[llvm] [AMDGPU] Fix SIFoldOperands miscompiling values that leave a divergent loop (PR #203256)

[Jay Foad via llvm-commits]

jayfoad wrote:

Interesting. This is the temporal divergence case. I wonder why we have not found this earlier.

I wonder if would make more sense to implement the fix based on MachineUniformityAnalysis instead of LoopInfo.

https://github.com/llvm/llvm-project/pull/203256