16#ifndef LLVM_CODEGEN_BASICTTIIMPL_H
17#define LLVM_CODEGEN_BASICTTIIMPL_H
87 T *thisT() {
return static_cast<T *
>(
this); }
96 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
100 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
119 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
121 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
134 "Can only extract subvectors from vectors");
136 assert((!isa<FixedVectorType>(VTy) ||
137 (
Index + NumSubElts) <=
138 (
int)cast<FixedVectorType>(VTy)->getNumElements()) &&
139 "SK_ExtractSubvector index out of range");
145 for (
int i = 0; i != NumSubElts; ++i) {
147 thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
149 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, SubVTy,
162 "Can only insert subvectors into vectors");
164 assert((!isa<FixedVectorType>(VTy) ||
165 (
Index + NumSubElts) <=
166 (
int)cast<FixedVectorType>(VTy)->getNumElements()) &&
167 "SK_InsertSubvector index out of range");
173 for (
int i = 0; i != NumSubElts; ++i) {
174 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, SubVTy,
177 thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
CostKind,
178 i +
Index,
nullptr,
nullptr);
185 return static_cast<const T *
>(
this)->getST();
190 return static_cast<const T *
>(
this)->getTLI();
212 bool IsGatherScatter,
216 if (isa<ScalableVectorType>(DataTy))
219 auto *VT = cast<FixedVectorType>(DataTy);
220 unsigned VF = VT->getNumElements();
236 VF * thisT()->getMemoryOpCost(Opcode, VT->getElementType(), Alignment,
242 Opcode == Instruction::Store,
CostKind);
256 VF * (thisT()->getCFInstrCost(Instruction::Br,
CostKind) +
257 thisT()->getCFInstrCost(Instruction::PHI,
CostKind));
260 return AddrExtractCost + MemoryOpCost + PackingCost + ConditionalCost;
275 unsigned *
Fast)
const {
318 std::pair<const Value *, unsigned>
341 bool HasBaseReg, int64_t Scale,
unsigned AddrSpace,
343 int64_t ScalableOffset = 0) {
358 Type *ScalarValTy)
const {
359 auto &&IsSupportedByTarget = [
this, ScalarMemTy, ScalarValTy](
unsigned VF) {
362 if (getTLI()->isOperationLegal(
ISD::STORE, VT) ||
372 while (VF > 2 && IsSupportedByTarget(VF))
412 int64_t Scale,
unsigned AddrSpace) {
451 unsigned &JumpTableSize,
461 unsigned N = SI.getNumCases();
466 bool IsJTAllowed = TLI->
areJTsAllowed(SI.getParent()->getParent());
472 APInt MaxCaseVal = SI.case_begin()->getCaseValue()->getValue();
473 APInt MinCaseVal = MaxCaseVal;
474 for (
auto CI : SI.cases()) {
475 const APInt &CaseVal = CI.getCaseValue()->getValue();
476 if (CaseVal.
sgt(MaxCaseVal))
477 MaxCaseVal = CaseVal;
478 if (CaseVal.
slt(MinCaseVal))
479 MinCaseVal = CaseVal;
485 for (
auto I : SI.cases())
486 Dests.
insert(
I.getCaseSuccessor());
495 if (
N < 2 || N < TLI->getMinimumJumpTableEntries())
498 (MaxCaseVal - MinCaseVal)
499 .getLimitedValue(std::numeric_limits<uint64_t>::max() - 1) + 1;
502 JumpTableSize =
Range;
518 if (!
TM.isPositionIndependent())
528 Triple TargetTriple =
TM.getTargetTriple();
566 case Instruction::SDiv:
567 case Instruction::SRem:
568 case Instruction::UDiv:
569 case Instruction::URem: {
618 else if (ST->getSchedModel().LoopMicroOpBufferSize > 0)
619 MaxOps = ST->getSchedModel().LoopMicroOpBufferSize;
626 if (isa<CallInst>(
I) || isa<InvokeInst>(
I)) {
636 <<
"advising against unrolling the loop because it "
689 std::optional<Value *>
692 bool &KnownBitsComputed) {
703 IC,
II, DemandedElts, UndefElts, UndefElts2, UndefElts3,
707 virtual std::optional<unsigned>
709 return std::optional<unsigned>(
713 virtual std::optional<unsigned>
715 std::optional<unsigned> TargetResult =
733 unsigned NumStridedMemAccesses,
734 unsigned NumPrefetches,
735 bool HasCall)
const {
737 NumPrefetches, HasCall);
769 const APInt &DemandedElts,
770 bool Insert,
bool Extract,
774 if (isa<ScalableVectorType>(InTy))
776 auto *Ty = cast<FixedVectorType>(InTy);
779 "Vector size mismatch");
783 for (
int i = 0, e = Ty->getNumElements(); i < e; ++i) {
784 if (!DemandedElts[i])
787 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, Ty,
790 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
801 if (isa<ScalableVectorType>(InTy))
803 auto *Ty = cast<FixedVectorType>(InTy);
806 return thisT()->getScalarizationOverhead(Ty, DemandedElts, Insert, Extract,
817 assert(Args.size() == Tys.
size() &&
"Expected matching Args and Tys");
821 for (
int I = 0,
E = Args.size();
I !=
E;
I++) {
829 if (!isa<Constant>(
A) && UniqueOperands.
insert(
A).second) {
830 if (
auto *VecTy = dyn_cast<VectorType>(Ty))
887 if (MTy == LK.second)
901 ArrayRef<const Value *> Args = std::nullopt,
902 const Instruction *CxtI =
nullptr) {
904 const TargetLoweringBase *TLI = getTLI();
905 int ISD = TLI->InstructionOpcodeToISD(Opcode);
906 assert(ISD &&
"Invalid opcode");
919 InstructionCost OpCost = (IsFloat ? 2 : 1);
921 if (TLI->isOperationLegalOrPromote(ISD, LT.second)) {
924 return LT.first * OpCost;
927 if (!TLI->isOperationExpand(ISD,
LT.second)) {
930 return LT.first * 2 * OpCost;
942 unsigned DivOpc = IsSigned ? Instruction::SDiv : Instruction::UDiv;
943 InstructionCost DivCost = thisT()->getArithmeticInstrCost(
944 DivOpc, Ty,
CostKind, Opd1Info, Opd2Info);
945 InstructionCost MulCost =
946 thisT()->getArithmeticInstrCost(Instruction::Mul, Ty,
CostKind);
947 InstructionCost SubCost =
948 thisT()->getArithmeticInstrCost(Instruction::Sub, Ty,
CostKind);
949 return DivCost + MulCost + SubCost;
954 if (isa<ScalableVectorType>(Ty))
960 if (
auto *VTy = dyn_cast<FixedVectorType>(Ty)) {
961 InstructionCost
Cost = thisT()->getArithmeticInstrCost(
966 SmallVector<Type *> Tys(
Args.size(), Ty);
989 (
Index + Mask.size()) <= (
size_t)NumSrcElts) {
997 Mask, NumSrcElts, NumSubElts,
Index)) {
998 if (
Index + NumSubElts > NumSrcElts)
1031 if (
auto *FVT = dyn_cast<FixedVectorType>(Tp))
1032 return getBroadcastShuffleOverhead(FVT,
CostKind);
1040 if (
auto *FVT = dyn_cast<FixedVectorType>(Tp))
1041 return getPermuteShuffleOverhead(FVT,
CostKind);
1045 cast<FixedVectorType>(SubTp));
1048 cast<FixedVectorType>(SubTp));
1062 assert(ISD &&
"Invalid opcode");
1066 TypeSize SrcSize = SrcLT.second.getSizeInBits();
1067 TypeSize DstSize = DstLT.second.getSizeInBits();
1068 bool IntOrPtrSrc = Src->isIntegerTy() || Src->isPointerTy();
1069 bool IntOrPtrDst = Dst->isIntegerTy() || Dst->isPointerTy();
1074 case Instruction::Trunc:
1079 case Instruction::BitCast:
1082 if (SrcLT.first == DstLT.first && IntOrPtrSrc == IntOrPtrDst &&
1086 case Instruction::FPExt:
1087 if (
I && getTLI()->isExtFree(
I))
1090 case Instruction::ZExt:
1091 if (TLI->
isZExtFree(SrcLT.second, DstLT.second))
1094 case Instruction::SExt:
1095 if (
I && getTLI()->isExtFree(
I))
1105 if (DstLT.first == SrcLT.first &&
1110 case Instruction::AddrSpaceCast:
1112 Dst->getPointerAddressSpace()))
1117 auto *SrcVTy = dyn_cast<VectorType>(Src);
1118 auto *DstVTy = dyn_cast<VectorType>(Dst);
1121 if (SrcLT.first == DstLT.first &&
1126 if (!SrcVTy && !DstVTy) {
1137 if (DstVTy && SrcVTy) {
1139 if (SrcLT.first == DstLT.first && SrcSize == DstSize) {
1142 if (Opcode == Instruction::ZExt)
1146 if (Opcode == Instruction::SExt)
1147 return SrcLT.first * 2;
1153 return SrcLT.first * 1;
1166 if ((SplitSrc || SplitDst) && SrcVTy->getElementCount().isVector() &&
1167 DstVTy->getElementCount().isVector()) {
1170 T *
TTI =
static_cast<T *
>(
this);
1173 (!SplitSrc || !SplitDst) ?
TTI->getVectorSplitCost() : 0;
1180 if (isa<ScalableVectorType>(DstVTy))
1185 unsigned Num = cast<FixedVectorType>(DstVTy)->getNumElements();
1187 Opcode, Dst->getScalarType(), Src->getScalarType(), CCH,
CostKind,
I);
1200 if (Opcode == Instruction::BitCast) {
1216 return thisT()->getVectorInstrCost(Instruction::ExtractElement, VecTy,
1233 assert(ISD &&
"Invalid opcode");
1242 assert(CondTy &&
"CondTy must exist");
1248 if (!(ValTy->
isVectorTy() && !LT.second.isVector()) &&
1252 return LT.first * 1;
1258 if (
auto *ValVTy = dyn_cast<VectorType>(ValTy)) {
1259 if (isa<ScalableVectorType>(ValTy))
1262 unsigned Num = cast<FixedVectorType>(ValVTy)->getNumElements();
1266 Opcode, ValVTy->getScalarType(), CondTy, VecPred,
CostKind,
I);
1288 Value *Op0 =
nullptr;
1289 Value *Op1 =
nullptr;
1290 if (
auto *IE = dyn_cast<InsertElementInst>(&
I)) {
1291 Op0 = IE->getOperand(0);
1292 Op1 = IE->getOperand(1);
1294 return thisT()->getVectorInstrCost(
I.getOpcode(), Val,
CostKind,
Index, Op0,
1300 const APInt &DemandedDstElts,
1303 "Unexpected size of DemandedDstElts.");
1321 Cost += thisT()->getScalarizationOverhead(SrcVT, DemandedSrcElts,
1324 Cost += thisT()->getScalarizationOverhead(ReplicatedVT, DemandedDstElts,
1336 assert(!Src->isVoidTy() &&
"Invalid type");
1338 if (getTLI()->getValueType(
DL, Src,
true) == MVT::Other)
1353 LT.second.getSizeInBits())) {
1359 if (Opcode == Instruction::Store)
1368 cast<VectorType>(Src), Opcode != Instruction::Store,
1369 Opcode == Instruction::Store,
CostKind);
1380 return getCommonMaskedMemoryOpCost(Opcode, DataTy, Alignment,
true,
false,
1385 const Value *
Ptr,
bool VariableMask,
1389 return getCommonMaskedMemoryOpCost(Opcode, DataTy, Alignment, VariableMask,
1394 const Value *
Ptr,
bool VariableMask,
1401 return thisT()->getGatherScatterOpCost(Opcode, DataTy,
Ptr, VariableMask,
1408 bool UseMaskForCond =
false,
bool UseMaskForGaps =
false) {
1411 if (isa<ScalableVectorType>(VecTy))
1414 auto *VT = cast<FixedVectorType>(VecTy);
1416 unsigned NumElts = VT->getNumElements();
1417 assert(Factor > 1 && NumElts % Factor == 0 &&
"Invalid interleave factor");
1419 unsigned NumSubElts = NumElts / Factor;
1424 if (UseMaskForCond || UseMaskForGaps)
1425 Cost = thisT()->getMaskedMemoryOpCost(Opcode, VecTy, Alignment,
1434 unsigned VecTySize = thisT()->getDataLayout().getTypeStoreSize(VecTy);
1454 unsigned NumLegalInsts =
divideCeil(VecTySize, VecTyLTSize);
1458 unsigned NumEltsPerLegalInst =
divideCeil(NumElts, NumLegalInsts);
1461 BitVector UsedInsts(NumLegalInsts,
false);
1462 for (
unsigned Index : Indices)
1463 for (
unsigned Elt = 0; Elt < NumSubElts; ++Elt)
1464 UsedInsts.
set((
Index + Elt * Factor) / NumEltsPerLegalInst);
1473 "Interleaved memory op has too many members");
1479 for (
unsigned Index : Indices) {
1480 assert(
Index < Factor &&
"Invalid index for interleaved memory op");
1481 for (
unsigned Elm = 0; Elm < NumSubElts; Elm++)
1482 DemandedLoadStoreElts.
setBit(
Index + Elm * Factor);
1485 if (Opcode == Instruction::Load) {
1495 SubVT, DemandedAllSubElts,
1497 Cost += Indices.
size() * InsSubCost;
1498 Cost += thisT()->getScalarizationOverhead(VT, DemandedLoadStoreElts,
1516 SubVT, DemandedAllSubElts,
1518 Cost += ExtSubCost * Indices.
size();
1519 Cost += thisT()->getScalarizationOverhead(VT, DemandedLoadStoreElts,
1524 if (!UseMaskForCond)
1529 Cost += thisT()->getReplicationShuffleCost(
1530 I8Type, Factor, NumSubElts,
1531 UseMaskForGaps ? DemandedLoadStoreElts : DemandedAllResultElts,
1539 if (UseMaskForGaps) {
1541 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::And, MaskVT,
1566 (
RetTy->isVectorTy() ? cast<VectorType>(
RetTy)->getElementCount()
1575 case Intrinsic::powi:
1576 if (
auto *RHSC = dyn_cast<ConstantInt>(Args[1])) {
1577 bool ShouldOptForSize =
I->getParent()->getParent()->hasOptSize();
1579 ShouldOptForSize)) {
1583 unsigned ActiveBits =
Exponent.getActiveBits();
1584 unsigned PopCount =
Exponent.popcount();
1586 thisT()->getArithmeticInstrCost(
1588 if (RHSC->isNegative())
1589 Cost += thisT()->getArithmeticInstrCost(Instruction::FDiv,
RetTy,
1595 case Intrinsic::cttz:
1601 case Intrinsic::ctlz:
1607 case Intrinsic::memcpy:
1608 return thisT()->getMemcpyCost(ICA.
getInst());
1610 case Intrinsic::masked_scatter: {
1611 const Value *Mask = Args[3];
1612 bool VarMask = !isa<Constant>(Mask);
1613 Align Alignment = cast<ConstantInt>(Args[2])->getAlignValue();
1614 return thisT()->getGatherScatterOpCost(Instruction::Store,
1618 case Intrinsic::masked_gather: {
1619 const Value *Mask = Args[2];
1620 bool VarMask = !isa<Constant>(Mask);
1621 Align Alignment = cast<ConstantInt>(Args[1])->getAlignValue();
1622 return thisT()->getGatherScatterOpCost(Instruction::Load,
RetTy, Args[0],
1625 case Intrinsic::experimental_vp_strided_store: {
1628 const Value *Mask = Args[3];
1629 const Value *EVL = Args[4];
1630 bool VarMask = !isa<Constant>(Mask) || !isa<Constant>(EVL);
1631 Align Alignment =
I->getParamAlign(1).valueOrOne();
1632 return thisT()->getStridedMemoryOpCost(Instruction::Store,
1633 Data->getType(),
Ptr, VarMask,
1636 case Intrinsic::experimental_vp_strided_load: {
1638 const Value *Mask = Args[2];
1639 const Value *EVL = Args[3];
1640 bool VarMask = !isa<Constant>(Mask) || !isa<Constant>(EVL);
1641 Align Alignment =
I->getParamAlign(0).valueOrOne();
1642 return thisT()->getStridedMemoryOpCost(Instruction::Load,
RetTy,
Ptr,
1645 case Intrinsic::experimental_stepvector: {
1646 if (isa<ScalableVectorType>(
RetTy))
1651 case Intrinsic::vector_extract: {
1654 if (isa<ScalableVectorType>(
RetTy))
1656 unsigned Index = cast<ConstantInt>(Args[1])->getZExtValue();
1657 return thisT()->getShuffleCost(
1661 case Intrinsic::vector_insert: {
1664 if (isa<ScalableVectorType>(Args[1]->
getType()))
1666 unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
1667 return thisT()->getShuffleCost(
1671 case Intrinsic::vector_reverse: {
1672 return thisT()->getShuffleCost(
1676 case Intrinsic::vector_splice: {
1677 unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
1678 return thisT()->getShuffleCost(
1682 case Intrinsic::vector_reduce_add:
1683 case Intrinsic::vector_reduce_mul:
1684 case Intrinsic::vector_reduce_and:
1685 case Intrinsic::vector_reduce_or:
1686 case Intrinsic::vector_reduce_xor:
1687 case Intrinsic::vector_reduce_smax:
1688 case Intrinsic::vector_reduce_smin:
1689 case Intrinsic::vector_reduce_fmax:
1690 case Intrinsic::vector_reduce_fmin:
1691 case Intrinsic::vector_reduce_fmaximum:
1692 case Intrinsic::vector_reduce_fminimum:
1693 case Intrinsic::vector_reduce_umax:
1694 case Intrinsic::vector_reduce_umin: {
1698 case Intrinsic::vector_reduce_fadd:
1699 case Intrinsic::vector_reduce_fmul: {
1701 IID,
RetTy, {Args[0]->getType(), Args[1]->
getType()}, FMF,
I, 1);
1704 case Intrinsic::fshl:
1705 case Intrinsic::fshr: {
1706 const Value *
X = Args[0];
1707 const Value *
Y = Args[1];
1708 const Value *Z = Args[2];
1721 thisT()->getArithmeticInstrCost(BinaryOperator::Or,
RetTy,
CostKind);
1723 thisT()->getArithmeticInstrCost(BinaryOperator::Sub,
RetTy,
CostKind);
1724 Cost += thisT()->getArithmeticInstrCost(
1727 Cost += thisT()->getArithmeticInstrCost(
1732 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::URem,
RetTy,
1736 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1738 thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
1741 thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
1746 case Intrinsic::get_active_lane_mask: {
1752 if (!getTLI()->shouldExpandGetActiveLaneMask(ResVT, ArgType)) {
1762 thisT()->getTypeBasedIntrinsicInstrCost(Attrs,
CostKind);
1763 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, ExpRetTy,
RetTy,
1767 case Intrinsic::experimental_cttz_elts: {
1772 if (!getTLI()->shouldExpandCttzElements(ArgType))
1780 bool ZeroIsPoison = !cast<ConstantInt>(Args[1])->isZero();
1782 if (isa<ScalableVectorType>(ICA.
getArgTypes()[0]) &&
I &&
I->getCaller())
1791 NewEltTy, cast<VectorType>(Args[0]->
getType())->getElementCount());
1796 thisT()->getIntrinsicInstrCost(StepVecAttrs,
CostKind);
1799 thisT()->getArithmeticInstrCost(Instruction::Sub, NewVecTy,
CostKind);
1800 Cost += thisT()->getCastInstrCost(Instruction::SExt, NewVecTy,
1804 thisT()->getArithmeticInstrCost(Instruction::And, NewVecTy,
CostKind);
1807 NewEltTy, NewVecTy, FMF,
I, 1);
1808 Cost += thisT()->getTypeBasedIntrinsicInstrCost(ReducAttrs,
CostKind);
1810 thisT()->getArithmeticInstrCost(Instruction::Sub, NewEltTy,
CostKind);
1822 std::optional<unsigned> FOp =
1825 if (ICA.
getID() == Intrinsic::vp_load) {
1827 if (
auto *VPI = dyn_cast_or_null<VPIntrinsic>(ICA.
getInst()))
1828 Alignment = VPI->getPointerAlignment().valueOrOne();
1832 dyn_cast<PointerType>(ICA.
getArgs()[0]->getType()))
1833 AS = PtrTy->getAddressSpace();
1834 return thisT()->getMemoryOpCost(*FOp, ICA.
getReturnType(), Alignment,
1837 if (ICA.
getID() == Intrinsic::vp_store) {
1839 if (
auto *VPI = dyn_cast_or_null<VPIntrinsic>(ICA.
getInst()))
1840 Alignment = VPI->getPointerAlignment().valueOrOne();
1844 dyn_cast<PointerType>(ICA.
getArgs()[1]->getType()))
1845 AS = PtrTy->getAddressSpace();
1846 return thisT()->getMemoryOpCost(*FOp, Args[0]->
getType(), Alignment,
1850 return thisT()->getArithmeticInstrCost(*FOp, ICA.
getReturnType(),
1855 std::optional<Intrinsic::ID> FID =
1860 "Expected VPIntrinsic to have Mask and Vector Length args and "
1868 *FID != Intrinsic::vector_reduce_fadd &&
1869 *FID != Intrinsic::vector_reduce_fmul)
1874 return thisT()->getIntrinsicInstrCost(NewICA,
CostKind);
1883 ScalarizationCost = 0;
1884 if (!
RetTy->isVoidTy())
1886 cast<VectorType>(
RetTy),
1888 ScalarizationCost +=
1894 return thisT()->getTypeBasedIntrinsicInstrCost(Attrs,
CostKind);
1915 unsigned VecTyIndex = 0;
1916 if (IID == Intrinsic::vector_reduce_fadd ||
1917 IID == Intrinsic::vector_reduce_fmul)
1919 assert(Tys.
size() > VecTyIndex &&
"Unexpected IntrinsicCostAttributes");
1920 VecOpTy = dyn_cast<VectorType>(Tys[VecTyIndex]);
1929 if (isa<ScalableVectorType>(
RetTy) ||
any_of(Tys, [](
const Type *Ty) {
1930 return isa<ScalableVectorType>(Ty);
1936 SkipScalarizationCost ? ScalarizationCostPassed : 0;
1937 unsigned ScalarCalls = 1;
1939 if (
auto *RetVTy = dyn_cast<VectorType>(
RetTy)) {
1940 if (!SkipScalarizationCost)
1943 ScalarCalls = std::max(ScalarCalls,
1944 cast<FixedVectorType>(RetVTy)->getNumElements());
1945 ScalarRetTy =
RetTy->getScalarType();
1948 for (
Type *Ty : Tys) {
1949 if (
auto *VTy = dyn_cast<VectorType>(Ty)) {
1950 if (!SkipScalarizationCost)
1953 ScalarCalls = std::max(ScalarCalls,
1954 cast<FixedVectorType>(VTy)->getNumElements());
1959 if (ScalarCalls == 1)
1964 thisT()->getIntrinsicInstrCost(ScalarAttrs,
CostKind);
1966 return ScalarCalls * ScalarCost + ScalarizationCost;
1970 case Intrinsic::sqrt:
1973 case Intrinsic::sin:
1976 case Intrinsic::cos:
1979 case Intrinsic::tan:
1982 case Intrinsic::exp:
1985 case Intrinsic::exp2:
1988 case Intrinsic::exp10:
1991 case Intrinsic::log:
1994 case Intrinsic::log10:
1997 case Intrinsic::log2:
2000 case Intrinsic::fabs:
2003 case Intrinsic::canonicalize:
2006 case Intrinsic::minnum:
2009 case Intrinsic::maxnum:
2012 case Intrinsic::minimum:
2015 case Intrinsic::maximum:
2018 case Intrinsic::copysign:
2021 case Intrinsic::floor:
2024 case Intrinsic::ceil:
2027 case Intrinsic::trunc:
2030 case Intrinsic::nearbyint:
2033 case Intrinsic::rint:
2036 case Intrinsic::lrint:
2039 case Intrinsic::llrint:
2042 case Intrinsic::round:
2045 case Intrinsic::roundeven:
2048 case Intrinsic::pow:
2051 case Intrinsic::fma:
2054 case Intrinsic::fmuladd:
2057 case Intrinsic::experimental_constrained_fmuladd:
2061 case Intrinsic::lifetime_start:
2062 case Intrinsic::lifetime_end:
2063 case Intrinsic::sideeffect:
2064 case Intrinsic::pseudoprobe:
2065 case Intrinsic::arithmetic_fence:
2067 case Intrinsic::masked_store: {
2069 Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
2070 return thisT()->getMaskedMemoryOpCost(Instruction::Store, Ty, TyAlign, 0,
2073 case Intrinsic::masked_load: {
2075 Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
2076 return thisT()->getMaskedMemoryOpCost(Instruction::Load, Ty, TyAlign, 0,
2079 case Intrinsic::vector_reduce_add:
2080 case Intrinsic::vector_reduce_mul:
2081 case Intrinsic::vector_reduce_and:
2082 case Intrinsic::vector_reduce_or:
2083 case Intrinsic::vector_reduce_xor:
2084 return thisT()->getArithmeticReductionCost(
2087 case Intrinsic::vector_reduce_fadd:
2088 case Intrinsic::vector_reduce_fmul:
2089 return thisT()->getArithmeticReductionCost(
2091 case Intrinsic::vector_reduce_smax:
2092 case Intrinsic::vector_reduce_smin:
2093 case Intrinsic::vector_reduce_umax:
2094 case Intrinsic::vector_reduce_umin:
2095 case Intrinsic::vector_reduce_fmax:
2096 case Intrinsic::vector_reduce_fmin:
2097 case Intrinsic::vector_reduce_fmaximum:
2098 case Intrinsic::vector_reduce_fminimum:
2101 case Intrinsic::abs: {
2103 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2106 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
2108 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
2111 Cost += thisT()->getArithmeticInstrCost(
2115 case Intrinsic::smax:
2116 case Intrinsic::smin:
2117 case Intrinsic::umax:
2118 case Intrinsic::umin: {
2120 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2121 bool IsUnsigned = IID == Intrinsic::umax || IID == Intrinsic::umin;
2125 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
2127 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
2131 case Intrinsic::sadd_sat:
2132 case Intrinsic::ssub_sat: {
2133 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2137 ? Intrinsic::sadd_with_overflow
2138 : Intrinsic::ssub_with_overflow;
2145 nullptr, ScalarizationCostPassed);
2146 Cost += thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2147 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
2149 Cost += 2 * thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy,
2153 case Intrinsic::uadd_sat:
2154 case Intrinsic::usub_sat: {
2155 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2159 ? Intrinsic::uadd_with_overflow
2160 : Intrinsic::usub_with_overflow;
2164 nullptr, ScalarizationCostPassed);
2165 Cost += thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2167 thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
2171 case Intrinsic::smul_fix:
2172 case Intrinsic::umul_fix: {
2173 unsigned ExtSize =
RetTy->getScalarSizeInBits() * 2;
2174 Type *ExtTy =
RetTy->getWithNewBitWidth(ExtSize);
2177 IID == Intrinsic::smul_fix ? Instruction::SExt : Instruction::ZExt;
2183 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2184 Cost += 2 * thisT()->getCastInstrCost(Instruction::Trunc,
RetTy, ExtTy,
2186 Cost += thisT()->getArithmeticInstrCost(Instruction::LShr,
RetTy,
2196 case Intrinsic::sadd_with_overflow:
2197 case Intrinsic::ssub_with_overflow: {
2198 Type *SumTy =
RetTy->getContainedType(0);
2199 Type *OverflowTy =
RetTy->getContainedType(1);
2200 unsigned Opcode = IID == Intrinsic::sadd_with_overflow
2201 ? BinaryOperator::Add
2202 : BinaryOperator::Sub;
2209 Cost += thisT()->getArithmeticInstrCost(Opcode, SumTy,
CostKind);
2210 Cost += 2 * thisT()->getCmpSelInstrCost(
2211 Instruction::ICmp, SumTy, OverflowTy,
2213 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::Xor, OverflowTy,
2217 case Intrinsic::uadd_with_overflow:
2218 case Intrinsic::usub_with_overflow: {
2219 Type *SumTy =
RetTy->getContainedType(0);
2220 Type *OverflowTy =
RetTy->getContainedType(1);
2221 unsigned Opcode = IID == Intrinsic::uadd_with_overflow
2222 ? BinaryOperator::Add
2223 : BinaryOperator::Sub;
2229 Cost += thisT()->getArithmeticInstrCost(Opcode, SumTy,
CostKind);
2231 thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, SumTy, OverflowTy,
2235 case Intrinsic::smul_with_overflow:
2236 case Intrinsic::umul_with_overflow: {
2237 Type *MulTy =
RetTy->getContainedType(0);
2238 Type *OverflowTy =
RetTy->getContainedType(1);
2241 bool IsSigned = IID == Intrinsic::smul_with_overflow;
2243 unsigned ExtOp = IsSigned ? Instruction::SExt : Instruction::ZExt;
2247 Cost += 2 * thisT()->getCastInstrCost(ExtOp, ExtTy, MulTy, CCH,
CostKind);
2249 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2250 Cost += 2 * thisT()->getCastInstrCost(Instruction::Trunc, MulTy, ExtTy,
2252 Cost += thisT()->getArithmeticInstrCost(Instruction::LShr, ExtTy,
2258 Cost += thisT()->getArithmeticInstrCost(Instruction::AShr, MulTy,
2263 Cost += thisT()->getCmpSelInstrCost(
2267 case Intrinsic::fptosi_sat:
2268 case Intrinsic::fptoui_sat: {
2271 Type *FromTy = Tys[0];
2272 bool IsSigned = IID == Intrinsic::fptosi_sat;
2277 Cost += thisT()->getIntrinsicInstrCost(Attrs1,
CostKind);
2280 Cost += thisT()->getIntrinsicInstrCost(Attrs2,
CostKind);
2281 Cost += thisT()->getCastInstrCost(
2282 IsSigned ? Instruction::FPToSI : Instruction::FPToUI,
RetTy, FromTy,
2285 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2286 Cost += thisT()->getCmpSelInstrCost(
2288 Cost += thisT()->getCmpSelInstrCost(
2293 case Intrinsic::ctpop:
2299 case Intrinsic::ctlz:
2302 case Intrinsic::cttz:
2305 case Intrinsic::bswap:
2308 case Intrinsic::bitreverse:
2317 if (IID == Intrinsic::fabs && LT.second.isFloatingPoint() &&
2327 return (LT.first * 2);
2329 return (LT.first * 1);
2333 return (LT.first * 2);
2338 if (IID == Intrinsic::fmuladd)
2339 return thisT()->getArithmeticInstrCost(BinaryOperator::FMul,
RetTy,
2341 thisT()->getArithmeticInstrCost(BinaryOperator::FAdd,
RetTy,
2343 if (IID == Intrinsic::experimental_constrained_fmuladd) {
2345 Intrinsic::experimental_constrained_fmul,
RetTy, Tys);
2347 Intrinsic::experimental_constrained_fadd,
RetTy, Tys);
2348 return thisT()->getIntrinsicInstrCost(FMulAttrs,
CostKind) +
2349 thisT()->getIntrinsicInstrCost(FAddAttrs,
CostKind);
2355 if (
auto *RetVTy = dyn_cast<VectorType>(
RetTy)) {
2357 if (isa<ScalableVectorType>(
RetTy) ||
any_of(Tys, [](
const Type *Ty) {
2358 return isa<ScalableVectorType>(Ty);
2363 SkipScalarizationCost
2364 ? ScalarizationCostPassed
2368 unsigned ScalarCalls = cast<FixedVectorType>(RetVTy)->getNumElements();
2370 for (
Type *Ty : Tys) {
2377 thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2378 for (
Type *Ty : Tys) {
2379 if (
auto *VTy = dyn_cast<VectorType>(Ty)) {
2383 ScalarCalls = std::max(ScalarCalls,
2384 cast<FixedVectorType>(VTy)->getNumElements());
2387 return ScalarCalls * ScalarCost + ScalarizationCost;
2391 return SingleCallCost;
2413 return LT.first.isValid() ? *LT.first.getValue() : 0;
2446 if (isa<ScalableVectorType>(Ty))
2450 unsigned NumVecElts = cast<FixedVectorType>(Ty)->getNumElements();
2451 if ((Opcode == Instruction::Or || Opcode == Instruction::And) &&
2461 return thisT()->getCastInstrCost(Instruction::BitCast, ValTy, Ty,
2463 thisT()->getCmpSelInstrCost(Instruction::ICmp, ValTy,
2467 unsigned NumReduxLevels =
Log2_32(NumVecElts);
2470 std::pair<InstructionCost, MVT> LT = thisT()->getTypeLegalizationCost(Ty);
2471 unsigned LongVectorCount = 0;
2473 LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
2474 while (NumVecElts > MVTLen) {
2480 ArithCost += thisT()->getArithmeticInstrCost(Opcode, SubTy,
CostKind);
2485 NumReduxLevels -= LongVectorCount;
2497 NumReduxLevels * thisT()->getArithmeticInstrCost(Opcode, Ty,
CostKind);
2498 return ShuffleCost + ArithCost +
2499 thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
2523 if (isa<ScalableVectorType>(Ty))
2526 auto *VTy = cast<FixedVectorType>(Ty);
2533 return ExtractCost + ArithCost;
2537 std::optional<FastMathFlags> FMF,
2539 assert(Ty &&
"Unknown reduction vector type");
2552 if (isa<ScalableVectorType>(Ty))
2556 unsigned NumVecElts = cast<FixedVectorType>(Ty)->getNumElements();
2557 unsigned NumReduxLevels =
Log2_32(NumVecElts);
2560 std::pair<InstructionCost, MVT> LT = thisT()->getTypeLegalizationCost(Ty);
2561 unsigned LongVectorCount = 0;
2563 LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
2564 while (NumVecElts > MVTLen) {
2578 NumReduxLevels -= LongVectorCount;
2591 return ShuffleCost + MinMaxCost +
2592 thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
2604 thisT()->getArithmeticReductionCost(Opcode, ExtTy, FMF,
CostKind);
2606 IsUnsigned ? Instruction::ZExt : Instruction::SExt, ExtTy, Ty,
2609 return RedCost + ExtCost;
2620 Instruction::Add, ExtTy, std::nullopt,
CostKind);
2622 IsUnsigned ? Instruction::ZExt : Instruction::SExt, ExtTy, Ty,
2626 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2628 return RedCost + MulCost + 2 * ExtCost;
This file implements a class to represent arbitrary precision integral constant values and operations...
This file implements the BitVector class.
static GCRegistry::Add< ErlangGC > A("erlang", "erlang-compatible garbage collector")
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This file contains the declarations for the subclasses of Constant, which represent the different fla...
static cl::opt< TargetTransformInfo::TargetCostKind > CostKind("cost-kind", cl::desc("Target cost kind"), cl::init(TargetTransformInfo::TCK_RecipThroughput), cl::values(clEnumValN(TargetTransformInfo::TCK_RecipThroughput, "throughput", "Reciprocal throughput"), clEnumValN(TargetTransformInfo::TCK_Latency, "latency", "Instruction latency"), clEnumValN(TargetTransformInfo::TCK_CodeSize, "code-size", "Code size"), clEnumValN(TargetTransformInfo::TCK_SizeAndLatency, "size-latency", "Code size and latency")))
static GCMetadataPrinterRegistry::Add< ErlangGCPrinter > X("erlang", "erlang-compatible garbage collector")
mir Rename Register Operands
static const Function * getCalledFunction(const Value *V, bool &IsNoBuiltin)
ConstantRange Range(APInt(BitWidth, Low), APInt(BitWidth, High))
uint64_t IntrinsicInst * II
static GCMetadataPrinterRegistry::Add< OcamlGCMetadataPrinter > Y("ocaml", "ocaml 3.10-compatible collector")
const char LLVMTargetMachineRef TM
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
This file defines the SmallPtrSet class.
This file defines the SmallVector class.
static SymbolRef::Type getType(const Symbol *Sym)
This file describes how to lower LLVM code to machine code.
Class for arbitrary precision integers.
static APInt getAllOnes(unsigned numBits)
Return an APInt of a specified width with all bits set.
void setBit(unsigned BitPosition)
Set the given bit to 1 whose position is given as "bitPosition".
bool sgt(const APInt &RHS) const
Signed greater than comparison.
unsigned getBitWidth() const
Return the number of bits in the APInt.
bool slt(const APInt &RHS) const
Signed less than comparison.
static APInt getZero(unsigned numBits)
Get the '0' value for the specified bit-width.
an instruction to allocate memory on the stack
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
ArrayRef< T > drop_front(size_t N=1) const
Drop the first N elements of the array.
size_t size() const
size - Get the array size.
ArrayRef< T > drop_back(size_t N=1) const
Drop the last N elements of the array.
A cache of @llvm.assume calls within a function.
LLVM Basic Block Representation.
Base class which can be used to help build a TTI implementation.
bool isTypeLegal(Type *Ty)
InstructionCost getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA, TTI::TargetCostKind CostKind)
Get intrinsic cost based on arguments.
bool isValidAddrSpaceCast(unsigned FromAS, unsigned ToAS) const
virtual unsigned getPrefetchDistance() const
InstructionCost getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy, unsigned Factor, ArrayRef< unsigned > Indices, Align Alignment, unsigned AddressSpace, TTI::TargetCostKind CostKind, bool UseMaskForCond=false, bool UseMaskForGaps=false)
InstructionCost getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy, CmpInst::Predicate VecPred, TTI::TargetCostKind CostKind, const Instruction *I=nullptr)
void getUnrollingPreferences(Loop *L, ScalarEvolution &SE, TTI::UnrollingPreferences &UP, OptimizationRemarkEmitter *ORE)
bool preferToKeepConstantsAttached(const Instruction &Inst, const Function &Fn) const
unsigned getMaxInterleaveFactor(ElementCount VF)
unsigned getNumberOfParts(Type *Tp)
InstructionCost getMaskedMemoryOpCost(unsigned Opcode, Type *DataTy, Align Alignment, unsigned AddressSpace, TTI::TargetCostKind CostKind)
InstructionCost getExtractWithExtendCost(unsigned Opcode, Type *Dst, VectorType *VecTy, unsigned Index)
TypeSize getRegisterBitWidth(TargetTransformInfo::RegisterKind K) const
std::optional< unsigned > getVScaleForTuning() const
InstructionCost getOrderedReductionCost(unsigned Opcode, VectorType *Ty, TTI::TargetCostKind CostKind)
Try to calculate the cost of performing strict (in-order) reductions, which involves doing a sequence...
bool isNumRegsMajorCostOfLSR()
bool isTruncateFree(Type *Ty1, Type *Ty2)
InstructionCost getVectorInstrCost(unsigned Opcode, Type *Val, TTI::TargetCostKind CostKind, unsigned Index, Value *Op0, Value *Op1)
bool isHardwareLoopProfitable(Loop *L, ScalarEvolution &SE, AssumptionCache &AC, TargetLibraryInfo *LibInfo, HardwareLoopInfo &HWLoopInfo)
InstructionCost getTreeReductionCost(unsigned Opcode, VectorType *Ty, TTI::TargetCostKind CostKind)
Try to calculate arithmetic and shuffle op costs for reduction intrinsics.
bool preferPredicateOverEpilogue(TailFoldingInfo *TFI)
virtual bool shouldPrefetchAddressSpace(unsigned AS) const
InstructionCost getStridedMemoryOpCost(unsigned Opcode, Type *DataTy, const Value *Ptr, bool VariableMask, Align Alignment, TTI::TargetCostKind CostKind, const Instruction *I)
bool isLegalICmpImmediate(int64_t imm)
bool isProfitableToHoist(Instruction *I)
virtual unsigned getMaxPrefetchIterationsAhead() const
InstructionCost getVectorInstrCost(const Instruction &I, Type *Val, TTI::TargetCostKind CostKind, unsigned Index)
std::optional< unsigned > getMaxVScale() const
TTI::ShuffleKind improveShuffleKindFromMask(TTI::ShuffleKind Kind, ArrayRef< int > Mask, VectorType *Ty, int &Index, VectorType *&SubTy) const
InstructionCost getExtendedReductionCost(unsigned Opcode, bool IsUnsigned, Type *ResTy, VectorType *Ty, FastMathFlags FMF, TTI::TargetCostKind CostKind)
unsigned getRegUsageForType(Type *Ty)
bool shouldBuildRelLookupTables() const
InstructionCost getMinMaxReductionCost(Intrinsic::ID IID, VectorType *Ty, FastMathFlags FMF, TTI::TargetCostKind CostKind)
Try to calculate op costs for min/max reduction operations.
unsigned getCallerAllocaCost(const CallBase *CB, const AllocaInst *AI) const
InstructionCost getMemoryOpCost(unsigned Opcode, Type *Src, MaybeAlign Alignment, unsigned AddressSpace, TTI::TargetCostKind CostKind, TTI::OperandValueInfo OpInfo={TTI::OK_AnyValue, TTI::OP_None}, const Instruction *I=nullptr)
InstructionCost getGatherScatterOpCost(unsigned Opcode, Type *DataTy, const Value *Ptr, bool VariableMask, Align Alignment, TTI::TargetCostKind CostKind, const Instruction *I=nullptr)
unsigned getEstimatedNumberOfCaseClusters(const SwitchInst &SI, unsigned &JumpTableSize, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI)
bool isIndexedLoadLegal(TTI::MemIndexedMode M, Type *Ty, const DataLayout &DL) const
bool shouldDropLSRSolutionIfLessProfitable() const
InstructionCost getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, ArrayRef< int > Mask, TTI::TargetCostKind CostKind, int Index, VectorType *SubTp, ArrayRef< const Value * > Args=std::nullopt, const Instruction *CxtI=nullptr)
bool isLSRCostLess(TTI::LSRCost C1, TTI::LSRCost C2)
std::optional< Value * > simplifyDemandedUseBitsIntrinsic(InstCombiner &IC, IntrinsicInst &II, APInt DemandedMask, KnownBits &Known, bool &KnownBitsComputed)
bool shouldFoldTerminatingConditionAfterLSR() const
virtual unsigned getMinPrefetchStride(unsigned NumMemAccesses, unsigned NumStridedMemAccesses, unsigned NumPrefetches, bool HasCall) const
bool hasBranchDivergence(const Function *F=nullptr)
bool isIndexedStoreLegal(TTI::MemIndexedMode M, Type *Ty, const DataLayout &DL) const
unsigned getAssumedAddrSpace(const Value *V) const
InstructionCost getOperandsScalarizationOverhead(ArrayRef< const Value * > Args, ArrayRef< Type * > Tys, TTI::TargetCostKind CostKind)
Estimate the overhead of scalarizing an instructions unique non-constant operands.
InstructionCost getAddressComputationCost(Type *Ty, ScalarEvolution *, const SCEV *)
InstructionCost getScalarizationOverhead(VectorType *InTy, const APInt &DemandedElts, bool Insert, bool Extract, TTI::TargetCostKind CostKind)
Estimate the overhead of scalarizing an instruction.
int64_t getPreferredLargeGEPBaseOffset(int64_t MinOffset, int64_t MaxOffset)
InstructionCost getGEPCost(Type *PointeeType, const Value *Ptr, ArrayRef< const Value * > Operands, Type *AccessType, TTI::TargetCostKind CostKind)
bool isFCmpOrdCheaperThanFCmpZero(Type *Ty)
virtual std::optional< unsigned > getCacheSize(TargetTransformInfo::CacheLevel Level) const
InstructionCost getScalingFactorCost(Type *Ty, GlobalValue *BaseGV, StackOffset BaseOffset, bool HasBaseReg, int64_t Scale, unsigned AddrSpace)
bool isAlwaysUniform(const Value *V)
TailFoldingStyle getPreferredTailFoldingStyle(bool IVUpdateMayOverflow=true)
bool allowsMisalignedMemoryAccesses(LLVMContext &Context, unsigned BitWidth, unsigned AddressSpace, Align Alignment, unsigned *Fast) const
unsigned getStoreMinimumVF(unsigned VF, Type *ScalarMemTy, Type *ScalarValTy) const
InstructionCost getScalarizationOverhead(VectorType *InTy, bool Insert, bool Extract, TTI::TargetCostKind CostKind)
Helper wrapper for the DemandedElts variant of getScalarizationOverhead.
virtual std::optional< unsigned > getCacheAssociativity(TargetTransformInfo::CacheLevel Level) const
virtual bool enableWritePrefetching() const
Value * rewriteIntrinsicWithAddressSpace(IntrinsicInst *II, Value *OldV, Value *NewV) const
void getPeelingPreferences(Loop *L, ScalarEvolution &SE, TTI::PeelingPreferences &PP)
InstructionCost getMulAccReductionCost(bool IsUnsigned, Type *ResTy, VectorType *Ty, TTI::TargetCostKind CostKind)
InstructionCost getCFInstrCost(unsigned Opcode, TTI::TargetCostKind CostKind, const Instruction *I=nullptr)
bool collectFlatAddressOperands(SmallVectorImpl< int > &OpIndexes, Intrinsic::ID IID) const
InstructionCost getCallInstrCost(Function *F, Type *RetTy, ArrayRef< Type * > Tys, TTI::TargetCostKind CostKind)
Compute a cost of the given call instruction.
InstructionCost getArithmeticInstrCost(unsigned Opcode, Type *Ty, TTI::TargetCostKind CostKind, TTI::OperandValueInfo Opd1Info={TTI::OK_AnyValue, TTI::OP_None}, TTI::OperandValueInfo Opd2Info={TTI::OK_AnyValue, TTI::OP_None}, ArrayRef< const Value * > Args=std::nullopt, const Instruction *CxtI=nullptr)
InstructionCost getArithmeticReductionCost(unsigned Opcode, VectorType *Ty, std::optional< FastMathFlags > FMF, TTI::TargetCostKind CostKind)
InstructionCost getFPOpCost(Type *Ty)
InstructionCost getVectorSplitCost()
std::pair< InstructionCost, MVT > getTypeLegalizationCost(Type *Ty) const
Estimate the cost of type-legalization and the legalized type.
bool haveFastSqrt(Type *Ty)
std::pair< const Value *, unsigned > getPredicatedAddrSpace(const Value *V) const
unsigned getInliningThresholdMultiplier() const
InstructionCost getReplicationShuffleCost(Type *EltTy, int ReplicationFactor, int VF, const APInt &DemandedDstElts, TTI::TargetCostKind CostKind)
virtual ~BasicTTIImplBase()=default
bool isLegalAddScalableImmediate(int64_t Imm)
InstructionCost getScalarizationOverhead(VectorType *RetTy, ArrayRef< const Value * > Args, ArrayRef< Type * > Tys, TTI::TargetCostKind CostKind)
Estimate the overhead of scalarizing the inputs and outputs of an instruction, with return type RetTy...
bool isVScaleKnownToBeAPowerOfTwo() const
std::optional< Instruction * > instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II)
bool addrspacesMayAlias(unsigned AS0, unsigned AS1) const
bool isLegalAddImmediate(int64_t imm)
bool shouldBuildLookupTables()
unsigned getFlatAddressSpace()
InstructionCost getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src, TTI::CastContextHint CCH, TTI::TargetCostKind CostKind, const Instruction *I=nullptr)
virtual unsigned getCacheLineSize() const
bool isNoopAddrSpaceCast(unsigned FromAS, unsigned ToAS) const
bool isSourceOfDivergence(const Value *V)
int getInlinerVectorBonusPercent() const
InstructionCost getTypeBasedIntrinsicInstrCost(const IntrinsicCostAttributes &ICA, TTI::TargetCostKind CostKind)
Get intrinsic cost based on argument types.
std::optional< Value * > simplifyDemandedVectorEltsIntrinsic(InstCombiner &IC, IntrinsicInst &II, APInt DemandedElts, APInt &UndefElts, APInt &UndefElts2, APInt &UndefElts3, std::function< void(Instruction *, unsigned, APInt, APInt &)> SimplifyAndSetOp)
bool isLegalAddressingMode(Type *Ty, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale, unsigned AddrSpace, Instruction *I=nullptr, int64_t ScalableOffset=0)
bool isSingleThreaded() const
BasicTTIImplBase(const TargetMachine *TM, const DataLayout &DL)
unsigned adjustInliningThreshold(const CallBase *CB)
bool isProfitableLSRChainElement(Instruction *I)
Concrete BasicTTIImpl that can be used if no further customization is needed.
size_type count() const
count - Returns the number of bits which are set.
BlockFrequencyInfo pass uses BlockFrequencyInfoImpl implementation to estimate IR basic block frequen...
Base class for all callable instructions (InvokeInst and CallInst) Holds everything related to callin...
static Type * makeCmpResultType(Type *opnd_type)
Create a result type for fcmp/icmp.
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
@ ICMP_UGT
unsigned greater than
@ ICMP_SGT
signed greater than
@ ICMP_ULT
unsigned less than
@ FCMP_UNO
1 0 0 0 True if unordered: isnan(X) | isnan(Y)
This class represents a range of values.
A parsed version of the target data layout string in and methods for querying it.
TypeSize getTypeStoreSizeInBits(Type *Ty) const
Returns the maximum number of bits that may be overwritten by storing the specified type; always a mu...
unsigned getIndexSizeInBits(unsigned AS) const
Size in bits of index used for address calculation in getelementptr.
constexpr bool isVector() const
One or more elements.
static constexpr ElementCount getFixed(ScalarTy MinVal)
constexpr bool isScalar() const
Exactly one element.
Convenience struct for specifying and reasoning about fast-math flags.
Class to represent fixed width SIMD vectors.
unsigned getNumElements() const
static FixedVectorType * get(Type *ElementType, unsigned NumElts)
bool isTargetIntrinsic() const
isTargetIntrinsic - Returns true if this function is an intrinsic and the intrinsic is specific to a ...
AttributeList getAttributes() const
Return the attribute list for this Function.
The core instruction combiner logic.
static InstructionCost getInvalid(CostType Val=0)
std::optional< CostType > getValue() const
This function is intended to be used as sparingly as possible, since the class provides the full rang...
unsigned getOpcode() const
Returns a member of one of the enums like Instruction::Add.
static IntegerType * get(LLVMContext &C, unsigned NumBits)
This static method is the primary way of constructing an IntegerType.
FastMathFlags getFlags() const
const SmallVectorImpl< Type * > & getArgTypes() const
Type * getReturnType() const
bool skipScalarizationCost() const
const SmallVectorImpl< const Value * > & getArgs() const
InstructionCost getScalarizationCost() const
const IntrinsicInst * getInst() const
Intrinsic::ID getID() const
bool isTypeBasedOnly() const
A wrapper class for inspecting calls to intrinsic functions.
This is an important class for using LLVM in a threaded context.
Represents a single loop in the control flow graph.
virtual bool shouldPrefetchAddressSpace(unsigned AS) const
virtual unsigned getMinPrefetchStride(unsigned NumMemAccesses, unsigned NumStridedMemAccesses, unsigned NumPrefetches, bool HasCall) const
Return the minimum stride necessary to trigger software prefetching.
virtual bool enableWritePrefetching() const
virtual unsigned getMaxPrefetchIterationsAhead() const
Return the maximum prefetch distance in terms of loop iterations.
virtual unsigned getPrefetchDistance() const
Return the preferred prefetch distance in terms of instructions.
virtual std::optional< unsigned > getCacheAssociativity(unsigned Level) const
Return the cache associatvity for the given level of cache.
virtual std::optional< unsigned > getCacheLineSize(unsigned Level) const
Return the target cache line size in bytes at a given level.
TypeSize getStoreSize() const
Return the number of bytes overwritten by a store of the specified value type.
static PointerType * get(Type *ElementType, unsigned AddressSpace)
This constructs a pointer to an object of the specified type in a numbered address space.
Analysis providing profile information.
This class represents an analyzed expression in the program.
The main scalar evolution driver.
static bool isZeroEltSplatMask(ArrayRef< int > Mask, int NumSrcElts)
Return true if this shuffle mask chooses all elements with the same value as the first element of exa...
static bool isSpliceMask(ArrayRef< int > Mask, int NumSrcElts, int &Index)
Return true if this shuffle mask is a splice mask, concatenating the two inputs together and then ext...
static bool isSelectMask(ArrayRef< int > Mask, int NumSrcElts)
Return true if this shuffle mask chooses elements from its source vectors without lane crossings.
static bool isExtractSubvectorMask(ArrayRef< int > Mask, int NumSrcElts, int &Index)
Return true if this shuffle mask is an extract subvector mask.
static bool isReverseMask(ArrayRef< int > Mask, int NumSrcElts)
Return true if this shuffle mask swaps the order of elements from exactly one source vector.
static bool isTransposeMask(ArrayRef< int > Mask, int NumSrcElts)
Return true if this shuffle mask is a transpose mask.
static bool isInsertSubvectorMask(ArrayRef< int > Mask, int NumSrcElts, int &NumSubElts, int &Index)
Return true if this shuffle mask is an insert subvector mask.
std::pair< iterator, bool > insert(PtrType Ptr)
Inserts Ptr if and only if there is no element in the container equal to Ptr.
SmallPtrSet - This class implements a set which is optimized for holding SmallSize or less elements.
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
void push_back(const T &Elt)
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
StackOffset holds a fixed and a scalable offset in bytes.
static StackOffset getScalable(int64_t Scalable)
static StackOffset getFixed(int64_t Fixed)
static StructType * create(LLVMContext &Context, StringRef Name)
This creates an identified struct.
Provides information about what library functions are available for the current target.
This base class for TargetLowering contains the SelectionDAG-independent parts that can be used from ...
bool isOperationExpand(unsigned Op, EVT VT) const
Return true if the specified operation is illegal on this target or unlikely to be made legal with cu...
int InstructionOpcodeToISD(unsigned Opcode) const
Get the ISD node that corresponds to the Instruction class opcode.
bool isIndexedStoreLegal(unsigned IdxMode, EVT VT) const
Return true if the specified indexed load is legal on this target.
EVT getValueType(const DataLayout &DL, Type *Ty, bool AllowUnknown=false) const
Return the EVT corresponding to this LLVM type.
LegalizeAction
This enum indicates whether operations are valid for a target, and if not, what action should be used...
virtual bool isLegalICmpImmediate(int64_t) const
Return true if the specified immediate is legal icmp immediate, that is the target has icmp instructi...
const TargetMachine & getTargetMachine() const
virtual bool isZExtFree(Type *FromTy, Type *ToTy) const
Return true if any actual instruction that defines a value of type FromTy implicitly zero-extends the...
@ TypeScalarizeScalableVector
virtual bool isSuitableForJumpTable(const SwitchInst *SI, uint64_t NumCases, uint64_t Range, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) const
Return true if lowering to a jump table is suitable for a set of case clusters which may contain NumC...
virtual bool areJTsAllowed(const Function *Fn) const
Return true if lowering to a jump table is allowed.
bool isOperationLegalOrPromote(unsigned Op, EVT VT, bool LegalOnly=false) const
Return true if the specified operation is legal on this target or can be made legal using promotion.
virtual unsigned getNumRegisters(LLVMContext &Context, EVT VT, std::optional< MVT > RegisterVT=std::nullopt) const
Return the number of registers that this ValueType will eventually require.
virtual bool isCheapToSpeculateCttz(Type *Ty) const
Return true if it is cheap to speculate a call to intrinsic cttz.
bool isTruncStoreLegal(EVT ValVT, EVT MemVT) const
Return true if the specified store with truncation is legal on this target.
unsigned getBitWidthForCttzElements(Type *RetTy, ElementCount EC, bool ZeroIsPoison, const ConstantRange *VScaleRange) const
Return the minimum number of bits required to hold the maximum possible number of trailing zero vecto...
virtual bool allowsMisalignedMemoryAccesses(EVT, unsigned AddrSpace=0, Align Alignment=Align(1), MachineMemOperand::Flags Flags=MachineMemOperand::MONone, unsigned *=nullptr) const
Determine if the target supports unaligned memory accesses.
virtual bool isTruncateFree(Type *FromTy, Type *ToTy) const
Return true if it's free to truncate a value of type FromTy to type ToTy.
virtual EVT getTypeToTransformTo(LLVMContext &Context, EVT VT) const
For types supported by the target, this is an identity function.
bool isTypeLegal(EVT VT) const
Return true if the target has native support for the specified value type.
bool isSuitableForBitTests(unsigned NumDests, unsigned NumCmps, const APInt &Low, const APInt &High, const DataLayout &DL) const
Return true if lowering to a bit test is suitable for a set of case clusters which contains NumDests ...
virtual bool isLegalAddImmediate(int64_t) const
Return true if the specified immediate is legal add immediate, that is the target has add instruction...
virtual bool isFreeAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const
Returns true if a cast from SrcAS to DestAS is "cheap", such that e.g.
LegalizeAction getTruncStoreAction(EVT ValVT, EVT MemVT) const
Return how this store with truncation should be treated: either it is legal, needs to be promoted to ...
LegalizeAction getLoadExtAction(unsigned ExtType, EVT ValVT, EVT MemVT) const
Return how this load with extension should be treated: either it is legal, needs to be promoted to a ...
virtual bool isIntDivCheap(EVT VT, AttributeList Attr) const
Return true if integer divide is usually cheaper than a sequence of several shifts,...
bool isOperationLegalOrCustom(unsigned Op, EVT VT, bool LegalOnly=false) const
Return true if the specified operation is legal on this target or can be made legal with custom lower...
virtual bool isProfitableToHoist(Instruction *I) const
bool isIndexedLoadLegal(unsigned IdxMode, EVT VT) const
Return true if the specified indexed load is legal on this target.
bool isLoadExtLegal(unsigned ExtType, EVT ValVT, EVT MemVT) const
Return true if the specified load with extension is legal on this target.
virtual bool isCheapToSpeculateCtlz(Type *Ty) const
Return true if it is cheap to speculate a call to intrinsic ctlz.
virtual int64_t getPreferredLargeGEPBaseOffset(int64_t MinOffset, int64_t MaxOffset) const
Return the prefered common base offset.
LegalizeKind getTypeConversion(LLVMContext &Context, EVT VT) const
Return pair that represents the legalization kind (first) that needs to happen to EVT (second) in ord...
LegalizeTypeAction getTypeAction(LLVMContext &Context, EVT VT) const
Return how we should legalize values of this type, either it is already legal (return 'Legal') or we ...
virtual bool isLegalAddScalableImmediate(int64_t) const
Return true if adding the specified scalable immediate is legal, that is the target has add instructi...
bool isBeneficialToExpandPowI(int64_t Exponent, bool OptForSize) const
Return true if it is beneficial to expand an @llvm.powi.
virtual bool isFAbsFree(EVT VT) const
Return true if an fabs operation is free to the point where it is never worthwhile to replace it with...
virtual bool isLegalAddressingMode(const DataLayout &DL, const AddrMode &AM, Type *Ty, unsigned AddrSpace, Instruction *I=nullptr) const
Return true if the addressing mode represented by AM is legal for this target, for a load/store of th...
bool isOperationLegalOrCustomOrPromote(unsigned Op, EVT VT, bool LegalOnly=false) const
Return true if the specified operation is legal on this target or can be made legal with custom lower...
std::pair< LegalizeTypeAction, EVT > LegalizeKind
LegalizeKind holds the legalization kind that needs to happen to EVT in order to type-legalize it.
Primary interface to the complete machine description for the target machine.
virtual std::pair< const Value *, unsigned > getPredicatedAddrSpace(const Value *V) const
If the specified predicate checks whether a generic pointer falls within a specified address space,...
virtual bool isNoopAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const
Returns true if a cast between SrcAS and DestAS is a noop.
virtual unsigned getAssumedAddrSpace(const Value *V) const
If the specified generic pointer could be assumed as a pointer to a specific address space,...
ThreadModel::Model ThreadModel
ThreadModel - This flag specifies the type of threading model to assume for things like atomics.
TargetSubtargetInfo - Generic base class for all target subtargets.
virtual bool useAA() const
Enable use of alias analysis during code generation (during MI scheduling, DAGCombine,...
Triple - Helper class for working with autoconf configuration names.
ArchType getArch() const
Get the parsed architecture type of this triple.
bool isArch64Bit() const
Test whether the architecture is 64-bit.
bool isOSDarwin() const
Is this a "Darwin" OS (macOS, iOS, tvOS, watchOS, XROS, or DriverKit).
static constexpr TypeSize getFixed(ScalarTy ExactSize)
The instances of the Type class are immutable: once they are created, they are never changed.
bool isVectorTy() const
True if this is an instance of VectorType.
bool isIntOrIntVectorTy() const
Return true if this is an integer type or a vector of integer types.
static IntegerType * getInt1Ty(LLVMContext &C)
static IntegerType * getIntNTy(LLVMContext &C, unsigned N)
unsigned getScalarSizeInBits() const LLVM_READONLY
If this is a vector type, return the getPrimitiveSizeInBits value for the element type.
Type * getWithNewBitWidth(unsigned NewBitWidth) const
Given an integer or vector type, change the lane bitwidth to NewBitwidth, whilst keeping the old numb...
LLVMContext & getContext() const
Return the LLVMContext in which this type was uniqued.
static IntegerType * getInt8Ty(LLVMContext &C)
bool isPtrOrPtrVectorTy() const
Return true if this is a pointer type or a vector of pointer types.
bool isFPOrFPVectorTy() const
Return true if this is a FP type or a vector of FP.
Type * getScalarType() const
If this is a vector type, return the element type, otherwise return 'this'.
Value * getOperand(unsigned i) const
static bool isVPBinOp(Intrinsic::ID ID)
static std::optional< unsigned > getFunctionalOpcodeForVP(Intrinsic::ID ID)
static std::optional< Intrinsic::ID > getFunctionalIntrinsicIDForVP(Intrinsic::ID ID)
static bool isVPIntrinsic(Intrinsic::ID)
static bool isVPReduction(Intrinsic::ID ID)
LLVM Value Representation.
Type * getType() const
All values are typed, get the type of this value.
Base class of all SIMD vector types.
static VectorType * getHalfElementsVectorType(VectorType *VTy)
This static method returns a VectorType with half as many elements as the input type and the same ele...
ElementCount getElementCount() const
Return an ElementCount instance to represent the (possibly scalable) number of elements in the vector...
static VectorType * get(Type *ElementType, ElementCount EC)
This static method is the primary way to construct an VectorType.
Type * getElementType() const
static constexpr bool isKnownLT(const FixedOrScalableQuantity &LHS, const FixedOrScalableQuantity &RHS)
constexpr bool isScalable() const
Returns whether the quantity is scaled by a runtime quantity (vscale).
constexpr ScalarTy getKnownMinValue() const
Returns the minimum value this quantity can represent.
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
constexpr char Args[]
Key for Kernel::Metadata::mArgs.
APInt ScaleBitMask(const APInt &A, unsigned NewBitWidth, bool MatchAllBits=false)
Splat/Merge neighboring bits to widen/narrow the bitmask represented by.
@ Fast
Attempts to make calls as fast as possible (e.g.
@ C
The default llvm calling convention, compatible with C.
@ BSWAP
Byte Swap and Counting operators.
@ FMA
FMA - Perform a * b + c with no intermediate rounding step.
@ FADD
Simple binary floating point operators.
@ SDIVREM
SDIVREM/UDIVREM - Divide two integers and produce both a quotient and remainder result.
@ BRIND
BRIND - Indirect branch.
@ BR_JT
BR_JT - Jumptable branch.
@ FCANONICALIZE
Returns platform specific canonical encoding of a floating point number.
@ SELECT
Select(COND, TRUEVAL, FALSEVAL).
@ FMINNUM
FMINNUM/FMAXNUM - Perform floating-point minimum or maximum on two values.
@ VSELECT
Select with a vector condition (op #0) and two vector operands (ops #1 and #2), returning a vector re...
@ FMINIMUM
FMINIMUM/FMAXIMUM - NaN-propagating minimum/maximum that also treat -0.0 as less than 0....
@ FCOPYSIGN
FCOPYSIGN(X, Y) - Return the value of X with the sign of Y.
MemIndexedMode
MemIndexedMode enum - This enum defines the load / store indexed addressing modes.
DiagnosticInfoOptimizationBase::Argument NV
This is an optimization pass for GlobalISel generic memory operations.
Intrinsic::ID getMinMaxReductionIntrinsicOp(Intrinsic::ID RdxID)
Returns the min/max intrinsic used when expanding a min/max reduction.
uint64_t divideCeil(uint64_t Numerator, uint64_t Denominator)
Returns the integer ceil(Numerator / Denominator).
unsigned getArithmeticReductionInstruction(Intrinsic::ID RdxID)
Returns the arithmetic instruction opcode used when expanding a reduction.
bool any_of(R &&range, UnaryPredicate P)
Provide wrappers to std::any_of which take ranges instead of having to pass begin/end explicitly.
unsigned Log2_32(uint32_t Value)
Return the floor log base 2 of the specified value, -1 if the value is zero.
constexpr bool isPowerOf2_32(uint32_t Value)
Return true if the argument is a power of two > 0.
ConstantRange getVScaleRange(const Function *F, unsigned BitWidth)
Determine the possible constant range of vscale with the given bit width, based on the vscale_range f...
constexpr unsigned BitWidth
cl::opt< unsigned > PartialUnrollingThreshold
This struct is a compact representation of a valid (non-zero power of two) alignment.
bool isSimple() const
Test if the given EVT is simple (as opposed to being extended).
ElementCount getVectorElementCount() const
static EVT getEVT(Type *Ty, bool HandleUnknown=false)
Return the value type corresponding to the specified type.
MVT getSimpleVT() const
Return the SimpleValueType held in the specified simple EVT.
static EVT getIntegerVT(LLVMContext &Context, unsigned BitWidth)
Returns the EVT that represents an integer with the given number of bits.
Attributes of a target dependent hardware loop.
This struct is a compact representation of a valid (power of two) or undefined (0) alignment.
This represents an addressing mode of: BaseGV + BaseOffs + BaseReg + Scale*ScaleReg + ScalableOffset*...