[llvm] Late temporal divergence lowering for SDAG (PR #67033)

[Sameer Sahasrabuddhe via llvm-commits]

================

----------------
ssahasra wrote:

Does the test have to be so big? It's really difficult to see which value is of interest in the LLVM IR. So if the FileCheck patterns are auto updated in the future, it will be difficult to determine if the tested thing is still correct. Also, in the future, it is possible that the test breaks for some completely different reason, that is not related to uniformity.

I would strongly recommended a much simpler hand-coded test which demonstrates what the actual pattern of interest.

https://github.com/llvm/llvm-project/pull/67033