unsigned A = MI->getOperand(0).getReg();
unsigned M = MI->getOperand(3).getImm();
NewMI = BuildMI(MF, MI->getDebugLoc(), get(X86::PSHUFDri))
- .addReg(A, true, false, false, isDead)
- .addReg(B, false, false, isKill).addImm(M);
+ .addReg(A, RegState::Define | getDeadRegState(isDead))
+ .addReg(B, getKillRegState(isKill)).addImm(M);
break;
}
case X86::SHL64ri: {
if (ShAmt == 0 || ShAmt >= 4) return 0;
NewMI = BuildMI(MF, MI->getDebugLoc(), get(X86::LEA64r))
- .addReg(Dest, true, false, false, isDead)
- .addReg(0).addImm(1 << ShAmt).addReg(Src, false, false, isKill).addImm(0);
+ .addReg(Dest, RegState::Define | getDeadRegState(isDead))
+ .addReg(0).addImm(1 << ShAmt)
+ .addReg(Src, getKillRegState(isKill))
+ .addImm(0);
break;
}
case X86::SHL32ri: {
unsigned Opc = TM.getSubtarget<X86Subtarget>().is64Bit() ?
X86::LEA64_32r : X86::LEA32r;
NewMI = BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead)
+ .addReg(Dest, RegState::Define | getDeadRegState(isDead))
.addReg(0).addImm(1 << ShAmt)
- .addReg(Src, false, false, isKill).addImm(0);
+ .addReg(Src, getKillRegState(isKill)).addImm(0);
break;
}
case X86::SHL16ri: {
BuildMI(*MFI, MBBI, MI->getDebugLoc(), get(X86::IMPLICIT_DEF), leaInReg);
MachineInstr *InsMI =
BuildMI(*MFI, MBBI, MI->getDebugLoc(), get(X86::INSERT_SUBREG),leaInReg)
- .addReg(leaInReg).addReg(Src, false, false, isKill)
+ .addReg(leaInReg)
+ .addReg(Src, getKillRegState(isKill))
.addImm(X86::SUBREG_16BIT);
NewMI = BuildMI(*MFI, MBBI, MI->getDebugLoc(), get(Opc), leaOutReg)
.addReg(0).addImm(1 << ShAmt)
- .addReg(leaInReg, false, false, true).addImm(0);
+ .addReg(leaInReg, RegState::Kill)
+ .addImm(0);
MachineInstr *ExtMI =
BuildMI(*MFI, MBBI, MI->getDebugLoc(), get(X86::EXTRACT_SUBREG))
- .addReg(Dest, true, false, false, isDead)
- .addReg(leaOutReg, false, false, true).addImm(X86::SUBREG_16BIT);
+ .addReg(Dest, RegState::Define | getDeadRegState(isDead))
+ .addReg(leaOutReg, RegState::Kill)
+ .addImm(X86::SUBREG_16BIT);
if (LV) {
// Update live variables
return ExtMI;
} else {
NewMI = BuildMI(MF, MI->getDebugLoc(), get(X86::LEA16r))
- .addReg(Dest, true, false, false, isDead)
+ .addReg(Dest, RegState::Define | getDeadRegState(isDead))
.addReg(0).addImm(1 << ShAmt)
- .addReg(Src, false, false, isKill).addImm(0);
+ .addReg(Src, getKillRegState(isKill))
+ .addImm(0);
}
break;
}
unsigned Opc = MIOpc == X86::INC64r ? X86::LEA64r
: (is64Bit ? X86::LEA64_32r : X86::LEA32r);
NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, 1);
break;
}
if (DisableLEA16) return 0;
assert(MI->getNumOperands() >= 2 && "Unknown inc instruction!");
NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA16r))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, 1);
break;
case X86::DEC64r:
unsigned Opc = MIOpc == X86::DEC64r ? X86::LEA64r
: (is64Bit ? X86::LEA64_32r : X86::LEA32r);
NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, -1);
break;
}
if (DisableLEA16) return 0;
assert(MI->getNumOperands() >= 2 && "Unknown dec instruction!");
NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA16r))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, -1);
break;
case X86::ADD64rr:
unsigned Src2 = MI->getOperand(2).getReg();
bool isKill2 = MI->getOperand(2).isKill();
NewMI = addRegReg(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, Src2, isKill2);
if (LV && isKill2)
LV->replaceKillInstruction(Src2, MI, NewMI);
unsigned Src2 = MI->getOperand(2).getReg();
bool isKill2 = MI->getOperand(2).isKill();
NewMI = addRegReg(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA16r))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, Src2, isKill2);
if (LV && isKill2)
LV->replaceKillInstruction(Src2, MI, NewMI);
assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");
if (MI->getOperand(2).isImm())
NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA64r))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, MI->getOperand(2).getImm());
break;
case X86::ADD32ri:
if (MI->getOperand(2).isImm()) {
unsigned Opc = is64Bit ? X86::LEA64_32r : X86::LEA32r;
NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, MI->getOperand(2).getImm());
}
break;
assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");
if (MI->getOperand(2).isImm())
NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA16r))
- .addReg(Dest, true, false, false, isDead),
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
Src, isKill, MI->getOperand(2).getImm());
break;
case X86::SHL16ri:
: (MIOpc == X86::SHL32ri
? (is64Bit ? X86::LEA64_32r : X86::LEA32r) : X86::LEA16r);
NewMI = addFullAddress(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead), AM);
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)), AM);
if (isKill)
NewMI->getOperand(3).setIsKill(true);
}
if (I->getOpcode() == X86::JMP) {
if (!AllowModify) {
TBB = I->getOperand(0).getMBB();
- return false;
+ continue;
}
// If the block has any instructions after a JMP, delete them.
/// 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,
} else if (CommonRC == &X86::GR16RegClass) {
Opc = X86::MOV16rr;
} else if (CommonRC == &X86::GR8RegClass) {
- // Copying two or from a physical H register on x86-64 requires a NOREX
+ // 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::MOV32rr;
} else if (CommonRC == &X86::GR16_ABCDRegClass) {
Opc = X86::MOV16rr;
- } else if (CommonRC == &X86::GR8_ABCDRegClass) {
+ } else if (CommonRC == &X86::GR8_ABCD_LRegClass) {
Opc = X86::MOV8rr;
+ } 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 (CommonRC == &X86::GR32_NOREXRegClass) {
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;
+ // 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_ABCDRegClass) {
Opc = X86::MOV32mr;
} else if (RC == &X86::GR16_ABCDRegClass) {
Opc = X86::MOV16mr;
- } else if (RC == &X86::GR8_ABCDRegClass) {
+ } 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)
- .addReg(SrcReg, false, false, isKill);
+ .addReg(SrcReg, getKillRegState(isKill));
}
void X86InstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
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)
MIB.addOperand(Addr[i]);
- MIB.addReg(SrcReg, false, false, isKill);
+ MIB.addReg(SrcReg, getKillRegState(isKill));
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;
+ // 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_ABCDRegClass) {
Opc = X86::MOV32rm;
} else if (RC == &X86::GR16_ABCDRegClass) {
Opc = X86::MOV16rm;
- } else if (RC == &X86::GR8_ABCDRegClass) {
+ } 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)
MachineFunction &MF = *MBB.getParent();
X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
- X86FI->setCalleeSavedFrameSize(CSI.size() * SlotSize);
+ unsigned CalleeFrameSize = 0;
unsigned Opc = is64Bit ? X86::PUSH64r : X86::PUSH32r;
for (unsigned i = CSI.size(); i != 0; --i) {
unsigned Reg = CSI[i-1].getReg();
+ const TargetRegisterClass *RegClass = CSI[i-1].getRegClass();
// Add the callee-saved register as live-in. It's killed at the spill.
MBB.addLiveIn(Reg);
- BuildMI(MBB, MI, DL, get(Opc))
- .addReg(Reg, /*isDef=*/false, /*isImp=*/false, /*isKill=*/true);
+ if (RegClass != &X86::VR128RegClass) {
+ CalleeFrameSize += SlotSize;
+ BuildMI(MBB, MI, DL, get(Opc))
+ .addReg(Reg, RegState::Kill);
+ } else {
+ storeRegToStackSlot(MBB, MI, Reg, true, CSI[i-1].getFrameIdx(), RegClass);
+ }
}
+
+ X86FI->setCalleeSavedFrameSize(CalleeFrameSize);
return true;
}
unsigned Opc = is64Bit ? X86::POP64r : X86::POP32r;
for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
unsigned Reg = CSI[i].getReg();
- BuildMI(MBB, MI, DL, get(Opc), Reg);
+ const TargetRegisterClass *RegClass = CSI[i].getRegClass();
+ if (RegClass != &X86::VR128RegClass) {
+ BuildMI(MBB, MI, DL, get(Opc), Reg);
+ } else {
+ loadRegFromStackSlot(MBB, MI, Reg, CSI[i].getFrameIdx(), RegClass);
+ }
}
return true;
}
MachineInstrBuilder MIB(DataMI);
if (FoldedStore)
- MIB.addReg(Reg, true);
+ MIB.addReg(Reg, RegState::Define);
for (unsigned i = 0, e = BeforeOps.size(); i != e; ++i)
MIB.addOperand(BeforeOps[i]);
if (FoldedLoad)
MIB.addOperand(AfterOps[i]);
for (unsigned i = 0, e = ImpOps.size(); i != e; ++i) {
MachineOperand &MO = ImpOps[i];
- MIB.addReg(MO.getReg(), MO.isDef(), true, MO.isKill(), MO.isDead());
+ MIB.addReg(MO.getReg(),
+ getDefRegState(MO.isDef()) |
+ RegState::Implicit |
+ getKillRegState(MO.isKill()) |
+ getDeadRegState(MO.isDead()));
}
// Change CMP32ri r, 0 back to TEST32rr r, r, etc.
unsigned NewOpc = 0;
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 ((!Is64BitMode || DispForReloc || BaseReg != 0) &&
+ IndexReg.getReg() == 0 &&
+ (BaseReg == 0 || X86RegisterInfo::getX86RegNum(BaseReg) != N86::ESP)) {
if (BaseReg == 0) { // Just a displacement?
// Emit special case [disp32] encoding
++FinalSize;
// Emit the lock opcode prefix as needed.
if (Desc->TSFlags & X86II::LOCK) ++FinalSize;
- // Emit segment overrid opcode prefix as needed.
+ // Emit segment override opcode prefix as needed.
switch (Desc->TSFlags & X86II::SegOvrMask) {
case X86II::FS:
case X86II::GS:
case X86II::T8: // 0F 38
++FinalSize;
break;
- case X86II::TA: // 0F 3A
+ case X86II::TA: // 0F 3A
++FinalSize;
break;
}
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();
+ if (Desc->getOpcode() == X86::LFENCE ||
+ Desc->getOpcode() == X86::MFENCE) {
+ // Special handling of lfence and mfence;
+ FinalSize += sizeRegModRMByte();
+ } else if (Desc->getOpcode() == X86::MONITOR ||
+ Desc->getOpcode() == X86::MWAIT) {
+ // Special handling of monitor and mwait.
+ FinalSize += sizeRegModRMByte() + 1; // +1 for the opcode.
+ } 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++);
projects / oota-llvm.git / blobdiff