AArch64ISelDAGToDAG.cpp [plain text]
#include "AArch64TargetMachine.h"
#include "MCTargetDesc/AArch64AddressingModes.h"
#include "llvm/ADT/APSInt.h"
#include "llvm/CodeGen/SelectionDAGISel.h"
#include "llvm/IR/Function.h" // To access function attributes.
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "aarch64-isel"
namespace {
class AArch64DAGToDAGISel : public SelectionDAGISel {
AArch64TargetMachine &TM;
const AArch64Subtarget *Subtarget;
bool ForCodeSize;
public:
explicit AArch64DAGToDAGISel(AArch64TargetMachine &tm,
CodeGenOpt::Level OptLevel)
: SelectionDAGISel(tm, OptLevel), TM(tm), Subtarget(nullptr),
ForCodeSize(false) {}
const char *getPassName() const override {
return "AArch64 Instruction Selection";
}
bool runOnMachineFunction(MachineFunction &MF) override {
ForCodeSize =
MF.getFunction()->hasFnAttribute(Attribute::OptimizeForSize) ||
MF.getFunction()->hasFnAttribute(Attribute::MinSize);
Subtarget = &MF.getSubtarget<AArch64Subtarget>();
return SelectionDAGISel::runOnMachineFunction(MF);
}
SDNode *Select(SDNode *Node) override;
bool SelectInlineAsmMemoryOperand(const SDValue &Op,
char ConstraintCode,
std::vector<SDValue> &OutOps) override;
SDNode *SelectMLAV64LaneV128(SDNode *N);
SDNode *SelectMULLV64LaneV128(unsigned IntNo, SDNode *N);
bool SelectArithExtendedRegister(SDValue N, SDValue &Reg, SDValue &Shift);
bool SelectArithImmed(SDValue N, SDValue &Val, SDValue &Shift);
bool SelectNegArithImmed(SDValue N, SDValue &Val, SDValue &Shift);
bool SelectArithShiftedRegister(SDValue N, SDValue &Reg, SDValue &Shift) {
return SelectShiftedRegister(N, false, Reg, Shift);
}
bool SelectLogicalShiftedRegister(SDValue N, SDValue &Reg, SDValue &Shift) {
return SelectShiftedRegister(N, true, Reg, Shift);
}
bool SelectAddrModeIndexed8(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeIndexed(N, 1, Base, OffImm);
}
bool SelectAddrModeIndexed16(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeIndexed(N, 2, Base, OffImm);
}
bool SelectAddrModeIndexed32(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeIndexed(N, 4, Base, OffImm);
}
bool SelectAddrModeIndexed64(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeIndexed(N, 8, Base, OffImm);
}
bool SelectAddrModeIndexed128(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeIndexed(N, 16, Base, OffImm);
}
bool SelectAddrModeUnscaled8(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeUnscaled(N, 1, Base, OffImm);
}
bool SelectAddrModeUnscaled16(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeUnscaled(N, 2, Base, OffImm);
}
bool SelectAddrModeUnscaled32(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeUnscaled(N, 4, Base, OffImm);
}
bool SelectAddrModeUnscaled64(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeUnscaled(N, 8, Base, OffImm);
}
bool SelectAddrModeUnscaled128(SDValue N, SDValue &Base, SDValue &OffImm) {
return SelectAddrModeUnscaled(N, 16, Base, OffImm);
}
template<int Width>
bool SelectAddrModeWRO(SDValue N, SDValue &Base, SDValue &Offset,
SDValue &SignExtend, SDValue &DoShift) {
return SelectAddrModeWRO(N, Width / 8, Base, Offset, SignExtend, DoShift);
}
template<int Width>
bool SelectAddrModeXRO(SDValue N, SDValue &Base, SDValue &Offset,
SDValue &SignExtend, SDValue &DoShift) {
return SelectAddrModeXRO(N, Width / 8, Base, Offset, SignExtend, DoShift);
}
SDValue createDTuple(ArrayRef<SDValue> Vecs);
SDValue createQTuple(ArrayRef<SDValue> Vecs);
SDValue createTuple(ArrayRef<SDValue> Vecs, unsigned RegClassIDs[],
unsigned SubRegs[]);
SDNode *SelectTable(SDNode *N, unsigned NumVecs, unsigned Opc, bool isExt);
SDNode *SelectIndexedLoad(SDNode *N, bool &Done);
SDNode *SelectLoad(SDNode *N, unsigned NumVecs, unsigned Opc,
unsigned SubRegIdx);
SDNode *SelectPostLoad(SDNode *N, unsigned NumVecs, unsigned Opc,
unsigned SubRegIdx);
SDNode *SelectLoadLane(SDNode *N, unsigned NumVecs, unsigned Opc);
SDNode *SelectPostLoadLane(SDNode *N, unsigned NumVecs, unsigned Opc);
SDNode *SelectStore(SDNode *N, unsigned NumVecs, unsigned Opc);
SDNode *SelectPostStore(SDNode *N, unsigned NumVecs, unsigned Opc);
SDNode *SelectStoreLane(SDNode *N, unsigned NumVecs, unsigned Opc);
SDNode *SelectPostStoreLane(SDNode *N, unsigned NumVecs, unsigned Opc);
SDNode *SelectBitfieldExtractOp(SDNode *N);
SDNode *SelectBitfieldInsertOp(SDNode *N);
SDNode *SelectLIBM(SDNode *N);
#include "AArch64GenDAGISel.inc"
private:
bool SelectShiftedRegister(SDValue N, bool AllowROR, SDValue &Reg,
SDValue &Shift);
bool SelectAddrModeIndexed(SDValue N, unsigned Size, SDValue &Base,
SDValue &OffImm);
bool SelectAddrModeUnscaled(SDValue N, unsigned Size, SDValue &Base,
SDValue &OffImm);
bool SelectAddrModeWRO(SDValue N, unsigned Size, SDValue &Base,
SDValue &Offset, SDValue &SignExtend,
SDValue &DoShift);
bool SelectAddrModeXRO(SDValue N, unsigned Size, SDValue &Base,
SDValue &Offset, SDValue &SignExtend,
SDValue &DoShift);
bool isWorthFolding(SDValue V) const;
bool SelectExtendedSHL(SDValue N, unsigned Size, bool WantExtend,
SDValue &Offset, SDValue &SignExtend);
template<unsigned RegWidth>
bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos) {
return SelectCVTFixedPosOperand(N, FixedPos, RegWidth);
}
bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos, unsigned Width);
};
}
static bool isIntImmediate(const SDNode *N, uint64_t &Imm) {
if (const ConstantSDNode *C = dyn_cast<const ConstantSDNode>(N)) {
Imm = C->getZExtValue();
return true;
}
return false;
}
static bool isIntImmediate(SDValue N, uint64_t &Imm) {
return isIntImmediate(N.getNode(), Imm);
}
static bool isOpcWithIntImmediate(const SDNode *N, unsigned Opc,
uint64_t &Imm) {
return N->getOpcode() == Opc &&
isIntImmediate(N->getOperand(1).getNode(), Imm);
}
bool AArch64DAGToDAGISel::SelectInlineAsmMemoryOperand(
const SDValue &Op, char ConstraintCode, std::vector<SDValue> &OutOps) {
assert(ConstraintCode == 'm' && "unexpected asm memory constraint");
OutOps.push_back(Op);
return false;
}
bool AArch64DAGToDAGISel::SelectArithImmed(SDValue N, SDValue &Val,
SDValue &Shift) {
if (!isa<ConstantSDNode>(N.getNode()))
return false;
uint64_t Immed = cast<ConstantSDNode>(N.getNode())->getZExtValue();
unsigned ShiftAmt;
if (Immed >> 12 == 0) {
ShiftAmt = 0;
} else if ((Immed & 0xfff) == 0 && Immed >> 24 == 0) {
ShiftAmt = 12;
Immed = Immed >> 12;
} else
return false;
unsigned ShVal = AArch64_AM::getShifterImm(AArch64_AM::LSL, ShiftAmt);
Val = CurDAG->getTargetConstant(Immed, MVT::i32);
Shift = CurDAG->getTargetConstant(ShVal, MVT::i32);
return true;
}
bool AArch64DAGToDAGISel::SelectNegArithImmed(SDValue N, SDValue &Val,
SDValue &Shift) {
if (!isa<ConstantSDNode>(N.getNode()))
return false;
uint64_t Immed = cast<ConstantSDNode>(N.getNode())->getZExtValue();
if (Immed == 0)
return false;
if (N.getValueType() == MVT::i32)
Immed = ~((uint32_t)Immed) + 1;
else
Immed = ~Immed + 1ULL;
if (Immed & 0xFFFFFFFFFF000000ULL)
return false;
Immed &= 0xFFFFFFULL;
return SelectArithImmed(CurDAG->getConstant(Immed, MVT::i32), Val, Shift);
}
static AArch64_AM::ShiftExtendType getShiftTypeForNode(SDValue N) {
switch (N.getOpcode()) {
default:
return AArch64_AM::InvalidShiftExtend;
case ISD::SHL:
return AArch64_AM::LSL;
case ISD::SRL:
return AArch64_AM::LSR;
case ISD::SRA:
return AArch64_AM::ASR;
case ISD::ROTR:
return AArch64_AM::ROR;
}
}
bool AArch64DAGToDAGISel::isWorthFolding(SDValue V) const {
if (ForCodeSize || V.hasOneUse())
return true;
return false;
}
bool AArch64DAGToDAGISel::SelectShiftedRegister(SDValue N, bool AllowROR,
SDValue &Reg, SDValue &Shift) {
AArch64_AM::ShiftExtendType ShType = getShiftTypeForNode(N);
if (ShType == AArch64_AM::InvalidShiftExtend)
return false;
if (!AllowROR && ShType == AArch64_AM::ROR)
return false;
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
unsigned BitSize = N.getValueType().getSizeInBits();
unsigned Val = RHS->getZExtValue() & (BitSize - 1);
unsigned ShVal = AArch64_AM::getShifterImm(ShType, Val);
Reg = N.getOperand(0);
Shift = CurDAG->getTargetConstant(ShVal, MVT::i32);
return isWorthFolding(N);
}
return false;
}
static AArch64_AM::ShiftExtendType
getExtendTypeForNode(SDValue N, bool IsLoadStore = false) {
if (N.getOpcode() == ISD::SIGN_EXTEND ||
N.getOpcode() == ISD::SIGN_EXTEND_INREG) {
EVT SrcVT;
if (N.getOpcode() == ISD::SIGN_EXTEND_INREG)
SrcVT = cast<VTSDNode>(N.getOperand(1))->getVT();
else
SrcVT = N.getOperand(0).getValueType();
if (!IsLoadStore && SrcVT == MVT::i8)
return AArch64_AM::SXTB;
else if (!IsLoadStore && SrcVT == MVT::i16)
return AArch64_AM::SXTH;
else if (SrcVT == MVT::i32)
return AArch64_AM::SXTW;
assert(SrcVT != MVT::i64 && "extend from 64-bits?");
return AArch64_AM::InvalidShiftExtend;
} else if (N.getOpcode() == ISD::ZERO_EXTEND ||
N.getOpcode() == ISD::ANY_EXTEND) {
EVT SrcVT = N.getOperand(0).getValueType();
if (!IsLoadStore && SrcVT == MVT::i8)
return AArch64_AM::UXTB;
else if (!IsLoadStore && SrcVT == MVT::i16)
return AArch64_AM::UXTH;
else if (SrcVT == MVT::i32)
return AArch64_AM::UXTW;
assert(SrcVT != MVT::i64 && "extend from 64-bits?");
return AArch64_AM::InvalidShiftExtend;
} else if (N.getOpcode() == ISD::AND) {
ConstantSDNode *CSD = dyn_cast<ConstantSDNode>(N.getOperand(1));
if (!CSD)
return AArch64_AM::InvalidShiftExtend;
uint64_t AndMask = CSD->getZExtValue();
switch (AndMask) {
default:
return AArch64_AM::InvalidShiftExtend;
case 0xFF:
return !IsLoadStore ? AArch64_AM::UXTB : AArch64_AM::InvalidShiftExtend;
case 0xFFFF:
return !IsLoadStore ? AArch64_AM::UXTH : AArch64_AM::InvalidShiftExtend;
case 0xFFFFFFFF:
return AArch64_AM::UXTW;
}
}
return AArch64_AM::InvalidShiftExtend;
}
static bool checkHighLaneIndex(SDNode *DL, SDValue &LaneOp, int &LaneIdx) {
if (DL->getOpcode() != AArch64ISD::DUPLANE16 &&
DL->getOpcode() != AArch64ISD::DUPLANE32)
return false;
SDValue SV = DL->getOperand(0);
if (SV.getOpcode() != ISD::INSERT_SUBVECTOR)
return false;
SDValue EV = SV.getOperand(1);
if (EV.getOpcode() != ISD::EXTRACT_SUBVECTOR)
return false;
ConstantSDNode *DLidx = cast<ConstantSDNode>(DL->getOperand(1).getNode());
ConstantSDNode *EVidx = cast<ConstantSDNode>(EV.getOperand(1).getNode());
LaneIdx = DLidx->getSExtValue() + EVidx->getSExtValue();
LaneOp = EV.getOperand(0);
return true;
}
static bool checkV64LaneV128(SDValue Op0, SDValue Op1, SDValue &StdOp,
SDValue &LaneOp, int &LaneIdx) {
if (!checkHighLaneIndex(Op0.getNode(), LaneOp, LaneIdx)) {
std::swap(Op0, Op1);
if (!checkHighLaneIndex(Op0.getNode(), LaneOp, LaneIdx))
return false;
}
StdOp = Op1;
return true;
}
SDNode *AArch64DAGToDAGISel::SelectMLAV64LaneV128(SDNode *N) {
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
SDValue MLAOp1; SDValue MLAOp2; int LaneIdx = -1;
if (Op1.getOpcode() != ISD::MUL ||
!checkV64LaneV128(Op1.getOperand(0), Op1.getOperand(1), MLAOp1, MLAOp2,
LaneIdx)) {
std::swap(Op0, Op1);
if (Op1.getOpcode() != ISD::MUL ||
!checkV64LaneV128(Op1.getOperand(0), Op1.getOperand(1), MLAOp1, MLAOp2,
LaneIdx))
return nullptr;
}
SDValue LaneIdxVal = CurDAG->getTargetConstant(LaneIdx, MVT::i64);
SDValue Ops[] = { Op0, MLAOp1, MLAOp2, LaneIdxVal };
unsigned MLAOpc = ~0U;
switch (N->getSimpleValueType(0).SimpleTy) {
default:
llvm_unreachable("Unrecognized MLA.");
case MVT::v4i16:
MLAOpc = AArch64::MLAv4i16_indexed;
break;
case MVT::v8i16:
MLAOpc = AArch64::MLAv8i16_indexed;
break;
case MVT::v2i32:
MLAOpc = AArch64::MLAv2i32_indexed;
break;
case MVT::v4i32:
MLAOpc = AArch64::MLAv4i32_indexed;
break;
}
return CurDAG->getMachineNode(MLAOpc, SDLoc(N), N->getValueType(0), Ops);
}
SDNode *AArch64DAGToDAGISel::SelectMULLV64LaneV128(unsigned IntNo, SDNode *N) {
SDValue SMULLOp0;
SDValue SMULLOp1;
int LaneIdx;
if (!checkV64LaneV128(N->getOperand(1), N->getOperand(2), SMULLOp0, SMULLOp1,
LaneIdx))
return nullptr;
SDValue LaneIdxVal = CurDAG->getTargetConstant(LaneIdx, MVT::i64);
SDValue Ops[] = { SMULLOp0, SMULLOp1, LaneIdxVal };
unsigned SMULLOpc = ~0U;
if (IntNo == Intrinsic::aarch64_neon_smull) {
switch (N->getSimpleValueType(0).SimpleTy) {
default:
llvm_unreachable("Unrecognized SMULL.");
case MVT::v4i32:
SMULLOpc = AArch64::SMULLv4i16_indexed;
break;
case MVT::v2i64:
SMULLOpc = AArch64::SMULLv2i32_indexed;
break;
}
} else if (IntNo == Intrinsic::aarch64_neon_umull) {
switch (N->getSimpleValueType(0).SimpleTy) {
default:
llvm_unreachable("Unrecognized SMULL.");
case MVT::v4i32:
SMULLOpc = AArch64::UMULLv4i16_indexed;
break;
case MVT::v2i64:
SMULLOpc = AArch64::UMULLv2i32_indexed;
break;
}
} else
llvm_unreachable("Unrecognized intrinsic.");
return CurDAG->getMachineNode(SMULLOpc, SDLoc(N), N->getValueType(0), Ops);
}
static SDValue narrowIfNeeded(SelectionDAG *CurDAG, SDValue N) {
if (N.getValueType() == MVT::i32)
return N;
SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
MachineSDNode *Node = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
SDLoc(N), MVT::i32, N, SubReg);
return SDValue(Node, 0);
}
bool AArch64DAGToDAGISel::SelectArithExtendedRegister(SDValue N, SDValue &Reg,
SDValue &Shift) {
unsigned ShiftVal = 0;
AArch64_AM::ShiftExtendType Ext;
if (N.getOpcode() == ISD::SHL) {
ConstantSDNode *CSD = dyn_cast<ConstantSDNode>(N.getOperand(1));
if (!CSD)
return false;
ShiftVal = CSD->getZExtValue();
if (ShiftVal > 4)
return false;
Ext = getExtendTypeForNode(N.getOperand(0));
if (Ext == AArch64_AM::InvalidShiftExtend)
return false;
Reg = N.getOperand(0).getOperand(0);
} else {
Ext = getExtendTypeForNode(N);
if (Ext == AArch64_AM::InvalidShiftExtend)
return false;
Reg = N.getOperand(0);
}
assert(Ext != AArch64_AM::UXTX && Ext != AArch64_AM::SXTX);
Reg = narrowIfNeeded(CurDAG, Reg);
Shift = CurDAG->getTargetConstant(getArithExtendImm(Ext, ShiftVal), MVT::i32);
return isWorthFolding(N);
}
static bool isWorthFoldingADDlow(SDValue N) {
for (auto Use : N->uses()) {
if (Use->getOpcode() != ISD::LOAD && Use->getOpcode() != ISD::STORE &&
Use->getOpcode() != ISD::ATOMIC_LOAD &&
Use->getOpcode() != ISD::ATOMIC_STORE)
return false;
if (cast<MemSDNode>(Use)->getOrdering() > Monotonic)
return false;
}
return true;
}
bool AArch64DAGToDAGISel::SelectAddrModeIndexed(SDValue N, unsigned Size,
SDValue &Base, SDValue &OffImm) {
const TargetLowering *TLI = getTargetLowering();
if (N.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(N)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy());
OffImm = CurDAG->getTargetConstant(0, MVT::i64);
return true;
}
if (N.getOpcode() == AArch64ISD::ADDlow && isWorthFoldingADDlow(N)) {
GlobalAddressSDNode *GAN =
dyn_cast<GlobalAddressSDNode>(N.getOperand(1).getNode());
Base = N.getOperand(0);
OffImm = N.getOperand(1);
if (!GAN)
return true;
const GlobalValue *GV = GAN->getGlobal();
unsigned Alignment = GV->getAlignment();
const DataLayout *DL = TLI->getDataLayout();
Type *Ty = GV->getType()->getElementType();
if (Alignment == 0 && Ty->isSized())
Alignment = DL->getABITypeAlignment(Ty);
if (Alignment >= Size)
return true;
}
if (CurDAG->isBaseWithConstantOffset(N)) {
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
int64_t RHSC = (int64_t)RHS->getZExtValue();
unsigned Scale = Log2_32(Size);
if ((RHSC & (Size - 1)) == 0 && RHSC >= 0 && RHSC < (0x1000 << Scale)) {
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy());
}
OffImm = CurDAG->getTargetConstant(RHSC >> Scale, MVT::i64);
return true;
}
}
}
if (SelectAddrModeUnscaled(N, Size, Base, OffImm))
return false;
Base = N;
OffImm = CurDAG->getTargetConstant(0, MVT::i64);
return true;
}
bool AArch64DAGToDAGISel::SelectAddrModeUnscaled(SDValue N, unsigned Size,
SDValue &Base,
SDValue &OffImm) {
if (!CurDAG->isBaseWithConstantOffset(N))
return false;
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(N.getOperand(1))) {
int64_t RHSC = RHS->getSExtValue();
if ((RHSC & (Size - 1)) == 0 && RHSC >= 0 &&
RHSC < (0x1000 << Log2_32(Size)))
return false;
if (RHSC >= -256 && RHSC < 256) {
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
const TargetLowering *TLI = getTargetLowering();
Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy());
}
OffImm = CurDAG->getTargetConstant(RHSC, MVT::i64);
return true;
}
}
return false;
}
static SDValue Widen(SelectionDAG *CurDAG, SDValue N) {
SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
SDValue ImpDef = SDValue(
CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, SDLoc(N), MVT::i64),
0);
MachineSDNode *Node = CurDAG->getMachineNode(
TargetOpcode::INSERT_SUBREG, SDLoc(N), MVT::i64, ImpDef, N, SubReg);
return SDValue(Node, 0);
}
bool AArch64DAGToDAGISel::SelectExtendedSHL(SDValue N, unsigned Size,
bool WantExtend, SDValue &Offset,
SDValue &SignExtend) {
assert(N.getOpcode() == ISD::SHL && "Invalid opcode.");
ConstantSDNode *CSD = dyn_cast<ConstantSDNode>(N.getOperand(1));
if (!CSD || (CSD->getZExtValue() & 0x7) != CSD->getZExtValue())
return false;
if (WantExtend) {
AArch64_AM::ShiftExtendType Ext =
getExtendTypeForNode(N.getOperand(0), true);
if (Ext == AArch64_AM::InvalidShiftExtend)
return false;
Offset = narrowIfNeeded(CurDAG, N.getOperand(0).getOperand(0));
SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, MVT::i32);
} else {
Offset = N.getOperand(0);
SignExtend = CurDAG->getTargetConstant(0, MVT::i32);
}
unsigned LegalShiftVal = Log2_32(Size);
unsigned ShiftVal = CSD->getZExtValue();
if (ShiftVal != 0 && ShiftVal != LegalShiftVal)
return false;
if (isWorthFolding(N))
return true;
return false;
}
bool AArch64DAGToDAGISel::SelectAddrModeWRO(SDValue N, unsigned Size,
SDValue &Base, SDValue &Offset,
SDValue &SignExtend,
SDValue &DoShift) {
if (N.getOpcode() != ISD::ADD)
return false;
SDValue LHS = N.getOperand(0);
SDValue RHS = N.getOperand(1);
if (isa<ConstantSDNode>(LHS) || isa<ConstantSDNode>(RHS))
return false;
const SDNode *Node = N.getNode();
for (SDNode *UI : Node->uses()) {
if (!isa<MemSDNode>(*UI))
return false;
}
bool IsExtendedRegisterWorthFolding = isWorthFolding(N);
if (IsExtendedRegisterWorthFolding && RHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(RHS, Size, true, Offset, SignExtend)) {
Base = LHS;
DoShift = CurDAG->getTargetConstant(true, MVT::i32);
return true;
}
if (IsExtendedRegisterWorthFolding && LHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(LHS, Size, true, Offset, SignExtend)) {
Base = RHS;
DoShift = CurDAG->getTargetConstant(true, MVT::i32);
return true;
}
DoShift = CurDAG->getTargetConstant(false, MVT::i32);
AArch64_AM::ShiftExtendType Ext = AArch64_AM::InvalidShiftExtend;
if (IsExtendedRegisterWorthFolding &&
(Ext = getExtendTypeForNode(LHS, true)) !=
AArch64_AM::InvalidShiftExtend) {
Base = RHS;
Offset = narrowIfNeeded(CurDAG, LHS.getOperand(0));
SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, MVT::i32);
if (isWorthFolding(LHS))
return true;
}
if (IsExtendedRegisterWorthFolding &&
(Ext = getExtendTypeForNode(RHS, true)) !=
AArch64_AM::InvalidShiftExtend) {
Base = LHS;
Offset = narrowIfNeeded(CurDAG, RHS.getOperand(0));
SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, MVT::i32);
if (isWorthFolding(RHS))
return true;
}
return false;
}
static bool isPreferredADD(int64_t ImmOff) {
if ((ImmOff & 0xfffffffffffff000LL) == 0x0LL)
return true;
if ((ImmOff & 0xffffffffff000fffLL) == 0x0LL)
return (ImmOff & 0xffffffffff00ffffLL) != 0x0LL &&
(ImmOff & 0xffffffffffff0fffLL) != 0x0LL;
return false;
}
bool AArch64DAGToDAGISel::SelectAddrModeXRO(SDValue N, unsigned Size,
SDValue &Base, SDValue &Offset,
SDValue &SignExtend,
SDValue &DoShift) {
if (N.getOpcode() != ISD::ADD)
return false;
SDValue LHS = N.getOperand(0);
SDValue RHS = N.getOperand(1);
const SDNode *Node = N.getNode();
for (SDNode *UI : Node->uses()) {
if (!isa<MemSDNode>(*UI))
return false;
}
if (isa<ConstantSDNode>(RHS)) {
int64_t ImmOff = (int64_t)dyn_cast<ConstantSDNode>(RHS)->getZExtValue();
unsigned Scale = Log2_32(Size);
if ((ImmOff % Size == 0 && ImmOff >= 0 && ImmOff < (0x1000 << Scale)) ||
isPreferredADD(ImmOff) || isPreferredADD(-ImmOff))
return false;
SDLoc DL(N.getNode());
SDValue Ops[] = { RHS };
SDNode *MOVI =
CurDAG->getMachineNode(AArch64::MOVi64imm, DL, MVT::i64, Ops);
SDValue MOVIV = SDValue(MOVI, 0);
N = CurDAG->getNode(ISD::ADD, DL, MVT::i64, LHS, MOVIV);
}
bool IsExtendedRegisterWorthFolding = isWorthFolding(N);
if (IsExtendedRegisterWorthFolding && RHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(RHS, Size, false, Offset, SignExtend)) {
Base = LHS;
DoShift = CurDAG->getTargetConstant(true, MVT::i32);
return true;
}
if (IsExtendedRegisterWorthFolding && LHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(LHS, Size, false, Offset, SignExtend)) {
Base = RHS;
DoShift = CurDAG->getTargetConstant(true, MVT::i32);
return true;
}
Base = LHS;
Offset = RHS;
SignExtend = CurDAG->getTargetConstant(false, MVT::i32);
DoShift = CurDAG->getTargetConstant(false, MVT::i32);
return true;
}
SDValue AArch64DAGToDAGISel::createDTuple(ArrayRef<SDValue> Regs) {
static unsigned RegClassIDs[] = {
AArch64::DDRegClassID, AArch64::DDDRegClassID, AArch64::DDDDRegClassID};
static unsigned SubRegs[] = { AArch64::dsub0, AArch64::dsub1,
AArch64::dsub2, AArch64::dsub3 };
return createTuple(Regs, RegClassIDs, SubRegs);
}
SDValue AArch64DAGToDAGISel::createQTuple(ArrayRef<SDValue> Regs) {
static unsigned RegClassIDs[] = {
AArch64::QQRegClassID, AArch64::QQQRegClassID, AArch64::QQQQRegClassID};
static unsigned SubRegs[] = { AArch64::qsub0, AArch64::qsub1,
AArch64::qsub2, AArch64::qsub3 };
return createTuple(Regs, RegClassIDs, SubRegs);
}
SDValue AArch64DAGToDAGISel::createTuple(ArrayRef<SDValue> Regs,
unsigned RegClassIDs[],
unsigned SubRegs[]) {
if (Regs.size() == 1)
return Regs[0];
assert(Regs.size() >= 2 && Regs.size() <= 4);
SDLoc DL(Regs[0].getNode());
SmallVector<SDValue, 4> Ops;
Ops.push_back(
CurDAG->getTargetConstant(RegClassIDs[Regs.size() - 2], MVT::i32));
for (unsigned i = 0; i < Regs.size(); ++i) {
Ops.push_back(Regs[i]);
Ops.push_back(CurDAG->getTargetConstant(SubRegs[i], MVT::i32));
}
SDNode *N =
CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, DL, MVT::Untyped, Ops);
return SDValue(N, 0);
}
SDNode *AArch64DAGToDAGISel::SelectTable(SDNode *N, unsigned NumVecs,
unsigned Opc, bool isExt) {
SDLoc dl(N);
EVT VT = N->getValueType(0);
unsigned ExtOff = isExt;
unsigned Vec0Off = ExtOff + 1;
SmallVector<SDValue, 4> Regs(N->op_begin() + Vec0Off,
N->op_begin() + Vec0Off + NumVecs);
SDValue RegSeq = createQTuple(Regs);
SmallVector<SDValue, 6> Ops;
if (isExt)
Ops.push_back(N->getOperand(1));
Ops.push_back(RegSeq);
Ops.push_back(N->getOperand(NumVecs + ExtOff + 1));
return CurDAG->getMachineNode(Opc, dl, VT, Ops);
}
SDNode *AArch64DAGToDAGISel::SelectIndexedLoad(SDNode *N, bool &Done) {
LoadSDNode *LD = cast<LoadSDNode>(N);
if (LD->isUnindexed())
return nullptr;
EVT VT = LD->getMemoryVT();
EVT DstVT = N->getValueType(0);
ISD::MemIndexedMode AM = LD->getAddressingMode();
bool IsPre = AM == ISD::PRE_INC || AM == ISD::PRE_DEC;
unsigned Opcode = 0;
ISD::LoadExtType ExtType = LD->getExtensionType();
bool InsertTo64 = false;
if (VT == MVT::i64)
Opcode = IsPre ? AArch64::LDRXpre : AArch64::LDRXpost;
else if (VT == MVT::i32) {
if (ExtType == ISD::NON_EXTLOAD)
Opcode = IsPre ? AArch64::LDRWpre : AArch64::LDRWpost;
else if (ExtType == ISD::SEXTLOAD)
Opcode = IsPre ? AArch64::LDRSWpre : AArch64::LDRSWpost;
else {
Opcode = IsPre ? AArch64::LDRWpre : AArch64::LDRWpost;
InsertTo64 = true;
DstVT = MVT::i32;
}
} else if (VT == MVT::i16) {
if (ExtType == ISD::SEXTLOAD) {
if (DstVT == MVT::i64)
Opcode = IsPre ? AArch64::LDRSHXpre : AArch64::LDRSHXpost;
else
Opcode = IsPre ? AArch64::LDRSHWpre : AArch64::LDRSHWpost;
} else {
Opcode = IsPre ? AArch64::LDRHHpre : AArch64::LDRHHpost;
InsertTo64 = DstVT == MVT::i64;
DstVT = MVT::i32;
}
} else if (VT == MVT::i8) {
if (ExtType == ISD::SEXTLOAD) {
if (DstVT == MVT::i64)
Opcode = IsPre ? AArch64::LDRSBXpre : AArch64::LDRSBXpost;
else
Opcode = IsPre ? AArch64::LDRSBWpre : AArch64::LDRSBWpost;
} else {
Opcode = IsPre ? AArch64::LDRBBpre : AArch64::LDRBBpost;
InsertTo64 = DstVT == MVT::i64;
DstVT = MVT::i32;
}
} else if (VT == MVT::f32) {
Opcode = IsPre ? AArch64::LDRSpre : AArch64::LDRSpost;
} else if (VT == MVT::f64 || VT.is64BitVector()) {
Opcode = IsPre ? AArch64::LDRDpre : AArch64::LDRDpost;
} else if (VT.is128BitVector()) {
Opcode = IsPre ? AArch64::LDRQpre : AArch64::LDRQpost;
} else
return nullptr;
SDValue Chain = LD->getChain();
SDValue Base = LD->getBasePtr();
ConstantSDNode *OffsetOp = cast<ConstantSDNode>(LD->getOffset());
int OffsetVal = (int)OffsetOp->getZExtValue();
SDValue Offset = CurDAG->getTargetConstant(OffsetVal, MVT::i64);
SDValue Ops[] = { Base, Offset, Chain };
SDNode *Res = CurDAG->getMachineNode(Opcode, SDLoc(N), MVT::i64, DstVT,
MVT::Other, Ops);
Done = true;
SDValue LoadedVal = SDValue(Res, 1);
if (InsertTo64) {
SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
LoadedVal =
SDValue(CurDAG->getMachineNode(
AArch64::SUBREG_TO_REG, SDLoc(N), MVT::i64,
CurDAG->getTargetConstant(0, MVT::i64), LoadedVal, SubReg),
0);
}
ReplaceUses(SDValue(N, 0), LoadedVal);
ReplaceUses(SDValue(N, 1), SDValue(Res, 0));
ReplaceUses(SDValue(N, 2), SDValue(Res, 2));
return nullptr;
}
SDNode *AArch64DAGToDAGISel::SelectLoad(SDNode *N, unsigned NumVecs,
unsigned Opc, unsigned SubRegIdx) {
SDLoc dl(N);
EVT VT = N->getValueType(0);
SDValue Chain = N->getOperand(0);
SmallVector<SDValue, 6> Ops;
Ops.push_back(N->getOperand(2)); Ops.push_back(Chain);
std::vector<EVT> ResTys;
ResTys.push_back(MVT::Untyped);
ResTys.push_back(MVT::Other);
SDNode *Ld = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
SDValue SuperReg = SDValue(Ld, 0);
for (unsigned i = 0; i < NumVecs; ++i)
ReplaceUses(SDValue(N, i),
CurDAG->getTargetExtractSubreg(SubRegIdx + i, dl, VT, SuperReg));
ReplaceUses(SDValue(N, NumVecs), SDValue(Ld, 1));
return nullptr;
}
SDNode *AArch64DAGToDAGISel::SelectPostLoad(SDNode *N, unsigned NumVecs,
unsigned Opc, unsigned SubRegIdx) {
SDLoc dl(N);
EVT VT = N->getValueType(0);
SDValue Chain = N->getOperand(0);
SmallVector<SDValue, 6> Ops;
Ops.push_back(N->getOperand(1)); Ops.push_back(N->getOperand(2)); Ops.push_back(Chain);
std::vector<EVT> ResTys;
ResTys.push_back(MVT::i64); ResTys.push_back(MVT::Untyped);
ResTys.push_back(MVT::Other);
SDNode *Ld = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
ReplaceUses(SDValue(N, NumVecs), SDValue(Ld, 0));
SDValue SuperReg = SDValue(Ld, 1);
if (NumVecs == 1)
ReplaceUses(SDValue(N, 0), SuperReg);
else
for (unsigned i = 0; i < NumVecs; ++i)
ReplaceUses(SDValue(N, i),
CurDAG->getTargetExtractSubreg(SubRegIdx + i, dl, VT, SuperReg));
ReplaceUses(SDValue(N, NumVecs + 1), SDValue(Ld, 2));
return nullptr;
}
SDNode *AArch64DAGToDAGISel::SelectStore(SDNode *N, unsigned NumVecs,
unsigned Opc) {
SDLoc dl(N);
EVT VT = N->getOperand(2)->getValueType(0);
bool Is128Bit = VT.getSizeInBits() == 128;
SmallVector<SDValue, 4> Regs(N->op_begin() + 2, N->op_begin() + 2 + NumVecs);
SDValue RegSeq = Is128Bit ? createQTuple(Regs) : createDTuple(Regs);
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(N->getOperand(NumVecs + 2));
Ops.push_back(N->getOperand(0));
SDNode *St = CurDAG->getMachineNode(Opc, dl, N->getValueType(0), Ops);
return St;
}
SDNode *AArch64DAGToDAGISel::SelectPostStore(SDNode *N, unsigned NumVecs,
unsigned Opc) {
SDLoc dl(N);
EVT VT = N->getOperand(2)->getValueType(0);
SmallVector<EVT, 2> ResTys;
ResTys.push_back(MVT::i64); ResTys.push_back(MVT::Other);
bool Is128Bit = VT.getSizeInBits() == 128;
SmallVector<SDValue, 4> Regs(N->op_begin() + 1, N->op_begin() + 1 + NumVecs);
SDValue RegSeq = Is128Bit ? createQTuple(Regs) : createDTuple(Regs);
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(N->getOperand(NumVecs + 1)); Ops.push_back(N->getOperand(NumVecs + 2)); Ops.push_back(N->getOperand(0)); SDNode *St = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
return St;
}
namespace {
class WidenVector {
SelectionDAG &DAG;
public:
WidenVector(SelectionDAG &DAG) : DAG(DAG) {}
SDValue operator()(SDValue V64Reg) {
EVT VT = V64Reg.getValueType();
unsigned NarrowSize = VT.getVectorNumElements();
MVT EltTy = VT.getVectorElementType().getSimpleVT();
MVT WideTy = MVT::getVectorVT(EltTy, 2 * NarrowSize);
SDLoc DL(V64Reg);
SDValue Undef =
SDValue(DAG.getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, WideTy), 0);
return DAG.getTargetInsertSubreg(AArch64::dsub, DL, WideTy, Undef, V64Reg);
}
};
}
static SDValue NarrowVector(SDValue V128Reg, SelectionDAG &DAG) {
EVT VT = V128Reg.getValueType();
unsigned WideSize = VT.getVectorNumElements();
MVT EltTy = VT.getVectorElementType().getSimpleVT();
MVT NarrowTy = MVT::getVectorVT(EltTy, WideSize / 2);
return DAG.getTargetExtractSubreg(AArch64::dsub, SDLoc(V128Reg), NarrowTy,
V128Reg);
}
SDNode *AArch64DAGToDAGISel::SelectLoadLane(SDNode *N, unsigned NumVecs,
unsigned Opc) {
SDLoc dl(N);
EVT VT = N->getValueType(0);
bool Narrow = VT.getSizeInBits() == 64;
SmallVector<SDValue, 4> Regs(N->op_begin() + 2, N->op_begin() + 2 + NumVecs);
if (Narrow)
std::transform(Regs.begin(), Regs.end(), Regs.begin(),
WidenVector(*CurDAG));
SDValue RegSeq = createQTuple(Regs);
std::vector<EVT> ResTys;
ResTys.push_back(MVT::Untyped);
ResTys.push_back(MVT::Other);
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 2))->getZExtValue();
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(CurDAG->getTargetConstant(LaneNo, MVT::i64));
Ops.push_back(N->getOperand(NumVecs + 3));
Ops.push_back(N->getOperand(0));
SDNode *Ld = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
SDValue SuperReg = SDValue(Ld, 0);
EVT WideVT = RegSeq.getOperand(1)->getValueType(0);
static unsigned QSubs[] = { AArch64::qsub0, AArch64::qsub1, AArch64::qsub2,
AArch64::qsub3 };
for (unsigned i = 0; i < NumVecs; ++i) {
SDValue NV = CurDAG->getTargetExtractSubreg(QSubs[i], dl, WideVT, SuperReg);
if (Narrow)
NV = NarrowVector(NV, *CurDAG);
ReplaceUses(SDValue(N, i), NV);
}
ReplaceUses(SDValue(N, NumVecs), SDValue(Ld, 1));
return Ld;
}
SDNode *AArch64DAGToDAGISel::SelectPostLoadLane(SDNode *N, unsigned NumVecs,
unsigned Opc) {
SDLoc dl(N);
EVT VT = N->getValueType(0);
bool Narrow = VT.getSizeInBits() == 64;
SmallVector<SDValue, 4> Regs(N->op_begin() + 1, N->op_begin() + 1 + NumVecs);
if (Narrow)
std::transform(Regs.begin(), Regs.end(), Regs.begin(),
WidenVector(*CurDAG));
SDValue RegSeq = createQTuple(Regs);
std::vector<EVT> ResTys;
ResTys.push_back(MVT::i64); ResTys.push_back(RegSeq->getValueType(0));
ResTys.push_back(MVT::Other);
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 1))->getZExtValue();
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(CurDAG->getTargetConstant(LaneNo, MVT::i64)); Ops.push_back(N->getOperand(NumVecs + 2)); Ops.push_back(N->getOperand(NumVecs + 3)); Ops.push_back(N->getOperand(0));
SDNode *Ld = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
ReplaceUses(SDValue(N, NumVecs), SDValue(Ld, 0));
SDValue SuperReg = SDValue(Ld, 1);
if (NumVecs == 1) {
ReplaceUses(SDValue(N, 0),
Narrow ? NarrowVector(SuperReg, *CurDAG) : SuperReg);
} else {
EVT WideVT = RegSeq.getOperand(1)->getValueType(0);
static unsigned QSubs[] = { AArch64::qsub0, AArch64::qsub1, AArch64::qsub2,
AArch64::qsub3 };
for (unsigned i = 0; i < NumVecs; ++i) {
SDValue NV = CurDAG->getTargetExtractSubreg(QSubs[i], dl, WideVT,
SuperReg);
if (Narrow)
NV = NarrowVector(NV, *CurDAG);
ReplaceUses(SDValue(N, i), NV);
}
}
ReplaceUses(SDValue(N, NumVecs + 1), SDValue(Ld, 2));
return Ld;
}
SDNode *AArch64DAGToDAGISel::SelectStoreLane(SDNode *N, unsigned NumVecs,
unsigned Opc) {
SDLoc dl(N);
EVT VT = N->getOperand(2)->getValueType(0);
bool Narrow = VT.getSizeInBits() == 64;
SmallVector<SDValue, 4> Regs(N->op_begin() + 2, N->op_begin() + 2 + NumVecs);
if (Narrow)
std::transform(Regs.begin(), Regs.end(), Regs.begin(),
WidenVector(*CurDAG));
SDValue RegSeq = createQTuple(Regs);
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 2))->getZExtValue();
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(CurDAG->getTargetConstant(LaneNo, MVT::i64));
Ops.push_back(N->getOperand(NumVecs + 3));
Ops.push_back(N->getOperand(0));
SDNode *St = CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops);
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
cast<MachineSDNode>(St)->setMemRefs(MemOp, MemOp + 1);
return St;
}
SDNode *AArch64DAGToDAGISel::SelectPostStoreLane(SDNode *N, unsigned NumVecs,
unsigned Opc) {
SDLoc dl(N);
EVT VT = N->getOperand(2)->getValueType(0);
bool Narrow = VT.getSizeInBits() == 64;
SmallVector<SDValue, 4> Regs(N->op_begin() + 1, N->op_begin() + 1 + NumVecs);
if (Narrow)
std::transform(Regs.begin(), Regs.end(), Regs.begin(),
WidenVector(*CurDAG));
SDValue RegSeq = createQTuple(Regs);
SmallVector<EVT, 2> ResTys;
ResTys.push_back(MVT::i64); ResTys.push_back(MVT::Other);
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 1))->getZExtValue();
SmallVector<SDValue, 6> Ops;
Ops.push_back(RegSeq);
Ops.push_back(CurDAG->getTargetConstant(LaneNo, MVT::i64));
Ops.push_back(N->getOperand(NumVecs + 2)); Ops.push_back(N->getOperand(NumVecs + 3)); Ops.push_back(N->getOperand(0));
SDNode *St = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
cast<MachineSDNode>(St)->setMemRefs(MemOp, MemOp + 1);
return St;
}
static bool isBitfieldExtractOpFromAnd(SelectionDAG *CurDAG, SDNode *N,
unsigned &Opc, SDValue &Opd0,
unsigned &LSB, unsigned &MSB,
unsigned NumberOfIgnoredLowBits,
bool BiggerPattern) {
assert(N->getOpcode() == ISD::AND &&
"N must be a AND operation to call this function");
EVT VT = N->getValueType(0);
assert((VT == MVT::i32 || VT == MVT::i64) &&
"Type checking must have been done before calling this function");
uint64_t And_imm = 0;
if (!isOpcWithIntImmediate(N, ISD::AND, And_imm))
return false;
const SDNode *Op0 = N->getOperand(0).getNode();
And_imm |= (1 << NumberOfIgnoredLowBits) - 1;
if (And_imm & (And_imm + 1))
return false;
bool ClampMSB = false;
uint64_t Srl_imm = 0;
if (VT == MVT::i64 && Op0->getOpcode() == ISD::ANY_EXTEND &&
isOpcWithIntImmediate(Op0->getOperand(0).getNode(), ISD::SRL, Srl_imm)) {
Opd0 = Widen(CurDAG, Op0->getOperand(0).getOperand(0));
ClampMSB = true;
} else if (VT == MVT::i32 && Op0->getOpcode() == ISD::TRUNCATE &&
isOpcWithIntImmediate(Op0->getOperand(0).getNode(), ISD::SRL,
Srl_imm)) {
Opd0 = Op0->getOperand(0).getOperand(0);
VT = Opd0->getValueType(0);
} else if (isOpcWithIntImmediate(Op0, ISD::SRL, Srl_imm)) {
Opd0 = Op0->getOperand(0);
} else if (BiggerPattern) {
Opd0 = N->getOperand(0);
} else
return false;
if (!BiggerPattern && (Srl_imm <= 0 || Srl_imm >= VT.getSizeInBits()))
return false;
LSB = Srl_imm;
MSB = Srl_imm + (VT == MVT::i32 ? CountTrailingOnes_32(And_imm)
: CountTrailingOnes_64(And_imm)) -
1;
if (ClampMSB)
MSB = MSB > 31 ? 31 : MSB;
Opc = VT == MVT::i32 ? AArch64::UBFMWri : AArch64::UBFMXri;
return true;
}
static bool isSeveralBitsExtractOpFromShr(SDNode *N, unsigned &Opc,
SDValue &Opd0, unsigned &LSB,
unsigned &MSB) {
if (N->getOpcode() != ISD::SRL)
return false;
uint64_t And_mask = 0;
if (!isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::AND, And_mask))
return false;
Opd0 = N->getOperand(0).getOperand(0);
uint64_t Srl_imm = 0;
if (!isIntImmediate(N->getOperand(1), Srl_imm))
return false;
unsigned BitWide = 64 - CountLeadingOnes_64(~(And_mask >> Srl_imm));
if (BitWide && isMask_64(And_mask >> Srl_imm)) {
if (N->getValueType(0) == MVT::i32)
Opc = AArch64::UBFMWri;
else
Opc = AArch64::UBFMXri;
LSB = Srl_imm;
MSB = BitWide + Srl_imm - 1;
return true;
}
return false;
}
static bool isBitfieldExtractOpFromShr(SDNode *N, unsigned &Opc, SDValue &Opd0,
unsigned &LSB, unsigned &MSB,
bool BiggerPattern) {
assert((N->getOpcode() == ISD::SRA || N->getOpcode() == ISD::SRL) &&
"N must be a SHR/SRA operation to call this function");
EVT VT = N->getValueType(0);
assert((VT == MVT::i32 || VT == MVT::i64) &&
"Type checking must have been done before calling this function");
if (isSeveralBitsExtractOpFromShr(N, Opc, Opd0, LSB, MSB))
return true;
uint64_t Shl_imm = 0;
uint64_t Trunc_bits = 0;
if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::SHL, Shl_imm)) {
Opd0 = N->getOperand(0).getOperand(0);
} else if (VT == MVT::i32 && N->getOpcode() == ISD::SRL &&
N->getOperand(0).getNode()->getOpcode() == ISD::TRUNCATE) {
Opd0 = N->getOperand(0).getOperand(0);
Trunc_bits = Opd0->getValueType(0).getSizeInBits() - VT.getSizeInBits();
VT = Opd0->getValueType(0);
assert(VT == MVT::i64 && "the promoted type should be i64");
} else if (BiggerPattern) {
Opd0 = N->getOperand(0);
} else
return false;
if (Shl_imm >= VT.getSizeInBits())
return false;
uint64_t Srl_imm = 0;
if (!isIntImmediate(N->getOperand(1), Srl_imm))
return false;
assert(Srl_imm > 0 && Srl_imm < VT.getSizeInBits() &&
"bad amount in shift node!");
unsigned Width = VT.getSizeInBits() - Trunc_bits - Srl_imm - 1;
int sLSB = Srl_imm - Shl_imm;
if (sLSB < 0)
return false;
LSB = sLSB;
MSB = LSB + Width;
if (VT == MVT::i32)
Opc = N->getOpcode() == ISD::SRA ? AArch64::SBFMWri : AArch64::UBFMWri;
else
Opc = N->getOpcode() == ISD::SRA ? AArch64::SBFMXri : AArch64::UBFMXri;
return true;
}
static bool isBitfieldExtractOp(SelectionDAG *CurDAG, SDNode *N, unsigned &Opc,
SDValue &Opd0, unsigned &LSB, unsigned &MSB,
unsigned NumberOfIgnoredLowBits = 0,
bool BiggerPattern = false) {
if (N->getValueType(0) != MVT::i32 && N->getValueType(0) != MVT::i64)
return false;
switch (N->getOpcode()) {
default:
if (!N->isMachineOpcode())
return false;
break;
case ISD::AND:
return isBitfieldExtractOpFromAnd(CurDAG, N, Opc, Opd0, LSB, MSB,
NumberOfIgnoredLowBits, BiggerPattern);
case ISD::SRL:
case ISD::SRA:
return isBitfieldExtractOpFromShr(N, Opc, Opd0, LSB, MSB, BiggerPattern);
}
unsigned NOpc = N->getMachineOpcode();
switch (NOpc) {
default:
return false;
case AArch64::SBFMWri:
case AArch64::UBFMWri:
case AArch64::SBFMXri:
case AArch64::UBFMXri:
Opc = NOpc;
Opd0 = N->getOperand(0);
LSB = cast<ConstantSDNode>(N->getOperand(1).getNode())->getZExtValue();
MSB = cast<ConstantSDNode>(N->getOperand(2).getNode())->getZExtValue();
return true;
}
return false;
}
SDNode *AArch64DAGToDAGISel::SelectBitfieldExtractOp(SDNode *N) {
unsigned Opc, LSB, MSB;
SDValue Opd0;
if (!isBitfieldExtractOp(CurDAG, N, Opc, Opd0, LSB, MSB))
return nullptr;
EVT VT = N->getValueType(0);
if ((Opc == AArch64::SBFMXri || Opc == AArch64::UBFMXri) && VT == MVT::i32) {
SDValue Ops64[] = {Opd0, CurDAG->getTargetConstant(LSB, MVT::i64),
CurDAG->getTargetConstant(MSB, MVT::i64)};
SDNode *BFM = CurDAG->getMachineNode(Opc, SDLoc(N), MVT::i64, Ops64);
SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
MachineSDNode *Node =
CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, SDLoc(N), MVT::i32,
SDValue(BFM, 0), SubReg);
return Node;
}
SDValue Ops[] = {Opd0, CurDAG->getTargetConstant(LSB, VT),
CurDAG->getTargetConstant(MSB, VT)};
return CurDAG->SelectNodeTo(N, Opc, VT, Ops);
}
static bool isBitfieldDstMask(uint64_t DstMask, APInt BitsToBeInserted,
unsigned NumberOfIgnoredHighBits, EVT VT) {
assert((VT == MVT::i32 || VT == MVT::i64) &&
"i32 or i64 mask type expected!");
unsigned BitWidth = VT.getSizeInBits() - NumberOfIgnoredHighBits;
APInt SignificantDstMask = APInt(BitWidth, DstMask);
APInt SignificantBitsToBeInserted = BitsToBeInserted.zextOrTrunc(BitWidth);
return (SignificantDstMask & SignificantBitsToBeInserted) == 0 &&
(SignificantDstMask | SignificantBitsToBeInserted).isAllOnesValue();
}
static void getUsefulBits(SDValue Op, APInt &UsefulBits, unsigned Depth = 0);
static void getUsefulBitsFromAndWithImmediate(SDValue Op, APInt &UsefulBits,
unsigned Depth) {
uint64_t Imm =
cast<const ConstantSDNode>(Op.getOperand(1).getNode())->getZExtValue();
Imm = AArch64_AM::decodeLogicalImmediate(Imm, UsefulBits.getBitWidth());
UsefulBits &= APInt(UsefulBits.getBitWidth(), Imm);
getUsefulBits(Op, UsefulBits, Depth + 1);
}
static void getUsefulBitsFromBitfieldMoveOpd(SDValue Op, APInt &UsefulBits,
uint64_t Imm, uint64_t MSB,
unsigned Depth) {
APInt OpUsefulBits(UsefulBits);
OpUsefulBits = 1;
if (MSB >= Imm) {
OpUsefulBits = OpUsefulBits.shl(MSB - Imm + 1);
--OpUsefulBits;
getUsefulBits(Op, OpUsefulBits, Depth + 1);
OpUsefulBits = OpUsefulBits.shl(Imm);
} else {
OpUsefulBits = OpUsefulBits.shl(MSB + 1);
--OpUsefulBits;
OpUsefulBits = OpUsefulBits.shl(OpUsefulBits.getBitWidth() - Imm);
getUsefulBits(Op, OpUsefulBits, Depth + 1);
OpUsefulBits = OpUsefulBits.lshr(OpUsefulBits.getBitWidth() - Imm);
}
UsefulBits &= OpUsefulBits;
}
static void getUsefulBitsFromUBFM(SDValue Op, APInt &UsefulBits,
unsigned Depth) {
uint64_t Imm =
cast<const ConstantSDNode>(Op.getOperand(1).getNode())->getZExtValue();
uint64_t MSB =
cast<const ConstantSDNode>(Op.getOperand(2).getNode())->getZExtValue();
getUsefulBitsFromBitfieldMoveOpd(Op, UsefulBits, Imm, MSB, Depth);
}
static void getUsefulBitsFromOrWithShiftedReg(SDValue Op, APInt &UsefulBits,
unsigned Depth) {
uint64_t ShiftTypeAndValue =
cast<const ConstantSDNode>(Op.getOperand(2).getNode())->getZExtValue();
APInt Mask(UsefulBits);
Mask.clearAllBits();
Mask.flipAllBits();
if (AArch64_AM::getShiftType(ShiftTypeAndValue) == AArch64_AM::LSL) {
uint64_t ShiftAmt = AArch64_AM::getShiftValue(ShiftTypeAndValue);
Mask = Mask.shl(ShiftAmt);
getUsefulBits(Op, Mask, Depth + 1);
Mask = Mask.lshr(ShiftAmt);
} else if (AArch64_AM::getShiftType(ShiftTypeAndValue) == AArch64_AM::LSR) {
uint64_t ShiftAmt = AArch64_AM::getShiftValue(ShiftTypeAndValue);
Mask = Mask.lshr(ShiftAmt);
getUsefulBits(Op, Mask, Depth + 1);
Mask = Mask.shl(ShiftAmt);
} else
return;
UsefulBits &= Mask;
}
static void getUsefulBitsFromBFM(SDValue Op, SDValue Orig, APInt &UsefulBits,
unsigned Depth) {
uint64_t Imm =
cast<const ConstantSDNode>(Op.getOperand(2).getNode())->getZExtValue();
uint64_t MSB =
cast<const ConstantSDNode>(Op.getOperand(3).getNode())->getZExtValue();
if (Op.getOperand(1) == Orig)
return getUsefulBitsFromBitfieldMoveOpd(Op, UsefulBits, Imm, MSB, Depth);
APInt OpUsefulBits(UsefulBits);
OpUsefulBits = 1;
if (MSB >= Imm) {
OpUsefulBits = OpUsefulBits.shl(MSB - Imm + 1);
--OpUsefulBits;
UsefulBits &= ~OpUsefulBits;
getUsefulBits(Op, UsefulBits, Depth + 1);
} else {
OpUsefulBits = OpUsefulBits.shl(MSB + 1);
--OpUsefulBits;
UsefulBits = ~(OpUsefulBits.shl(OpUsefulBits.getBitWidth() - Imm));
getUsefulBits(Op, UsefulBits, Depth + 1);
}
}
static void getUsefulBitsForUse(SDNode *UserNode, APInt &UsefulBits,
SDValue Orig, unsigned Depth) {
if (!UserNode->isMachineOpcode())
return;
switch (UserNode->getMachineOpcode()) {
default:
return;
case AArch64::ANDSWri:
case AArch64::ANDSXri:
case AArch64::ANDWri:
case AArch64::ANDXri:
return getUsefulBitsFromAndWithImmediate(SDValue(UserNode, 0), UsefulBits,
Depth);
case AArch64::UBFMWri:
case AArch64::UBFMXri:
return getUsefulBitsFromUBFM(SDValue(UserNode, 0), UsefulBits, Depth);
case AArch64::ORRWrs:
case AArch64::ORRXrs:
if (UserNode->getOperand(1) != Orig)
return;
return getUsefulBitsFromOrWithShiftedReg(SDValue(UserNode, 0), UsefulBits,
Depth);
case AArch64::BFMWri:
case AArch64::BFMXri:
return getUsefulBitsFromBFM(SDValue(UserNode, 0), Orig, UsefulBits, Depth);
}
}
static void getUsefulBits(SDValue Op, APInt &UsefulBits, unsigned Depth) {
if (Depth >= 6)
return;
if (!Depth) {
unsigned Bitwidth = Op.getValueType().getScalarType().getSizeInBits();
UsefulBits = APInt(Bitwidth, 0);
UsefulBits.flipAllBits();
}
APInt UsersUsefulBits(UsefulBits.getBitWidth(), 0);
for (SDNode *Node : Op.getNode()->uses()) {
APInt UsefulBitsForUse = APInt(UsefulBits);
getUsefulBitsForUse(Node, UsefulBitsForUse, Op, Depth);
UsersUsefulBits |= UsefulBitsForUse;
}
UsefulBits &= UsersUsefulBits;
}
static SDValue getLeftShift(SelectionDAG *CurDAG, SDValue Op, int ShlAmount) {
if (ShlAmount == 0)
return Op;
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
unsigned UBFMOpc = BitWidth == 32 ? AArch64::UBFMWri : AArch64::UBFMXri;
SDNode *ShiftNode;
if (ShlAmount > 0) {
ShiftNode = CurDAG->getMachineNode(
UBFMOpc, SDLoc(Op), VT, Op,
CurDAG->getTargetConstant(BitWidth - ShlAmount, VT),
CurDAG->getTargetConstant(BitWidth - 1 - ShlAmount, VT));
} else {
assert(ShlAmount < 0 && "expected right shift");
int ShrAmount = -ShlAmount;
ShiftNode = CurDAG->getMachineNode(
UBFMOpc, SDLoc(Op), VT, Op, CurDAG->getTargetConstant(ShrAmount, VT),
CurDAG->getTargetConstant(BitWidth - 1, VT));
}
return SDValue(ShiftNode, 0);
}
static bool isBitfieldPositioningOp(SelectionDAG *CurDAG, SDValue Op,
SDValue &Src, int &ShiftAmount,
int &MaskWidth) {
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
(void)BitWidth;
assert(BitWidth == 32 || BitWidth == 64);
APInt KnownZero, KnownOne;
CurDAG->computeKnownBits(Op, KnownZero, KnownOne);
uint64_t NonZeroBits = (~KnownZero).getZExtValue();
uint64_t AndImm;
if (isOpcWithIntImmediate(Op.getNode(), ISD::AND, AndImm)) {
assert((~APInt(BitWidth, AndImm) & ~KnownZero) == 0);
Op = Op.getOperand(0);
}
uint64_t ShlImm;
if (!isOpcWithIntImmediate(Op.getNode(), ISD::SHL, ShlImm))
return false;
Op = Op.getOperand(0);
if (!isShiftedMask_64(NonZeroBits))
return false;
ShiftAmount = countTrailingZeros(NonZeroBits);
MaskWidth = CountTrailingOnes_64(NonZeroBits >> ShiftAmount);
Src = getLeftShift(CurDAG, Op, ShlImm - ShiftAmount);
return true;
}
static bool isBitfieldInsertOpFromOr(SDNode *N, unsigned &Opc, SDValue &Dst,
SDValue &Src, unsigned &ImmR,
unsigned &ImmS, SelectionDAG *CurDAG) {
assert(N->getOpcode() == ISD::OR && "Expect a OR operation");
EVT VT = N->getValueType(0);
if (VT == MVT::i32)
Opc = AArch64::BFMWri;
else if (VT == MVT::i64)
Opc = AArch64::BFMXri;
else
return false;
APInt UsefulBits;
getUsefulBits(SDValue(N, 0), UsefulBits);
unsigned NumberOfIgnoredLowBits = UsefulBits.countTrailingZeros();
unsigned NumberOfIgnoredHighBits = UsefulBits.countLeadingZeros();
SDValue OrOpd1Val = N->getOperand(1);
SDNode *OrOpd0 = N->getOperand(0).getNode();
SDNode *OrOpd1 = N->getOperand(1).getNode();
for (int i = 0; i < 2;
++i, std::swap(OrOpd0, OrOpd1), OrOpd1Val = N->getOperand(0)) {
unsigned BFXOpc;
int DstLSB, Width;
if (isBitfieldExtractOp(CurDAG, OrOpd0, BFXOpc, Src, ImmR, ImmS,
NumberOfIgnoredLowBits, true)) {
if ((BFXOpc != AArch64::UBFMXri && VT == MVT::i64) ||
(BFXOpc != AArch64::UBFMWri && VT == MVT::i32))
continue;
DstLSB = 0;
Width = ImmS - ImmR + 1;
if (Width <= 0)
continue;
} else if (isBitfieldPositioningOp(CurDAG, SDValue(OrOpd0, 0), Src,
DstLSB, Width)) {
ImmR = (VT.getSizeInBits() - DstLSB) % VT.getSizeInBits();
ImmS = Width - 1;
} else
continue;
EVT VT = OrOpd1->getValueType(0);
assert((VT == MVT::i32 || VT == MVT::i64) && "unexpected OR operand");
APInt KnownZero, KnownOne;
CurDAG->computeKnownBits(OrOpd1Val, KnownZero, KnownOne);
APInt BitsToBeInserted =
APInt::getBitsSet(KnownZero.getBitWidth(), DstLSB, DstLSB + Width);
if ((BitsToBeInserted & ~KnownZero) != 0)
continue;
uint64_t Imm;
if (isOpcWithIntImmediate(OrOpd1, ISD::AND, Imm) &&
isBitfieldDstMask(Imm, BitsToBeInserted, NumberOfIgnoredHighBits, VT))
Dst = OrOpd1->getOperand(0);
else
Dst = OrOpd1Val;
return true;
}
return false;
}
SDNode *AArch64DAGToDAGISel::SelectBitfieldInsertOp(SDNode *N) {
if (N->getOpcode() != ISD::OR)
return nullptr;
unsigned Opc;
unsigned LSB, MSB;
SDValue Opd0, Opd1;
if (!isBitfieldInsertOpFromOr(N, Opc, Opd0, Opd1, LSB, MSB, CurDAG))
return nullptr;
EVT VT = N->getValueType(0);
SDValue Ops[] = { Opd0,
Opd1,
CurDAG->getTargetConstant(LSB, VT),
CurDAG->getTargetConstant(MSB, VT) };
return CurDAG->SelectNodeTo(N, Opc, VT, Ops);
}
SDNode *AArch64DAGToDAGISel::SelectLIBM(SDNode *N) {
EVT VT = N->getValueType(0);
unsigned Variant;
unsigned Opc;
unsigned FRINTXOpcs[] = { AArch64::FRINTXSr, AArch64::FRINTXDr };
if (VT == MVT::f32) {
Variant = 0;
} else if (VT == MVT::f64) {
Variant = 1;
} else
return nullptr;
unsigned FRINTXOpc = FRINTXOpcs[Variant];
switch (N->getOpcode()) {
default:
return nullptr; case ISD::FCEIL: {
unsigned FRINTPOpcs[] = { AArch64::FRINTPSr, AArch64::FRINTPDr };
Opc = FRINTPOpcs[Variant];
break;
}
case ISD::FFLOOR: {
unsigned FRINTMOpcs[] = { AArch64::FRINTMSr, AArch64::FRINTMDr };
Opc = FRINTMOpcs[Variant];
break;
}
case ISD::FTRUNC: {
unsigned FRINTZOpcs[] = { AArch64::FRINTZSr, AArch64::FRINTZDr };
Opc = FRINTZOpcs[Variant];
break;
}
case ISD::FROUND: {
unsigned FRINTAOpcs[] = { AArch64::FRINTASr, AArch64::FRINTADr };
Opc = FRINTAOpcs[Variant];
break;
}
}
SDLoc dl(N);
SDValue In = N->getOperand(0);
SmallVector<SDValue, 2> Ops;
Ops.push_back(In);
if (!TM.Options.UnsafeFPMath) {
SDNode *FRINTX = CurDAG->getMachineNode(FRINTXOpc, dl, VT, MVT::Glue, In);
Ops.push_back(SDValue(FRINTX, 1));
}
return CurDAG->getMachineNode(Opc, dl, VT, Ops);
}
bool
AArch64DAGToDAGISel::SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos,
unsigned RegWidth) {
APFloat FVal(0.0);
if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(N))
FVal = CN->getValueAPF();
else if (LoadSDNode *LN = dyn_cast<LoadSDNode>(N)) {
if (LN->getOperand(1).getOpcode() != AArch64ISD::ADDlow ||
!isa<ConstantPoolSDNode>(LN->getOperand(1)->getOperand(1)))
return false;
ConstantPoolSDNode *CN =
dyn_cast<ConstantPoolSDNode>(LN->getOperand(1)->getOperand(1));
FVal = cast<ConstantFP>(CN->getConstVal())->getValueAPF();
} else
return false;
bool IsExact;
APSInt IntVal(65, true);
FVal.convertToInteger(IntVal, APFloat::rmTowardZero, &IsExact);
if (!IsExact || !IntVal.isPowerOf2()) return false;
unsigned FBits = IntVal.logBase2();
if (FBits == 0 || FBits > RegWidth) return false;
FixedPos = CurDAG->getTargetConstant(FBits, MVT::i32);
return true;
}
SDNode *AArch64DAGToDAGISel::Select(SDNode *Node) {
DEBUG(errs() << "Selecting: ");
DEBUG(Node->dump(CurDAG));
DEBUG(errs() << "\n");
if (Node->isMachineOpcode()) {
DEBUG(errs() << "== "; Node->dump(CurDAG); errs() << "\n");
Node->setNodeId(-1);
return nullptr;
}
SDNode *ResNode = nullptr;
EVT VT = Node->getValueType(0);
switch (Node->getOpcode()) {
default:
break;
case ISD::ADD:
if (SDNode *I = SelectMLAV64LaneV128(Node))
return I;
break;
case ISD::LOAD: {
bool Done = false;
SDNode *I = SelectIndexedLoad(Node, Done);
if (Done)
return I;
break;
}
case ISD::SRL:
case ISD::AND:
case ISD::SRA:
if (SDNode *I = SelectBitfieldExtractOp(Node))
return I;
break;
case ISD::OR:
if (SDNode *I = SelectBitfieldInsertOp(Node))
return I;
break;
case ISD::EXTRACT_VECTOR_ELT: {
ConstantSDNode *LaneNode = cast<ConstantSDNode>(Node->getOperand(1));
if (LaneNode->getZExtValue() != 0)
break;
if (VT != Node->getOperand(0).getValueType().getVectorElementType())
break;
unsigned SubReg;
switch (Node->getOperand(0)
.getValueType()
.getVectorElementType()
.getSizeInBits()) {
default:
llvm_unreachable("Unexpected vector element type!");
case 64:
SubReg = AArch64::dsub;
break;
case 32:
SubReg = AArch64::ssub;
break;
case 16:
SubReg = AArch64::hsub;
break;
case 8:
llvm_unreachable("unexpected zext-requiring extract element!");
}
SDValue Extract = CurDAG->getTargetExtractSubreg(SubReg, SDLoc(Node), VT,
Node->getOperand(0));
DEBUG(dbgs() << "ISEL: Custom selection!\n=> ");
DEBUG(Extract->dumpr(CurDAG));
DEBUG(dbgs() << "\n");
return Extract.getNode();
}
case ISD::Constant: {
ConstantSDNode *ConstNode = cast<ConstantSDNode>(Node);
if (ConstNode->isNullValue()) {
if (VT == MVT::i32)
return CurDAG->getCopyFromReg(CurDAG->getEntryNode(), SDLoc(Node),
AArch64::WZR, MVT::i32).getNode();
else if (VT == MVT::i64)
return CurDAG->getCopyFromReg(CurDAG->getEntryNode(), SDLoc(Node),
AArch64::XZR, MVT::i64).getNode();
}
break;
}
case ISD::FrameIndex: {
int FI = cast<FrameIndexSDNode>(Node)->getIndex();
unsigned Shifter = AArch64_AM::getShifterImm(AArch64_AM::LSL, 0);
const TargetLowering *TLI = getTargetLowering();
SDValue TFI = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy());
SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, MVT::i32),
CurDAG->getTargetConstant(Shifter, MVT::i32) };
return CurDAG->SelectNodeTo(Node, AArch64::ADDXri, MVT::i64, Ops);
}
case ISD::INTRINSIC_W_CHAIN: {
unsigned IntNo = cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue();
switch (IntNo) {
default:
break;
case Intrinsic::aarch64_ldaxp:
case Intrinsic::aarch64_ldxp: {
unsigned Op =
IntNo == Intrinsic::aarch64_ldaxp ? AArch64::LDAXPX : AArch64::LDXPX;
SDValue MemAddr = Node->getOperand(2);
SDLoc DL(Node);
SDValue Chain = Node->getOperand(0);
SDNode *Ld = CurDAG->getMachineNode(Op, DL, MVT::i64, MVT::i64,
MVT::Other, MemAddr, Chain);
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(Node)->getMemOperand();
cast<MachineSDNode>(Ld)->setMemRefs(MemOp, MemOp + 1);
return Ld;
}
case Intrinsic::aarch64_stlxp:
case Intrinsic::aarch64_stxp: {
unsigned Op =
IntNo == Intrinsic::aarch64_stlxp ? AArch64::STLXPX : AArch64::STXPX;
SDLoc DL(Node);
SDValue Chain = Node->getOperand(0);
SDValue ValLo = Node->getOperand(2);
SDValue ValHi = Node->getOperand(3);
SDValue MemAddr = Node->getOperand(4);
SmallVector<SDValue, 7> Ops;
Ops.push_back(ValLo);
Ops.push_back(ValHi);
Ops.push_back(MemAddr);
Ops.push_back(Chain);
SDNode *St = CurDAG->getMachineNode(Op, DL, MVT::i32, MVT::Other, Ops);
MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
MemOp[0] = cast<MemIntrinsicSDNode>(Node)->getMemOperand();
cast<MachineSDNode>(St)->setMemRefs(MemOp, MemOp + 1);
return St;
}
case Intrinsic::aarch64_neon_ld1x2:
if (VT == MVT::v8i8)
return SelectLoad(Node, 2, AArch64::LD1Twov8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 2, AArch64::LD1Twov16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 2, AArch64::LD1Twov4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 2, AArch64::LD1Twov8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 2, AArch64::LD1Twov2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 2, AArch64::LD1Twov4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 2, AArch64::LD1Twov1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 2, AArch64::LD1Twov2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld1x3:
if (VT == MVT::v8i8)
return SelectLoad(Node, 3, AArch64::LD1Threev8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 3, AArch64::LD1Threev16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 3, AArch64::LD1Threev4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 3, AArch64::LD1Threev8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 3, AArch64::LD1Threev2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 3, AArch64::LD1Threev4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 3, AArch64::LD1Threev1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 3, AArch64::LD1Threev2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld1x4:
if (VT == MVT::v8i8)
return SelectLoad(Node, 4, AArch64::LD1Fourv8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 4, AArch64::LD1Fourv16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 4, AArch64::LD1Fourv4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 4, AArch64::LD1Fourv8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 4, AArch64::LD1Fourv2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 4, AArch64::LD1Fourv4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 4, AArch64::LD1Fourv1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 4, AArch64::LD1Fourv2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld2:
if (VT == MVT::v8i8)
return SelectLoad(Node, 2, AArch64::LD2Twov8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 2, AArch64::LD2Twov16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 2, AArch64::LD2Twov4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 2, AArch64::LD2Twov8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 2, AArch64::LD2Twov2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 2, AArch64::LD2Twov4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 2, AArch64::LD1Twov1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 2, AArch64::LD2Twov2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld3:
if (VT == MVT::v8i8)
return SelectLoad(Node, 3, AArch64::LD3Threev8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 3, AArch64::LD3Threev16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 3, AArch64::LD3Threev4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 3, AArch64::LD3Threev8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 3, AArch64::LD3Threev2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 3, AArch64::LD3Threev4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 3, AArch64::LD1Threev1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 3, AArch64::LD3Threev2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld4:
if (VT == MVT::v8i8)
return SelectLoad(Node, 4, AArch64::LD4Fourv8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 4, AArch64::LD4Fourv16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 4, AArch64::LD4Fourv4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 4, AArch64::LD4Fourv8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 4, AArch64::LD4Fourv2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 4, AArch64::LD4Fourv4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 4, AArch64::LD1Fourv1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 4, AArch64::LD4Fourv2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld2r:
if (VT == MVT::v8i8)
return SelectLoad(Node, 2, AArch64::LD2Rv8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 2, AArch64::LD2Rv16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 2, AArch64::LD2Rv4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 2, AArch64::LD2Rv8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 2, AArch64::LD2Rv2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 2, AArch64::LD2Rv4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 2, AArch64::LD2Rv1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 2, AArch64::LD2Rv2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld3r:
if (VT == MVT::v8i8)
return SelectLoad(Node, 3, AArch64::LD3Rv8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 3, AArch64::LD3Rv16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 3, AArch64::LD3Rv4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 3, AArch64::LD3Rv8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 3, AArch64::LD3Rv2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 3, AArch64::LD3Rv4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 3, AArch64::LD3Rv1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 3, AArch64::LD3Rv2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld4r:
if (VT == MVT::v8i8)
return SelectLoad(Node, 4, AArch64::LD4Rv8b, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectLoad(Node, 4, AArch64::LD4Rv16b, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectLoad(Node, 4, AArch64::LD4Rv4h, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectLoad(Node, 4, AArch64::LD4Rv8h, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectLoad(Node, 4, AArch64::LD4Rv2s, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectLoad(Node, 4, AArch64::LD4Rv4s, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectLoad(Node, 4, AArch64::LD4Rv1d, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectLoad(Node, 4, AArch64::LD4Rv2d, AArch64::qsub0);
break;
case Intrinsic::aarch64_neon_ld2lane:
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectLoadLane(Node, 2, AArch64::LD2i8);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectLoadLane(Node, 2, AArch64::LD2i16);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectLoadLane(Node, 2, AArch64::LD2i32);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectLoadLane(Node, 2, AArch64::LD2i64);
break;
case Intrinsic::aarch64_neon_ld3lane:
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectLoadLane(Node, 3, AArch64::LD3i8);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectLoadLane(Node, 3, AArch64::LD3i16);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectLoadLane(Node, 3, AArch64::LD3i32);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectLoadLane(Node, 3, AArch64::LD3i64);
break;
case Intrinsic::aarch64_neon_ld4lane:
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectLoadLane(Node, 4, AArch64::LD4i8);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectLoadLane(Node, 4, AArch64::LD4i16);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectLoadLane(Node, 4, AArch64::LD4i32);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectLoadLane(Node, 4, AArch64::LD4i64);
break;
}
} break;
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo = cast<ConstantSDNode>(Node->getOperand(0))->getZExtValue();
switch (IntNo) {
default:
break;
case Intrinsic::aarch64_neon_tbl2:
return SelectTable(Node, 2, VT == MVT::v8i8 ? AArch64::TBLv8i8Two
: AArch64::TBLv16i8Two,
false);
case Intrinsic::aarch64_neon_tbl3:
return SelectTable(Node, 3, VT == MVT::v8i8 ? AArch64::TBLv8i8Three
: AArch64::TBLv16i8Three,
false);
case Intrinsic::aarch64_neon_tbl4:
return SelectTable(Node, 4, VT == MVT::v8i8 ? AArch64::TBLv8i8Four
: AArch64::TBLv16i8Four,
false);
case Intrinsic::aarch64_neon_tbx2:
return SelectTable(Node, 2, VT == MVT::v8i8 ? AArch64::TBXv8i8Two
: AArch64::TBXv16i8Two,
true);
case Intrinsic::aarch64_neon_tbx3:
return SelectTable(Node, 3, VT == MVT::v8i8 ? AArch64::TBXv8i8Three
: AArch64::TBXv16i8Three,
true);
case Intrinsic::aarch64_neon_tbx4:
return SelectTable(Node, 4, VT == MVT::v8i8 ? AArch64::TBXv8i8Four
: AArch64::TBXv16i8Four,
true);
case Intrinsic::aarch64_neon_smull:
case Intrinsic::aarch64_neon_umull:
if (SDNode *N = SelectMULLV64LaneV128(IntNo, Node))
return N;
break;
}
break;
}
case ISD::INTRINSIC_VOID: {
unsigned IntNo = cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue();
if (Node->getNumOperands() >= 3)
VT = Node->getOperand(2)->getValueType(0);
switch (IntNo) {
default:
break;
case Intrinsic::aarch64_neon_st1x2: {
if (VT == MVT::v8i8)
return SelectStore(Node, 2, AArch64::ST1Twov8b);
else if (VT == MVT::v16i8)
return SelectStore(Node, 2, AArch64::ST1Twov16b);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectStore(Node, 2, AArch64::ST1Twov4h);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectStore(Node, 2, AArch64::ST1Twov8h);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectStore(Node, 2, AArch64::ST1Twov2s);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectStore(Node, 2, AArch64::ST1Twov4s);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectStore(Node, 2, AArch64::ST1Twov2d);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectStore(Node, 2, AArch64::ST1Twov1d);
break;
}
case Intrinsic::aarch64_neon_st1x3: {
if (VT == MVT::v8i8)
return SelectStore(Node, 3, AArch64::ST1Threev8b);
else if (VT == MVT::v16i8)
return SelectStore(Node, 3, AArch64::ST1Threev16b);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectStore(Node, 3, AArch64::ST1Threev4h);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectStore(Node, 3, AArch64::ST1Threev8h);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectStore(Node, 3, AArch64::ST1Threev2s);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectStore(Node, 3, AArch64::ST1Threev4s);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectStore(Node, 3, AArch64::ST1Threev2d);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectStore(Node, 3, AArch64::ST1Threev1d);
break;
}
case Intrinsic::aarch64_neon_st1x4: {
if (VT == MVT::v8i8)
return SelectStore(Node, 4, AArch64::ST1Fourv8b);
else if (VT == MVT::v16i8)
return SelectStore(Node, 4, AArch64::ST1Fourv16b);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectStore(Node, 4, AArch64::ST1Fourv4h);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectStore(Node, 4, AArch64::ST1Fourv8h);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectStore(Node, 4, AArch64::ST1Fourv2s);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectStore(Node, 4, AArch64::ST1Fourv4s);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectStore(Node, 4, AArch64::ST1Fourv2d);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectStore(Node, 4, AArch64::ST1Fourv1d);
break;
}
case Intrinsic::aarch64_neon_st2: {
if (VT == MVT::v8i8)
return SelectStore(Node, 2, AArch64::ST2Twov8b);
else if (VT == MVT::v16i8)
return SelectStore(Node, 2, AArch64::ST2Twov16b);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectStore(Node, 2, AArch64::ST2Twov4h);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectStore(Node, 2, AArch64::ST2Twov8h);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectStore(Node, 2, AArch64::ST2Twov2s);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectStore(Node, 2, AArch64::ST2Twov4s);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectStore(Node, 2, AArch64::ST2Twov2d);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectStore(Node, 2, AArch64::ST1Twov1d);
break;
}
case Intrinsic::aarch64_neon_st3: {
if (VT == MVT::v8i8)
return SelectStore(Node, 3, AArch64::ST3Threev8b);
else if (VT == MVT::v16i8)
return SelectStore(Node, 3, AArch64::ST3Threev16b);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectStore(Node, 3, AArch64::ST3Threev4h);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectStore(Node, 3, AArch64::ST3Threev8h);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectStore(Node, 3, AArch64::ST3Threev2s);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectStore(Node, 3, AArch64::ST3Threev4s);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectStore(Node, 3, AArch64::ST3Threev2d);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectStore(Node, 3, AArch64::ST1Threev1d);
break;
}
case Intrinsic::aarch64_neon_st4: {
if (VT == MVT::v8i8)
return SelectStore(Node, 4, AArch64::ST4Fourv8b);
else if (VT == MVT::v16i8)
return SelectStore(Node, 4, AArch64::ST4Fourv16b);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectStore(Node, 4, AArch64::ST4Fourv4h);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectStore(Node, 4, AArch64::ST4Fourv8h);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectStore(Node, 4, AArch64::ST4Fourv2s);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectStore(Node, 4, AArch64::ST4Fourv4s);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectStore(Node, 4, AArch64::ST4Fourv2d);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectStore(Node, 4, AArch64::ST1Fourv1d);
break;
}
case Intrinsic::aarch64_neon_st2lane: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectStoreLane(Node, 2, AArch64::ST2i8);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectStoreLane(Node, 2, AArch64::ST2i16);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectStoreLane(Node, 2, AArch64::ST2i32);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectStoreLane(Node, 2, AArch64::ST2i64);
break;
}
case Intrinsic::aarch64_neon_st3lane: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectStoreLane(Node, 3, AArch64::ST3i8);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectStoreLane(Node, 3, AArch64::ST3i16);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectStoreLane(Node, 3, AArch64::ST3i32);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectStoreLane(Node, 3, AArch64::ST3i64);
break;
}
case Intrinsic::aarch64_neon_st4lane: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectStoreLane(Node, 4, AArch64::ST4i8);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectStoreLane(Node, 4, AArch64::ST4i16);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectStoreLane(Node, 4, AArch64::ST4i32);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectStoreLane(Node, 4, AArch64::ST4i64);
break;
}
}
}
case AArch64ISD::LD2post: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 2, AArch64::LD2Twov8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 2, AArch64::LD2Twov16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 2, AArch64::LD2Twov4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 2, AArch64::LD2Twov8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 2, AArch64::LD2Twov2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 2, AArch64::LD2Twov4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 2, AArch64::LD1Twov1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 2, AArch64::LD2Twov2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD3post: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 3, AArch64::LD3Threev8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 3, AArch64::LD3Threev16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 3, AArch64::LD3Threev4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 3, AArch64::LD3Threev8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 3, AArch64::LD3Threev2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 3, AArch64::LD3Threev4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 3, AArch64::LD1Threev1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 3, AArch64::LD3Threev2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD4post: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 4, AArch64::LD4Fourv2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD1x2post: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 2, AArch64::LD1Twov8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 2, AArch64::LD1Twov16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 2, AArch64::LD1Twov4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 2, AArch64::LD1Twov8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 2, AArch64::LD1Twov2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 2, AArch64::LD1Twov4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 2, AArch64::LD1Twov1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 2, AArch64::LD1Twov2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD1x3post: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 3, AArch64::LD1Threev8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 3, AArch64::LD1Threev16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 3, AArch64::LD1Threev4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 3, AArch64::LD1Threev8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 3, AArch64::LD1Threev2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 3, AArch64::LD1Threev4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 3, AArch64::LD1Threev1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 3, AArch64::LD1Threev2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD1x4post: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 4, AArch64::LD1Fourv2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD1DUPpost: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 1, AArch64::LD1Rv8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 1, AArch64::LD1Rv16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 1, AArch64::LD1Rv4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 1, AArch64::LD1Rv8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 1, AArch64::LD1Rv2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 1, AArch64::LD1Rv4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 1, AArch64::LD1Rv1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 1, AArch64::LD1Rv2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD2DUPpost: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 2, AArch64::LD2Rv8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 2, AArch64::LD2Rv16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 2, AArch64::LD2Rv4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 2, AArch64::LD2Rv8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 2, AArch64::LD2Rv2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 2, AArch64::LD2Rv4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 2, AArch64::LD2Rv1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 2, AArch64::LD2Rv2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD3DUPpost: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 3, AArch64::LD3Rv8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 3, AArch64::LD3Rv16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 3, AArch64::LD3Rv4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 3, AArch64::LD3Rv8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 3, AArch64::LD3Rv2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 3, AArch64::LD3Rv4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 3, AArch64::LD3Rv1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 3, AArch64::LD3Rv2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD4DUPpost: {
if (VT == MVT::v8i8)
return SelectPostLoad(Node, 4, AArch64::LD4Rv8b_POST, AArch64::dsub0);
else if (VT == MVT::v16i8)
return SelectPostLoad(Node, 4, AArch64::LD4Rv16b_POST, AArch64::qsub0);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostLoad(Node, 4, AArch64::LD4Rv4h_POST, AArch64::dsub0);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostLoad(Node, 4, AArch64::LD4Rv8h_POST, AArch64::qsub0);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostLoad(Node, 4, AArch64::LD4Rv2s_POST, AArch64::dsub0);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostLoad(Node, 4, AArch64::LD4Rv4s_POST, AArch64::qsub0);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostLoad(Node, 4, AArch64::LD4Rv1d_POST, AArch64::dsub0);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostLoad(Node, 4, AArch64::LD4Rv2d_POST, AArch64::qsub0);
break;
}
case AArch64ISD::LD1LANEpost: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostLoadLane(Node, 1, AArch64::LD1i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostLoadLane(Node, 1, AArch64::LD1i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostLoadLane(Node, 1, AArch64::LD1i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostLoadLane(Node, 1, AArch64::LD1i64_POST);
break;
}
case AArch64ISD::LD2LANEpost: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostLoadLane(Node, 2, AArch64::LD2i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostLoadLane(Node, 2, AArch64::LD2i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostLoadLane(Node, 2, AArch64::LD2i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostLoadLane(Node, 2, AArch64::LD2i64_POST);
break;
}
case AArch64ISD::LD3LANEpost: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostLoadLane(Node, 3, AArch64::LD3i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostLoadLane(Node, 3, AArch64::LD3i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostLoadLane(Node, 3, AArch64::LD3i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostLoadLane(Node, 3, AArch64::LD3i64_POST);
break;
}
case AArch64ISD::LD4LANEpost: {
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostLoadLane(Node, 4, AArch64::LD4i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostLoadLane(Node, 4, AArch64::LD4i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostLoadLane(Node, 4, AArch64::LD4i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostLoadLane(Node, 4, AArch64::LD4i64_POST);
break;
}
case AArch64ISD::ST2post: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v8i8)
return SelectPostStore(Node, 2, AArch64::ST2Twov8b_POST);
else if (VT == MVT::v16i8)
return SelectPostStore(Node, 2, AArch64::ST2Twov16b_POST);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostStore(Node, 2, AArch64::ST2Twov4h_POST);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostStore(Node, 2, AArch64::ST2Twov8h_POST);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostStore(Node, 2, AArch64::ST2Twov2s_POST);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostStore(Node, 2, AArch64::ST2Twov4s_POST);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostStore(Node, 2, AArch64::ST2Twov2d_POST);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostStore(Node, 2, AArch64::ST1Twov1d_POST);
break;
}
case AArch64ISD::ST3post: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v8i8)
return SelectPostStore(Node, 3, AArch64::ST3Threev8b_POST);
else if (VT == MVT::v16i8)
return SelectPostStore(Node, 3, AArch64::ST3Threev16b_POST);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostStore(Node, 3, AArch64::ST3Threev4h_POST);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostStore(Node, 3, AArch64::ST3Threev8h_POST);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostStore(Node, 3, AArch64::ST3Threev2s_POST);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostStore(Node, 3, AArch64::ST3Threev4s_POST);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostStore(Node, 3, AArch64::ST3Threev2d_POST);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostStore(Node, 3, AArch64::ST1Threev1d_POST);
break;
}
case AArch64ISD::ST4post: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v8i8)
return SelectPostStore(Node, 4, AArch64::ST4Fourv8b_POST);
else if (VT == MVT::v16i8)
return SelectPostStore(Node, 4, AArch64::ST4Fourv16b_POST);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostStore(Node, 4, AArch64::ST4Fourv4h_POST);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostStore(Node, 4, AArch64::ST4Fourv8h_POST);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostStore(Node, 4, AArch64::ST4Fourv2s_POST);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostStore(Node, 4, AArch64::ST4Fourv4s_POST);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostStore(Node, 4, AArch64::ST4Fourv2d_POST);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostStore(Node, 4, AArch64::ST1Fourv1d_POST);
break;
}
case AArch64ISD::ST1x2post: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v8i8)
return SelectPostStore(Node, 2, AArch64::ST1Twov8b_POST);
else if (VT == MVT::v16i8)
return SelectPostStore(Node, 2, AArch64::ST1Twov16b_POST);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostStore(Node, 2, AArch64::ST1Twov4h_POST);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostStore(Node, 2, AArch64::ST1Twov8h_POST);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostStore(Node, 2, AArch64::ST1Twov2s_POST);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostStore(Node, 2, AArch64::ST1Twov4s_POST);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostStore(Node, 2, AArch64::ST1Twov1d_POST);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostStore(Node, 2, AArch64::ST1Twov2d_POST);
break;
}
case AArch64ISD::ST1x3post: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v8i8)
return SelectPostStore(Node, 3, AArch64::ST1Threev8b_POST);
else if (VT == MVT::v16i8)
return SelectPostStore(Node, 3, AArch64::ST1Threev16b_POST);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostStore(Node, 3, AArch64::ST1Threev4h_POST);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostStore(Node, 3, AArch64::ST1Threev8h_POST);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostStore(Node, 3, AArch64::ST1Threev2s_POST);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostStore(Node, 3, AArch64::ST1Threev4s_POST);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostStore(Node, 3, AArch64::ST1Threev1d_POST);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostStore(Node, 3, AArch64::ST1Threev2d_POST);
break;
}
case AArch64ISD::ST1x4post: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v8i8)
return SelectPostStore(Node, 4, AArch64::ST1Fourv8b_POST);
else if (VT == MVT::v16i8)
return SelectPostStore(Node, 4, AArch64::ST1Fourv16b_POST);
else if (VT == MVT::v4i16 || VT == MVT::v4f16)
return SelectPostStore(Node, 4, AArch64::ST1Fourv4h_POST);
else if (VT == MVT::v8i16 || VT == MVT::v8f16)
return SelectPostStore(Node, 4, AArch64::ST1Fourv8h_POST);
else if (VT == MVT::v2i32 || VT == MVT::v2f32)
return SelectPostStore(Node, 4, AArch64::ST1Fourv2s_POST);
else if (VT == MVT::v4i32 || VT == MVT::v4f32)
return SelectPostStore(Node, 4, AArch64::ST1Fourv4s_POST);
else if (VT == MVT::v1i64 || VT == MVT::v1f64)
return SelectPostStore(Node, 4, AArch64::ST1Fourv1d_POST);
else if (VT == MVT::v2i64 || VT == MVT::v2f64)
return SelectPostStore(Node, 4, AArch64::ST1Fourv2d_POST);
break;
}
case AArch64ISD::ST2LANEpost: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostStoreLane(Node, 2, AArch64::ST2i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostStoreLane(Node, 2, AArch64::ST2i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostStoreLane(Node, 2, AArch64::ST2i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostStoreLane(Node, 2, AArch64::ST2i64_POST);
break;
}
case AArch64ISD::ST3LANEpost: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostStoreLane(Node, 3, AArch64::ST3i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostStoreLane(Node, 3, AArch64::ST3i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostStoreLane(Node, 3, AArch64::ST3i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostStoreLane(Node, 3, AArch64::ST3i64_POST);
break;
}
case AArch64ISD::ST4LANEpost: {
VT = Node->getOperand(1).getValueType();
if (VT == MVT::v16i8 || VT == MVT::v8i8)
return SelectPostStoreLane(Node, 4, AArch64::ST4i8_POST);
else if (VT == MVT::v8i16 || VT == MVT::v4i16 || VT == MVT::v4f16 ||
VT == MVT::v8f16)
return SelectPostStoreLane(Node, 4, AArch64::ST4i16_POST);
else if (VT == MVT::v4i32 || VT == MVT::v2i32 || VT == MVT::v4f32 ||
VT == MVT::v2f32)
return SelectPostStoreLane(Node, 4, AArch64::ST4i32_POST);
else if (VT == MVT::v2i64 || VT == MVT::v1i64 || VT == MVT::v2f64 ||
VT == MVT::v1f64)
return SelectPostStoreLane(Node, 4, AArch64::ST4i64_POST);
break;
}
case ISD::FCEIL:
case ISD::FFLOOR:
case ISD::FTRUNC:
case ISD::FROUND:
if (SDNode *I = SelectLIBM(Node))
return I;
break;
}
ResNode = SelectCode(Node);
DEBUG(errs() << "=> ");
if (ResNode == nullptr || ResNode == Node)
DEBUG(Node->dump(CurDAG));
else
DEBUG(ResNode->dump(CurDAG));
DEBUG(errs() << "\n");
return ResNode;
}
FunctionPass *llvm::createAArch64ISelDag(AArch64TargetMachine &TM,
CodeGenOpt::Level OptLevel) {
return new AArch64DAGToDAGISel(TM, OptLevel);
}