[llvm] [SLP]Fix graph traversal in getSpillCost (PR #124984)
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Thu Jan 30 11:37:30 PST 2025
alexey-bataev wrote:
> Hm, I may have noticed one other thing. I think you want to replace "&*PrevInstIt != PrevInst" with !isVectorized(&*PrevInstIt) in the NumCall++ condition. It doesn't seem to cause a measurable test diff in tree, but I think the current code could over-count calls which are all part of one vector bundle.
>
> This change applied on it's own to the prior version of the code causes the same changes in the tests as this review. So it's not purely a nop.
Added the check for non-vectorized/vectorized as function call calls
https://github.com/llvm/llvm-project/pull/124984
More information about the llvm-commits
mailing list