16#ifndef LLVM_CODEGEN_BASICTTIIMPL_H
17#define LLVM_CODEGEN_BASICTTIIMPL_H
85 T *thisT() {
return static_cast<T *
>(
this); }
94 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
98 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
117 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
119 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
132 "Can only extract subvectors from vectors");
134 assert((!isa<FixedVectorType>(VTy) ||
135 (
Index + NumSubElts) <=
136 (
int)cast<FixedVectorType>(VTy)->getNumElements()) &&
137 "SK_ExtractSubvector index out of range");
143 for (
int i = 0; i != NumSubElts; ++i) {
145 thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
147 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, SubVTy,
160 "Can only insert subvectors into vectors");
162 assert((!isa<FixedVectorType>(VTy) ||
163 (
Index + NumSubElts) <=
164 (
int)cast<FixedVectorType>(VTy)->getNumElements()) &&
165 "SK_InsertSubvector index out of range");
171 for (
int i = 0; i != NumSubElts; ++i) {
172 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, SubVTy,
175 thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
CostKind,
176 i +
Index,
nullptr,
nullptr);
183 return static_cast<const T *
>(
this)->getST();
188 return static_cast<const T *
>(
this)->getTLI();
210 bool IsGatherScatter,
213 if (isa<ScalableVectorType>(DataTy))
216 auto *VT = cast<FixedVectorType>(DataTy);
226 VT->getNumElements()),
230 VT->getNumElements() *
237 Opcode == Instruction::Store,
CostKind);
248 VT->getNumElements() *
250 Instruction::ExtractElement,
252 VT->getNumElements()),
258 return LoadCost + PackingCost + ConditionalCost;
273 unsigned *
Fast)
const {
310 std::pair<const Value *, unsigned>
329 bool HasBaseReg, int64_t Scale,
340 Type *ScalarValTy)
const {
341 auto &&IsSupportedByTarget = [
this, ScalarMemTy, ScalarValTy](
unsigned VF) {
344 if (getTLI()->isOperationLegal(
ISD::STORE, VT) ||
354 while (VF > 2 && IsSupportedByTarget(VF))
384 int64_t BaseOffset,
bool HasBaseReg,
385 int64_t Scale,
unsigned AddrSpace) {
423 unsigned &JumpTableSize,
433 unsigned N =
SI.getNumCases();
444 APInt MaxCaseVal =
SI.case_begin()->getCaseValue()->getValue();
445 APInt MinCaseVal = MaxCaseVal;
446 for (
auto CI :
SI.cases()) {
447 const APInt &CaseVal = CI.getCaseValue()->getValue();
448 if (CaseVal.
sgt(MaxCaseVal))
449 MaxCaseVal = CaseVal;
450 if (CaseVal.
slt(MinCaseVal))
451 MinCaseVal = CaseVal;
457 for (
auto I :
SI.cases())
458 Dests.
insert(
I.getCaseSuccessor());
467 if (
N < 2 || N < TLI->getMinimumJumpTableEntries())
470 (MaxCaseVal - MinCaseVal)
471 .getLimitedValue(std::numeric_limits<uint64_t>::max() - 1) + 1;
474 JumpTableSize = Range;
490 if (!
TM.isPositionIndependent())
500 Triple TargetTriple =
TM.getTargetTriple();
568 else if (ST->getSchedModel().LoopMicroOpBufferSize > 0)
569 MaxOps = ST->getSchedModel().LoopMicroOpBufferSize;
576 if (isa<CallInst>(
I) || isa<InvokeInst>(
I)) {
586 <<
"advising against unrolling the loop because it "
643 std::optional<Value *>
646 bool &KnownBitsComputed) {
657 IC, II, DemandedElts, UndefElts, UndefElts2, UndefElts3,
661 virtual std::optional<unsigned>
663 return std::optional<unsigned>(
667 virtual std::optional<unsigned>
669 std::optional<unsigned> TargetResult =
687 unsigned NumStridedMemAccesses,
688 unsigned NumPrefetches,
689 bool HasCall)
const {
691 NumPrefetches, HasCall);
722 const APInt &DemandedElts,
723 bool Insert,
bool Extract,
727 if (isa<ScalableVectorType>(InTy))
729 auto *Ty = cast<FixedVectorType>(InTy);
732 "Vector size mismatch");
736 for (
int i = 0, e = Ty->getNumElements(); i < e; ++i) {
737 if (!DemandedElts[i])
740 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, Ty,
743 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
754 if (isa<ScalableVectorType>(InTy))
756 auto *Ty = cast<FixedVectorType>(InTy);
759 return thisT()->getScalarizationOverhead(Ty, DemandedElts, Insert, Extract,
770 assert(Args.size() == Tys.
size() &&
"Expected matching Args and Tys");
774 for (
int I = 0,
E = Args.size();
I !=
E;
I++) {
782 if (!isa<Constant>(
A) && UniqueOperands.
insert(
A).second) {
783 if (
auto *VecTy = dyn_cast<VectorType>(Ty))
840 if (MTy == LK.second)
854 ArrayRef<const Value *> Args = ArrayRef<const Value *>(),
855 const Instruction *CxtI =
nullptr) {
857 const TargetLoweringBase *TLI = getTLI();
858 int ISD = TLI->InstructionOpcodeToISD(Opcode);
859 assert(ISD &&
"Invalid opcode");
872 InstructionCost OpCost = (IsFloat ? 2 : 1);
874 if (TLI->isOperationLegalOrPromote(ISD, LT.second)) {
877 return LT.first * OpCost;
880 if (!TLI->isOperationExpand(ISD,
LT.second)) {
883 return LT.first * 2 * OpCost;
895 unsigned DivOpc = IsSigned ? Instruction::SDiv : Instruction::UDiv;
896 InstructionCost DivCost = thisT()->getArithmeticInstrCost(
897 DivOpc, Ty,
CostKind, Opd1Info, Opd2Info);
898 InstructionCost MulCost =
899 thisT()->getArithmeticInstrCost(Instruction::Mul, Ty,
CostKind);
900 InstructionCost SubCost =
901 thisT()->getArithmeticInstrCost(Instruction::Sub, Ty,
CostKind);
902 return DivCost + MulCost + SubCost;
907 if (isa<ScalableVectorType>(Ty))
913 if (
auto *VTy = dyn_cast<FixedVectorType>(Ty)) {
914 InstructionCost
Cost = thisT()->getArithmeticInstrCost(
919 SmallVector<Type *> Tys(
Args.size(), Ty);
930 int Limit = Mask.size() * 2;
934 any_of(Mask, [Limit](
int I) {
return I >= Limit; }))
972 if (
auto *FVT = dyn_cast<FixedVectorType>(Tp))
973 return getBroadcastShuffleOverhead(FVT,
CostKind);
981 if (
auto *FVT = dyn_cast<FixedVectorType>(Tp))
982 return getPermuteShuffleOverhead(FVT,
CostKind);
986 cast<FixedVectorType>(SubTp));
989 cast<FixedVectorType>(SubTp));
1003 assert(ISD &&
"Invalid opcode");
1007 TypeSize SrcSize = SrcLT.second.getSizeInBits();
1008 TypeSize DstSize = DstLT.second.getSizeInBits();
1009 bool IntOrPtrSrc = Src->isIntegerTy() || Src->isPointerTy();
1010 bool IntOrPtrDst = Dst->isIntegerTy() || Dst->isPointerTy();
1015 case Instruction::Trunc:
1020 case Instruction::BitCast:
1023 if (SrcLT.first == DstLT.first && IntOrPtrSrc == IntOrPtrDst &&
1027 case Instruction::FPExt:
1028 if (
I && getTLI()->isExtFree(
I))
1031 case Instruction::ZExt:
1032 if (TLI->
isZExtFree(SrcLT.second, DstLT.second))
1035 case Instruction::SExt:
1036 if (
I && getTLI()->isExtFree(
I))
1046 if (DstLT.first == SrcLT.first &&
1051 case Instruction::AddrSpaceCast:
1053 Dst->getPointerAddressSpace()))
1058 auto *SrcVTy = dyn_cast<VectorType>(Src);
1059 auto *DstVTy = dyn_cast<VectorType>(Dst);
1062 if (SrcLT.first == DstLT.first &&
1067 if (!SrcVTy && !DstVTy) {
1078 if (DstVTy && SrcVTy) {
1080 if (SrcLT.first == DstLT.first && SrcSize == DstSize) {
1083 if (Opcode == Instruction::ZExt)
1087 if (Opcode == Instruction::SExt)
1088 return SrcLT.first * 2;
1094 return SrcLT.first * 1;
1107 if ((SplitSrc || SplitDst) && SrcVTy->getElementCount().isVector() &&
1108 DstVTy->getElementCount().isVector()) {
1111 T *
TTI =
static_cast<T *
>(
this);
1114 (!SplitSrc || !SplitDst) ?
TTI->getVectorSplitCost() : 0;
1121 if (isa<ScalableVectorType>(DstVTy))
1126 unsigned Num = cast<FixedVectorType>(DstVTy)->getNumElements();
1128 Opcode, Dst->getScalarType(), Src->getScalarType(), CCH,
CostKind,
I);
1141 if (Opcode == Instruction::BitCast) {
1157 return thisT()->getVectorInstrCost(Instruction::ExtractElement, VecTy,
1174 assert(ISD &&
"Invalid opcode");
1183 assert(CondTy &&
"CondTy must exist");
1189 if (!(ValTy->
isVectorTy() && !LT.second.isVector()) &&
1193 return LT.first * 1;
1199 if (
auto *ValVTy = dyn_cast<VectorType>(ValTy)) {
1200 if (isa<ScalableVectorType>(ValTy))
1203 unsigned Num = cast<FixedVectorType>(ValVTy)->getNumElements();
1207 Opcode, ValVTy->getScalarType(), CondTy, VecPred,
CostKind,
I);
1229 Value *Op0 =
nullptr;
1230 Value *Op1 =
nullptr;
1231 if (
auto *IE = dyn_cast<InsertElementInst>(&
I)) {
1232 Op0 = IE->getOperand(0);
1233 Op1 = IE->getOperand(1);
1235 return thisT()->getVectorInstrCost(
I.getOpcode(), Val,
CostKind,
Index, Op0,
1241 const APInt &DemandedDstElts,
1244 "Unexpected size of DemandedDstElts.");
1262 Cost += thisT()->getScalarizationOverhead(SrcVT, DemandedSrcElts,
1265 Cost += thisT()->getScalarizationOverhead(ReplicatedVT, DemandedDstElts,
1277 assert(!Src->isVoidTy() &&
"Invalid type");
1279 if (getTLI()->getValueType(
DL, Src,
true) ==
MVT::Other)
1294 LT.second.getSizeInBits())) {
1300 if (Opcode == Instruction::Store)
1309 cast<VectorType>(Src), Opcode != Instruction::Store,
1310 Opcode == Instruction::Store,
CostKind);
1320 return getCommonMaskedMemoryOpCost(Opcode, DataTy, Alignment,
true,
false,
1325 const Value *
Ptr,
bool VariableMask,
1329 return getCommonMaskedMemoryOpCost(Opcode, DataTy, Alignment, VariableMask,
1336 bool UseMaskForCond =
false,
bool UseMaskForGaps =
false) {
1339 if (isa<ScalableVectorType>(VecTy))
1342 auto *VT = cast<FixedVectorType>(VecTy);
1344 unsigned NumElts = VT->getNumElements();
1345 assert(Factor > 1 && NumElts % Factor == 0 &&
"Invalid interleave factor");
1347 unsigned NumSubElts = NumElts / Factor;
1352 if (UseMaskForCond || UseMaskForGaps)
1353 Cost = thisT()->getMaskedMemoryOpCost(Opcode, VecTy, Alignment,
1362 unsigned VecTySize = thisT()->getDataLayout().getTypeStoreSize(VecTy);
1379 if (
Cost.isValid() && VecTySize > VecTyLTSize) {
1382 unsigned NumLegalInsts =
divideCeil(VecTySize, VecTyLTSize);
1386 unsigned NumEltsPerLegalInst =
divideCeil(NumElts, NumLegalInsts);
1389 BitVector UsedInsts(NumLegalInsts,
false);
1390 for (
unsigned Index : Indices)
1391 for (
unsigned Elt = 0; Elt < NumSubElts; ++Elt)
1392 UsedInsts.
set((
Index + Elt * Factor) / NumEltsPerLegalInst);
1401 "Interleaved memory op has too many members");
1407 for (
unsigned Index : Indices) {
1408 assert(
Index < Factor &&
"Invalid index for interleaved memory op");
1409 for (
unsigned Elm = 0; Elm < NumSubElts; Elm++)
1410 DemandedLoadStoreElts.
setBit(
Index + Elm * Factor);
1413 if (Opcode == Instruction::Load) {
1423 SubVT, DemandedAllSubElts,
1425 Cost += Indices.
size() * InsSubCost;
1426 Cost += thisT()->getScalarizationOverhead(VT, DemandedLoadStoreElts,
1444 SubVT, DemandedAllSubElts,
1446 Cost += ExtSubCost * Indices.
size();
1447 Cost += thisT()->getScalarizationOverhead(VT, DemandedLoadStoreElts,
1452 if (!UseMaskForCond)
1457 Cost += thisT()->getReplicationShuffleCost(
1458 I8Type, Factor, NumSubElts,
1459 UseMaskForGaps ? DemandedLoadStoreElts : DemandedAllResultElts,
1467 if (UseMaskForGaps) {
1469 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::And, MaskVT,
1494 (
RetTy->isVectorTy() ? cast<VectorType>(
RetTy)->getElementCount()
1503 case Intrinsic::powi:
1504 if (
auto *RHSC = dyn_cast<ConstantInt>(Args[1])) {
1505 bool ShouldOptForSize =
I->getParent()->getParent()->hasOptSize();
1507 ShouldOptForSize)) {
1511 unsigned ActiveBits =
Exponent.getActiveBits();
1512 unsigned PopCount =
Exponent.popcount();
1514 thisT()->getArithmeticInstrCost(
1516 if (RHSC->getSExtValue() < 0)
1517 Cost += thisT()->getArithmeticInstrCost(Instruction::FDiv,
RetTy,
1523 case Intrinsic::cttz:
1529 case Intrinsic::ctlz:
1535 case Intrinsic::memcpy:
1536 return thisT()->getMemcpyCost(ICA.
getInst());
1538 case Intrinsic::masked_scatter: {
1539 const Value *Mask = Args[3];
1540 bool VarMask = !isa<Constant>(Mask);
1541 Align Alignment = cast<ConstantInt>(Args[2])->getAlignValue();
1542 return thisT()->getGatherScatterOpCost(Instruction::Store,
1546 case Intrinsic::masked_gather: {
1547 const Value *Mask = Args[2];
1548 bool VarMask = !isa<Constant>(Mask);
1549 Align Alignment = cast<ConstantInt>(Args[1])->getAlignValue();
1550 return thisT()->getGatherScatterOpCost(Instruction::Load,
RetTy, Args[0],
1553 case Intrinsic::experimental_stepvector: {
1554 if (isa<ScalableVectorType>(
RetTy))
1559 case Intrinsic::vector_extract: {
1562 if (isa<ScalableVectorType>(
RetTy))
1564 unsigned Index = cast<ConstantInt>(Args[1])->getZExtValue();
1565 return thisT()->getShuffleCost(
1569 case Intrinsic::vector_insert: {
1572 if (isa<ScalableVectorType>(Args[1]->
getType()))
1574 unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
1575 return thisT()->getShuffleCost(
1579 case Intrinsic::experimental_vector_reverse: {
1580 return thisT()->getShuffleCost(
1584 case Intrinsic::experimental_vector_splice: {
1585 unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
1586 return thisT()->getShuffleCost(
1590 case Intrinsic::vector_reduce_add:
1591 case Intrinsic::vector_reduce_mul:
1592 case Intrinsic::vector_reduce_and:
1593 case Intrinsic::vector_reduce_or:
1594 case Intrinsic::vector_reduce_xor:
1595 case Intrinsic::vector_reduce_smax:
1596 case Intrinsic::vector_reduce_smin:
1597 case Intrinsic::vector_reduce_fmax:
1598 case Intrinsic::vector_reduce_fmin:
1599 case Intrinsic::vector_reduce_umax:
1600 case Intrinsic::vector_reduce_umin: {
1604 case Intrinsic::vector_reduce_fadd:
1605 case Intrinsic::vector_reduce_fmul: {
1607 IID,
RetTy, {Args[0]->getType(), Args[1]->
getType()}, FMF,
I, 1);
1610 case Intrinsic::fshl:
1611 case Intrinsic::fshr: {
1612 const Value *
X = Args[0];
1613 const Value *
Y = Args[1];
1614 const Value *Z = Args[2];
1627 thisT()->getArithmeticInstrCost(BinaryOperator::Or,
RetTy,
CostKind);
1629 thisT()->getArithmeticInstrCost(BinaryOperator::Sub,
RetTy,
CostKind);
1630 Cost += thisT()->getArithmeticInstrCost(
1633 Cost += thisT()->getArithmeticInstrCost(
1638 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::URem,
RetTy,
1642 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1644 thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
1647 thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
1652 case Intrinsic::get_active_lane_mask: {
1658 if (!getTLI()->shouldExpandGetActiveLaneMask(ResVT, ArgType)) {
1668 thisT()->getTypeBasedIntrinsicInstrCost(Attrs,
CostKind);
1669 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, ExpRetTy,
RetTy,
1680 ScalarizationCost = 0;
1681 if (!
RetTy->isVoidTy())
1683 cast<VectorType>(
RetTy),
1685 ScalarizationCost +=
1691 return thisT()->getTypeBasedIntrinsicInstrCost(Attrs,
CostKind);
1712 unsigned VecTyIndex = 0;
1713 if (IID == Intrinsic::vector_reduce_fadd ||
1714 IID == Intrinsic::vector_reduce_fmul)
1716 assert(Tys.
size() > VecTyIndex &&
"Unexpected IntrinsicCostAttributes");
1717 VecOpTy = dyn_cast<VectorType>(Tys[VecTyIndex]);
1726 if (isa<ScalableVectorType>(
RetTy) ||
any_of(Tys, [](
const Type *Ty) {
1727 return isa<ScalableVectorType>(Ty);
1733 SkipScalarizationCost ? ScalarizationCostPassed : 0;
1734 unsigned ScalarCalls = 1;
1736 if (
auto *RetVTy = dyn_cast<VectorType>(
RetTy)) {
1737 if (!SkipScalarizationCost)
1740 ScalarCalls = std::max(ScalarCalls,
1741 cast<FixedVectorType>(RetVTy)->getNumElements());
1742 ScalarRetTy =
RetTy->getScalarType();
1745 for (
unsigned i = 0, ie = Tys.
size(); i != ie; ++i) {
1747 if (
auto *VTy = dyn_cast<VectorType>(Ty)) {
1748 if (!SkipScalarizationCost)
1751 ScalarCalls = std::max(ScalarCalls,
1752 cast<FixedVectorType>(VTy)->getNumElements());
1757 if (ScalarCalls == 1)
1762 thisT()->getIntrinsicInstrCost(ScalarAttrs,
CostKind);
1764 return ScalarCalls * ScalarCost + ScalarizationCost;
1768 case Intrinsic::sqrt:
1771 case Intrinsic::sin:
1774 case Intrinsic::cos:
1777 case Intrinsic::exp:
1780 case Intrinsic::exp2:
1783 case Intrinsic::log:
1786 case Intrinsic::log10:
1789 case Intrinsic::log2:
1792 case Intrinsic::fabs:
1795 case Intrinsic::canonicalize:
1798 case Intrinsic::minnum:
1801 case Intrinsic::maxnum:
1804 case Intrinsic::minimum:
1807 case Intrinsic::maximum:
1810 case Intrinsic::copysign:
1813 case Intrinsic::floor:
1816 case Intrinsic::ceil:
1819 case Intrinsic::trunc:
1822 case Intrinsic::nearbyint:
1825 case Intrinsic::rint:
1828 case Intrinsic::round:
1831 case Intrinsic::roundeven:
1834 case Intrinsic::pow:
1837 case Intrinsic::fma:
1840 case Intrinsic::fmuladd:
1843 case Intrinsic::experimental_constrained_fmuladd:
1847 case Intrinsic::lifetime_start:
1848 case Intrinsic::lifetime_end:
1849 case Intrinsic::sideeffect:
1850 case Intrinsic::pseudoprobe:
1851 case Intrinsic::arithmetic_fence:
1853 case Intrinsic::masked_store: {
1855 Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
1856 return thisT()->getMaskedMemoryOpCost(Instruction::Store, Ty, TyAlign, 0,
1859 case Intrinsic::masked_load: {
1861 Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
1862 return thisT()->getMaskedMemoryOpCost(Instruction::Load, Ty, TyAlign, 0,
1865 case Intrinsic::vector_reduce_add:
1866 return thisT()->getArithmeticReductionCost(Instruction::Add, VecOpTy,
1868 case Intrinsic::vector_reduce_mul:
1869 return thisT()->getArithmeticReductionCost(Instruction::Mul, VecOpTy,
1871 case Intrinsic::vector_reduce_and:
1872 return thisT()->getArithmeticReductionCost(Instruction::And, VecOpTy,
1874 case Intrinsic::vector_reduce_or:
1875 return thisT()->getArithmeticReductionCost(Instruction::Or, VecOpTy,
1877 case Intrinsic::vector_reduce_xor:
1878 return thisT()->getArithmeticReductionCost(Instruction::Xor, VecOpTy,
1880 case Intrinsic::vector_reduce_fadd:
1881 return thisT()->getArithmeticReductionCost(Instruction::FAdd, VecOpTy,
1883 case Intrinsic::vector_reduce_fmul:
1884 return thisT()->getArithmeticReductionCost(Instruction::FMul, VecOpTy,
1886 case Intrinsic::vector_reduce_smax:
1887 case Intrinsic::vector_reduce_smin:
1888 case Intrinsic::vector_reduce_fmax:
1889 case Intrinsic::vector_reduce_fmin:
1890 return thisT()->getMinMaxReductionCost(
1893 case Intrinsic::vector_reduce_umax:
1894 case Intrinsic::vector_reduce_umin:
1895 return thisT()->getMinMaxReductionCost(
1898 case Intrinsic::abs: {
1900 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1903 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
1905 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
1908 Cost += thisT()->getArithmeticInstrCost(
1912 case Intrinsic::smax:
1913 case Intrinsic::smin:
1914 case Intrinsic::umax:
1915 case Intrinsic::umin: {
1917 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1918 bool IsUnsigned = IID == Intrinsic::umax || IID == Intrinsic::umin;
1922 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
1924 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
1928 case Intrinsic::sadd_sat:
1929 case Intrinsic::ssub_sat: {
1930 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1934 ? Intrinsic::sadd_with_overflow
1935 : Intrinsic::ssub_with_overflow;
1942 nullptr, ScalarizationCostPassed);
1943 Cost += thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
1944 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
1946 Cost += 2 * thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy,
1950 case Intrinsic::uadd_sat:
1951 case Intrinsic::usub_sat: {
1952 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1956 ? Intrinsic::uadd_with_overflow
1957 : Intrinsic::usub_with_overflow;
1961 nullptr, ScalarizationCostPassed);
1962 Cost += thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
1964 thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
1968 case Intrinsic::smul_fix:
1969 case Intrinsic::umul_fix: {
1970 unsigned ExtSize =
RetTy->getScalarSizeInBits() * 2;
1971 Type *ExtTy =
RetTy->getWithNewBitWidth(ExtSize);
1974 IID == Intrinsic::smul_fix ? Instruction::SExt : Instruction::ZExt;
1980 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
1981 Cost += 2 * thisT()->getCastInstrCost(Instruction::Trunc,
RetTy, ExtTy,
1983 Cost += thisT()->getArithmeticInstrCost(Instruction::LShr,
RetTy,
1993 case Intrinsic::sadd_with_overflow:
1994 case Intrinsic::ssub_with_overflow: {
1995 Type *SumTy =
RetTy->getContainedType(0);
1996 Type *OverflowTy =
RetTy->getContainedType(1);
1997 unsigned Opcode = IID == Intrinsic::sadd_with_overflow
1998 ? BinaryOperator::Add
1999 : BinaryOperator::Sub;
2006 Cost += thisT()->getArithmeticInstrCost(Opcode, SumTy,
CostKind);
2007 Cost += 2 * thisT()->getCmpSelInstrCost(
2008 Instruction::ICmp, SumTy, OverflowTy,
2010 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::Xor, OverflowTy,
2014 case Intrinsic::uadd_with_overflow:
2015 case Intrinsic::usub_with_overflow: {
2016 Type *SumTy =
RetTy->getContainedType(0);
2017 Type *OverflowTy =
RetTy->getContainedType(1);
2018 unsigned Opcode = IID == Intrinsic::uadd_with_overflow
2019 ? BinaryOperator::Add
2020 : BinaryOperator::Sub;
2026 Cost += thisT()->getArithmeticInstrCost(Opcode, SumTy,
CostKind);
2028 thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, SumTy, OverflowTy,
2032 case Intrinsic::smul_with_overflow:
2033 case Intrinsic::umul_with_overflow: {
2034 Type *MulTy =
RetTy->getContainedType(0);
2035 Type *OverflowTy =
RetTy->getContainedType(1);
2038 bool IsSigned = IID == Intrinsic::smul_with_overflow;
2040 unsigned ExtOp = IsSigned ? Instruction::SExt : Instruction::ZExt;
2044 Cost += 2 * thisT()->getCastInstrCost(ExtOp, ExtTy, MulTy, CCH,
CostKind);
2046 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2047 Cost += 2 * thisT()->getCastInstrCost(Instruction::Trunc, MulTy, ExtTy,
2049 Cost += thisT()->getArithmeticInstrCost(Instruction::LShr, ExtTy,
2055 Cost += thisT()->getArithmeticInstrCost(Instruction::AShr, MulTy,
2060 Cost += thisT()->getCmpSelInstrCost(
2064 case Intrinsic::fptosi_sat:
2065 case Intrinsic::fptoui_sat: {
2068 Type *FromTy = Tys[0];
2069 bool IsSigned = IID == Intrinsic::fptosi_sat;
2074 Cost += thisT()->getIntrinsicInstrCost(Attrs1,
CostKind);
2077 Cost += thisT()->getIntrinsicInstrCost(Attrs2,
CostKind);
2078 Cost += thisT()->getCastInstrCost(
2079 IsSigned ? Instruction::FPToSI : Instruction::FPToUI,
RetTy, FromTy,
2082 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2083 Cost += thisT()->getCmpSelInstrCost(
2085 Cost += thisT()->getCmpSelInstrCost(
2090 case Intrinsic::ctpop:
2096 case Intrinsic::ctlz:
2099 case Intrinsic::cttz:
2102 case Intrinsic::bswap:
2105 case Intrinsic::bitreverse:
2114 if (IID == Intrinsic::fabs && LT.second.isFloatingPoint() &&
2124 return (LT.first * 2);
2126 return (LT.first * 1);
2130 return (LT.first * 2);
2135 if (IID == Intrinsic::fmuladd)
2136 return thisT()->getArithmeticInstrCost(BinaryOperator::FMul,
RetTy,
2138 thisT()->getArithmeticInstrCost(BinaryOperator::FAdd,
RetTy,
2140 if (IID == Intrinsic::experimental_constrained_fmuladd) {
2142 Intrinsic::experimental_constrained_fmul,
RetTy, Tys);
2144 Intrinsic::experimental_constrained_fadd,
RetTy, Tys);
2145 return thisT()->getIntrinsicInstrCost(FMulAttrs,
CostKind) +
2146 thisT()->getIntrinsicInstrCost(FAddAttrs,
CostKind);
2152 if (
auto *RetVTy = dyn_cast<VectorType>(
RetTy)) {
2154 if (isa<ScalableVectorType>(
RetTy) ||
any_of(Tys, [](
const Type *Ty) {
2155 return isa<ScalableVectorType>(Ty);
2160 SkipScalarizationCost
2161 ? ScalarizationCostPassed
2165 unsigned ScalarCalls = cast<FixedVectorType>(RetVTy)->getNumElements();
2167 for (
unsigned i = 0, ie = Tys.
size(); i != ie; ++i) {
2175 thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2176 for (
unsigned i = 0, ie = Tys.
size(); i != ie; ++i) {
2177 if (
auto *VTy = dyn_cast<VectorType>(Tys[i])) {
2181 ScalarCalls = std::max(ScalarCalls,
2182 cast<FixedVectorType>(VTy)->getNumElements());
2185 return ScalarCalls * ScalarCost + ScalarizationCost;
2189 return SingleCallCost;
2211 return LT.first.isValid() ? *LT.first.getValue() : 0;
2244 if (isa<ScalableVectorType>(Ty))
2248 unsigned NumVecElts = cast<FixedVectorType>(Ty)->getNumElements();
2249 if ((Opcode == Instruction::Or || Opcode == Instruction::And) &&
2259 return thisT()->getCastInstrCost(Instruction::BitCast, ValTy, Ty,
2261 thisT()->getCmpSelInstrCost(Instruction::ICmp, ValTy,
2265 unsigned NumReduxLevels =
Log2_32(NumVecElts);
2268 std::pair<InstructionCost, MVT> LT = thisT()->getTypeLegalizationCost(Ty);
2269 unsigned LongVectorCount = 0;
2271 LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
2272 while (NumVecElts > MVTLen) {
2278 ArithCost += thisT()->getArithmeticInstrCost(Opcode, SubTy,
CostKind);
2283 NumReduxLevels -= LongVectorCount;
2295 NumReduxLevels * thisT()->getArithmeticInstrCost(Opcode, Ty,
CostKind);
2296 return ShuffleCost + ArithCost +
2297 thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
2321 if (isa<ScalableVectorType>(Ty))
2324 auto *VTy = cast<FixedVectorType>(Ty);
2331 return ExtractCost + ArithCost;
2335 std::optional<FastMathFlags> FMF,
2337 assert(Ty &&
"Unknown reduction vector type");
2350 if (isa<ScalableVectorType>(Ty))
2355 unsigned NumVecElts = cast<FixedVectorType>(Ty)->getNumElements();
2356 unsigned NumReduxLevels =
Log2_32(NumVecElts);
2359 CmpOpcode = Instruction::FCmp;
2362 "expecting floating point or integer type for min/max reduction");
2363 CmpOpcode = Instruction::ICmp;
2367 std::pair<InstructionCost, MVT> LT = thisT()->getTypeLegalizationCost(Ty);
2368 unsigned LongVectorCount = 0;
2370 LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
2371 while (NumVecElts > MVTLen) {
2380 thisT()->getCmpSelInstrCost(CmpOpcode, SubTy, CondTy,
2382 thisT()->getCmpSelInstrCost(Instruction::Select, SubTy, CondTy,
2388 NumReduxLevels -= LongVectorCount;
2399 (thisT()->getCmpSelInstrCost(CmpOpcode, Ty, CondTy,
2401 thisT()->getCmpSelInstrCost(Instruction::Select, Ty, CondTy,
2405 return ShuffleCost + MinMaxCost +
2406 thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
2412 std::optional<FastMathFlags> FMF,
2418 thisT()->getArithmeticReductionCost(Opcode, ExtTy, FMF,
CostKind);
2420 IsUnsigned ? Instruction::ZExt : Instruction::SExt, ExtTy, Ty,
2423 return RedCost + ExtCost;
2434 Instruction::Add, ExtTy, std::nullopt,
CostKind);
2436 IsUnsigned ? Instruction::ZExt : Instruction::SExt, ExtTy, Ty,
2440 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2442 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)
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.
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
size_t size() const
size - Get the array size.
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.
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)
InstructionCost getScalingFactorCost(Type *Ty, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale, unsigned AddrSpace)
void getUnrollingPreferences(Loop *L, ScalarEvolution &SE, TTI::UnrollingPreferences &UP, OptimizationRemarkEmitter *ORE)
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 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=ArrayRef< const Value * >(), const Instruction *CxtI=nullptr)
InstructionCost getTreeReductionCost(unsigned Opcode, VectorType *Ty, TTI::TargetCostKind CostKind)
Try to calculate arithmetic and shuffle op costs for reduction intrinsics.
virtual bool shouldPrefetchAddressSpace(unsigned AS) const
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
int getInlinerVectorBonusPercent()
unsigned getRegUsageForType(Type *Ty)
bool shouldBuildRelLookupTables() 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 getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, ArrayRef< int > Mask, TTI::TargetCostKind CostKind, int Index, VectorType *SubTp, ArrayRef< const Value * > Args=std::nullopt)
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 isLSRCostLess(TTI::LSRCost C1, TTI::LSRCost C2)
std::optional< Value * > simplifyDemandedUseBitsIntrinsic(InstCombiner &IC, IntrinsicInst &II, APInt DemandedMask, KnownBits &Known, bool &KnownBitsComputed)
virtual unsigned getMinPrefetchStride(unsigned NumMemAccesses, unsigned NumStridedMemAccesses, unsigned NumPrefetches, bool HasCall) const
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.
bool isFCmpOrdCheaperThanFCmpZero(Type *Ty)
unsigned getInliningThresholdMultiplier()
virtual std::optional< unsigned > getCacheSize(TargetTransformInfo::CacheLevel Level) const
bool isAlwaysUniform(const Value *V)
bool isLegalAddressingMode(Type *Ty, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale, unsigned AddrSpace, Instruction *I=nullptr)
TailFoldingStyle getPreferredTailFoldingStyle(bool IVUpdateMayOverflow=true)
bool useGPUDivergenceAnalysis()
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)
bool hasBranchDivergence()
InstructionCost getMinMaxReductionCost(VectorType *Ty, VectorType *CondTy, bool IsUnsigned, TTI::TargetCostKind CostKind)
Try to calculate op costs for min/max reduction operations.
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 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
bool preferPredicateOverEpilogue(Loop *L, LoopInfo *LI, ScalarEvolution &SE, AssumptionCache &AC, TargetLibraryInfo *TLI, DominatorTree *DT, LoopVectorizationLegality *LVL, InterleavedAccessInfo *IAI)
TTI::ShuffleKind improveShuffleKindFromMask(TTI::ShuffleKind Kind, ArrayRef< int > Mask) const
InstructionCost getReplicationShuffleCost(Type *EltTy, int ReplicationFactor, int VF, const APInt &DemandedDstElts, TTI::TargetCostKind CostKind)
virtual ~BasicTTIImplBase()=default
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...
InstructionCost getGEPCost(Type *PointeeType, const Value *Ptr, ArrayRef< const Value * > Operands, TTI::TargetCostKind CostKind)
std::optional< Instruction * > instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II)
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)
InstructionCost getExtendedReductionCost(unsigned Opcode, bool IsUnsigned, Type *ResTy, VectorType *Ty, std::optional< FastMathFlags > FMF, TTI::TargetCostKind CostKind)
virtual unsigned getCacheLineSize() const
bool isNoopAddrSpaceCast(unsigned FromAS, unsigned ToAS) const
bool isSourceOfDivergence(const Value *V)
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 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)
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.
Concrete subclass of DominatorTreeBase that is used to compute a normal dominator tree.
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 ...
The core instruction combiner logic.
static InstructionCost getInvalid(CostType Val=0)
static IntegerType * get(LLVMContext &C, unsigned NumBits)
This static method is the primary way of constructing an IntegerType.
Drive the analysis of interleaved memory accesses in the loop.
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.
LoopVectorizationLegality checks if it is legal to vectorize a loop, and to what vectorization factor...
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 isSelectMask(ArrayRef< int > Mask)
Return true if this shuffle mask chooses elements from its source vectors without lane crossings.
static bool isReverseMask(ArrayRef< int > Mask)
Return true if this shuffle mask swaps the order of elements from exactly one source vector.
static bool isZeroEltSplatMask(ArrayRef< int > Mask)
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 &Index)
Return true if this shuffle mask is a splice mask, concatenating the two inputs together and then ext...
static bool isTransposeMask(ArrayRef< int > Mask)
Return true if this shuffle mask is a transpose 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.
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...
bool isBeneficialToExpandPowI(int Exponent, bool OptForSize) const
Return true if it is beneficial to expand an @llvm.powi.
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.
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 ...
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.
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 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, 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)
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'.
LLVM Value Representation.
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...
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).
#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.
uint64_t divideCeil(uint64_t Numerator, uint64_t Denominator)
Returns the integer ceil(Numerator / Denominator).
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.
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).
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 If BaseGV is null...