aboutsummaryrefslogtreecommitdiff
path: root/contrib/llvm-project/llvm/lib/Transforms/Vectorize
diff options
context:
space:
mode:
authorDimitry Andric <dim@FreeBSD.org>2022-02-18 22:41:20 +0000
committerDimitry Andric <dim@FreeBSD.org>2022-05-14 11:45:57 +0000
commitd56accc7c3dcc897489b6a07834763a03b9f3d68 (patch)
tree918f41a708218122215937f4ab4e68b1a942da68 /contrib/llvm-project/llvm/lib/Transforms/Vectorize
parent8885dff6cec52378084211fdd9366a73833eceee (diff)
parent7eff647615f93a9aaff1997e1880b195dc3aabe6 (diff)
Diffstat (limited to 'contrib/llvm-project/llvm/lib/Transforms/Vectorize')
-rw-r--r--contrib/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp35
-rw-r--r--contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp179
2 files changed, 38 insertions, 176 deletions
diff --git a/contrib/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/contrib/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 3290439ecd07..21c16f07e237 100644
--- a/contrib/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/contrib/llvm-project/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1701,6 +1701,11 @@ public:
private:
unsigned NumPredStores = 0;
+ /// Convenience function that returns the value of vscale_range iff
+ /// vscale_range.min == vscale_range.max or otherwise returns the value
+ /// returned by the corresponding TLI method.
+ Optional<unsigned> getVScaleForTuning() const;
+
/// \return An upper bound for the vectorization factors for both
/// fixed and scalable vectorization, where the minimum-known number of
/// elements is a power-of-2 larger than zero. If scalable vectorization is
@@ -5600,6 +5605,18 @@ ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
return MaxVF;
}
+Optional<unsigned> LoopVectorizationCostModel::getVScaleForTuning() const {
+ if (TheFunction->hasFnAttribute(Attribute::VScaleRange)) {
+ auto Attr = TheFunction->getFnAttribute(Attribute::VScaleRange);
+ auto Min = Attr.getVScaleRangeMin();
+ auto Max = Attr.getVScaleRangeMax();
+ if (Max && Min == Max)
+ return Max;
+ }
+
+ return TTI.getVScaleForTuning();
+}
+
bool LoopVectorizationCostModel::isMoreProfitable(
const VectorizationFactor &A, const VectorizationFactor &B) const {
InstructionCost CostA = A.Cost;
@@ -5624,7 +5641,7 @@ bool LoopVectorizationCostModel::isMoreProfitable(
// Improve estimate for the vector width if it is scalable.
unsigned EstimatedWidthA = A.Width.getKnownMinValue();
unsigned EstimatedWidthB = B.Width.getKnownMinValue();
- if (Optional<unsigned> VScale = TTI.getVScaleForTuning()) {
+ if (Optional<unsigned> VScale = getVScaleForTuning()) {
if (A.Width.isScalable())
EstimatedWidthA *= VScale.getValue();
if (B.Width.isScalable())
@@ -5673,7 +5690,7 @@ VectorizationFactor LoopVectorizationCostModel::selectVectorizationFactor(
#ifndef NDEBUG
unsigned AssumedMinimumVscale = 1;
- if (Optional<unsigned> VScale = TTI.getVScaleForTuning())
+ if (Optional<unsigned> VScale = getVScaleForTuning())
AssumedMinimumVscale = VScale.getValue();
unsigned Width =
Candidate.Width.isScalable()
@@ -5885,8 +5902,20 @@ LoopVectorizationCostModel::selectEpilogueVectorizationFactor(
return Result;
}
+ // If MainLoopVF = vscale x 2, and vscale is expected to be 4, then we know
+ // the main loop handles 8 lanes per iteration. We could still benefit from
+ // vectorizing the epilogue loop with VF=4.
+ ElementCount EstimatedRuntimeVF = MainLoopVF;
+ if (MainLoopVF.isScalable()) {
+ EstimatedRuntimeVF = ElementCount::getFixed(MainLoopVF.getKnownMinValue());
+ if (Optional<unsigned> VScale = getVScaleForTuning())
+ EstimatedRuntimeVF *= VScale.getValue();
+ }
+
for (auto &NextVF : ProfitableVFs)
- if (ElementCount::isKnownLT(NextVF.Width, MainLoopVF) &&
+ if (((!NextVF.Width.isScalable() && MainLoopVF.isScalable() &&
+ ElementCount::isKnownLT(NextVF.Width, EstimatedRuntimeVF)) ||
+ ElementCount::isKnownLT(NextVF.Width, MainLoopVF)) &&
(Result.Width.isScalar() || isMoreProfitable(NextVF, Result)) &&
LVP.hasPlanWithVF(NextVF.Width))
Result = NextVF;
diff --git a/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 15b349f53fd9..25bf69729c70 100644
--- a/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -471,36 +471,17 @@ static bool isValidForAlternation(unsigned Opcode) {
return true;
}
-static InstructionsState getSameOpcode(ArrayRef<Value *> VL,
- unsigned BaseIndex = 0);
-
-/// Checks if the provided operands of 2 cmp instructions are compatible, i.e.
-/// compatible instructions or constants, or just some other regular values.
-static bool areCompatibleCmpOps(Value *BaseOp0, Value *BaseOp1, Value *Op0,
- Value *Op1) {
- return (isConstant(BaseOp0) && isConstant(Op0)) ||
- (isConstant(BaseOp1) && isConstant(Op1)) ||
- (!isa<Instruction>(BaseOp0) && !isa<Instruction>(Op0) &&
- !isa<Instruction>(BaseOp1) && !isa<Instruction>(Op1)) ||
- getSameOpcode({BaseOp0, Op0}).getOpcode() ||
- getSameOpcode({BaseOp1, Op1}).getOpcode();
-}
-
/// \returns analysis of the Instructions in \p VL described in
/// InstructionsState, the Opcode that we suppose the whole list
/// could be vectorized even if its structure is diverse.
static InstructionsState getSameOpcode(ArrayRef<Value *> VL,
- unsigned BaseIndex) {
+ unsigned BaseIndex = 0) {
// Make sure these are all Instructions.
if (llvm::any_of(VL, [](Value *V) { return !isa<Instruction>(V); }))
return InstructionsState(VL[BaseIndex], nullptr, nullptr);
bool IsCastOp = isa<CastInst>(VL[BaseIndex]);
bool IsBinOp = isa<BinaryOperator>(VL[BaseIndex]);
- bool IsCmpOp = isa<CmpInst>(VL[BaseIndex]);
- CmpInst::Predicate BasePred =
- IsCmpOp ? cast<CmpInst>(VL[BaseIndex])->getPredicate()
- : CmpInst::BAD_ICMP_PREDICATE;
unsigned Opcode = cast<Instruction>(VL[BaseIndex])->getOpcode();
unsigned AltOpcode = Opcode;
unsigned AltIndex = BaseIndex;
@@ -533,57 +514,6 @@ static InstructionsState getSameOpcode(ArrayRef<Value *> VL,
continue;
}
}
- } else if (IsCmpOp && isa<CmpInst>(VL[Cnt])) {
- auto *BaseInst = cast<Instruction>(VL[BaseIndex]);
- auto *Inst = cast<Instruction>(VL[Cnt]);
- Type *Ty0 = BaseInst->getOperand(0)->getType();
- Type *Ty1 = Inst->getOperand(0)->getType();
- if (Ty0 == Ty1) {
- Value *BaseOp0 = BaseInst->getOperand(0);
- Value *BaseOp1 = BaseInst->getOperand(1);
- Value *Op0 = Inst->getOperand(0);
- Value *Op1 = Inst->getOperand(1);
- CmpInst::Predicate CurrentPred =
- cast<CmpInst>(VL[Cnt])->getPredicate();
- CmpInst::Predicate SwappedCurrentPred =
- CmpInst::getSwappedPredicate(CurrentPred);
- // Check for compatible operands. If the corresponding operands are not
- // compatible - need to perform alternate vectorization.
- if (InstOpcode == Opcode) {
- if (BasePred == CurrentPred &&
- areCompatibleCmpOps(BaseOp0, BaseOp1, Op0, Op1))
- continue;
- if (BasePred == SwappedCurrentPred &&
- areCompatibleCmpOps(BaseOp0, BaseOp1, Op1, Op0))
- continue;
- if (E == 2 &&
- (BasePred == CurrentPred || BasePred == SwappedCurrentPred))
- continue;
- auto *AltInst = cast<CmpInst>(VL[AltIndex]);
- CmpInst::Predicate AltPred = AltInst->getPredicate();
- Value *AltOp0 = AltInst->getOperand(0);
- Value *AltOp1 = AltInst->getOperand(1);
- // Check if operands are compatible with alternate operands.
- if (AltPred == CurrentPred &&
- areCompatibleCmpOps(AltOp0, AltOp1, Op0, Op1))
- continue;
- if (AltPred == SwappedCurrentPred &&
- areCompatibleCmpOps(AltOp0, AltOp1, Op1, Op0))
- continue;
- }
- if (BaseIndex == AltIndex) {
- assert(isValidForAlternation(Opcode) &&
- isValidForAlternation(InstOpcode) &&
- "Cast isn't safe for alternation, logic needs to be updated!");
- AltIndex = Cnt;
- continue;
- }
- auto *AltInst = cast<CmpInst>(VL[AltIndex]);
- CmpInst::Predicate AltPred = AltInst->getPredicate();
- if (BasePred == CurrentPred || BasePred == SwappedCurrentPred ||
- AltPred == CurrentPred || AltPred == SwappedCurrentPred)
- continue;
- }
} else if (InstOpcode == Opcode || InstOpcode == AltOpcode)
continue;
return InstructionsState(VL[BaseIndex], nullptr, nullptr);
@@ -4424,41 +4354,9 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
LLVM_DEBUG(dbgs() << "SLP: added a ShuffleVector op.\n");
// Reorder operands if reordering would enable vectorization.
- auto *CI = dyn_cast<CmpInst>(VL0);
- if (isa<BinaryOperator>(VL0) || CI) {
+ if (isa<BinaryOperator>(VL0)) {
ValueList Left, Right;
- if (!CI || all_of(VL, [](Value *V) {
- return cast<CmpInst>(V)->isCommutative();
- })) {
- reorderInputsAccordingToOpcode(VL, Left, Right, *DL, *SE, *this);
- } else {
- CmpInst::Predicate P0 = CI->getPredicate();
- CmpInst::Predicate AltP0 = cast<CmpInst>(S.AltOp)->getPredicate();
- CmpInst::Predicate AltP0Swapped = CmpInst::getSwappedPredicate(AltP0);
- Value *BaseOp0 = VL0->getOperand(0);
- Value *BaseOp1 = VL0->getOperand(1);
- // Collect operands - commute if it uses the swapped predicate or
- // alternate operation.
- for (Value *V : VL) {
- auto *Cmp = cast<CmpInst>(V);
- Value *LHS = Cmp->getOperand(0);
- Value *RHS = Cmp->getOperand(1);
- CmpInst::Predicate CurrentPred = CI->getPredicate();
- CmpInst::Predicate CurrentPredSwapped =
- CmpInst::getSwappedPredicate(CurrentPred);
- if (P0 == AltP0 || P0 == AltP0Swapped) {
- if ((P0 == CurrentPred &&
- !areCompatibleCmpOps(BaseOp0, BaseOp1, LHS, RHS)) ||
- (P0 == CurrentPredSwapped &&
- !areCompatibleCmpOps(BaseOp0, BaseOp1, RHS, LHS)))
- std::swap(LHS, RHS);
- } else if (!areCompatibleCmpOps(BaseOp0, BaseOp1, LHS, RHS)) {
- std::swap(LHS, RHS);
- }
- Left.push_back(LHS);
- Right.push_back(RHS);
- }
- }
+ reorderInputsAccordingToOpcode(VL, Left, Right, *DL, *SE, *this);
TE->setOperand(0, Left);
TE->setOperand(1, Right);
buildTree_rec(Left, Depth + 1, {TE, 0});
@@ -5390,8 +5288,7 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
((Instruction::isBinaryOp(E->getOpcode()) &&
Instruction::isBinaryOp(E->getAltOpcode())) ||
(Instruction::isCast(E->getOpcode()) &&
- Instruction::isCast(E->getAltOpcode())) ||
- (isa<CmpInst>(VL0) && isa<CmpInst>(E->getAltOp()))) &&
+ Instruction::isCast(E->getAltOpcode()))) &&
"Invalid Shuffle Vector Operand");
InstructionCost ScalarCost = 0;
if (NeedToShuffleReuses) {
@@ -5439,14 +5336,6 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
VecCost = TTI->getArithmeticInstrCost(E->getOpcode(), VecTy, CostKind);
VecCost += TTI->getArithmeticInstrCost(E->getAltOpcode(), VecTy,
CostKind);
- } else if (auto *CI0 = dyn_cast<CmpInst>(VL0)) {
- VecCost = TTI->getCmpSelInstrCost(E->getOpcode(), ScalarTy,
- Builder.getInt1Ty(),
- CI0->getPredicate(), CostKind, VL0);
- VecCost += TTI->getCmpSelInstrCost(
- E->getOpcode(), ScalarTy, Builder.getInt1Ty(),
- cast<CmpInst>(E->getAltOp())->getPredicate(), CostKind,
- E->getAltOp());
} else {
Type *Src0SclTy = E->getMainOp()->getOperand(0)->getType();
Type *Src1SclTy = E->getAltOp()->getOperand(0)->getType();
@@ -5463,29 +5352,6 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
E->Scalars, E->ReorderIndices, E->ReuseShuffleIndices,
[E](Instruction *I) {
assert(E->isOpcodeOrAlt(I) && "Unexpected main/alternate opcode");
- if (auto *CI0 = dyn_cast<CmpInst>(E->getMainOp())) {
- auto *AltCI0 = cast<CmpInst>(E->getAltOp());
- auto *CI = cast<CmpInst>(I);
- CmpInst::Predicate P0 = CI0->getPredicate();
- CmpInst::Predicate AltP0 = AltCI0->getPredicate();
- CmpInst::Predicate AltP0Swapped =
- CmpInst::getSwappedPredicate(AltP0);
- CmpInst::Predicate CurrentPred = CI->getPredicate();
- CmpInst::Predicate CurrentPredSwapped =
- CmpInst::getSwappedPredicate(CurrentPred);
- if (P0 == AltP0 || P0 == AltP0Swapped) {
- // Alternate cmps have same/swapped predicate as main cmps but
- // different order of compatible operands.
- return !(
- (P0 == CurrentPred &&
- areCompatibleCmpOps(CI0->getOperand(0), CI0->getOperand(1),
- I->getOperand(0), I->getOperand(1))) ||
- (P0 == CurrentPredSwapped &&
- areCompatibleCmpOps(CI0->getOperand(0), CI0->getOperand(1),
- I->getOperand(1), I->getOperand(0))));
- }
- return CurrentPred != P0 && CurrentPredSwapped != P0;
- }
return I->getOpcode() == E->getAltOpcode();
},
Mask);
@@ -6968,12 +6834,11 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
((Instruction::isBinaryOp(E->getOpcode()) &&
Instruction::isBinaryOp(E->getAltOpcode())) ||
(Instruction::isCast(E->getOpcode()) &&
- Instruction::isCast(E->getAltOpcode())) ||
- (isa<CmpInst>(VL0) && isa<CmpInst>(E->getAltOp()))) &&
+ Instruction::isCast(E->getAltOpcode()))) &&
"Invalid Shuffle Vector Operand");
Value *LHS = nullptr, *RHS = nullptr;
- if (Instruction::isBinaryOp(E->getOpcode()) || isa<CmpInst>(VL0)) {
+ if (Instruction::isBinaryOp(E->getOpcode())) {
setInsertPointAfterBundle(E);
LHS = vectorizeTree(E->getOperand(0));
RHS = vectorizeTree(E->getOperand(1));
@@ -6993,15 +6858,6 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
static_cast<Instruction::BinaryOps>(E->getOpcode()), LHS, RHS);
V1 = Builder.CreateBinOp(
static_cast<Instruction::BinaryOps>(E->getAltOpcode()), LHS, RHS);
- } else if (auto *CI0 = dyn_cast<CmpInst>(VL0)) {
- V0 = Builder.CreateCmp(CI0->getPredicate(), LHS, RHS);
- auto *AltCI = cast<CmpInst>(E->getAltOp());
- CmpInst::Predicate AltPred = AltCI->getPredicate();
- unsigned AltIdx =
- std::distance(E->Scalars.begin(), find(E->Scalars, AltCI));
- if (AltCI->getOperand(0) != E->getOperand(0)[AltIdx])
- AltPred = CmpInst::getSwappedPredicate(AltPred);
- V1 = Builder.CreateCmp(AltPred, LHS, RHS);
} else {
V0 = Builder.CreateCast(
static_cast<Instruction::CastOps>(E->getOpcode()), LHS, VecTy);
@@ -7026,29 +6882,6 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
E->Scalars, E->ReorderIndices, E->ReuseShuffleIndices,
[E](Instruction *I) {
assert(E->isOpcodeOrAlt(I) && "Unexpected main/alternate opcode");
- if (auto *CI0 = dyn_cast<CmpInst>(E->getMainOp())) {
- auto *AltCI0 = cast<CmpInst>(E->getAltOp());
- auto *CI = cast<CmpInst>(I);
- CmpInst::Predicate P0 = CI0->getPredicate();
- CmpInst::Predicate AltP0 = AltCI0->getPredicate();
- CmpInst::Predicate AltP0Swapped =
- CmpInst::getSwappedPredicate(AltP0);
- CmpInst::Predicate CurrentPred = CI->getPredicate();
- CmpInst::Predicate CurrentPredSwapped =
- CmpInst::getSwappedPredicate(CurrentPred);
- if (P0 == AltP0 || P0 == AltP0Swapped) {
- // Alternate cmps have same/swapped predicate as main cmps but
- // different order of compatible operands.
- return !(
- (P0 == CurrentPred &&
- areCompatibleCmpOps(CI0->getOperand(0), CI0->getOperand(1),
- I->getOperand(0), I->getOperand(1))) ||
- (P0 == CurrentPredSwapped &&
- areCompatibleCmpOps(CI0->getOperand(0), CI0->getOperand(1),
- I->getOperand(1), I->getOperand(0))));
- }
- return CurrentPred != P0 && CurrentPredSwapped != P0;
- }
return I->getOpcode() == E->getAltOpcode();
},
Mask, &OpScalars, &AltScalars);