P4 to Git Change 1497344 by gandryey@gera-w8 on 2017/12/21 11:12:49

SWDEV-79445 - OCL generic changes and code clean-up
	- Rename flushCUCaches() to addBarrier().

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#43 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#68 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#38 edit


[ROCm/clr commit: 013d49e0a6]

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

@@ -646,7 +646,7 @@ bool Resource::create(MemoryType memType, CreateParams* params) {
         return false;
       }
       Pal::BufferViewInfo viewInfo = {};
-      viewInfo.gpuAddr = memRef_->iMem()->Desc().gpuVirtAddr + offset();
+      viewInfo.gpuAddr = vmAddress() + offset();
       viewInfo.range = memRef_->iMem()->Desc().size;
       viewInfo.stride = elementSize();
       viewInfo.swizzledFormat.format = format;
@@ -760,7 +760,7 @@ bool Resource::create(MemoryType memType, CreateParams* params) {
         mapCount_++;
       }
       Pal::BufferViewInfo viewInfo = {};
-      viewInfo.gpuAddr = memRef_->iMem()->Desc().gpuVirtAddr + offset();
+      viewInfo.gpuAddr = vmAddress() + offset();
       viewInfo.range = memRef_->iMem()->Desc().size;
       viewInfo.stride = elementSize();
       viewInfo.swizzledFormat.format = format;
@@ -1058,7 +1058,7 @@ bool Resource::create(MemoryType memType, CreateParams* params) {
     desc_.cardMemory_ = false;
     if ((nullptr != params) && (nullptr != params->owner_) &&
         (nullptr != params->owner_->getSvmPtr())) {
-      params->owner_->setSvmPtr(reinterpret_cast<void*>(memRef_->iMem()->Desc().gpuVirtAddr));
+      params->owner_->setSvmPtr(reinterpret_cast<void*>(vmAddress()));
     }
     return true;
   }
@@ -1265,7 +1265,7 @@ bool Resource::partialMemCopyTo(VirtualGPU& gpu, const amd::Coord3D& srcOrigin,
 
   if (gpu.validateSdmaOverlap(*this, dstResource)) {
     // Note: PAL should insert a NOP into the command buffer for synchronization
-    gpu.flushCUCaches();
+    gpu.addBarrier();
   }
 
   Pal::ImageLayout imgLayout = {};

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

@@ -416,7 +416,7 @@ void VirtualGPU::MemoryDependency::validate(VirtualGPU& gpu, const Memory* memor
 
   if (maxMemObjectsInQueue_ == 0) {
     // Flush cache
-    gpu.flushCUCaches();
+    gpu.addBarrier();
     return;
   }
 
@@ -452,7 +452,7 @@ void VirtualGPU::MemoryDependency::validate(VirtualGPU& gpu, const Memory* memor
   if (flushL1Cache) {
     // Flush cache
     if (!gpu.profiling()) {
-        gpu.flushCUCaches();
+        gpu.addBarrier();
     }
 
     // Clear memory dependency state
@@ -1985,7 +1985,7 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
     // Note: This a workaround for incorrect results reported with release_mem packet,
     // when the packet can be processed later after this dispatch and including extra time
     if (profiling() || state_.profileEnabled_) {
-      flushCUCaches();
+      addBarrier();
     }
     eventEnd(MainEngine, gpuEvent);
 
@@ -2127,7 +2127,7 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
                         gpuDefQueue->schedParamIdx_,
                         gpuDefQueue->vqHeader_->aql_slot_num / (DeviceQueueMaskSize * maskGroups_));
       const static bool FlushL2 = true;
-      gpuDefQueue->flushCUCaches(FlushL2);
+      gpuDefQueue->addBarrier(FlushL2);
 
       // Get the address of PM4 template and add write it to params
       //! @note DMA flush must not occur between patch and the scheduler
@@ -2955,7 +2955,7 @@ bool VirtualGPU::processMemObjectsHSA(const amd::Kernel& kernel, const_address p
       if (!supportFineGrainedSystem) {
         return false;
       } else if (sync) {
-        flushCUCaches();
+        addBarrier();
         // Clear memory dependency state
         const static bool All = true;
         memoryDependency().clear(!All);
@@ -3001,7 +3001,7 @@ bool VirtualGPU::processMemObjectsHSA(const amd::Kernel& kernel, const_address p
         svmMem =
             amd::SvmManager::FindSvmBuffer(*reinterpret_cast<void* const*>(params + desc.offset_));
         if (!svmMem) {
-          flushCUCaches();
+          addBarrier();
           // Clear memory dependency state
           const static bool All = true;
           memoryDependency().clear(!All);

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

@@ -433,7 +433,7 @@ class VirtualGPU : public device::VirtualDevice {
   //! Returns queue, associated with VirtualGPU
   Queue& queue(EngineType id) const { return *queues_[id]; }
 
-  void flushCUCaches(bool flushL2 = false) const {
+  void addBarrier(bool flushL2 = false) const {
     Pal::BarrierInfo barrier = {};
     barrier.pipePointWaitCount = 1;
     Pal::HwPipePoint point = Pal::HwPipePostCs;