[PATCH] D142015: [LV] Plan with and without FoldTailByMasking

Florian Hahn via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Apr 4 09:17:19 PDT 2023


fhahn added a comment.

In D142015#4223465 <https://reviews.llvm.org/D142015#4223465>, @dmgreen wrote:

> OK This now attempt to use the scheme @fhahn suggested where there are multiple cost models that get created with TailFold=true and TailFold=false.  It should hopefully work with allowing planning with and without tail folding (this patch), plus the predicated epilog vectorization in the followup.
>
> Some of what was in CostModel has moved into the Planner. This patch adds a reference from the vplan to the costmodel to make sure the correct model is used with the correct plan. Hopefully that can be reduced and removed eventually. It does manage to remove the VFCandidates array which is a nice little simplification.

I updated D143938 <https://reviews.llvm.org/D143938> and D142669 <https://reviews.llvm.org/D142669> recently and I think they should be in good shape now and should be ready for review. Rebasing the patch on top of them should hopefully simplify the diff. There are a few additional dependencies to remove a few more places that rely on the cost model after construction.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D142015/new/

https://reviews.llvm.org/D142015



More information about the llvm-commits mailing list