P4 to Git Change 1275745 by gujin@gujin-sc-win on 2016/06/02 13:34:16

SWDEV-76911 - Set output pipeline of multi-shader compilation in the same way driver sets. Fill output shader pointers only if they will be generated based on the input shaders, otherwise NULL.

	ReviewBoardURL = http://dxreview.amd.com/r/20059/

Affected files ...

... //depot/stg/sc/Src/Dev/TestEngine.cpp#512 edit

rocclr/runtime/device/gpu/gpuvirtual.cpp

@@ -472,7 +472,7 @@ VirtualGPU::create(bool profiling, uint rtCUs, uint  deviceQueueSize,
                 }
                 //!@todo This is not a generic solution and
                 // may have issues with > 8 queues
-                idx = index() % (dev().engines().numComputeRings() +
+                idx = index() % (dev().engines().numComputeRings() + 
                         dev().engines().numComputeRingsRT());
             }
             // hwRing_ should be set 0 if forced to have single scratch buffer
@@ -1839,23 +1839,11 @@ VirtualGPU::submitKernelInternalHSA(
             pKernelInfo = &kernelInfo;
         }
 
-        // Set up the dispatch information
-        KernelDispatchInfo dispatchInfo;
-        dispatchInfo.aqlPacket = aqlPkt;
-        dispatchInfo.mems = vmMems();
-        dispatchInfo.numMems = cal_.memCount_;
-        dispatchInfo.scratch = scratch;
-        dispatchInfo.scratchOffset = scratchOffset;
-        dispatchInfo.cpuAqlCode = hsaKernel.cpuAqlCode();
-        dispatchInfo.hsaQueueVA = hsaQueueMem_->vmAddress();
-        dispatchInfo.kernelInfo = pKernelInfo;
-        dispatchInfo.wavesPerSH = hsaKernel.getWavesPerSH(this);
-        dispatchInfo.lastDoppSubmission = kernel.parameters().getExecNewVcop();
-
         GpuEvent    gpuEvent;
         // Run AQL dispatch in HW
         eventBegin(MainEngine);
-        cs()->AqlDispatch(&dispatchInfo);
+        cs()->AqlDispatch(aqlPkt, vmMems(), cal_.memCount_, scratch, scratchOffset,
+            hsaKernel.cpuAqlCode(), hsaQueueMem_->vmAddress(), pKernelInfo, hsaKernel.getWavesPerSH(this));
         eventEnd(MainEngine, gpuEvent);
 
         if (dbgManager && (NULL != dbgManager->postDispatchCallBackFunc())) {
@@ -2155,7 +2143,7 @@ VirtualGPU::submitKernelInternal(
             }
 
             // Execute the kernel
-            if (gpuKernel.run(*this, &gpuEvent, lastRun, kernel.parameters().getExecNewVcop())) {
+            if (gpuKernel.run(*this, &gpuEvent, lastRun)) {
                 //! @todo A flush is necessary to make sure
                 // that 2 consecutive runs won't access to the same
                 // private/local memory. CAL has to generate cache flush