P4 to Git Change 1966866 by vsytchen@vsytchen-ocl-win10 on 2019/07/12 16:08:01

SWDEV-192353 - 12% Performance drop observed while running ROC_OCL_Perf_CompubenchCL_GPU_W64 only on Vega10 with Win7

	1. Reduce CmdAllocator sub-allocation size to 4KB.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#137 edit


[ROCm/clr commit: 3669ba964e]

projects/clr/rocclr/runtime/device/pal/palvirtual.cpp

@@ -785,7 +785,7 @@ bool VirtualGPU::create(bool profiling, uint deviceQueueSize, uint rtCUs,
 
   createInfo.allocInfo[Pal::GpuScratchMemAlloc].allocHeap = Pal::GpuHeapInvisible;
   createInfo.allocInfo[Pal::GpuScratchMemAlloc].allocSize = 64 * Ki;
-  createInfo.allocInfo[Pal::GpuScratchMemAlloc].suballocSize = 64 * Ki;
+  createInfo.allocInfo[Pal::GpuScratchMemAlloc].suballocSize = 4 * Ki;
 
   Pal::Result result;
   size_t cmdAllocSize = dev().iDev()->GetCmdAllocatorSize(createInfo, &result);