SWDEV-324411 - Use blit kernel for copyBufferRect if atomic is not supported

Change-Id: I2e110fd3418117ee9c7ede379244d2c6c4f248b7


[ROCm/clr commit: ed7737564e]

projects/clr/rocclr/device/rocm/rocblit.cpp

@@ -1585,8 +1585,8 @@ bool KernelBlitManager::copyBufferRect(device::Memory& srcMemory, device::Memory
   bool rejected = false;
 
   // Fall into the ROC path for rejected transfers
-  if (setup_.disableCopyBufferRect_ ||
+  if (dev().info().pcie_atomics_ && (setup_.disableCopyBufferRect_ ||
-      srcMemory.isHostMemDirectAccess() || dstMemory.isHostMemDirectAccess()) {
+      srcMemory.isHostMemDirectAccess() || dstMemory.isHostMemDirectAccess())) {
     result = DmaBlitManager::copyBufferRect(srcMemory, dstMemory, srcRectIn, dstRectIn, sizeIn, entire);
 
     if (result) {

projects/clr/rocclr/device/rocm/rocvirtual.cpp

@@ -2781,7 +2781,7 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
           break;
         }
         case amd::KernelParameterDescriptor::HiddenHostcallBuffer: {
-          if (roc_device_.info().pcie_atomics_) {
+          if (dev().info().pcie_atomics_) {
             uintptr_t buffer = reinterpret_cast<uintptr_t>(
               roc_device_.getOrCreateHostcallBuffer(gpu_queue_, coopGroups, cuMask_));
             if (!buffer) {
@@ -2936,8 +2936,8 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
 
     // Check for group memory overflow
     //! @todo Check should be in HSA - here we should have at most an assert
-    assert(roc_device_.info().localMemSizePerCU_ > 0);
+    assert(dev().info().localMemSizePerCU_ > 0);
-    if (ldsUsage > roc_device_.info().localMemSizePerCU_) {
+    if (ldsUsage > dev().info().localMemSizePerCU_) {
       LogError("No local memory available\n");
       return false;
     }