#include "X86MachineFunctionInfo.h"
#include "X86Subtarget.h"
#include "X86TargetMachine.h"
+#include "llvm/GlobalVariable.h"
#include "llvm/DerivedTypes.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Target/TargetAsmInfo.h"
-
using namespace llvm;
-// FIXME: This should be some header
-static const int X86AddrNumOperands = 4;
-
namespace {
cl::opt<bool>
NoFusing("disable-spill-fusing",
{ X86::JMP64r, X86::JMP64m, 1 },
{ X86::MOV16ri, X86::MOV16mi, 0 },
{ X86::MOV16rr, X86::MOV16mr, 0 },
- { X86::MOV16to16_, X86::MOV16_mr, 0 },
{ X86::MOV32ri, X86::MOV32mi, 0 },
{ X86::MOV32rr, X86::MOV32mr, 0 },
- { X86::MOV32to32_, X86::MOV32_mr, 0 },
{ X86::MOV64ri32, X86::MOV64mi32, 0 },
{ X86::MOV64rr, X86::MOV64mr, 0 },
{ X86::MOV8ri, X86::MOV8mi, 0 },
{ X86::MOV8rr, X86::MOV8mr, 0 },
+ { X86::MOV8rr_NOREX, X86::MOV8mr_NOREX, 0 },
{ X86::MOVAPDrr, X86::MOVAPDmr, 0 },
{ X86::MOVAPSrr, X86::MOVAPSmr, 0 },
{ X86::MOVDQArr, X86::MOVDQAmr, 0 },
{ X86::Int_UCOMISDrr, X86::Int_UCOMISDrm },
{ X86::Int_UCOMISSrr, X86::Int_UCOMISSrm },
{ X86::MOV16rr, X86::MOV16rm },
- { X86::MOV16to16_, X86::MOV16_rm },
{ X86::MOV32rr, X86::MOV32rm },
- { X86::MOV32to32_, X86::MOV32_rm },
{ X86::MOV64rr, X86::MOV64rm },
{ X86::MOV64toPQIrr, X86::MOVQI2PQIrm },
{ X86::MOV64toSDrr, X86::MOV64toSDrm },
{ X86::MOVZPQILo2PQIrr, X86::MOVZPQILo2PQIrm },
{ X86::MOVZX16rr8, X86::MOVZX16rm8 },
{ X86::MOVZX32rr16, X86::MOVZX32rm16 },
+ { X86::MOVZX32_NOREXrr8, X86::MOVZX32_NOREXrm8 },
{ X86::MOVZX32rr8, X86::MOVZX32rm8 },
{ X86::MOVZX64rr16, X86::MOVZX64rm16 },
{ X86::MOVZX64rr32, X86::MOVZX64rm32 },
default:
return false;
case X86::MOV8rr:
+ case X86::MOV8rr_NOREX:
case X86::MOV16rr:
case X86::MOV32rr:
case X86::MOV64rr:
- case X86::MOV16to16_:
- case X86::MOV32to32_:
case X86::MOVSSrr:
case X86::MOVSDrr:
default: break;
case X86::MOV8rm:
case X86::MOV16rm:
- case X86::MOV16_rm:
case X86::MOV32rm:
- case X86::MOV32_rm:
case X86::MOV64rm:
case X86::LD_Fp64m:
case X86::MOVSSrm:
default: break;
case X86::MOV8mr:
case X86::MOV16mr:
- case X86::MOV16_mr:
case X86::MOV32mr:
- case X86::MOV32_mr:
case X86::MOV64mr:
case X86::ST_FpP64m:
case X86::MOVSSmr:
static inline bool isGVStub(GlobalValue *GV, X86TargetMachine &TM) {
return TM.getSubtarget<X86Subtarget>().GVRequiresExtraLoad(GV, TM, false);
}
+
+/// CanRematLoadWithDispOperand - Return true if a load with the specified
+/// operand is a candidate for remat: for this to be true we need to know that
+/// the load will always return the same value, even if moved.
+static bool CanRematLoadWithDispOperand(const MachineOperand &MO,
+ X86TargetMachine &TM) {
+ // Loads from constant pool entries can be remat'd.
+ if (MO.isCPI()) return true;
+
+ // We can remat globals in some cases.
+ if (MO.isGlobal()) {
+ // If this is a load of a stub, not of the global, we can remat it. This
+ // access will always return the address of the global.
+ if (isGVStub(MO.getGlobal(), TM))
+ return true;
+
+ // If the global itself is constant, we can remat the load.
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(MO.getGlobal()))
+ if (GV->isConstant())
+ return true;
+ }
+ return false;
+}
bool
X86InstrInfo::isReallyTriviallyReMaterializable(const MachineInstr *MI) const {
default: break;
case X86::MOV8rm:
case X86::MOV16rm:
- case X86::MOV16_rm:
case X86::MOV32rm:
- case X86::MOV32_rm:
case X86::MOV64rm:
case X86::LD_Fp64m:
case X86::MOVSSrm:
if (MI->getOperand(1).isReg() &&
MI->getOperand(2).isImm() &&
MI->getOperand(3).isReg() && MI->getOperand(3).getReg() == 0 &&
- (MI->getOperand(4).isCPI() ||
- (MI->getOperand(4).isGlobal() &&
- isGVStub(MI->getOperand(4).getGlobal(), TM)))) {
+ CanRematLoadWithDispOperand(MI->getOperand(4), TM)) {
unsigned BaseReg = MI->getOperand(1).getReg();
- if (BaseReg == 0)
+ if (BaseReg == 0 || BaseReg == X86::RIP)
return true;
// Allow re-materialization of PIC load.
if (!ReMatPICStubLoad && MI->getOperand(4).isGlobal())
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;
}
assert(MI->getNumOperands() >= 2 && "Unknown inc instruction!");
unsigned Opc = MIOpc == X86::INC64r ? X86::LEA64r
: (is64Bit ? X86::LEA64_32r : X86::LEA32r);
- NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
- Src, isKill, 1);
+ NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
+ Src, isKill, 1);
break;
}
case X86::INC16r:
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:
assert(MI->getNumOperands() >= 2 && "Unknown dec instruction!");
unsigned Opc = MIOpc == X86::DEC64r ? X86::LEA64r
: (is64Bit ? X86::LEA64_32r : X86::LEA32r);
- NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
- Src, isKill, -1);
+ NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
+ Src, isKill, -1);
break;
}
case X86::DEC16r:
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);
case X86::ADD64ri8:
assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");
if (MI->getOperand(2).isImm())
- NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA64r))
- .addReg(Dest, true, false, false, isDead),
- Src, isKill, MI->getOperand(2).getImm());
+ NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA64r))
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
+ Src, isKill, MI->getOperand(2).getImm());
break;
case X86::ADD32ri:
case X86::ADD32ri8:
assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");
if (MI->getOperand(2).isImm()) {
unsigned Opc = is64Bit ? X86::LEA64_32r : X86::LEA32r;
- NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
- .addReg(Dest, true, false, false, isDead),
- Src, isKill, MI->getOperand(2).getImm());
+ NewMI = addLeaRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
+ .addReg(Dest, RegState::Define |
+ getDeadRegState(isDead)),
+ Src, isKill, MI->getOperand(2).getImm());
}
break;
case X86::ADD16ri:
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);
}
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;
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.
return Count;
}
+/// isHReg - Test if the given register is a physical h register.
+static bool isHReg(unsigned Reg) {
+ return X86::GR8_ABCD_HRegClass.contains(Reg);
+}
+
bool X86InstrInfo::copyRegToReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned DestReg, unsigned SrcReg,
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) {
+ } 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 (CommonRC == &X86::GR64_ABCDRegClass) {
+ Opc = X86::MOV64rr;
+ } else if (CommonRC == &X86::GR32_ABCDRegClass) {
+ Opc = X86::MOV32rr;
+ } else if (CommonRC == &X86::GR16_ABCDRegClass) {
+ Opc = X86::MOV16rr;
+ } else if (CommonRC == &X86::GR8_ABCD_LRegClass) {
Opc = X86::MOV8rr;
- } else if (DestRC == &X86::GR32_RegClass) {
- Opc = X86::MOV32_rr;
- } else if (DestRC == &X86::GR16_RegClass) {
- Opc = X86::MOV16_rr;
- } else if (DestRC == &X86::RFP32RegClass) {
+ } 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) {
+ Opc = X86::MOV32rr;
+ } else if (CommonRC == &X86::GR16_NOREXRegClass) {
+ Opc = X86::MOV16rr;
+ } else if (CommonRC == &X86::GR8_NOREXRegClass) {
+ Opc = X86::MOV8rr;
+ } 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 true;
}
}
-
+
// Moving from ST(0) turns into FpGET_ST0_32 etc.
if (SrcRC == &X86::RSTRegClass) {
// Copying from ST(0)/ST(1).
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) {
+ // 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_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) {
+ Opc = X86::MOV32mr;
+ } else if (RC == &X86::GR16_NOREXRegClass) {
+ Opc = X86::MOV16mr;
+ } else if (RC == &X86::GR8_NOREXRegClass) {
Opc = X86::MOV8mr;
- } else if (RC == &X86::GR32_RegClass) {
- Opc = X86::MOV32_mr;
- } else if (RC == &X86::GR16_RegClass) {
- Opc = X86::MOV16_mr;
} else if (RC == &X86::RFP80RegClass) {
Opc = X86::ST_FpP80m; // pops
} else if (RC == &X86::RFP64RegClass) {
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) {
+ // 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_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) {
+ Opc = X86::MOV32rm;
+ } else if (RC == &X86::GR16_NOREXRegClass) {
+ Opc = X86::MOV16rm;
+ } else if (RC == &X86::GR8_NOREXRegClass) {
Opc = X86::MOV8rm;
- } else if (RC == &X86::GR32_RegClass) {
- Opc = X86::MOV32_rm;
- } else if (RC == &X86::GR16_RegClass) {
- Opc = X86::MOV16_rm;
} else if (RC == &X86::RFP80RegClass) {
Opc = X86::LD_Fp80m;
} else if (RC == &X86::RFP64RegClass) {
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;
}
for (unsigned i = 0; i != NumAddrOps; ++i)
MIB.addOperand(MOs[i]);
if (NumAddrOps < 4) // FrameIndex only
- MIB.addImm(1).addReg(0).addImm(0);
+ addOffset(MIB, 0);
// Loop over the rest of the ri operands, converting them over.
unsigned NumOps = MI->getDesc().getNumOperands()-2;
for (unsigned i = 0; i != NumAddrOps; ++i)
MIB.addOperand(MOs[i]);
if (NumAddrOps < 4) // FrameIndex only
- MIB.addImm(1).addReg(0).addImm(0);
+ addOffset(MIB, 0);
} else {
MIB.addOperand(MO);
}
for (unsigned i = 0; i != NumAddrOps; ++i)
MIB.addOperand(MOs[i]);
if (NumAddrOps < 4) // FrameIndex only
- MIB.addImm(1).addReg(0).addImm(0);
+ addOffset(MIB, 0);
return MIB.addImm(0);
}
} else if (Ops.size() != 1)
return NULL;
- SmallVector<MachineOperand,4> MOs;
+ SmallVector<MachineOperand,X86AddrNumOperands> MOs;
if (LoadMI->getOpcode() == X86::V_SET0 ||
LoadMI->getOpcode() == X86::V_SETALLONES) {
// Folding a V_SET0 or V_SETALLONES as a load, to ease register pressure.
MOs.push_back(MachineOperand::CreateImm(1));
MOs.push_back(MachineOperand::CreateReg(0, false));
MOs.push_back(MachineOperand::CreateCPI(CPI, 0));
+ MOs.push_back(MachineOperand::CreateReg(0, false));
} else {
// Folding a normal load. Just copy the load's address operands.
unsigned NumOps = LoadMI->getDesc().getNumOperands();
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);
}
case X86II::MRM4m: case X86II::MRM5m:
case X86II::MRM6m: case X86II::MRM7m:
case X86II::MRMDestMem: {
- unsigned e = isTwoAddr ? 5 : 4;
+ unsigned e = (isTwoAddr ? X86AddrNumOperands+1 : X86AddrNumOperands);
i = isTwoAddr ? 1 : 0;
if (NumOps > e && isX86_64ExtendedReg(MI.getOperand(e)))
REX |= 1 << 2;
unsigned BaseReg = Base.getReg();
// Is a SIB byte needed?
- if (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!");
FinalSize += sizeConstant(X86InstrInfo::sizeOfImm(Desc));
break;
}
- case X86::TLS_tp:
- case X86::TLS_gs_ri:
- FinalSize += 2;
- FinalSize += sizeGlobalAddress(false);
- break;
}
CurOp = NumOps;
break;
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++);
bool IsPIC = (TM.getRelocationModel() == Reloc::PIC_);
bool Is64BitMode = TM.getSubtargetImpl()->is64Bit();
unsigned Size = GetInstSizeWithDesc(*MI, &Desc, IsPIC, Is64BitMode);
- if (Desc.getOpcode() == X86::MOVPC32r) {
+ if (Desc.getOpcode() == X86::MOVPC32r)
Size += GetInstSizeWithDesc(*MI, &get(X86::POP32r), IsPIC, Is64BitMode);
- }
return Size;
}
const TargetInstrInfo *TII = TM.getInstrInfo();
// Operand of MovePCtoStack is completely ignored by asm printer. It's
// only used in JIT code emission as displacement to pc.
- BuildMI(FirstMBB, MBBI, DL, TII->get(X86::MOVPC32r), PC)
- .addImm(0);
+ BuildMI(FirstMBB, MBBI, DL, TII->get(X86::MOVPC32r), PC).addImm(0);
// If we're using vanilla 'GOT' PIC style, we should use relative addressing
- // not to pc, but to _GLOBAL_ADDRESS_TABLE_ external
+ // not to pc, but to _GLOBAL_OFFSET_TABLE_ external.
if (TM.getRelocationModel() == Reloc::PIC_ &&
TM.getSubtarget<X86Subtarget>().isPICStyleGOT()) {
- GlobalBaseReg =
- RegInfo.createVirtualRegister(X86::GR32RegisterClass);
+ GlobalBaseReg = RegInfo.createVirtualRegister(X86::GR32RegisterClass);
+ // Generate addl $__GLOBAL_OFFSET_TABLE_ + [.-piclabel], %some_register
BuildMI(FirstMBB, MBBI, DL, TII->get(X86::ADD32ri), GlobalBaseReg)
- .addReg(PC).addExternalSymbol("_GLOBAL_OFFSET_TABLE_");
+ .addReg(PC).addExternalSymbol("_GLOBAL_OFFSET_TABLE_", 0,
+ X86II::MO_GOT_ABSOLUTE_ADDRESS);
} else {
GlobalBaseReg = PC;
}