SWDEV-193956 - [hipclang-vdi-rocm][perf]

~45% to 50% of Performance drop on rocBLAS_int8 test

Add support for active waits without blocking the host thread.

Change-Id: Ie7bb48dcafcb4c93d448bf74749b829b626c3578


[ROCm/clr commit: 0fc433e076]

projects/clr/rocclr/platform/command.cpp

@@ -190,12 +190,17 @@ bool Event::awaitCompletion() {
     }
 
     ClPrint(LOG_DEBUG, LOG_WAIT, "waiting for event %p to complete, current status %d", this, status_);
+    if (command().queue()->vdev()->ActiveWait()) {
+      while (status_ > CL_COMPLETE) {
+        amd::Os::yield();
+      }
+    } else {
+      ScopedLock lock(lock_);
 
-    ScopedLock lock(lock_);
-
-    // Wait until the status becomes CL_COMPLETE or negative.
-    while (status_ > CL_COMPLETE) {
-      lock_.wait();
+      // Wait until the status becomes CL_COMPLETE or negative.
+      while (status_ > CL_COMPLETE) {
+        lock_.wait();
+      }
     }
 
     ClPrint(LOG_DEBUG, LOG_WAIT, "event %p wait completed", this);