[llvm] [VPlan] Expand VPWidenIntOrFpInductionRecipe into separate recipes (PR #118638)

[Nikolay Panchenko via llvm-commits]

npanchen wrote:

> This will allow us to use EVL as the increment value in the backedge value to deal with vsetvl's 6.3.2, instead of VF, e.g. something like
> My thinking is that it shouldn't matter if it's a vp.add or a regular widened add since the phi value should be correct across all lanes for the next iteration?

I see, I did misunderstand you then. Yes, than it should be correct to use regular add there. However, it does make more sense to stick with same approach: either emit vp-intrinsics always out of vectorizer or to have a minimal subset + postprocessing pass after vectorizer.  Current vision is to do former.
By the way,  unary and binary evl-recipes are already supported https://github.com/llvm/llvm-project/pull/93854, so it should be fine to use that, right ?

https://github.com/llvm/llvm-project/pull/118638