[llvm] [AMDGPU] Add pattern to select scalar ops for fshr with uniform operands (PR #165295)

[Akash Dutta via llvm-commits]

akadutta wrote:

> > > @aleksandar-amd has also been working on this, see #143551.
> > 
> > 
> > Thank you for pointing me to this PR. I took a look at it. To me, the 2 don't seem to overlap. This one simply adds a new set of scalar instructions for the fshr operation with uniform operands, whereas #143551 targets the rotr operation. Or am I missing something?
> 
> fshr and rotr are very similar operations. Yes #143551 only handles rotr, but with very small changes it could handle fshr as well (or instead). It is more sophisticated than the current patch because it tries to optimize the common case where the shift/rotate amount is constant.

Yes. That makes sense. Thank you for explaining.

https://github.com/llvm/llvm-project/pull/165295