[LV] Add scalar load/stores to VPReplicateRecipe::computeCost #153218
Conversation
|
@llvm/pr-subscribers-vectorizers @llvm/pr-subscribers-llvm-transforms Author: David Sherwood (david-arm) ChangesAvoid calling getLegacyCost for single scalar loads and stores where the cost is trivial to calculate. Full diff: https://github.com/llvm/llvm-project/pull/153218.diff 1 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index e34cab117f321..b301aee6dad23 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -2975,6 +2975,22 @@ InstructionCost VPReplicateRecipe::computeCost(ElementCount VF,
Op2Info, Operands, UI, &Ctx.TLI) *
(isSingleScalar() ? 1 : VF.getFixedValue());
}
+ case Instruction::Load:
+ case Instruction::Store: {
+ if (isSingleScalar()) {
+ Type *ValTy = getLoadStoreType(UI);
+ Type *ScalarPtrTy = getLoadStorePointerOperand(UI)->getType();
+ const Align Alignment = getLoadStoreAlignment(UI);
+ unsigned AS = getLoadStoreAddressSpace(UI);
+ TTI::OperandValueInfo OpInfo = TTI::getOperandInfo(UI->getOperand(0));
+ InstructionCost ScalarMemOpCost = Ctx.TTI.getMemoryOpCost(
+ UI->getOpcode(), ValTy, Alignment, AS, CostKind, OpInfo, UI);
+ return ScalarMemOpCost + Ctx.TTI.getAddressComputationCost(ScalarPtrTy);
+ }
+ // TODO: See getMemInstScalarizationCost for how to handle vector and
+ // predicated cases.
+ break;
+ }
}
return Ctx.getLegacyCost(UI, VF);
|
|
Gentle ping. :) |
| UI->getOpcode(), ValTy, Alignment, AS, CostKind, OpInfo, UI); | ||
| return ScalarMemOpCost + Ctx.TTI.getAddressComputationCost(ScalarPtrTy); | ||
| } | ||
| // TODO: See getMemInstScalarizationCost for how to handle vector and |
There was a problem hiding this comment.
| // TODO: See getMemInstScalarizationCost for how to handle vector and | |
| // TODO: See getMemInstScalarizationCost for how to handle replicating and |
This code here won't handle vector cases?
| Type *ValTy = getLoadStoreType(UI); | ||
| Type *ScalarPtrTy = getLoadStorePointerOperand(UI)->getType(); |
There was a problem hiding this comment.
Can we get this from the VPlan type analysis?
Avoid calling getLegacyCost for single scalar loads and stores where the cost is trivial to calculate.
a4e0e92 to
fd7bf4b
Compare
| // TODO: Is this right? Operand 0 has a different meaning for loads and | ||
| // stores. |
There was a problem hiding this comment.
It is indeed confusing and the argument isn't documented, but it looks like in RISCV TTI and X86 TTI (https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/X86/X86TargetTransformInfo.cpp#L5241) the argument is used as it is passed here: stored value for stores, address for loads. AArch64 and ARM TTIs don't use it.
So it seems it is used as expected, even though the documentation of the interface could be improved
Avoid calling getLegacyCost for single scalar loads and stores where the cost is trivial to calculate.