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

[Alexander Kornienko via llvm-commits]

alexfh wrote:

The test case can easily be over-reduced. The interestingness test I used is that a known-good version of `opt -passes=loop-vectorize` succeeds and the same `opt -passes=loop-vectorize` after this commit crashes. Any advice on what I could do to verify there are no dead loads? `opt -passes=slp-vectorizer,verify` still succeeds on both reduced inputs.

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