From 06d9ce1a021ba4d43dcd5693fce769f97be8db13 Mon Sep 17 00:00:00 2001
From: Peter Boyle <paboyle@ph.ed.ac.uk>
Date: Thu, 4 Aug 2022 13:35:56 -0400
Subject: [PATCH] Synch ranks on node here for GPU - GPU memcopy

---
 Grid/communicator/Communicator_mpi3.cc | 1 +
 1 file changed, 1 insertion(+)

diff --git a/Grid/communicator/Communicator_mpi3.cc b/Grid/communicator/Communicator_mpi3.cc
index 5c009890..fef4ea1f 100644
--- a/Grid/communicator/Communicator_mpi3.cc
+++ b/Grid/communicator/Communicator_mpi3.cc
@@ -406,6 +406,7 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
 void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list,int dir)
 {
   acceleratorCopySynchronise();
+  StencilBarrier();// Synch shared memory on a single nodes
 
   int nreq=list.size();