default:
return false;
case X86::MOV8rr:
+ case X86::MOV8rr_NOREX:
case X86::MOV16rr:
case X86::MOV32rr:
case X86::MOV64rr:
case X86::CMOVG64rr: Opc = X86::CMOVLE64rr; break;
case X86::CMOVS16rr: Opc = X86::CMOVNS16rr; break;
case X86::CMOVS32rr: Opc = X86::CMOVNS32rr; break;
- case X86::CMOVS64rr: Opc = X86::CMOVNS32rr; break;
+ case X86::CMOVS64rr: Opc = X86::CMOVNS64rr; break;
case X86::CMOVNS16rr: Opc = X86::CMOVS16rr; break;
case X86::CMOVNS32rr: Opc = X86::CMOVS32rr; break;
case X86::CMOVNS64rr: Opc = X86::CMOVS64rr; break;
case X86::CMOVP16rr: Opc = X86::CMOVNP16rr; break;
case X86::CMOVP32rr: Opc = X86::CMOVNP32rr; break;
- case X86::CMOVP64rr: Opc = X86::CMOVNP32rr; break;
+ case X86::CMOVP64rr: Opc = X86::CMOVNP64rr; break;
case X86::CMOVNP16rr: Opc = X86::CMOVP16rr; break;
case X86::CMOVNP32rr: Opc = X86::CMOVP32rr; break;
case X86::CMOVNP64rr: Opc = X86::CMOVP64rr; break;
case X86::CMOVO16rr: Opc = X86::CMOVNO16rr; break;
case X86::CMOVO32rr: Opc = X86::CMOVNO32rr; break;
- case X86::CMOVO64rr: Opc = X86::CMOVNO32rr; break;
+ case X86::CMOVO64rr: Opc = X86::CMOVNO64rr; break;
case X86::CMOVNO16rr: Opc = X86::CMOVO16rr; break;
case X86::CMOVNO32rr: Opc = X86::CMOVO32rr; break;
case X86::CMOVNO64rr: Opc = X86::CMOVO64rr; break;
/// isHReg - Test if the given register is a physical h register.
static bool isHReg(unsigned Reg) {
- return Reg == X86::AH || Reg == X86::BH || Reg == X86::CH || Reg == X86::DH;
+ return X86::GR8_ABCD_HRegClass.contains(Reg);
}
bool X86InstrInfo::copyRegToReg(MachineBasicBlock &MBB,
DebugLoc DL = DebugLoc::getUnknownLoc();
if (MI != MBB.end()) DL = MI->getDebugLoc();
- if (DestRC == SrcRC) {
+ // Determine if DstRC and SrcRC have a common superclass in common.
+ const TargetRegisterClass *CommonRC = DestRC;
+ if (DestRC == SrcRC)
+ /* Source and destination have the same register class. */;
+ else if (CommonRC->hasSuperClass(SrcRC))
+ CommonRC = SrcRC;
+ else if (!DestRC->hasSubClass(SrcRC))
+ CommonRC = 0;
+
+ if (CommonRC) {
unsigned Opc;
- if (DestRC == &X86::GR64RegClass) {
+ if (CommonRC == &X86::GR64RegClass) {
Opc = X86::MOV64rr;
- } else if (DestRC == &X86::GR32RegClass) {
+ } else if (CommonRC == &X86::GR32RegClass) {
Opc = X86::MOV32rr;
- } else if (DestRC == &X86::GR16RegClass) {
+ } else if (CommonRC == &X86::GR16RegClass) {
Opc = X86::MOV16rr;
- } else if (DestRC == &X86::GR8RegClass) {
- // Copying two or from a physical H register requires a NOREX move.
- // Otherwise use a normal move.
- if (isHReg(DestReg) || isHReg(SrcReg))
+ } else if (CommonRC == &X86::GR8RegClass) {
+ // Copying to or from a physical H register on x86-64 requires a NOREX
+ // move. Otherwise use a normal move.
+ if ((isHReg(DestReg) || isHReg(SrcReg)) &&
+ TM.getSubtarget<X86Subtarget>().is64Bit())
Opc = X86::MOV8rr_NOREX;
else
Opc = X86::MOV8rr;
- } else if (DestRC == &X86::GR64_RegClass) {
+ } else if (CommonRC == &X86::GR64_ABCDRegClass) {
Opc = X86::MOV64rr;
- } else if (DestRC == &X86::GR32_RegClass) {
+ } else if (CommonRC == &X86::GR32_ABCDRegClass) {
Opc = X86::MOV32rr;
- } else if (DestRC == &X86::GR16_RegClass) {
+ } else if (CommonRC == &X86::GR16_ABCDRegClass) {
Opc = X86::MOV16rr;
- } else if (DestRC == &X86::GR8_RegClass) {
+ } else if (CommonRC == &X86::GR8_ABCD_LRegClass) {
Opc = X86::MOV8rr;
- } else if (DestRC == &X86::GR64_NOREXRegClass) {
+ } else if (CommonRC == &X86::GR8_ABCD_HRegClass) {
+ if (TM.getSubtarget<X86Subtarget>().is64Bit())
+ Opc = X86::MOV8rr_NOREX;
+ else
+ Opc = X86::MOV8rr;
+ } else if (CommonRC == &X86::GR64_NOREXRegClass) {
Opc = X86::MOV64rr;
- } else if (DestRC == &X86::GR32_NOREXRegClass) {
+ } else if (CommonRC == &X86::GR32_NOREXRegClass) {
Opc = X86::MOV32rr;
- } else if (DestRC == &X86::GR16_NOREXRegClass) {
+ } else if (CommonRC == &X86::GR16_NOREXRegClass) {
Opc = X86::MOV16rr;
- } else if (DestRC == &X86::GR8_NOREXRegClass) {
+ } else if (CommonRC == &X86::GR8_NOREXRegClass) {
Opc = X86::MOV8rr;
- } else if (DestRC == &X86::RFP32RegClass) {
+ } else if (CommonRC == &X86::RFP32RegClass) {
Opc = X86::MOV_Fp3232;
- } else if (DestRC == &X86::RFP64RegClass || DestRC == &X86::RSTRegClass) {
+ } else if (CommonRC == &X86::RFP64RegClass || CommonRC == &X86::RSTRegClass) {
Opc = X86::MOV_Fp6464;
- } else if (DestRC == &X86::RFP80RegClass) {
+ } else if (CommonRC == &X86::RFP80RegClass) {
Opc = X86::MOV_Fp8080;
- } else if (DestRC == &X86::FR32RegClass) {
+ } else if (CommonRC == &X86::FR32RegClass) {
Opc = X86::FsMOVAPSrr;
- } else if (DestRC == &X86::FR64RegClass) {
+ } else if (CommonRC == &X86::FR64RegClass) {
Opc = X86::FsMOVAPDrr;
- } else if (DestRC == &X86::VR128RegClass) {
+ } else if (CommonRC == &X86::VR128RegClass) {
Opc = X86::MOVAPSrr;
- } else if (DestRC == &X86::VR64RegClass) {
+ } else if (CommonRC == &X86::VR64RegClass) {
Opc = X86::MMX_MOVQ64rr;
} else {
return false;
return false;
}
-static unsigned getStoreRegOpcode(const TargetRegisterClass *RC,
- bool isStackAligned) {
+static unsigned getStoreRegOpcode(unsigned SrcReg,
+ const TargetRegisterClass *RC,
+ bool isStackAligned,
+ TargetMachine &TM) {
unsigned Opc = 0;
if (RC == &X86::GR64RegClass) {
Opc = X86::MOV64mr;
} else if (RC == &X86::GR16RegClass) {
Opc = X86::MOV16mr;
} else if (RC == &X86::GR8RegClass) {
- Opc = X86::MOV8mr;
- } else if (RC == &X86::GR64_RegClass) {
+ // Copying to or from a physical H register on x86-64 requires a NOREX
+ // move. Otherwise use a normal move.
+ if (isHReg(SrcReg) &&
+ TM.getSubtarget<X86Subtarget>().is64Bit())
+ Opc = X86::MOV8mr_NOREX;
+ else
+ Opc = X86::MOV8mr;
+ } else if (RC == &X86::GR64_ABCDRegClass) {
Opc = X86::MOV64mr;
- } else if (RC == &X86::GR32_RegClass) {
+ } else if (RC == &X86::GR32_ABCDRegClass) {
Opc = X86::MOV32mr;
- } else if (RC == &X86::GR16_RegClass) {
+ } else if (RC == &X86::GR16_ABCDRegClass) {
Opc = X86::MOV16mr;
- } else if (RC == &X86::GR8_RegClass) {
+ } else if (RC == &X86::GR8_ABCD_LRegClass) {
Opc = X86::MOV8mr;
+ } else if (RC == &X86::GR8_ABCD_HRegClass) {
+ if (TM.getSubtarget<X86Subtarget>().is64Bit())
+ Opc = X86::MOV8mr_NOREX;
+ else
+ Opc = X86::MOV8mr;
} else if (RC == &X86::GR64_NOREXRegClass) {
Opc = X86::MOV64mr;
} else if (RC == &X86::GR32_NOREXRegClass) {
const MachineFunction &MF = *MBB.getParent();
bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
- unsigned Opc = getStoreRegOpcode(RC, isAligned);
+ unsigned Opc = getStoreRegOpcode(SrcReg, RC, isAligned, TM);
DebugLoc DL = DebugLoc::getUnknownLoc();
if (MI != MBB.end()) DL = MI->getDebugLoc();
addFrameReference(BuildMI(MBB, MI, DL, get(Opc)), FrameIdx)
SmallVectorImpl<MachineInstr*> &NewMIs) const {
bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
- unsigned Opc = getStoreRegOpcode(RC, isAligned);
+ unsigned Opc = getStoreRegOpcode(SrcReg, RC, isAligned, TM);
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc));
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
NewMIs.push_back(MIB);
}
-static unsigned getLoadRegOpcode(const TargetRegisterClass *RC,
- bool isStackAligned) {
+static unsigned getLoadRegOpcode(unsigned DestReg,
+ const TargetRegisterClass *RC,
+ bool isStackAligned,
+ const TargetMachine &TM) {
unsigned Opc = 0;
if (RC == &X86::GR64RegClass) {
Opc = X86::MOV64rm;
} else if (RC == &X86::GR16RegClass) {
Opc = X86::MOV16rm;
} else if (RC == &X86::GR8RegClass) {
- Opc = X86::MOV8rm;
- } else if (RC == &X86::GR64_RegClass) {
+ // Copying to or from a physical H register on x86-64 requires a NOREX
+ // move. Otherwise use a normal move.
+ if (isHReg(DestReg) &&
+ TM.getSubtarget<X86Subtarget>().is64Bit())
+ Opc = X86::MOV8rm_NOREX;
+ else
+ Opc = X86::MOV8rm;
+ } else if (RC == &X86::GR64_ABCDRegClass) {
Opc = X86::MOV64rm;
- } else if (RC == &X86::GR32_RegClass) {
+ } else if (RC == &X86::GR32_ABCDRegClass) {
Opc = X86::MOV32rm;
- } else if (RC == &X86::GR16_RegClass) {
+ } else if (RC == &X86::GR16_ABCDRegClass) {
Opc = X86::MOV16rm;
- } else if (RC == &X86::GR8_RegClass) {
+ } else if (RC == &X86::GR8_ABCD_LRegClass) {
Opc = X86::MOV8rm;
+ } else if (RC == &X86::GR8_ABCD_HRegClass) {
+ if (TM.getSubtarget<X86Subtarget>().is64Bit())
+ Opc = X86::MOV8rm_NOREX;
+ else
+ Opc = X86::MOV8rm;
} else if (RC == &X86::GR64_NOREXRegClass) {
Opc = X86::MOV64rm;
} else if (RC == &X86::GR32_NOREXRegClass) {
const MachineFunction &MF = *MBB.getParent();
bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
- unsigned Opc = getLoadRegOpcode(RC, isAligned);
+ unsigned Opc = getLoadRegOpcode(DestReg, RC, isAligned, TM);
DebugLoc DL = DebugLoc::getUnknownLoc();
if (MI != MBB.end()) DL = MI->getDebugLoc();
addFrameReference(BuildMI(MBB, MI, DL, get(Opc), DestReg), FrameIdx);
SmallVectorImpl<MachineInstr*> &NewMIs) const {
bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
- unsigned Opc = getLoadRegOpcode(RC, isAligned);
+ unsigned Opc = getLoadRegOpcode(DestReg, RC, isAligned, TM);
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc), DestReg);
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MVT VT = *RC->vt_begin();
bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
- Load = DAG.getTargetNode(getLoadRegOpcode(RC, isAligned), dl,
- VT, MVT::Other,
- &AddrOps[0], AddrOps.size());
+ Load = DAG.getTargetNode(getLoadRegOpcode(0, RC, isAligned, TM), dl,
+ VT, MVT::Other, &AddrOps[0], AddrOps.size());
NewNodes.push_back(Load);
}
AddrOps.push_back(Chain);
bool isAligned = (RI.getStackAlignment() >= 16) ||
RI.needsStackRealignment(MF);
- SDNode *Store = DAG.getTargetNode(getStoreRegOpcode(DstRC, isAligned), dl,
- MVT::Other, &AddrOps[0], AddrOps.size());
+ SDNode *Store = DAG.getTargetNode(getStoreRegOpcode(0, DstRC,
+ isAligned, TM),
+ dl, MVT::Other,
+ &AddrOps[0], AddrOps.size());
NewNodes.push_back(Store);
}
unsigned BaseReg = Base.getReg();
// Is a SIB byte needed?
- if ((!Is64BitMode || DispForReloc) && IndexReg.getReg() == 0 &&
- (BaseReg == 0 || X86RegisterInfo::getX86RegNum(BaseReg) != N86::ESP)) {
+ if (IndexReg.getReg() == 0 && (!Is64BitMode || BaseReg != 0) &&
+ (BaseReg == 0 || X86RegisterInfo::getX86RegNum(BaseReg) != N86::ESP)) {
if (BaseReg == 0) { // Just a displacement?
// Emit special case [disp32] encoding
++FinalSize;
unsigned CurOp = 0;
if (NumOps > 1 && Desc->getOperandConstraint(1, TOI::TIED_TO) != -1)
CurOp++;
+ else if (NumOps > 2 && Desc->getOperandConstraint(NumOps-1, TOI::TIED_TO)== 0)
+ // Skip the last source operand that is tied_to the dest reg. e.g. LXADD32
+ --NumOps;
switch (Desc->TSFlags & X86II::FormMask) {
default: assert(0 && "Unknown FormMask value in X86 MachineCodeEmitter!");
case X86II::MRMDestMem: {
++FinalSize;
FinalSize += getMemModRMByteSize(MI, CurOp, IsPIC, Is64BitMode);
- CurOp += 5;
+ CurOp += X86AddrNumOperands + 1;
if (CurOp != NumOps) {
++CurOp;
FinalSize += sizeConstant(X86InstrInfo::sizeOfImm(Desc));
break;
case X86II::MRMSrcMem: {
+ int AddrOperands;
+ if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r ||
+ Opcode == X86::LEA16r || Opcode == X86::LEA32r)
+ AddrOperands = X86AddrNumOperands - 1; // No segment register
+ else
+ AddrOperands = X86AddrNumOperands;
++FinalSize;
FinalSize += getMemModRMByteSize(MI, CurOp+1, IsPIC, Is64BitMode);
- CurOp += 5;
+ CurOp += AddrOperands + 1;
if (CurOp != NumOps) {
++CurOp;
FinalSize += sizeConstant(X86InstrInfo::sizeOfImm(Desc));
case X86II::MRM4r: case X86II::MRM5r:
case X86II::MRM6r: case X86II::MRM7r:
++FinalSize;
- ++CurOp;
- FinalSize += sizeRegModRMByte();
+ // Special handling of lfence and mfence.
+ if (Desc->getOpcode() == X86::LFENCE ||
+ Desc->getOpcode() == X86::MFENCE)
+ FinalSize += sizeRegModRMByte();
+ else {
+ ++CurOp;
+ FinalSize += sizeRegModRMByte();
+ }
if (CurOp != NumOps) {
const MachineOperand &MO1 = MI.getOperand(CurOp++);
++FinalSize;
FinalSize += getMemModRMByteSize(MI, CurOp, IsPIC, Is64BitMode);
- CurOp += 4;
+ CurOp += X86AddrNumOperands;
if (CurOp != NumOps) {
const MachineOperand &MO = MI.getOperand(CurOp++);