[llvm] [AMDGPU] Canonicalize scalar bf16 more efficiently (PR #213789)

[Lukas Sommer via llvm-commits]

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

LGTM. Do we have the for GlobalISel already or are you planning on adding it there, too?

https://github.com/llvm/llvm-project/pull/213789