void HexagonMCChecker::init(MCInst const& MCI) {
const MCInstrDesc& MCID = HexagonMCInstrInfo::getDesc(MCII, MCI);
unsigned PredReg = Hexagon::NoRegister;
bool isTrue = false;
// Get used registers.
for (unsigned i = MCID.getNumDefs(); i < MCID.getNumOperands(); ++i)
if (MCI.getOperand(i).isReg()) {
unsigned R = MCI.getOperand(i).getReg();
if (HexagonMCInstrInfo::isPredicated(MCII, MCI) && isPredicateRegister(R)) {
// Note an used predicate register.
PredReg = R;
isTrue = HexagonMCInstrInfo::isPredicatedTrue(MCII, MCI);
// Note use of new predicate register.
if (HexagonMCInstrInfo::isPredicatedNew(MCII, MCI))
NewPreds.insert(PredReg);
}
else
// Note register use. Super-registers are not tracked directly,
// but their components.
for(MCRegAliasIterator SRI(R, &RI, !MCSubRegIterator(R, &RI).isValid());
SRI.isValid();
++SRI)
if (!MCSubRegIterator(*SRI, &RI).isValid())
// Skip super-registers used indirectly.
Uses.insert(*SRI);
}
// Get implicit register definitions.
if (const MCPhysReg *ImpDef = MCID.getImplicitDefs())
for (; *ImpDef; ++ImpDef) {
unsigned R = *ImpDef;
if (Hexagon::R31 != R && MCID.isCall())
// Any register other than the LR and the PC are actually volatile ones
// as defined by the ABI, not modified implicitly by the call insn.
continue;
if (Hexagon::PC == R)
// Branches are the only insns that can change the PC,
// otherwise a read-only register.
continue;
if (Hexagon::USR_OVF == R)
// Many insns change the USR implicitly, but only one or another flag.
// The instruction table models the USR.OVF flag, which can be implicitly
// modified more than once, but cannot be modified in the same packet
// with an instruction that modifies is explicitly. Deal with such situ-
// ations individually.
SoftDefs.insert(R);
else if (isPredicateRegister(R) &&
HexagonMCInstrInfo::isPredicateLate(MCII, MCI))
// Include implicit late predicates.
LatePreds.insert(R);
else
Defs[R].insert(PredSense(PredReg, isTrue));
}
// Figure out explicit register definitions.
for (unsigned i = 0; i < MCID.getNumDefs(); ++i) {
unsigned R = MCI.getOperand(i).getReg(),
S = Hexagon::NoRegister;
// Note register definitions, direct ones as well as indirect side-effects.
// Super-registers are not tracked directly, but their components.
for(MCRegAliasIterator SRI(R, &RI, !MCSubRegIterator(R, &RI).isValid());
SRI.isValid();
++SRI) {
if (MCSubRegIterator(*SRI, &RI).isValid())
// Skip super-registers defined indirectly.
continue;
if (R == *SRI) {
if (S == R)
// Avoid scoring the defined register multiple times.
continue;
else
// Note that the defined register has already been scored.
S = R;
}
if (Hexagon::P3_0 != R && Hexagon::P3_0 == *SRI)
// P3:0 is a special case, since multiple predicate register definitions
// in a packet is allowed as the equivalent of their logical "and".
// Only an explicit definition of P3:0 is noted as such; if a
// side-effect, then note as a soft definition.
SoftDefs.insert(*SRI);
else if (HexagonMCInstrInfo::isPredicateLate(MCII, MCI) && isPredicateRegister(*SRI))
// Some insns produce predicates too late to be used in the same packet.
LatePreds.insert(*SRI);
else if (i == 0 && llvm::HexagonMCInstrInfo::getType(MCII, MCI) == HexagonII::TypeCVI_VM_CUR_LD)
// Current loads should be used in the same packet.
// TODO: relies on the impossibility of a current and a temporary loads
// in the same packet.
CurDefs.insert(*SRI), Defs[*SRI].insert(PredSense(PredReg, isTrue));
else if (i == 0 && llvm::HexagonMCInstrInfo::getType(MCII, MCI) == HexagonII::TypeCVI_VM_TMP_LD)
// Temporary loads should be used in the same packet, but don't commit
// results, so it should be disregarded if another insn changes the same
// register.
// TODO: relies on the impossibility of a current and a temporary loads
// in the same packet.
TmpDefs.insert(*SRI);
else if (i <= 1 && llvm::HexagonMCInstrInfo::hasNewValue2(MCII, MCI) )
// vshuff(Vx, Vy, Rx) <- Vx(0) and Vy(1) are both source and
// destination registers with this instruction. same for vdeal(Vx,Vy,Rx)
Uses.insert(*SRI);
else
Defs[*SRI].insert(PredSense(PredReg, isTrue));
}
}
// Figure out register definitions that produce new values.
if (HexagonMCInstrInfo::hasNewValue(MCII, MCI)) {
unsigned R = HexagonMCInstrInfo::getNewValueOperand(MCII, MCI).getReg();
if (HexagonMCInstrInfo::isCompound(MCII, MCI))
compoundRegisterMap(R); // Compound insns have a limited register range.
for(MCRegAliasIterator SRI(R, &RI, !MCSubRegIterator(R, &RI).isValid());
SRI.isValid();
++SRI)
if (!MCSubRegIterator(*SRI, &RI).isValid())
// No super-registers defined indirectly.
NewDefs[*SRI].push_back(NewSense::Def(PredReg, HexagonMCInstrInfo::isPredicatedTrue(MCII, MCI),
HexagonMCInstrInfo::isFloat(MCII, MCI)));
// For fairly unique 2-dot-new producers, example:
// vdeal(V1, V9, R0) V1.new and V9.new can be used by consumers.
if (HexagonMCInstrInfo::hasNewValue2(MCII, MCI)) {
unsigned R2 = HexagonMCInstrInfo::getNewValueOperand2(MCII, MCI).getReg();
for(MCRegAliasIterator SRI(R2, &RI, !MCSubRegIterator(R2, &RI).isValid());
SRI.isValid();
++SRI)
if (!MCSubRegIterator(*SRI, &RI).isValid())
NewDefs[*SRI].push_back(NewSense::Def(PredReg, HexagonMCInstrInfo::isPredicatedTrue(MCII, MCI),
HexagonMCInstrInfo::isFloat(MCII, MCI)));
}
}
// Figure out definitions of new predicate registers.
if (HexagonMCInstrInfo::isPredicatedNew(MCII, MCI))
for (unsigned i = MCID.getNumDefs(); i < MCID.getNumOperands(); ++i)
if (MCI.getOperand(i).isReg()) {
unsigned P = MCI.getOperand(i).getReg();
if (isPredicateRegister(P))
NewPreds.insert(P);
}
// Figure out uses of new values.
if (HexagonMCInstrInfo::isNewValue(MCII, MCI)) {
unsigned N = HexagonMCInstrInfo::getNewValueOperand(MCII, MCI).getReg();
if (!MCSubRegIterator(N, &RI).isValid()) {
// Super-registers cannot use new values.
if (MCID.isBranch())
NewUses[N] = NewSense::Jmp(llvm::HexagonMCInstrInfo::getType(MCII, MCI) == HexagonII::TypeNV);
else
NewUses[N] = NewSense::Use(PredReg, HexagonMCInstrInfo::isPredicatedTrue(MCII, MCI));
}
}
}
// Note that MIMG format provides no information about VADDR size.
// Consequently, decoded instructions always show address
// as if it has 1 dword, which could be not really so.
DecodeStatus AMDGPUDisassembler::convertMIMGInst(MCInst &MI) const {
int VDstIdx = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
AMDGPU::OpName::vdst);
int VDataIdx = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
AMDGPU::OpName::vdata);
int DMaskIdx = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
AMDGPU::OpName::dmask);
int TFEIdx = AMDGPU::getNamedOperandIdx(MI.getOpcode(),
AMDGPU::OpName::tfe);
assert(VDataIdx != -1);
assert(DMaskIdx != -1);
assert(TFEIdx != -1);
bool isAtomic = (VDstIdx != -1);
unsigned DMask = MI.getOperand(DMaskIdx).getImm() & 0xf;
if (DMask == 0)
return MCDisassembler::Success;
unsigned DstSize = countPopulation(DMask);
if (DstSize == 1)
return MCDisassembler::Success;
bool D16 = MCII->get(MI.getOpcode()).TSFlags & SIInstrFlags::D16;
if (D16 && AMDGPU::hasPackedD16(STI)) {
DstSize = (DstSize + 1) / 2;
}
// FIXME: Add tfe support
if (MI.getOperand(TFEIdx).getImm())
return MCDisassembler::Success;
int NewOpcode = -1;
if (isAtomic) {
if (DMask == 0x1 || DMask == 0x3 || DMask == 0xF) {
NewOpcode = AMDGPU::getMaskedMIMGAtomicOp(*MCII, MI.getOpcode(), DstSize);
}
if (NewOpcode == -1) return MCDisassembler::Success;
} else {
NewOpcode = AMDGPU::getMaskedMIMGOp(*MCII, MI.getOpcode(), DstSize);
assert(NewOpcode != -1 && "could not find matching mimg channel instruction");
}
auto RCID = MCII->get(NewOpcode).OpInfo[VDataIdx].RegClass;
// Get first subregister of VData
unsigned Vdata0 = MI.getOperand(VDataIdx).getReg();
unsigned VdataSub0 = MRI.getSubReg(Vdata0, AMDGPU::sub0);
Vdata0 = (VdataSub0 != 0)? VdataSub0 : Vdata0;
// Widen the register to the correct number of enabled channels.
auto NewVdata = MRI.getMatchingSuperReg(Vdata0, AMDGPU::sub0,
&MRI.getRegClass(RCID));
if (NewVdata == AMDGPU::NoRegister) {
// It's possible to encode this such that the low register + enabled
// components exceeds the register count.
return MCDisassembler::Success;
}
MI.setOpcode(NewOpcode);
// vaddr will be always appear as a single VGPR. This will look different than
// how it is usually emitted because the number of register components is not
// in the instruction encoding.
MI.getOperand(VDataIdx) = MCOperand::createReg(NewVdata);
if (isAtomic) {
// Atomic operations have an additional operand (a copy of data)
MI.getOperand(VDstIdx) = MCOperand::createReg(NewVdata);
}
return MCDisassembler::Success;
}
void HexagonAsmPrinter::HexagonProcessInstruction(MCInst &Inst,
const MachineInstr &MI) {
MCInst &MappedInst = static_cast <MCInst &>(Inst);
const MCRegisterInfo *RI = OutStreamer->getContext().getRegisterInfo();
switch (Inst.getOpcode()) {
default: return;
case Hexagon::A2_iconst: {
Inst.setOpcode(Hexagon::A2_addi);
MCOperand Reg = Inst.getOperand(0);
MCOperand S16 = Inst.getOperand(1);
HexagonMCInstrInfo::setMustNotExtend(*S16.getExpr());
HexagonMCInstrInfo::setS23_2_reloc(*S16.getExpr());
Inst.clear();
Inst.addOperand(Reg);
Inst.addOperand(MCOperand::createReg(Hexagon::R0));
Inst.addOperand(S16);
break;
}
// "$dst = CONST64(#$src1)",
case Hexagon::CONST64:
if (!OutStreamer->hasRawTextSupport()) {
const MCOperand &Imm = MappedInst.getOperand(1);
MCSectionSubPair Current = OutStreamer->getCurrentSection();
MCSymbol *Sym = smallData(*this, MI, *OutStreamer, Imm, 8);
OutStreamer->SwitchSection(Current.first, Current.second);
MCInst TmpInst;
MCOperand &Reg = MappedInst.getOperand(0);
TmpInst.setOpcode(Hexagon::L2_loadrdgp);
TmpInst.addOperand(Reg);
TmpInst.addOperand(MCOperand::createExpr(
MCSymbolRefExpr::create(Sym, OutContext)));
MappedInst = TmpInst;
}
break;
case Hexagon::CONST32:
if (!OutStreamer->hasRawTextSupport()) {
MCOperand &Imm = MappedInst.getOperand(1);
MCSectionSubPair Current = OutStreamer->getCurrentSection();
MCSymbol *Sym = smallData(*this, MI, *OutStreamer, Imm, 4);
OutStreamer->SwitchSection(Current.first, Current.second);
MCInst TmpInst;
MCOperand &Reg = MappedInst.getOperand(0);
TmpInst.setOpcode(Hexagon::L2_loadrigp);
TmpInst.addOperand(Reg);
TmpInst.addOperand(MCOperand::createExpr(HexagonMCExpr::create(
MCSymbolRefExpr::create(Sym, OutContext), OutContext)));
MappedInst = TmpInst;
}
break;
// C2_pxfer_map maps to C2_or instruction. Though, it's possible to use
// C2_or during instruction selection itself but it results
// into suboptimal code.
case Hexagon::C2_pxfer_map: {
MCOperand &Ps = Inst.getOperand(1);
MappedInst.setOpcode(Hexagon::C2_or);
MappedInst.addOperand(Ps);
return;
}
// Vector reduce complex multiply by scalar, Rt & 1 map to :hi else :lo
// The insn is mapped from the 4 operand to the 3 operand raw form taking
// 3 register pairs.
case Hexagon::M2_vrcmpys_acc_s1: {
MCOperand &Rt = Inst.getOperand(3);
assert (Rt.isReg() && "Expected register and none was found");
unsigned Reg = RI->getEncodingValue(Rt.getReg());
if (Reg & 1)
MappedInst.setOpcode(Hexagon::M2_vrcmpys_acc_s1_h);
else
MappedInst.setOpcode(Hexagon::M2_vrcmpys_acc_s1_l);
Rt.setReg(getHexagonRegisterPair(Rt.getReg(), RI));
return;
}
case Hexagon::M2_vrcmpys_s1: {
MCOperand &Rt = Inst.getOperand(2);
assert (Rt.isReg() && "Expected register and none was found");
unsigned Reg = RI->getEncodingValue(Rt.getReg());
if (Reg & 1)
MappedInst.setOpcode(Hexagon::M2_vrcmpys_s1_h);
else
MappedInst.setOpcode(Hexagon::M2_vrcmpys_s1_l);
Rt.setReg(getHexagonRegisterPair(Rt.getReg(), RI));
return;
}
case Hexagon::M2_vrcmpys_s1rp: {
MCOperand &Rt = Inst.getOperand(2);
assert (Rt.isReg() && "Expected register and none was found");
unsigned Reg = RI->getEncodingValue(Rt.getReg());
if (Reg & 1)
MappedInst.setOpcode(Hexagon::M2_vrcmpys_s1rp_h);
else
MappedInst.setOpcode(Hexagon::M2_vrcmpys_s1rp_l);
Rt.setReg(getHexagonRegisterPair(Rt.getReg(), RI));
return;
}
case Hexagon::A4_boundscheck: {
MCOperand &Rs = Inst.getOperand(1);
assert (Rs.isReg() && "Expected register and none was found");
unsigned Reg = RI->getEncodingValue(Rs.getReg());
if (Reg & 1) // Odd mapped to raw:hi, regpair is rodd:odd-1, like r3:2
MappedInst.setOpcode(Hexagon::A4_boundscheck_hi);
else // raw:lo
MappedInst.setOpcode(Hexagon::A4_boundscheck_lo);
Rs.setReg(getHexagonRegisterPair(Rs.getReg(), RI));
return;
}
case Hexagon::S5_asrhub_rnd_sat_goodsyntax: {
MCOperand &MO = MappedInst.getOperand(2);
int64_t Imm;
MCExpr const *Expr = MO.getExpr();
bool Success = Expr->evaluateAsAbsolute(Imm);
assert (Success && "Expected immediate and none was found");
(void)Success;
MCInst TmpInst;
if (Imm == 0) {
TmpInst.setOpcode(Hexagon::S2_vsathub);
TmpInst.addOperand(MappedInst.getOperand(0));
TmpInst.addOperand(MappedInst.getOperand(1));
MappedInst = TmpInst;
return;
}
TmpInst.setOpcode(Hexagon::S5_asrhub_rnd_sat);
TmpInst.addOperand(MappedInst.getOperand(0));
TmpInst.addOperand(MappedInst.getOperand(1));
const MCExpr *One = MCConstantExpr::create(1, OutContext);
const MCExpr *Sub = MCBinaryExpr::createSub(Expr, One, OutContext);
TmpInst.addOperand(
MCOperand::createExpr(HexagonMCExpr::create(Sub, OutContext)));
MappedInst = TmpInst;
return;
}
case Hexagon::S5_vasrhrnd_goodsyntax:
case Hexagon::S2_asr_i_p_rnd_goodsyntax: {
MCOperand &MO2 = MappedInst.getOperand(2);
MCExpr const *Expr = MO2.getExpr();
int64_t Imm;
bool Success = Expr->evaluateAsAbsolute(Imm);
assert (Success && "Expected immediate and none was found");
(void)Success;
MCInst TmpInst;
if (Imm == 0) {
TmpInst.setOpcode(Hexagon::A2_combinew);
TmpInst.addOperand(MappedInst.getOperand(0));
MCOperand &MO1 = MappedInst.getOperand(1);
unsigned High = RI->getSubReg(MO1.getReg(), Hexagon::subreg_hireg);
unsigned Low = RI->getSubReg(MO1.getReg(), Hexagon::subreg_loreg);
// Add a new operand for the second register in the pair.
TmpInst.addOperand(MCOperand::createReg(High));
TmpInst.addOperand(MCOperand::createReg(Low));
MappedInst = TmpInst;
return;
}
if (Inst.getOpcode() == Hexagon::S2_asr_i_p_rnd_goodsyntax)
TmpInst.setOpcode(Hexagon::S2_asr_i_p_rnd);
else
TmpInst.setOpcode(Hexagon::S5_vasrhrnd);
TmpInst.addOperand(MappedInst.getOperand(0));
TmpInst.addOperand(MappedInst.getOperand(1));
const MCExpr *One = MCConstantExpr::create(1, OutContext);
const MCExpr *Sub = MCBinaryExpr::createSub(Expr, One, OutContext);
TmpInst.addOperand(
MCOperand::createExpr(HexagonMCExpr::create(Sub, OutContext)));
MappedInst = TmpInst;
return;
}
// if ("#u5==0") Assembler mapped to: "Rd=Rs"; else Rd=asr(Rs,#u5-1):rnd
case Hexagon::S2_asr_i_r_rnd_goodsyntax: {
MCOperand &MO = Inst.getOperand(2);
MCExpr const *Expr = MO.getExpr();
int64_t Imm;
bool Success = Expr->evaluateAsAbsolute(Imm);
assert (Success && "Expected immediate and none was found");
(void)Success;
MCInst TmpInst;
if (Imm == 0) {
TmpInst.setOpcode(Hexagon::A2_tfr);
TmpInst.addOperand(MappedInst.getOperand(0));
TmpInst.addOperand(MappedInst.getOperand(1));
MappedInst = TmpInst;
return;
}
TmpInst.setOpcode(Hexagon::S2_asr_i_r_rnd);
TmpInst.addOperand(MappedInst.getOperand(0));
TmpInst.addOperand(MappedInst.getOperand(1));
const MCExpr *One = MCConstantExpr::create(1, OutContext);
const MCExpr *Sub = MCBinaryExpr::createSub(Expr, One, OutContext);
TmpInst.addOperand(
MCOperand::createExpr(HexagonMCExpr::create(Sub, OutContext)));
MappedInst = TmpInst;
return;
}
// Translate a "$Rdd = #imm" to "$Rdd = combine(#[-1,0], #imm)"
case Hexagon::A2_tfrpi: {
MCInst TmpInst;
MCOperand &Rdd = MappedInst.getOperand(0);
MCOperand &MO = MappedInst.getOperand(1);
TmpInst.setOpcode(Hexagon::A2_combineii);
TmpInst.addOperand(Rdd);
int64_t Imm;
bool Success = MO.getExpr()->evaluateAsAbsolute(Imm);
if (Success && Imm < 0) {
const MCExpr *MOne = MCConstantExpr::create(-1, OutContext);
TmpInst.addOperand(MCOperand::createExpr(HexagonMCExpr::create(MOne, OutContext)));
} else {
const MCExpr *Zero = MCConstantExpr::create(0, OutContext);
TmpInst.addOperand(MCOperand::createExpr(HexagonMCExpr::create(Zero, OutContext)));
}
TmpInst.addOperand(MO);
MappedInst = TmpInst;
return;
}
// Translate a "$Rdd = $Rss" to "$Rdd = combine($Rs, $Rt)"
case Hexagon::A2_tfrp: {
MCOperand &MO = MappedInst.getOperand(1);
unsigned High = RI->getSubReg(MO.getReg(), Hexagon::subreg_hireg);
unsigned Low = RI->getSubReg(MO.getReg(), Hexagon::subreg_loreg);
MO.setReg(High);
// Add a new operand for the second register in the pair.
MappedInst.addOperand(MCOperand::createReg(Low));
MappedInst.setOpcode(Hexagon::A2_combinew);
return;
}
case Hexagon::A2_tfrpt:
case Hexagon::A2_tfrpf: {
MCOperand &MO = MappedInst.getOperand(2);
unsigned High = RI->getSubReg(MO.getReg(), Hexagon::subreg_hireg);
unsigned Low = RI->getSubReg(MO.getReg(), Hexagon::subreg_loreg);
MO.setReg(High);
// Add a new operand for the second register in the pair.
MappedInst.addOperand(MCOperand::createReg(Low));
MappedInst.setOpcode((Inst.getOpcode() == Hexagon::A2_tfrpt)
? Hexagon::C2_ccombinewt
: Hexagon::C2_ccombinewf);
return;
}
case Hexagon::A2_tfrptnew:
case Hexagon::A2_tfrpfnew: {
MCOperand &MO = MappedInst.getOperand(2);
unsigned High = RI->getSubReg(MO.getReg(), Hexagon::subreg_hireg);
unsigned Low = RI->getSubReg(MO.getReg(), Hexagon::subreg_loreg);
MO.setReg(High);
// Add a new operand for the second register in the pair.
MappedInst.addOperand(MCOperand::createReg(Low));
MappedInst.setOpcode((Inst.getOpcode() == Hexagon::A2_tfrptnew)
? Hexagon::C2_ccombinewnewt
: Hexagon::C2_ccombinewnewf);
return;
}
case Hexagon::M2_mpysmi: {
MCOperand &Imm = MappedInst.getOperand(2);
MCExpr const *Expr = Imm.getExpr();
int64_t Value;
bool Success = Expr->evaluateAsAbsolute(Value);
assert(Success);
(void)Success;
if (Value < 0 && Value > -256) {
MappedInst.setOpcode(Hexagon::M2_mpysin);
Imm.setExpr(HexagonMCExpr::create(
MCUnaryExpr::createMinus(Expr, OutContext), OutContext));
} else
MappedInst.setOpcode(Hexagon::M2_mpysip);
return;
}
case Hexagon::A2_addsp: {
MCOperand &Rt = Inst.getOperand(1);
assert (Rt.isReg() && "Expected register and none was found");
unsigned Reg = RI->getEncodingValue(Rt.getReg());
if (Reg & 1)
MappedInst.setOpcode(Hexagon::A2_addsph);
else
MappedInst.setOpcode(Hexagon::A2_addspl);
Rt.setReg(getHexagonRegisterPair(Rt.getReg(), RI));
return;
}
case Hexagon::HEXAGON_V6_vd0_pseudo:
case Hexagon::HEXAGON_V6_vd0_pseudo_128B: {
MCInst TmpInst;
assert (Inst.getOperand(0).isReg() &&
"Expected register and none was found");
TmpInst.setOpcode(Hexagon::V6_vxor);
TmpInst.addOperand(Inst.getOperand(0));
TmpInst.addOperand(Inst.getOperand(0));
TmpInst.addOperand(Inst.getOperand(0));
MappedInst = TmpInst;
return;
}
}
}
/// \brief Simplify things like MOV32rm to MOV32o32a.
static void SimplifyShortMoveForm(X86AsmPrinter &Printer, MCInst &Inst,
unsigned Opcode) {
// Don't make these simplifications in 64-bit mode; other assemblers don't
// perform them because they make the code larger.
if (Printer.getSubtarget().is64Bit())
return;
bool IsStore = Inst.getOperand(0).isReg() && Inst.getOperand(1).isReg();
unsigned AddrBase = IsStore;
unsigned RegOp = IsStore ? 0 : 5;
unsigned AddrOp = AddrBase + 3;
assert(Inst.getNumOperands() == 6 && Inst.getOperand(RegOp).isReg() &&
Inst.getOperand(AddrBase + X86::AddrBaseReg).isReg() &&
Inst.getOperand(AddrBase + X86::AddrScaleAmt).isImm() &&
Inst.getOperand(AddrBase + X86::AddrIndexReg).isReg() &&
Inst.getOperand(AddrBase + X86::AddrSegmentReg).isReg() &&
(Inst.getOperand(AddrOp).isExpr() ||
Inst.getOperand(AddrOp).isImm()) &&
"Unexpected instruction!");
// Check whether the destination register can be fixed.
unsigned Reg = Inst.getOperand(RegOp).getReg();
if (Reg != X86::AL && Reg != X86::AX && Reg != X86::EAX && Reg != X86::RAX)
return;
// Check whether this is an absolute address.
// FIXME: We know TLVP symbol refs aren't, but there should be a better way
// to do this here.
bool Absolute = true;
if (Inst.getOperand(AddrOp).isExpr()) {
const MCExpr *MCE = Inst.getOperand(AddrOp).getExpr();
if (const MCSymbolRefExpr *SRE = dyn_cast<MCSymbolRefExpr>(MCE))
if (SRE->getKind() == MCSymbolRefExpr::VK_TLVP)
Absolute = false;
}
if (Absolute &&
(Inst.getOperand(AddrBase + X86::AddrBaseReg).getReg() != 0 ||
Inst.getOperand(AddrBase + X86::AddrScaleAmt).getImm() != 1 ||
Inst.getOperand(AddrBase + X86::AddrIndexReg).getReg() != 0))
return;
// If so, rewrite the instruction.
MCOperand Saved = Inst.getOperand(AddrOp);
MCOperand Seg = Inst.getOperand(AddrBase + X86::AddrSegmentReg);
Inst = MCInst();
Inst.setOpcode(Opcode);
Inst.addOperand(Saved);
Inst.addOperand(Seg);
}
/*
* Disassemble a function, using the LLVM MC disassembler.
*
* See also:
* - http://blog.llvm.org/2010/01/x86-disassembler.html
* - http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html
*/
static size_t
disassemble(const void* func, llvm::raw_ostream & Out)
{
using namespace llvm;
const uint8_t *bytes = (const uint8_t *)func;
/*
* Limit disassembly to this extent
*/
const uint64_t extent = 96 * 1024;
uint64_t max_pc = 0;
/*
* Initialize all used objects.
*/
std::string Triple = sys::getDefaultTargetTriple();
std::string Error;
const Target *T = TargetRegistry::lookupTarget(Triple, Error);
#if HAVE_LLVM >= 0x0304
OwningPtr<const MCAsmInfo> AsmInfo(T->createMCAsmInfo(*T->createMCRegInfo(Triple), Triple));
#else
OwningPtr<const MCAsmInfo> AsmInfo(T->createMCAsmInfo(Triple));
#endif
if (!AsmInfo) {
Out << "error: no assembly info for target " << Triple << "\n";
Out.flush();
return 0;
}
unsigned int AsmPrinterVariant = AsmInfo->getAssemblerDialect();
OwningPtr<const MCRegisterInfo> MRI(T->createMCRegInfo(Triple));
if (!MRI) {
Out << "error: no register info for target " << Triple.c_str() << "\n";
Out.flush();
return 0;
}
OwningPtr<const MCInstrInfo> MII(T->createMCInstrInfo());
if (!MII) {
Out << "error: no instruction info for target " << Triple.c_str() << "\n";
Out.flush();
return 0;
}
#if HAVE_LLVM >= 0x0305
OwningPtr<const MCSubtargetInfo> STI(T->createMCSubtargetInfo(Triple, sys::getHostCPUName(), ""));
OwningPtr<MCContext> MCCtx(new MCContext(AsmInfo.get(), MRI.get(), 0));
OwningPtr<const MCDisassembler> DisAsm(T->createMCDisassembler(*STI, *MCCtx));
#else
OwningPtr<const MCSubtargetInfo> STI(T->createMCSubtargetInfo(Triple, sys::getHostCPUName(), ""));
OwningPtr<const MCDisassembler> DisAsm(T->createMCDisassembler(*STI));
#endif
if (!DisAsm) {
Out << "error: no disassembler for target " << Triple << "\n";
Out.flush();
return 0;
}
OwningPtr<MCInstPrinter> Printer(
T->createMCInstPrinter(AsmPrinterVariant, *AsmInfo, *MII, *MRI, *STI));
if (!Printer) {
Out << "error: no instruction printer for target " << Triple.c_str() << "\n";
Out.flush();
return 0;
}
TargetOptions options;
#if defined(DEBUG)
options.JITEmitDebugInfo = true;
#endif
#if defined(PIPE_ARCH_X86)
options.StackAlignmentOverride = 4;
#endif
#if defined(DEBUG) || defined(PROFILE)
options.NoFramePointerElim = true;
#endif
OwningPtr<TargetMachine> TM(T->createTargetMachine(Triple, sys::getHostCPUName(), "", options));
#if HAVE_LLVM >= 0x0306
const TargetInstrInfo *TII = TM->getSubtargetImpl()->getInstrInfo();
#else
const TargetInstrInfo *TII = TM->getInstrInfo();
#endif
/*
* Wrap the data in a MemoryObject
*/
BufferMemoryObject memoryObject((const uint8_t *)bytes, extent);
uint64_t pc;
pc = 0;
while (true) {
MCInst Inst;
uint64_t Size;
/*
* Print address. We use addresses relative to the start of the function,
* so that between runs.
*/
Out << llvm::format("%6lu:\t", (unsigned long)pc);
if (!DisAsm->getInstruction(Inst, Size, memoryObject,
pc,
nulls(), nulls())) {
Out << "invalid";
pc += 1;
}
/*
* Output the bytes in hexidecimal format.
*/
if (0) {
unsigned i;
for (i = 0; i < Size; ++i) {
Out << llvm::format("%02x ", ((const uint8_t*)bytes)[pc + i]);
}
for (; i < 16; ++i) {
Out << " ";
}
}
/*
* Print the instruction.
*/
Printer->printInst(&Inst, Out, "");
/*
* Advance.
*/
pc += Size;
const MCInstrDesc &TID = TII->get(Inst.getOpcode());
/*
* Keep track of forward jumps to a nearby address.
*/
if (TID.isBranch()) {
for (unsigned i = 0; i < Inst.getNumOperands(); ++i) {
const MCOperand &operand = Inst.getOperand(i);
if (operand.isImm()) {
uint64_t jump;
/*
* FIXME: Handle both relative and absolute addresses correctly.
* EDInstInfo actually has this info, but operandTypes and
* operandFlags enums are not exposed in the public interface.
*/
if (1) {
/*
* PC relative addr.
*/
jump = pc + operand.getImm();
} else {
/*
* Absolute addr.
*/
jump = (uint64_t)operand.getImm();
}
/*
* Output the address relative to the function start, given
* that MC will print the addresses relative the current pc.
*/
Out << "\t\t; " << jump;
/*
* Ignore far jumps given it could be actually a tail return to
* a random address.
*/
if (jump > max_pc &&
jump < extent) {
max_pc = jump;
}
}
}
}
Out << "\n";
/*
* Stop disassembling on return statements, if there is no record of a
* jump to a successive address.
*/
if (TID.isReturn()) {
if (pc > max_pc) {
break;
}
}
}
/*
* Print GDB command, useful to verify output.
*/
if (0) {
_debug_printf("disassemble %p %p\n", bytes, bytes + pc);
}
Out << "\n";
Out.flush();
return pc;
}
void WebAssemblyMCCodeEmitter::encodeInstruction(
const MCInst &MI, raw_ostream &OS, SmallVectorImpl<MCFixup> &Fixups,
const MCSubtargetInfo &STI) const {
uint64_t Start = OS.tell();
uint64_t Binary = getBinaryCodeForInstr(MI, Fixups, STI);
assert(Binary < UINT8_MAX && "Multi-byte opcodes not supported yet");
OS << uint8_t(Binary);
// For br_table instructions, encode the size of the table. In the MCInst,
// there's an index operand, one operand for each table entry, and the
// default operand.
if (MI.getOpcode() == WebAssembly::BR_TABLE_I32 ||
MI.getOpcode() == WebAssembly::BR_TABLE_I64)
encodeULEB128(MI.getNumOperands() - 2, OS);
const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
for (unsigned i = 0, e = MI.getNumOperands(); i < e; ++i) {
const MCOperand &MO = MI.getOperand(i);
if (MO.isReg()) {
/* nothing to encode */
} else if (MO.isImm()) {
if (i < Desc.getNumOperands()) {
assert(Desc.TSFlags == 0 &&
"WebAssembly non-variable_ops don't use TSFlags");
const MCOperandInfo &Info = Desc.OpInfo[i];
if (Info.OperandType == WebAssembly::OPERAND_I32IMM) {
encodeSLEB128(int32_t(MO.getImm()), OS);
} else if (Info.OperandType == WebAssembly::OPERAND_I64IMM) {
encodeSLEB128(int64_t(MO.getImm()), OS);
} else if (Info.OperandType == WebAssembly::OPERAND_GLOBAL) {
llvm_unreachable("wasm globals should only be accessed symbolicly");
} else if (Info.OperandType == WebAssembly::OPERAND_SIGNATURE) {
encodeSLEB128(int64_t(MO.getImm()), OS);
} else {
encodeULEB128(uint64_t(MO.getImm()), OS);
}
} else {
assert(Desc.TSFlags == (WebAssemblyII::VariableOpIsImmediate |
WebAssemblyII::VariableOpImmediateIsLabel));
encodeULEB128(uint64_t(MO.getImm()), OS);
}
} else if (MO.isFPImm()) {
assert(i < Desc.getNumOperands() &&
"Unexpected floating-point immediate as a non-fixed operand");
assert(Desc.TSFlags == 0 &&
"WebAssembly variable_ops floating point ops don't use TSFlags");
const MCOperandInfo &Info = Desc.OpInfo[i];
if (Info.OperandType == WebAssembly::OPERAND_F32IMM) {
// TODO: MC converts all floating point immediate operands to double.
// This is fine for numeric values, but may cause NaNs to change bits.
float f = float(MO.getFPImm());
support::endian::Writer<support::little>(OS).write<float>(f);
} else {
assert(Info.OperandType == WebAssembly::OPERAND_F64IMM);
double d = MO.getFPImm();
support::endian::Writer<support::little>(OS).write<double>(d);
}
} else if (MO.isExpr()) {
const MCOperandInfo &Info = Desc.OpInfo[i];
llvm::MCFixupKind FixupKind;
size_t PaddedSize;
if (Info.OperandType == WebAssembly::OPERAND_I32IMM) {
FixupKind = MCFixupKind(WebAssembly::fixup_code_sleb128_i32);
PaddedSize = 5;
} else if (Info.OperandType == WebAssembly::OPERAND_I64IMM) {
FixupKind = MCFixupKind(WebAssembly::fixup_code_sleb128_i64);
PaddedSize = 10;
} else if (Info.OperandType == WebAssembly::OPERAND_FUNCTION32 ||
Info.OperandType == WebAssembly::OPERAND_OFFSET32 ||
Info.OperandType == WebAssembly::OPERAND_TYPEINDEX) {
FixupKind = MCFixupKind(WebAssembly::fixup_code_uleb128_i32);
PaddedSize = 5;
} else if (Info.OperandType == WebAssembly::OPERAND_GLOBAL) {
FixupKind = MCFixupKind(WebAssembly::fixup_code_global_index);
PaddedSize = 5;
} else {
llvm_unreachable("unexpected symbolic operand kind");
}
Fixups.push_back(MCFixup::create(
OS.tell() - Start, MO.getExpr(),
FixupKind, MI.getLoc()));
++MCNumFixups;
encodeULEB128(0, OS, PaddedSize - 1);
} else {
llvm_unreachable("unexpected operand kind");
}
}
++MCNumEmitted; // Keep track of the # of mi's emitted.
}
static bool isReg(const MCInst &MI, unsigned OpNo) {
assert(MI.getOperand(OpNo).isReg() && "Register operand expected.");
return MI.getOperand(OpNo).getReg() == R;
}
bool SparcAsmParser::expandSET(MCInst &Inst, SMLoc IDLoc,
SmallVectorImpl<MCInst> &Instructions) {
MCOperand MCRegOp = Inst.getOperand(0);
MCOperand MCValOp = Inst.getOperand(1);
assert(MCRegOp.isReg());
assert(MCValOp.isImm() || MCValOp.isExpr());
// the imm operand can be either an expression or an immediate.
bool IsImm = Inst.getOperand(1).isImm();
int64_t RawImmValue = IsImm ? MCValOp.getImm() : 0;
// Allow either a signed or unsigned 32-bit immediate.
if (RawImmValue < -2147483648LL || RawImmValue > 4294967295LL) {
return Error(IDLoc,
"set: argument must be between -2147483648 and 4294967295");
}
// If the value was expressed as a large unsigned number, that's ok.
// We want to see if it "looks like" a small signed number.
int32_t ImmValue = RawImmValue;
// For 'set' you can't use 'or' with a negative operand on V9 because
// that would splat the sign bit across the upper half of the destination
// register, whereas 'set' is defined to zero the high 32 bits.
bool IsEffectivelyImm13 =
IsImm && ((is64Bit() ? 0 : -4096) <= ImmValue && ImmValue < 4096);
const MCExpr *ValExpr;
if (IsImm)
ValExpr = MCConstantExpr::create(ImmValue, getContext());
else
ValExpr = MCValOp.getExpr();
MCOperand PrevReg = MCOperand::createReg(Sparc::G0);
// If not just a signed imm13 value, then either we use a 'sethi' with a
// following 'or', or a 'sethi' by itself if there are no more 1 bits.
// In either case, start with the 'sethi'.
if (!IsEffectivelyImm13) {
MCInst TmpInst;
const MCExpr *Expr = adjustPICRelocation(SparcMCExpr::VK_Sparc_HI, ValExpr);
TmpInst.setLoc(IDLoc);
TmpInst.setOpcode(SP::SETHIi);
TmpInst.addOperand(MCRegOp);
TmpInst.addOperand(MCOperand::createExpr(Expr));
Instructions.push_back(TmpInst);
PrevReg = MCRegOp;
}
// The low bits require touching in 3 cases:
// * A non-immediate value will always require both instructions.
// * An effectively imm13 value needs only an 'or' instruction.
// * Otherwise, an immediate that is not effectively imm13 requires the
// 'or' only if bits remain after clearing the 22 bits that 'sethi' set.
// If the low bits are known zeros, there's nothing to do.
// In the second case, and only in that case, must we NOT clear
// bits of the immediate value via the %lo() assembler function.
// Note also, the 'or' instruction doesn't mind a large value in the case
// where the operand to 'set' was 0xFFFFFzzz - it does exactly what you mean.
if (!IsImm || IsEffectivelyImm13 || (ImmValue & 0x3ff)) {
MCInst TmpInst;
const MCExpr *Expr;
if (IsEffectivelyImm13)
Expr = ValExpr;
else
Expr = adjustPICRelocation(SparcMCExpr::VK_Sparc_LO, ValExpr);
TmpInst.setLoc(IDLoc);
TmpInst.setOpcode(SP::ORri);
TmpInst.addOperand(MCRegOp);
TmpInst.addOperand(PrevReg);
TmpInst.addOperand(MCOperand::createExpr(Expr));
Instructions.push_back(TmpInst);
}
return false;
}
bool X86MCInstrAnalysis::isDependencyBreaking(const MCSubtargetInfo &STI,
const MCInst &Inst) const {
if (STI.getCPU() == "btver2") {
// Reference: Agner Fog's microarchitecture.pdf - Section 20 "AMD Bobcat and
// Jaguar pipeline", subsection 8 "Dependency-breaking instructions".
switch (Inst.getOpcode()) {
default:
return false;
case X86::SUB32rr:
case X86::SUB64rr:
case X86::SBB32rr:
case X86::SBB64rr:
case X86::XOR32rr:
case X86::XOR64rr:
case X86::XORPSrr:
case X86::XORPDrr:
case X86::VXORPSrr:
case X86::VXORPDrr:
case X86::ANDNPSrr:
case X86::VANDNPSrr:
case X86::ANDNPDrr:
case X86::VANDNPDrr:
case X86::PXORrr:
case X86::VPXORrr:
case X86::PANDNrr:
case X86::VPANDNrr:
case X86::PSUBBrr:
case X86::PSUBWrr:
case X86::PSUBDrr:
case X86::PSUBQrr:
case X86::VPSUBBrr:
case X86::VPSUBWrr:
case X86::VPSUBDrr:
case X86::VPSUBQrr:
case X86::PCMPEQBrr:
case X86::PCMPEQWrr:
case X86::PCMPEQDrr:
case X86::PCMPEQQrr:
case X86::VPCMPEQBrr:
case X86::VPCMPEQWrr:
case X86::VPCMPEQDrr:
case X86::VPCMPEQQrr:
case X86::PCMPGTBrr:
case X86::PCMPGTWrr:
case X86::PCMPGTDrr:
case X86::PCMPGTQrr:
case X86::VPCMPGTBrr:
case X86::VPCMPGTWrr:
case X86::VPCMPGTDrr:
case X86::VPCMPGTQrr:
case X86::MMX_PXORirr:
case X86::MMX_PANDNirr:
case X86::MMX_PSUBBirr:
case X86::MMX_PSUBDirr:
case X86::MMX_PSUBQirr:
case X86::MMX_PSUBWirr:
case X86::MMX_PCMPGTBirr:
case X86::MMX_PCMPGTDirr:
case X86::MMX_PCMPGTWirr:
case X86::MMX_PCMPEQBirr:
case X86::MMX_PCMPEQDirr:
case X86::MMX_PCMPEQWirr:
return Inst.getOperand(1).getReg() == Inst.getOperand(2).getReg();
case X86::CMP32rr:
case X86::CMP64rr:
return Inst.getOperand(0).getReg() == Inst.getOperand(1).getReg();
}
}
return false;
}
/// \brief Simplify things like MOV32rm to MOV32o32a.
static void SimplifyShortMoveForm(MCInst &Inst, unsigned Opcode) {
bool IsStore = Inst.getOperand(0).isReg() && Inst.getOperand(1).isReg();
unsigned AddrBase = IsStore;
unsigned RegOp = IsStore ? 0 : 5;
unsigned AddrOp = AddrBase + 3;
assert(Inst.getNumOperands() == 6 && Inst.getOperand(RegOp).isReg() &&
Inst.getOperand(AddrBase + 0).isReg() && // base
Inst.getOperand(AddrBase + 1).isImm() && // scale
Inst.getOperand(AddrBase + 2).isReg() && // index register
(Inst.getOperand(AddrOp).isExpr() || // address
Inst.getOperand(AddrOp).isImm())&&
Inst.getOperand(AddrBase + 4).isReg() && // segment
"Unexpected instruction!");
// Check whether the destination register can be fixed.
unsigned Reg = Inst.getOperand(RegOp).getReg();
if (Reg != X86::AL && Reg != X86::AX && Reg != X86::EAX && Reg != X86::RAX)
return;
// Check whether this is an absolute address.
// FIXME: We know TLVP symbol refs aren't, but there should be a better way
// to do this here.
bool Absolute = true;
if (Inst.getOperand(AddrOp).isExpr()) {
const MCExpr *MCE = Inst.getOperand(AddrOp).getExpr();
if (const MCSymbolRefExpr *SRE = dyn_cast<MCSymbolRefExpr>(MCE))
if (SRE->getKind() == MCSymbolRefExpr::VK_TLVP)
Absolute = false;
}
if (Absolute &&
(Inst.getOperand(AddrBase + 0).getReg() != 0 ||
Inst.getOperand(AddrBase + 2).getReg() != 0 ||
Inst.getOperand(AddrBase + 4).getReg() != 0 ||
Inst.getOperand(AddrBase + 1).getImm() != 1))
return;
// If so, rewrite the instruction.
MCOperand Saved = Inst.getOperand(AddrOp);
Inst = MCInst();
Inst.setOpcode(Opcode);
Inst.addOperand(Saved);
}
void X86MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
OutMI.setOpcode(MI->getOpcode());
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
MCOperand MCOp;
switch (MO.getType()) {
default:
MI->dump();
llvm_unreachable("unknown operand type");
case MachineOperand::MO_Register:
// Ignore all implicit register operands.
if (MO.isImplicit()) continue;
MCOp = MCOperand::CreateReg(MO.getReg());
break;
case MachineOperand::MO_Immediate:
MCOp = MCOperand::CreateImm(MO.getImm());
break;
case MachineOperand::MO_MachineBasicBlock:
MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
MO.getMBB()->getSymbol(), Ctx));
break;
case MachineOperand::MO_GlobalAddress:
MCOp = LowerSymbolOperand(MO, GetSymbolFromOperand(MO));
break;
case MachineOperand::MO_ExternalSymbol:
MCOp = LowerSymbolOperand(MO, GetSymbolFromOperand(MO));
break;
case MachineOperand::MO_JumpTableIndex:
MCOp = LowerSymbolOperand(MO, AsmPrinter.GetJTISymbol(MO.getIndex()));
break;
case MachineOperand::MO_ConstantPoolIndex:
MCOp = LowerSymbolOperand(MO, AsmPrinter.GetCPISymbol(MO.getIndex()));
break;
case MachineOperand::MO_BlockAddress:
MCOp = LowerSymbolOperand(MO,
AsmPrinter.GetBlockAddressSymbol(MO.getBlockAddress()));
break;
}
OutMI.addOperand(MCOp);
}
// Handle a few special cases to eliminate operand modifiers.
switch (OutMI.getOpcode()) {
case X86::LEA64_32r: // Handle 'subreg rewriting' for the lea64_32mem operand.
lower_lea64_32mem(&OutMI, 1);
break;
case X86::MOVZX16rr8: LowerSubReg32_Op0(OutMI, X86::MOVZX32rr8); break;
case X86::MOVZX16rm8: LowerSubReg32_Op0(OutMI, X86::MOVZX32rm8); break;
case X86::MOVSX16rr8: LowerSubReg32_Op0(OutMI, X86::MOVSX32rr8); break;
case X86::MOVSX16rm8: LowerSubReg32_Op0(OutMI, X86::MOVSX32rm8); break;
case X86::MOVZX64rr32: LowerSubReg32_Op0(OutMI, X86::MOV32rr); break;
case X86::MOVZX64rm32: LowerSubReg32_Op0(OutMI, X86::MOV32rm); break;
case X86::MOV64ri64i32: LowerSubReg32_Op0(OutMI, X86::MOV32ri); break;
case X86::MOVZX64rr8: LowerSubReg32_Op0(OutMI, X86::MOVZX32rr8); break;
case X86::MOVZX64rm8: LowerSubReg32_Op0(OutMI, X86::MOVZX32rm8); break;
case X86::MOVZX64rr16: LowerSubReg32_Op0(OutMI, X86::MOVZX32rr16); break;
case X86::MOVZX64rm16: LowerSubReg32_Op0(OutMI, X86::MOVZX32rm16); break;
case X86::SETB_C8r: LowerUnaryToTwoAddr(OutMI, X86::SBB8rr); break;
case X86::SETB_C16r: LowerUnaryToTwoAddr(OutMI, X86::SBB16rr); break;
case X86::SETB_C32r: LowerUnaryToTwoAddr(OutMI, X86::SBB32rr); break;
case X86::SETB_C64r: LowerUnaryToTwoAddr(OutMI, X86::SBB64rr); break;
case X86::MOV8r0: LowerUnaryToTwoAddr(OutMI, X86::XOR8rr); break;
case X86::MOV32r0: LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); break;
case X86::MMX_V_SET0: LowerUnaryToTwoAddr(OutMI, X86::MMX_PXORrr); break;
case X86::MMX_V_SETALLONES:
LowerUnaryToTwoAddr(OutMI, X86::MMX_PCMPEQDrr); break;
case X86::FsFLD0SS: LowerUnaryToTwoAddr(OutMI, X86::PXORrr); break;
case X86::FsFLD0SD: LowerUnaryToTwoAddr(OutMI, X86::PXORrr); break;
case X86::V_SET0PS: LowerUnaryToTwoAddr(OutMI, X86::XORPSrr); break;
case X86::V_SET0PD: LowerUnaryToTwoAddr(OutMI, X86::XORPDrr); break;
case X86::V_SET0PI: LowerUnaryToTwoAddr(OutMI, X86::PXORrr); break;
case X86::V_SETALLONES: LowerUnaryToTwoAddr(OutMI, X86::PCMPEQDrr); break;
case X86::MOV16r0:
LowerSubReg32_Op0(OutMI, X86::MOV32r0); // MOV16r0 -> MOV32r0
LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); // MOV32r0 -> XOR32rr
break;
case X86::MOV64r0:
LowerSubReg32_Op0(OutMI, X86::MOV32r0); // MOV64r0 -> MOV32r0
LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); // MOV32r0 -> XOR32rr
break;
// TAILJMPr, TAILJMPr64, CALL64r, CALL64pcrel32 - These instructions have
// register inputs modeled as normal uses instead of implicit uses. As such,
// truncate off all but the first operand (the callee). FIXME: Change isel.
case X86::TAILJMPr:
case X86::TAILJMPr64:
case X86::CALL64r:
case X86::CALL64pcrel32: {
unsigned Opcode = OutMI.getOpcode();
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
// TAILJMPd, TAILJMPd64 - Lower to the correct jump instructions.
case X86::TAILJMPd:
case X86::TAILJMPd64: {
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(X86::TAILJMP_1);
OutMI.addOperand(Saved);
break;
}
// The assembler backend wants to see branches in their small form and relax
// them to their large form. The JIT can only handle the large form because
// it does not do relaxation. For now, translate the large form to the
// small one here.
case X86::JMP_4: OutMI.setOpcode(X86::JMP_1); break;
case X86::JO_4: OutMI.setOpcode(X86::JO_1); break;
case X86::JNO_4: OutMI.setOpcode(X86::JNO_1); break;
case X86::JB_4: OutMI.setOpcode(X86::JB_1); break;
case X86::JAE_4: OutMI.setOpcode(X86::JAE_1); break;
case X86::JE_4: OutMI.setOpcode(X86::JE_1); break;
case X86::JNE_4: OutMI.setOpcode(X86::JNE_1); break;
case X86::JBE_4: OutMI.setOpcode(X86::JBE_1); break;
case X86::JA_4: OutMI.setOpcode(X86::JA_1); break;
case X86::JS_4: OutMI.setOpcode(X86::JS_1); break;
case X86::JNS_4: OutMI.setOpcode(X86::JNS_1); break;
case X86::JP_4: OutMI.setOpcode(X86::JP_1); break;
case X86::JNP_4: OutMI.setOpcode(X86::JNP_1); break;
case X86::JL_4: OutMI.setOpcode(X86::JL_1); break;
case X86::JGE_4: OutMI.setOpcode(X86::JGE_1); break;
case X86::JLE_4: OutMI.setOpcode(X86::JLE_1); break;
case X86::JG_4: OutMI.setOpcode(X86::JG_1); break;
// We don't currently select the correct instruction form for instructions
// which have a short %eax, etc. form. Handle this by custom lowering, for
// now.
//
// Note, we are currently not handling the following instructions:
// MOV64ao8, MOV64o8a
// XCHG16ar, XCHG32ar, XCHG64ar
case X86::MOV8mr_NOREX:
case X86::MOV8mr: SimplifyShortMoveForm(OutMI, X86::MOV8ao8); break;
case X86::MOV8rm_NOREX:
case X86::MOV8rm: SimplifyShortMoveForm(OutMI, X86::MOV8o8a); break;
case X86::MOV16mr: SimplifyShortMoveForm(OutMI, X86::MOV16ao16); break;
case X86::MOV16rm: SimplifyShortMoveForm(OutMI, X86::MOV16o16a); break;
case X86::MOV32mr: SimplifyShortMoveForm(OutMI, X86::MOV32ao32); break;
case X86::MOV32rm: SimplifyShortMoveForm(OutMI, X86::MOV32o32a); break;
case X86::MOV64mr: SimplifyShortMoveForm(OutMI, X86::MOV64ao64); break;
case X86::MOV64rm: SimplifyShortMoveForm(OutMI, X86::MOV64o64a); break;
case X86::ADC8ri: SimplifyShortImmForm(OutMI, X86::ADC8i8); break;
case X86::ADC16ri: SimplifyShortImmForm(OutMI, X86::ADC16i16); break;
case X86::ADC32ri: SimplifyShortImmForm(OutMI, X86::ADC32i32); break;
case X86::ADC64ri32: SimplifyShortImmForm(OutMI, X86::ADC64i32); break;
case X86::ADD8ri: SimplifyShortImmForm(OutMI, X86::ADD8i8); break;
case X86::ADD16ri: SimplifyShortImmForm(OutMI, X86::ADD16i16); break;
case X86::ADD32ri: SimplifyShortImmForm(OutMI, X86::ADD32i32); break;
case X86::ADD64ri32: SimplifyShortImmForm(OutMI, X86::ADD64i32); break;
case X86::AND8ri: SimplifyShortImmForm(OutMI, X86::AND8i8); break;
case X86::AND16ri: SimplifyShortImmForm(OutMI, X86::AND16i16); break;
case X86::AND32ri: SimplifyShortImmForm(OutMI, X86::AND32i32); break;
case X86::AND64ri32: SimplifyShortImmForm(OutMI, X86::AND64i32); break;
case X86::CMP8ri: SimplifyShortImmForm(OutMI, X86::CMP8i8); break;
case X86::CMP16ri: SimplifyShortImmForm(OutMI, X86::CMP16i16); break;
case X86::CMP32ri: SimplifyShortImmForm(OutMI, X86::CMP32i32); break;
case X86::CMP64ri32: SimplifyShortImmForm(OutMI, X86::CMP64i32); break;
case X86::OR8ri: SimplifyShortImmForm(OutMI, X86::OR8i8); break;
case X86::OR16ri: SimplifyShortImmForm(OutMI, X86::OR16i16); break;
case X86::OR32ri: SimplifyShortImmForm(OutMI, X86::OR32i32); break;
case X86::OR64ri32: SimplifyShortImmForm(OutMI, X86::OR64i32); break;
case X86::SBB8ri: SimplifyShortImmForm(OutMI, X86::SBB8i8); break;
case X86::SBB16ri: SimplifyShortImmForm(OutMI, X86::SBB16i16); break;
case X86::SBB32ri: SimplifyShortImmForm(OutMI, X86::SBB32i32); break;
case X86::SBB64ri32: SimplifyShortImmForm(OutMI, X86::SBB64i32); break;
case X86::SUB8ri: SimplifyShortImmForm(OutMI, X86::SUB8i8); break;
case X86::SUB16ri: SimplifyShortImmForm(OutMI, X86::SUB16i16); break;
case X86::SUB32ri: SimplifyShortImmForm(OutMI, X86::SUB32i32); break;
case X86::SUB64ri32: SimplifyShortImmForm(OutMI, X86::SUB64i32); break;
case X86::TEST8ri: SimplifyShortImmForm(OutMI, X86::TEST8i8); break;
case X86::TEST16ri: SimplifyShortImmForm(OutMI, X86::TEST16i16); break;
case X86::TEST32ri: SimplifyShortImmForm(OutMI, X86::TEST32i32); break;
case X86::TEST64ri32: SimplifyShortImmForm(OutMI, X86::TEST64i32); break;
case X86::XOR8ri: SimplifyShortImmForm(OutMI, X86::XOR8i8); break;
case X86::XOR16ri: SimplifyShortImmForm(OutMI, X86::XOR16i16); break;
case X86::XOR32ri: SimplifyShortImmForm(OutMI, X86::XOR32i32); break;
case X86::XOR64ri32: SimplifyShortImmForm(OutMI, X86::XOR64i32); break;
}
}
unsigned Disassembler::decodeInstruction(unsigned Address,
MachineBasicBlock *Block) {
// Disassemble instruction
const MCDisassembler *DA = MC->getMCDisassembler();
uint64_t InstSize;
MCInst *Inst = new MCInst();
StringRef Bytes;
if (!(DA->getInstruction(*Inst, InstSize, *CurSectionMemory, Address,
nulls(), nulls()))) {
printError("Unknown instruction encountered, instruction decode failed!");
return 1;
// Instructions[Address] = NULL;
// Block->push_back(NULL);
// TODO: Replace with default size for each target.
// return 1;
// outs() << format("%8" PRIx64 ":\t", SectAddr + Index);
// Dism->rawBytesToString(StringRef(Bytes.data() + Index, Size));
// outs() << " unkn\n";
}
Instructions[Address] = Inst;
// Recover Instruction information
const MCInstrInfo *MII = MC->getMCInstrInfo();
MCInstrDesc *MCID = new MCInstrDesc(MII->get(Inst->getOpcode()));
MCID->Size = InstSize;
// Check if the instruction can load to program counter and mark it as a Ret
// FIXME: Better analysis would be to see if the PC value references memory
// sent as a parameter or set locally in the function, but that would need to
// happen after decompilation. In either case, this is definitely a BB
// terminator or branch!
if (MCID->mayLoad()
&& MCID->mayAffectControlFlow(*Inst, *MC->getMCRegisterInfo())) {
MCID->Flags |= (1 << MCID::Return);
MCID->Flags |= (1 << MCID::Terminator);
}
// Recover MachineInstr representation
DebugLoc *Location = setDebugLoc(Address);
MachineInstrBuilder MIB = BuildMI(Block, *Location, *MCID);
unsigned int numDefs = MCID->getNumDefs();
for (unsigned int i = 0; i < Inst->getNumOperands(); i++) {
MCOperand MCO = Inst->getOperand(i);
// FIXME: This hack is a workaround for the assert in MachineInstr.cpp:653,
// where OpNo >= MCID->getNumOperands()...
if (i >= MCID->getNumOperands() && !(MCID->isVariadic()))
break;
if (MCO.isReg()) {
unsigned flags = 0;
// Defs always start at the beginning of the operands list,
// unfortunately BuildMI doesn't set default define flags so we have
// to do it manually here.
// NOTE: This should always be true, but might not be if operands list
// is not populated correctly by the MC Backend for the target.
if (i < numDefs) {
flags |= RegState::Define;
}
// NOTE: No need to worry about imp defs and uses, as these are already
// specificed in the MCID attached to the MachineInst object.
MIB.addReg(MCO.getReg(), flags);
continue;
}
if (MCO.isImm()) {
MIB.addImm(MCO.getImm());
continue;
}
//else if (MCO.isFPImm()) MIB.addFPImm(MCO.getFPImm());
if (MCO.isExpr()) {
MCOperandInfo MCOpInfo = MCID->OpInfo[i];
switch (MCOpInfo.OperandType) {
case MCOI::OPERAND_MEMORY:
case MCOI::OPERAND_PCREL:
case MCOI::OPERAND_UNKNOWN:
default:
printError("Unknown how to handle this Expression at this time.");
}
}
printError("Unknown how to handle Operand!");
}
// NOTE: I tried MCOpInfo here, and it appearst o be NULL
// ... at least for ARM.
unsigned flags = 0;
if (MCID->mayLoad())
flags |= MachineMemOperand::MOLoad;
if (MCID->mayStore())
flags |= MachineMemOperand::MOStore;
if (flags != 0) {
// Constant* cInt = ConstantInt::get(Type::getInt64Ty(ctx), MCO.getImm());
// Value *Val = ConstantExpr::getIntToPtr(cInt,
// PointerType::getUnqual(Type::getInt32Ty(ctx)));
// FIXME: note size of 4 is known to be bad for
// some targets
//Copy & paste set getImm to zero
MachineMemOperand* MMO = new MachineMemOperand(
MachinePointerInfo(), flags, 4, 0); //MCO.getImm()
MIB.addMemOperand(MMO);
//outs() << "Name: " << MII->getName(Inst->getOpcode()) << " Flags: " << flags << "\n";
}
// Note: I don't know why they decided instruction size needed to be 64 bits,
// but the following conversion shouldn't be an issue.
return ((unsigned)InstSize);
}
void ARM64InstPrinter::printInst(const MCInst *MI, raw_ostream &O,
StringRef Annot) {
// Check for special encodings and print the cannonical alias instead.
unsigned Opcode = MI->getOpcode();
// Move to/from SP is encoded as add of #0.
// "add <Wd|WSP>, <Wn|WSP>, #0" is "mov <Wd|WSP>, <Wn|WSP>"
// "add <Xd|SP>, <Xd|SP>, #0" is "mov <Xd|SP>, <Xn|SP>"
if (Opcode == ARM64::ADDXri || Opcode == ARM64::ADDWri) {
const MCOperand &Op0 = MI->getOperand(0);
const MCOperand &Op1 = MI->getOperand(1);
if (MI->getOperand(2).isImm() &&
MI->getOperand(2).getImm() == 0 &&
MI->getOperand(3).getImm() == 0 &&
(Op0.getReg() == ARM64::SP || Op0.getReg() == ARM64::WSP ||
Op1.getReg() == ARM64::SP || Op1.getReg() == ARM64::WSP)) {
O << '\t' << "mov" << '\t'
<< getRegisterName(Op0.getReg()) << ", "
<< getRegisterName(Op1.getReg());
printAnnotation(O, Annot);
return;
}
}
// Register copy is encoded as ORR.
// "orr Wd, WZR, Wn" is "mov Wd, Wn"
// "orr Xd, XZR, Xn" is "mov Xd, Xn"
if (Opcode == ARM64::ORRXrs || Opcode == ARM64::ORRWrs) {
const MCOperand &Op0 = MI->getOperand(0);
const MCOperand &Op1 = MI->getOperand(1);
const MCOperand &Op2 = MI->getOperand(2);
if ((Op1.getReg() == ARM64::WZR || Op1.getReg() == ARM64::XZR) &&
MI->getOperand(3).getImm() == 0) {
O << '\t' << "mov" << '\t'
<< getRegisterName(Op0.getReg()) << ", "
<< getRegisterName(Op2.getReg());
printAnnotation(O, Annot);
return;
}
}
if (Opcode == ARM64::SYS || Opcode == ARM64::SYSxt)
if (printSysAlias(MI, O)) {
printAnnotation(O, Annot);
return;
}
// TBZ/TBNZ should print the register operand as a Wreg if the bit
// number is < 32.
if ((Opcode == ARM64::TBNZ || Opcode == ARM64::TBZ) &&
MI->getOperand(1).getImm() < 32) {
MCInst newMI = *MI;
unsigned Reg = MI->getOperand(0).getReg();
newMI.getOperand(0).setReg(getWRegFromXReg(Reg));
printInstruction(&newMI, O);
printAnnotation(O, Annot);
return;
}
// SBFM/UBFM should print to a nicer aliased form if possible.
if (Opcode == ARM64::SBFMXri || Opcode == ARM64::SBFMWri ||
Opcode == ARM64::UBFMXri || Opcode == ARM64::UBFMWri) {
const MCOperand &Op0 = MI->getOperand(0);
const MCOperand &Op1 = MI->getOperand(1);
const MCOperand &Op2 = MI->getOperand(2);
const MCOperand &Op3 = MI->getOperand(3);
if (Op2.isImm() && Op2.getImm() == 0 && Op3.isImm()) {
bool IsSigned = (Opcode == ARM64::SBFMXri || Opcode == ARM64::SBFMWri);
const char *AsmMnemonic = 0;
switch (Op3.getImm()) {
default: break;
case 7:
AsmMnemonic = IsSigned ? "sxtb" : "uxtb";
break;
case 15:
AsmMnemonic = IsSigned ? "sxth" : "uxth";
break;
case 31:
AsmMnemonic = IsSigned ? "sxtw" : "uxtw";
break;
}
if (AsmMnemonic) {
O << '\t' << AsmMnemonic << '\t'
<< getRegisterName(Op0.getReg())
<< ", " << getRegisterName(Op1.getReg());
printAnnotation(O, Annot);
return;
}
}
// All immediate shifts are aliases, implemented using the Bitfield
// instruction. In all cases the immediate shift amount shift must be in
// the range 0 to (reg.size -1).
if (Op2.isImm() && Op3.isImm()) {
const char *AsmMnemonic = 0;
int shift = 0;
int64_t immr = Op2.getImm();
int64_t imms = Op3.getImm();
if (Opcode == ARM64::UBFMWri && imms != 0x1F && ((imms + 1) == immr)) {
AsmMnemonic = "lsl";
shift = 31 - imms;
} else if (Opcode == ARM64::UBFMXri && imms != 0x3f &&
((imms + 1 == immr))) {
AsmMnemonic = "lsl";
shift = 63 - imms;
} else if (Opcode == ARM64::UBFMWri && imms == 0x1f) {
AsmMnemonic = "lsr";
shift = immr;
} else if (Opcode == ARM64::UBFMXri && imms == 0x3f) {
AsmMnemonic = "lsr";
shift = immr;
} else if (Opcode == ARM64::SBFMWri && imms == 0x1f) {
AsmMnemonic = "asr";
shift = immr;
} else if (Opcode == ARM64::SBFMXri && imms == 0x3f) {
AsmMnemonic = "asr";
shift = immr;
}
if (AsmMnemonic) {
O << '\t' << AsmMnemonic << '\t'
<< getRegisterName(Op0.getReg())
<< ", " << getRegisterName(Op1.getReg())
<< ", #" << shift;
printAnnotation(O, Annot);
return;
}
}
}
// Symbolic operands for MOVZ, MOVN and MOVK already imply a shift
// (e.g. :gottprel_g1: is always going to be "lsl #16") so it should not be
// printed.
if ((Opcode == ARM64::MOVZXi || Opcode == ARM64::MOVZWi ||
Opcode == ARM64::MOVNXi || Opcode == ARM64::MOVNWi) &&
MI->getOperand(1).isExpr()) {
if (Opcode == ARM64::MOVZXi || Opcode == ARM64::MOVZWi)
O << "\tmovz\t";
else
O << "\tmovn\t";
O << getRegisterName(MI->getOperand(0).getReg()) << ", #"
<< *MI->getOperand(1).getExpr();
return;
}
if ((Opcode == ARM64::MOVKXi || Opcode == ARM64::MOVKWi) &&
MI->getOperand(2).isExpr()) {
O << "\tmovk\t"
<< getRegisterName(MI->getOperand(0).getReg()) << ", #"
<< *MI->getOperand(2).getExpr();
return;
}
// HINT #[0-5] should print as nop, yield, wfe, wfi, sev, and sevl,
// respectively.
if (Opcode == ARM64::HINT && MI->getOperand(0).getImm() < 6) {
static const char *Mnemonic[] = {
"nop", "yield", "wfe", "wfi", "sev", "sevl"
};
O << '\t' << Mnemonic[MI->getOperand(0).getImm()];
printAnnotation(O, Annot);
return;
}
// ANDS WZR, Wn, #imm ==> TST Wn, #imm
// ANDS XZR, Xn, #imm ==> TST Xn, #imm
if (Opcode == ARM64::ANDSWri && MI->getOperand(0).getReg() == ARM64::WZR) {
O << "\ttst\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printLogicalImm32(MI, 2, O);
return;
}
if (Opcode == ARM64::ANDSXri && MI->getOperand(0).getReg() == ARM64::XZR) {
O << "\ttst\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printLogicalImm64(MI, 2, O);
return;
}
// ANDS WZR, Wn, Wm{, lshift #imm} ==> TST Wn{, lshift #imm}
// ANDS XZR, Xn, Xm{, lshift #imm} ==> TST Xn{, lshift #imm}
if ((Opcode == ARM64::ANDSWrs && MI->getOperand(0).getReg() == ARM64::WZR) ||
(Opcode == ARM64::ANDSXrs && MI->getOperand(0).getReg() == ARM64::XZR)) {
O << "\ttst\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printShiftedRegister(MI, 2, O);
return;
}
// SUBS WZR, Wn, #imm ==> CMP Wn, #imm
// SUBS XZR, Xn, #imm ==> CMP Xn, #imm
if ((Opcode == ARM64::SUBSWri && MI->getOperand(0).getReg() == ARM64::WZR) ||
(Opcode == ARM64::SUBSXri && MI->getOperand(0).getReg() == ARM64::XZR)) {
O << "\tcmp\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printAddSubImm(MI, 2 ,O);
return;
}
// SUBS WZR, Wn, Wm{, lshift #imm} ==> CMP Wn, Wm{, lshift #imm}
// SUBS XZR, Xn, Xm{, lshift #imm} ==> CMP Xn, Xm{, lshift #imm}
if ((Opcode == ARM64::SUBSWrs && MI->getOperand(0).getReg() == ARM64::WZR) ||
(Opcode == ARM64::SUBSXrs && MI->getOperand(0).getReg() == ARM64::XZR)) {
O << "\tcmp\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printShiftedRegister(MI, 2, O);
return;
}
// SUBS XZR, Xn, Wm, uxtb #imm ==> CMP Xn, uxtb #imm
if (Opcode == ARM64::SUBSXrx && MI->getOperand(0).getReg() == ARM64::XZR) {
O << "\tcmp\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printExtendedRegister(MI, 2, O);
return;
}
// SUBS XZR, Xn, Xm, uxtx #imm ==> CMP Xn, uxtb #imm
if (Opcode == ARM64::SUBSXrx64 && MI->getOperand(0).getReg() == ARM64::XZR) {
O << "\tcmp\t" << getRegisterName(MI->getOperand(1).getReg()) << ", "
<< getRegisterName(MI->getOperand(2).getReg());
printExtend(MI, 3, O);
return;
}
// ADDS WZR, Wn, #imm ==> CMN Wn, #imm
// ADDS XZR, Xn, #imm ==> CMN Xn, #imm
if ((Opcode == ARM64::ADDSWri && MI->getOperand(0).getReg() == ARM64::WZR) ||
(Opcode == ARM64::ADDSXri && MI->getOperand(0).getReg() == ARM64::XZR)) {
O << "\tcmn\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printAddSubImm(MI, 2 ,O);
return;
}
// ADDS WZR, Wn, Wm{, lshift #imm} ==> CMN Wn, Wm{, lshift #imm}
// ADDS XZR, Xn, Xm{, lshift #imm} ==> CMN Xn, Xm{, lshift #imm}
if ((Opcode == ARM64::ADDSWrs && MI->getOperand(0).getReg() == ARM64::WZR) ||
(Opcode == ARM64::ADDSXrs && MI->getOperand(0).getReg() == ARM64::XZR)) {
O << "\tcmn\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printShiftedRegister(MI, 2, O);
return;
}
// ADDS XZR, Xn, Wm, uxtb #imm ==> CMN Xn, uxtb #imm
if (Opcode == ARM64::ADDSXrx && MI->getOperand(0).getReg() == ARM64::XZR) {
O << "\tcmn\t" << getRegisterName(MI->getOperand(1).getReg()) << ", ";
printExtendedRegister(MI, 2, O);
return;
}
// ADDS XZR, Xn, Xm, uxtx #imm ==> CMN Xn, uxtb #imm
if (Opcode == ARM64::ADDSXrx64 && MI->getOperand(0).getReg() == ARM64::XZR) {
O << "\tcmn\t" << getRegisterName(MI->getOperand(1).getReg()) << ", "
<< getRegisterName(MI->getOperand(2).getReg());
printExtend(MI, 3, O);
return;
}
// ORN Wd, WZR, Wm{, lshift #imm} ==> MVN Wd, Wm{, lshift #imm}
// ORN Xd, XZR, Xm{, lshift #imm} ==> MVN Xd, Xm{, lshift #imm}
if ((Opcode == ARM64::ORNWrs && MI->getOperand(1).getReg() == ARM64::WZR) ||
(Opcode == ARM64::ORNXrs && MI->getOperand(1).getReg() == ARM64::XZR)) {
O << "\tmvn\t" << getRegisterName(MI->getOperand(0).getReg()) << ", ";
printShiftedRegister(MI, 2, O);
return;
}
printInstruction(MI, O);
printAnnotation(O, Annot);
}
void X86MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
OutMI.setOpcode(MI->getOpcode());
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
MCOperand MCOp;
switch (MO.getType()) {
default:
MI->dump();
llvm_unreachable("unknown operand type");
case MachineOperand::MO_Register:
// Ignore all implicit register operands.
if (MO.isImplicit()) continue;
MCOp = MCOperand::CreateReg(MO.getReg());
break;
case MachineOperand::MO_Immediate:
MCOp = MCOperand::CreateImm(MO.getImm());
break;
case MachineOperand::MO_MachineBasicBlock:
case MachineOperand::MO_GlobalAddress:
case MachineOperand::MO_ExternalSymbol:
MCOp = LowerSymbolOperand(MO, GetSymbolFromOperand(MO));
break;
case MachineOperand::MO_JumpTableIndex:
MCOp = LowerSymbolOperand(MO, AsmPrinter.GetJTISymbol(MO.getIndex()));
break;
case MachineOperand::MO_ConstantPoolIndex:
MCOp = LowerSymbolOperand(MO, AsmPrinter.GetCPISymbol(MO.getIndex()));
break;
case MachineOperand::MO_BlockAddress:
MCOp = LowerSymbolOperand(MO,
AsmPrinter.GetBlockAddressSymbol(MO.getBlockAddress()));
break;
case MachineOperand::MO_RegisterMask:
// Ignore call clobbers.
continue;
}
OutMI.addOperand(MCOp);
}
// Handle a few special cases to eliminate operand modifiers.
ReSimplify:
switch (OutMI.getOpcode()) {
case X86::LEA64_32r:
case X86::LEA64r:
case X86::LEA16r:
case X86::LEA32r:
// LEA should have a segment register, but it must be empty.
assert(OutMI.getNumOperands() == 1+X86::AddrNumOperands &&
"Unexpected # of LEA operands");
assert(OutMI.getOperand(1+X86::AddrSegmentReg).getReg() == 0 &&
"LEA has segment specified!");
break;
case X86::MOV32ri64:
OutMI.setOpcode(X86::MOV32ri);
break;
// Commute operands to get a smaller encoding by using VEX.R instead of VEX.B
// if one of the registers is extended, but other isn't.
case X86::VMOVAPDrr:
case X86::VMOVAPDYrr:
case X86::VMOVAPSrr:
case X86::VMOVAPSYrr:
case X86::VMOVDQArr:
case X86::VMOVDQAYrr:
case X86::VMOVDQUrr:
case X86::VMOVDQUYrr:
case X86::VMOVUPDrr:
case X86::VMOVUPDYrr:
case X86::VMOVUPSrr:
case X86::VMOVUPSYrr: {
if (!X86II::isX86_64ExtendedReg(OutMI.getOperand(0).getReg()) &&
X86II::isX86_64ExtendedReg(OutMI.getOperand(1).getReg())) {
unsigned NewOpc;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::VMOVAPDrr: NewOpc = X86::VMOVAPDrr_REV; break;
case X86::VMOVAPDYrr: NewOpc = X86::VMOVAPDYrr_REV; break;
case X86::VMOVAPSrr: NewOpc = X86::VMOVAPSrr_REV; break;
case X86::VMOVAPSYrr: NewOpc = X86::VMOVAPSYrr_REV; break;
case X86::VMOVDQArr: NewOpc = X86::VMOVDQArr_REV; break;
case X86::VMOVDQAYrr: NewOpc = X86::VMOVDQAYrr_REV; break;
case X86::VMOVDQUrr: NewOpc = X86::VMOVDQUrr_REV; break;
case X86::VMOVDQUYrr: NewOpc = X86::VMOVDQUYrr_REV; break;
case X86::VMOVUPDrr: NewOpc = X86::VMOVUPDrr_REV; break;
case X86::VMOVUPDYrr: NewOpc = X86::VMOVUPDYrr_REV; break;
case X86::VMOVUPSrr: NewOpc = X86::VMOVUPSrr_REV; break;
case X86::VMOVUPSYrr: NewOpc = X86::VMOVUPSYrr_REV; break;
}
OutMI.setOpcode(NewOpc);
}
break;
}
case X86::VMOVSDrr:
case X86::VMOVSSrr: {
if (!X86II::isX86_64ExtendedReg(OutMI.getOperand(0).getReg()) &&
X86II::isX86_64ExtendedReg(OutMI.getOperand(2).getReg())) {
unsigned NewOpc;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::VMOVSDrr: NewOpc = X86::VMOVSDrr_REV; break;
case X86::VMOVSSrr: NewOpc = X86::VMOVSSrr_REV; break;
}
OutMI.setOpcode(NewOpc);
}
break;
}
// TAILJMPr64, CALL64r, CALL64pcrel32 - These instructions have register
// inputs modeled as normal uses instead of implicit uses. As such, truncate
// off all but the first operand (the callee). FIXME: Change isel.
case X86::TAILJMPr64:
case X86::CALL64r:
case X86::CALL64pcrel32: {
unsigned Opcode = OutMI.getOpcode();
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
case X86::EH_RETURN:
case X86::EH_RETURN64: {
OutMI = MCInst();
OutMI.setOpcode(getRetOpcode(AsmPrinter.getSubtarget()));
break;
}
// TAILJMPd, TAILJMPd64 - Lower to the correct jump instructions.
case X86::TAILJMPr:
case X86::TAILJMPd:
case X86::TAILJMPd64: {
unsigned Opcode;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::TAILJMPr: Opcode = X86::JMP32r; break;
case X86::TAILJMPd:
case X86::TAILJMPd64: Opcode = X86::JMP_1; break;
}
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
// These are pseudo-ops for OR to help with the OR->ADD transformation. We do
// this with an ugly goto in case the resultant OR uses EAX and needs the
// short form.
case X86::ADD16rr_DB: OutMI.setOpcode(X86::OR16rr); goto ReSimplify;
case X86::ADD32rr_DB: OutMI.setOpcode(X86::OR32rr); goto ReSimplify;
case X86::ADD64rr_DB: OutMI.setOpcode(X86::OR64rr); goto ReSimplify;
case X86::ADD16ri_DB: OutMI.setOpcode(X86::OR16ri); goto ReSimplify;
case X86::ADD32ri_DB: OutMI.setOpcode(X86::OR32ri); goto ReSimplify;
case X86::ADD64ri32_DB: OutMI.setOpcode(X86::OR64ri32); goto ReSimplify;
case X86::ADD16ri8_DB: OutMI.setOpcode(X86::OR16ri8); goto ReSimplify;
case X86::ADD32ri8_DB: OutMI.setOpcode(X86::OR32ri8); goto ReSimplify;
case X86::ADD64ri8_DB: OutMI.setOpcode(X86::OR64ri8); goto ReSimplify;
// The assembler backend wants to see branches in their small form and relax
// them to their large form. The JIT can only handle the large form because
// it does not do relaxation. For now, translate the large form to the
// small one here.
case X86::JMP_4: OutMI.setOpcode(X86::JMP_1); break;
case X86::JO_4: OutMI.setOpcode(X86::JO_1); break;
case X86::JNO_4: OutMI.setOpcode(X86::JNO_1); break;
case X86::JB_4: OutMI.setOpcode(X86::JB_1); break;
case X86::JAE_4: OutMI.setOpcode(X86::JAE_1); break;
case X86::JE_4: OutMI.setOpcode(X86::JE_1); break;
case X86::JNE_4: OutMI.setOpcode(X86::JNE_1); break;
case X86::JBE_4: OutMI.setOpcode(X86::JBE_1); break;
case X86::JA_4: OutMI.setOpcode(X86::JA_1); break;
case X86::JS_4: OutMI.setOpcode(X86::JS_1); break;
case X86::JNS_4: OutMI.setOpcode(X86::JNS_1); break;
case X86::JP_4: OutMI.setOpcode(X86::JP_1); break;
case X86::JNP_4: OutMI.setOpcode(X86::JNP_1); break;
case X86::JL_4: OutMI.setOpcode(X86::JL_1); break;
case X86::JGE_4: OutMI.setOpcode(X86::JGE_1); break;
case X86::JLE_4: OutMI.setOpcode(X86::JLE_1); break;
case X86::JG_4: OutMI.setOpcode(X86::JG_1); break;
// Atomic load and store require a separate pseudo-inst because Acquire
// implies mayStore and Release implies mayLoad; fix these to regular MOV
// instructions here
case X86::ACQUIRE_MOV8rm: OutMI.setOpcode(X86::MOV8rm); goto ReSimplify;
case X86::ACQUIRE_MOV16rm: OutMI.setOpcode(X86::MOV16rm); goto ReSimplify;
case X86::ACQUIRE_MOV32rm: OutMI.setOpcode(X86::MOV32rm); goto ReSimplify;
case X86::ACQUIRE_MOV64rm: OutMI.setOpcode(X86::MOV64rm); goto ReSimplify;
case X86::RELEASE_MOV8mr: OutMI.setOpcode(X86::MOV8mr); goto ReSimplify;
case X86::RELEASE_MOV16mr: OutMI.setOpcode(X86::MOV16mr); goto ReSimplify;
case X86::RELEASE_MOV32mr: OutMI.setOpcode(X86::MOV32mr); goto ReSimplify;
case X86::RELEASE_MOV64mr: OutMI.setOpcode(X86::MOV64mr); goto ReSimplify;
case X86::RELEASE_MOV8mi: OutMI.setOpcode(X86::MOV8mi); goto ReSimplify;
case X86::RELEASE_MOV16mi: OutMI.setOpcode(X86::MOV16mi); goto ReSimplify;
case X86::RELEASE_MOV32mi: OutMI.setOpcode(X86::MOV32mi); goto ReSimplify;
case X86::RELEASE_MOV64mi32: OutMI.setOpcode(X86::MOV64mi32); goto ReSimplify;
case X86::RELEASE_ADD8mi: OutMI.setOpcode(X86::ADD8mi); goto ReSimplify;
case X86::RELEASE_ADD32mi: OutMI.setOpcode(X86::ADD32mi); goto ReSimplify;
case X86::RELEASE_ADD64mi32: OutMI.setOpcode(X86::ADD64mi32); goto ReSimplify;
case X86::RELEASE_AND8mi: OutMI.setOpcode(X86::AND8mi); goto ReSimplify;
case X86::RELEASE_AND32mi: OutMI.setOpcode(X86::AND32mi); goto ReSimplify;
case X86::RELEASE_AND64mi32: OutMI.setOpcode(X86::AND64mi32); goto ReSimplify;
case X86::RELEASE_OR8mi: OutMI.setOpcode(X86::OR8mi); goto ReSimplify;
case X86::RELEASE_OR32mi: OutMI.setOpcode(X86::OR32mi); goto ReSimplify;
case X86::RELEASE_OR64mi32: OutMI.setOpcode(X86::OR64mi32); goto ReSimplify;
case X86::RELEASE_XOR8mi: OutMI.setOpcode(X86::XOR8mi); goto ReSimplify;
case X86::RELEASE_XOR32mi: OutMI.setOpcode(X86::XOR32mi); goto ReSimplify;
case X86::RELEASE_XOR64mi32: OutMI.setOpcode(X86::XOR64mi32); goto ReSimplify;
case X86::RELEASE_INC8m: OutMI.setOpcode(X86::INC8m); goto ReSimplify;
case X86::RELEASE_INC16m: OutMI.setOpcode(X86::INC16m); goto ReSimplify;
case X86::RELEASE_INC32m: OutMI.setOpcode(X86::INC32m); goto ReSimplify;
case X86::RELEASE_INC64m: OutMI.setOpcode(X86::INC64m); goto ReSimplify;
case X86::RELEASE_DEC8m: OutMI.setOpcode(X86::DEC8m); goto ReSimplify;
case X86::RELEASE_DEC16m: OutMI.setOpcode(X86::DEC16m); goto ReSimplify;
case X86::RELEASE_DEC32m: OutMI.setOpcode(X86::DEC32m); goto ReSimplify;
case X86::RELEASE_DEC64m: OutMI.setOpcode(X86::DEC64m); goto ReSimplify;
// We don't currently select the correct instruction form for instructions
// which have a short %eax, etc. form. Handle this by custom lowering, for
// now.
//
// Note, we are currently not handling the following instructions:
// MOV64ao8, MOV64o8a
// XCHG16ar, XCHG32ar, XCHG64ar
case X86::MOV8mr_NOREX:
case X86::MOV8mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV8ao8); break;
case X86::MOV8rm_NOREX:
case X86::MOV8rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV8o8a); break;
case X86::MOV16mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV16ao16); break;
case X86::MOV16rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV16o16a); break;
case X86::MOV32mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV32ao32); break;
case X86::MOV32rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV32o32a); break;
case X86::ADC8ri: SimplifyShortImmForm(OutMI, X86::ADC8i8); break;
case X86::ADC16ri: SimplifyShortImmForm(OutMI, X86::ADC16i16); break;
case X86::ADC32ri: SimplifyShortImmForm(OutMI, X86::ADC32i32); break;
case X86::ADC64ri32: SimplifyShortImmForm(OutMI, X86::ADC64i32); break;
case X86::ADD8ri: SimplifyShortImmForm(OutMI, X86::ADD8i8); break;
case X86::ADD16ri: SimplifyShortImmForm(OutMI, X86::ADD16i16); break;
case X86::ADD32ri: SimplifyShortImmForm(OutMI, X86::ADD32i32); break;
case X86::ADD64ri32: SimplifyShortImmForm(OutMI, X86::ADD64i32); break;
case X86::AND8ri: SimplifyShortImmForm(OutMI, X86::AND8i8); break;
case X86::AND16ri: SimplifyShortImmForm(OutMI, X86::AND16i16); break;
case X86::AND32ri: SimplifyShortImmForm(OutMI, X86::AND32i32); break;
case X86::AND64ri32: SimplifyShortImmForm(OutMI, X86::AND64i32); break;
case X86::CMP8ri: SimplifyShortImmForm(OutMI, X86::CMP8i8); break;
case X86::CMP16ri: SimplifyShortImmForm(OutMI, X86::CMP16i16); break;
case X86::CMP32ri: SimplifyShortImmForm(OutMI, X86::CMP32i32); break;
case X86::CMP64ri32: SimplifyShortImmForm(OutMI, X86::CMP64i32); break;
case X86::OR8ri: SimplifyShortImmForm(OutMI, X86::OR8i8); break;
case X86::OR16ri: SimplifyShortImmForm(OutMI, X86::OR16i16); break;
case X86::OR32ri: SimplifyShortImmForm(OutMI, X86::OR32i32); break;
case X86::OR64ri32: SimplifyShortImmForm(OutMI, X86::OR64i32); break;
case X86::SBB8ri: SimplifyShortImmForm(OutMI, X86::SBB8i8); break;
case X86::SBB16ri: SimplifyShortImmForm(OutMI, X86::SBB16i16); break;
case X86::SBB32ri: SimplifyShortImmForm(OutMI, X86::SBB32i32); break;
case X86::SBB64ri32: SimplifyShortImmForm(OutMI, X86::SBB64i32); break;
case X86::SUB8ri: SimplifyShortImmForm(OutMI, X86::SUB8i8); break;
case X86::SUB16ri: SimplifyShortImmForm(OutMI, X86::SUB16i16); break;
case X86::SUB32ri: SimplifyShortImmForm(OutMI, X86::SUB32i32); break;
case X86::SUB64ri32: SimplifyShortImmForm(OutMI, X86::SUB64i32); break;
case X86::TEST8ri: SimplifyShortImmForm(OutMI, X86::TEST8i8); break;
case X86::TEST16ri: SimplifyShortImmForm(OutMI, X86::TEST16i16); break;
case X86::TEST32ri: SimplifyShortImmForm(OutMI, X86::TEST32i32); break;
case X86::TEST64ri32: SimplifyShortImmForm(OutMI, X86::TEST64i32); break;
case X86::XOR8ri: SimplifyShortImmForm(OutMI, X86::XOR8i8); break;
case X86::XOR16ri: SimplifyShortImmForm(OutMI, X86::XOR16i16); break;
case X86::XOR32ri: SimplifyShortImmForm(OutMI, X86::XOR32i32); break;
case X86::XOR64ri32: SimplifyShortImmForm(OutMI, X86::XOR64i32); break;
// Try to shrink some forms of movsx.
case X86::MOVSX16rr8:
case X86::MOVSX32rr16:
case X86::MOVSX64rr32:
SimplifyMOVSX(OutMI);
break;
}
}
void X86MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
OutMI.setOpcode(MI->getOpcode());
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
MCOperand MCOp;
switch (MO.getType()) {
default:
MI->dump();
llvm_unreachable("unknown operand type");
case MachineOperand::MO_Register:
// Ignore all implicit register operands.
if (MO.isImplicit()) continue;
MCOp = MCOperand::CreateReg(MO.getReg());
break;
case MachineOperand::MO_Immediate:
MCOp = MCOperand::CreateImm(MO.getImm());
break;
case MachineOperand::MO_MachineBasicBlock:
MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
MO.getMBB()->getSymbol(), Ctx));
break;
case MachineOperand::MO_GlobalAddress:
case MachineOperand::MO_ExternalSymbol:
MCOp = LowerSymbolOperand(MO, GetSymbolFromOperand(MO));
break;
case MachineOperand::MO_JumpTableIndex:
MCOp = LowerSymbolOperand(MO, AsmPrinter.GetJTISymbol(MO.getIndex()));
break;
case MachineOperand::MO_ConstantPoolIndex:
MCOp = LowerSymbolOperand(MO, AsmPrinter.GetCPISymbol(MO.getIndex()));
break;
case MachineOperand::MO_BlockAddress:
MCOp = LowerSymbolOperand(MO,
AsmPrinter.GetBlockAddressSymbol(MO.getBlockAddress()));
break;
case MachineOperand::MO_RegisterMask:
// Ignore call clobbers.
continue;
}
OutMI.addOperand(MCOp);
}
// Handle a few special cases to eliminate operand modifiers.
ReSimplify:
switch (OutMI.getOpcode()) {
case X86::LEA64_32r: // Handle 'subreg rewriting' for the lea64_32mem operand.
lower_lea64_32mem(&OutMI, 1);
// FALL THROUGH.
case X86::LEA64r:
case X86::LEA16r:
case X86::LEA32r:
// LEA should have a segment register, but it must be empty.
assert(OutMI.getNumOperands() == 1+X86::AddrNumOperands &&
"Unexpected # of LEA operands");
assert(OutMI.getOperand(1+X86::AddrSegmentReg).getReg() == 0 &&
"LEA has segment specified!");
break;
case X86::MOVZX64rr32: LowerSubReg32_Op0(OutMI, X86::MOV32rr); break;
case X86::MOVZX64rm32: LowerSubReg32_Op0(OutMI, X86::MOV32rm); break;
case X86::MOV64ri64i32: LowerSubReg32_Op0(OutMI, X86::MOV32ri); break;
case X86::MOVZX64rr8: LowerSubReg32_Op0(OutMI, X86::MOVZX32rr8); break;
case X86::MOVZX64rm8: LowerSubReg32_Op0(OutMI, X86::MOVZX32rm8); break;
case X86::MOVZX64rr16: LowerSubReg32_Op0(OutMI, X86::MOVZX32rr16); break;
case X86::MOVZX64rm16: LowerSubReg32_Op0(OutMI, X86::MOVZX32rm16); break;
case X86::SETB_C8r: LowerUnaryToTwoAddr(OutMI, X86::SBB8rr); break;
case X86::SETB_C16r: LowerUnaryToTwoAddr(OutMI, X86::SBB16rr); break;
case X86::SETB_C32r: LowerUnaryToTwoAddr(OutMI, X86::SBB32rr); break;
case X86::SETB_C64r: LowerUnaryToTwoAddr(OutMI, X86::SBB64rr); break;
case X86::MOV8r0: LowerUnaryToTwoAddr(OutMI, X86::XOR8rr); break;
case X86::MOV32r0: LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); break;
case X86::MOV16r0:
LowerSubReg32_Op0(OutMI, X86::MOV32r0); // MOV16r0 -> MOV32r0
LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); // MOV32r0 -> XOR32rr
break;
case X86::MOV64r0:
LowerSubReg32_Op0(OutMI, X86::MOV32r0); // MOV64r0 -> MOV32r0
LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); // MOV32r0 -> XOR32rr
break;
// TAILJMPr64, CALL64r, CALL64pcrel32 - These instructions have register
// inputs modeled as normal uses instead of implicit uses. As such, truncate
// off all but the first operand (the callee). FIXME: Change isel.
case X86::TAILJMPr64:
case X86::CALL64r:
case X86::CALL64pcrel32: {
unsigned Opcode = OutMI.getOpcode();
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
case X86::EH_RETURN:
case X86::EH_RETURN64: {
OutMI = MCInst();
OutMI.setOpcode(X86::RET);
break;
}
// TAILJMPd, TAILJMPd64 - Lower to the correct jump instructions.
case X86::TAILJMPr:
case X86::TAILJMPd:
case X86::TAILJMPd64: {
unsigned Opcode;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::TAILJMPr: Opcode = X86::JMP32r; break;
case X86::TAILJMPd:
case X86::TAILJMPd64: Opcode = X86::JMP_1; break;
}
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
// These are pseudo-ops for OR to help with the OR->ADD transformation. We do
// this with an ugly goto in case the resultant OR uses EAX and needs the
// short form.
case X86::ADD16rr_DB: OutMI.setOpcode(X86::OR16rr); goto ReSimplify;
case X86::ADD32rr_DB: OutMI.setOpcode(X86::OR32rr); goto ReSimplify;
case X86::ADD64rr_DB: OutMI.setOpcode(X86::OR64rr); goto ReSimplify;
case X86::ADD16ri_DB: OutMI.setOpcode(X86::OR16ri); goto ReSimplify;
case X86::ADD32ri_DB: OutMI.setOpcode(X86::OR32ri); goto ReSimplify;
case X86::ADD64ri32_DB: OutMI.setOpcode(X86::OR64ri32); goto ReSimplify;
case X86::ADD16ri8_DB: OutMI.setOpcode(X86::OR16ri8); goto ReSimplify;
case X86::ADD32ri8_DB: OutMI.setOpcode(X86::OR32ri8); goto ReSimplify;
case X86::ADD64ri8_DB: OutMI.setOpcode(X86::OR64ri8); goto ReSimplify;
// The assembler backend wants to see branches in their small form and relax
// them to their large form. The JIT can only handle the large form because
// it does not do relaxation. For now, translate the large form to the
// small one here.
case X86::JMP_4: OutMI.setOpcode(X86::JMP_1); break;
case X86::JO_4: OutMI.setOpcode(X86::JO_1); break;
case X86::JNO_4: OutMI.setOpcode(X86::JNO_1); break;
case X86::JB_4: OutMI.setOpcode(X86::JB_1); break;
case X86::JAE_4: OutMI.setOpcode(X86::JAE_1); break;
case X86::JE_4: OutMI.setOpcode(X86::JE_1); break;
case X86::JNE_4: OutMI.setOpcode(X86::JNE_1); break;
case X86::JBE_4: OutMI.setOpcode(X86::JBE_1); break;
case X86::JA_4: OutMI.setOpcode(X86::JA_1); break;
case X86::JS_4: OutMI.setOpcode(X86::JS_1); break;
case X86::JNS_4: OutMI.setOpcode(X86::JNS_1); break;
case X86::JP_4: OutMI.setOpcode(X86::JP_1); break;
case X86::JNP_4: OutMI.setOpcode(X86::JNP_1); break;
case X86::JL_4: OutMI.setOpcode(X86::JL_1); break;
case X86::JGE_4: OutMI.setOpcode(X86::JGE_1); break;
case X86::JLE_4: OutMI.setOpcode(X86::JLE_1); break;
case X86::JG_4: OutMI.setOpcode(X86::JG_1); break;
// Atomic load and store require a separate pseudo-inst because Acquire
// implies mayStore and Release implies mayLoad; fix these to regular MOV
// instructions here
case X86::ACQUIRE_MOV8rm: OutMI.setOpcode(X86::MOV8rm); goto ReSimplify;
case X86::ACQUIRE_MOV16rm: OutMI.setOpcode(X86::MOV16rm); goto ReSimplify;
case X86::ACQUIRE_MOV32rm: OutMI.setOpcode(X86::MOV32rm); goto ReSimplify;
case X86::ACQUIRE_MOV64rm: OutMI.setOpcode(X86::MOV64rm); goto ReSimplify;
case X86::RELEASE_MOV8mr: OutMI.setOpcode(X86::MOV8mr); goto ReSimplify;
case X86::RELEASE_MOV16mr: OutMI.setOpcode(X86::MOV16mr); goto ReSimplify;
case X86::RELEASE_MOV32mr: OutMI.setOpcode(X86::MOV32mr); goto ReSimplify;
case X86::RELEASE_MOV64mr: OutMI.setOpcode(X86::MOV64mr); goto ReSimplify;
// We don't currently select the correct instruction form for instructions
// which have a short %eax, etc. form. Handle this by custom lowering, for
// now.
//
// Note, we are currently not handling the following instructions:
// MOV64ao8, MOV64o8a
// XCHG16ar, XCHG32ar, XCHG64ar
case X86::MOV8mr_NOREX:
case X86::MOV8mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV8ao8); break;
case X86::MOV8rm_NOREX:
case X86::MOV8rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV8o8a); break;
case X86::MOV16mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV16ao16); break;
case X86::MOV16rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV16o16a); break;
case X86::MOV32mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV32ao32); break;
case X86::MOV32rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV32o32a); break;
case X86::ADC8ri: SimplifyShortImmForm(OutMI, X86::ADC8i8); break;
case X86::ADC16ri: SimplifyShortImmForm(OutMI, X86::ADC16i16); break;
case X86::ADC32ri: SimplifyShortImmForm(OutMI, X86::ADC32i32); break;
case X86::ADC64ri32: SimplifyShortImmForm(OutMI, X86::ADC64i32); break;
case X86::ADD8ri: SimplifyShortImmForm(OutMI, X86::ADD8i8); break;
case X86::ADD16ri: SimplifyShortImmForm(OutMI, X86::ADD16i16); break;
case X86::ADD32ri: SimplifyShortImmForm(OutMI, X86::ADD32i32); break;
case X86::ADD64ri32: SimplifyShortImmForm(OutMI, X86::ADD64i32); break;
case X86::AND8ri: SimplifyShortImmForm(OutMI, X86::AND8i8); break;
case X86::AND16ri: SimplifyShortImmForm(OutMI, X86::AND16i16); break;
case X86::AND32ri: SimplifyShortImmForm(OutMI, X86::AND32i32); break;
case X86::AND64ri32: SimplifyShortImmForm(OutMI, X86::AND64i32); break;
case X86::CMP8ri: SimplifyShortImmForm(OutMI, X86::CMP8i8); break;
case X86::CMP16ri: SimplifyShortImmForm(OutMI, X86::CMP16i16); break;
case X86::CMP32ri: SimplifyShortImmForm(OutMI, X86::CMP32i32); break;
case X86::CMP64ri32: SimplifyShortImmForm(OutMI, X86::CMP64i32); break;
case X86::OR8ri: SimplifyShortImmForm(OutMI, X86::OR8i8); break;
case X86::OR16ri: SimplifyShortImmForm(OutMI, X86::OR16i16); break;
case X86::OR32ri: SimplifyShortImmForm(OutMI, X86::OR32i32); break;
case X86::OR64ri32: SimplifyShortImmForm(OutMI, X86::OR64i32); break;
case X86::SBB8ri: SimplifyShortImmForm(OutMI, X86::SBB8i8); break;
case X86::SBB16ri: SimplifyShortImmForm(OutMI, X86::SBB16i16); break;
case X86::SBB32ri: SimplifyShortImmForm(OutMI, X86::SBB32i32); break;
case X86::SBB64ri32: SimplifyShortImmForm(OutMI, X86::SBB64i32); break;
case X86::SUB8ri: SimplifyShortImmForm(OutMI, X86::SUB8i8); break;
case X86::SUB16ri: SimplifyShortImmForm(OutMI, X86::SUB16i16); break;
case X86::SUB32ri: SimplifyShortImmForm(OutMI, X86::SUB32i32); break;
case X86::SUB64ri32: SimplifyShortImmForm(OutMI, X86::SUB64i32); break;
case X86::TEST8ri: SimplifyShortImmForm(OutMI, X86::TEST8i8); break;
case X86::TEST16ri: SimplifyShortImmForm(OutMI, X86::TEST16i16); break;
case X86::TEST32ri: SimplifyShortImmForm(OutMI, X86::TEST32i32); break;
case X86::TEST64ri32: SimplifyShortImmForm(OutMI, X86::TEST64i32); break;
case X86::XOR8ri: SimplifyShortImmForm(OutMI, X86::XOR8i8); break;
case X86::XOR16ri: SimplifyShortImmForm(OutMI, X86::XOR16i16); break;
case X86::XOR32ri: SimplifyShortImmForm(OutMI, X86::XOR32i32); break;
case X86::XOR64ri32: SimplifyShortImmForm(OutMI, X86::XOR64i32); break;
case X86::MORESTACK_RET:
OutMI.setOpcode(X86::RET);
break;
case X86::MORESTACK_RET_RESTORE_R10: {
MCInst retInst;
OutMI.setOpcode(X86::MOV64rr);
OutMI.addOperand(MCOperand::CreateReg(X86::R10));
OutMI.addOperand(MCOperand::CreateReg(X86::RAX));
retInst.setOpcode(X86::RET);
AsmPrinter.OutStreamer.EmitInstruction(retInst);
break;
}
}
}
/// EmitInstruction -- Print out a single PowerPC MI in Darwin syntax to
/// the current output stream.
///
void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
MCInst TmpInst;
// Lower multi-instruction pseudo operations.
switch (MI->getOpcode()) {
default: break;
case TargetOpcode::DBG_VALUE: {
if (!isVerbose() || !OutStreamer.hasRawTextSupport()) return;
SmallString<32> Str;
raw_svector_ostream O(Str);
unsigned NOps = MI->getNumOperands();
assert(NOps==4);
O << '\t' << MAI->getCommentString() << "DEBUG_VALUE: ";
// cast away const; DIetc do not take const operands for some reason.
DIVariable V(const_cast<MDNode *>(MI->getOperand(NOps-1).getMetadata()));
O << V.getName();
O << " <- ";
// Frame address. Currently handles register +- offset only.
assert(MI->getOperand(0).isReg() && MI->getOperand(1).isImm());
O << '['; printOperand(MI, 0, O); O << '+'; printOperand(MI, 1, O);
O << ']';
O << "+";
printOperand(MI, NOps-2, O);
OutStreamer.EmitRawText(O.str());
return;
}
case PPC::MovePCtoLR:
case PPC::MovePCtoLR8: {
// Transform %LR = MovePCtoLR
// Into this, where the label is the PIC base:
// bl L1$pb
// L1$pb:
MCSymbol *PICBase = MF->getPICBaseSymbol();
// Emit the 'bl'.
OutStreamer.EmitInstruction(MCInstBuilder(PPC::BL_Darwin) // Darwin vs SVR4 doesn't matter here.
// FIXME: We would like an efficient form for this, so we don't have to do
// a lot of extra uniquing.
.addExpr(MCSymbolRefExpr::Create(PICBase, OutContext)));
// Emit the label.
OutStreamer.EmitLabel(PICBase);
return;
}
case PPC::LDtocJTI:
case PPC::LDtocCPT:
case PPC::LDtoc: {
// Transform %X3 = LDtoc <ga:@min1>, %X2
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LD, and the global address operand to be a
// reference to the TOC entry we will synthesize later.
TmpInst.setOpcode(PPC::LD);
const MachineOperand &MO = MI->getOperand(1);
// Map symbol -> label of TOC entry
assert(MO.isGlobal() || MO.isCPI() || MO.isJTI());
MCSymbol *MOSymbol = 0;
if (MO.isGlobal())
MOSymbol = Mang->getSymbol(MO.getGlobal());
else if (MO.isCPI())
MOSymbol = GetCPISymbol(MO.getIndex());
else if (MO.isJTI())
MOSymbol = GetJTISymbol(MO.getIndex());
MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol);
const MCExpr *Exp =
MCSymbolRefExpr::Create(TOCEntry, MCSymbolRefExpr::VK_PPC_TOC_ENTRY,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
OutStreamer.EmitInstruction(TmpInst);
return;
}
case PPC::ADDIStocHA: {
// Transform %Xd = ADDIStocHA %X2, <ga:@sym>
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to ADDIS8. If the global address is external,
// has common linkage, is a function address, or is a jump table
// address, then generate a TOC entry and reference that. Otherwise
// reference the symbol directly.
TmpInst.setOpcode(PPC::ADDIS8);
const MachineOperand &MO = MI->getOperand(2);
assert((MO.isGlobal() || MO.isCPI() || MO.isJTI()) &&
"Invalid operand for ADDIStocHA!");
MCSymbol *MOSymbol = 0;
bool IsExternal = false;
bool IsFunction = false;
bool IsCommon = false;
if (MO.isGlobal()) {
const GlobalValue *GValue = MO.getGlobal();
MOSymbol = Mang->getSymbol(GValue);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GValue);
IsExternal = GVar && !GVar->hasInitializer();
IsCommon = GVar && GValue->hasCommonLinkage();
IsFunction = !GVar;
} else if (MO.isCPI())
MOSymbol = GetCPISymbol(MO.getIndex());
else if (MO.isJTI())
MOSymbol = GetJTISymbol(MO.getIndex());
if (IsExternal || IsFunction || IsCommon || MO.isJTI())
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC16_HA,
OutContext);
TmpInst.getOperand(2) = MCOperand::CreateExpr(Exp);
OutStreamer.EmitInstruction(TmpInst);
return;
}
case PPC::LDtocL: {
// Transform %Xd = LDtocL <ga:@sym>, %Xs
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LDrs, which is a form of LD with the offset
// specified by a SymbolLo. If the global address is external, has
// common linkage, or is a jump table address, then reference the
// associated TOC entry. Otherwise reference the symbol directly.
TmpInst.setOpcode(PPC::LDrs);
const MachineOperand &MO = MI->getOperand(1);
assert((MO.isGlobal() || MO.isJTI()) && "Invalid operand for LDtocL!");
MCSymbol *MOSymbol = 0;
if (MO.isJTI())
MOSymbol = lookUpOrCreateTOCEntry(GetJTISymbol(MO.getIndex()));
else {
const GlobalValue *GValue = MO.getGlobal();
MOSymbol = Mang->getSymbol(GValue);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GValue);
if (!GVar || !GVar->hasInitializer() || GValue->hasCommonLinkage())
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
}
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC16_LO,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
OutStreamer.EmitInstruction(TmpInst);
return;
}
case PPC::ADDItocL: {
// Transform %Xd = ADDItocL %Xs, <ga:@sym>
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to ADDI8L. If the global address is external, then
// generate a TOC entry and reference that. Otherwise reference the
// symbol directly.
TmpInst.setOpcode(PPC::ADDI8L);
const MachineOperand &MO = MI->getOperand(2);
assert((MO.isGlobal() || MO.isCPI()) && "Invalid operand for ADDItocL");
MCSymbol *MOSymbol = 0;
bool IsExternal = false;
bool IsFunction = false;
if (MO.isGlobal()) {
const GlobalValue *GValue = MO.getGlobal();
MOSymbol = Mang->getSymbol(GValue);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GValue);
IsExternal = GVar && !GVar->hasInitializer();
IsFunction = !GVar;
} else if (MO.isCPI())
MOSymbol = GetCPISymbol(MO.getIndex());
if (IsFunction || IsExternal)
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC16_LO,
OutContext);
TmpInst.getOperand(2) = MCOperand::CreateExpr(Exp);
OutStreamer.EmitInstruction(TmpInst);
return;
}
case PPC::ADDISgotTprelHA: {
// Transform: %Xd = ADDISgotTprelHA %X2, <ga:@sym>
// Into: %Xd = ADDIS8 %X2, sym@got@tlsgd@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymGotTprel =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TPREL16_HA,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X2)
.addExpr(SymGotTprel));
return;
}
case PPC::LDgotTprelL: {
// Transform %Xd = LDgotTprelL <ga:@sym>, %Xs
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LDrs, which is a form of LD with the offset
// specified by a SymbolLo.
TmpInst.setOpcode(PPC::LDrs);
const MachineOperand &MO = MI->getOperand(1);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TPREL16_LO,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
OutStreamer.EmitInstruction(TmpInst);
return;
}
case PPC::ADDIStlsgdHA: {
// Transform: %Xd = ADDIStlsgdHA %X2, <ga:@sym>
// Into: %Xd = ADDIS8 %X2, sym@got@tlsgd@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymGotTlsGD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSGD16_HA,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X2)
.addExpr(SymGotTlsGD));
return;
}
case PPC::ADDItlsgdL: {
// Transform: %Xd = ADDItlsgdL %Xs, <ga:@sym>
// Into: %Xd = ADDI8L %Xs, sym@got@tlsgd@l
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymGotTlsGD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSGD16_LO,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDI8L)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg())
.addExpr(SymGotTlsGD));
return;
}
case PPC::GETtlsADDR: {
// Transform: %X3 = GETtlsADDR %X3, <ga:@sym>
// Into: BL8_NOP_ELF_TLSGD __tls_get_addr(sym@tlsgd)
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
StringRef Name = "__tls_get_addr";
MCSymbol *TlsGetAddr = OutContext.GetOrCreateSymbol(Name);
const MCSymbolRefExpr *TlsRef =
MCSymbolRefExpr::Create(TlsGetAddr, MCSymbolRefExpr::VK_None, OutContext);
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymVar =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TLSGD,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::BL8_NOP_ELF_TLSGD)
.addExpr(TlsRef)
.addExpr(SymVar));
return;
}
case PPC::ADDIStlsldHA: {
// Transform: %Xd = ADDIStlsldHA %X2, <ga:@sym>
// Into: %Xd = ADDIS8 %X2, sym@got@tlsld@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymGotTlsLD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_HA,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X2)
.addExpr(SymGotTlsLD));
return;
}
case PPC::ADDItlsldL: {
// Transform: %Xd = ADDItlsldL %Xs, <ga:@sym>
// Into: %Xd = ADDI8L %Xs, sym@got@tlsld@l
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymGotTlsLD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_LO,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDI8L)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg())
.addExpr(SymGotTlsLD));
return;
}
case PPC::GETtlsldADDR: {
// Transform: %X3 = GETtlsldADDR %X3, <ga:@sym>
// Into: BL8_NOP_ELF_TLSLD __tls_get_addr(sym@tlsld)
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
StringRef Name = "__tls_get_addr";
MCSymbol *TlsGetAddr = OutContext.GetOrCreateSymbol(Name);
const MCSymbolRefExpr *TlsRef =
MCSymbolRefExpr::Create(TlsGetAddr, MCSymbolRefExpr::VK_None, OutContext);
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymVar =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TLSLD,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::BL8_NOP_ELF_TLSLD)
.addExpr(TlsRef)
.addExpr(SymVar));
return;
}
case PPC::ADDISdtprelHA: {
// Transform: %Xd = ADDISdtprelHA %X3, <ga:@sym>
// Into: %Xd = ADDIS8 %X3, sym@dtprel@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymDtprel =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_DTPREL16_HA,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X3)
.addExpr(SymDtprel));
return;
}
case PPC::ADDIdtprelL: {
// Transform: %Xd = ADDIdtprelL %Xs, <ga:@sym>
// Into: %Xd = ADDI8L %Xs, sym@dtprel@l
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = Mang->getSymbol(GValue);
const MCExpr *SymDtprel =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_DTPREL16_LO,
OutContext);
OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDI8L)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg())
.addExpr(SymDtprel));
return;
}
case PPC::MFCRpseud:
case PPC::MFCR8pseud:
// Transform: %R3 = MFCRpseud %CR7
// Into: %R3 = MFCR ;; cr7
OutStreamer.AddComment(PPCInstPrinter::
getRegisterName(MI->getOperand(1).getReg()));
OutStreamer.EmitInstruction(MCInstBuilder(Subtarget.isPPC64() ? PPC::MFCR8 : PPC::MFCR)
.addReg(MI->getOperand(0).getReg()));
return;
case PPC::SYNC:
// In Book E sync is called msync, handle this special case here...
if (Subtarget.isBookE()) {
OutStreamer.EmitRawText(StringRef("\tmsync"));
return;
}
}
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
OutStreamer.EmitInstruction(TmpInst);
}
void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
X86MCInstLower MCInstLowering(Mang, *MF, *this);
switch (MI->getOpcode()) {
case TargetOpcode::DBG_VALUE:
if (isVerbose() && OutStreamer.hasRawTextSupport()) {
std::string TmpStr;
raw_string_ostream OS(TmpStr);
PrintDebugValueComment(MI, OS);
OutStreamer.EmitRawText(StringRef(OS.str()));
}
return;
// Emit nothing here but a comment if we can.
case X86::Int_MemBarrier:
if (OutStreamer.hasRawTextSupport())
OutStreamer.EmitRawText(StringRef("\t#MEMBARRIER"));
return;
case X86::EH_RETURN:
case X86::EH_RETURN64: {
// Lower these as normal, but add some comments.
unsigned Reg = MI->getOperand(0).getReg();
OutStreamer.AddComment(StringRef("eh_return, addr: %") +
X86ATTInstPrinter::getRegisterName(Reg));
break;
}
case X86::TAILJMPr:
case X86::TAILJMPd:
case X86::TAILJMPd64:
// Lower these as normal, but add some comments.
OutStreamer.AddComment("TAILCALL");
break;
case X86::TLS_addr32:
case X86::TLS_addr64:
case X86::TLS_base_addr32:
case X86::TLS_base_addr64:
return LowerTlsAddr(OutStreamer, MCInstLowering, *MI);
case X86::MOVPC32r: {
MCInst TmpInst;
// This is a pseudo op for a two instruction sequence with a label, which
// looks like:
// call "L1$pb"
// "L1$pb":
// popl %esi
// Emit the call.
MCSymbol *PICBase = MF->getPICBaseSymbol();
TmpInst.setOpcode(X86::CALLpcrel32);
// FIXME: We would like an efficient form for this, so we don't have to do a
// lot of extra uniquing.
TmpInst.addOperand(MCOperand::CreateExpr(MCSymbolRefExpr::Create(PICBase,
OutContext)));
OutStreamer.EmitInstruction(TmpInst);
// Emit the label.
OutStreamer.EmitLabel(PICBase);
// popl $reg
TmpInst.setOpcode(X86::POP32r);
TmpInst.getOperand(0) = MCOperand::CreateReg(MI->getOperand(0).getReg());
OutStreamer.EmitInstruction(TmpInst);
return;
}
case X86::ADD32ri: {
// Lower the MO_GOT_ABSOLUTE_ADDRESS form of ADD32ri.
if (MI->getOperand(2).getTargetFlags() != X86II::MO_GOT_ABSOLUTE_ADDRESS)
break;
// Okay, we have something like:
// EAX = ADD32ri EAX, MO_GOT_ABSOLUTE_ADDRESS(@MYGLOBAL)
// For this, we want to print something like:
// MYGLOBAL + (. - PICBASE)
// However, we can't generate a ".", so just emit a new label here and refer
// to it.
MCSymbol *DotSym = OutContext.CreateTempSymbol();
OutStreamer.EmitLabel(DotSym);
// Now that we have emitted the label, lower the complex operand expression.
MCSymbol *OpSym = MCInstLowering.GetSymbolFromOperand(MI->getOperand(2));
const MCExpr *DotExpr = MCSymbolRefExpr::Create(DotSym, OutContext);
const MCExpr *PICBase =
MCSymbolRefExpr::Create(MF->getPICBaseSymbol(), OutContext);
DotExpr = MCBinaryExpr::CreateSub(DotExpr, PICBase, OutContext);
DotExpr = MCBinaryExpr::CreateAdd(MCSymbolRefExpr::Create(OpSym,OutContext),
DotExpr, OutContext);
MCInst TmpInst;
TmpInst.setOpcode(X86::ADD32ri);
TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
TmpInst.addOperand(MCOperand::CreateExpr(DotExpr));
OutStreamer.EmitInstruction(TmpInst);
return;
}
}
MCInst TmpInst;
MCInstLowering.Lower(MI, TmpInst);
OutStreamer.EmitInstruction(TmpInst);
}
SmallVector<DuplexCandidate, 8>
HexagonMCInstrInfo::getDuplexPossibilties(MCInstrInfo const &MCII,
MCSubtargetInfo const &STI,
MCInst const &MCB) {
assert(isBundle(MCB));
SmallVector<DuplexCandidate, 8> duplexToTry;
// Use an "order matters" version of isDuplexPair.
unsigned numInstrInPacket = MCB.getNumOperands();
for (unsigned distance = 1; distance < numInstrInPacket; ++distance) {
for (unsigned j = HexagonMCInstrInfo::bundleInstructionsOffset,
k = j + distance;
(j < numInstrInPacket) && (k < numInstrInPacket); ++j, ++k) {
// Check if reversible.
bool bisReversable = true;
if (isStoreInst(MCB.getOperand(j).getInst()->getOpcode()) &&
isStoreInst(MCB.getOperand(k).getInst()->getOpcode())) {
DEBUG(dbgs() << "skip out of order write pair: " << k << "," << j
<< "\n");
bisReversable = false;
}
if (HexagonMCInstrInfo::isMemReorderDisabled(MCB)) // }:mem_noshuf
bisReversable = false;
// Try in order.
if (isOrderedDuplexPair(
MCII, *MCB.getOperand(k).getInst(),
HexagonMCInstrInfo::hasExtenderForIndex(MCB, k - 1),
*MCB.getOperand(j).getInst(),
HexagonMCInstrInfo::hasExtenderForIndex(MCB, j - 1),
bisReversable, STI)) {
// Get iClass.
unsigned iClass = iClassOfDuplexPair(
getDuplexCandidateGroup(*MCB.getOperand(k).getInst()),
getDuplexCandidateGroup(*MCB.getOperand(j).getInst()));
// Save off pairs for duplex checking.
duplexToTry.push_back(DuplexCandidate(j, k, iClass));
DEBUG(dbgs() << "adding pair: " << j << "," << k << ":"
<< MCB.getOperand(j).getInst()->getOpcode() << ","
<< MCB.getOperand(k).getInst()->getOpcode() << "\n");
continue;
} else {
DEBUG(dbgs() << "skipping pair: " << j << "," << k << ":"
<< MCB.getOperand(j).getInst()->getOpcode() << ","
<< MCB.getOperand(k).getInst()->getOpcode() << "\n");
}
// Try reverse.
if (bisReversable) {
if (isOrderedDuplexPair(
MCII, *MCB.getOperand(j).getInst(),
HexagonMCInstrInfo::hasExtenderForIndex(MCB, j - 1),
*MCB.getOperand(k).getInst(),
HexagonMCInstrInfo::hasExtenderForIndex(MCB, k - 1),
bisReversable, STI)) {
// Get iClass.
unsigned iClass = iClassOfDuplexPair(
getDuplexCandidateGroup(*MCB.getOperand(j).getInst()),
getDuplexCandidateGroup(*MCB.getOperand(k).getInst()));
// Save off pairs for duplex checking.
duplexToTry.push_back(DuplexCandidate(k, j, iClass));
DEBUG(dbgs() << "adding pair:" << k << "," << j << ":"
<< MCB.getOperand(j).getInst()->getOpcode() << ","
<< MCB.getOperand(k).getInst()->getOpcode() << "\n");
} else {
DEBUG(dbgs() << "skipping pair: " << k << "," << j << ":"
<< MCB.getOperand(j).getInst()->getOpcode() << ","
<< MCB.getOperand(k).getInst()->getOpcode() << "\n");
}
}
}
}
return duplexToTry;
}
/// EmitInstruction -- Print out a single PowerPC MI in Darwin syntax to
/// the current output stream.
///
void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
MCInst TmpInst;
bool isPPC64 = Subtarget.isPPC64();
// Lower multi-instruction pseudo operations.
switch (MI->getOpcode()) {
default: break;
case TargetOpcode::DBG_VALUE:
llvm_unreachable("Should be handled target independently");
case PPC::MovePCtoLR:
case PPC::MovePCtoLR8: {
// Transform %LR = MovePCtoLR
// Into this, where the label is the PIC base:
// bl L1$pb
// L1$pb:
MCSymbol *PICBase = MF->getPICBaseSymbol();
// Emit the 'bl'.
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::BL)
// FIXME: We would like an efficient form for this, so we don't have to do
// a lot of extra uniquing.
.addExpr(MCSymbolRefExpr::Create(PICBase, OutContext)));
// Emit the label.
OutStreamer.EmitLabel(PICBase);
return;
}
case PPC::LDtocJTI:
case PPC::LDtocCPT:
case PPC::LDtoc: {
// Transform %X3 = LDtoc <ga:@min1>, %X2
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LD, and the global address operand to be a
// reference to the TOC entry we will synthesize later.
TmpInst.setOpcode(PPC::LD);
const MachineOperand &MO = MI->getOperand(1);
// Map symbol -> label of TOC entry
assert(MO.isGlobal() || MO.isCPI() || MO.isJTI());
MCSymbol *MOSymbol = 0;
if (MO.isGlobal())
MOSymbol = getSymbol(MO.getGlobal());
else if (MO.isCPI())
MOSymbol = GetCPISymbol(MO.getIndex());
else if (MO.isJTI())
MOSymbol = GetJTISymbol(MO.getIndex());
MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol);
const MCExpr *Exp =
MCSymbolRefExpr::Create(TOCEntry, MCSymbolRefExpr::VK_PPC_TOC,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
EmitToStreamer(OutStreamer, TmpInst);
return;
}
case PPC::ADDIStocHA: {
// Transform %Xd = ADDIStocHA %X2, <ga:@sym>
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to ADDIS8. If the global address is external,
// has common linkage, is a function address, or is a jump table
// address, then generate a TOC entry and reference that. Otherwise
// reference the symbol directly.
TmpInst.setOpcode(PPC::ADDIS8);
const MachineOperand &MO = MI->getOperand(2);
assert((MO.isGlobal() || MO.isCPI() || MO.isJTI()) &&
"Invalid operand for ADDIStocHA!");
MCSymbol *MOSymbol = 0;
bool IsExternal = false;
bool IsFunction = false;
bool IsCommon = false;
bool IsAvailExt = false;
if (MO.isGlobal()) {
const GlobalValue *GValue = MO.getGlobal();
const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
const GlobalValue *RealGValue = GAlias ?
GAlias->resolveAliasedGlobal(false) : GValue;
MOSymbol = getSymbol(RealGValue);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
IsExternal = GVar && !GVar->hasInitializer();
IsCommon = GVar && RealGValue->hasCommonLinkage();
IsFunction = !GVar;
IsAvailExt = GVar && RealGValue->hasAvailableExternallyLinkage();
} else if (MO.isCPI())
MOSymbol = GetCPISymbol(MO.getIndex());
else if (MO.isJTI())
MOSymbol = GetJTISymbol(MO.getIndex());
if (IsExternal || IsFunction || IsCommon || IsAvailExt || MO.isJTI() ||
TM.getCodeModel() == CodeModel::Large)
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC_HA,
OutContext);
TmpInst.getOperand(2) = MCOperand::CreateExpr(Exp);
EmitToStreamer(OutStreamer, TmpInst);
return;
}
case PPC::LDtocL: {
// Transform %Xd = LDtocL <ga:@sym>, %Xs
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LD. If the global address is external, has
// common linkage, or is a jump table address, then reference the
// associated TOC entry. Otherwise reference the symbol directly.
TmpInst.setOpcode(PPC::LD);
const MachineOperand &MO = MI->getOperand(1);
assert((MO.isGlobal() || MO.isJTI() || MO.isCPI()) &&
"Invalid operand for LDtocL!");
MCSymbol *MOSymbol = 0;
if (MO.isJTI())
MOSymbol = lookUpOrCreateTOCEntry(GetJTISymbol(MO.getIndex()));
else if (MO.isCPI()) {
MOSymbol = GetCPISymbol(MO.getIndex());
if (TM.getCodeModel() == CodeModel::Large)
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
}
else if (MO.isGlobal()) {
const GlobalValue *GValue = MO.getGlobal();
const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
const GlobalValue *RealGValue = GAlias ?
GAlias->resolveAliasedGlobal(false) : GValue;
MOSymbol = getSymbol(RealGValue);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
if (!GVar || !GVar->hasInitializer() || RealGValue->hasCommonLinkage() ||
RealGValue->hasAvailableExternallyLinkage() ||
TM.getCodeModel() == CodeModel::Large)
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
}
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC_LO,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
EmitToStreamer(OutStreamer, TmpInst);
return;
}
case PPC::ADDItocL: {
// Transform %Xd = ADDItocL %Xs, <ga:@sym>
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to ADDI8. If the global address is external, then
// generate a TOC entry and reference that. Otherwise reference the
// symbol directly.
TmpInst.setOpcode(PPC::ADDI8);
const MachineOperand &MO = MI->getOperand(2);
assert((MO.isGlobal() || MO.isCPI()) && "Invalid operand for ADDItocL");
MCSymbol *MOSymbol = 0;
bool IsExternal = false;
bool IsFunction = false;
if (MO.isGlobal()) {
const GlobalValue *GValue = MO.getGlobal();
const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
const GlobalValue *RealGValue = GAlias ?
GAlias->resolveAliasedGlobal(false) : GValue;
MOSymbol = getSymbol(RealGValue);
const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
IsExternal = GVar && !GVar->hasInitializer();
IsFunction = !GVar;
} else if (MO.isCPI())
MOSymbol = GetCPISymbol(MO.getIndex());
if (IsFunction || IsExternal || TM.getCodeModel() == CodeModel::Large)
MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC_LO,
OutContext);
TmpInst.getOperand(2) = MCOperand::CreateExpr(Exp);
EmitToStreamer(OutStreamer, TmpInst);
return;
}
case PPC::ADDISgotTprelHA: {
// Transform: %Xd = ADDISgotTprelHA %X2, <ga:@sym>
// Into: %Xd = ADDIS8 %X2, sym@got@tlsgd@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymGotTprel =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TPREL_HA,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X2)
.addExpr(SymGotTprel));
return;
}
case PPC::LDgotTprelL:
case PPC::LDgotTprelL32: {
// Transform %Xd = LDgotTprelL <ga:@sym>, %Xs
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LD.
TmpInst.setOpcode(isPPC64 ? PPC::LD : PPC::LWZ);
const MachineOperand &MO = MI->getOperand(1);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *Exp =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TPREL_LO,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
EmitToStreamer(OutStreamer, TmpInst);
return;
}
case PPC::PPC32GOT: {
MCSymbol *GOTSymbol = OutContext.GetOrCreateSymbol(StringRef("_GLOBAL_OFFSET_TABLE_"));
const MCExpr *SymGotTlsL =
MCSymbolRefExpr::Create(GOTSymbol, MCSymbolRefExpr::VK_PPC_LO,
OutContext);
const MCExpr *SymGotTlsHA =
MCSymbolRefExpr::Create(GOTSymbol, MCSymbolRefExpr::VK_PPC_HA,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::LI)
.addReg(MI->getOperand(0).getReg())
.addExpr(SymGotTlsL));
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDIS)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(0).getReg())
.addExpr(SymGotTlsHA));
return;
}
case PPC::ADDIStlsgdHA: {
// Transform: %Xd = ADDIStlsgdHA %X2, <ga:@sym>
// Into: %Xd = ADDIS8 %X2, sym@got@tlsgd@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymGotTlsGD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSGD_HA,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X2)
.addExpr(SymGotTlsGD));
return;
}
case PPC::ADDItlsgdL: {
// Transform: %Xd = ADDItlsgdL %Xs, <ga:@sym>
// Into: %Xd = ADDI8 %Xs, sym@got@tlsgd@l
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymGotTlsGD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSGD_LO,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDI8)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg())
.addExpr(SymGotTlsGD));
return;
}
case PPC::GETtlsADDR: {
// Transform: %X3 = GETtlsADDR %X3, <ga:@sym>
// Into: BL8_NOP_TLS __tls_get_addr(sym@tlsgd)
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
StringRef Name = "__tls_get_addr";
MCSymbol *TlsGetAddr = OutContext.GetOrCreateSymbol(Name);
const MCSymbolRefExpr *TlsRef =
MCSymbolRefExpr::Create(TlsGetAddr, MCSymbolRefExpr::VK_None, OutContext);
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymVar =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TLSGD,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::BL8_NOP_TLS)
.addExpr(TlsRef)
.addExpr(SymVar));
return;
}
case PPC::ADDIStlsldHA: {
// Transform: %Xd = ADDIStlsldHA %X2, <ga:@sym>
// Into: %Xd = ADDIS8 %X2, sym@got@tlsld@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymGotTlsLD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSLD_HA,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X2)
.addExpr(SymGotTlsLD));
return;
}
case PPC::ADDItlsldL: {
// Transform: %Xd = ADDItlsldL %Xs, <ga:@sym>
// Into: %Xd = ADDI8 %Xs, sym@got@tlsld@l
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymGotTlsLD =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSLD_LO,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDI8)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg())
.addExpr(SymGotTlsLD));
return;
}
case PPC::GETtlsldADDR: {
// Transform: %X3 = GETtlsldADDR %X3, <ga:@sym>
// Into: BL8_NOP_TLS __tls_get_addr(sym@tlsld)
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
StringRef Name = "__tls_get_addr";
MCSymbol *TlsGetAddr = OutContext.GetOrCreateSymbol(Name);
const MCSymbolRefExpr *TlsRef =
MCSymbolRefExpr::Create(TlsGetAddr, MCSymbolRefExpr::VK_None, OutContext);
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymVar =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TLSLD,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::BL8_NOP_TLS)
.addExpr(TlsRef)
.addExpr(SymVar));
return;
}
case PPC::ADDISdtprelHA: {
// Transform: %Xd = ADDISdtprelHA %X3, <ga:@sym>
// Into: %Xd = ADDIS8 %X3, sym@dtprel@ha
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymDtprel =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_DTPREL_HA,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDIS8)
.addReg(MI->getOperand(0).getReg())
.addReg(PPC::X3)
.addExpr(SymDtprel));
return;
}
case PPC::ADDIdtprelL: {
// Transform: %Xd = ADDIdtprelL %Xs, <ga:@sym>
// Into: %Xd = ADDI8 %Xs, sym@dtprel@l
assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
const MachineOperand &MO = MI->getOperand(2);
const GlobalValue *GValue = MO.getGlobal();
MCSymbol *MOSymbol = getSymbol(GValue);
const MCExpr *SymDtprel =
MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_DTPREL_LO,
OutContext);
EmitToStreamer(OutStreamer, MCInstBuilder(PPC::ADDI8)
.addReg(MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg())
.addExpr(SymDtprel));
return;
}
case PPC::MFOCRF:
case PPC::MFOCRF8:
if (!Subtarget.hasMFOCRF()) {
// Transform: %R3 = MFOCRF %CR7
// Into: %R3 = MFCR ;; cr7
unsigned NewOpcode =
MI->getOpcode() == PPC::MFOCRF ? PPC::MFCR : PPC::MFCR8;
OutStreamer.AddComment(PPCInstPrinter::
getRegisterName(MI->getOperand(1).getReg()));
EmitToStreamer(OutStreamer, MCInstBuilder(NewOpcode)
.addReg(MI->getOperand(0).getReg()));
return;
}
break;
case PPC::MTOCRF:
case PPC::MTOCRF8:
if (!Subtarget.hasMFOCRF()) {
// Transform: %CR7 = MTOCRF %R3
// Into: MTCRF mask, %R3 ;; cr7
unsigned NewOpcode =
MI->getOpcode() == PPC::MTOCRF ? PPC::MTCRF : PPC::MTCRF8;
unsigned Mask = 0x80 >> OutContext.getRegisterInfo()
->getEncodingValue(MI->getOperand(0).getReg());
OutStreamer.AddComment(PPCInstPrinter::
getRegisterName(MI->getOperand(0).getReg()));
EmitToStreamer(OutStreamer, MCInstBuilder(NewOpcode)
.addImm(Mask)
.addReg(MI->getOperand(1).getReg()));
return;
}
break;
case PPC::LD:
case PPC::STD:
case PPC::LWA_32:
case PPC::LWA: {
// Verify alignment is legal, so we don't create relocations
// that can't be supported.
// FIXME: This test is currently disabled for Darwin. The test
// suite shows a handful of test cases that fail this check for
// Darwin. Those need to be investigated before this sanity test
// can be enabled for those subtargets.
if (!Subtarget.isDarwin()) {
unsigned OpNum = (MI->getOpcode() == PPC::STD) ? 2 : 1;
const MachineOperand &MO = MI->getOperand(OpNum);
if (MO.isGlobal() && MO.getGlobal()->getAlignment() < 4)
llvm_unreachable("Global must be word-aligned for LD, STD, LWA!");
}
// Now process the instruction normally.
break;
}
}
unsigned HexagonMCInstrInfo::getDuplexCandidateGroup(MCInst const &MCI) {
unsigned DstReg, PredReg, SrcReg, Src1Reg, Src2Reg;
switch (MCI.getOpcode()) {
default:
return HexagonII::HSIG_None;
//
// Group L1:
//
// Rd = memw(Rs+#u4:2)
// Rd = memub(Rs+#u4:0)
case Hexagon::L2_loadri_io:
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
// Special case this one from Group L2.
// Rd = memw(r29+#u5:2)
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg)) {
if (HexagonMCInstrInfo::isIntReg(SrcReg) &&
Hexagon::R29 == SrcReg && inRange<5, 2>(MCI, 2)) {
return HexagonII::HSIG_L2;
}
// Rd = memw(Rs+#u4:2)
if (HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
inRange<4, 2>(MCI, 2)) {
return HexagonII::HSIG_L1;
}
}
break;
case Hexagon::L2_loadrub_io:
// Rd = memub(Rs+#u4:0)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
inRange<4>(MCI, 2)) {
return HexagonII::HSIG_L1;
}
break;
//
// Group L2:
//
// Rd = memh/memuh(Rs+#u3:1)
// Rd = memb(Rs+#u3:0)
// Rd = memw(r29+#u5:2) - Handled above.
// Rdd = memd(r29+#u5:3)
// deallocframe
// [if ([!]p0[.new])] dealloc_return
// [if ([!]p0[.new])] jumpr r31
case Hexagon::L2_loadrh_io:
case Hexagon::L2_loadruh_io:
// Rd = memh/memuh(Rs+#u3:1)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
inRange<3, 1>(MCI, 2)) {
return HexagonII::HSIG_L2;
}
break;
case Hexagon::L2_loadrb_io:
// Rd = memb(Rs+#u3:0)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
inRange<3>(MCI, 2)) {
return HexagonII::HSIG_L2;
}
break;
case Hexagon::L2_loadrd_io:
// Rdd = memd(r29+#u5:3)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isDblRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntReg(SrcReg) && Hexagon::R29 == SrcReg &&
inRange<5, 3>(MCI, 2)) {
return HexagonII::HSIG_L2;
}
break;
case Hexagon::L4_return:
case Hexagon::L2_deallocframe:
return HexagonII::HSIG_L2;
case Hexagon::EH_RETURN_JMPR:
case Hexagon::J2_jumpr:
case Hexagon::PS_jmpret:
// jumpr r31
// Actual form JMPR %PC<imp-def>, %R31<imp-use>, %R0<imp-use,internal>.
DstReg = MCI.getOperand(0).getReg();
if (Hexagon::R31 == DstReg)
return HexagonII::HSIG_L2;
break;
case Hexagon::J2_jumprt:
case Hexagon::J2_jumprf:
case Hexagon::J2_jumprtnew:
case Hexagon::J2_jumprfnew:
case Hexagon::J2_jumprtnewpt:
case Hexagon::J2_jumprfnewpt:
case Hexagon::PS_jmprett:
case Hexagon::PS_jmpretf:
case Hexagon::PS_jmprettnew:
case Hexagon::PS_jmpretfnew:
case Hexagon::PS_jmprettnewpt:
case Hexagon::PS_jmpretfnewpt:
DstReg = MCI.getOperand(1).getReg();
SrcReg = MCI.getOperand(0).getReg();
// [if ([!]p0[.new])] jumpr r31
if ((HexagonMCInstrInfo::isPredReg(SrcReg) && (Hexagon::P0 == SrcReg)) &&
(Hexagon::R31 == DstReg)) {
return HexagonII::HSIG_L2;
}
break;
case Hexagon::L4_return_t:
case Hexagon::L4_return_f:
case Hexagon::L4_return_tnew_pnt:
case Hexagon::L4_return_fnew_pnt:
case Hexagon::L4_return_tnew_pt:
case Hexagon::L4_return_fnew_pt:
// [if ([!]p0[.new])] dealloc_return
SrcReg = MCI.getOperand(0).getReg();
if (Hexagon::P0 == SrcReg) {
return HexagonII::HSIG_L2;
}
break;
//
// Group S1:
//
// memw(Rs+#u4:2) = Rt
// memb(Rs+#u4:0) = Rt
case Hexagon::S2_storeri_io:
// Special case this one from Group S2.
// memw(r29+#u5:2) = Rt
Src1Reg = MCI.getOperand(0).getReg();
Src2Reg = MCI.getOperand(2).getReg();
if (HexagonMCInstrInfo::isIntReg(Src1Reg) &&
HexagonMCInstrInfo::isIntRegForSubInst(Src2Reg) &&
Hexagon::R29 == Src1Reg && inRange<5, 2>(MCI, 1)) {
return HexagonII::HSIG_S2;
}
// memw(Rs+#u4:2) = Rt
if (HexagonMCInstrInfo::isIntRegForSubInst(Src1Reg) &&
HexagonMCInstrInfo::isIntRegForSubInst(Src2Reg) &&
inRange<4, 2>(MCI, 1)) {
return HexagonII::HSIG_S1;
}
break;
case Hexagon::S2_storerb_io:
// memb(Rs+#u4:0) = Rt
Src1Reg = MCI.getOperand(0).getReg();
Src2Reg = MCI.getOperand(2).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(Src1Reg) &&
HexagonMCInstrInfo::isIntRegForSubInst(Src2Reg) &&
inRange<4>(MCI, 1)) {
return HexagonII::HSIG_S1;
}
break;
//
// Group S2:
//
// memh(Rs+#u3:1) = Rt
// memw(r29+#u5:2) = Rt
// memd(r29+#s6:3) = Rtt
// memw(Rs+#u4:2) = #U1
// memb(Rs+#u4) = #U1
// allocframe(#u5:3)
case Hexagon::S2_storerh_io:
// memh(Rs+#u3:1) = Rt
Src1Reg = MCI.getOperand(0).getReg();
Src2Reg = MCI.getOperand(2).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(Src1Reg) &&
HexagonMCInstrInfo::isIntRegForSubInst(Src2Reg) &&
inRange<3, 1>(MCI, 1)) {
return HexagonII::HSIG_S2;
}
break;
case Hexagon::S2_storerd_io:
// memd(r29+#s6:3) = Rtt
Src1Reg = MCI.getOperand(0).getReg();
Src2Reg = MCI.getOperand(2).getReg();
if (HexagonMCInstrInfo::isDblRegForSubInst(Src2Reg) &&
HexagonMCInstrInfo::isIntReg(Src1Reg) && Hexagon::R29 == Src1Reg &&
inSRange<6, 3>(MCI, 1)) {
return HexagonII::HSIG_S2;
}
break;
case Hexagon::S4_storeiri_io:
// memw(Rs+#u4:2) = #U1
Src1Reg = MCI.getOperand(0).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(Src1Reg) &&
inRange<4, 2>(MCI, 1) && inRange<1>(MCI, 2)) {
return HexagonII::HSIG_S2;
}
break;
case Hexagon::S4_storeirb_io:
// memb(Rs+#u4) = #U1
Src1Reg = MCI.getOperand(0).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(Src1Reg) &&
inRange<4>(MCI, 1) && inRange<1>(MCI, 2)) {
return HexagonII::HSIG_S2;
}
break;
case Hexagon::S2_allocframe:
if (inRange<5, 3>(MCI, 0))
return HexagonII::HSIG_S2;
break;
//
// Group A:
//
// Rx = add(Rx,#s7)
// Rd = Rs
// Rd = #u6
// Rd = #-1
// if ([!]P0[.new]) Rd = #0
// Rd = add(r29,#u6:2)
// Rx = add(Rx,Rs)
// P0 = cmp.eq(Rs,#u2)
// Rdd = combine(#0,Rs)
// Rdd = combine(Rs,#0)
// Rdd = combine(#u2,#U2)
// Rd = add(Rs,#1)
// Rd = add(Rs,#-1)
// Rd = sxth/sxtb/zxtb/zxth(Rs)
// Rd = and(Rs,#1)
case Hexagon::A2_addi:
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg)) {
// Rd = add(r29,#u6:2)
if (HexagonMCInstrInfo::isIntReg(SrcReg) && Hexagon::R29 == SrcReg &&
inRange<6, 2>(MCI, 2)) {
return HexagonII::HSIG_A;
}
// Rx = add(Rx,#s7)
if (DstReg == SrcReg) {
return HexagonII::HSIG_A;
}
// Rd = add(Rs,#1)
// Rd = add(Rs,#-1)
if (HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
(minConstant(MCI, 2) == 1 || minConstant(MCI, 2) == -1)) {
return HexagonII::HSIG_A;
}
}
break;
case Hexagon::A2_add:
// Rx = add(Rx,Rs)
DstReg = MCI.getOperand(0).getReg();
Src1Reg = MCI.getOperand(1).getReg();
Src2Reg = MCI.getOperand(2).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) && (DstReg == Src1Reg) &&
HexagonMCInstrInfo::isIntRegForSubInst(Src2Reg)) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A2_andir:
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
(minConstant(MCI, 2) == 1 || minConstant(MCI, 2) == 255)) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A2_tfr:
// Rd = Rs
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg)) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A2_tfrsi:
DstReg = MCI.getOperand(0).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg)) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::C2_cmoveit:
case Hexagon::C2_cmovenewit:
case Hexagon::C2_cmoveif:
case Hexagon::C2_cmovenewif:
// if ([!]P0[.new]) Rd = #0
// Actual form:
// %R16<def> = C2_cmovenewit %P0<internal>, 0, %R16<imp-use,undef>;
DstReg = MCI.getOperand(0).getReg(); // Rd
PredReg = MCI.getOperand(1).getReg(); // P0
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
Hexagon::P0 == PredReg && minConstant(MCI, 2) == 0) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::C2_cmpeqi:
// P0 = cmp.eq(Rs,#u2)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (Hexagon::P0 == DstReg &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
inRange<2>(MCI, 2)) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A2_combineii:
case Hexagon::A4_combineii:
// Rdd = combine(#u2,#U2)
DstReg = MCI.getOperand(0).getReg();
if (HexagonMCInstrInfo::isDblRegForSubInst(DstReg) &&
inRange<2>(MCI, 1) && inRange<2>(MCI, 2)) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A4_combineri:
// Rdd = combine(Rs,#0)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isDblRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
minConstant(MCI, 2) == 0) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A4_combineir:
// Rdd = combine(#0,Rs)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(2).getReg();
if (HexagonMCInstrInfo::isDblRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg) &&
minConstant(MCI, 1) == 0) {
return HexagonII::HSIG_A;
}
break;
case Hexagon::A2_sxtb:
case Hexagon::A2_sxth:
case Hexagon::A2_zxtb:
case Hexagon::A2_zxth:
// Rd = sxth/sxtb/zxtb/zxth(Rs)
DstReg = MCI.getOperand(0).getReg();
SrcReg = MCI.getOperand(1).getReg();
if (HexagonMCInstrInfo::isIntRegForSubInst(DstReg) &&
HexagonMCInstrInfo::isIntRegForSubInst(SrcReg)) {
return HexagonII::HSIG_A;
}
break;
}
return HexagonII::HSIG_None;
}
static void populateWrites(InstrDesc &ID, const MCInst &MCI,
const MCInstrDesc &MCDesc,
const MCSchedClassDesc &SCDesc,
const MCSubtargetInfo &STI) {
// Set if writes through this opcode may update super registers.
// TODO: on x86-64, a 4 byte write of a general purpose register always
// fully updates the super-register.
// More in general, (at least on x86) not all register writes perform
// a partial (super-)register update.
// For example, an AVX instruction that writes on a XMM register implicitly
// zeroes the upper half of every aliasing super-register.
//
// For now, we pessimistically assume that writes are all potentially
// partial register updates. This is a good default for most targets, execept
// for those like x86 which implement a special semantic for certain opcodes.
// At least on x86, this may lead to an inaccurate prediction of the
// instruction level parallelism.
bool FullyUpdatesSuperRegisters = false;
// Now Populate Writes.
// This algorithm currently works under the strong (and potentially incorrect)
// assumption that information related to register def/uses can be obtained
// from MCInstrDesc.
//
// However class MCInstrDesc is used to describe MachineInstr objects and not
// MCInst objects. To be more specific, MCInstrDesc objects are opcode
// descriptors that are automatically generated via tablegen based on the
// instruction set information available from the target .td files. That
// means, the number of (explicit) definitions according to MCInstrDesc always
// matches the cardinality of the `(outs)` set in tablegen.
//
// By constructions, definitions must appear first in the operand sequence of
// a MachineInstr. Also, the (outs) sequence is preserved (example: the first
// element in the outs set is the first operand in the corresponding
// MachineInstr). That's the reason why MCInstrDesc only needs to declare the
// total number of register definitions, and not where those definitions are
// in the machine operand sequence.
//
// Unfortunately, it is not safe to use the information from MCInstrDesc to
// also describe MCInst objects. An MCInst object can be obtained from a
// MachineInstr through a lowering step which may restructure the operand
// sequence (and even remove or introduce new operands). So, there is a high
// risk that the lowering step breaks the assumptions that register
// definitions are always at the beginning of the machine operand sequence.
//
// This is a fundamental problem, and it is still an open problem. Essentially
// we have to find a way to correlate def/use operands of a MachineInstr to
// operands of an MCInst. Otherwise, we cannot correctly reconstruct data
// dependencies, nor we can correctly interpret the scheduling model, which
// heavily uses machine operand indices to define processor read-advance
// information, and to identify processor write resources. Essentially, we
// either need something like a MCInstrDesc, but for MCInst, or a way
// to map MCInst operands back to MachineInstr operands.
//
// Unfortunately, we don't have that information now. So, this prototype
// currently work under the strong assumption that we can always safely trust
// the content of an MCInstrDesc. For example, we can query a MCInstrDesc to
// obtain the number of explicit and implicit register defintions. We also
// assume that register definitions always come first in the operand sequence.
// This last assumption usually makes sense for MachineInstr, where register
// definitions always appear at the beginning of the operands sequence. In
// reality, these assumptions could be broken by the lowering step, which can
// decide to lay out operands in a different order than the original order of
// operand as specified by the MachineInstr.
//
// Things get even more complicated in the presence of "optional" register
// definitions. For MachineInstr, optional register definitions are always at
// the end of the operand sequence. Some ARM instructions that may update the
// status flags specify that register as a optional operand. Since we don't
// have operand descriptors for MCInst, we assume for now that the optional
// definition is always the last operand of a MCInst. Again, this assumption
// may be okay for most targets. However, there is no guarantee that targets
// would respect that.
//
// In conclusion: these are for now the strong assumptions made by the tool:
// * The number of explicit and implicit register definitions in a MCInst
// matches the number of explicit and implicit definitions according to
// the opcode descriptor (MCInstrDesc).
// * Register definitions take precedence over register uses in the operands
// list.
// * If an opcode specifies an optional definition, then the optional
// definition is always the last operand in the sequence, and it can be
// set to zero (i.e. "no register").
//
// These assumptions work quite well for most out-of-order in-tree targets
// like x86. This is mainly because the vast majority of instructions is
// expanded to MCInst using a straightforward lowering logic that preserves
// the ordering of the operands.
//
// In the longer term, we need to find a proper solution for this issue.
unsigned NumExplicitDefs = MCDesc.getNumDefs();
unsigned NumImplicitDefs = MCDesc.getNumImplicitDefs();
unsigned NumWriteLatencyEntries = SCDesc.NumWriteLatencyEntries;
unsigned TotalDefs = NumExplicitDefs + NumImplicitDefs;
if (MCDesc.hasOptionalDef())
TotalDefs++;
ID.Writes.resize(TotalDefs);
// Iterate over the operands list, and skip non-register operands.
// The first NumExplictDefs register operands are expected to be register
// definitions.
unsigned CurrentDef = 0;
unsigned i = 0;
for (; i < MCI.getNumOperands() && CurrentDef < NumExplicitDefs; ++i) {
const MCOperand &Op = MCI.getOperand(i);
if (!Op.isReg())
continue;
WriteDescriptor &Write = ID.Writes[CurrentDef];
Write.OpIndex = i;
if (CurrentDef < NumWriteLatencyEntries) {
const MCWriteLatencyEntry &WLE =
*STI.getWriteLatencyEntry(&SCDesc, CurrentDef);
// Conservatively default to MaxLatency.
Write.Latency = WLE.Cycles == -1 ? ID.MaxLatency : WLE.Cycles;
Write.SClassOrWriteResourceID = WLE.WriteResourceID;
} else {
// Assign a default latency for this write.
Write.Latency = ID.MaxLatency;
Write.SClassOrWriteResourceID = 0;
}
Write.FullyUpdatesSuperRegs = FullyUpdatesSuperRegisters;
Write.IsOptionalDef = false;
LLVM_DEBUG({
dbgs() << "\t\tOpIdx=" << Write.OpIndex << ", Latency=" << Write.Latency
<< ", WriteResourceID=" << Write.SClassOrWriteResourceID << '\n';
});
CurrentDef++;
}
/// non-Symmetrical. See if these two instructions are fit for duplex pair.
bool HexagonMCInstrInfo::isOrderedDuplexPair(MCInstrInfo const &MCII,
MCInst const &MIa, bool ExtendedA,
MCInst const &MIb, bool ExtendedB,
bool bisReversable,
MCSubtargetInfo const &STI) {
// Slot 1 cannot be extended in duplexes PRM 10.5
if (ExtendedA)
return false;
// Only A2_addi and A2_tfrsi can be extended in duplex form PRM 10.5
if (ExtendedB) {
unsigned Opcode = MIb.getOpcode();
if ((Opcode != Hexagon::A2_addi) && (Opcode != Hexagon::A2_tfrsi))
return false;
}
unsigned MIaG = HexagonMCInstrInfo::getDuplexCandidateGroup(MIa),
MIbG = HexagonMCInstrInfo::getDuplexCandidateGroup(MIb);
static std::map<unsigned, unsigned> subinstOpcodeMap(std::begin(opcodeData),
std::end(opcodeData));
// If a duplex contains 2 insns in the same group, the insns must be
// ordered such that the numerically smaller opcode is in slot 1.
if ((MIaG != HexagonII::HSIG_None) && (MIaG == MIbG) && bisReversable) {
MCInst SubInst0 = HexagonMCInstrInfo::deriveSubInst(MIa);
MCInst SubInst1 = HexagonMCInstrInfo::deriveSubInst(MIb);
unsigned zeroedSubInstS0 =
subinstOpcodeMap.find(SubInst0.getOpcode())->second;
unsigned zeroedSubInstS1 =
subinstOpcodeMap.find(SubInst1.getOpcode())->second;
if (zeroedSubInstS0 < zeroedSubInstS1)
// subinstS0 (maps to slot 0) must be greater than
// subinstS1 (maps to slot 1)
return false;
}
// allocframe must always be in slot 0
if (MIb.getOpcode() == Hexagon::S2_allocframe)
return false;
if ((MIaG != HexagonII::HSIG_None) && (MIbG != HexagonII::HSIG_None)) {
// Prevent 2 instructions with extenders from duplexing
// Note that MIb (slot1) can be extended and MIa (slot0)
// can never be extended
if (subInstWouldBeExtended(MIa))
return false;
// If duplexing produces an extender, but the original did not
// have an extender, do not duplex.
if (subInstWouldBeExtended(MIb) && !ExtendedB)
return false;
}
// If jumpr r31 appears, it must be in slot 0, and never slot 1 (MIb).
if (MIbG == HexagonII::HSIG_L2) {
if ((MIb.getNumOperands() > 1) && MIb.getOperand(1).isReg() &&
(MIb.getOperand(1).getReg() == Hexagon::R31))
return false;
if ((MIb.getNumOperands() > 0) && MIb.getOperand(0).isReg() &&
(MIb.getOperand(0).getReg() == Hexagon::R31))
return false;
}
if (STI.getCPU().equals_lower("hexagonv4") ||
STI.getCPU().equals_lower("hexagonv5") ||
STI.getCPU().equals_lower("hexagonv55") ||
STI.getCPU().equals_lower("hexagonv60")) {
// If a store appears, it must be in slot 0 (MIa) 1st, and then slot 1 (MIb);
// therefore, not duplexable if slot 1 is a store, and slot 0 is not.
if ((MIbG == HexagonII::HSIG_S1) || (MIbG == HexagonII::HSIG_S2)) {
if ((MIaG != HexagonII::HSIG_S1) && (MIaG != HexagonII::HSIG_S2))
return false;
}
}
return (isDuplexPairMatch(MIaG, MIbG));
}
void X86MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
OutMI.setOpcode(MI->getOpcode());
for (const MachineOperand &MO : MI->operands())
if (auto MaybeMCOp = LowerMachineOperand(MI, MO))
OutMI.addOperand(MaybeMCOp.getValue());
// Handle a few special cases to eliminate operand modifiers.
ReSimplify:
switch (OutMI.getOpcode()) {
case X86::LEA64_32r:
case X86::LEA64r:
case X86::LEA16r:
case X86::LEA32r:
// LEA should have a segment register, but it must be empty.
assert(OutMI.getNumOperands() == 1+X86::AddrNumOperands &&
"Unexpected # of LEA operands");
assert(OutMI.getOperand(1+X86::AddrSegmentReg).getReg() == 0 &&
"LEA has segment specified!");
break;
case X86::MOV32ri64:
OutMI.setOpcode(X86::MOV32ri);
break;
// Commute operands to get a smaller encoding by using VEX.R instead of VEX.B
// if one of the registers is extended, but other isn't.
case X86::VMOVAPDrr:
case X86::VMOVAPDYrr:
case X86::VMOVAPSrr:
case X86::VMOVAPSYrr:
case X86::VMOVDQArr:
case X86::VMOVDQAYrr:
case X86::VMOVDQUrr:
case X86::VMOVDQUYrr:
case X86::VMOVUPDrr:
case X86::VMOVUPDYrr:
case X86::VMOVUPSrr:
case X86::VMOVUPSYrr: {
if (!X86II::isX86_64ExtendedReg(OutMI.getOperand(0).getReg()) &&
X86II::isX86_64ExtendedReg(OutMI.getOperand(1).getReg())) {
unsigned NewOpc;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::VMOVAPDrr: NewOpc = X86::VMOVAPDrr_REV; break;
case X86::VMOVAPDYrr: NewOpc = X86::VMOVAPDYrr_REV; break;
case X86::VMOVAPSrr: NewOpc = X86::VMOVAPSrr_REV; break;
case X86::VMOVAPSYrr: NewOpc = X86::VMOVAPSYrr_REV; break;
case X86::VMOVDQArr: NewOpc = X86::VMOVDQArr_REV; break;
case X86::VMOVDQAYrr: NewOpc = X86::VMOVDQAYrr_REV; break;
case X86::VMOVDQUrr: NewOpc = X86::VMOVDQUrr_REV; break;
case X86::VMOVDQUYrr: NewOpc = X86::VMOVDQUYrr_REV; break;
case X86::VMOVUPDrr: NewOpc = X86::VMOVUPDrr_REV; break;
case X86::VMOVUPDYrr: NewOpc = X86::VMOVUPDYrr_REV; break;
case X86::VMOVUPSrr: NewOpc = X86::VMOVUPSrr_REV; break;
case X86::VMOVUPSYrr: NewOpc = X86::VMOVUPSYrr_REV; break;
}
OutMI.setOpcode(NewOpc);
}
break;
}
case X86::VMOVSDrr:
case X86::VMOVSSrr: {
if (!X86II::isX86_64ExtendedReg(OutMI.getOperand(0).getReg()) &&
X86II::isX86_64ExtendedReg(OutMI.getOperand(2).getReg())) {
unsigned NewOpc;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::VMOVSDrr: NewOpc = X86::VMOVSDrr_REV; break;
case X86::VMOVSSrr: NewOpc = X86::VMOVSSrr_REV; break;
}
OutMI.setOpcode(NewOpc);
}
break;
}
// TAILJMPr64, CALL64r, CALL64pcrel32 - These instructions have register
// inputs modeled as normal uses instead of implicit uses. As such, truncate
// off all but the first operand (the callee). FIXME: Change isel.
case X86::TAILJMPr64:
case X86::TAILJMPr64_REX:
case X86::CALL64r:
case X86::CALL64pcrel32: {
unsigned Opcode = OutMI.getOpcode();
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
case X86::EH_RETURN:
case X86::EH_RETURN64: {
OutMI = MCInst();
OutMI.setOpcode(getRetOpcode(AsmPrinter.getSubtarget()));
break;
}
// TAILJMPd, TAILJMPd64 - Lower to the correct jump instructions.
case X86::TAILJMPr:
case X86::TAILJMPd:
case X86::TAILJMPd64: {
unsigned Opcode;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::TAILJMPr: Opcode = X86::JMP32r; break;
case X86::TAILJMPd:
case X86::TAILJMPd64: Opcode = X86::JMP_1; break;
}
MCOperand Saved = OutMI.getOperand(0);
OutMI = MCInst();
OutMI.setOpcode(Opcode);
OutMI.addOperand(Saved);
break;
}
case X86::DEC16r:
case X86::DEC32r:
case X86::INC16r:
case X86::INC32r:
// If we aren't in 64-bit mode we can use the 1-byte inc/dec instructions.
if (!AsmPrinter.getSubtarget().is64Bit()) {
unsigned Opcode;
switch (OutMI.getOpcode()) {
default: llvm_unreachable("Invalid opcode");
case X86::DEC16r: Opcode = X86::DEC16r_alt; break;
case X86::DEC32r: Opcode = X86::DEC32r_alt; break;
case X86::INC16r: Opcode = X86::INC16r_alt; break;
case X86::INC32r: Opcode = X86::INC32r_alt; break;
}
OutMI.setOpcode(Opcode);
}
break;
// These are pseudo-ops for OR to help with the OR->ADD transformation. We do
// this with an ugly goto in case the resultant OR uses EAX and needs the
// short form.
case X86::ADD16rr_DB: OutMI.setOpcode(X86::OR16rr); goto ReSimplify;
case X86::ADD32rr_DB: OutMI.setOpcode(X86::OR32rr); goto ReSimplify;
case X86::ADD64rr_DB: OutMI.setOpcode(X86::OR64rr); goto ReSimplify;
case X86::ADD16ri_DB: OutMI.setOpcode(X86::OR16ri); goto ReSimplify;
case X86::ADD32ri_DB: OutMI.setOpcode(X86::OR32ri); goto ReSimplify;
case X86::ADD64ri32_DB: OutMI.setOpcode(X86::OR64ri32); goto ReSimplify;
case X86::ADD16ri8_DB: OutMI.setOpcode(X86::OR16ri8); goto ReSimplify;
case X86::ADD32ri8_DB: OutMI.setOpcode(X86::OR32ri8); goto ReSimplify;
case X86::ADD64ri8_DB: OutMI.setOpcode(X86::OR64ri8); goto ReSimplify;
// Atomic load and store require a separate pseudo-inst because Acquire
// implies mayStore and Release implies mayLoad; fix these to regular MOV
// instructions here
case X86::ACQUIRE_MOV8rm: OutMI.setOpcode(X86::MOV8rm); goto ReSimplify;
case X86::ACQUIRE_MOV16rm: OutMI.setOpcode(X86::MOV16rm); goto ReSimplify;
case X86::ACQUIRE_MOV32rm: OutMI.setOpcode(X86::MOV32rm); goto ReSimplify;
case X86::ACQUIRE_MOV64rm: OutMI.setOpcode(X86::MOV64rm); goto ReSimplify;
case X86::RELEASE_MOV8mr: OutMI.setOpcode(X86::MOV8mr); goto ReSimplify;
case X86::RELEASE_MOV16mr: OutMI.setOpcode(X86::MOV16mr); goto ReSimplify;
case X86::RELEASE_MOV32mr: OutMI.setOpcode(X86::MOV32mr); goto ReSimplify;
case X86::RELEASE_MOV64mr: OutMI.setOpcode(X86::MOV64mr); goto ReSimplify;
case X86::RELEASE_MOV8mi: OutMI.setOpcode(X86::MOV8mi); goto ReSimplify;
case X86::RELEASE_MOV16mi: OutMI.setOpcode(X86::MOV16mi); goto ReSimplify;
case X86::RELEASE_MOV32mi: OutMI.setOpcode(X86::MOV32mi); goto ReSimplify;
case X86::RELEASE_MOV64mi32: OutMI.setOpcode(X86::MOV64mi32); goto ReSimplify;
case X86::RELEASE_ADD8mi: OutMI.setOpcode(X86::ADD8mi); goto ReSimplify;
case X86::RELEASE_ADD32mi: OutMI.setOpcode(X86::ADD32mi); goto ReSimplify;
case X86::RELEASE_ADD64mi32: OutMI.setOpcode(X86::ADD64mi32); goto ReSimplify;
case X86::RELEASE_AND8mi: OutMI.setOpcode(X86::AND8mi); goto ReSimplify;
case X86::RELEASE_AND32mi: OutMI.setOpcode(X86::AND32mi); goto ReSimplify;
case X86::RELEASE_AND64mi32: OutMI.setOpcode(X86::AND64mi32); goto ReSimplify;
case X86::RELEASE_OR8mi: OutMI.setOpcode(X86::OR8mi); goto ReSimplify;
case X86::RELEASE_OR32mi: OutMI.setOpcode(X86::OR32mi); goto ReSimplify;
case X86::RELEASE_OR64mi32: OutMI.setOpcode(X86::OR64mi32); goto ReSimplify;
case X86::RELEASE_XOR8mi: OutMI.setOpcode(X86::XOR8mi); goto ReSimplify;
case X86::RELEASE_XOR32mi: OutMI.setOpcode(X86::XOR32mi); goto ReSimplify;
case X86::RELEASE_XOR64mi32: OutMI.setOpcode(X86::XOR64mi32); goto ReSimplify;
case X86::RELEASE_INC8m: OutMI.setOpcode(X86::INC8m); goto ReSimplify;
case X86::RELEASE_INC16m: OutMI.setOpcode(X86::INC16m); goto ReSimplify;
case X86::RELEASE_INC32m: OutMI.setOpcode(X86::INC32m); goto ReSimplify;
case X86::RELEASE_INC64m: OutMI.setOpcode(X86::INC64m); goto ReSimplify;
case X86::RELEASE_DEC8m: OutMI.setOpcode(X86::DEC8m); goto ReSimplify;
case X86::RELEASE_DEC16m: OutMI.setOpcode(X86::DEC16m); goto ReSimplify;
case X86::RELEASE_DEC32m: OutMI.setOpcode(X86::DEC32m); goto ReSimplify;
case X86::RELEASE_DEC64m: OutMI.setOpcode(X86::DEC64m); goto ReSimplify;
// We don't currently select the correct instruction form for instructions
// which have a short %eax, etc. form. Handle this by custom lowering, for
// now.
//
// Note, we are currently not handling the following instructions:
// MOV64ao8, MOV64o8a
// XCHG16ar, XCHG32ar, XCHG64ar
case X86::MOV8mr_NOREX:
case X86::MOV8mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV8o32a); break;
case X86::MOV8rm_NOREX:
case X86::MOV8rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV8ao32); break;
case X86::MOV16mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV16o32a); break;
case X86::MOV16rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV16ao32); break;
case X86::MOV32mr: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV32o32a); break;
case X86::MOV32rm: SimplifyShortMoveForm(AsmPrinter, OutMI, X86::MOV32ao32); break;
case X86::ADC8ri: SimplifyShortImmForm(OutMI, X86::ADC8i8); break;
case X86::ADC16ri: SimplifyShortImmForm(OutMI, X86::ADC16i16); break;
case X86::ADC32ri: SimplifyShortImmForm(OutMI, X86::ADC32i32); break;
case X86::ADC64ri32: SimplifyShortImmForm(OutMI, X86::ADC64i32); break;
case X86::ADD8ri: SimplifyShortImmForm(OutMI, X86::ADD8i8); break;
case X86::ADD16ri: SimplifyShortImmForm(OutMI, X86::ADD16i16); break;
case X86::ADD32ri: SimplifyShortImmForm(OutMI, X86::ADD32i32); break;
case X86::ADD64ri32: SimplifyShortImmForm(OutMI, X86::ADD64i32); break;
case X86::AND8ri: SimplifyShortImmForm(OutMI, X86::AND8i8); break;
case X86::AND16ri: SimplifyShortImmForm(OutMI, X86::AND16i16); break;
case X86::AND32ri: SimplifyShortImmForm(OutMI, X86::AND32i32); break;
case X86::AND64ri32: SimplifyShortImmForm(OutMI, X86::AND64i32); break;
case X86::CMP8ri: SimplifyShortImmForm(OutMI, X86::CMP8i8); break;
case X86::CMP16ri: SimplifyShortImmForm(OutMI, X86::CMP16i16); break;
case X86::CMP32ri: SimplifyShortImmForm(OutMI, X86::CMP32i32); break;
case X86::CMP64ri32: SimplifyShortImmForm(OutMI, X86::CMP64i32); break;
case X86::OR8ri: SimplifyShortImmForm(OutMI, X86::OR8i8); break;
case X86::OR16ri: SimplifyShortImmForm(OutMI, X86::OR16i16); break;
case X86::OR32ri: SimplifyShortImmForm(OutMI, X86::OR32i32); break;
case X86::OR64ri32: SimplifyShortImmForm(OutMI, X86::OR64i32); break;
case X86::SBB8ri: SimplifyShortImmForm(OutMI, X86::SBB8i8); break;
case X86::SBB16ri: SimplifyShortImmForm(OutMI, X86::SBB16i16); break;
case X86::SBB32ri: SimplifyShortImmForm(OutMI, X86::SBB32i32); break;
case X86::SBB64ri32: SimplifyShortImmForm(OutMI, X86::SBB64i32); break;
case X86::SUB8ri: SimplifyShortImmForm(OutMI, X86::SUB8i8); break;
case X86::SUB16ri: SimplifyShortImmForm(OutMI, X86::SUB16i16); break;
case X86::SUB32ri: SimplifyShortImmForm(OutMI, X86::SUB32i32); break;
case X86::SUB64ri32: SimplifyShortImmForm(OutMI, X86::SUB64i32); break;
case X86::TEST8ri: SimplifyShortImmForm(OutMI, X86::TEST8i8); break;
case X86::TEST16ri: SimplifyShortImmForm(OutMI, X86::TEST16i16); break;
case X86::TEST32ri: SimplifyShortImmForm(OutMI, X86::TEST32i32); break;
case X86::TEST64ri32: SimplifyShortImmForm(OutMI, X86::TEST64i32); break;
case X86::XOR8ri: SimplifyShortImmForm(OutMI, X86::XOR8i8); break;
case X86::XOR16ri: SimplifyShortImmForm(OutMI, X86::XOR16i16); break;
case X86::XOR32ri: SimplifyShortImmForm(OutMI, X86::XOR32i32); break;
case X86::XOR64ri32: SimplifyShortImmForm(OutMI, X86::XOR64i32); break;
// Try to shrink some forms of movsx.
case X86::MOVSX16rr8:
case X86::MOVSX32rr16:
case X86::MOVSX64rr32:
SimplifyMOVSX(OutMI);
break;
}
}
MCInst HexagonMCInstrInfo::deriveSubInst(MCInst const &Inst) {
MCInst Result;
bool Absolute;
int64_t Value;
switch (Inst.getOpcode()) {
default:
// dbgs() << "opcode: "<< Inst->getOpcode() << "\n";
llvm_unreachable("Unimplemented subinstruction \n");
break;
case Hexagon::A2_addi:
Absolute = Inst.getOperand(2).getExpr()->evaluateAsAbsolute(Value);
if (Absolute) {
if (Value == 1) {
Result.setOpcode(Hexagon::SA1_inc);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break;
} // 1,2 SUBInst $Rd = add($Rs, #1)
if (Value == -1) {
Result.setOpcode(Hexagon::SA1_dec);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break;
} // 1,2 SUBInst $Rd = add($Rs,#-1)
if (Inst.getOperand(1).getReg() == Hexagon::R29) {
Result.setOpcode(Hexagon::SA1_addsp);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break;
} // 1,3 SUBInst $Rd = add(r29, #$u6_2)
}
Result.setOpcode(Hexagon::SA1_addi);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rx = add($Rx, #$s7)
case Hexagon::A2_add:
Result.setOpcode(Hexagon::SA1_addrx);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rx = add($_src_, $Rs)
case Hexagon::S2_allocframe:
Result.setOpcode(Hexagon::SS2_allocframe);
addOps(Result, Inst, 0);
break; // 1 SUBInst allocframe(#$u5_3)
case Hexagon::A2_andir:
if (minConstant(Inst, 2) == 255) {
Result.setOpcode(Hexagon::SA1_zxtb);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 $Rd = and($Rs, #255)
} else {
Result.setOpcode(Hexagon::SA1_and1);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rd = and($Rs, #1)
}
case Hexagon::C2_cmpeqi:
Result.setOpcode(Hexagon::SA1_cmpeqi);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 2,3 SUBInst p0 = cmp.eq($Rs, #$u2)
case Hexagon::A4_combineii:
case Hexagon::A2_combineii:
Absolute = Inst.getOperand(1).getExpr()->evaluateAsAbsolute(Value);
assert(Absolute);(void)Absolute;
if (Value == 1) {
Result.setOpcode(Hexagon::SA1_combine1i);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 1,3 SUBInst $Rdd = combine(#1, #$u2)
}
if (Value == 3) {
Result.setOpcode(Hexagon::SA1_combine3i);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 1,3 SUBInst $Rdd = combine(#3, #$u2)
}
if (Value == 0) {
Result.setOpcode(Hexagon::SA1_combine0i);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 1,3 SUBInst $Rdd = combine(#0, #$u2)
}
if (Value == 2) {
Result.setOpcode(Hexagon::SA1_combine2i);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 1,3 SUBInst $Rdd = combine(#2, #$u2)
}
case Hexagon::A4_combineir:
Result.setOpcode(Hexagon::SA1_combinezr);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 1,3 SUBInst $Rdd = combine(#0, $Rs)
case Hexagon::A4_combineri:
Result.setOpcode(Hexagon::SA1_combinerz);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rdd = combine($Rs, #0)
case Hexagon::L4_return_tnew_pnt:
case Hexagon::L4_return_tnew_pt:
Result.setOpcode(Hexagon::SL2_return_tnew);
break; // none SUBInst if (p0.new) dealloc_return:nt
case Hexagon::L4_return_fnew_pnt:
case Hexagon::L4_return_fnew_pt:
Result.setOpcode(Hexagon::SL2_return_fnew);
break; // none SUBInst if (!p0.new) dealloc_return:nt
case Hexagon::L4_return_f:
Result.setOpcode(Hexagon::SL2_return_f);
break; // none SUBInst if (!p0) dealloc_return
case Hexagon::L4_return_t:
Result.setOpcode(Hexagon::SL2_return_t);
break; // none SUBInst if (p0) dealloc_return
case Hexagon::L4_return:
Result.setOpcode(Hexagon::SL2_return);
break; // none SUBInst dealloc_return
case Hexagon::L2_deallocframe:
Result.setOpcode(Hexagon::SL2_deallocframe);
break; // none SUBInst deallocframe
case Hexagon::EH_RETURN_JMPR:
case Hexagon::J2_jumpr:
case Hexagon::PS_jmpret:
Result.setOpcode(Hexagon::SL2_jumpr31);
break; // none SUBInst jumpr r31
case Hexagon::J2_jumprf:
case Hexagon::PS_jmpretf:
Result.setOpcode(Hexagon::SL2_jumpr31_f);
break; // none SUBInst if (!p0) jumpr r31
case Hexagon::J2_jumprfnew:
case Hexagon::J2_jumprfnewpt:
case Hexagon::PS_jmpretfnewpt:
case Hexagon::PS_jmpretfnew:
Result.setOpcode(Hexagon::SL2_jumpr31_fnew);
break; // none SUBInst if (!p0.new) jumpr:nt r31
case Hexagon::J2_jumprt:
case Hexagon::PS_jmprett:
Result.setOpcode(Hexagon::SL2_jumpr31_t);
break; // none SUBInst if (p0) jumpr r31
case Hexagon::J2_jumprtnew:
case Hexagon::J2_jumprtnewpt:
case Hexagon::PS_jmprettnewpt:
case Hexagon::PS_jmprettnew:
Result.setOpcode(Hexagon::SL2_jumpr31_tnew);
break; // none SUBInst if (p0.new) jumpr:nt r31
case Hexagon::L2_loadrb_io:
Result.setOpcode(Hexagon::SL2_loadrb_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rd = memb($Rs + #$u3_0)
case Hexagon::L2_loadrd_io:
Result.setOpcode(Hexagon::SL2_loadrd_sp);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 1,3 SUBInst $Rdd = memd(r29 + #$u5_3)
case Hexagon::L2_loadrh_io:
Result.setOpcode(Hexagon::SL2_loadrh_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rd = memh($Rs + #$u3_1)
case Hexagon::L2_loadrub_io:
Result.setOpcode(Hexagon::SL1_loadrub_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rd = memub($Rs + #$u4_0)
case Hexagon::L2_loadruh_io:
Result.setOpcode(Hexagon::SL2_loadruh_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rd = memuh($Rs + #$u3_1)
case Hexagon::L2_loadri_io:
if (Inst.getOperand(1).getReg() == Hexagon::R29) {
Result.setOpcode(Hexagon::SL2_loadri_sp);
addOps(Result, Inst, 0);
addOps(Result, Inst, 2);
break; // 2 1,3 SUBInst $Rd = memw(r29 + #$u5_2)
} else {
Result.setOpcode(Hexagon::SL1_loadri_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst $Rd = memw($Rs + #$u4_2)
}
case Hexagon::S4_storeirb_io:
Absolute = Inst.getOperand(2).getExpr()->evaluateAsAbsolute(Value);
assert(Absolute);(void)Absolute;
if (Value == 0) {
Result.setOpcode(Hexagon::SS2_storebi0);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst memb($Rs + #$u4_0)=#0
} else if (Value == 1) {
Result.setOpcode(Hexagon::SS2_storebi1);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 2 1,2 SUBInst memb($Rs + #$u4_0)=#1
}
case Hexagon::S2_storerb_io:
Result.setOpcode(Hexagon::SS1_storeb_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst memb($Rs + #$u4_0) = $Rt
case Hexagon::S2_storerd_io:
Result.setOpcode(Hexagon::SS2_stored_sp);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 2,3 SUBInst memd(r29 + #$s6_3) = $Rtt
case Hexagon::S2_storerh_io:
Result.setOpcode(Hexagon::SS2_storeh_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1,2,3 SUBInst memb($Rs + #$u4_0) = $Rt
case Hexagon::S4_storeiri_io:
Absolute = Inst.getOperand(2).getExpr()->evaluateAsAbsolute(Value);
assert(Absolute);(void)Absolute;
if (Value == 0) {
Result.setOpcode(Hexagon::SS2_storewi0);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 3 1,2 SUBInst memw($Rs + #$u4_2)=#0
} else if (Value == 1) {
Result.setOpcode(Hexagon::SS2_storewi1);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 3 1,2 SUBInst memw($Rs + #$u4_2)=#1
} else if (Inst.getOperand(0).getReg() == Hexagon::R29) {
Result.setOpcode(Hexagon::SS2_storew_sp);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2);
break; // 1 2,3 SUBInst memw(r29 + #$u5_2) = $Rt
}
case Hexagon::S2_storeri_io:
if (Inst.getOperand(0).getReg() == Hexagon::R29) {
Result.setOpcode(Hexagon::SS2_storew_sp);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2); // 1,2,3 SUBInst memw(sp + #$u5_2) = $Rt
} else {
Result.setOpcode(Hexagon::SS1_storew_io);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
addOps(Result, Inst, 2); // 1,2,3 SUBInst memw($Rs + #$u4_2) = $Rt
}
break;
case Hexagon::A2_sxtb:
Result.setOpcode(Hexagon::SA1_sxtb);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rd = sxtb($Rs)
case Hexagon::A2_sxth:
Result.setOpcode(Hexagon::SA1_sxth);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rd = sxth($Rs)
case Hexagon::A2_tfr:
Result.setOpcode(Hexagon::SA1_tfr);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rd = $Rs
case Hexagon::C2_cmovenewif:
Result.setOpcode(Hexagon::SA1_clrfnew);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 2 SUBInst if (!p0.new) $Rd = #0
case Hexagon::C2_cmovenewit:
Result.setOpcode(Hexagon::SA1_clrtnew);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 2 SUBInst if (p0.new) $Rd = #0
case Hexagon::C2_cmoveif:
Result.setOpcode(Hexagon::SA1_clrf);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 2 SUBInst if (!p0) $Rd = #0
case Hexagon::C2_cmoveit:
Result.setOpcode(Hexagon::SA1_clrt);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 2 SUBInst if (p0) $Rd = #0
case Hexagon::A2_tfrsi:
Absolute = Inst.getOperand(1).getExpr()->evaluateAsAbsolute(Value);
if (Absolute && Value == -1) {
Result.setOpcode(Hexagon::SA1_setin1);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 2 1 SUBInst $Rd = #-1
} else {
Result.setOpcode(Hexagon::SA1_seti);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rd = #$u6
}
case Hexagon::A2_zxtb:
Result.setOpcode(Hexagon::SA1_zxtb);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 $Rd = and($Rs, #255)
case Hexagon::A2_zxth:
Result.setOpcode(Hexagon::SA1_zxth);
addOps(Result, Inst, 0);
addOps(Result, Inst, 1);
break; // 1,2 SUBInst $Rd = zxth($Rs)
}
return Result;
}
/// EmitInstruction -- Print out a single PowerPC MI in Darwin syntax to
/// the current output stream.
///
void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
MCInst TmpInst;
// Lower multi-instruction pseudo operations.
switch (MI->getOpcode()) {
default: break;
case TargetOpcode::DBG_VALUE: {
if (!isVerbose() || !OutStreamer.hasRawTextSupport()) return;
SmallString<32> Str;
raw_svector_ostream O(Str);
unsigned NOps = MI->getNumOperands();
assert(NOps==4);
O << '\t' << MAI->getCommentString() << "DEBUG_VALUE: ";
// cast away const; DIetc do not take const operands for some reason.
DIVariable V(const_cast<MDNode *>(MI->getOperand(NOps-1).getMetadata()));
O << V.getName();
O << " <- ";
// Frame address. Currently handles register +- offset only.
assert(MI->getOperand(0).isReg() && MI->getOperand(1).isImm());
O << '['; printOperand(MI, 0, O); O << '+'; printOperand(MI, 1, O);
O << ']';
O << "+";
printOperand(MI, NOps-2, O);
OutStreamer.EmitRawText(O.str());
return;
}
case PPC::MovePCtoLR:
case PPC::MovePCtoLR8: {
// Transform %LR = MovePCtoLR
// Into this, where the label is the PIC base:
// bl L1$pb
// L1$pb:
MCSymbol *PICBase = MF->getPICBaseSymbol();
// Emit the 'bl'.
TmpInst.setOpcode(PPC::BL_Darwin); // Darwin vs SVR4 doesn't matter here.
// FIXME: We would like an efficient form for this, so we don't have to do
// a lot of extra uniquing.
TmpInst.addOperand(MCOperand::CreateExpr(MCSymbolRefExpr::
Create(PICBase, OutContext)));
OutStreamer.EmitInstruction(TmpInst);
// Emit the label.
OutStreamer.EmitLabel(PICBase);
return;
}
case PPC::LDtoc: {
// Transform %X3 = LDtoc <ga:@min1>, %X2
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
// Change the opcode to LD, and the global address operand to be a
// reference to the TOC entry we will synthesize later.
TmpInst.setOpcode(PPC::LD);
const MachineOperand &MO = MI->getOperand(1);
assert(MO.isGlobal());
// Map symbol -> label of TOC entry.
MCSymbol *&TOCEntry = TOC[Mang->getSymbol(MO.getGlobal())];
if (TOCEntry == 0)
TOCEntry = GetTempSymbol("C", TOCLabelID++);
const MCExpr *Exp =
MCSymbolRefExpr::Create(TOCEntry, MCSymbolRefExpr::VK_PPC_TOC,
OutContext);
TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
OutStreamer.EmitInstruction(TmpInst);
return;
}
case PPC::MFCRpseud:
case PPC::MFCR8pseud:
// Transform: %R3 = MFCRpseud %CR7
// Into: %R3 = MFCR ;; cr7
OutStreamer.AddComment(PPCInstPrinter::
getRegisterName(MI->getOperand(1).getReg()));
TmpInst.setOpcode(Subtarget.isPPC64() ? PPC::MFCR8 : PPC::MFCR);
TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
OutStreamer.EmitInstruction(TmpInst);
return;
case PPC::SYNC:
// In Book E sync is called msync, handle this special case here...
if (Subtarget.isBookE()) {
OutStreamer.EmitRawText(StringRef("\tmsync"));
return;
}
}
LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
OutStreamer.EmitInstruction(TmpInst);
}
//===----------------------------------------------------------------------===//
void OR1KAsmPrinter::customEmitInstruction(const MachineInstr *MI) {
OR1KMCInstLower MCInstLowering(OutContext, *Mang, *this);
unsigned Opcode = MI->getOpcode();
MCSubtargetInfo STI = getSubtargetInfo();
switch (Opcode) {
default: break;
case OR1K::MOVHI:
case OR1K::ORI: {
MCSymbolRefExpr::VariantKind Kind = MCSymbolRefExpr::VK_None;
if (Opcode == OR1K::MOVHI &&
MI->getOperand(1).getTargetFlags() == OR1KII::MO_GOTPCHI)
Kind = MCSymbolRefExpr::VK_OR1K_GOTPCHI;
else if (Opcode == OR1K::ORI &&
MI->getOperand(2).getTargetFlags() == OR1KII::MO_GOTPCLO)
Kind = MCSymbolRefExpr::VK_OR1K_GOTPCLO;
else
break;
// We want to print something like:
// MYGLOBAL + (. - PICBASE)
// However, we can't generate a ".", so just emit a new label here and refer
// to it.
MCSymbol *DotSym = OutContext.CreateTempSymbol();
const MCExpr *DotExpr = MCSymbolRefExpr::Create(DotSym, OutContext);
const MCExpr *PICBase =
MCSymbolRefExpr::Create(MF->getPICBaseSymbol(), OutContext);
OutStreamer.EmitLabel(DotSym);
// Now that we have emitted the label, lower the complex operand expression.
MachineOperand MO = (MI->getOpcode() == OR1K::MOVHI) ?
MI->getOperand(1) : MI->getOperand(2);
MCSymbol *OpSym = MCInstLowering.GetExternalSymbolSymbol(MO);
DotExpr = MCBinaryExpr::CreateSub(DotExpr, PICBase, OutContext);
DotExpr = MCBinaryExpr::CreateAdd(MCSymbolRefExpr::Create(OpSym, Kind,
OutContext),
DotExpr, OutContext);
MCInst TmpInst;
TmpInst.setOpcode(MI->getOpcode());
TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
if (MI->getOpcode() == OR1K::ORI)
TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
TmpInst.addOperand(MCOperand::CreateExpr(DotExpr));
OutStreamer.EmitInstruction(TmpInst, STI);
return;
}
case OR1K::GETPC: {
MCInst TmpInst;
// This is a pseudo op for a two instruction sequence with a label, which
// looks like:
// l.jal .L1$pb
// l.nop
// .L1$pb:
// Emit the call.
MCSymbol *PICBase = MF->getPICBaseSymbol();
TmpInst.setOpcode(OR1K::JAL);
// FIXME: We would like an efficient form for this, so we don't have to do a
// lot of extra uniquing.
TmpInst.addOperand(MCOperand::CreateExpr(
MCSymbolRefExpr::Create(PICBase,OutContext)));
OutStreamer.EmitInstruction(TmpInst, STI);
// Emit delay-slot nop
// FIXME: omit on no-delay-slot targets
TmpInst.setOpcode(OR1K::NOP);
TmpInst.getOperand(0) = MCOperand::CreateImm(0);
OutStreamer.EmitInstruction(TmpInst, STI);
// Emit the label.
OutStreamer.EmitLabel(PICBase);
return;
}
}
MCInst TmpInst;
MCInstLowering.Lower(MI, TmpInst);
OutStreamer.EmitInstruction(TmpInst, STI);
}
void HexagonMCChecker::init(MCInst const &MCI) {
const MCInstrDesc &MCID = HexagonMCInstrInfo::getDesc(MCII, MCI);
unsigned PredReg = Hexagon::NoRegister;
bool isTrue = false;
// Get used registers.
for (unsigned i = MCID.getNumDefs(); i < MCID.getNumOperands(); ++i)
if (MCI.getOperand(i).isReg())
initReg(MCI, MCI.getOperand(i).getReg(), PredReg, isTrue);
for (unsigned i = 0; i < MCID.getNumImplicitUses(); ++i)
initReg(MCI, MCID.getImplicitUses()[i], PredReg, isTrue);
// Get implicit register definitions.
if (const MCPhysReg *ImpDef = MCID.getImplicitDefs())
for (; *ImpDef; ++ImpDef) {
unsigned R = *ImpDef;
if (Hexagon::R31 != R && MCID.isCall())
// Any register other than the LR and the PC are actually volatile ones
// as defined by the ABI, not modified implicitly by the call insn.
continue;
if (Hexagon::PC == R)
// Branches are the only insns that can change the PC,
// otherwise a read-only register.
continue;
if (Hexagon::USR_OVF == R)
// Many insns change the USR implicitly, but only one or another flag.
// The instruction table models the USR.OVF flag, which can be
// implicitly modified more than once, but cannot be modified in the
// same packet with an instruction that modifies is explicitly. Deal
// with such situations individually.
SoftDefs.insert(R);
else if (isPredicateRegister(R) &&
HexagonMCInstrInfo::isPredicateLate(MCII, MCI))
// Include implicit late predicates.
LatePreds.insert(R);
else
Defs[R].insert(PredSense(PredReg, isTrue));
}
// Figure out explicit register definitions.
for (unsigned i = 0; i < MCID.getNumDefs(); ++i) {
unsigned R = MCI.getOperand(i).getReg(), S = Hexagon::NoRegister;
// USR has subregisters (while C8 does not for technical reasons), so
// reset R to USR, since we know how to handle multiple defs of USR,
// taking into account its subregisters.
if (R == Hexagon::C8)
R = Hexagon::USR;
// Note register definitions, direct ones as well as indirect side-effects.
// Super-registers are not tracked directly, but their components.
for (MCRegAliasIterator SRI(R, &RI, !MCSubRegIterator(R, &RI).isValid());
SRI.isValid(); ++SRI) {
if (MCSubRegIterator(*SRI, &RI).isValid())
// Skip super-registers defined indirectly.
continue;
if (R == *SRI) {
if (S == R)
// Avoid scoring the defined register multiple times.
continue;
else
// Note that the defined register has already been scored.
S = R;
}
if (Hexagon::P3_0 != R && Hexagon::P3_0 == *SRI)
// P3:0 is a special case, since multiple predicate register definitions
// in a packet is allowed as the equivalent of their logical "and".
// Only an explicit definition of P3:0 is noted as such; if a
// side-effect, then note as a soft definition.
SoftDefs.insert(*SRI);
else if (HexagonMCInstrInfo::isPredicateLate(MCII, MCI) &&
isPredicateRegister(*SRI))
// Some insns produce predicates too late to be used in the same packet.
LatePreds.insert(*SRI);
else if (i == 0 && HexagonMCInstrInfo::getType(MCII, MCI) ==
HexagonII::TypeCVI_VM_TMP_LD)
// Temporary loads should be used in the same packet, but don't commit
// results, so it should be disregarded if another insn changes the same
// register.
// TODO: relies on the impossibility of a current and a temporary loads
// in the same packet.
TmpDefs.insert(*SRI);
else if (i <= 1 && HexagonMCInstrInfo::hasNewValue2(MCII, MCI))
// vshuff(Vx, Vy, Rx) <- Vx(0) and Vy(1) are both source and
// destination registers with this instruction. same for vdeal(Vx,Vy,Rx)
Uses.insert(*SRI);
else
Defs[*SRI].insert(PredSense(PredReg, isTrue));
}
}
// Figure out definitions of new predicate registers.
if (HexagonMCInstrInfo::isPredicatedNew(MCII, MCI))
for (unsigned i = MCID.getNumDefs(); i < MCID.getNumOperands(); ++i)
if (MCI.getOperand(i).isReg()) {
unsigned P = MCI.getOperand(i).getReg();
if (isPredicateRegister(P))
NewPreds.insert(P);
}
}
/// LowerUnaryToTwoAddr - R = setb -> R = sbb R, R
static void LowerUnaryToTwoAddr(MCInst &OutMI, unsigned NewOpc) {
OutMI.setOpcode(NewOpc);
OutMI.addOperand(OutMI.getOperand(0));
OutMI.addOperand(OutMI.getOperand(0));
}
static bool getMCRDeprecationInfo(MCInst &MI, MCSubtargetInfo &STI,
std::string &Info) {
if (STI.getFeatureBits()[llvm::ARM::HasV7Ops] &&
(MI.getOperand(0).isImm() && MI.getOperand(0).getImm() == 15) &&
(MI.getOperand(1).isImm() && MI.getOperand(1).getImm() == 0) &&
// Checks for the deprecated CP15ISB encoding:
// mcr p15, #0, rX, c7, c5, #4
(MI.getOperand(3).isImm() && MI.getOperand(3).getImm() == 7)) {
if ((MI.getOperand(5).isImm() && MI.getOperand(5).getImm() == 4)) {
if (MI.getOperand(4).isImm() && MI.getOperand(4).getImm() == 5) {
Info = "deprecated since v7, use 'isb'";
return true;
}
// Checks for the deprecated CP15DSB encoding:
// mcr p15, #0, rX, c7, c10, #4
if (MI.getOperand(4).isImm() && MI.getOperand(4).getImm() == 10) {
Info = "deprecated since v7, use 'dsb'";
return true;
}
}
// Checks for the deprecated CP15DMB encoding:
// mcr p15, #0, rX, c7, c10, #5
if (MI.getOperand(4).isImm() && MI.getOperand(4).getImm() == 10 &&
(MI.getOperand(5).isImm() && MI.getOperand(5).getImm() == 5)) {
Info = "deprecated since v7, use 'dmb'";
return true;
}
}
return false;
}
void HexagonMCInstrInfo::setOuterLoop(MCInst &MCI) {
assert(isBundle(MCI));
MCOperand &Operand = MCI.getOperand(0);
Operand.setImm(Operand.getImm() | outerLoopMask);
}
