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

[Jay Foad via llvm-branch-commits]

https://github.com/jayfoad commented:

Looks reasonable to me now.

It also looks like this could be cleaned up: now that `LowerGlobalAddress` handles `G->getOffset()` for all LOCAL/REGION addresses, maybe `isOffsetFoldingLegal` should return true for all LOCAL/REGION addresses too instead of just for named barriers? But that should probably be a separate PR, especially if it affects codegen for normal LDS accesses.

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