From f29d59aa00bc03f349f913e8daf984f41acbdb03 Mon Sep 17 00:00:00 2001
From: Arm Patinyasakdikul <arm.patinyasakdikul@amd.com>
Date: Thu, 10 Apr 2025 10:44:16 -0500
Subject: [PATCH] Add device synchronization before destroying proxy thread.
 (#1631)

This commit ensures that GPU finishes all kernel before destroying
communicator thread.

[ROCm/rccl commit: 52654e23010cb3ea9476fc73a470d35c012e1f66]
---
 projects/rccl/src/proxy.cc | 1 +
 1 file changed, 1 insertion(+)

diff --git a/projects/rccl/src/proxy.cc b/projects/rccl/src/proxy.cc
index 90f6a7568f..a0b86889ca 100644
--- a/projects/rccl/src/proxy.cc
+++ b/projects/rccl/src/proxy.cc
@@ -1702,6 +1702,7 @@ void* ncclProxyService(void* _args) {
   }
 
   // Wait for all operations to complete and stop progress thread before freeing any resource
+  hipDeviceSynchronize();
   if (ncclProxyProgressDestroy(proxyState) != ncclSuccess) {
     WARN("[Proxy Service] proxyDestroy failed");
   }