[llvm] [LoopVectorize] Fix double-application of FindIV reduction expression in epilogue (PR #219362)

[Florian Hahn via llvm-commits]

fhahn wrote:

> @fhahn I tested the issue reproducer on the upstream(main), and this doesn't crash anymore... so we can say this issue is already fixed in the current upstream.. Should we still continue with this PR ?

This is surprising, I am not aware of any changes that fixed the underlying issue specifically.  Perhaps another change is causing different vectorization decisions, but the underlying issue is still present?

https://github.com/llvm/llvm-project/pull/219362