[llvm] [AMDGPU] CodeGen for 64-bit buffer atomic cmpswap intrinsics (PR #70475)

[Piotr Sobczak via llvm-commits]

https://github.com/piotrAMD approved this pull request.

LGTM, thanks.

Looks we'll need to add support for *_atomic_fcmpswap/*_atomic_cmpswap_f32 too at some point? Can't see any codegen tests for them.

https://github.com/llvm/llvm-project/pull/70475