[llvm] [AMDGPU] Fold constant offsets into named barrier addresses (PR #205216)

[Jay Foad via llvm-commits]

================
@@ -9837,6 +9838,13 @@ SDValue SITargetLowering::lowerBUILD_VECTOR(SDValue Op,
 
 bool SITargetLowering::isOffsetFoldingLegal(
     const GlobalAddressSDNode *GA) const {
+  // Named barriers have fixed, non-relocated LDS addresses, so a constant
----------------
jayfoad wrote:

I don't think there's any need for this PR to be blocked on the BARRIER AS thing.

https://github.com/llvm/llvm-project/pull/205216