P4 to Git Change 1182228 by gandryey@gera-dev-w7 on 2015/08/19 12:22:39

EPR #419072 - [OpenCL2.0] Enable 16MB large on device queues
	- Add mask_groups argument into the library for the empty slot spreading

Affected files ...

... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/devenq.h#11 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/enqueue.cl#10 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/eprep.cl#6 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/events.cl#4 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/schedule.cl#11 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#520 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusched.hpp#18 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#373 edit

rocclr/runtime/device/gpu/gpuvirtual.cpp

@@ -358,6 +358,7 @@ VirtualGPU::createVirtualQueue(uint deviceQueueSize)
     header->aql_slot_mask   = vaBase + slotMaskOffs;
     header->wait_size       = dev().settings().numWaitEvents_;
     header->arg_size        = dev().info().maxParameterSize_ + 64;
+    header->mask_groups     = maskGroups_;
     vqHeader_ = new AmdVQueueHeader;
     if (NULL == vqHeader_) {
         return false;
@@ -1938,7 +1939,6 @@ VirtualGPU::submitKernelInternalHSA(
             param->parentAQL = vmParentWrap;
             param->dedicatedQueue = dev().settings().useDeviceQueue_;
             param->useATC = dev().settings().svmFineGrainSystem_;
-            param->mask_groups = maskGroups_;
 
             // Fill the scratch buffer information
             if (hsaKernel.prog().maxScratchRegs() > 0) {