MachineInstr *
LanaiInstrInfo::optimizeSelect(MachineInstr &MI,
SmallPtrSetImpl<MachineInstr *> &SeenMIs,
bool PreferFalse) const {
assert(MI.getOpcode() == Lanai::SELECT && "unknown select instruction");
MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
MachineInstr *DefMI = canFoldIntoSelect(MI.getOperand(1).getReg(), MRI, this);
bool Invert = !DefMI;
if (!DefMI)
DefMI = canFoldIntoSelect(MI.getOperand(2).getReg(), MRI, this);
if (!DefMI)
return nullptr;
// Find new register class to use.
MachineOperand FalseReg = MI.getOperand(Invert ? 1 : 2);
unsigned DestReg = MI.getOperand(0).getReg();
const TargetRegisterClass *PreviousClass = MRI.getRegClass(FalseReg.getReg());
if (!MRI.constrainRegClass(DestReg, PreviousClass))
return nullptr;
// Create a new predicated version of DefMI.
MachineInstrBuilder NewMI =
BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), DefMI->getDesc(), DestReg);
// Copy all the DefMI operands, excluding its (null) predicate.
const MCInstrDesc &DefDesc = DefMI->getDesc();
for (unsigned i = 1, e = DefDesc.getNumOperands();
i != e && !DefDesc.OpInfo[i].isPredicate(); ++i)
NewMI.addOperand(DefMI->getOperand(i));
unsigned CondCode = MI.getOperand(3).getImm();
if (Invert)
NewMI.addImm(getOppositeCondition(LPCC::CondCode(CondCode)));
else
NewMI.addImm(CondCode);
NewMI.copyImplicitOps(MI);
// The output register value when the predicate is false is an implicit
// register operand tied to the first def. The tie makes the register
// allocator ensure the FalseReg is allocated the same register as operand 0.
FalseReg.setImplicit();
NewMI.addOperand(FalseReg);
NewMI->tieOperands(0, NewMI->getNumOperands() - 1);
// Update SeenMIs set: register newly created MI and erase removed DefMI.
SeenMIs.insert(NewMI);
SeenMIs.erase(DefMI);
// If MI is inside a loop, and DefMI is outside the loop, then kill flags on
// DefMI would be invalid when transferred inside the loop. Checking for a
// loop is expensive, but at least remove kill flags if they are in different
// BBs.
if (DefMI->getParent() != MI.getParent())
NewMI->clearKillInfo();
// The caller will erase MI, but not DefMI.
DefMI->eraseFromParent();
return NewMI;
}
/// ExpandVLD - Translate VLD pseudo instructions with Q, QQ or QQQQ register
/// operands to real VLD instructions with D register operands.
void ARMExpandPseudo::ExpandVLD(MachineBasicBlock::iterator &MBBI) {
MachineInstr &MI = *MBBI;
MachineBasicBlock &MBB = *MI.getParent();
const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
assert(TableEntry && TableEntry->IsLoad && "NEONLdStTable lookup failed");
NEONRegSpacing RegSpc = TableEntry->RegSpacing;
unsigned NumRegs = TableEntry->NumRegs;
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
TII->get(TableEntry->RealOpc));
unsigned OpIdx = 0;
bool DstIsDead = MI.getOperand(OpIdx).isDead();
unsigned DstReg = MI.getOperand(OpIdx++).getReg();
unsigned D0, D1, D2, D3;
GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3);
MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead))
.addReg(D1, RegState::Define | getDeadRegState(DstIsDead));
if (NumRegs > 2)
MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead));
if (NumRegs > 3)
MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead));
if (TableEntry->HasWriteBack)
MIB.addOperand(MI.getOperand(OpIdx++));
// Copy the addrmode6 operands.
MIB.addOperand(MI.getOperand(OpIdx++));
MIB.addOperand(MI.getOperand(OpIdx++));
// Copy the am6offset operand.
if (TableEntry->HasWriteBack)
MIB.addOperand(MI.getOperand(OpIdx++));
// For an instruction writing double-spaced subregs, the pseudo instruction
// has an extra operand that is a use of the super-register. Record the
// operand index and skip over it.
unsigned SrcOpIdx = 0;
if (RegSpc == EvenDblSpc || RegSpc == OddDblSpc)
SrcOpIdx = OpIdx++;
// Copy the predicate operands.
MIB.addOperand(MI.getOperand(OpIdx++));
MIB.addOperand(MI.getOperand(OpIdx++));
// Copy the super-register source operand used for double-spaced subregs over
// to the new instruction as an implicit operand.
if (SrcOpIdx != 0) {
MachineOperand MO = MI.getOperand(SrcOpIdx);
MO.setImplicit(true);
MIB.addOperand(MO);
}
// Add an implicit def for the super-register.
MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead));
TransferImpOps(MI, MIB, MIB);
MI.eraseFromParent();
}
/// TransferImpOps - Transfer implicit operands on the pseudo instruction to
/// the instructions created from the expansion.
void ARMExpandPseudo::TransferImpOps(MachineInstr &OldMI,
MachineInstrBuilder &UseMI,
MachineInstrBuilder &DefMI) {
const TargetInstrDesc &Desc = OldMI.getDesc();
for (unsigned i = Desc.getNumOperands(), e = OldMI.getNumOperands();
i != e; ++i) {
const MachineOperand &MO = OldMI.getOperand(i);
assert(MO.isReg() && MO.getReg());
if (MO.isUse())
UseMI.addOperand(MO);
else
DefMI.addOperand(MO);
}
}
void HexagonEarlyIfConversion::predicateInstr(MachineBasicBlock *ToB,
MachineBasicBlock::iterator At, MachineInstr *MI,
unsigned PredR, bool IfTrue) {
DebugLoc DL;
if (At != ToB->end())
DL = At->getDebugLoc();
else if (!ToB->empty())
DL = ToB->back().getDebugLoc();
unsigned Opc = MI->getOpcode();
if (isPredicableStore(MI)) {
unsigned COpc = getCondStoreOpcode(Opc, IfTrue);
assert(COpc);
MachineInstrBuilder MIB = BuildMI(*ToB, At, DL, HII->get(COpc));
MachineInstr::mop_iterator MOI = MI->operands_begin();
if (HII->isPostIncrement(*MI)) {
MIB.addOperand(*MOI);
++MOI;
}
MIB.addReg(PredR);
for (const MachineOperand &MO : make_range(MOI, MI->operands_end()))
MIB.addOperand(MO);
// Set memory references.
MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
MIB.setMemRefs(MMOBegin, MMOEnd);
MI->eraseFromParent();
return;
}
if (Opc == Hexagon::J2_jump) {
MachineBasicBlock *TB = MI->getOperand(0).getMBB();
const MCInstrDesc &D = HII->get(IfTrue ? Hexagon::J2_jumpt
: Hexagon::J2_jumpf);
BuildMI(*ToB, At, DL, D)
.addReg(PredR)
.addMBB(TB);
MI->eraseFromParent();
return;
}
// Print the offending instruction unconditionally as we are about to
// abort.
dbgs() << *MI;
llvm_unreachable("Unexpected instruction");
}
void IA64InstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
bool isKill,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc = 0;
if (RC == IA64::FPRegisterClass) {
Opc = IA64::STF8;
} else if (RC == IA64::GRRegisterClass) {
Opc = IA64::ST8;
} else if (RC == IA64::PRRegisterClass) {
Opc = IA64::ST1;
} else {
assert(0 &&
"sorry, I don't know how to store this sort of reg\n");
}
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);
NewMIs.push_back(MIB);
return;
}
/// \brief Replace loop instructions with the constant extended version.
void HexagonFixupHwLoops::useExtLoopInstr(MachineFunction &MF,
MachineBasicBlock::iterator &MII) {
const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
MachineBasicBlock *MBB = MII->getParent();
DebugLoc DL = MII->getDebugLoc();
MachineInstrBuilder MIB;
unsigned newOp;
switch (MII->getOpcode()) {
case Hexagon::J2_loop0r:
newOp = Hexagon::J2_loop0rext;
break;
case Hexagon::J2_loop0i:
newOp = Hexagon::J2_loop0iext;
break;
case Hexagon::J2_loop1r:
newOp = Hexagon::J2_loop1rext;
break;
case Hexagon::J2_loop1i:
newOp = Hexagon::J2_loop1iext;
break;
default:
llvm_unreachable("Invalid Hardware Loop Instruction.");
}
MIB = BuildMI(*MBB, MII, DL, TII->get(newOp));
for (unsigned i = 0; i < MII->getNumOperands(); ++i)
MIB.addOperand(MII->getOperand(i));
}
/// Insert a conditional branch at the end of \p MBB to \p NewDestBB, using the
/// inverse condition of branch \p OldBr.
/// \returns The number of bytes added to the block.
unsigned AArch64BranchRelaxation::insertInvertedConditionalBranch(
MachineBasicBlock &SrcMBB,
MachineBasicBlock::iterator InsPt,
const DebugLoc &DL,
const MachineInstr &OldBr,
MachineBasicBlock &NewDestBB) const {
unsigned OppositeCondOpc = getOppositeConditionOpcode(OldBr.getOpcode());
MachineInstrBuilder MIB =
BuildMI(SrcMBB, InsPt, DL, TII->get(OppositeCondOpc))
.addOperand(OldBr.getOperand(0));
unsigned Opc = OldBr.getOpcode();
if (Opc == AArch64::TBZW || Opc == AArch64::TBNZW ||
Opc == AArch64::TBZX || Opc == AArch64::TBNZX)
MIB.addOperand(OldBr.getOperand(1));
if (OldBr.getOpcode() == AArch64::Bcc)
invertBccCondition(*MIB);
MIB.addMBB(&NewDestBB);
return TII->getInstSizeInBytes(*MIB);
}
/// Generate a predicated version of MI (where the condition is given via
/// PredR and Cond) at the point indicated by Where.
void HexagonExpandCondsets::predicateAt(const MachineOperand &DefOp,
MachineInstr &MI,
MachineBasicBlock::iterator Where,
const MachineOperand &PredOp, bool Cond,
std::set<unsigned> &UpdRegs) {
// The problem with updating live intervals is that we can move one def
// past another def. In particular, this can happen when moving an A2_tfrt
// over an A2_tfrf defining the same register. From the point of view of
// live intervals, these two instructions are two separate definitions,
// and each one starts another live segment. LiveIntervals's "handleMove"
// does not allow such moves, so we need to handle it ourselves. To avoid
// invalidating liveness data while we are using it, the move will be
// implemented in 4 steps: (1) add a clone of the instruction MI at the
// target location, (2) update liveness, (3) delete the old instruction,
// and (4) update liveness again.
MachineBasicBlock &B = *MI.getParent();
DebugLoc DL = Where->getDebugLoc(); // "Where" points to an instruction.
unsigned Opc = MI.getOpcode();
unsigned PredOpc = HII->getCondOpcode(Opc, !Cond);
MachineInstrBuilder MB = BuildMI(B, Where, DL, HII->get(PredOpc));
unsigned Ox = 0, NP = MI.getNumOperands();
// Skip all defs from MI first.
while (Ox < NP) {
MachineOperand &MO = MI.getOperand(Ox);
if (!MO.isReg() || !MO.isDef())
break;
Ox++;
}
// Add the new def, then the predicate register, then the rest of the
// operands.
MB.addReg(DefOp.getReg(), getRegState(DefOp), DefOp.getSubReg());
MB.addReg(PredOp.getReg(), PredOp.isUndef() ? RegState::Undef : 0,
PredOp.getSubReg());
while (Ox < NP) {
MachineOperand &MO = MI.getOperand(Ox);
if (!MO.isReg() || !MO.isImplicit())
MB.addOperand(MO);
Ox++;
}
MachineFunction &MF = *B.getParent();
MachineInstr::mmo_iterator I = MI.memoperands_begin();
unsigned NR = std::distance(I, MI.memoperands_end());
MachineInstr::mmo_iterator MemRefs = MF.allocateMemRefsArray(NR);
for (unsigned i = 0; i < NR; ++i)
MemRefs[i] = *I++;
MB.setMemRefs(MemRefs, MemRefs+NR);
MachineInstr *NewI = MB;
NewI->clearKillInfo();
LIS->InsertMachineInstrInMaps(*NewI);
for (auto &Op : NewI->operands())
if (Op.isReg())
UpdRegs.insert(Op.getReg());
}
/// ExpandVST - Translate VST pseudo instructions with Q, QQ or QQQQ register
/// operands to real VST instructions with D register operands.
void ARMExpandPseudo::ExpandVST(MachineBasicBlock::iterator &MBBI) {
MachineInstr &MI = *MBBI;
MachineBasicBlock &MBB = *MI.getParent();
const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode());
assert(TableEntry && !TableEntry->IsLoad && "NEONLdStTable lookup failed");
NEONRegSpacing RegSpc = TableEntry->RegSpacing;
unsigned NumRegs = TableEntry->NumRegs;
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(),
TII->get(TableEntry->RealOpc));
unsigned OpIdx = 0;
if (TableEntry->HasWriteBack)
MIB.addOperand(MI.getOperand(OpIdx++));
// Copy the addrmode6 operands.
MIB.addOperand(MI.getOperand(OpIdx++));
MIB.addOperand(MI.getOperand(OpIdx++));
// Copy the am6offset operand.
if (TableEntry->HasWriteBack)
MIB.addOperand(MI.getOperand(OpIdx++));
bool SrcIsKill = MI.getOperand(OpIdx).isKill();
unsigned SrcReg = MI.getOperand(OpIdx++).getReg();
unsigned D0, D1, D2, D3;
GetDSubRegs(SrcReg, RegSpc, TRI, D0, D1, D2, D3);
MIB.addReg(D0).addReg(D1);
if (NumRegs > 2)
MIB.addReg(D2);
if (NumRegs > 3)
MIB.addReg(D3);
// Copy the predicate operands.
MIB.addOperand(MI.getOperand(OpIdx++));
MIB.addOperand(MI.getOperand(OpIdx++));
if (SrcIsKill)
// Add an implicit kill for the super-reg.
(*MIB).addRegisterKilled(SrcReg, TRI, true);
TransferImpOps(MI, MIB, MIB);
MI.eraseFromParent();
}
bool HexagonOptAddrMode::changeStore(MachineInstr *OldMI, MachineOperand ImmOp,
unsigned ImmOpNum) {
bool Changed = false;
unsigned OpStart;
unsigned OpEnd = OldMI->getNumOperands();
MachineBasicBlock *BB = OldMI->getParent();
auto UsePos = MachineBasicBlock::iterator(OldMI);
MachineBasicBlock::instr_iterator InsertPt = UsePos.getInstrIterator();
++InsertPt;
MachineInstrBuilder MIB;
if (ImmOpNum == 0) {
if (HII->getAddrMode(OldMI) == HexagonII::BaseRegOffset) {
short NewOpCode = HII->getBaseWithLongOffset(OldMI);
assert(NewOpCode >= 0 && "Invalid New opcode\n");
MIB = BuildMI(*BB, InsertPt, OldMI->getDebugLoc(), HII->get(NewOpCode));
MIB.addOperand(OldMI->getOperand(1));
MIB.addOperand(OldMI->getOperand(2));
MIB.addOperand(ImmOp);
MIB.addOperand(OldMI->getOperand(3));
OpStart = 4;
} else if (HII->getAddrMode(OldMI) == HexagonII::BaseImmOffset) {
short NewOpCode = HII->getAbsoluteForm(OldMI);
assert(NewOpCode >= 0 && "Invalid New opcode\n");
MIB = BuildMI(*BB, InsertPt, OldMI->getDebugLoc(), HII->get(NewOpCode));
const GlobalValue *GV = ImmOp.getGlobal();
int64_t Offset = ImmOp.getOffset() + OldMI->getOperand(1).getImm();
MIB.addGlobalAddress(GV, Offset, ImmOp.getTargetFlags());
MIB.addOperand(OldMI->getOperand(2));
OpStart = 3;
}
Changed = true;
DEBUG(dbgs() << "[Changing]: " << *OldMI << "\n");
DEBUG(dbgs() << "[TO]: " << MIB << "\n");
} else if (ImmOpNum == 1 && OldMI->getOperand(2).getImm() == 0) {
short NewOpCode = HII->xformRegToImmOffset(OldMI);
assert(NewOpCode >= 0 && "Invalid New opcode\n");
MIB = BuildMI(*BB, InsertPt, OldMI->getDebugLoc(), HII->get(NewOpCode));
MIB.addOperand(OldMI->getOperand(0));
MIB.addOperand(ImmOp);
MIB.addOperand(OldMI->getOperand(1));
OpStart = 2;
Changed = true;
DEBUG(dbgs() << "[Changing]: " << *OldMI << "\n");
DEBUG(dbgs() << "[TO]: " << MIB << "\n");
}
if (Changed)
for (unsigned i = OpStart; i < OpEnd; ++i)
MIB.addOperand(OldMI->getOperand(i));
return Changed;
}
unsigned
AArch64InstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
const SmallVectorImpl<MachineOperand> &Cond,
DebugLoc DL) const {
if (FBB == 0 && Cond.empty()) {
BuildMI(&MBB, DL, get(AArch64::Bimm)).addMBB(TBB);
return 1;
} else if (FBB == 0) {
MachineInstrBuilder MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
for (int i = 1, e = Cond.size(); i != e; ++i)
MIB.addOperand(Cond[i]);
MIB.addMBB(TBB);
return 1;
}
MachineInstrBuilder MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
for (int i = 1, e = Cond.size(); i != e; ++i)
MIB.addOperand(Cond[i]);
MIB.addMBB(TBB);
BuildMI(&MBB, DL, get(AArch64::Bimm)).addMBB(FBB);
return 2;
}
/// ExpandVTBL - Translate VTBL and VTBX pseudo instructions with Q or QQ
/// register operands to real instructions with D register operands.
void ARMExpandPseudo::ExpandVTBL(MachineBasicBlock::iterator &MBBI,
unsigned Opc, bool IsExt, unsigned NumRegs) {
MachineInstr &MI = *MBBI;
MachineBasicBlock &MBB = *MI.getParent();
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc));
unsigned OpIdx = 0;
// Transfer the destination register operand.
MIB.addOperand(MI.getOperand(OpIdx++));
if (IsExt)
MIB.addOperand(MI.getOperand(OpIdx++));
bool SrcIsKill = MI.getOperand(OpIdx).isKill();
unsigned SrcReg = MI.getOperand(OpIdx++).getReg();
unsigned D0, D1, D2, D3;
GetDSubRegs(SrcReg, SingleSpc, TRI, D0, D1, D2, D3);
MIB.addReg(D0).addReg(D1);
if (NumRegs > 2)
MIB.addReg(D2);
if (NumRegs > 3)
MIB.addReg(D3);
// Copy the other source register operand.
MIB.addOperand(MI.getOperand(OpIdx++));
// Copy the predicate operands.
MIB.addOperand(MI.getOperand(OpIdx++));
MIB.addOperand(MI.getOperand(OpIdx++));
if (SrcIsKill)
// Add an implicit kill for the super-reg.
(*MIB).addRegisterKilled(SrcReg, TRI, true);
TransferImpOps(MI, MIB, MIB);
MI.eraseFromParent();
}
void SparcInstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc = 0;
if (RC == SP::IntRegsRegisterClass)
Opc = SP::LDri;
else if (RC == SP::FPRegsRegisterClass)
Opc = SP::LDFri;
else if (RC == SP::DFPRegsRegisterClass)
Opc = SP::LDDFri;
else
assert(0 && "Can't load this register");
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc), DestReg);
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB.addOperand(Addr[i]);
NewMIs.push_back(MIB);
return;
}
void MipsInstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc;
if (RC == Mips::CPURegsRegisterClass)
Opc = Mips::LW;
else if (RC == Mips::FGR32RegisterClass)
Opc = Mips::LWC1;
else {
assert(RC == Mips::AFGR64RegisterClass);
Opc = Mips::LDC1;
}
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc), DestReg);
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB.addOperand(Addr[i]);
NewMIs.push_back(MIB);
return;
}
void SparcInstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
bool isKill,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc = 0;
DebugLoc DL = DebugLoc::getUnknownLoc();
if (RC == SP::IntRegsRegisterClass)
Opc = SP::STri;
else if (RC == SP::FPRegsRegisterClass)
Opc = SP::STFri;
else if (RC == SP::DFPRegsRegisterClass)
Opc = SP::STDFri;
else
assert(0 && "Can't load this register");
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);
NewMIs.push_back(MIB);
return;
}
void MipsInstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
bool isKill, SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC, SmallVectorImpl<MachineInstr*> &NewMIs) const
{
unsigned Opc;
if (RC == Mips::CPURegsRegisterClass)
Opc = Mips::SW;
else if (RC == Mips::FGR32RegisterClass)
Opc = Mips::SWC1;
else {
assert(RC == Mips::AFGR64RegisterClass);
Opc = Mips::SDC1;
}
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc))
.addReg(SrcReg, getKillRegState(isKill));
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB.addOperand(Addr[i]);
NewMIs.push_back(MIB);
return;
}
void IA64InstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const {
unsigned Opc = 0;
if (RC == IA64::FPRegisterClass) {
Opc = IA64::LDF8;
} else if (RC == IA64::GRRegisterClass) {
Opc = IA64::LD8;
} else if (RC == IA64::PRRegisterClass) {
Opc = IA64::LD1;
} else {
assert(0 &&
"sorry, I don't know how to load this sort of reg\n");
}
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc), DestReg);
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB.addOperand(Addr[i]);
NewMIs.push_back(MIB);
return;
}
/*!
\note We are really pessimistic here about what kind of a load we're doing.
*/
void SPUInstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs)
const {
cerr << "loadRegToAddr() invoked!\n";
abort();
if (Addr[0].isFI()) {
/* do what loadRegFromStackSlot does here... */
} else {
unsigned Opc = 0;
if (RC == SPU::R8CRegisterClass) {
/* do brilliance here */
} else if (RC == SPU::R16CRegisterClass) {
/* Opc = PPC::LWZ; */
} else if (RC == SPU::R32CRegisterClass) {
/* Opc = PPC::LD; */
} else if (RC == SPU::R32FPRegisterClass) {
/* Opc = PPC::LFD; */
} else if (RC == SPU::R64FPRegisterClass) {
/* Opc = PPC::LFS; */
} else if (RC == SPU::VECREGRegisterClass) {
/* Opc = PPC::LVX; */
} else if (RC == SPU::GPRCRegisterClass) {
/* Opc = something else! */
} else {
assert(0 && "Unknown regclass!");
abort();
}
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc), DestReg);
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB.addOperand(Addr[i]);
NewMIs.push_back(MIB);
}
}
void SPUInstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
bool isKill,
SmallVectorImpl<MachineOperand> &Addr,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs) const {
cerr << "storeRegToAddr() invoked!\n";
abort();
if (Addr[0].isFI()) {
/* do what storeRegToStackSlot does here */
} else {
unsigned Opc = 0;
if (RC == SPU::GPRCRegisterClass) {
/* Opc = PPC::STW; */
} else if (RC == SPU::R16CRegisterClass) {
/* Opc = PPC::STD; */
} else if (RC == SPU::R32CRegisterClass) {
/* Opc = PPC::STFD; */
} else if (RC == SPU::R32FPRegisterClass) {
/* Opc = PPC::STFD; */
} else if (RC == SPU::R64FPRegisterClass) {
/* Opc = PPC::STFS; */
} else if (RC == SPU::VECREGRegisterClass) {
/* Opc = PPC::STVX; */
} else {
assert(0 && "Unknown regclass!");
abort();
}
DebugLoc DL = DebugLoc::getUnknownLoc();
MachineInstrBuilder MIB = BuildMI(MF, DL, get(Opc))
.addReg(SrcReg, false, false, isKill);
for (unsigned i = 0, e = Addr.size(); i != e; ++i)
MIB.addOperand(Addr[i]);
NewMIs.push_back(MIB);
}
}
/// fixupConditionalBranch - Fix up a conditional branch whose destination is
/// too far away to fit in its displacement field. It is converted to an inverse
/// conditional branch + an unconditional branch to the destination.
bool AArch64BranchRelaxation::fixupConditionalBranch(MachineInstr *MI) {
MachineBasicBlock *DestBB = getDestBlock(MI);
// Add an unconditional branch to the destination and invert the branch
// condition to jump over it:
// tbz L1
// =>
// tbnz L2
// b L1
// L2:
// If the branch is at the end of its MBB and that has a fall-through block,
// direct the updated conditional branch to the fall-through block. Otherwise,
// split the MBB before the next instruction.
MachineBasicBlock *MBB = MI->getParent();
MachineInstr *BMI = &MBB->back();
bool NeedSplit = (BMI != MI) || !BBHasFallthrough(MBB);
if (BMI != MI) {
if (std::next(MachineBasicBlock::iterator(MI)) ==
std::prev(MBB->getLastNonDebugInstr()) &&
BMI->getOpcode() == AArch64::B) {
// Last MI in the BB is an unconditional branch. Can we simply invert the
// condition and swap destinations:
// beq L1
// b L2
// =>
// bne L2
// b L1
MachineBasicBlock *NewDest = BMI->getOperand(0).getMBB();
if (isBlockInRange(MI, NewDest,
getBranchDisplacementBits(MI->getOpcode()))) {
DEBUG(dbgs() << " Invert condition and swap its destination with "
<< *BMI);
BMI->getOperand(0).setMBB(DestBB);
unsigned OpNum = (MI->getOpcode() == AArch64::TBZW ||
MI->getOpcode() == AArch64::TBNZW ||
MI->getOpcode() == AArch64::TBZX ||
MI->getOpcode() == AArch64::TBNZX)
? 2
: 1;
MI->getOperand(OpNum).setMBB(NewDest);
MI->setDesc(TII->get(getOppositeConditionOpcode(MI->getOpcode())));
if (MI->getOpcode() == AArch64::Bcc)
invertBccCondition(MI);
return true;
}
}
}
if (NeedSplit) {
// Analyze the branch so we know how to update the successor lists.
MachineBasicBlock *TBB, *FBB;
SmallVector<MachineOperand, 2> Cond;
TII->AnalyzeBranch(*MBB, TBB, FBB, Cond, false);
MachineBasicBlock *NewBB = splitBlockBeforeInstr(MI);
// No need for the branch to the next block. We're adding an unconditional
// branch to the destination.
int delta = TII->GetInstSizeInBytes(&MBB->back());
BlockInfo[MBB->getNumber()].Size -= delta;
MBB->back().eraseFromParent();
// BlockInfo[SplitBB].Offset is wrong temporarily, fixed below
// Update the successor lists according to the transformation to follow.
// Do it here since if there's no split, no update is needed.
MBB->replaceSuccessor(FBB, NewBB);
NewBB->addSuccessor(FBB);
}
MachineBasicBlock *NextBB = std::next(MachineFunction::iterator(MBB));
DEBUG(dbgs() << " Insert B to BB#" << DestBB->getNumber()
<< ", invert condition and change dest. to BB#"
<< NextBB->getNumber() << "\n");
// Insert a new conditional branch and a new unconditional branch.
MachineInstrBuilder MIB = BuildMI(
MBB, DebugLoc(), TII->get(getOppositeConditionOpcode(MI->getOpcode())))
.addOperand(MI->getOperand(0));
if (MI->getOpcode() == AArch64::TBZW || MI->getOpcode() == AArch64::TBNZW ||
MI->getOpcode() == AArch64::TBZX || MI->getOpcode() == AArch64::TBNZX)
MIB.addOperand(MI->getOperand(1));
if (MI->getOpcode() == AArch64::Bcc)
invertBccCondition(MIB);
MIB.addMBB(NextBB);
BlockInfo[MBB->getNumber()].Size += TII->GetInstSizeInBytes(&MBB->back());
BuildMI(MBB, DebugLoc(), TII->get(AArch64::B)).addMBB(DestBB);
BlockInfo[MBB->getNumber()].Size += TII->GetInstSizeInBytes(&MBB->back());
// Remove the old conditional branch. It may or may not still be in MBB.
BlockInfo[MI->getParent()->getNumber()].Size -= TII->GetInstSizeInBytes(MI);
MI->eraseFromParent();
// Finally, keep the block offsets up to date.
adjustBlockOffsets(*MBB);
return true;
}
void
AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
AArch64MachineFunctionInfo *FuncInfo =
MF.getInfo<AArch64MachineFunctionInfo>();
MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
DebugLoc DL = MBBI->getDebugLoc();
const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
MachineFrameInfo &MFI = *MF.getFrameInfo();
unsigned RetOpcode = MBBI->getOpcode();
// Initial and residual are named for consitency with the prologue. Note that
// in the epilogue, the residual adjustment is executed first.
uint64_t NumInitialBytes = FuncInfo->getInitialStackAdjust();
uint64_t NumResidualBytes = MFI.getStackSize() - NumInitialBytes;
uint64_t ArgumentPopSize = 0;
if (RetOpcode == AArch64::TC_RETURNdi ||
RetOpcode == AArch64::TC_RETURNxi) {
MachineOperand &JumpTarget = MBBI->getOperand(0);
MachineOperand &StackAdjust = MBBI->getOperand(1);
MachineInstrBuilder MIB;
if (RetOpcode == AArch64::TC_RETURNdi) {
MIB = BuildMI(MBB, MBBI, DL, TII.get(AArch64::TAIL_Bimm));
if (JumpTarget.isGlobal()) {
MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
JumpTarget.getTargetFlags());
} else {
assert(JumpTarget.isSymbol() && "unexpected tail call destination");
MIB.addExternalSymbol(JumpTarget.getSymbolName(),
JumpTarget.getTargetFlags());
}
} else {
assert(RetOpcode == AArch64::TC_RETURNxi && JumpTarget.isReg()
&& "Unexpected tail call");
MIB = BuildMI(MBB, MBBI, DL, TII.get(AArch64::TAIL_BRx));
MIB.addReg(JumpTarget.getReg(), RegState::Kill);
}
// Add the extra operands onto the new tail call instruction even though
// they're not used directly (so that liveness is tracked properly etc).
for (unsigned i = 2, e = MBBI->getNumOperands(); i != e; ++i)
MIB->addOperand(MBBI->getOperand(i));
// Delete the pseudo instruction TC_RETURN.
MachineInstr *NewMI = std::prev(MBBI);
MBB.erase(MBBI);
MBBI = NewMI;
// For a tail-call in a callee-pops-arguments environment, some or all of
// the stack may actually be in use for the call's arguments, this is
// calculated during LowerCall and consumed here...
ArgumentPopSize = StackAdjust.getImm();
} else {
// ... otherwise the amount to pop is *all* of the argument space,
// conveniently stored in the MachineFunctionInfo by
// LowerFormalArguments. This will, of course, be zero for the C calling
// convention.
ArgumentPopSize = FuncInfo->getArgumentStackToRestore();
}
assert(NumInitialBytes % 16 == 0 && NumResidualBytes % 16 == 0
&& "refusing to adjust stack by misaligned amt");
// We may need to address callee-saved registers differently, so find out the
// bound on the frame indices.
const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo();
int MinCSFI = 0;
int MaxCSFI = -1;
if (CSI.size()) {
MinCSFI = CSI[0].getFrameIdx();
MaxCSFI = CSI[CSI.size() - 1].getFrameIdx();
}
// The "residual" stack update comes first from this direction and guarantees
// that SP is NumInitialBytes below its value on function entry, either by a
// direct update or restoring it from the frame pointer.
if (NumInitialBytes + ArgumentPopSize != 0) {
emitSPUpdate(MBB, MBBI, DL, TII, AArch64::X16,
NumInitialBytes + ArgumentPopSize);
--MBBI;
}
// MBBI now points to the instruction just past the last callee-saved
// restoration (either RET/B if NumInitialBytes == 0, or the "ADD sp, sp"
// otherwise).
// Now we need to find out where to put the bulk of the stack adjustment
MachineBasicBlock::iterator FirstEpilogue = MBBI;
while (MBBI != MBB.begin()) {
--MBBI;
unsigned FrameOp;
for (FrameOp = 0; FrameOp < MBBI->getNumOperands(); ++FrameOp) {
if (MBBI->getOperand(FrameOp).isFI())
break;
}
// If this instruction doesn't have a frame index we've reached the end of
// the callee-save restoration.
if (FrameOp == MBBI->getNumOperands())
break;
// Likewise if it *is* a local reference, but not to a callee-saved object.
int FrameIdx = MBBI->getOperand(FrameOp).getIndex();
if (FrameIdx < MinCSFI || FrameIdx > MaxCSFI)
break;
FirstEpilogue = MBBI;
}
if (MF.getFrameInfo()->hasVarSizedObjects()) {
int64_t StaticFrameBase;
StaticFrameBase = -(NumInitialBytes + FuncInfo->getFramePointerOffset());
emitRegUpdate(MBB, FirstEpilogue, DL, TII,
AArch64::XSP, AArch64::X29, AArch64::NoRegister,
StaticFrameBase);
} else {
emitSPUpdate(MBB, FirstEpilogue, DL,TII, AArch64::X16, NumResidualBytes);
}
}
void SSACCmpConv::convert(SmallVectorImpl<MachineBasicBlock *> &RemovedBlocks) {
DEBUG(dbgs() << "Merging BB#" << CmpBB->getNumber() << " into BB#"
<< Head->getNumber() << ":\n" << *CmpBB);
// All CmpBB instructions are moved into Head, and CmpBB is deleted.
// Update the CFG first.
updateTailPHIs();
Head->removeSuccessor(CmpBB);
CmpBB->removeSuccessor(Tail);
Head->transferSuccessorsAndUpdatePHIs(CmpBB);
DebugLoc TermDL = Head->getFirstTerminator()->getDebugLoc();
TII->RemoveBranch(*Head);
// If the Head terminator was one of the cbz / tbz branches with built-in
// compare, we need to insert an explicit compare instruction in its place.
if (HeadCond[0].getImm() == -1) {
++NumCompBranches;
unsigned Opc = 0;
switch (HeadCond[1].getImm()) {
case AArch64::CBZW:
case AArch64::CBNZW:
Opc = AArch64::SUBSWri;
break;
case AArch64::CBZX:
case AArch64::CBNZX:
Opc = AArch64::SUBSXri;
break;
default:
llvm_unreachable("Cannot convert Head branch");
}
const MCInstrDesc &MCID = TII->get(Opc);
// Create a dummy virtual register for the SUBS def.
unsigned DestReg =
MRI->createVirtualRegister(TII->getRegClass(MCID, 0, TRI, *MF));
// Insert a SUBS Rn, #0 instruction instead of the cbz / cbnz.
BuildMI(*Head, Head->end(), TermDL, MCID)
.addReg(DestReg, RegState::Define | RegState::Dead)
.addOperand(HeadCond[2])
.addImm(0)
.addImm(0);
// SUBS uses the GPR*sp register classes.
MRI->constrainRegClass(HeadCond[2].getReg(),
TII->getRegClass(MCID, 1, TRI, *MF));
}
Head->splice(Head->end(), CmpBB, CmpBB->begin(), CmpBB->end());
// Now replace CmpMI with a ccmp instruction that also considers the incoming
// flags.
unsigned Opc = 0;
unsigned FirstOp = 1; // First CmpMI operand to copy.
bool isZBranch = false; // CmpMI is a cbz/cbnz instruction.
switch (CmpMI->getOpcode()) {
default:
llvm_unreachable("Unknown compare opcode");
case AArch64::SUBSWri: Opc = AArch64::CCMPWi; break;
case AArch64::SUBSWrr: Opc = AArch64::CCMPWr; break;
case AArch64::SUBSXri: Opc = AArch64::CCMPXi; break;
case AArch64::SUBSXrr: Opc = AArch64::CCMPXr; break;
case AArch64::ADDSWri: Opc = AArch64::CCMNWi; break;
case AArch64::ADDSWrr: Opc = AArch64::CCMNWr; break;
case AArch64::ADDSXri: Opc = AArch64::CCMNXi; break;
case AArch64::ADDSXrr: Opc = AArch64::CCMNXr; break;
case AArch64::FCMPSrr: Opc = AArch64::FCCMPSrr; FirstOp = 0; break;
case AArch64::FCMPDrr: Opc = AArch64::FCCMPDrr; FirstOp = 0; break;
case AArch64::FCMPESrr: Opc = AArch64::FCCMPESrr; FirstOp = 0; break;
case AArch64::FCMPEDrr: Opc = AArch64::FCCMPEDrr; FirstOp = 0; break;
case AArch64::CBZW:
case AArch64::CBNZW:
Opc = AArch64::CCMPWi;
FirstOp = 0;
isZBranch = true;
break;
case AArch64::CBZX:
case AArch64::CBNZX:
Opc = AArch64::CCMPXi;
FirstOp = 0;
isZBranch = true;
break;
}
// The ccmp instruction should set the flags according to the comparison when
// Head would have branched to CmpBB.
// The NZCV immediate operand should provide flags for the case where Head
// would have branched to Tail. These flags should cause the new Head
// terminator to branch to tail.
unsigned NZCV = AArch64CC::getNZCVToSatisfyCondCode(CmpBBTailCC);
const MCInstrDesc &MCID = TII->get(Opc);
MRI->constrainRegClass(CmpMI->getOperand(FirstOp).getReg(),
TII->getRegClass(MCID, 0, TRI, *MF));
if (CmpMI->getOperand(FirstOp + 1).isReg())
MRI->constrainRegClass(CmpMI->getOperand(FirstOp + 1).getReg(),
TII->getRegClass(MCID, 1, TRI, *MF));
MachineInstrBuilder MIB =
BuildMI(*Head, CmpMI, CmpMI->getDebugLoc(), MCID)
.addOperand(CmpMI->getOperand(FirstOp)); // Register Rn
if (isZBranch)
MIB.addImm(0); // cbz/cbnz Rn -> ccmp Rn, #0
else
MIB.addOperand(CmpMI->getOperand(FirstOp + 1)); // Register Rm / Immediate
MIB.addImm(NZCV).addImm(HeadCmpBBCC);
// If CmpMI was a terminator, we need a new conditional branch to replace it.
// This now becomes a Head terminator.
if (isZBranch) {
bool isNZ = CmpMI->getOpcode() == AArch64::CBNZW ||
CmpMI->getOpcode() == AArch64::CBNZX;
BuildMI(*Head, CmpMI, CmpMI->getDebugLoc(), TII->get(AArch64::Bcc))
.addImm(isNZ ? AArch64CC::NE : AArch64CC::EQ)
.addOperand(CmpMI->getOperand(1)); // Branch target.
}
CmpMI->eraseFromParent();
Head->updateTerminator();
RemovedBlocks.push_back(CmpBB);
CmpBB->eraseFromParent();
DEBUG(dbgs() << "Result:\n" << *Head);
++NumConverted;
}
MachineInstrBuilder
MipsInstrInfo::genInstrWithNewOpc(unsigned NewOpc,
MachineBasicBlock::iterator I) const {
MachineInstrBuilder MIB;
// Certain branches have two forms: e.g beq $1, $zero, dst vs beqz $1, dest
// Pick the zero form of the branch for readable assembly and for greater
// branch distance in non-microMIPS mode.
// FIXME: Certain atomic sequences on mips64 generate 32bit references to
// Mips::ZERO, which is incorrect. This test should be updated to use
// Subtarget.getABI().GetZeroReg() when those atomic sequences and others
// are fixed.
bool BranchWithZeroOperand =
(I->isBranch() && !I->isPseudo() && I->getOperand(1).isReg() &&
(I->getOperand(1).getReg() == Mips::ZERO ||
I->getOperand(1).getReg() == Mips::ZERO_64));
if (BranchWithZeroOperand) {
switch (NewOpc) {
case Mips::BEQC:
NewOpc = Mips::BEQZC;
break;
case Mips::BNEC:
NewOpc = Mips::BNEZC;
break;
case Mips::BGEC:
NewOpc = Mips::BGEZC;
break;
case Mips::BLTC:
NewOpc = Mips::BLTZC;
break;
}
}
MIB = BuildMI(*I->getParent(), I, I->getDebugLoc(), get(NewOpc));
// For MIPSR6 JI*C requires an immediate 0 as an operand, JIALC(64) an
// immediate 0 as an operand and requires the removal of it's %RA<imp-def>
// implicit operand as copying the implicit operations of the instructio we're
// looking at will give us the correct flags.
if (NewOpc == Mips::JIC || NewOpc == Mips::JIALC || NewOpc == Mips::JIC64 ||
NewOpc == Mips::JIALC64) {
if (NewOpc == Mips::JIALC || NewOpc == Mips::JIALC64)
MIB->RemoveOperand(0);
for (unsigned J = 0, E = I->getDesc().getNumOperands(); J < E; ++J) {
MIB.addOperand(I->getOperand(J));
}
MIB.addImm(0);
} else if (BranchWithZeroOperand) {
// For MIPSR6 and microMIPS branches with an explicit zero operand, copy
// everything after the zero.
MIB.addOperand(I->getOperand(0));
for (unsigned J = 2, E = I->getDesc().getNumOperands(); J < E; ++J) {
MIB.addOperand(I->getOperand(J));
}
} else {
// All other cases copy all other operands.
for (unsigned J = 0, E = I->getDesc().getNumOperands(); J < E; ++J) {
MIB.addOperand(I->getOperand(J));
}
}
MIB.copyImplicitOps(*I);
MIB.setMemRefs(I->memoperands_begin(), I->memoperands_end());
return MIB;
}
void X86FrameInfo::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
const X86RegisterInfo *RegInfo = TM.getRegisterInfo();
const X86InstrInfo &TII = *TM.getInstrInfo();
MachineBasicBlock::iterator MBBI = prior(MBB.end());
unsigned RetOpcode = MBBI->getOpcode();
DebugLoc DL = MBBI->getDebugLoc();
bool Is64Bit = STI.is64Bit();
unsigned StackAlign = getStackAlignment();
unsigned SlotSize = RegInfo->getSlotSize();
unsigned FramePtr = RegInfo->getFrameRegister(MF);
unsigned StackPtr = RegInfo->getStackRegister();
switch (RetOpcode) {
default:
llvm_unreachable("Can only insert epilog into returning blocks");
case X86::RET:
case X86::RETI:
case X86::TCRETURNdi:
case X86::TCRETURNri:
case X86::TCRETURNmi:
case X86::TCRETURNdi64:
case X86::TCRETURNri64:
case X86::TCRETURNmi64:
case X86::EH_RETURN:
case X86::EH_RETURN64:
break; // These are ok
}
// Get the number of bytes to allocate from the FrameInfo.
uint64_t StackSize = MFI->getStackSize();
uint64_t MaxAlign = MFI->getMaxAlignment();
unsigned CSSize = X86FI->getCalleeSavedFrameSize();
uint64_t NumBytes = 0;
// If we're forcing a stack realignment we can't rely on just the frame
// info, we need to know the ABI stack alignment as well in case we
// have a call out. Otherwise just make sure we have some alignment - we'll
// go with the minimum.
if (ForceStackAlign) {
if (MFI->hasCalls())
MaxAlign = (StackAlign > MaxAlign) ? StackAlign : MaxAlign;
else
MaxAlign = MaxAlign ? MaxAlign : 4;
}
if (hasFP(MF)) {
// Calculate required stack adjustment.
uint64_t FrameSize = StackSize - SlotSize;
if (RegInfo->needsStackRealignment(MF))
FrameSize = (FrameSize + MaxAlign - 1)/MaxAlign*MaxAlign;
NumBytes = FrameSize - CSSize;
// Pop EBP.
BuildMI(MBB, MBBI, DL,
TII.get(Is64Bit ? X86::POP64r : X86::POP32r), FramePtr);
} else {
NumBytes = StackSize - CSSize;
}
// Skip the callee-saved pop instructions.
MachineBasicBlock::iterator LastCSPop = MBBI;
while (MBBI != MBB.begin()) {
MachineBasicBlock::iterator PI = prior(MBBI);
unsigned Opc = PI->getOpcode();
if (Opc != X86::POP32r && Opc != X86::POP64r &&
!PI->getDesc().isTerminator())
break;
--MBBI;
}
DL = MBBI->getDebugLoc();
// If there is an ADD32ri or SUB32ri of ESP immediately before this
// instruction, merge the two instructions.
if (NumBytes || MFI->hasVarSizedObjects())
mergeSPUpdatesUp(MBB, MBBI, StackPtr, &NumBytes);
// If dynamic alloca is used, then reset esp to point to the last callee-saved
// slot before popping them off! Same applies for the case, when stack was
// realigned.
if (RegInfo->needsStackRealignment(MF)) {
// We cannot use LEA here, because stack pointer was realigned. We need to
// deallocate local frame back.
if (CSSize) {
emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII);
MBBI = prior(LastCSPop);
}
BuildMI(MBB, MBBI, DL,
TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),
StackPtr).addReg(FramePtr);
} else if (MFI->hasVarSizedObjects()) {
if (CSSize) {
unsigned Opc = Is64Bit ? X86::LEA64r : X86::LEA32r;
MachineInstr *MI =
addRegOffset(BuildMI(MF, DL, TII.get(Opc), StackPtr),
FramePtr, false, -CSSize);
MBB.insert(MBBI, MI);
} else {
BuildMI(MBB, MBBI, DL,
TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr), StackPtr)
.addReg(FramePtr);
}
} else if (NumBytes) {
// Adjust stack pointer back: ESP += numbytes.
emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII);
}
// We're returning from function via eh_return.
if (RetOpcode == X86::EH_RETURN || RetOpcode == X86::EH_RETURN64) {
MBBI = prior(MBB.end());
MachineOperand &DestAddr = MBBI->getOperand(0);
assert(DestAddr.isReg() && "Offset should be in register!");
BuildMI(MBB, MBBI, DL,
TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),
StackPtr).addReg(DestAddr.getReg());
} else if (RetOpcode == X86::TCRETURNri || RetOpcode == X86::TCRETURNdi ||
RetOpcode == X86::TCRETURNmi ||
RetOpcode == X86::TCRETURNri64 || RetOpcode == X86::TCRETURNdi64 ||
RetOpcode == X86::TCRETURNmi64) {
bool isMem = RetOpcode == X86::TCRETURNmi || RetOpcode == X86::TCRETURNmi64;
// Tail call return: adjust the stack pointer and jump to callee.
MBBI = prior(MBB.end());
MachineOperand &JumpTarget = MBBI->getOperand(0);
MachineOperand &StackAdjust = MBBI->getOperand(isMem ? 5 : 1);
assert(StackAdjust.isImm() && "Expecting immediate value.");
// Adjust stack pointer.
int StackAdj = StackAdjust.getImm();
int MaxTCDelta = X86FI->getTCReturnAddrDelta();
int Offset = 0;
assert(MaxTCDelta <= 0 && "MaxTCDelta should never be positive");
// Incoporate the retaddr area.
Offset = StackAdj-MaxTCDelta;
assert(Offset >= 0 && "Offset should never be negative");
if (Offset) {
// Check for possible merge with preceeding ADD instruction.
Offset += mergeSPUpdates(MBB, MBBI, StackPtr, true);
emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, TII);
}
// Jump to label or value in register.
if (RetOpcode == X86::TCRETURNdi || RetOpcode == X86::TCRETURNdi64) {
MachineInstrBuilder MIB =
BuildMI(MBB, MBBI, DL, TII.get((RetOpcode == X86::TCRETURNdi)
? X86::TAILJMPd : X86::TAILJMPd64));
if (JumpTarget.isGlobal())
MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
JumpTarget.getTargetFlags());
else {
assert(JumpTarget.isSymbol());
MIB.addExternalSymbol(JumpTarget.getSymbolName(),
JumpTarget.getTargetFlags());
}
} else if (RetOpcode == X86::TCRETURNmi || RetOpcode == X86::TCRETURNmi64) {
MachineInstrBuilder MIB =
BuildMI(MBB, MBBI, DL, TII.get((RetOpcode == X86::TCRETURNmi)
? X86::TAILJMPm : X86::TAILJMPm64));
for (unsigned i = 0; i != 5; ++i)
MIB.addOperand(MBBI->getOperand(i));
} else if (RetOpcode == X86::TCRETURNri64) {
BuildMI(MBB, MBBI, DL, TII.get(X86::TAILJMPr64)).
addReg(JumpTarget.getReg(), RegState::Kill);
} else {
BuildMI(MBB, MBBI, DL, TII.get(X86::TAILJMPr)).
addReg(JumpTarget.getReg(), RegState::Kill);
}
MachineInstr *NewMI = prior(MBBI);
for (unsigned i = 2, e = MBBI->getNumOperands(); i != e; ++i)
NewMI->addOperand(MBBI->getOperand(i));
// Delete the pseudo instruction TCRETURN.
MBB.erase(MBBI);
} else if ((RetOpcode == X86::RET || RetOpcode == X86::RETI) &&
(X86FI->getTCReturnAddrDelta() < 0)) {
// Add the return addr area delta back since we are not tail calling.
int delta = -1*X86FI->getTCReturnAddrDelta();
MBBI = prior(MBB.end());
// Check for possible merge with preceeding ADD instruction.
delta += mergeSPUpdates(MBB, MBBI, StackPtr, true);
emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, TII);
}
}
/// If \p MBBI is a pseudo instruction, this method expands
/// it to the corresponding (sequence of) actual instruction(s).
/// \returns true if \p MBBI has been expanded.
bool X86ExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MBBI) {
MachineInstr &MI = *MBBI;
unsigned Opcode = MI.getOpcode();
DebugLoc DL = MBBI->getDebugLoc();
switch (Opcode) {
default:
return false;
case X86::TCRETURNdi:
case X86::TCRETURNri:
case X86::TCRETURNmi:
case X86::TCRETURNdi64:
case X86::TCRETURNri64:
case X86::TCRETURNmi64: {
bool isMem = Opcode == X86::TCRETURNmi || Opcode == X86::TCRETURNmi64;
MachineOperand &JumpTarget = MBBI->getOperand(0);
MachineOperand &StackAdjust = MBBI->getOperand(isMem ? 5 : 1);
assert(StackAdjust.isImm() && "Expecting immediate value.");
// Adjust stack pointer.
int StackAdj = StackAdjust.getImm();
if (StackAdj) {
// Check for possible merge with preceding ADD instruction.
StackAdj += X86FL->mergeSPUpdates(MBB, MBBI, true);
X86FL->emitSPUpdate(MBB, MBBI, StackAdj, /*InEpilogue=*/true);
}
// Jump to label or value in register.
bool IsWin64 = STI->isTargetWin64();
if (Opcode == X86::TCRETURNdi || Opcode == X86::TCRETURNdi64) {
unsigned Op = (Opcode == X86::TCRETURNdi)
? X86::TAILJMPd
: (IsWin64 ? X86::TAILJMPd64_REX : X86::TAILJMPd64);
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(Op));
if (JumpTarget.isGlobal())
MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
JumpTarget.getTargetFlags());
else {
assert(JumpTarget.isSymbol());
MIB.addExternalSymbol(JumpTarget.getSymbolName(),
JumpTarget.getTargetFlags());
}
} else if (Opcode == X86::TCRETURNmi || Opcode == X86::TCRETURNmi64) {
unsigned Op = (Opcode == X86::TCRETURNmi)
? X86::TAILJMPm
: (IsWin64 ? X86::TAILJMPm64_REX : X86::TAILJMPm64);
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(Op));
for (unsigned i = 0; i != 5; ++i)
MIB.addOperand(MBBI->getOperand(i));
} else if (Opcode == X86::TCRETURNri64) {
BuildMI(MBB, MBBI, DL,
TII->get(IsWin64 ? X86::TAILJMPr64_REX : X86::TAILJMPr64))
.addReg(JumpTarget.getReg(), RegState::Kill);
} else {
BuildMI(MBB, MBBI, DL, TII->get(X86::TAILJMPr))
.addReg(JumpTarget.getReg(), RegState::Kill);
}
MachineInstr *NewMI = std::prev(MBBI);
NewMI->copyImplicitOps(*MBBI->getParent()->getParent(), *MBBI);
// Delete the pseudo instruction TCRETURN.
MBB.erase(MBBI);
return true;
}
case X86::EH_RETURN:
case X86::EH_RETURN64: {
MachineOperand &DestAddr = MBBI->getOperand(0);
assert(DestAddr.isReg() && "Offset should be in register!");
const bool Uses64BitFramePtr =
STI->isTarget64BitLP64() || STI->isTargetNaCl64();
unsigned StackPtr = TRI->getStackRegister();
BuildMI(MBB, MBBI, DL,
TII->get(Uses64BitFramePtr ? X86::MOV64rr : X86::MOV32rr), StackPtr)
.addReg(DestAddr.getReg());
// The EH_RETURN pseudo is really removed during the MC Lowering.
return true;
}
case X86::IRET: {
// Adjust stack to erase error code
int64_t StackAdj = MBBI->getOperand(0).getImm();
X86FL->emitSPUpdate(MBB, MBBI, StackAdj, true);
// Replace pseudo with machine iret
BuildMI(MBB, MBBI, DL,
TII->get(STI->is64Bit() ? X86::IRET64 : X86::IRET32));
MBB.erase(MBBI);
return true;
}
case X86::RET: {
// Adjust stack to erase error code
int64_t StackAdj = MBBI->getOperand(0).getImm();
MachineInstrBuilder MIB;
if (StackAdj == 0) {
MIB = BuildMI(MBB, MBBI, DL,
TII->get(STI->is64Bit() ? X86::RETQ : X86::RETL));
} else if (isUInt<16>(StackAdj)) {
MIB = BuildMI(MBB, MBBI, DL,
TII->get(STI->is64Bit() ? X86::RETIQ : X86::RETIL))
.addImm(StackAdj);
} else {
assert(!STI->is64Bit() &&
"shouldn't need to do this for x86_64 targets!");
// A ret can only handle immediates as big as 2**16-1. If we need to pop
// off bytes before the return address, we must do it manually.
BuildMI(MBB, MBBI, DL, TII->get(X86::POP32r)).addReg(X86::ECX, RegState::Define);
X86FL->emitSPUpdate(MBB, MBBI, StackAdj, /*InEpilogue=*/true);
BuildMI(MBB, MBBI, DL, TII->get(X86::PUSH32r)).addReg(X86::ECX);
MIB = BuildMI(MBB, MBBI, DL, TII->get(X86::RETL));
}
for (unsigned I = 1, E = MBBI->getNumOperands(); I != E; ++I)
MIB.addOperand(MBBI->getOperand(I));
MBB.erase(MBBI);
return true;
}
case X86::EH_RESTORE: {
// Restore ESP and EBP, and optionally ESI if required.
bool IsSEH = isAsynchronousEHPersonality(classifyEHPersonality(
MBB.getParent()->getFunction()->getPersonalityFn()));
X86FL->restoreWin32EHStackPointers(MBB, MBBI, DL, /*RestoreSP=*/IsSEH);
MBBI->eraseFromParent();
return true;
}
}
llvm_unreachable("Previous switch has a fallthrough?");
}
bool
Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI,
const ReduceEntry &Entry,
bool LiveCPSR, bool IsSelfLoop) {
if (ReduceLimit != -1 && ((int)NumNarrows >= ReduceLimit))
return false;
if (!MinimizeSize && !OptimizeSize && Entry.AvoidMovs &&
STI->avoidMOVsShifterOperand())
// Don't issue movs with shifter operand for some CPUs unless we
// are optimizing / minimizing for size.
return false;
unsigned Limit = ~0U;
if (Entry.Imm1Limit)
Limit = (1 << Entry.Imm1Limit) - 1;
const MCInstrDesc &MCID = MI->getDesc();
for (unsigned i = 0, e = MCID.getNumOperands(); i != e; ++i) {
if (MCID.OpInfo[i].isPredicate())
continue;
const MachineOperand &MO = MI->getOperand(i);
if (MO.isReg()) {
unsigned Reg = MO.getReg();
if (!Reg || Reg == ARM::CPSR)
continue;
if (Entry.LowRegs1 && !isARMLowRegister(Reg))
return false;
} else if (MO.isImm() &&
!MCID.OpInfo[i].isPredicate()) {
if (((unsigned)MO.getImm()) > Limit)
return false;
}
}
// Check if it's possible / necessary to transfer the predicate.
const MCInstrDesc &NewMCID = TII->get(Entry.NarrowOpc1);
unsigned PredReg = 0;
ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg);
bool SkipPred = false;
if (Pred != ARMCC::AL) {
if (!NewMCID.isPredicable())
// Can't transfer predicate, fail.
return false;
} else {
SkipPred = !NewMCID.isPredicable();
}
bool HasCC = false;
bool CCDead = false;
if (MCID.hasOptionalDef()) {
unsigned NumOps = MCID.getNumOperands();
HasCC = (MI->getOperand(NumOps-1).getReg() == ARM::CPSR);
if (HasCC && MI->getOperand(NumOps-1).isDead())
CCDead = true;
}
if (!VerifyPredAndCC(MI, Entry, false, Pred, LiveCPSR, HasCC, CCDead))
return false;
// Avoid adding a false dependency on partial flag update by some 16-bit
// instructions which has the 's' bit set.
if (Entry.PartFlag && NewMCID.hasOptionalDef() && HasCC &&
canAddPseudoFlagDep(MI, IsSelfLoop))
return false;
// Add the 16-bit instruction.
DebugLoc dl = MI->getDebugLoc();
MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, NewMCID);
MIB.addOperand(MI->getOperand(0));
if (NewMCID.hasOptionalDef()) {
if (HasCC)
AddDefaultT1CC(MIB, CCDead);
else
AddNoT1CC(MIB);
}
// Transfer the rest of operands.
unsigned NumOps = MCID.getNumOperands();
for (unsigned i = 1, e = MI->getNumOperands(); i != e; ++i) {
if (i < NumOps && MCID.OpInfo[i].isOptionalDef())
continue;
if ((MCID.getOpcode() == ARM::t2RSBSri ||
MCID.getOpcode() == ARM::t2RSBri ||
MCID.getOpcode() == ARM::t2SXTB ||
MCID.getOpcode() == ARM::t2SXTH ||
MCID.getOpcode() == ARM::t2UXTB ||
MCID.getOpcode() == ARM::t2UXTH) && i == 2)
// Skip the zero immediate operand, it's now implicit.
continue;
bool isPred = (i < NumOps && MCID.OpInfo[i].isPredicate());
if (SkipPred && isPred)
continue;
const MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isImplicit() && MO.getReg() == ARM::CPSR)
// Skip implicit def of CPSR. Either it's modeled as an optional
// def now or it's already an implicit def on the new instruction.
continue;
MIB.addOperand(MO);
}
if (!MCID.isPredicable() && NewMCID.isPredicable())
AddDefaultPred(MIB);
// Transfer MI flags.
MIB.setMIFlags(MI->getFlags());
DEBUG(errs() << "Converted 32-bit: " << *MI << " to 16-bit: " << *MIB);
MBB.erase_instr(MI);
++NumNarrows;
return true;
}
bool
Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
const ReduceEntry &Entry,
bool LiveCPSR, bool IsSelfLoop) {
if (ReduceLimit2Addr != -1 && ((int)Num2Addrs >= ReduceLimit2Addr))
return false;
if (!MinimizeSize && !OptimizeSize && Entry.AvoidMovs &&
STI->avoidMOVsShifterOperand())
// Don't issue movs with shifter operand for some CPUs unless we
// are optimizing / minimizing for size.
return false;
unsigned Reg0 = MI->getOperand(0).getReg();
unsigned Reg1 = MI->getOperand(1).getReg();
// t2MUL is "special". The tied source operand is second, not first.
if (MI->getOpcode() == ARM::t2MUL) {
unsigned Reg2 = MI->getOperand(2).getReg();
// Early exit if the regs aren't all low regs.
if (!isARMLowRegister(Reg0) || !isARMLowRegister(Reg1)
|| !isARMLowRegister(Reg2))
return false;
if (Reg0 != Reg2) {
// If the other operand also isn't the same as the destination, we
// can't reduce.
if (Reg1 != Reg0)
return false;
// Try to commute the operands to make it a 2-address instruction.
MachineInstr *CommutedMI = TII->commuteInstruction(MI);
if (!CommutedMI)
return false;
}
} else if (Reg0 != Reg1) {
// Try to commute the operands to make it a 2-address instruction.
unsigned CommOpIdx1, CommOpIdx2;
if (!TII->findCommutedOpIndices(MI, CommOpIdx1, CommOpIdx2) ||
CommOpIdx1 != 1 || MI->getOperand(CommOpIdx2).getReg() != Reg0)
return false;
MachineInstr *CommutedMI = TII->commuteInstruction(MI);
if (!CommutedMI)
return false;
}
if (Entry.LowRegs2 && !isARMLowRegister(Reg0))
return false;
if (Entry.Imm2Limit) {
unsigned Imm = MI->getOperand(2).getImm();
unsigned Limit = (1 << Entry.Imm2Limit) - 1;
if (Imm > Limit)
return false;
} else {
unsigned Reg2 = MI->getOperand(2).getReg();
if (Entry.LowRegs2 && !isARMLowRegister(Reg2))
return false;
}
// Check if it's possible / necessary to transfer the predicate.
const MCInstrDesc &NewMCID = TII->get(Entry.NarrowOpc2);
unsigned PredReg = 0;
ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg);
bool SkipPred = false;
if (Pred != ARMCC::AL) {
if (!NewMCID.isPredicable())
// Can't transfer predicate, fail.
return false;
} else {
SkipPred = !NewMCID.isPredicable();
}
bool HasCC = false;
bool CCDead = false;
const MCInstrDesc &MCID = MI->getDesc();
if (MCID.hasOptionalDef()) {
unsigned NumOps = MCID.getNumOperands();
HasCC = (MI->getOperand(NumOps-1).getReg() == ARM::CPSR);
if (HasCC && MI->getOperand(NumOps-1).isDead())
CCDead = true;
}
if (!VerifyPredAndCC(MI, Entry, true, Pred, LiveCPSR, HasCC, CCDead))
return false;
// Avoid adding a false dependency on partial flag update by some 16-bit
// instructions which has the 's' bit set.
if (Entry.PartFlag && NewMCID.hasOptionalDef() && HasCC &&
canAddPseudoFlagDep(MI, IsSelfLoop))
return false;
// Add the 16-bit instruction.
DebugLoc dl = MI->getDebugLoc();
MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, NewMCID);
MIB.addOperand(MI->getOperand(0));
if (NewMCID.hasOptionalDef()) {
if (HasCC)
AddDefaultT1CC(MIB, CCDead);
else
AddNoT1CC(MIB);
}
// Transfer the rest of operands.
unsigned NumOps = MCID.getNumOperands();
for (unsigned i = 1, e = MI->getNumOperands(); i != e; ++i) {
if (i < NumOps && MCID.OpInfo[i].isOptionalDef())
continue;
if (SkipPred && MCID.OpInfo[i].isPredicate())
continue;
MIB.addOperand(MI->getOperand(i));
}
// Transfer MI flags.
MIB.setMIFlags(MI->getFlags());
DEBUG(errs() << "Converted 32-bit: " << *MI << " to 16-bit: " << *MIB);
MBB.erase_instr(MI);
++Num2Addrs;
return true;
}
bool
Thumb2SizeReduce::ReduceLoadStore(MachineBasicBlock &MBB, MachineInstr *MI,
const ReduceEntry &Entry) {
if (ReduceLimitLdSt != -1 && ((int)NumLdSts >= ReduceLimitLdSt))
return false;
unsigned Scale = 1;
bool HasImmOffset = false;
bool HasShift = false;
bool HasOffReg = true;
bool isLdStMul = false;
unsigned Opc = Entry.NarrowOpc1;
unsigned OpNum = 3; // First 'rest' of operands.
uint8_t ImmLimit = Entry.Imm1Limit;
switch (Entry.WideOpc) {
default:
llvm_unreachable("Unexpected Thumb2 load / store opcode!");
case ARM::t2LDRi12:
case ARM::t2STRi12:
if (MI->getOperand(1).getReg() == ARM::SP) {
Opc = Entry.NarrowOpc2;
ImmLimit = Entry.Imm2Limit;
HasOffReg = false;
}
Scale = 4;
HasImmOffset = true;
HasOffReg = false;
break;
case ARM::t2LDRBi12:
case ARM::t2STRBi12:
HasImmOffset = true;
HasOffReg = false;
break;
case ARM::t2LDRHi12:
case ARM::t2STRHi12:
Scale = 2;
HasImmOffset = true;
HasOffReg = false;
break;
case ARM::t2LDRs:
case ARM::t2LDRBs:
case ARM::t2LDRHs:
case ARM::t2LDRSBs:
case ARM::t2LDRSHs:
case ARM::t2STRs:
case ARM::t2STRBs:
case ARM::t2STRHs:
HasShift = true;
OpNum = 4;
break;
case ARM::t2LDMIA:
case ARM::t2LDMDB: {
unsigned BaseReg = MI->getOperand(0).getReg();
if (!isARMLowRegister(BaseReg) || Entry.WideOpc != ARM::t2LDMIA)
return false;
// For the non-writeback version (this one), the base register must be
// one of the registers being loaded.
bool isOK = false;
for (unsigned i = 4; i < MI->getNumOperands(); ++i) {
if (MI->getOperand(i).getReg() == BaseReg) {
isOK = true;
break;
}
}
if (!isOK)
return false;
OpNum = 0;
isLdStMul = true;
break;
}
case ARM::t2LDMIA_RET: {
unsigned BaseReg = MI->getOperand(1).getReg();
if (BaseReg != ARM::SP)
return false;
Opc = Entry.NarrowOpc2; // tPOP_RET
OpNum = 2;
isLdStMul = true;
break;
}
case ARM::t2LDMIA_UPD:
case ARM::t2LDMDB_UPD:
case ARM::t2STMIA_UPD:
case ARM::t2STMDB_UPD: {
OpNum = 0;
unsigned BaseReg = MI->getOperand(1).getReg();
if (BaseReg == ARM::SP &&
(Entry.WideOpc == ARM::t2LDMIA_UPD ||
Entry.WideOpc == ARM::t2STMDB_UPD)) {
Opc = Entry.NarrowOpc2; // tPOP or tPUSH
OpNum = 2;
} else if (!isARMLowRegister(BaseReg) ||
(Entry.WideOpc != ARM::t2LDMIA_UPD &&
Entry.WideOpc != ARM::t2STMIA_UPD)) {
return false;
}
isLdStMul = true;
break;
}
}
unsigned OffsetReg = 0;
bool OffsetKill = false;
if (HasShift) {
OffsetReg = MI->getOperand(2).getReg();
OffsetKill = MI->getOperand(2).isKill();
if (MI->getOperand(3).getImm())
// Thumb1 addressing mode doesn't support shift.
return false;
}
unsigned OffsetImm = 0;
if (HasImmOffset) {
OffsetImm = MI->getOperand(2).getImm();
unsigned MaxOffset = ((1 << ImmLimit) - 1) * Scale;
if ((OffsetImm & (Scale - 1)) || OffsetImm > MaxOffset)
// Make sure the immediate field fits.
return false;
}
// Add the 16-bit load / store instruction.
DebugLoc dl = MI->getDebugLoc();
MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, TII->get(Opc));
if (!isLdStMul) {
MIB.addOperand(MI->getOperand(0));
MIB.addOperand(MI->getOperand(1));
if (HasImmOffset)
MIB.addImm(OffsetImm / Scale);
assert((!HasShift || OffsetReg) && "Invalid so_reg load / store address!");
if (HasOffReg)
MIB.addReg(OffsetReg, getKillRegState(OffsetKill));
}
// Transfer the rest of operands.
for (unsigned e = MI->getNumOperands(); OpNum != e; ++OpNum)
MIB.addOperand(MI->getOperand(OpNum));
// Transfer memoperands.
MIB->setMemRefs(MI->memoperands_begin(), MI->memoperands_end());
// Transfer MI flags.
MIB.setMIFlags(MI->getFlags());
DEBUG(errs() << "Converted 32-bit: " << *MI << " to 16-bit: " << *MIB);
MBB.erase_instr(MI);
++NumLdSts;
return true;
}
void XCoreFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
const XCoreInstrInfo &TII =
*static_cast<const XCoreInstrInfo*>(MF.getTarget().getInstrInfo());
XCoreFunctionInfo *XFI = MF.getInfo<XCoreFunctionInfo>();
DebugLoc dl = MBBI->getDebugLoc();
// Work out frame sizes.
// We will adjust the SP in stages towards the final FrameSize.
int RemainingAdj = MFI->getStackSize();
assert(RemainingAdj%4 == 0 && "Misaligned frame size");
RemainingAdj /= 4;
bool restoreLR = XFI->getUsesLR();
bool UseRETSP = restoreLR && RemainingAdj
&& (MFI->getObjectOffset(XFI->getLRSpillSlot()) == 0);
if (UseRETSP)
restoreLR = false;
bool FP = hasFP(MF);
if (FP) // Restore the stack pointer.
BuildMI(MBB, MBBI, dl, TII.get(XCore::SETSP_1r)).addReg(FramePtr);
// If necessary, restore LR and FP from the stack, as we EXTSP.
SmallVector<std::pair<unsigned,int>,2> SpillList;
GetSpillList(SpillList, MFI, XFI, restoreLR, FP);
unsigned i = SpillList.size();
while (i--) {
unsigned SpilledReg = SpillList[i].first;
int SpillOffset = SpillList[i].second;
assert(SpillOffset % 4 == 0 && "Misaligned stack offset");
assert(SpillOffset <= 0 && "Unexpected positive stack offset");
int OffsetFromTop = - SpillOffset/4;
IfNeededLDAWSP(MBB, MBBI, dl, TII, OffsetFromTop, RemainingAdj);
int Offset = RemainingAdj - OffsetFromTop;
int Opcode = isImmU6(Offset) ? XCore::LDWSP_ru6 : XCore::LDWSP_lru6;
BuildMI(MBB, MBBI, dl, TII.get(Opcode), SpilledReg).addImm(Offset);
}
if (RemainingAdj) {
// Complete all but one of the remaining Stack adjustments.
IfNeededLDAWSP(MBB, MBBI, dl, TII, 0, RemainingAdj);
if (UseRETSP) {
// Fold prologue into return instruction
assert(MBBI->getOpcode() == XCore::RETSP_u6
|| MBBI->getOpcode() == XCore::RETSP_lu6);
int Opcode = isImmU6(RemainingAdj) ? XCore::RETSP_u6 : XCore::RETSP_lu6;
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(Opcode))
.addImm(RemainingAdj);
for (unsigned i = 3, e = MBBI->getNumOperands(); i < e; ++i)
MIB->addOperand(MBBI->getOperand(i)); // copy any variadic operands
MBB.erase(MBBI); // Erase the previous return instruction.
} else {
int Opcode = isImmU6(RemainingAdj) ? XCore::LDAWSP_ru6 :
XCore::LDAWSP_lru6;
BuildMI(MBB, MBBI, dl, TII.get(Opcode), XCore::SP).addImm(RemainingAdj);
// Don't erase the return instruction.
}
} // else Don't erase the return instruction.
}
void XCoreFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
const XCoreInstrInfo &TII = *MF.getSubtarget<XCoreSubtarget>().getInstrInfo();
XCoreFunctionInfo *XFI = MF.getInfo<XCoreFunctionInfo>();
DebugLoc dl = MBBI->getDebugLoc();
unsigned RetOpcode = MBBI->getOpcode();
// Work out frame sizes.
// We will adjust the SP in stages towards the final FrameSize.
int RemainingAdj = MFI->getStackSize();
assert(RemainingAdj%4 == 0 && "Misaligned frame size");
RemainingAdj /= 4;
if (RetOpcode == XCore::EH_RETURN) {
// 'Restore' the exception info the unwinder has placed into the stack
// slots.
SmallVector<StackSlotInfo,2> SpillList;
GetEHSpillList(SpillList, MFI, XFI, MF.getSubtarget().getTargetLowering());
RestoreSpillList(MBB, MBBI, dl, TII, RemainingAdj, SpillList);
// Return to the landing pad.
unsigned EhStackReg = MBBI->getOperand(0).getReg();
unsigned EhHandlerReg = MBBI->getOperand(1).getReg();
BuildMI(MBB, MBBI, dl, TII.get(XCore::SETSP_1r)).addReg(EhStackReg);
BuildMI(MBB, MBBI, dl, TII.get(XCore::BAU_1r)).addReg(EhHandlerReg);
MBB.erase(MBBI); // Erase the previous return instruction.
return;
}
bool restoreLR = XFI->hasLRSpillSlot();
bool UseRETSP = restoreLR && RemainingAdj
&& (MFI->getObjectOffset(XFI->getLRSpillSlot()) == 0);
if (UseRETSP)
restoreLR = false;
bool FP = hasFP(MF);
if (FP) // Restore the stack pointer.
BuildMI(MBB, MBBI, dl, TII.get(XCore::SETSP_1r)).addReg(FramePtr);
// If necessary, restore LR and FP from the stack, as we EXTSP.
SmallVector<StackSlotInfo,2> SpillList;
GetSpillList(SpillList, MFI, XFI, restoreLR, FP);
RestoreSpillList(MBB, MBBI, dl, TII, RemainingAdj, SpillList);
if (RemainingAdj) {
// Complete all but one of the remaining Stack adjustments.
IfNeededLDAWSP(MBB, MBBI, dl, TII, 0, RemainingAdj);
if (UseRETSP) {
// Fold prologue into return instruction
assert(RetOpcode == XCore::RETSP_u6
|| RetOpcode == XCore::RETSP_lu6);
int Opcode = isImmU6(RemainingAdj) ? XCore::RETSP_u6 : XCore::RETSP_lu6;
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(Opcode))
.addImm(RemainingAdj);
for (unsigned i = 3, e = MBBI->getNumOperands(); i < e; ++i)
MIB->addOperand(MBBI->getOperand(i)); // copy any variadic operands
MBB.erase(MBBI); // Erase the previous return instruction.
} else {
int Opcode = isImmU6(RemainingAdj) ? XCore::LDAWSP_ru6 :
XCore::LDAWSP_lru6;
BuildMI(MBB, MBBI, dl, TII.get(Opcode), XCore::SP).addImm(RemainingAdj);
// Don't erase the return instruction.
}
} // else Don't erase the return instruction.
}
