}
bool runOnMachineFunction(MachineFunction &MF) override {
- ForCodeSize =
- MF.getFunction()->hasFnAttribute(Attribute::OptimizeForSize) ||
- MF.getFunction()->hasFnAttribute(Attribute::MinSize);
+ ForCodeSize = MF.getFunction()->optForSize();
Subtarget = &MF.getSubtarget<AArch64Subtarget>();
return SelectionDAGISel::runOnMachineFunction(MF);
}
SDNode *SelectBitfieldInsertOp(SDNode *N);
SDNode *SelectLIBM(SDNode *N);
+ SDNode *SelectFPConvertWithRound(SDNode *N);
+
+ SDNode *SelectReadRegister(SDNode *N);
+ SDNode *SelectWriteRegister(SDNode *N);
// Include the pieces autogenerated from the target description.
#include "AArch64GenDAGISel.inc"
}
bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos, unsigned Width);
+
+ SDNode *GenerateInexactFlagIfNeeded(const SDValue &In, unsigned InTyVariant,
+ SDLoc DL);
};
} // end anonymous namespace
return false;
unsigned ShVal = AArch64_AM::getShifterImm(AArch64_AM::LSL, ShiftAmt);
- Val = CurDAG->getTargetConstant(Immed, MVT::i32);
- Shift = CurDAG->getTargetConstant(ShVal, MVT::i32);
+ SDLoc dl(N);
+ Val = CurDAG->getTargetConstant(Immed, dl, MVT::i32);
+ Shift = CurDAG->getTargetConstant(ShVal, dl, MVT::i32);
return true;
}
return false;
Immed &= 0xFFFFFFULL;
- return SelectArithImmed(CurDAG->getConstant(Immed, MVT::i32), Val, Shift);
+ return SelectArithImmed(CurDAG->getConstant(Immed, SDLoc(N), MVT::i32), Val,
+ Shift);
}
/// getShiftTypeForNode - Translate a shift node to the corresponding
unsigned ShVal = AArch64_AM::getShifterImm(ShType, Val);
Reg = N.getOperand(0);
- Shift = CurDAG->getTargetConstant(ShVal, MVT::i32);
+ Shift = CurDAG->getTargetConstant(ShVal, SDLoc(N), MVT::i32);
return isWorthFolding(N);
}
/// is a lane in the upper half of a 128-bit vector. Recognize and select this
/// so that we don't emit unnecessary lane extracts.
SDNode *AArch64DAGToDAGISel::SelectMLAV64LaneV128(SDNode *N) {
+ SDLoc dl(N);
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
SDValue MLAOp1; // Will hold ordinary multiplicand for MLA.
return nullptr;
}
- SDValue LaneIdxVal = CurDAG->getTargetConstant(LaneIdx, MVT::i64);
+ SDValue LaneIdxVal = CurDAG->getTargetConstant(LaneIdx, dl, MVT::i64);
SDValue Ops[] = { Op0, MLAOp1, MLAOp2, LaneIdxVal };
break;
}
- return CurDAG->getMachineNode(MLAOpc, SDLoc(N), N->getValueType(0), Ops);
+ return CurDAG->getMachineNode(MLAOpc, dl, N->getValueType(0), Ops);
}
SDNode *AArch64DAGToDAGISel::SelectMULLV64LaneV128(unsigned IntNo, SDNode *N) {
+ SDLoc dl(N);
SDValue SMULLOp0;
SDValue SMULLOp1;
int LaneIdx;
LaneIdx))
return nullptr;
- SDValue LaneIdxVal = CurDAG->getTargetConstant(LaneIdx, MVT::i64);
+ SDValue LaneIdxVal = CurDAG->getTargetConstant(LaneIdx, dl, MVT::i64);
SDValue Ops[] = { SMULLOp0, SMULLOp1, LaneIdxVal };
} else
llvm_unreachable("Unrecognized intrinsic.");
- return CurDAG->getMachineNode(SMULLOpc, SDLoc(N), N->getValueType(0), Ops);
+ return CurDAG->getMachineNode(SMULLOpc, dl, N->getValueType(0), Ops);
}
/// Instructions that accept extend modifiers like UXTW expect the register
if (N.getValueType() == MVT::i32)
return N;
- SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
+ SDLoc dl(N);
+ SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, dl, MVT::i32);
MachineSDNode *Node = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
- SDLoc(N), MVT::i32, N, SubReg);
+ dl, MVT::i32, N, SubReg);
return SDValue(Node, 0);
}
// (harmlessly) synthesize one by injected an EXTRACT_SUBREG here.
assert(Ext != AArch64_AM::UXTX && Ext != AArch64_AM::SXTX);
Reg = narrowIfNeeded(CurDAG, Reg);
- Shift = CurDAG->getTargetConstant(getArithExtendImm(Ext, ShiftVal), MVT::i32);
+ Shift = CurDAG->getTargetConstant(getArithExtendImm(Ext, ShiftVal), SDLoc(N),
+ MVT::i32);
return isWorthFolding(N);
}
/// reference, which determines the scale.
bool AArch64DAGToDAGISel::SelectAddrModeIndexed(SDValue N, unsigned Size,
SDValue &Base, SDValue &OffImm) {
+ SDLoc dl(N);
+ const DataLayout &DL = CurDAG->getDataLayout();
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);
+ Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy(DL));
+ OffImm = CurDAG->getTargetConstant(0, dl, MVT::i64);
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);
+ Alignment = DL.getABITypeAlignment(Ty);
if (Alignment >= Size)
return true;
Base = N.getOperand(0);
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
- Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy());
+ Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy(DL));
}
- OffImm = CurDAG->getTargetConstant(RHSC >> Scale, MVT::i64);
+ OffImm = CurDAG->getTargetConstant(RHSC >> Scale, dl, MVT::i64);
return true;
}
}
// add x0, Xbase, #offset
// ldr x0, [x0]
Base = N;
- OffImm = CurDAG->getTargetConstant(0, MVT::i64);
+ OffImm = CurDAG->getTargetConstant(0, dl, MVT::i64);
return true;
}
if (Base.getOpcode() == ISD::FrameIndex) {
int FI = cast<FrameIndexSDNode>(Base)->getIndex();
const TargetLowering *TLI = getTargetLowering();
- Base = CurDAG->getTargetFrameIndex(FI, TLI->getPointerTy());
+ Base = CurDAG->getTargetFrameIndex(
+ FI, TLI->getPointerTy(CurDAG->getDataLayout()));
}
- OffImm = CurDAG->getTargetConstant(RHSC, MVT::i64);
+ OffImm = CurDAG->getTargetConstant(RHSC, SDLoc(N), MVT::i64);
return true;
}
}
}
static SDValue Widen(SelectionDAG *CurDAG, SDValue N) {
- SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
+ SDLoc dl(N);
+ SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, dl, MVT::i32);
SDValue ImpDef = SDValue(
- CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, SDLoc(N), MVT::i64),
- 0);
+ CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, MVT::i64), 0);
MachineSDNode *Node = CurDAG->getMachineNode(
- TargetOpcode::INSERT_SUBREG, SDLoc(N), MVT::i64, ImpDef, N, SubReg);
+ TargetOpcode::INSERT_SUBREG, dl, MVT::i64, ImpDef, N, SubReg);
return SDValue(Node, 0);
}
if (!CSD || (CSD->getZExtValue() & 0x7) != CSD->getZExtValue())
return false;
+ SDLoc dl(N);
if (WantExtend) {
AArch64_AM::ShiftExtendType Ext =
getExtendTypeForNode(N.getOperand(0), true);
return false;
Offset = narrowIfNeeded(CurDAG, N.getOperand(0).getOperand(0));
- SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, MVT::i32);
+ SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, dl,
+ MVT::i32);
} else {
Offset = N.getOperand(0);
- SignExtend = CurDAG->getTargetConstant(0, MVT::i32);
+ SignExtend = CurDAG->getTargetConstant(0, dl, MVT::i32);
}
unsigned LegalShiftVal = Log2_32(Size);
return false;
SDValue LHS = N.getOperand(0);
SDValue RHS = N.getOperand(1);
+ SDLoc dl(N);
// We don't want to match immediate adds here, because they are better lowered
// to the register-immediate addressing modes.
if (IsExtendedRegisterWorthFolding && RHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(RHS, Size, true, Offset, SignExtend)) {
Base = LHS;
- DoShift = CurDAG->getTargetConstant(true, MVT::i32);
+ DoShift = CurDAG->getTargetConstant(true, dl, MVT::i32);
return true;
}
if (IsExtendedRegisterWorthFolding && LHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(LHS, Size, true, Offset, SignExtend)) {
Base = RHS;
- DoShift = CurDAG->getTargetConstant(true, MVT::i32);
+ DoShift = CurDAG->getTargetConstant(true, dl, MVT::i32);
return true;
}
// There was no shift, whatever else we find.
- DoShift = CurDAG->getTargetConstant(false, MVT::i32);
+ DoShift = CurDAG->getTargetConstant(false, dl, MVT::i32);
AArch64_AM::ShiftExtendType Ext = AArch64_AM::InvalidShiftExtend;
// Try to match an unshifted extend on the LHS.
AArch64_AM::InvalidShiftExtend) {
Base = RHS;
Offset = narrowIfNeeded(CurDAG, LHS.getOperand(0));
- SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, MVT::i32);
+ SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, dl,
+ MVT::i32);
if (isWorthFolding(LHS))
return true;
}
AArch64_AM::InvalidShiftExtend) {
Base = LHS;
Offset = narrowIfNeeded(CurDAG, RHS.getOperand(0));
- SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, MVT::i32);
+ SignExtend = CurDAG->getTargetConstant(Ext == AArch64_AM::SXTW, dl,
+ MVT::i32);
if (isWorthFolding(RHS))
return true;
}
return false;
SDValue LHS = N.getOperand(0);
SDValue RHS = N.getOperand(1);
+ SDLoc DL(N);
// Check if this particular node is reused in any non-memory related
// operation. If yes, do not try to fold this node into the address
// MOV X0, WideImmediate
// LDR X2, [BaseReg, X0]
if (isa<ConstantSDNode>(RHS)) {
- int64_t ImmOff = (int64_t)dyn_cast<ConstantSDNode>(RHS)->getZExtValue();
+ int64_t ImmOff = (int64_t)cast<ConstantSDNode>(RHS)->getZExtValue();
unsigned Scale = Log2_32(Size);
// Skip the immediate can be seleced by load/store addressing mode.
// Also skip the immediate can be encoded by a single ADD (SUB is also
isPreferredADD(ImmOff) || isPreferredADD(-ImmOff))
return false;
- SDLoc DL(N.getNode());
SDValue Ops[] = { RHS };
SDNode *MOVI =
CurDAG->getMachineNode(AArch64::MOVi64imm, DL, MVT::i64, Ops);
if (IsExtendedRegisterWorthFolding && RHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(RHS, Size, false, Offset, SignExtend)) {
Base = LHS;
- DoShift = CurDAG->getTargetConstant(true, MVT::i32);
+ DoShift = CurDAG->getTargetConstant(true, DL, MVT::i32);
return true;
}
if (IsExtendedRegisterWorthFolding && LHS.getOpcode() == ISD::SHL &&
SelectExtendedSHL(LHS, Size, false, Offset, SignExtend)) {
Base = RHS;
- DoShift = CurDAG->getTargetConstant(true, MVT::i32);
+ DoShift = CurDAG->getTargetConstant(true, DL, MVT::i32);
return true;
}
// Match any non-shifted, non-extend, non-immediate add expression.
Base = LHS;
Offset = RHS;
- SignExtend = CurDAG->getTargetConstant(false, MVT::i32);
- DoShift = CurDAG->getTargetConstant(false, MVT::i32);
+ SignExtend = CurDAG->getTargetConstant(false, DL, MVT::i32);
+ DoShift = CurDAG->getTargetConstant(false, DL, MVT::i32);
// Reg1 + Reg2 is free: no check needed.
return true;
}
assert(Regs.size() >= 2 && Regs.size() <= 4);
- SDLoc DL(Regs[0].getNode());
+ SDLoc DL(Regs[0]);
SmallVector<SDValue, 4> Ops;
// First operand of REG_SEQUENCE is the desired RegClass.
Ops.push_back(
- CurDAG->getTargetConstant(RegClassIDs[Regs.size() - 2], MVT::i32));
+ CurDAG->getTargetConstant(RegClassIDs[Regs.size() - 2], DL, MVT::i32));
// Then we get pairs of source & subregister-position for the components.
for (unsigned i = 0; i < Regs.size(); ++i) {
Ops.push_back(Regs[i]);
- Ops.push_back(CurDAG->getTargetConstant(SubRegs[i], MVT::i32));
+ Ops.push_back(CurDAG->getTargetConstant(SubRegs[i], DL, MVT::i32));
}
SDNode *N =
// it into an i64.
DstVT = MVT::i32;
}
+ } else if (VT == MVT::f16) {
+ Opcode = IsPre ? AArch64::LDRHpre : AArch64::LDRHpost;
} else if (VT == MVT::f32) {
Opcode = IsPre ? AArch64::LDRSpre : AArch64::LDRSpost;
} else if (VT == MVT::f64 || VT.is64BitVector()) {
SDValue Base = LD->getBasePtr();
ConstantSDNode *OffsetOp = cast<ConstantSDNode>(LD->getOffset());
int OffsetVal = (int)OffsetOp->getZExtValue();
- SDValue Offset = CurDAG->getTargetConstant(OffsetVal, MVT::i64);
+ SDLoc dl(N);
+ SDValue Offset = CurDAG->getTargetConstant(OffsetVal, dl, MVT::i64);
SDValue Ops[] = { Base, Offset, Chain };
- SDNode *Res = CurDAG->getMachineNode(Opcode, SDLoc(N), MVT::i64, DstVT,
+ SDNode *Res = CurDAG->getMachineNode(Opcode, dl, MVT::i64, DstVT,
MVT::Other, Ops);
// Either way, we're replacing the node, so tell the caller that.
Done = true;
SDValue LoadedVal = SDValue(Res, 1);
if (InsertTo64) {
- SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
+ SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, dl, MVT::i32);
LoadedVal =
SDValue(CurDAG->getMachineNode(
- AArch64::SUBREG_TO_REG, SDLoc(N), MVT::i64,
- CurDAG->getTargetConstant(0, MVT::i64), LoadedVal, SubReg),
+ AArch64::SUBREG_TO_REG, dl, MVT::i64,
+ CurDAG->getTargetConstant(0, dl, MVT::i64), LoadedVal,
+ SubReg),
0);
}
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 2))->getZExtValue();
- SDValue Ops[] = {RegSeq, CurDAG->getTargetConstant(LaneNo, MVT::i64),
+ SDValue Ops[] = {RegSeq, CurDAG->getTargetConstant(LaneNo, dl, MVT::i64),
N->getOperand(NumVecs + 3), N->getOperand(0)};
SDNode *Ld = CurDAG->getMachineNode(Opc, dl, ResTys, Ops);
SDValue SuperReg = SDValue(Ld, 0);
SDValue RegSeq = createQTuple(Regs);
const EVT ResTys[] = {MVT::i64, // Type of the write back register
- MVT::Untyped, MVT::Other};
+ RegSeq->getValueType(0), MVT::Other};
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 1))->getZExtValue();
SDValue Ops[] = {RegSeq,
- CurDAG->getTargetConstant(LaneNo, MVT::i64), // Lane Number
+ CurDAG->getTargetConstant(LaneNo, dl,
+ MVT::i64), // Lane Number
N->getOperand(NumVecs + 2), // Base register
N->getOperand(NumVecs + 3), // Incremental
N->getOperand(0)};
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 2))->getZExtValue();
- SDValue Ops[] = {RegSeq, CurDAG->getTargetConstant(LaneNo, MVT::i64),
+ SDValue Ops[] = {RegSeq, CurDAG->getTargetConstant(LaneNo, dl, MVT::i64),
N->getOperand(NumVecs + 3), N->getOperand(0)};
SDNode *St = CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops);
unsigned LaneNo =
cast<ConstantSDNode>(N->getOperand(NumVecs + 1))->getZExtValue();
- SDValue Ops[] = {RegSeq, CurDAG->getTargetConstant(LaneNo, MVT::i64),
+ SDValue Ops[] = {RegSeq, CurDAG->getTargetConstant(LaneNo, dl, MVT::i64),
N->getOperand(NumVecs + 2), // Base Register
N->getOperand(NumVecs + 3), // Incremental
N->getOperand(0)};
}
static bool isBitfieldExtractOpFromShr(SDNode *N, unsigned &Opc, SDValue &Opd0,
- unsigned &LSB, unsigned &MSB,
+ unsigned &Immr, unsigned &Imms,
bool BiggerPattern) {
assert((N->getOpcode() == ISD::SRA || N->getOpcode() == ISD::SRL) &&
"N must be a SHR/SRA operation to call this function");
"Type checking must have been done before calling this function");
// Check for AND + SRL doing several bits extract.
- if (isSeveralBitsExtractOpFromShr(N, Opc, Opd0, LSB, MSB))
+ if (isSeveralBitsExtractOpFromShr(N, Opc, Opd0, Immr, Imms))
return true;
// we're looking for a shift of a shift
assert(Srl_imm > 0 && Srl_imm < VT.getSizeInBits() &&
"bad amount in shift node!");
- // Note: The width operand is encoded as width-1.
- 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;
+ int immr = Srl_imm - Shl_imm;
+ Immr = immr < 0 ? immr + VT.getSizeInBits() : immr;
+ Imms = VT.getSizeInBits() - Shl_imm - Trunc_bits - 1;
// SRA requires a signed extraction
if (VT == MVT::i32)
Opc = N->getOpcode() == ISD::SRA ? AArch64::SBFMWri : AArch64::UBFMWri;
}
static bool isBitfieldExtractOp(SelectionDAG *CurDAG, SDNode *N, unsigned &Opc,
- SDValue &Opd0, unsigned &LSB, unsigned &MSB,
+ SDValue &Opd0, unsigned &Immr, unsigned &Imms,
unsigned NumberOfIgnoredLowBits = 0,
bool BiggerPattern = false) {
if (N->getValueType(0) != MVT::i32 && N->getValueType(0) != MVT::i64)
return false;
break;
case ISD::AND:
- return isBitfieldExtractOpFromAnd(CurDAG, N, Opc, Opd0, LSB, MSB,
+ return isBitfieldExtractOpFromAnd(CurDAG, N, Opc, Opd0, Immr, Imms,
NumberOfIgnoredLowBits, BiggerPattern);
case ISD::SRL:
case ISD::SRA:
- return isBitfieldExtractOpFromShr(N, Opc, Opd0, LSB, MSB, BiggerPattern);
+ return isBitfieldExtractOpFromShr(N, Opc, Opd0, Immr, Imms, BiggerPattern);
}
unsigned NOpc = N->getMachineOpcode();
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();
+ Immr = cast<ConstantSDNode>(N->getOperand(1).getNode())->getZExtValue();
+ Imms = cast<ConstantSDNode>(N->getOperand(2).getNode())->getZExtValue();
return true;
}
// Unreachable
}
SDNode *AArch64DAGToDAGISel::SelectBitfieldExtractOp(SDNode *N) {
- unsigned Opc, LSB, MSB;
+ unsigned Opc, Immr, Imms;
SDValue Opd0;
- if (!isBitfieldExtractOp(CurDAG, N, Opc, Opd0, LSB, MSB))
+ if (!isBitfieldExtractOp(CurDAG, N, Opc, Opd0, Immr, Imms))
return nullptr;
EVT VT = N->getValueType(0);
+ SDLoc dl(N);
// If the bit extract operation is 64bit but the original type is 32bit, we
// need to add one EXTRACT_SUBREG.
if ((Opc == AArch64::SBFMXri || Opc == AArch64::UBFMXri) && VT == MVT::i32) {
- SDValue Ops64[] = {Opd0, CurDAG->getTargetConstant(LSB, MVT::i64),
- CurDAG->getTargetConstant(MSB, MVT::i64)};
+ SDValue Ops64[] = {Opd0, CurDAG->getTargetConstant(Immr, dl, MVT::i64),
+ CurDAG->getTargetConstant(Imms, dl, MVT::i64)};
- SDNode *BFM = CurDAG->getMachineNode(Opc, SDLoc(N), MVT::i64, Ops64);
- SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32);
+ SDNode *BFM = CurDAG->getMachineNode(Opc, dl, MVT::i64, Ops64);
+ SDValue SubReg = CurDAG->getTargetConstant(AArch64::sub_32, dl, MVT::i32);
MachineSDNode *Node =
- CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, SDLoc(N), MVT::i32,
+ CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG, dl, MVT::i32,
SDValue(BFM, 0), SubReg);
return Node;
}
- SDValue Ops[] = {Opd0, CurDAG->getTargetConstant(LSB, VT),
- CurDAG->getTargetConstant(MSB, VT)};
+ SDValue Ops[] = {Opd0, CurDAG->getTargetConstant(Immr, dl, VT),
+ CurDAG->getTargetConstant(Imms, dl, VT)};
return CurDAG->SelectNodeTo(N, Opc, VT, Ops);
}
return Op;
EVT VT = Op.getValueType();
+ SDLoc dl(Op);
unsigned BitWidth = VT.getSizeInBits();
unsigned UBFMOpc = BitWidth == 32 ? AArch64::UBFMWri : AArch64::UBFMXri;
if (ShlAmount > 0) {
// LSL wD, wN, #Amt == UBFM wD, wN, #32-Amt, #31-Amt
ShiftNode = CurDAG->getMachineNode(
- UBFMOpc, SDLoc(Op), VT, Op,
- CurDAG->getTargetConstant(BitWidth - ShlAmount, VT),
- CurDAG->getTargetConstant(BitWidth - 1 - ShlAmount, VT));
+ UBFMOpc, dl, VT, Op,
+ CurDAG->getTargetConstant(BitWidth - ShlAmount, dl, VT),
+ CurDAG->getTargetConstant(BitWidth - 1 - ShlAmount, dl, VT));
} else {
// LSR wD, wN, #Amt == UBFM wD, wN, #Amt, #32-1
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));
+ UBFMOpc, dl, VT, Op, CurDAG->getTargetConstant(ShrAmount, dl, VT),
+ CurDAG->getTargetConstant(BitWidth - 1, dl, VT));
}
return SDValue(ShiftNode, 0);
return nullptr;
EVT VT = N->getValueType(0);
+ SDLoc dl(N);
SDValue Ops[] = { Opd0,
Opd1,
- CurDAG->getTargetConstant(LSB, VT),
- CurDAG->getTargetConstant(MSB, VT) };
+ CurDAG->getTargetConstant(LSB, dl, VT),
+ CurDAG->getTargetConstant(MSB, dl, VT) };
return CurDAG->SelectNodeTo(N, Opc, VT, Ops);
}
+/// GenerateInexactFlagIfNeeded - Insert FRINTX instruction to generate inexact
+/// signal on round-to-integer operations if needed. C11 leaves it
+/// implementation-defined whether these operations trigger an inexact
+/// exception. IEEE says they don't. Unfortunately, Darwin decided they do so
+/// we sometimes have to insert a special instruction just to set the right bit
+/// in FPSR.
+SDNode *AArch64DAGToDAGISel::GenerateInexactFlagIfNeeded(const SDValue &In,
+ unsigned InTyVariant,
+ SDLoc DL) {
+ if (Subtarget->isTargetDarwin() && !TM.Options.UnsafeFPMath) {
+ // Pick the right FRINTX using InTyVariant needed to set the flags.
+ // InTyVariant is 0 for 32-bit and 1 for 64-bit.
+ unsigned FRINTXOpcs[] = { AArch64::FRINTXSr, AArch64::FRINTXDr };
+ return CurDAG->getMachineNode(FRINTXOpcs[InTyVariant], DL,
+ In.getValueType(), MVT::Glue, In);
+ }
+ return nullptr;
+}
+
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
return nullptr; // Unrecognized argument type. Fall back on default codegen.
- // Pick the FRINTX variant needed to set the flags.
- unsigned FRINTXOpc = FRINTXOpcs[Variant];
-
switch (N->getOpcode()) {
default:
return nullptr; // Unrecognized libm ISD node. Fall back on default codegen.
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));
- }
+ if (SDNode *FRINTXNode = GenerateInexactFlagIfNeeded(In, Variant, dl))
+ Ops.push_back(SDValue(FRINTXNode, 1));
return CurDAG->getMachineNode(Opc, dl, VT, Ops);
}
+/// SelectFPConvertWithRound - Try to combine FP rounding and
+/// FP-INT conversion.
+SDNode *AArch64DAGToDAGISel::SelectFPConvertWithRound(SDNode *N) {
+ SDNode *Op0 = N->getOperand(0).getNode();
+
+ // Return if the round op is used by other nodes, as this would result in two
+ // FRINTX, one each for round and convert.
+ if (!Op0->hasOneUse())
+ return nullptr;
+
+ unsigned InTyVariant;
+ EVT InTy = Op0->getValueType(0);
+ if (InTy == MVT::f32)
+ InTyVariant = 0;
+ else if (InTy == MVT::f64)
+ InTyVariant = 1;
+ else
+ return nullptr;
+
+ unsigned OutTyVariant;
+ EVT OutTy = N->getValueType(0);
+ if (OutTy == MVT::i32)
+ OutTyVariant = 0;
+ else if (OutTy == MVT::i64)
+ OutTyVariant = 1;
+ else
+ return nullptr;
+
+ assert((N->getOpcode() == ISD::FP_TO_SINT
+ || N->getOpcode() == ISD::FP_TO_UINT) && "Unexpected opcode!");
+ unsigned FpConVariant = N->getOpcode() == ISD::FP_TO_SINT ? 0 : 1;
+
+ unsigned Opc;
+ switch (Op0->getOpcode()) {
+ default:
+ return nullptr;
+ case ISD::FCEIL: {
+ unsigned FCVTPOpcs[2][2][2] = {
+ { { AArch64::FCVTPSUWSr, AArch64::FCVTPSUXSr },
+ { AArch64::FCVTPSUWDr, AArch64::FCVTPSUXDr } },
+ { { AArch64::FCVTPUUWSr, AArch64::FCVTPUUXSr },
+ { AArch64::FCVTPUUWDr, AArch64::FCVTPUUXDr } } };
+ Opc = FCVTPOpcs[FpConVariant][InTyVariant][OutTyVariant];
+ break;
+ }
+ case ISD::FFLOOR: {
+ unsigned FCVTMOpcs[2][2][2] = {
+ { { AArch64::FCVTMSUWSr, AArch64::FCVTMSUXSr },
+ { AArch64::FCVTMSUWDr, AArch64::FCVTMSUXDr } },
+ { { AArch64::FCVTMUUWSr, AArch64::FCVTMUUXSr },
+ { AArch64::FCVTMUUWDr, AArch64::FCVTMUUXDr } } };
+ Opc = FCVTMOpcs[FpConVariant][InTyVariant][OutTyVariant];
+ break;
+ }
+ case ISD::FTRUNC: {
+ unsigned FCVTZOpcs[2][2][2] = {
+ { { AArch64::FCVTZSUWSr, AArch64::FCVTZSUXSr },
+ { AArch64::FCVTZSUWDr, AArch64::FCVTZSUXDr } },
+ { { AArch64::FCVTZUUWSr, AArch64::FCVTZUUXSr },
+ { AArch64::FCVTZUUWDr, AArch64::FCVTZUUXDr } } };
+ Opc = FCVTZOpcs[FpConVariant][InTyVariant][OutTyVariant];
+ break;
+ }
+ case ISD::FROUND: {
+ unsigned FCVTAOpcs[2][2][2] = {
+ { { AArch64::FCVTASUWSr, AArch64::FCVTASUXSr },
+ { AArch64::FCVTASUWDr, AArch64::FCVTASUXDr } },
+ { { AArch64::FCVTAUUWSr, AArch64::FCVTAUUXSr },
+ { AArch64::FCVTAUUWDr, AArch64::FCVTAUUXDr } } };
+ Opc = FCVTAOpcs[FpConVariant][InTyVariant][OutTyVariant];
+ break;
+ }
+ }
+
+ SDLoc DL(N);
+ SDValue In = Op0->getOperand(0);
+ SmallVector<SDValue, 2> Ops;
+ Ops.push_back(In);
+
+ if (SDNode *FRINTXNode = GenerateInexactFlagIfNeeded(In, InTyVariant, DL))
+ Ops.push_back(SDValue(FRINTXNode, 1));
+
+ return CurDAG->getMachineNode(Opc, DL, OutTy, Ops);
+}
+
bool
AArch64DAGToDAGISel::SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos,
unsigned RegWidth) {
// finding FBits, but it must still be in range.
if (FBits == 0 || FBits > RegWidth) return false;
- FixedPos = CurDAG->getTargetConstant(FBits, MVT::i32);
+ FixedPos = CurDAG->getTargetConstant(FBits, SDLoc(N), MVT::i32);
return true;
}
+// Inspects a register string of the form o0:op1:CRn:CRm:op2 gets the fields
+// of the string and obtains the integer values from them and combines these
+// into a single value to be used in the MRS/MSR instruction.
+static int getIntOperandFromRegisterString(StringRef RegString) {
+ SmallVector<StringRef, 5> Fields;
+ RegString.split(Fields, ":");
+
+ if (Fields.size() == 1)
+ return -1;
+
+ assert(Fields.size() == 5
+ && "Invalid number of fields in read register string");
+
+ SmallVector<int, 5> Ops;
+ bool AllIntFields = true;
+
+ for (StringRef Field : Fields) {
+ unsigned IntField;
+ AllIntFields &= !Field.getAsInteger(10, IntField);
+ Ops.push_back(IntField);
+ }
+
+ assert(AllIntFields &&
+ "Unexpected non-integer value in special register string.");
+
+ // Need to combine the integer fields of the string into a single value
+ // based on the bit encoding of MRS/MSR instruction.
+ return (Ops[0] << 14) | (Ops[1] << 11) | (Ops[2] << 7) |
+ (Ops[3] << 3) | (Ops[4]);
+}
+
+// Lower the read_register intrinsic to an MRS instruction node if the special
+// register string argument is either of the form detailed in the ALCE (the
+// form described in getIntOperandsFromRegsterString) or is a named register
+// known by the MRS SysReg mapper.
+SDNode *AArch64DAGToDAGISel::SelectReadRegister(SDNode *N) {
+ const MDNodeSDNode *MD = dyn_cast<MDNodeSDNode>(N->getOperand(1));
+ const MDString *RegString = dyn_cast<MDString>(MD->getMD()->getOperand(0));
+ SDLoc DL(N);
+
+ int Reg = getIntOperandFromRegisterString(RegString->getString());
+ if (Reg != -1)
+ return CurDAG->getMachineNode(AArch64::MRS, DL, N->getSimpleValueType(0),
+ MVT::Other,
+ CurDAG->getTargetConstant(Reg, DL, MVT::i32),
+ N->getOperand(0));
+
+ // Use the sysreg mapper to map the remaining possible strings to the
+ // value for the register to be used for the instruction operand.
+ AArch64SysReg::MRSMapper mapper;
+ bool IsValidSpecialReg;
+ Reg = mapper.fromString(RegString->getString(),
+ Subtarget->getFeatureBits(),
+ IsValidSpecialReg);
+ if (IsValidSpecialReg)
+ return CurDAG->getMachineNode(AArch64::MRS, DL, N->getSimpleValueType(0),
+ MVT::Other,
+ CurDAG->getTargetConstant(Reg, DL, MVT::i32),
+ N->getOperand(0));
+
+ return nullptr;
+}
+
+// Lower the write_register intrinsic to an MSR instruction node if the special
+// register string argument is either of the form detailed in the ALCE (the
+// form described in getIntOperandsFromRegsterString) or is a named register
+// known by the MSR SysReg mapper.
+SDNode *AArch64DAGToDAGISel::SelectWriteRegister(SDNode *N) {
+ const MDNodeSDNode *MD = dyn_cast<MDNodeSDNode>(N->getOperand(1));
+ const MDString *RegString = dyn_cast<MDString>(MD->getMD()->getOperand(0));
+ SDLoc DL(N);
+
+ int Reg = getIntOperandFromRegisterString(RegString->getString());
+ if (Reg != -1)
+ return CurDAG->getMachineNode(AArch64::MSR, DL, MVT::Other,
+ CurDAG->getTargetConstant(Reg, DL, MVT::i32),
+ N->getOperand(2), N->getOperand(0));
+
+ // Check if the register was one of those allowed as the pstatefield value in
+ // the MSR (immediate) instruction. To accept the values allowed in the
+ // pstatefield for the MSR (immediate) instruction, we also require that an
+ // immediate value has been provided as an argument, we know that this is
+ // the case as it has been ensured by semantic checking.
+ AArch64PState::PStateMapper PMapper;
+ bool IsValidSpecialReg;
+ Reg = PMapper.fromString(RegString->getString(),
+ Subtarget->getFeatureBits(),
+ IsValidSpecialReg);
+ if (IsValidSpecialReg) {
+ assert (isa<ConstantSDNode>(N->getOperand(2))
+ && "Expected a constant integer expression.");
+ uint64_t Immed = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue();
+ return CurDAG->getMachineNode(AArch64::MSRpstate, DL, MVT::Other,
+ CurDAG->getTargetConstant(Reg, DL, MVT::i32),
+ CurDAG->getTargetConstant(Immed, DL, MVT::i16),
+ N->getOperand(0));
+ }
+
+ // Use the sysreg mapper to attempt to map the remaining possible strings
+ // to the value for the register to be used for the MSR (register)
+ // instruction operand.
+ AArch64SysReg::MSRMapper Mapper;
+ Reg = Mapper.fromString(RegString->getString(),
+ Subtarget->getFeatureBits(),
+ IsValidSpecialReg);
+
+ if (IsValidSpecialReg)
+ return CurDAG->getMachineNode(AArch64::MSR, DL, MVT::Other,
+ CurDAG->getTargetConstant(Reg, DL, MVT::i32),
+ N->getOperand(2), N->getOperand(0));
+
+ return nullptr;
+}
+
SDNode *AArch64DAGToDAGISel::Select(SDNode *Node) {
// Dump information about the Node being selected
DEBUG(errs() << "Selecting: ");
default:
break;
+ case ISD::READ_REGISTER:
+ if (SDNode *Res = SelectReadRegister(Node))
+ return Res;
+ break;
+
+ case ISD::WRITE_REGISTER:
+ if (SDNode *Res = SelectWriteRegister(Node))
+ return Res;
+ break;
+
case ISD::ADD:
if (SDNode *I = SelectMLAV64LaneV128(Node))
return I;
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) };
+ SDValue TFI = CurDAG->getTargetFrameIndex(
+ FI, TLI->getPointerTy(CurDAG->getDataLayout()));
+ SDLoc DL(Node);
+ SDValue Ops[] = { TFI, CurDAG->getTargetConstant(0, DL, MVT::i32),
+ CurDAG->getTargetConstant(Shifter, DL, MVT::i32) };
return CurDAG->SelectNodeTo(Node, AArch64::ADDXri, MVT::i64, Ops);
}
case ISD::INTRINSIC_W_CHAIN: {
break;
}
}
+ break;
}
case AArch64ISD::LD2post: {
if (VT == MVT::v8i8)
if (SDNode *I = SelectLIBM(Node))
return I;
break;
+
+ case ISD::FP_TO_SINT:
+ case ISD::FP_TO_UINT:
+ if (SDNode *I = SelectFPConvertWithRound(Node))
+ return I;
+ break;
}
// Select the default instruction
projects / oota-llvm.git / blobdiff