[llvm] [VPlan] Add initial VPScalarEvolution, use to get trip count SCEV (NFC) (PR #94464)
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Mon Jun 10 13:24:34 PDT 2024
fhahn wrote:
> This looks ok to me, but not yet clear how the new `VPScalarEvolution` class will be used inside `VPlanAnalysis`.
>
> Will it just be a helper class? If so, why not a static method? If not, then how will it be composed across different VPlans and the analysis part? I don't have a strong opinion here, just trying to understand the direction.
In the future, I am planning to extend it to support constructing SCEVs for any recipe (e.g. `VPInstruction::Add -> SE.getAddExpr({getSCEV(VPOp) for VPOp : Recipe->operands()})`).
Users are expected to be spread across new analyses and transforms. It could also be a single `getSCEVForVPValue(VPValue *VPV, ScalarEvolution &SE)` entry point instead of providing a class, which may be more convenient?
https://github.com/llvm/llvm-project/pull/94464
More information about the llvm-commits
mailing list