SWDEV-424603 - Force CPU wait if profiling

Some pytorch tests use a tracer plugin and rely on profiling information
to be reported right after hipDeviceSynchronize()

Change-Id: Ib021a1e7b1a30b3c24de72627c471810f7f7878d

rocclr/platform/commandqueue.cpp

@@ -124,6 +124,9 @@ void HostQueue::finish(bool cpu_wait) {
   // If command doesn't contain HW event and runtime didn't request CPU wait,
   // then force marker submit
   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);
   if (AMD_DIRECT_DISPATCH && (command != nullptr) && !cpu_wait) {
     void* hw_event =
       (command->NotifyEvent() != nullptr) ? command->NotifyEvent()->HwEvent() : command->HwEvent();