[llvm] [AMDGPU] Optimize rotate instruction selection patterns (PR #143551)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> > I do have some stashed changes for this third approach, though the implementation isn’t fully finished yet. If you believe this direction is acceptable, I can continue and complete the patch.
> 
> I can't be 100% sure but I think it is worth trying.

Is this PR still relevant? Did you ever create a new PR with the new approach?

https://github.com/llvm/llvm-project/pull/143551