[llvm] [LV] Invalidate SCEV values in the scalar loop after loop vectorization (PR #69886)

[Nikita Popov via llvm-commits]

nikic wrote:

Thanks for the explanation. As I understand it, the root issue is that SCEV looks through trivial LCSSA phi nodes, and we currently fail to fully invalidate when they become non-trivial. If this understand is correct, I believe it would be sufficient to invalidate SCEVUnknowns used by the LCSSA phi and defined in the loop. I've put up https://github.com/llvm/llvm-project/pull/69909 based on that idea.

https://github.com/llvm/llvm-project/pull/69886