16#ifndef LLVM_CODEGEN_BASICTTIIMPL_H
17#define LLVM_CODEGEN_BASICTTIIMPL_H
86 T *thisT() {
return static_cast<T *
>(
this); }
95 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
99 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
118 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
120 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
133 "Can only extract subvectors from vectors");
135 assert((!isa<FixedVectorType>(VTy) ||
136 (
Index + NumSubElts) <=
137 (
int)cast<FixedVectorType>(VTy)->getNumElements()) &&
138 "SK_ExtractSubvector index out of range");
144 for (
int i = 0; i != NumSubElts; ++i) {
146 thisT()->getVectorInstrCost(Instruction::ExtractElement, VTy,
148 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, SubVTy,
161 "Can only insert subvectors into vectors");
163 assert((!isa<FixedVectorType>(VTy) ||
164 (
Index + NumSubElts) <=
165 (
int)cast<FixedVectorType>(VTy)->getNumElements()) &&
166 "SK_InsertSubvector index out of range");
172 for (
int i = 0; i != NumSubElts; ++i) {
173 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, SubVTy,
176 thisT()->getVectorInstrCost(Instruction::InsertElement, VTy,
CostKind,
177 i +
Index,
nullptr,
nullptr);
184 return static_cast<const T *
>(
this)->getST();
189 return static_cast<const T *
>(
this)->getTLI();
211 bool IsGatherScatter,
214 if (isa<ScalableVectorType>(DataTy))
217 auto *VT = cast<FixedVectorType>(DataTy);
227 VT->getNumElements()),
231 VT->getNumElements() *
238 Opcode == Instruction::Store,
CostKind);
249 VT->getNumElements() *
251 Instruction::ExtractElement,
253 VT->getNumElements()),
259 return LoadCost + PackingCost + ConditionalCost;
274 unsigned *
Fast)
const {
317 std::pair<const Value *, unsigned>
336 bool HasBaseReg, int64_t Scale,
351 Type *ScalarValTy)
const {
352 auto &&IsSupportedByTarget = [
this, ScalarMemTy, ScalarValTy](
unsigned VF) {
355 if (getTLI()->isOperationLegal(
ISD::STORE, VT) ||
365 while (VF > 2 && IsSupportedByTarget(VF))
400 int64_t BaseOffset,
bool HasBaseReg,
401 int64_t Scale,
unsigned AddrSpace) {
439 unsigned &JumpTableSize,
449 unsigned N = SI.getNumCases();
454 bool IsJTAllowed = TLI->
areJTsAllowed(SI.getParent()->getParent());
460 APInt MaxCaseVal = SI.case_begin()->getCaseValue()->getValue();
461 APInt MinCaseVal = MaxCaseVal;
462 for (
auto CI : SI.cases()) {
463 const APInt &CaseVal = CI.getCaseValue()->getValue();
464 if (CaseVal.
sgt(MaxCaseVal))
465 MaxCaseVal = CaseVal;
466 if (CaseVal.
slt(MinCaseVal))
467 MinCaseVal = CaseVal;
473 for (
auto I : SI.cases())
474 Dests.
insert(
I.getCaseSuccessor());
483 if (
N < 2 || N < TLI->getMinimumJumpTableEntries())
486 (MaxCaseVal - MinCaseVal)
487 .getLimitedValue(std::numeric_limits<uint64_t>::max() - 1) + 1;
490 JumpTableSize = Range;
506 if (!
TM.isPositionIndependent())
516 Triple TargetTriple =
TM.getTargetTriple();
554 case Instruction::SDiv:
555 case Instruction::SRem:
556 case Instruction::UDiv:
557 case Instruction::URem: {
606 else if (ST->getSchedModel().LoopMicroOpBufferSize > 0)
607 MaxOps = ST->getSchedModel().LoopMicroOpBufferSize;
614 if (isa<CallInst>(
I) || isa<InvokeInst>(
I)) {
624 <<
"advising against unrolling the loop because it "
677 std::optional<Value *>
680 bool &KnownBitsComputed) {
691 IC, II, DemandedElts, UndefElts, UndefElts2, UndefElts3,
695 virtual std::optional<unsigned>
697 return std::optional<unsigned>(
701 virtual std::optional<unsigned>
703 std::optional<unsigned> TargetResult =
721 unsigned NumStridedMemAccesses,
722 unsigned NumPrefetches,
723 bool HasCall)
const {
725 NumPrefetches, HasCall);
757 const APInt &DemandedElts,
758 bool Insert,
bool Extract,
762 if (isa<ScalableVectorType>(InTy))
764 auto *Ty = cast<FixedVectorType>(InTy);
767 "Vector size mismatch");
771 for (
int i = 0, e = Ty->getNumElements(); i < e; ++i) {
772 if (!DemandedElts[i])
775 Cost += thisT()->getVectorInstrCost(Instruction::InsertElement, Ty,
778 Cost += thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
789 if (isa<ScalableVectorType>(InTy))
791 auto *Ty = cast<FixedVectorType>(InTy);
794 return thisT()->getScalarizationOverhead(Ty, DemandedElts, Insert, Extract,
805 assert(Args.size() == Tys.
size() &&
"Expected matching Args and Tys");
809 for (
int I = 0,
E = Args.size();
I !=
E;
I++) {
817 if (!isa<Constant>(
A) && UniqueOperands.
insert(
A).second) {
818 if (
auto *VecTy = dyn_cast<VectorType>(Ty))
875 if (MTy == LK.second)
889 ArrayRef<const Value *> Args = ArrayRef<const Value *>(),
890 const Instruction *CxtI =
nullptr) {
892 const TargetLoweringBase *TLI = getTLI();
893 int ISD = TLI->InstructionOpcodeToISD(Opcode);
894 assert(ISD &&
"Invalid opcode");
907 InstructionCost OpCost = (IsFloat ? 2 : 1);
909 if (TLI->isOperationLegalOrPromote(ISD, LT.second)) {
912 return LT.first * OpCost;
915 if (!TLI->isOperationExpand(ISD,
LT.second)) {
918 return LT.first * 2 * OpCost;
930 unsigned DivOpc = IsSigned ? Instruction::SDiv : Instruction::UDiv;
931 InstructionCost DivCost = thisT()->getArithmeticInstrCost(
932 DivOpc, Ty,
CostKind, Opd1Info, Opd2Info);
933 InstructionCost MulCost =
934 thisT()->getArithmeticInstrCost(Instruction::Mul, Ty,
CostKind);
935 InstructionCost SubCost =
936 thisT()->getArithmeticInstrCost(Instruction::Sub, Ty,
CostKind);
937 return DivCost + MulCost + SubCost;
942 if (isa<ScalableVectorType>(Ty))
948 if (
auto *VTy = dyn_cast<FixedVectorType>(Ty)) {
949 InstructionCost
Cost = thisT()->getArithmeticInstrCost(
954 SmallVector<Type *> Tys(
Args.size(), Ty);
977 (
Index + Mask.size()) <= (
size_t)NumSrcElts) {
985 Mask, NumSrcElts, NumSubElts,
Index)) {
986 if (
Index + NumSubElts > NumSrcElts)
1018 if (
auto *FVT = dyn_cast<FixedVectorType>(Tp))
1019 return getBroadcastShuffleOverhead(FVT,
CostKind);
1027 if (
auto *FVT = dyn_cast<FixedVectorType>(Tp))
1028 return getPermuteShuffleOverhead(FVT,
CostKind);
1032 cast<FixedVectorType>(SubTp));
1035 cast<FixedVectorType>(SubTp));
1049 assert(ISD &&
"Invalid opcode");
1053 TypeSize SrcSize = SrcLT.second.getSizeInBits();
1054 TypeSize DstSize = DstLT.second.getSizeInBits();
1055 bool IntOrPtrSrc = Src->isIntegerTy() || Src->isPointerTy();
1056 bool IntOrPtrDst = Dst->isIntegerTy() || Dst->isPointerTy();
1061 case Instruction::Trunc:
1066 case Instruction::BitCast:
1069 if (SrcLT.first == DstLT.first && IntOrPtrSrc == IntOrPtrDst &&
1073 case Instruction::FPExt:
1074 if (
I && getTLI()->isExtFree(
I))
1077 case Instruction::ZExt:
1078 if (TLI->
isZExtFree(SrcLT.second, DstLT.second))
1081 case Instruction::SExt:
1082 if (
I && getTLI()->isExtFree(
I))
1092 if (DstLT.first == SrcLT.first &&
1097 case Instruction::AddrSpaceCast:
1099 Dst->getPointerAddressSpace()))
1104 auto *SrcVTy = dyn_cast<VectorType>(Src);
1105 auto *DstVTy = dyn_cast<VectorType>(Dst);
1108 if (SrcLT.first == DstLT.first &&
1113 if (!SrcVTy && !DstVTy) {
1124 if (DstVTy && SrcVTy) {
1126 if (SrcLT.first == DstLT.first && SrcSize == DstSize) {
1129 if (Opcode == Instruction::ZExt)
1133 if (Opcode == Instruction::SExt)
1134 return SrcLT.first * 2;
1140 return SrcLT.first * 1;
1153 if ((SplitSrc || SplitDst) && SrcVTy->getElementCount().isVector() &&
1154 DstVTy->getElementCount().isVector()) {
1157 T *
TTI =
static_cast<T *
>(
this);
1160 (!SplitSrc || !SplitDst) ?
TTI->getVectorSplitCost() : 0;
1167 if (isa<ScalableVectorType>(DstVTy))
1172 unsigned Num = cast<FixedVectorType>(DstVTy)->getNumElements();
1174 Opcode, Dst->getScalarType(), Src->getScalarType(), CCH,
CostKind,
I);
1187 if (Opcode == Instruction::BitCast) {
1203 return thisT()->getVectorInstrCost(Instruction::ExtractElement, VecTy,
1220 assert(ISD &&
"Invalid opcode");
1229 assert(CondTy &&
"CondTy must exist");
1235 if (!(ValTy->
isVectorTy() && !LT.second.isVector()) &&
1239 return LT.first * 1;
1245 if (
auto *ValVTy = dyn_cast<VectorType>(ValTy)) {
1246 if (isa<ScalableVectorType>(ValTy))
1249 unsigned Num = cast<FixedVectorType>(ValVTy)->getNumElements();
1253 Opcode, ValVTy->getScalarType(), CondTy, VecPred,
CostKind,
I);
1275 Value *Op0 =
nullptr;
1276 Value *Op1 =
nullptr;
1277 if (
auto *IE = dyn_cast<InsertElementInst>(&
I)) {
1278 Op0 = IE->getOperand(0);
1279 Op1 = IE->getOperand(1);
1281 return thisT()->getVectorInstrCost(
I.getOpcode(), Val,
CostKind,
Index, Op0,
1287 const APInt &DemandedDstElts,
1290 "Unexpected size of DemandedDstElts.");
1308 Cost += thisT()->getScalarizationOverhead(SrcVT, DemandedSrcElts,
1311 Cost += thisT()->getScalarizationOverhead(ReplicatedVT, DemandedDstElts,
1323 assert(!Src->isVoidTy() &&
"Invalid type");
1325 if (getTLI()->getValueType(
DL, Src,
true) == MVT::Other)
1340 LT.second.getSizeInBits())) {
1346 if (Opcode == Instruction::Store)
1355 cast<VectorType>(Src), Opcode != Instruction::Store,
1356 Opcode == Instruction::Store,
CostKind);
1366 return getCommonMaskedMemoryOpCost(Opcode, DataTy, Alignment,
true,
false,
1371 const Value *
Ptr,
bool VariableMask,
1375 return getCommonMaskedMemoryOpCost(Opcode, DataTy, Alignment, VariableMask,
1380 const Value *
Ptr,
bool VariableMask,
1387 return thisT()->getGatherScatterOpCost(Opcode, DataTy,
Ptr, VariableMask,
1394 bool UseMaskForCond =
false,
bool UseMaskForGaps =
false) {
1397 if (isa<ScalableVectorType>(VecTy))
1400 auto *VT = cast<FixedVectorType>(VecTy);
1402 unsigned NumElts = VT->getNumElements();
1403 assert(Factor > 1 && NumElts % Factor == 0 &&
"Invalid interleave factor");
1405 unsigned NumSubElts = NumElts / Factor;
1410 if (UseMaskForCond || UseMaskForGaps)
1411 Cost = thisT()->getMaskedMemoryOpCost(Opcode, VecTy, Alignment,
1420 unsigned VecTySize = thisT()->getDataLayout().getTypeStoreSize(VecTy);
1440 unsigned NumLegalInsts =
divideCeil(VecTySize, VecTyLTSize);
1444 unsigned NumEltsPerLegalInst =
divideCeil(NumElts, NumLegalInsts);
1447 BitVector UsedInsts(NumLegalInsts,
false);
1448 for (
unsigned Index : Indices)
1449 for (
unsigned Elt = 0; Elt < NumSubElts; ++Elt)
1450 UsedInsts.
set((
Index + Elt * Factor) / NumEltsPerLegalInst);
1459 "Interleaved memory op has too many members");
1465 for (
unsigned Index : Indices) {
1466 assert(
Index < Factor &&
"Invalid index for interleaved memory op");
1467 for (
unsigned Elm = 0; Elm < NumSubElts; Elm++)
1468 DemandedLoadStoreElts.
setBit(
Index + Elm * Factor);
1471 if (Opcode == Instruction::Load) {
1481 SubVT, DemandedAllSubElts,
1483 Cost += Indices.
size() * InsSubCost;
1484 Cost += thisT()->getScalarizationOverhead(VT, DemandedLoadStoreElts,
1502 SubVT, DemandedAllSubElts,
1504 Cost += ExtSubCost * Indices.
size();
1505 Cost += thisT()->getScalarizationOverhead(VT, DemandedLoadStoreElts,
1510 if (!UseMaskForCond)
1515 Cost += thisT()->getReplicationShuffleCost(
1516 I8Type, Factor, NumSubElts,
1517 UseMaskForGaps ? DemandedLoadStoreElts : DemandedAllResultElts,
1525 if (UseMaskForGaps) {
1527 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::And, MaskVT,
1552 (
RetTy->isVectorTy() ? cast<VectorType>(
RetTy)->getElementCount()
1561 case Intrinsic::powi:
1562 if (
auto *RHSC = dyn_cast<ConstantInt>(Args[1])) {
1563 bool ShouldOptForSize =
I->getParent()->getParent()->hasOptSize();
1565 ShouldOptForSize)) {
1569 unsigned ActiveBits =
Exponent.getActiveBits();
1570 unsigned PopCount =
Exponent.popcount();
1572 thisT()->getArithmeticInstrCost(
1574 if (RHSC->isNegative())
1575 Cost += thisT()->getArithmeticInstrCost(Instruction::FDiv,
RetTy,
1581 case Intrinsic::cttz:
1587 case Intrinsic::ctlz:
1593 case Intrinsic::memcpy:
1594 return thisT()->getMemcpyCost(ICA.
getInst());
1596 case Intrinsic::masked_scatter: {
1597 const Value *Mask = Args[3];
1598 bool VarMask = !isa<Constant>(Mask);
1599 Align Alignment = cast<ConstantInt>(Args[2])->getAlignValue();
1600 return thisT()->getGatherScatterOpCost(Instruction::Store,
1604 case Intrinsic::masked_gather: {
1605 const Value *Mask = Args[2];
1606 bool VarMask = !isa<Constant>(Mask);
1607 Align Alignment = cast<ConstantInt>(Args[1])->getAlignValue();
1608 return thisT()->getGatherScatterOpCost(Instruction::Load,
RetTy, Args[0],
1611 case Intrinsic::experimental_vp_strided_store: {
1614 const Value *Mask = Args[3];
1615 const Value *EVL = Args[4];
1616 bool VarMask = !isa<Constant>(Mask) || !isa<Constant>(EVL);
1617 Align Alignment =
I->getParamAlign(1).valueOrOne();
1618 return thisT()->getStridedMemoryOpCost(Instruction::Store,
1619 Data->getType(),
Ptr, VarMask,
1622 case Intrinsic::experimental_vp_strided_load: {
1624 const Value *Mask = Args[2];
1625 const Value *EVL = Args[3];
1626 bool VarMask = !isa<Constant>(Mask) || !isa<Constant>(EVL);
1627 Align Alignment =
I->getParamAlign(0).valueOrOne();
1628 return thisT()->getStridedMemoryOpCost(Instruction::Load,
RetTy,
Ptr,
1631 case Intrinsic::experimental_stepvector: {
1632 if (isa<ScalableVectorType>(
RetTy))
1637 case Intrinsic::vector_extract: {
1640 if (isa<ScalableVectorType>(
RetTy))
1642 unsigned Index = cast<ConstantInt>(Args[1])->getZExtValue();
1643 return thisT()->getShuffleCost(
1647 case Intrinsic::vector_insert: {
1650 if (isa<ScalableVectorType>(Args[1]->
getType()))
1652 unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
1653 return thisT()->getShuffleCost(
1657 case Intrinsic::experimental_vector_reverse: {
1658 return thisT()->getShuffleCost(
1662 case Intrinsic::experimental_vector_splice: {
1663 unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
1664 return thisT()->getShuffleCost(
1668 case Intrinsic::vector_reduce_add:
1669 case Intrinsic::vector_reduce_mul:
1670 case Intrinsic::vector_reduce_and:
1671 case Intrinsic::vector_reduce_or:
1672 case Intrinsic::vector_reduce_xor:
1673 case Intrinsic::vector_reduce_smax:
1674 case Intrinsic::vector_reduce_smin:
1675 case Intrinsic::vector_reduce_fmax:
1676 case Intrinsic::vector_reduce_fmin:
1677 case Intrinsic::vector_reduce_fmaximum:
1678 case Intrinsic::vector_reduce_fminimum:
1679 case Intrinsic::vector_reduce_umax:
1680 case Intrinsic::vector_reduce_umin: {
1684 case Intrinsic::vector_reduce_fadd:
1685 case Intrinsic::vector_reduce_fmul: {
1687 IID,
RetTy, {Args[0]->getType(), Args[1]->
getType()}, FMF,
I, 1);
1690 case Intrinsic::fshl:
1691 case Intrinsic::fshr: {
1692 const Value *
X = Args[0];
1693 const Value *
Y = Args[1];
1694 const Value *Z = Args[2];
1707 thisT()->getArithmeticInstrCost(BinaryOperator::Or,
RetTy,
CostKind);
1709 thisT()->getArithmeticInstrCost(BinaryOperator::Sub,
RetTy,
CostKind);
1710 Cost += thisT()->getArithmeticInstrCost(
1713 Cost += thisT()->getArithmeticInstrCost(
1718 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::URem,
RetTy,
1722 Type *CondTy =
RetTy->getWithNewBitWidth(1);
1724 thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
1727 thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
1732 case Intrinsic::get_active_lane_mask: {
1738 if (!getTLI()->shouldExpandGetActiveLaneMask(ResVT, ArgType)) {
1748 thisT()->getTypeBasedIntrinsicInstrCost(Attrs,
CostKind);
1749 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, ExpRetTy,
RetTy,
1761 std::optional<unsigned> FOp =
1764 if (ICA.
getID() == Intrinsic::vp_load) {
1766 if (
auto *VPI = dyn_cast_or_null<VPIntrinsic>(ICA.
getInst()))
1767 Alignment = VPI->getPointerAlignment().valueOrOne();
1771 dyn_cast<PointerType>(ICA.
getArgs()[0]->getType()))
1772 AS = PtrTy->getAddressSpace();
1773 return thisT()->getMemoryOpCost(*FOp, ICA.
getReturnType(), Alignment,
1776 if (ICA.
getID() == Intrinsic::vp_store) {
1778 if (
auto *VPI = dyn_cast_or_null<VPIntrinsic>(ICA.
getInst()))
1779 Alignment = VPI->getPointerAlignment().valueOrOne();
1783 dyn_cast<PointerType>(ICA.
getArgs()[1]->getType()))
1784 AS = PtrTy->getAddressSpace();
1785 return thisT()->getMemoryOpCost(*FOp, Args[0]->
getType(), Alignment,
1789 return thisT()->getArithmeticInstrCost(*FOp, ICA.
getReturnType(),
1794 std::optional<Intrinsic::ID> FID =
1799 "Expected VPIntrinsic to have Mask and Vector Length args and "
1807 *FID != Intrinsic::vector_reduce_fadd &&
1808 *FID != Intrinsic::vector_reduce_fmul)
1813 return thisT()->getIntrinsicInstrCost(NewICA,
CostKind);
1822 ScalarizationCost = 0;
1823 if (!
RetTy->isVoidTy())
1825 cast<VectorType>(
RetTy),
1827 ScalarizationCost +=
1833 return thisT()->getTypeBasedIntrinsicInstrCost(Attrs,
CostKind);
1854 unsigned VecTyIndex = 0;
1855 if (IID == Intrinsic::vector_reduce_fadd ||
1856 IID == Intrinsic::vector_reduce_fmul)
1858 assert(Tys.
size() > VecTyIndex &&
"Unexpected IntrinsicCostAttributes");
1859 VecOpTy = dyn_cast<VectorType>(Tys[VecTyIndex]);
1868 if (isa<ScalableVectorType>(
RetTy) ||
any_of(Tys, [](
const Type *Ty) {
1869 return isa<ScalableVectorType>(Ty);
1875 SkipScalarizationCost ? ScalarizationCostPassed : 0;
1876 unsigned ScalarCalls = 1;
1878 if (
auto *RetVTy = dyn_cast<VectorType>(
RetTy)) {
1879 if (!SkipScalarizationCost)
1882 ScalarCalls = std::max(ScalarCalls,
1883 cast<FixedVectorType>(RetVTy)->getNumElements());
1884 ScalarRetTy =
RetTy->getScalarType();
1887 for (
unsigned i = 0, ie = Tys.
size(); i != ie; ++i) {
1889 if (
auto *VTy = dyn_cast<VectorType>(Ty)) {
1890 if (!SkipScalarizationCost)
1893 ScalarCalls = std::max(ScalarCalls,
1894 cast<FixedVectorType>(VTy)->getNumElements());
1899 if (ScalarCalls == 1)
1904 thisT()->getIntrinsicInstrCost(ScalarAttrs,
CostKind);
1906 return ScalarCalls * ScalarCost + ScalarizationCost;
1910 case Intrinsic::sqrt:
1913 case Intrinsic::sin:
1916 case Intrinsic::cos:
1919 case Intrinsic::exp:
1922 case Intrinsic::exp2:
1925 case Intrinsic::exp10:
1928 case Intrinsic::log:
1931 case Intrinsic::log10:
1934 case Intrinsic::log2:
1937 case Intrinsic::fabs:
1940 case Intrinsic::canonicalize:
1943 case Intrinsic::minnum:
1946 case Intrinsic::maxnum:
1949 case Intrinsic::minimum:
1952 case Intrinsic::maximum:
1955 case Intrinsic::copysign:
1958 case Intrinsic::floor:
1961 case Intrinsic::ceil:
1964 case Intrinsic::trunc:
1967 case Intrinsic::nearbyint:
1970 case Intrinsic::rint:
1973 case Intrinsic::lrint:
1976 case Intrinsic::llrint:
1979 case Intrinsic::round:
1982 case Intrinsic::roundeven:
1985 case Intrinsic::pow:
1988 case Intrinsic::fma:
1991 case Intrinsic::fmuladd:
1994 case Intrinsic::experimental_constrained_fmuladd:
1998 case Intrinsic::lifetime_start:
1999 case Intrinsic::lifetime_end:
2000 case Intrinsic::sideeffect:
2001 case Intrinsic::pseudoprobe:
2002 case Intrinsic::arithmetic_fence:
2004 case Intrinsic::masked_store: {
2006 Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
2007 return thisT()->getMaskedMemoryOpCost(Instruction::Store, Ty, TyAlign, 0,
2010 case Intrinsic::masked_load: {
2012 Align TyAlign = thisT()->DL.getABITypeAlign(Ty);
2013 return thisT()->getMaskedMemoryOpCost(Instruction::Load, Ty, TyAlign, 0,
2016 case Intrinsic::vector_reduce_add:
2017 case Intrinsic::vector_reduce_mul:
2018 case Intrinsic::vector_reduce_and:
2019 case Intrinsic::vector_reduce_or:
2020 case Intrinsic::vector_reduce_xor:
2021 return thisT()->getArithmeticReductionCost(
2024 case Intrinsic::vector_reduce_fadd:
2025 case Intrinsic::vector_reduce_fmul:
2026 return thisT()->getArithmeticReductionCost(
2028 case Intrinsic::vector_reduce_smax:
2029 case Intrinsic::vector_reduce_smin:
2030 case Intrinsic::vector_reduce_umax:
2031 case Intrinsic::vector_reduce_umin:
2032 case Intrinsic::vector_reduce_fmax:
2033 case Intrinsic::vector_reduce_fmin:
2034 case Intrinsic::vector_reduce_fmaximum:
2035 case Intrinsic::vector_reduce_fminimum:
2038 case Intrinsic::abs: {
2040 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2043 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
2045 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
2048 Cost += thisT()->getArithmeticInstrCost(
2052 case Intrinsic::smax:
2053 case Intrinsic::smin:
2054 case Intrinsic::umax:
2055 case Intrinsic::umin: {
2057 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2058 bool IsUnsigned = IID == Intrinsic::umax || IID == Intrinsic::umin;
2062 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
2064 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
2068 case Intrinsic::sadd_sat:
2069 case Intrinsic::ssub_sat: {
2070 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2074 ? Intrinsic::sadd_with_overflow
2075 : Intrinsic::ssub_with_overflow;
2082 nullptr, ScalarizationCostPassed);
2083 Cost += thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2084 Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp,
RetTy, CondTy,
2086 Cost += 2 * thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy,
2090 case Intrinsic::uadd_sat:
2091 case Intrinsic::usub_sat: {
2092 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2096 ? Intrinsic::uadd_with_overflow
2097 : Intrinsic::usub_with_overflow;
2101 nullptr, ScalarizationCostPassed);
2102 Cost += thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2104 thisT()->getCmpSelInstrCost(BinaryOperator::Select,
RetTy, CondTy,
2108 case Intrinsic::smul_fix:
2109 case Intrinsic::umul_fix: {
2110 unsigned ExtSize =
RetTy->getScalarSizeInBits() * 2;
2111 Type *ExtTy =
RetTy->getWithNewBitWidth(ExtSize);
2114 IID == Intrinsic::smul_fix ? Instruction::SExt : Instruction::ZExt;
2120 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2121 Cost += 2 * thisT()->getCastInstrCost(Instruction::Trunc,
RetTy, ExtTy,
2123 Cost += thisT()->getArithmeticInstrCost(Instruction::LShr,
RetTy,
2133 case Intrinsic::sadd_with_overflow:
2134 case Intrinsic::ssub_with_overflow: {
2135 Type *SumTy =
RetTy->getContainedType(0);
2136 Type *OverflowTy =
RetTy->getContainedType(1);
2137 unsigned Opcode = IID == Intrinsic::sadd_with_overflow
2138 ? BinaryOperator::Add
2139 : BinaryOperator::Sub;
2146 Cost += thisT()->getArithmeticInstrCost(Opcode, SumTy,
CostKind);
2147 Cost += 2 * thisT()->getCmpSelInstrCost(
2148 Instruction::ICmp, SumTy, OverflowTy,
2150 Cost += thisT()->getArithmeticInstrCost(BinaryOperator::Xor, OverflowTy,
2154 case Intrinsic::uadd_with_overflow:
2155 case Intrinsic::usub_with_overflow: {
2156 Type *SumTy =
RetTy->getContainedType(0);
2157 Type *OverflowTy =
RetTy->getContainedType(1);
2158 unsigned Opcode = IID == Intrinsic::uadd_with_overflow
2159 ? BinaryOperator::Add
2160 : BinaryOperator::Sub;
2166 Cost += thisT()->getArithmeticInstrCost(Opcode, SumTy,
CostKind);
2168 thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, SumTy, OverflowTy,
2172 case Intrinsic::smul_with_overflow:
2173 case Intrinsic::umul_with_overflow: {
2174 Type *MulTy =
RetTy->getContainedType(0);
2175 Type *OverflowTy =
RetTy->getContainedType(1);
2178 bool IsSigned = IID == Intrinsic::smul_with_overflow;
2180 unsigned ExtOp = IsSigned ? Instruction::SExt : Instruction::ZExt;
2184 Cost += 2 * thisT()->getCastInstrCost(ExtOp, ExtTy, MulTy, CCH,
CostKind);
2186 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2187 Cost += 2 * thisT()->getCastInstrCost(Instruction::Trunc, MulTy, ExtTy,
2189 Cost += thisT()->getArithmeticInstrCost(Instruction::LShr, ExtTy,
2195 Cost += thisT()->getArithmeticInstrCost(Instruction::AShr, MulTy,
2200 Cost += thisT()->getCmpSelInstrCost(
2204 case Intrinsic::fptosi_sat:
2205 case Intrinsic::fptoui_sat: {
2208 Type *FromTy = Tys[0];
2209 bool IsSigned = IID == Intrinsic::fptosi_sat;
2214 Cost += thisT()->getIntrinsicInstrCost(Attrs1,
CostKind);
2217 Cost += thisT()->getIntrinsicInstrCost(Attrs2,
CostKind);
2218 Cost += thisT()->getCastInstrCost(
2219 IsSigned ? Instruction::FPToSI : Instruction::FPToUI,
RetTy, FromTy,
2222 Type *CondTy =
RetTy->getWithNewBitWidth(1);
2223 Cost += thisT()->getCmpSelInstrCost(
2225 Cost += thisT()->getCmpSelInstrCost(
2230 case Intrinsic::ctpop:
2236 case Intrinsic::ctlz:
2239 case Intrinsic::cttz:
2242 case Intrinsic::bswap:
2245 case Intrinsic::bitreverse:
2254 if (IID == Intrinsic::fabs && LT.second.isFloatingPoint() &&
2264 return (LT.first * 2);
2266 return (LT.first * 1);
2270 return (LT.first * 2);
2275 if (IID == Intrinsic::fmuladd)
2276 return thisT()->getArithmeticInstrCost(BinaryOperator::FMul,
RetTy,
2278 thisT()->getArithmeticInstrCost(BinaryOperator::FAdd,
RetTy,
2280 if (IID == Intrinsic::experimental_constrained_fmuladd) {
2282 Intrinsic::experimental_constrained_fmul,
RetTy, Tys);
2284 Intrinsic::experimental_constrained_fadd,
RetTy, Tys);
2285 return thisT()->getIntrinsicInstrCost(FMulAttrs,
CostKind) +
2286 thisT()->getIntrinsicInstrCost(FAddAttrs,
CostKind);
2292 if (
auto *RetVTy = dyn_cast<VectorType>(
RetTy)) {
2294 if (isa<ScalableVectorType>(
RetTy) ||
any_of(Tys, [](
const Type *Ty) {
2295 return isa<ScalableVectorType>(Ty);
2300 SkipScalarizationCost
2301 ? ScalarizationCostPassed
2305 unsigned ScalarCalls = cast<FixedVectorType>(RetVTy)->getNumElements();
2307 for (
unsigned i = 0, ie = Tys.
size(); i != ie; ++i) {
2315 thisT()->getIntrinsicInstrCost(Attrs,
CostKind);
2316 for (
unsigned i = 0, ie = Tys.
size(); i != ie; ++i) {
2317 if (
auto *VTy = dyn_cast<VectorType>(Tys[i])) {
2321 ScalarCalls = std::max(ScalarCalls,
2322 cast<FixedVectorType>(VTy)->getNumElements());
2325 return ScalarCalls * ScalarCost + ScalarizationCost;
2329 return SingleCallCost;
2351 return LT.first.isValid() ? *LT.first.getValue() : 0;
2384 if (isa<ScalableVectorType>(Ty))
2388 unsigned NumVecElts = cast<FixedVectorType>(Ty)->getNumElements();
2389 if ((Opcode == Instruction::Or || Opcode == Instruction::And) &&
2399 return thisT()->getCastInstrCost(Instruction::BitCast, ValTy, Ty,
2401 thisT()->getCmpSelInstrCost(Instruction::ICmp, ValTy,
2405 unsigned NumReduxLevels =
Log2_32(NumVecElts);
2408 std::pair<InstructionCost, MVT> LT = thisT()->getTypeLegalizationCost(Ty);
2409 unsigned LongVectorCount = 0;
2411 LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
2412 while (NumVecElts > MVTLen) {
2418 ArithCost += thisT()->getArithmeticInstrCost(Opcode, SubTy,
CostKind);
2423 NumReduxLevels -= LongVectorCount;
2435 NumReduxLevels * thisT()->getArithmeticInstrCost(Opcode, Ty,
CostKind);
2436 return ShuffleCost + ArithCost +
2437 thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
2461 if (isa<ScalableVectorType>(Ty))
2464 auto *VTy = cast<FixedVectorType>(Ty);
2471 return ExtractCost + ArithCost;
2475 std::optional<FastMathFlags> FMF,
2477 assert(Ty &&
"Unknown reduction vector type");
2490 if (isa<ScalableVectorType>(Ty))
2494 unsigned NumVecElts = cast<FixedVectorType>(Ty)->getNumElements();
2495 unsigned NumReduxLevels =
Log2_32(NumVecElts);
2498 std::pair<InstructionCost, MVT> LT = thisT()->getTypeLegalizationCost(Ty);
2499 unsigned LongVectorCount = 0;
2501 LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
2502 while (NumVecElts > MVTLen) {
2516 NumReduxLevels -= LongVectorCount;
2529 return ShuffleCost + MinMaxCost +
2530 thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty,
2542 thisT()->getArithmeticReductionCost(Opcode, ExtTy, FMF,
CostKind);
2544 IsUnsigned ? Instruction::ZExt : Instruction::SExt, ExtTy, Ty,
2547 return RedCost + ExtCost;
2558 Instruction::Add, ExtTy, std::nullopt,
CostKind);
2560 IsUnsigned ? Instruction::ZExt : Instruction::SExt, ExtTy, Ty,
2564 thisT()->getArithmeticInstrCost(Instruction::Mul, ExtTy,
CostKind);
2566 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.
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)
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)
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 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.
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 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)
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
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 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 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
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 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.
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.
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.
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 ...
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)
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.
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...