SWDEV-79445 - OCL generic changes and code clean-up
- Add a workaround for incorrect profiling results, when release_mem can occur later in the pipeline and breaking the timeline sequence with dispatches. Ucode can't process RM packets and cs_partial_flush at the same time, thus RM packets are delayed up to 2 cs_patial_flush calls. A workaround adds CS partial flush after dispatch and switches to write_reg_mem for TS updates
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paltimestamp.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#66 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#37 edit
[ROCm/clr commit: 8f7976d7af]
foreman
e1316da844