Skip to content

[SLP] Compute a shuffle mask for getGatherCost #85330

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 6 commits into from
Mar 15, 2024
Merged

[SLP] Compute a shuffle mask for getGatherCost #85330

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
e330495
[SLP] Compute a shuffle mask for getGatherCost
preames Mar 14, 2024
cb5ced0
Update llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
preames Mar 14, 2024
b86bc2d
compute the correct shuffle mask
preames Mar 15, 2024
836ef4c
Update llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
preames Mar 15, 2024
7aa4fd2
Update llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
preames Mar 15, 2024
41152f6
Update llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
preames Mar 15, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 14 additions & 7 deletions llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10507,7 +10507,7 @@ InstructionCost BoUpSLP::getGatherCost(ArrayRef<Value *> VL,
// Check if the same elements are inserted several times and count them as
// shuffle candidates.
APInt ShuffledElements = APInt::getZero(VL.size());
DenseSet<Value *> UniqueElements;
DenseMap<Value *, unsigned> UniqueElements;
constexpr TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
InstructionCost Cost;
auto EstimateInsertCost = [&](unsigned I, Value *V) {
Expand All @@ -10516,27 +10516,34 @@ InstructionCost BoUpSLP::getGatherCost(ArrayRef<Value *> VL,
TTI->getVectorInstrCost(Instruction::InsertElement, VecTy, CostKind,
I, Constant::getNullValue(VecTy), V);
};
SmallVector<int> ShuffleMask(VL.size(), PoisonMaskElem);
for (unsigned I = 0, E = VL.size(); I < E; ++I) {
Value *V = VL[I];
// No need to shuffle duplicates for constants.
if ((ForPoisonSrc && isConstant(V)) || isa<UndefValue>(V)) {
ShuffledElements.setBit(I);
ShuffleMask[I] = isa<PoisonValue>(V) ? PoisonMaskElem : I;
continue;
}
if (!UniqueElements.insert(V).second) {
DuplicateNonConst = true;
ShuffledElements.setBit(I);

auto Res = UniqueElements.try_emplace(V, I);
if (Res.second) {
EstimateInsertCost(I, V);
ShuffleMask[I] = I;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This does not look correct. You generated identity mask here, which is not. Need to record the position of the first unique value and then use it here as index.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Clearly it is time for me to stop for the day. I wrote the correct code here - and then managed to commit the wrong version. Oops. Will push a corrected version tomorrow.

continue;
}
EstimateInsertCost(I, V);

DuplicateNonConst = true;
ShuffledElements.setBit(I);
ShuffleMask[I] = Res.first->second;
}
if (ForPoisonSrc)
Cost =
TTI->getScalarizationOverhead(VecTy, ~ShuffledElements, /*Insert*/ true,
/*Extract*/ false, CostKind);
if (DuplicateNonConst)
Cost +=
TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, VecTy);
Cost += TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc,
VecTy, ShuffleMask);
return Cost;
}

Expand Down