[llvm] [AMDGPU] Improve uniform argument handling in InstCombineIntrinsic (PR #105812)

[Sameer Sahasrabuddhe via llvm-commits]

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

LGTM, with an optional nit. I have been mulling over a similar idea. I would like to initiate some effort on a combiner pass that uses UniformityAnalysis in an AMD-specific way. I expect we would need to find one or more strategic places to run it ... it can benefit and also benefit from optimizations that transform the CFG.

https://github.com/llvm/llvm-project/pull/105812