[llvm] [AMDGPU] Add branching to vector rebroadcast tests. (PR #95122)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm commented:

I think you need to have a side-effecting expression / call inside each half of the conditions.

Also, I was really talking about an IR transform that you cannot see by codegening. For example if I modify one case:

https://godbolt.org/z/rG5s1sh7o

We still have all of these vector elements live across blocks, and a wide phi

https://github.com/llvm/llvm-project/pull/95122