[llvm] [LoopVectorize] Vectorize select-cmp reduction pattern for increasing integer induction variable (PR #67812)

[Florian Hahn via llvm-commits]

fhahn wrote:

It looks like the patch didn't correctly handle epilogue vectorization and caused a mis-compile in SPEC2017 on AArch64.

I pushed 0e528ac404e13ed2d952a2d83aaf8383293c851e to fix the issue and hopefully avoid a revert

https://github.com/llvm/llvm-project/pull/67812