[llvm] [AMDGPU] Fold constants in 2-elt vector canonicalization (PR #214384)

[Shilei Tian via llvm-commits]

https://github.com/shiltian commented:

Now the results look much better. LGTM but I'll defer the stamp to @arsenm.

https://github.com/llvm/llvm-project/pull/214384