[llvm] [AMDGPU] Support image atomic no return instructions (PR #150742)

[Jay Foad via llvm-commits]

================
@@ -9130,13 +9143,22 @@ SDValue SITargetLowering::lowerImage(SDValue Op,
     DAG.setNodeMemRefs(NewNode, {MemRef});
   }
 
+  if (BaseOpcode->NoReturn) {
+    if (BaseOpcode->Atomic) {
+      SmallVector<SDValue, 2> RetVals;
----------------
jayfoad wrote:

Don't need a SmallVector for this. You can use a C array. Or just call getMergeValues({X, Y}, DL).

https://github.com/llvm/llvm-project/pull/150742