P4 to Git Change 1973670 by gandryey@gera-win10 on 2019/07/25 11:27:11

SWDEV-79445 - OCL generic changes and code clean-up
	- Don't create an extra queue for DMA transfers when SDMA is disabled. That should allow to avoid useless sync operations.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palconstbuf.cpp#13 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palgpuopen.cpp#15 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#75 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paltimestamp.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paltimestamp.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#143 edit

rocclr/runtime/device/pal/palgpuopen.cpp

@@ -582,10 +582,11 @@ Pal::Result RgpCaptureMgr::EndRGPTrace(VirtualGPU* gpu) {
   // end the session itself
   if (result == Pal::Result::Success) {
     assert(trace_.gpa_session_ != nullptr);
+    EngineType engine = (gpu->dev().settings().disableSdma_) ? MainEngine : SdmaEngine;
     // Initiate SDMA copy
-    gpu->eventBegin(SdmaEngine);
-    result = trace_.gpa_session_->End(gpu->queue(SdmaEngine).iCmd());
-    gpu->eventEnd(SdmaEngine, trace_.end_event_);
+    gpu->eventBegin(engine);
+    result = trace_.gpa_session_->End(gpu->queue(engine).iCmd());
+    gpu->eventEnd(engine, trace_.end_event_);
   }
 
   // Submit the trace-end command buffer