SWDEV-279288 - Avoid profiling info for the sync barrier

- With direct disaptch profiling state is enabled to trigger the
callback on HSA signal. However ROCr has very low peformance on
the first call to get the profiling info. That impacts some tiny
performance tests.

Change-Id: Idacd1b10a473fcfb5feef3074b7191d35743f769

rocclr/device/rocm/rocvirtual.cpp

@@ -110,7 +110,10 @@ static unsigned extractAqlBits(unsigned v, unsigned pos, unsigned width) {
 
 // ================================================================================================
 void Timestamp::checkGpuTime() {
-  if (HwProfiling()) {
+  if (HwProfiling() &&
+      // Avoid profiling data for the sync barrier, in tiny performance tests the first call
+      // to ROCr is very slow and that also affects the overall performance of the callback thread
+      (command().GetBatchHead() == nullptr)) {
     uint64_t  start = std::numeric_limits<uint64_t>::max();
     uint64_t  end = 0;