SWDEV-538195 - Introduce threshold for handler submission (#723)

- When doing device/stream sync, we can submit a handler which may
  introduce some host side delays. Use DEBUG_CLR_BATCH_CPU_SYNC_SIZE to
  batch commands for host wait. Default for HIP is 8 commands.
- Investigation is underway in ROCr but need to address this for now in
  HIP runtime.

[ROCm/clr commit: 9b045922a8]

projects/clr/hipamd/src/hip_device.cpp

@@ -32,6 +32,7 @@ namespace hip {
 
 // ================================================================================================
 hip::Stream* Device::NullStream(bool wait) {
+  ClPrint(amd::LOG_DEBUG, amd::LOG_WAIT, "NullStream %p, wait %d", null_stream_, wait);
   if (null_stream_ == nullptr) {
     amd::ScopedLock lock(lock_);
     if (null_stream_ == nullptr) {
@@ -188,6 +189,7 @@ void Device::WaitActiveStreams(hip::Stream* blocking_stream, bool wait_null_stre
 
   if (wait_null_stream) {
     if (null_stream_) {
+      ClPrint(amd::LOG_DEBUG, amd::LOG_WAIT, "Waiting on nullstream %p", null_stream_);
       waitForStream(null_stream_);
     }
   } else {
@@ -198,6 +200,7 @@ void Device::WaitActiveStreams(hip::Stream* blocking_stream, bool wait_null_stre
         ((active_stream->Flags() & hipStreamNonBlocking) == 0) &&
         // and it's not the current stream
         (active_stream != blocking_stream)) {
+        ClPrint(amd::LOG_DEBUG, amd::LOG_WAIT, "Waiting on active stream %p", active_stream);
         // Get the last valid command
         waitForStream(active_stream);
       }