[llvm] [LV] Implement integer part of VPDerivedIV cost model (PR #198252)

[Elvina Yakubova via llvm-commits]

ElvinaYakubova wrote:

Thanks for the explanation and for pointing to the PRs, this helps. 
I agree there's no need to revert this PR over this one benchmark - the cost model direction makes sense. If you can tweak it a bit so rerolled saxpy patterns like this still vectorize or being a little less pessimistic on step-5 derived IVs), that would be great for us.

https://github.com/llvm/llvm-project/pull/198252