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

[Mel Chen via llvm-commits]

Mel-Chen wrote:

> Thanks for following up on all the comments, the last remaining thing to clarify from my side is what costs we currently assign to the new reduction kinds?
> 
> We should probably have a few target-dependent tests that check cost-modeling.

Is the cost of VPInstruction::ComputeReductionResult currently accounted for? FindLastIV should only require additional computation in the middle block, similar to AnyOf.


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