[SLP] Provide an universal interface for FixedVectorType::get. NFC. #96845
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-transforms Author: Han-Kuan Chen (HanKuanChen) ChangesSLP vectorizes scalar type to vector type. In the future, we will try to make SLP vectorizes vector type to vector type. We add a getWidenedType as a helper function. For example, SLP will make the following code %v0 = load i32, ptr %in0, align 4 into a load <4 x i32>. The ScalarTy is i32 and VF is 4. In the future, SLP will make the following code %v0 = load <4 x i32>, ptr %in0, align 4 into a load <16 x i32>. The ScalarTy is <4 x i32> and VF is 4. reference: https://discourse.llvm.org/t/rfc-make-slp-vectorizer-revectorize-vector-instructions/79436 Patch is 29.16 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/96845.diff 1 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 974f966d46e81..32f264535bccb 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -231,6 +231,11 @@ static bool isValidElementType(Type *Ty) {
!Ty->isPPC_FP128Ty();
}
+/// \returns the vector type of ScalarTy based on vectorization factor.
+static FixedVectorType *getWidenedType(Type *ScalarTy, unsigned VF) {
+ return FixedVectorType::get(ScalarTy, VF);
+}
+
/// \returns True if the value is a constant (but not globals/constant
/// expressions).
static bool isConstant(Value *V) {
@@ -1457,8 +1462,7 @@ class BoUpSLP {
if (getUnderlyingObject(LI1->getPointerOperand()) ==
getUnderlyingObject(LI2->getPointerOperand()) &&
R.TTI->isLegalMaskedGather(
- FixedVectorType::get(LI1->getType(), NumLanes),
- LI1->getAlign()))
+ getWidenedType(LI1->getType(), NumLanes), LI1->getAlign()))
return LookAheadHeuristics::ScoreMaskedGatherCandidate;
return CheckSameEntryOrFail();
}
@@ -4059,7 +4063,7 @@ BoUpSLP::findReusedOrderedScalars(const BoUpSLP::TreeEntry &TE) {
int NumScalars = GatheredScalars.size();
if (!isValidElementType(ScalarTy))
return std::nullopt;
- auto *VecTy = FixedVectorType::get(ScalarTy, NumScalars);
+ auto *VecTy = getWidenedType(ScalarTy, NumScalars);
int NumParts = TTI->getNumberOfParts(VecTy);
if (NumParts == 0 || NumParts >= NumScalars)
NumParts = 1;
@@ -4403,7 +4407,7 @@ BoUpSLP::LoadsState BoUpSLP::canVectorizeLoads(
}
Order.clear();
- auto *VecTy = FixedVectorType::get(ScalarTy, Sz);
+ auto *VecTy = getWidenedType(ScalarTy, Sz);
// Check the order of pointer operands or that all pointers are the same.
bool IsSorted = sortPtrAccesses(PointerOps, ScalarTy, *DL, *SE, Order);
// FIXME: Reordering isn't implemented for non-power-of-2 nodes yet.
@@ -4522,7 +4526,7 @@ BoUpSLP::LoadsState BoUpSLP::canVectorizeLoads(
/*VariableMask=*/false, CommonAlignment, CostKind) +
VectorGEPCost - ScalarGEPCost;
InstructionCost VecLdCost = 0;
- auto *SubVecTy = FixedVectorType::get(ScalarTy, VF);
+ auto *SubVecTy = getWidenedType(ScalarTy, VF);
for (auto [I, LS] : enumerate(States)) {
auto *LI0 = cast<LoadInst>(VL[I * VF]);
switch (LS) {
@@ -4792,8 +4796,8 @@ BoUpSLP::getReorderingData(const TreeEntry &TE, bool TopToBottom) {
}
}
if (Sz == 2 && TE.getVectorFactor() == 4 &&
- TTI->getNumberOfParts(FixedVectorType::get(
- TE.Scalars.front()->getType(), 2 * TE.getVectorFactor())) == 1)
+ TTI->getNumberOfParts(getWidenedType(TE.Scalars.front()->getType(),
+ 2 * TE.getVectorFactor())) == 1)
return std::nullopt;
if (!ShuffleVectorInst::isOneUseSingleSourceMask(TE.ReuseShuffleIndices,
Sz)) {
@@ -4965,7 +4969,7 @@ BoUpSLP::getReorderingData(const TreeEntry &TE, bool TopToBottom) {
find_if(TE.Scalars, [](Value *V) { return !isConstant(V); });
if (It == TE.Scalars.begin())
return OrdersType();
- auto *Ty = FixedVectorType::get(TE.Scalars.front()->getType(), Sz);
+ auto *Ty = getWidenedType(TE.Scalars.front()->getType(), Sz);
if (It != TE.Scalars.end()) {
OrdersType Order(Sz, Sz);
unsigned Idx = std::distance(TE.Scalars.begin(), It);
@@ -5101,7 +5105,7 @@ void BoUpSLP::reorderTopToBottom() {
// to take into account their order when looking for the most used order.
if (TE->isAltShuffle()) {
VectorType *VecTy =
- FixedVectorType::get(TE->Scalars[0]->getType(), TE->Scalars.size());
+ getWidenedType(TE->Scalars[0]->getType(), TE->Scalars.size());
unsigned Opcode0 = TE->getOpcode();
unsigned Opcode1 = TE->getAltOpcode();
SmallBitVector OpcodeMask(getAltInstrMask(TE->Scalars, Opcode0, Opcode1));
@@ -6018,7 +6022,7 @@ bool BoUpSLP::areAltOperandsProfitable(const InstructionsState &S,
unsigned Opcode1 = S.getAltOpcode();
SmallBitVector OpcodeMask(getAltInstrMask(VL, Opcode0, Opcode1));
// If this pattern is supported by the target then consider it profitable.
- if (TTI->isLegalAltInstr(FixedVectorType::get(S.MainOp->getType(), VL.size()),
+ if (TTI->isLegalAltInstr(getWidenedType(S.MainOp->getType(), VL.size()),
Opcode0, Opcode1, OpcodeMask))
return true;
SmallVector<ValueList> Operands;
@@ -7326,7 +7330,7 @@ unsigned BoUpSLP::canMapToVector(Type *T) const {
if (!isValidElementType(EltTy))
return 0;
- uint64_t VTSize = DL->getTypeStoreSizeInBits(FixedVectorType::get(EltTy, N));
+ uint64_t VTSize = DL->getTypeStoreSizeInBits(getWidenedType(EltTy, N));
if (VTSize < MinVecRegSize || VTSize > MaxVecRegSize ||
VTSize != DL->getTypeStoreSizeInBits(T))
return 0;
@@ -7990,7 +7994,7 @@ void BoUpSLP::transformNodes() {
if (E.State != TreeEntry::Vectorize)
break;
Type *ScalarTy = E.getMainOp()->getType();
- auto *VecTy = FixedVectorType::get(ScalarTy, E.Scalars.size());
+ auto *VecTy = getWidenedType(ScalarTy, E.Scalars.size());
Align CommonAlignment = computeCommonAlignment<LoadInst>(E.Scalars);
// Check if profitable to represent consecutive load + reverse as strided
// load with stride -1.
@@ -8017,7 +8021,7 @@ void BoUpSLP::transformNodes() {
case Instruction::Store: {
Type *ScalarTy =
cast<StoreInst>(E.getMainOp())->getValueOperand()->getType();
- auto *VecTy = FixedVectorType::get(ScalarTy, E.Scalars.size());
+ auto *VecTy = getWidenedType(ScalarTy, E.Scalars.size());
Align CommonAlignment = computeCommonAlignment<StoreInst>(E.Scalars);
// Check if profitable to represent consecutive load + reverse as strided
// load with stride -1.
@@ -8087,7 +8091,7 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
InstructionCost getBuildVectorCost(ArrayRef<Value *> VL, Value *Root) {
if ((!Root && allConstant(VL)) || all_of(VL, IsaPred<UndefValue>))
return TTI::TCC_Free;
- auto *VecTy = FixedVectorType::get(ScalarTy, VL.size());
+ auto *VecTy = getWidenedType(ScalarTy, VL.size());
InstructionCost GatherCost = 0;
SmallVector<Value *> Gathers(VL.begin(), VL.end());
// Improve gather cost for gather of loads, if we can group some of the
@@ -8185,7 +8189,7 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
LI->getAlign(), LI->getPointerAddressSpace(),
CostKind, TTI::OperandValueInfo(), LI);
}
- auto *LoadTy = FixedVectorType::get(VL.front()->getType(), VF);
+ auto *LoadTy = getWidenedType(VL.front()->getType(), VF);
for (const std::pair<unsigned, LoadsState> &P : VectorizedStarts) {
auto *LI = cast<LoadInst>(VL[P.first]);
Align Alignment = LI->getAlign();
@@ -8223,7 +8227,7 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
// TODO: improve checks if GEPs can be vectorized.
Value *Ptr0 = PointerOps.front();
Type *ScalarTy = Ptr0->getType();
- auto *VecTy = FixedVectorType::get(ScalarTy, VF);
+ auto *VecTy = getWidenedType(ScalarTy, VF);
auto [ScalarGEPCost, VectorGEPCost] =
getGEPCosts(TTI, PointerOps, Ptr0, Instruction::GetElementPtr,
CostKind, ScalarTy, VecTy);
@@ -8356,22 +8360,22 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
if (*ShuffleKinds[Part] != TTI::SK_PermuteSingleSrc ||
!ShuffleVectorInst::isIdentityMask(
MaskSlice, std::max<unsigned>(NumElts, MaskSlice.size())))
- Cost += ::getShuffleCost(TTI, *ShuffleKinds[Part],
- FixedVectorType::get(ScalarTy, NumElts),
- MaskSlice);
+ Cost +=
+ ::getShuffleCost(TTI, *ShuffleKinds[Part],
+ getWidenedType(ScalarTy, NumElts), MaskSlice);
continue;
}
if (*RegShuffleKind != TTI::SK_PermuteSingleSrc ||
!ShuffleVectorInst::isIdentityMask(SubMask, EltsPerVector)) {
- Cost += ::getShuffleCost(TTI, *RegShuffleKind,
- FixedVectorType::get(ScalarTy, EltsPerVector),
- SubMask);
+ Cost +=
+ ::getShuffleCost(TTI, *RegShuffleKind,
+ getWidenedType(ScalarTy, EltsPerVector), SubMask);
}
for (int Idx : Indices) {
Cost += ::getShuffleCost(TTI, TTI::SK_ExtractSubvector,
- FixedVectorType::get(ScalarTy, NumElts),
+ getWidenedType(ScalarTy, NumElts),
std::nullopt, CostKind, Idx,
- FixedVectorType::get(ScalarTy, EltsPerVector));
+ getWidenedType(ScalarTy, EltsPerVector));
}
}
return Cost;
@@ -8505,9 +8509,8 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
unsigned SrcSz = R.DL->getTypeSizeInBits(EScalarTy);
if (DstSz > SrcSz)
CastOpcode = IsSigned ? Instruction::SExt : Instruction::ZExt;
- return TTI.getCastInstrCost(CastOpcode,
- FixedVectorType::get(ScalarTy, VF),
- FixedVectorType::get(EScalarTy, VF),
+ return TTI.getCastInstrCost(CastOpcode, getWidenedType(ScalarTy, VF),
+ getWidenedType(EScalarTy, VF),
TTI::CastContextHint::None, CostKind);
}
return TTI::TCC_Free;
@@ -8562,8 +8565,8 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
ExtraCost += GetNodeMinBWAffectedCost(*E, E->getVectorFactor()) +
GetNodeMinBWAffectedCost(*E2, E2->getVectorFactor());
}
- V1 = Constant::getNullValue(FixedVectorType::get(ScalarTy, CommonVF));
- V2 = getAllOnesValue(*R.DL, FixedVectorType::get(ScalarTy, CommonVF));
+ V1 = Constant::getNullValue(getWidenedType(ScalarTy, CommonVF));
+ V2 = getAllOnesValue(*R.DL, getWidenedType(ScalarTy, CommonVF));
} else if (!V1 && P2.isNull()) {
// Shuffle single entry node.
const TreeEntry *E = P1.get<const TreeEntry *>();
@@ -8583,7 +8586,7 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
CommonVF = E->Scalars.size();
}
ExtraCost += GetNodeMinBWAffectedCost(*E, CommonVF);
- V1 = Constant::getNullValue(FixedVectorType::get(ScalarTy, CommonVF));
+ V1 = Constant::getNullValue(getWidenedType(ScalarTy, CommonVF));
// Not identity/broadcast? Try to see if the original vector is better.
if (!E->ReorderIndices.empty() && CommonVF == E->ReorderIndices.size() &&
CommonVF == CommonMask.size() &&
@@ -8628,10 +8631,10 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
CommonVF = VF;
}
ExtraCost += GetValueMinBWAffectedCost(V1);
- V1 = Constant::getNullValue(FixedVectorType::get(ScalarTy, CommonVF));
+ V1 = Constant::getNullValue(getWidenedType(ScalarTy, CommonVF));
ExtraCost += GetNodeMinBWAffectedCost(
*E2, std::min(CommonVF, E2->getVectorFactor()));
- V2 = getAllOnesValue(*R.DL, FixedVectorType::get(ScalarTy, CommonVF));
+ V2 = getAllOnesValue(*R.DL, getWidenedType(ScalarTy, CommonVF));
} else if (!V1 && V2) {
// Shuffle vector and tree node.
unsigned VF = cast<FixedVectorType>(V2->getType())->getNumElements();
@@ -8657,9 +8660,9 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
}
ExtraCost += GetNodeMinBWAffectedCost(
*E1, std::min(CommonVF, E1->getVectorFactor()));
- V1 = Constant::getNullValue(FixedVectorType::get(ScalarTy, CommonVF));
+ V1 = Constant::getNullValue(getWidenedType(ScalarTy, CommonVF));
ExtraCost += GetValueMinBWAffectedCost(V2);
- V2 = getAllOnesValue(*R.DL, FixedVectorType::get(ScalarTy, CommonVF));
+ V2 = getAllOnesValue(*R.DL, getWidenedType(ScalarTy, CommonVF));
} else {
assert(V1 && V2 && "Expected both vectors.");
unsigned VF = cast<FixedVectorType>(V1->getType())->getNumElements();
@@ -8673,17 +8676,17 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
ExtraCost +=
GetValueMinBWAffectedCost(V1) + GetValueMinBWAffectedCost(V2);
if (V1->getType() != V2->getType()) {
- V1 = Constant::getNullValue(FixedVectorType::get(ScalarTy, CommonVF));
- V2 = getAllOnesValue(*R.DL, FixedVectorType::get(ScalarTy, CommonVF));
+ V1 = Constant::getNullValue(getWidenedType(ScalarTy, CommonVF));
+ V2 = getAllOnesValue(*R.DL, getWidenedType(ScalarTy, CommonVF));
} else {
if (cast<VectorType>(V1->getType())->getElementType() != ScalarTy)
- V1 = Constant::getNullValue(FixedVectorType::get(ScalarTy, CommonVF));
+ V1 = Constant::getNullValue(getWidenedType(ScalarTy, CommonVF));
if (cast<VectorType>(V2->getType())->getElementType() != ScalarTy)
- V2 = getAllOnesValue(*R.DL, FixedVectorType::get(ScalarTy, CommonVF));
+ V2 = getAllOnesValue(*R.DL, getWidenedType(ScalarTy, CommonVF));
}
}
- InVectors.front() = Constant::getNullValue(
- FixedVectorType::get(ScalarTy, CommonMask.size()));
+ InVectors.front() =
+ Constant::getNullValue(getWidenedType(ScalarTy, CommonMask.size()));
if (InVectors.size() == 2)
InVectors.pop_back();
return ExtraCost + BaseShuffleAnalysis::createShuffle<InstructionCost>(
@@ -8792,8 +8795,8 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
SameNodesEstimated = false;
if (NumParts != 1 && UniqueBases.size() != 1) {
UseVecBaseAsInput = true;
- VecBase = Constant::getNullValue(
- FixedVectorType::get(ScalarTy, CommonMask.size()));
+ VecBase =
+ Constant::getNullValue(getWidenedType(ScalarTy, CommonMask.size()));
}
return VecBase;
}
@@ -8821,7 +8824,7 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
return;
}
assert(!CommonMask.empty() && "Expected non-empty common mask.");
- auto *MaskVecTy = FixedVectorType::get(ScalarTy, Mask.size());
+ auto *MaskVecTy = getWidenedType(ScalarTy, Mask.size());
unsigned NumParts = TTI.getNumberOfParts(MaskVecTy);
if (NumParts == 0 || NumParts >= Mask.size())
NumParts = 1;
@@ -8838,7 +8841,7 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
return;
}
assert(!CommonMask.empty() && "Expected non-empty common mask.");
- auto *MaskVecTy = FixedVectorType::get(ScalarTy, Mask.size());
+ auto *MaskVecTy = getWidenedType(ScalarTy, Mask.size());
unsigned NumParts = TTI.getNumberOfParts(MaskVecTy);
if (NumParts == 0 || NumParts >= Mask.size())
NumParts = 1;
@@ -9042,7 +9045,7 @@ static SmallVector<Type *> buildIntrinsicArgTypes(const CallInst *CI,
continue;
}
}
- ArgTys.push_back(FixedVectorType::get(Arg->getType(), VF));
+ ArgTys.push_back(getWidenedType(Arg->getType(), VF));
}
return ArgTys;
}
@@ -9071,9 +9074,9 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
Type *OrigScalarTy = ScalarTy;
if (It != MinBWs.end())
ScalarTy = IntegerType::get(F->getContext(), It->second.first);
- auto *VecTy = FixedVectorType::get(ScalarTy, VL.size());
+ auto *VecTy = getWidenedType(ScalarTy, VL.size());
unsigned EntryVF = E->getVectorFactor();
- auto *FinalVecTy = FixedVectorType::get(ScalarTy, EntryVF);
+ auto *FinalVecTy = getWidenedType(ScalarTy, EntryVF);
bool NeedToShuffleReuses = !E->ReuseShuffleIndices.empty();
if (E->State == TreeEntry::NeedToGather) {
@@ -9169,7 +9172,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
unsigned SrcBWSz = DL->getTypeSizeInBits(UserScalarTy);
unsigned VecOpcode;
auto *UserVecTy =
- FixedVectorType::get(UserScalarTy, E->getVectorFactor());
+ getWidenedType(UserScalarTy, E->getVectorFactor());
if (BWSz > SrcBWSz)
VecOpcode = Instruction::Trunc;
else
@@ -9241,7 +9244,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
NumElts = ATy->getNumElements();
else
NumElts = AggregateTy->getStructNumElements();
- SrcVecTy = FixedVectorType::get(OrigScalarTy, NumElts);
+ SrcVecTy = getWidenedType(OrigScalarTy, NumElts);
}
if (I->hasOneUse()) {
Instruction *Ext = I->user_back();
@@ -9335,7 +9338,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
// need to shift the vector.
// Do not calculate the cost if the actual size is the register size and
// we can merge this shuffle with the following SK_Select.
- auto *InsertVecTy = FixedVectorType::get(ScalarTy, InsertVecSz);
+ auto *InsertVecTy = getWidenedType(ScalarTy, InsertVecSz);
if (!IsIdentity)
Cost += TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc,
InsertVecTy, Mask);
@@ -9351,7 +9354,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
buildUseMask(NumElts, InsertMask, UseMask::UndefsAsMask));
if (!InMask.all() && NumScalars != NumElts && !IsWholeSubvector) {
if (InsertVecSz != VecSz) {
- auto *ActualVecTy = FixedVectorType::get(ScalarTy, VecSz);
+ auto *ActualVecTy = getWidenedType(ScalarTy, VecSz);
Cost += TTI->getShuffleCost(TTI::SK_InsertSubvector, ActualVecTy,
std::nullopt, CostKind, OffsetBeg - Offset,
InsertVecTy);
@@ -9385,7 +9388,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
case Instruction::BitCast: {
auto SrcIt = MinBWs.find(getOperandEntry(E, 0));
Type *SrcScalarTy = VL0->getOperand(0)->getType();
- auto *SrcVecTy = FixedVectorType::get(SrcScalarTy, VL.size());
+ auto *SrcVecTy = getWidenedType(SrcScalarTy, VL.size());
unsigned Opcode = ShuffleOrOp;
unsigned VecOpcode = Opcode;
if (!ScalarTy->isFloatingPointTy() && !SrcScalarTy->isFloatingPointTy() &&
@@ -9395,7 +9398,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
if (SrcIt != MinBWs.end()) {
SrcBWSz = SrcIt->second.first;
SrcScalarTy = IntegerType::get(F->getContext(), SrcBWSz);
- SrcVecTy = FixedVectorType::get(SrcScalarTy, VL.size());
+ SrcVecTy = getWidenedType(SrcScalarTy, VL.size());
}
unsigned BWSz = DL->getTypeSizeInBits(ScalarTy);
if (BWSz == SrcBWSz) {
@@ -9702,7 +9705,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
E->getAltOp());
} else {
Type *SrcSclTy = E->getMainOp()->getOperand(0)->getType();
- auto *SrcTy = FixedVectorType::get(SrcSclTy, VL.size());
+ auto *SrcTy = getWidenedType(SrcSclTy, VL.size());
if (SrcSclTy->isIntegerTy() && ScalarTy->isIntegerTy()) {
auto SrcIt = MinBWs.find(getOperandEntry(E, 0));
unsigned BWSz = DL->getTypeSizeInBits(ScalarTy);
@@ -9711,7 +9714,7 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
if (SrcIt != MinBWs.end()) {
SrcBWSz = SrcIt->second.first;
SrcSclTy = IntegerType::get(SrcSclTy->getContext(), SrcBWSz);
- SrcTy = FixedVectorType::get(SrcSclTy, VL.size());
+ SrcTy = getWidenedType(SrcSclTy, VL.size());
}
if (BWSz <= SrcBWSz) {
if (BWSz < SrcBWSz)
@@ -10048,7 +10051,7 @@ Instr...
[truncated]
|
HanKuanChen
force-pushed
the
slp-getWidenedType
branch
3 times, most recently
from
01b84cf to
acf642b
Compare
HanKuanChen
force-pushed
the
slp-getWidenedType
branch
from
acf642b to
ca980dd
Compare
SLP vectorizes scalar type to vector type. In the future, we will try to make SLP vectorizes vector type to vector type. We add a getWidenedType as a helper function. For example, SLP will make the following code %v0 = load i32, ptr %in0, align 4 %v1 = load i32, ptr %in1, align 4 %v2 = load i32, ptr %in2, align 4 %v3 = load i32, ptr %in3, align 4 into a load <4 x i32>. The ScalarTy is i32 and VF is 4. In the future, SLP will make the following code %v0 = load <4 x i32>, ptr %in0, align 4 %v1 = load <4 x i32>, ptr %in1, align 4 %v2 = load <4 x i32>, ptr %in2, align 4 %v3 = load <4 x i32>, ptr %in3, align 4 into a load <16 x i32>. The ScalarTy is <4 x i32> and VF is 4. reference: https://discourse.llvm.org/t/rfc-make-slp-vectorizer-revectorize-vector-instructions/79436
HanKuanChen
force-pushed
the
slp-getWidenedType
branch
from
ca980dd to
d7c87a5
Compare
aaryanshukla
pushed a commit
to aaryanshukla/llvm-project
that referenced
this pull request
Jul 14, 2024
…lvm#96845) SLP vectorizes scalar type to vector type. In the future, we will try to make SLP vectorizes vector type to vector type. We add a getWidenedType as a helper function. For example, SLP will make the following code %v0 = load i32, ptr %in0, align 4 %v1 = load i32, ptr %in1, align 4 %v2 = load i32, ptr %in2, align 4 %v3 = load i32, ptr %in3, align 4 into a load <4 x i32>. The ScalarTy is i32 and VF is 4. In the future, SLP will make the following code %v0 = load <4 x i32>, ptr %in0, align 4 %v1 = load <4 x i32>, ptr %in1, align 4 %v2 = load <4 x i32>, ptr %in2, align 4 %v3 = load <4 x i32>, ptr %in3, align 4 into a load <16 x i32>. The ScalarTy is <4 x i32> and VF is 4. reference: https://discourse.llvm.org/t/rfc-make-slp-vectorizer-revectorize-vector-instructions/79436
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
SLP vectorizes scalar type to vector type. In the future, we will try to make SLP vectorizes vector type to vector type. We add a getWidenedType as a helper function. For example, SLP will make the following code
%v0 = load i32, ptr %in0, align 4
%v1 = load i32, ptr %in1, align 4
%v2 = load i32, ptr %in2, align 4
%v3 = load i32, ptr %in3, align 4
into a load <4 x i32>. The ScalarTy is i32 and VF is 4. In the future, SLP will make the following code
%v0 = load <4 x i32>, ptr %in0, align 4
%v1 = load <4 x i32>, ptr %in1, align 4
%v2 = load <4 x i32>, ptr %in2, align 4
%v3 = load <4 x i32>, ptr %in3, align 4
into a load <16 x i32>. The ScalarTy is <4 x i32> and VF is 4.
reference: https://discourse.llvm.org/t/rfc-make-slp-vectorizer-revectorize-vector-instructions/79436