17 00:00:00 2001 From: David Addison Date: Thu, 12 Oct 2023 16:53:32 -0700 Subject: [PATCH] Added an MPI_Barrier() call after MPI_Bcast() for HCOLL issue [ROCm/rccl-tests commit: 1292b25553bd0384f2faa2965f9d82b99797a348] --- projects/rccl-tests/src/common.cu | 1 + 1 file changed, 1 insertion(+) diff --git a/projects/rccl-tests/src/common.cu b/projects/rccl-tests/src/common.cu index dcead4ddd4..8588047d78 100644 --- a/projects/rccl-tests/src/common.cu +++ b/projects/rccl-tests/src/common.cu @@ -924,6 +924,7 @@ testResult_t run() { } #ifdef MPI_SUPPORT MPI_Bcast(&ncclId, sizeof(ncclId), MPI_BYTE, 0, mpi_comm); + MPI_Barrier(MPI_COMM_WORLD); // Ensure Bcast is complete for HCOLL #endif int gpus[nGpus*nThreads]; cudaStream_t streams[nGpus*nThreads];