From fe7b36f3cbd81d888008da887b6497edf3f674dc Mon Sep 17 00:00:00 2001 From: German Andryeyev Date: Fri, 6 Oct 2023 13:39:41 -0400 Subject: [PATCH] SWDEV-424603 - Use OR for CPU wait request Make sure rocclr doesn't overwrite the client's request for a wait. Change-Id: I0addf18ea408b7f4ecaa1e04b2877cc0bbbfcc0d --- rocclr/platform/commandqueue.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/rocclr/platform/commandqueue.cpp b/rocclr/platform/commandqueue.cpp index ca66cdb7a8..a20aa74bca 100644 --- a/rocclr/platform/commandqueue.cpp +++ b/rocclr/platform/commandqueue.cpp @@ -126,7 +126,7 @@ void HostQueue::finish(bool cpu_wait) { bool force_marker = false; // Force CPU wait if profiler is enabled. Pytorch tests may use tracer's plugin and rely on // profiling information to be available right after finish. - cpu_wait = activity_prof::IsEnabled(OP_ID_DISPATCH); + cpu_wait |= activity_prof::IsEnabled(OP_ID_DISPATCH); if (AMD_DIRECT_DISPATCH && (command != nullptr) && !cpu_wait) { void* hw_event = (command->NotifyEvent() != nullptr) ? command->NotifyEvent()->HwEvent() : command->HwEvent();