[VPlan] Extract reverse mask from reverse accesses #155579
Conversation
Mel-Chen
requested review from
artagnon,
fhahn,
lukel97,
arcbbb,
alexey-bataev and
david-arm
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
| @@ -1338,248 +1338,49 @@ for.end: ; preds = %for.inc | |||
| ;} | |||
|
|
|||
| define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr nocapture readonly %trigger) local_unnamed_addr #0 { | |||
| ; AVX1-LABEL: @foo6( | |||
| // The legacy model have not computed the cost of reverse mask. | ||
| if (CostCtx.CM.Legal->isMaskRequired(UI)) | ||
| return true; |
There was a problem hiding this comment.
@fhahn This could lead to overestimating the cost.
Multiple VPWidenMemoryRecipes may be able to share a single reverse mask, but if each VPWidenMemoryRecipe computes the cost of the reverse mask separately, the cost would be overestimated.
| bool IsReverse = CostCtx.CM.getWideningDecision(UI, VF) == | ||
| LoopVectorizationCostModel::CM_Widen_Reverse; |
There was a problem hiding this comment.
@fhahn I think this is the most straightforward approach.
Currently, we could check whether the operand or user is a reverse operation, but in the future reverse operations might be simplified away, so relying on reverse operation is not a long-term approach.
The last option is to use the address for the check, since only reverse operations need to use VPVectorEndPointer.
What do you think?
llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse-mask4.ll
Outdated
Show resolved
Hide resolved
Following #146525, separate the reverse mask from reverse access recipes.
At the same time, remove the unused member variable
ReversefromVPWidenMemoryRecipe.This will help to reduce redundant reverse mask computations once VPlan-based CSE is supported.
Base on #146525