[llvm] [AMDGPU][gfx1250] Add wait_xcnt before any access that cannot be repeated (PR #168852)

[Jay Foad via llvm-commits]

================
@@ -776,6 +776,13 @@ std::optional<SIMemOpInfo> SIMemOpAccess::constructFromMIWithMMO(
     }
   }
 
+  // FIXME: The MMO of buffer atomic instructions does not always have an atomic
+  // ordering. We only need to handle VBUFFER atomics on GFX12+ so we can fix it
+  // here, but the lowering should really be cleaned up at some point.
+  if ((ST.getGeneration() >= GCNSubtarget::GFX12) && SIInstrInfo::isBUF(*MI) &&
----------------
jayfoad wrote:

Nit:
```suggestion
  if (ST.getGeneration() >= GCNSubtarget::GFX12 && SIInstrInfo::isBUF(*MI) &&
```

https://github.com/llvm/llvm-project/pull/168852