From c002dc86a364852859ca9642698fcfc5edf22c9d Mon Sep 17 00:00:00 2001 From: "Tang, Cheng" Date: Thu, 7 Oct 2021 10:09:16 -0700 Subject: [PATCH] set mpi group init flag after add group (#9293) --- .../orttraining/core/framework/communication/mpi/mpi_context.cc | 2 ++ 1 file changed, 2 insertions(+) diff --git a/orttraining/orttraining/core/framework/communication/mpi/mpi_context.cc b/orttraining/orttraining/core/framework/communication/mpi/mpi_context.cc index ef54afee7a..87da7477c8 100644 --- a/orttraining/orttraining/core/framework/communication/mpi/mpi_context.cc +++ b/orttraining/orttraining/core/framework/communication/mpi/mpi_context.cc @@ -169,6 +169,8 @@ void MPIContext::AddMPIGroup(WorkerGroupType group_type, WorkerGroup& group) { ORT_ENFORCE(this->mpi_groups_[group_type].communicator != MPI_COMM_NULL, "Failed to add new MPI group for worker group: ", DistributedRunContext::GetInstance().GetWorkerGroupName(group_type)); + //set the group initialized flag + this->mpi_groups_[group_type].is_group_initialized = true; #else ORT_THROW("ORT must be built with MPI to add ", DistributedRunContext::GetInstance().GetWorkerGroupName(group_type), " with group id: ", group.group_id); #endif