[llvm] [VPlan] First step towards VPlan cost modeling. (PR #92555)

[Florian Hahn via llvm-commits]

fhahn wrote:

@aeubanks thanks! Is it possible that the reproducers is over-reduced as it has an unused load. Would it be possible to get a reproducer where no instructions can trivially be removed? (e.g running `instcombine,loop-vectorize` when reducing?)

Also, are those failures from public code by any chance? If so I'd be happy to try building those locally to check.

https://github.com/llvm/llvm-project/pull/92555