C++ iterator::getOperand方法代码示例

bool X86CallFrameOptimization::adjustCallSequence(MachineFunction &MF,
                                                  const CallContext &Context) {
  // Ok, we can in fact do the transformation for this call.
  // Do not remove the FrameSetup instruction, but adjust the parameters.
  // PEI will end up finalizing the handling of this.
  MachineBasicBlock::iterator FrameSetup = Context.FrameSetup;
  MachineBasicBlock &MBB = *(FrameSetup->getParent());
  FrameSetup->getOperand(1).setImm(Context.ExpectedDist);

  DebugLoc DL = FrameSetup->getDebugLoc();
  // Now, iterate through the vector in reverse order, and replace the movs
  // with pushes. MOVmi/MOVmr doesn't have any defs, so no need to
  // replace uses.
  for (int Idx = (Context.ExpectedDist / 4) - 1; Idx >= 0; --Idx) {
    MachineBasicBlock::iterator MOV = *Context.MovVector[Idx];
    MachineOperand PushOp = MOV->getOperand(X86::AddrNumOperands);
    MachineBasicBlock::iterator Push = nullptr;
    if (MOV->getOpcode() == X86::MOV32mi) {
      unsigned PushOpcode = X86::PUSHi32;
      // If the operand is a small (8-bit) immediate, we can use a
      // PUSH instruction with a shorter encoding.
      // Note that isImm() may fail even though this is a MOVmi, because
      // the operand can also be a symbol.
      if (PushOp.isImm()) {
        int64_t Val = PushOp.getImm();
        if (isInt<8>(Val))
          PushOpcode = X86::PUSH32i8;
      }
      Push = BuildMI(MBB, Context.Call, DL, TII->get(PushOpcode))
          .addOperand(PushOp);
    } else {
      unsigned int Reg = PushOp.getReg();

      // If PUSHrmm is not slow on this target, try to fold the source of the
      // push into the instruction.
      bool SlowPUSHrmm = STI->isAtom() || STI->isSLM();

      // Check that this is legal to fold. Right now, we're extremely
      // conservative about that.
      MachineInstr *DefMov = nullptr;
      if (!SlowPUSHrmm && (DefMov = canFoldIntoRegPush(FrameSetup, Reg))) {
        Push = BuildMI(MBB, Context.Call, DL, TII->get(X86::PUSH32rmm));

        unsigned NumOps = DefMov->getDesc().getNumOperands();
        for (unsigned i = NumOps - X86::AddrNumOperands; i != NumOps; ++i)
          Push->addOperand(DefMov->getOperand(i));

        DefMov->eraseFromParent();
      } else {
        Push = BuildMI(MBB, Context.Call, DL, TII->get(X86::PUSH32r))
            .addReg(Reg)
            .getInstr();
      }
    }

    // For debugging, when using SP-based CFA, we need to adjust the CFA
    // offset after each push.
    // TODO: This is needed only if we require precise CFA.
    if (!TFL->hasFP(MF))
      TFL->BuildCFI(MBB, std::next(Push), DL, 
                    MCCFIInstruction::createAdjustCfaOffset(nullptr, 4));

    MBB.erase(MOV);
  }

  // The stack-pointer copy is no longer used in the call sequences.
  // There should not be any other users, but we can't commit to that, so:
  if (MRI->use_empty(Context.SPCopy->getOperand(0).getReg()))
    Context.SPCopy->eraseFromParent();

  // Once we've done this, we need to make sure PEI doesn't assume a reserved
  // frame.
  X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
  FuncInfo->setHasPushSequences(true);

  return true;
}

/// findSurvivorReg - Return the candidate register that is unused for the
/// longest after StargMII. UseMI is set to the instruction where the search
/// stopped.
///
/// No more than InstrLimit instructions are inspected.
///
unsigned RegScavenger::findSurvivorReg(MachineBasicBlock::iterator StartMI,
                                       BitVector &Candidates,
                                       unsigned InstrLimit,
                                       MachineBasicBlock::iterator &UseMI) {
  int Survivor = Candidates.find_first();
  assert(Survivor > 0 && "No candidates for scavenging");

  MachineBasicBlock::iterator ME = MBB->getFirstTerminator();
  assert(StartMI != ME && "MI already at terminator");
  MachineBasicBlock::iterator RestorePointMI = StartMI;
  MachineBasicBlock::iterator MI = StartMI;

  bool inVirtLiveRange = false;
  for (++MI; InstrLimit > 0 && MI != ME; ++MI, --InstrLimit) {
    if (MI->isDebugValue()) {
      ++InstrLimit; // Don't count debug instructions
      continue;
    }
    bool isVirtKillInsn = false;
    bool isVirtDefInsn = false;
    // Remove any candidates touched by instruction.
    for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
      const MachineOperand &MO = MI->getOperand(i);
      if (MO.isRegMask())
        Candidates.clearBitsNotInMask(MO.getRegMask());
      if (!MO.isReg() || MO.isUndef() || !MO.getReg())
        continue;
      if (TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
        if (MO.isDef())
          isVirtDefInsn = true;
        else if (MO.isKill())
          isVirtKillInsn = true;
        continue;
      }
      for (MCRegAliasIterator AI(MO.getReg(), TRI, true); AI.isValid(); ++AI)
        Candidates.reset(*AI);
    }
    // If we're not in a virtual reg's live range, this is a valid
    // restore point.
    if (!inVirtLiveRange) RestorePointMI = MI;

    // Update whether we're in the live range of a virtual register
    if (isVirtKillInsn) inVirtLiveRange = false;
    if (isVirtDefInsn) inVirtLiveRange = true;

    // Was our survivor untouched by this instruction?
    if (Candidates.test(Survivor))
      continue;

    // All candidates gone?
    if (Candidates.none())
      break;

    Survivor = Candidates.find_first();
  }
  // If we ran off the end, that's where we want to restore.
  if (MI == ME) RestorePointMI = ME;
  assert (RestorePointMI != StartMI &&
          "No available scavenger restore location!");

  // We ran out of candidates, so stop the search.
  UseMI = RestorePointMI;
  return Survivor;
}

X86CallFrameOptimization::InstClassification
X86CallFrameOptimization::classifyInstruction(
    MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
    const X86RegisterInfo &RegInfo, DenseSet<unsigned int> &UsedRegs) {
  if (MI == MBB.end())
    return Exit;

  // The instructions we actually care about are movs onto the stack or special
  // cases of constant-stores to stack
  switch (MI->getOpcode()) {
    case X86::AND16mi8:
    case X86::AND32mi8:
    case X86::AND64mi8: {
      MachineOperand ImmOp = MI->getOperand(X86::AddrNumOperands);
      return ImmOp.getImm() == 0 ? Convert : Exit;
    }
    case X86::OR16mi8:
    case X86::OR32mi8:
    case X86::OR64mi8: {
      MachineOperand ImmOp = MI->getOperand(X86::AddrNumOperands);
      return ImmOp.getImm() == -1 ? Convert : Exit;
    }
    case X86::MOV32mi:
    case X86::MOV32mr:
    case X86::MOV64mi32:
    case X86::MOV64mr:
      return Convert;
  }

  // Not all calling conventions have only stack MOVs between the stack
  // adjust and the call.

  // We want to tolerate other instructions, to cover more cases.
  // In particular:
  // a) PCrel calls, where we expect an additional COPY of the basereg.
  // b) Passing frame-index addresses.
  // c) Calling conventions that have inreg parameters. These generate
  //    both copies and movs into registers.
  // To avoid creating lots of special cases, allow any instruction
  // that does not write into memory, does not def or use the stack
  // pointer, and does not def any register that was used by a preceding
  // push.
  // (Reading from memory is allowed, even if referenced through a
  // frame index, since these will get adjusted properly in PEI)

  // The reason for the last condition is that the pushes can't replace
  // the movs in place, because the order must be reversed.
  // So if we have a MOV32mr that uses EDX, then an instruction that defs
  // EDX, and then the call, after the transformation the push will use
  // the modified version of EDX, and not the original one.
  // Since we are still in SSA form at this point, we only need to
  // make sure we don't clobber any *physical* registers that were
  // used by an earlier mov that will become a push.

  if (MI->isCall() || MI->mayStore())
    return Exit;

  for (const MachineOperand &MO : MI->operands()) {
    if (!MO.isReg())
      continue;
    unsigned int Reg = MO.getReg();
    if (!RegInfo.isPhysicalRegister(Reg))
      continue;
    if (RegInfo.regsOverlap(Reg, RegInfo.getStackRegister()))
      return Exit;
    if (MO.isDef()) {
      for (unsigned int U : UsedRegs)
        if (RegInfo.regsOverlap(Reg, U))
          return Exit;
    }
  }

  return Skip;
}

void SIInsertWaits::pushInstruction(MachineBasicBlock &MBB,
                                    MachineBasicBlock::iterator I,
                                    const Counters &Increment) {

  // Get the hardware counter increments and sum them up
  Counters Limit = ZeroCounts;
  unsigned Sum = 0;

  if (TII->mayAccessFlatAddressSpace(*I))
    IsFlatOutstanding = true;

  for (unsigned i = 0; i < 3; ++i) {
    LastIssued.Array[i] += Increment.Array[i];
    if (Increment.Array[i])
      Limit.Array[i] = LastIssued.Array[i];
    Sum += Increment.Array[i];
  }

  // If we don't increase anything then that's it
  if (Sum == 0) {
    LastOpcodeType = OTHER;
    return;
  }

  if (ST->getGeneration() >= SISubtarget::VOLCANIC_ISLANDS) {
    // Any occurrence of consecutive VMEM or SMEM instructions forms a VMEM
    // or SMEM clause, respectively.
    //
    // The temporary workaround is to break the clauses with S_NOP.
    //
    // The proper solution would be to allocate registers such that all source
    // and destination registers don't overlap, e.g. this is illegal:
    //   r0 = load r2
    //   r2 = load r0
    if (LastOpcodeType == VMEM && Increment.Named.VM) {
      // Insert a NOP to break the clause.
      BuildMI(MBB, I, DebugLoc(), TII->get(AMDGPU::S_NOP))
          .addImm(0);
      LastInstWritesM0 = false;
    }

    if (TII->isSMRD(*I))
      LastOpcodeType = SMEM;
    else if (Increment.Named.VM)
      LastOpcodeType = VMEM;
  }

  // Remember which export instructions we have seen
  if (Increment.Named.EXP) {
    ExpInstrTypesSeen |= TII->isEXP(*I) ? 1 : 2;
  }

  for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
    MachineOperand &Op = I->getOperand(i);
    if (!isOpRelevant(Op))
      continue;

    const TargetRegisterClass *RC = TII->getOpRegClass(*I, i);
    RegInterval Interval = getRegInterval(RC, Op);
    for (unsigned j = Interval.first; j < Interval.second; ++j) {

      // Remember which registers we define
      if (Op.isDef())
        DefinedRegs[j] = Limit;

      // and which one we are using
      if (Op.isUse())
        UsedRegs[j] = Limit;
    }
  }
}

// Finds compare instruction that corresponds to supported types of branching.
// Returns the instruction or nullptr on failures or detecting unsupported
// instructions.
MachineInstr *AArch64ConditionOptimizer::findSuitableCompare(
    MachineBasicBlock *MBB) {
  MachineBasicBlock::iterator I = MBB->getFirstTerminator();
  if (I == MBB->end())
    return nullptr;

  if (I->getOpcode() != AArch64::Bcc)
    return nullptr;

  // Since we may modify cmp of this MBB, make sure NZCV does not live out.
  for (auto SuccBB : MBB->successors())
    if (SuccBB->isLiveIn(AArch64::NZCV))
      return nullptr;

  // Now find the instruction controlling the terminator.
  for (MachineBasicBlock::iterator B = MBB->begin(); I != B;) {
    --I;
    assert(!I->isTerminator() && "Spurious terminator");
    // Check if there is any use of NZCV between CMP and Bcc.
    if (I->readsRegister(AArch64::NZCV))
      return nullptr;
    switch (I->getOpcode()) {
    // cmp is an alias for subs with a dead destination register.
    case AArch64::SUBSWri:
    case AArch64::SUBSXri:
    // cmn is an alias for adds with a dead destination register.
    case AArch64::ADDSWri:
    case AArch64::ADDSXri: {
      unsigned ShiftAmt = AArch64_AM::getShiftValue(I->getOperand(3).getImm());
      if (!I->getOperand(2).isImm()) {
        DEBUG(dbgs() << "Immediate of cmp is symbolic, " << *I << '\n');
        return nullptr;
      } else if (I->getOperand(2).getImm() << ShiftAmt >= 0xfff) {
        DEBUG(dbgs() << "Immediate of cmp may be out of range, " << *I << '\n');
        return nullptr;
      } else if (!MRI->use_empty(I->getOperand(0).getReg())) {
        DEBUG(dbgs() << "Destination of cmp is not dead, " << *I << '\n');
        return nullptr;
      }
      return I;
    }
    // Prevent false positive case like:
    // cmp      w19, #0
    // cinc     w0, w19, gt
    // ...
    // fcmp     d8, #0.0
    // b.gt     .LBB0_5
    case AArch64::FCMPDri:
    case AArch64::FCMPSri:
    case AArch64::FCMPESri:
    case AArch64::FCMPEDri:

    case AArch64::SUBSWrr:
    case AArch64::SUBSXrr:
    case AArch64::ADDSWrr:
    case AArch64::ADDSXrr:
    case AArch64::FCMPSrr:
    case AArch64::FCMPDrr:
    case AArch64::FCMPESrr:
    case AArch64::FCMPEDrr:
      // Skip comparison instructions without immediate operands.
      return nullptr;
    }
  }
  DEBUG(dbgs() << "Flags not defined in BB#" << MBB->getNumber() << '\n');
  return nullptr;
}

static void analyzeFrameIndexes(MachineFunction &MF) {
  if (MBDisableStackAdjust) return;

  MachineFrameInfo *MFI = MF.getFrameInfo();
  MBlazeFunctionInfo *MBlazeFI = MF.getInfo<MBlazeFunctionInfo>();
  const MachineRegisterInfo &MRI = MF.getRegInfo();

  MachineRegisterInfo::livein_iterator LII = MRI.livein_begin();
  MachineRegisterInfo::livein_iterator LIE = MRI.livein_end();
  const SmallVector<int, 16> &LiveInFI = MBlazeFI->getLiveIn();
  SmallVector<MachineInstr*, 16> EraseInstr;
  SmallVector<std::pair<int,int64_t>, 16> FrameRelocate;

  MachineBasicBlock *MBB = MF.getBlockNumbered(0);
  MachineBasicBlock::iterator MIB = MBB->begin();
  MachineBasicBlock::iterator MIE = MBB->end();

  int StackAdjust = 0;
  int StackOffset = -28;

  // In this loop we are searching frame indexes that corrospond to incoming
  // arguments that are already in the stack. We look for instruction sequences
  // like the following:
  //    
  //    LWI REG, FI1, 0
  //    ...
  //    SWI REG, FI2, 0
  //
  // As long as there are no defs of REG in the ... part, we can eliminate
  // the SWI instruction because the value has already been stored to the
  // stack by the caller. All we need to do is locate FI at the correct
  // stack location according to the calling convensions.
  //
  // Additionally, if the SWI operation kills the def of REG then we don't
  // need the LWI operation so we can erase it as well.
  for (unsigned i = 0, e = LiveInFI.size(); i < e; ++i) {
    for (MachineBasicBlock::iterator I=MIB; I != MIE; ++I) {
      if (I->getOpcode() != MBlaze::LWI || I->getNumOperands() != 3 ||
          !I->getOperand(1).isFI() || !I->getOperand(0).isReg() ||
          I->getOperand(1).getIndex() != LiveInFI[i]) continue;

      unsigned FIReg = I->getOperand(0).getReg();
      MachineBasicBlock::iterator SI = I;
      for (SI++; SI != MIE; ++SI) {
        if (!SI->getOperand(0).isReg() ||
            !SI->getOperand(1).isFI() ||
            SI->getOpcode() != MBlaze::SWI) continue;

        int FI = SI->getOperand(1).getIndex();
        if (SI->getOperand(0).getReg() != FIReg ||
            MFI->isFixedObjectIndex(FI) ||
            MFI->getObjectSize(FI) != 4) continue;

        if (SI->getOperand(0).isDef()) break;

        if (SI->getOperand(0).isKill()) {
          DEBUG(dbgs() << "LWI for FI#" << I->getOperand(1).getIndex() 
                       << " removed\n");
          EraseInstr.push_back(I);
        }

        EraseInstr.push_back(SI);
        DEBUG(dbgs() << "SWI for FI#" << FI << " removed\n");

        FrameRelocate.push_back(std::make_pair(FI,StackOffset));
        DEBUG(dbgs() << "FI#" << FI << " relocated to " << StackOffset << "\n");

        StackOffset -= 4;
        StackAdjust += 4;
        break;
      }
    }
  }

  // In this loop we are searching for frame indexes that corrospond to
  // incoming arguments that are in registers. We look for instruction
  // sequences like the following:
  //    
  //    ...  SWI REG, FI, 0
  // 
  // As long as the ... part does not define REG and if REG is an incoming
  // parameter register then we know that, according to ABI convensions, the
  // caller has allocated stack space for it already.  Instead of allocating
  // stack space on our frame, we record the correct location in the callers
  // frame.
  for (MachineRegisterInfo::livein_iterator LI = LII; LI != LIE; ++LI) {
    for (MachineBasicBlock::iterator I=MIB; I != MIE; ++I) {
      if (I->definesRegister(LI->first))
        break;

      if (I->getOpcode() != MBlaze::SWI || I->getNumOperands() != 3 ||
          !I->getOperand(1).isFI() || !I->getOperand(0).isReg() ||
          I->getOperand(1).getIndex() < 0) continue;

      if (I->getOperand(0).getReg() == LI->first) {
        int FI = I->getOperand(1).getIndex();
        MBlazeFI->recordLiveIn(FI);

        int FILoc = 0;
        switch (LI->first) {
//.........这里部分代码省略.........

MachineBasicBlock *
MachineBasicBlock::SplitCriticalEdge(MachineBasicBlock *Succ, Pass *P) {
  MachineFunction *MF = getParent();
  DebugLoc dl;  // FIXME: this is nowhere

  // We may need to update this's terminator, but we can't do that if
  // AnalyzeBranch fails. If this uses a jump table, we won't touch it.
  const TargetInstrInfo *TII = MF->getTarget().getInstrInfo();
  MachineBasicBlock *TBB = 0, *FBB = 0;
  SmallVector<MachineOperand, 4> Cond;
  if (TII->AnalyzeBranch(*this, TBB, FBB, Cond))
    return NULL;

  // Avoid bugpoint weirdness: A block may end with a conditional branch but
  // jumps to the same MBB is either case. We have duplicate CFG edges in that
  // case that we can't handle. Since this never happens in properly optimized
  // code, just skip those edges.
  if (TBB && TBB == FBB) {
    DEBUG(dbgs() << "Won't split critical edge after degenerate BB#"
                 << getNumber() << '\n');
    return NULL;
  }

  MachineBasicBlock *NMBB = MF->CreateMachineBasicBlock();
  MF->insert(llvm::next(MachineFunction::iterator(this)), NMBB);
  DEBUG(dbgs() << "Splitting critical edge:"
        " BB#" << getNumber()
        << " -- BB#" << NMBB->getNumber()
        << " -- BB#" << Succ->getNumber() << '\n');

  // On some targets like Mips, branches may kill virtual registers. Make sure
  // that LiveVariables is properly updated after updateTerminator replaces the
  // terminators.
  LiveVariables *LV = P->getAnalysisIfAvailable<LiveVariables>();

  // Collect a list of virtual registers killed by the terminators.
  SmallVector<unsigned, 4> KilledRegs;
  if (LV)
    for (iterator I = getFirstTerminator(), E = end(); I != E; ++I) {
      MachineInstr *MI = I;
      for (MachineInstr::mop_iterator OI = MI->operands_begin(),
           OE = MI->operands_end(); OI != OE; ++OI) {
        if (!OI->isReg() || !OI->isUse() || !OI->isKill() || OI->isUndef())
          continue;
        unsigned Reg = OI->getReg();
        if (TargetRegisterInfo::isVirtualRegister(Reg) &&
            LV->getVarInfo(Reg).removeKill(MI)) {
          KilledRegs.push_back(Reg);
          DEBUG(dbgs() << "Removing terminator kill: " << *MI);
          OI->setIsKill(false);
        }
      }
    }

  ReplaceUsesOfBlockWith(Succ, NMBB);
  updateTerminator();

  // Insert unconditional "jump Succ" instruction in NMBB if necessary.
  NMBB->addSuccessor(Succ);
  if (!NMBB->isLayoutSuccessor(Succ)) {
    Cond.clear();
    MF->getTarget().getInstrInfo()->InsertBranch(*NMBB, Succ, NULL, Cond, dl);
  }

  // Fix PHI nodes in Succ so they refer to NMBB instead of this
  for (MachineBasicBlock::iterator i = Succ->begin(), e = Succ->end();
       i != e && i->isPHI(); ++i)
    for (unsigned ni = 1, ne = i->getNumOperands(); ni != ne; ni += 2)
      if (i->getOperand(ni+1).getMBB() == this)
        i->getOperand(ni+1).setMBB(NMBB);

  // Inherit live-ins from the successor
  for (MachineBasicBlock::livein_iterator I = Succ->livein_begin(),
	 E = Succ->livein_end(); I != E; ++I)
    NMBB->addLiveIn(*I);

  // Update LiveVariables.
  if (LV) {
    // Restore kills of virtual registers that were killed by the terminators.
    while (!KilledRegs.empty()) {
      unsigned Reg = KilledRegs.pop_back_val();
      for (iterator I = end(), E = begin(); I != E;) {
        if (!(--I)->addRegisterKilled(Reg, NULL, /* addIfNotFound= */ false))
          continue;
        LV->getVarInfo(Reg).Kills.push_back(I);
        DEBUG(dbgs() << "Restored terminator kill: " << *I);
        break;
      }
    }
    // Update relevant live-through information.
    LV->addNewBlock(NMBB, this, Succ);
  }

  if (MachineDominatorTree *MDT =
      P->getAnalysisIfAvailable<MachineDominatorTree>()) {
    // Update dominator information.
    MachineDomTreeNode *SucccDTNode = MDT->getNode(Succ);

    bool IsNewIDom = true;
    for (const_pred_iterator PI = Succ->pred_begin(), E = Succ->pred_end();
//.........这里部分代码省略.........

void AArch64FrameLowering::emitPrologue(MachineFunction &MF) const {
  MachineBasicBlock &MBB = MF.front(); // Prologue goes in entry BB.
  MachineBasicBlock::iterator MBBI = MBB.begin();
  const MachineFrameInfo *MFI = MF.getFrameInfo();
  const Function *Fn = MF.getFunction();
  const AArch64RegisterInfo *RegInfo = static_cast<const AArch64RegisterInfo *>(
      MF.getSubtarget().getRegisterInfo());
  const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
  MachineModuleInfo &MMI = MF.getMMI();
  AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
  bool needsFrameMoves = MMI.hasDebugInfo() || Fn->needsUnwindTableEntry();
  bool HasFP = hasFP(MF);
  DebugLoc DL = MBB.findDebugLoc(MBBI);

  int NumBytes = (int)MFI->getStackSize();
  if (!AFI->hasStackFrame()) {
    assert(!HasFP && "unexpected function without stack frame but with FP");

    // All of the stack allocation is for locals.
    AFI->setLocalStackSize(NumBytes);

    // Label used to tie together the PROLOG_LABEL and the MachineMoves.
    MCSymbol *FrameLabel = MMI.getContext().CreateTempSymbol();

    // REDZONE: If the stack size is less than 128 bytes, we don't need
    // to actually allocate.
    if (NumBytes && !canUseRedZone(MF)) {
      emitFrameOffset(MBB, MBBI, DL, AArch64::SP, AArch64::SP, -NumBytes, TII,
                      MachineInstr::FrameSetup);

      // Encode the stack size of the leaf function.
      unsigned CFIIndex = MMI.addFrameInst(
          MCCFIInstruction::createDefCfaOffset(FrameLabel, -NumBytes));
      BuildMI(MBB, MBBI, DL, TII->get(TargetOpcode::CFI_INSTRUCTION))
          .addCFIIndex(CFIIndex);
    } else if (NumBytes) {
      ++NumRedZoneFunctions;
    }

    return;
  }

  // Only set up FP if we actually need to.
  int FPOffset = 0;
  if (HasFP) {
    // First instruction must a) allocate the stack  and b) have an immediate
    // that is a multiple of -2.
    assert((MBBI->getOpcode() == AArch64::STPXpre ||
            MBBI->getOpcode() == AArch64::STPDpre) &&
           MBBI->getOperand(3).getReg() == AArch64::SP &&
           MBBI->getOperand(4).getImm() < 0 &&
           (MBBI->getOperand(4).getImm() & 1) == 0);

    // Frame pointer is fp = sp - 16. Since the  STPXpre subtracts the space
    // required for the callee saved register area we get the frame pointer
    // by addding that offset - 16 = -getImm()*8 - 2*8 = -(getImm() + 2) * 8.
    FPOffset = -(MBBI->getOperand(4).getImm() + 2) * 8;
    assert(FPOffset >= 0 && "Bad Framepointer Offset");
  }

  // Move past the saves of the callee-saved registers.
  while (MBBI->getOpcode() == AArch64::STPXi ||
         MBBI->getOpcode() == AArch64::STPDi ||
         MBBI->getOpcode() == AArch64::STPXpre ||
         MBBI->getOpcode() == AArch64::STPDpre) {
    ++MBBI;
    NumBytes -= 16;
  }
  assert(NumBytes >= 0 && "Negative stack allocation size!?");
  if (HasFP) {
    // Issue    sub fp, sp, FPOffset or
    //          mov fp,sp          when FPOffset is zero.
    // Note: All stores of callee-saved registers are marked as "FrameSetup".
    // This code marks the instruction(s) that set the FP also.
    emitFrameOffset(MBB, MBBI, DL, AArch64::FP, AArch64::SP, FPOffset, TII,
                    MachineInstr::FrameSetup);
  }

  // All of the remaining stack allocations are for locals.
  AFI->setLocalStackSize(NumBytes);

  // Allocate space for the rest of the frame.
  if (NumBytes) {
    // If we're a leaf function, try using the red zone.
    if (!canUseRedZone(MF))
      emitFrameOffset(MBB, MBBI, DL, AArch64::SP, AArch64::SP, -NumBytes, TII,
                      MachineInstr::FrameSetup);
  }

  // If we need a base pointer, set it up here. It's whatever the value of the
  // stack pointer is at this point. Any variable size objects will be allocated
  // after this, so we can still use the base pointer to reference locals.
  //
  // FIXME: Clarify FrameSetup flags here.
  // Note: Use emitFrameOffset() like above for FP if the FrameSetup flag is
  // needed.
  //
  if (RegInfo->hasBasePointer(MF))
    TII->copyPhysReg(MBB, MBBI, DL, AArch64::X19, AArch64::SP, false);

//.........这里部分代码省略.........

MachineInstrBuilder
MipsInstrInfo::genInstrWithNewOpc(unsigned NewOpc,
                                  MachineBasicBlock::iterator I) const {
  MachineInstrBuilder MIB;

  // Certain branches have two forms: e.g beq $1, $zero, dest vs beqz $1, dest
  // Pick the zero form of the branch for readable assembly and for greater
  // branch distance in non-microMIPS mode.
  // Additional MIPSR6 does not permit the use of register $zero for compact
  // branches.
  // FIXME: Certain atomic sequences on mips64 generate 32bit references to
  // Mips::ZERO, which is incorrect. This test should be updated to use
  // Subtarget.getABI().GetZeroReg() when those atomic sequences and others
  // are fixed.
  int ZeroOperandPosition = -1;
  bool BranchWithZeroOperand = false;
  if (I->isBranch() && !I->isPseudo()) {
    auto TRI = I->getParent()->getParent()->getSubtarget().getRegisterInfo();
    ZeroOperandPosition = I->findRegisterUseOperandIdx(Mips::ZERO, false, TRI);
    BranchWithZeroOperand = ZeroOperandPosition != -1;
  }

  if (BranchWithZeroOperand) {
    switch (NewOpc) {
    case Mips::BEQC:
      NewOpc = Mips::BEQZC;
      break;
    case Mips::BNEC:
      NewOpc = Mips::BNEZC;
      break;
    case Mips::BGEC:
      NewOpc = Mips::BGEZC;
      break;
    case Mips::BLTC:
      NewOpc = Mips::BLTZC;
      break;
    case Mips::BEQC64:
      NewOpc = Mips::BEQZC64;
      break;
    case Mips::BNEC64:
      NewOpc = Mips::BNEZC64;
      break;
    }
  }

  MIB = BuildMI(*I->getParent(), I, I->getDebugLoc(), get(NewOpc));

  // For MIPSR6 JI*C requires an immediate 0 as an operand, JIALC(64) an
  // immediate 0 as an operand and requires the removal of it's %RA<imp-def>
  // implicit operand as copying the implicit operations of the instructio we're
  // looking at will give us the correct flags.
  if (NewOpc == Mips::JIC || NewOpc == Mips::JIALC || NewOpc == Mips::JIC64 ||
      NewOpc == Mips::JIALC64) {

    if (NewOpc == Mips::JIALC || NewOpc == Mips::JIALC64)
      MIB->RemoveOperand(0);

    for (unsigned J = 0, E = I->getDesc().getNumOperands(); J < E; ++J) {
      MIB.add(I->getOperand(J));
    }

    MIB.addImm(0);

  } else {
    for (unsigned J = 0, E = I->getDesc().getNumOperands(); J < E; ++J) {
      if (BranchWithZeroOperand && (unsigned)ZeroOperandPosition == J)
        continue;

      MIB.add(I->getOperand(J));
    }
  }

  MIB.copyImplicitOps(*I);

  MIB.setMemRefs(I->memoperands_begin(), I->memoperands_end());
  return MIB;
}

/// ComputeLocalLiveness - Computes liveness of registers within a basic
/// block, setting the killed/dead flags as appropriate.
void RALocal::ComputeLocalLiveness(MachineBasicBlock& MBB) {
  MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
  // Keep track of the most recently seen previous use or def of each reg, 
  // so that we can update them with dead/kill markers.
  DenseMap<unsigned, std::pair<MachineInstr*, unsigned> > LastUseDef;
  for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
       I != E; ++I) {
    if (I->isDebugValue())
      continue;
    
    for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
      MachineOperand &MO = I->getOperand(i);
      // Uses don't trigger any flags, but we need to save
      // them for later.  Also, we have to process these
      // _before_ processing the defs, since an instr
      // uses regs before it defs them.
      if (!MO.isReg() || !MO.getReg() || !MO.isUse())
        continue;
      
      LastUseDef[MO.getReg()] = std::make_pair(I, i);
      
      if (TargetRegisterInfo::isVirtualRegister(MO.getReg())) continue;
      
      const unsigned *Aliases = TRI->getAliasSet(MO.getReg());
      if (Aliases == 0)
        continue;
      
      while (*Aliases) {
        DenseMap<unsigned, std::pair<MachineInstr*, unsigned> >::iterator
          alias = LastUseDef.find(*Aliases);
        
        if (alias != LastUseDef.end() && alias->second.first != I)
          LastUseDef[*Aliases] = std::make_pair(I, i);
        
        ++Aliases;
      }
    }
    
    for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
      MachineOperand &MO = I->getOperand(i);
      // Defs others than 2-addr redefs _do_ trigger flag changes:
      //   - A def followed by a def is dead
      //   - A use followed by a def is a kill
      if (!MO.isReg() || !MO.getReg() || !MO.isDef()) continue;
      
      DenseMap<unsigned, std::pair<MachineInstr*, unsigned> >::iterator
        last = LastUseDef.find(MO.getReg());
      if (last != LastUseDef.end()) {
        // Check if this is a two address instruction.  If so, then
        // the def does not kill the use.
        if (last->second.first == I &&
            I->isRegTiedToUseOperand(i))
          continue;
        
        MachineOperand &lastUD =
                    last->second.first->getOperand(last->second.second);
        if (lastUD.isDef())
          lastUD.setIsDead(true);
        else
          lastUD.setIsKill(true);
      }
      
      LastUseDef[MO.getReg()] = std::make_pair(I, i);
    }
  }
  
  // Live-out (of the function) registers contain return values of the function,
  // so we need to make sure they are alive at return time.
  MachineBasicBlock::iterator Ret = MBB.getFirstTerminator();
  bool BBEndsInReturn = (Ret != MBB.end() && Ret->getDesc().isReturn());

  if (BBEndsInReturn)
    for (MachineRegisterInfo::liveout_iterator
         I = MF->getRegInfo().liveout_begin(),
         E = MF->getRegInfo().liveout_end(); I != E; ++I)
      if (!Ret->readsRegister(*I)) {
        Ret->addOperand(MachineOperand::CreateReg(*I, false, true));
        LastUseDef[*I] = std::make_pair(Ret, Ret->getNumOperands()-1);
      }
  
  // Finally, loop over the final use/def of each reg 
  // in the block and determine if it is dead.
  for (DenseMap<unsigned, std::pair<MachineInstr*, unsigned> >::iterator
       I = LastUseDef.begin(), E = LastUseDef.end(); I != E; ++I) {
    MachineInstr *MI = I->second.first;
    unsigned idx = I->second.second;
    MachineOperand &MO = MI->getOperand(idx);
    
    bool isPhysReg = TargetRegisterInfo::isPhysicalRegister(MO.getReg());
    
    // A crude approximation of "live-out" calculation
    bool usedOutsideBlock = isPhysReg ? false :   
          UsedInMultipleBlocks.test(MO.getReg() -  
                                    TargetRegisterInfo::FirstVirtualRegister);

    // If the machine BB ends in a return instruction, then the value isn't used
    // outside of the BB.
    if (!isPhysReg && (!usedOutsideBlock || BBEndsInReturn)) {
//.........这里部分代码省略.........

// Try to fuse comparison instruction Compare into a later branch.
// Return true on success and if Compare is therefore redundant.
bool SystemZElimCompare::fuseCompareOperations(
    MachineInstr &Compare, SmallVectorImpl<MachineInstr *> &CCUsers) {
  // See whether we have a single branch with which to fuse.
  if (CCUsers.size() != 1)
    return false;
  MachineInstr *Branch = CCUsers[0];
  SystemZII::FusedCompareType Type;
  switch (Branch->getOpcode()) {
  case SystemZ::BRC:
    Type = SystemZII::CompareAndBranch;
    break;
  case SystemZ::CondReturn:
    Type = SystemZII::CompareAndReturn;
    break;
  case SystemZ::CallBCR:
    Type = SystemZII::CompareAndSibcall;
    break;
  case SystemZ::CondTrap:
    Type = SystemZII::CompareAndTrap;
    break;
  default:
    return false;
  }

  // See whether we have a comparison that can be fused.
  unsigned FusedOpcode =
      TII->getFusedCompare(Compare.getOpcode(), Type, &Compare);
  if (!FusedOpcode)
    return false;

  // Make sure that the operands are available at the branch.
  // SrcReg2 is the register if the source operand is a register,
  // 0 if the source operand is immediate, and the base register
  // if the source operand is memory (index is not supported).
  unsigned SrcReg = Compare.getOperand(0).getReg();
  unsigned SrcReg2 =
      Compare.getOperand(1).isReg() ? Compare.getOperand(1).getReg() : 0;
  MachineBasicBlock::iterator MBBI = Compare, MBBE = Branch;
  for (++MBBI; MBBI != MBBE; ++MBBI)
    if (MBBI->modifiesRegister(SrcReg, TRI) ||
        (SrcReg2 && MBBI->modifiesRegister(SrcReg2, TRI)))
      return false;

  // Read the branch mask, target (if applicable), regmask (if applicable).
  MachineOperand CCMask(MBBI->getOperand(1));
  assert((CCMask.getImm() & ~SystemZ::CCMASK_ICMP) == 0 &&
         "Invalid condition-code mask for integer comparison");
  // This is only valid for CompareAndBranch.
  MachineOperand Target(MBBI->getOperand(
    Type == SystemZII::CompareAndBranch ? 2 : 0));
  const uint32_t *RegMask;
  if (Type == SystemZII::CompareAndSibcall)
    RegMask = MBBI->getOperand(2).getRegMask();

  // Clear out all current operands.
  int CCUse = MBBI->findRegisterUseOperandIdx(SystemZ::CC, false, TRI);
  assert(CCUse >= 0 && "BRC/BCR must use CC");
  Branch->RemoveOperand(CCUse);
  // Remove target (branch) or regmask (sibcall).
  if (Type == SystemZII::CompareAndBranch ||
      Type == SystemZII::CompareAndSibcall)
    Branch->RemoveOperand(2);
  Branch->RemoveOperand(1);
  Branch->RemoveOperand(0);

  // Rebuild Branch as a fused compare and branch.
  // SrcNOps is the number of MI operands of the compare instruction
  // that we need to copy over.
  unsigned SrcNOps = 2;
  if (FusedOpcode == SystemZ::CLT || FusedOpcode == SystemZ::CLGT)
    SrcNOps = 3;
  Branch->setDesc(TII->get(FusedOpcode));
  MachineInstrBuilder MIB(*Branch->getParent()->getParent(), Branch);
  for (unsigned I = 0; I < SrcNOps; I++)
    MIB.addOperand(Compare.getOperand(I));
  MIB.addOperand(CCMask);

  if (Type == SystemZII::CompareAndBranch) {
    // Only conditional branches define CC, as they may be converted back
    // to a non-fused branch because of a long displacement.  Conditional
    // returns don't have that problem.
    MIB.addOperand(Target)
       .addReg(SystemZ::CC, RegState::ImplicitDefine | RegState::Dead);
  }

  if (Type == SystemZII::CompareAndSibcall)
    MIB.addRegMask(RegMask);

  // Clear any intervening kills of SrcReg and SrcReg2.
  MBBI = Compare;
  for (++MBBI; MBBI != MBBE; ++MBBI) {
    MBBI->clearRegisterKills(SrcReg, TRI);
    if (SrcReg2)
      MBBI->clearRegisterKills(SrcReg2, TRI);
  }
  FusedComparisons += 1;
  return true;
}

/// If \p MBBI is a pseudo instruction, this method expands
/// it to the corresponding (sequence of) actual instruction(s).
/// \returns true if \p MBBI has been expanded.
bool X86ExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
                               MachineBasicBlock::iterator MBBI) {
  MachineInstr &MI = *MBBI;
  unsigned Opcode = MI.getOpcode();
  DebugLoc DL = MBBI->getDebugLoc();
  switch (Opcode) {
  default:
    return false;
  case X86::TCRETURNdi:
  case X86::TCRETURNri:
  case X86::TCRETURNmi:
  case X86::TCRETURNdi64:
  case X86::TCRETURNri64:
  case X86::TCRETURNmi64: {
    bool isMem = Opcode == X86::TCRETURNmi || Opcode == X86::TCRETURNmi64;
    MachineOperand &JumpTarget = MBBI->getOperand(0);
    MachineOperand &StackAdjust = MBBI->getOperand(isMem ? 5 : 1);
    assert(StackAdjust.isImm() && "Expecting immediate value.");

    // Adjust stack pointer.
    int StackAdj = StackAdjust.getImm();

    if (StackAdj) {
      // Check for possible merge with preceding ADD instruction.
      StackAdj += X86FL->mergeSPUpdates(MBB, MBBI, true);
      X86FL->emitSPUpdate(MBB, MBBI, StackAdj, /*InEpilogue=*/true);
    }

    // Jump to label or value in register.
    bool IsWin64 = STI->isTargetWin64();
    if (Opcode == X86::TCRETURNdi || Opcode == X86::TCRETURNdi64) {
      unsigned Op = (Opcode == X86::TCRETURNdi)
                        ? X86::TAILJMPd
                        : (IsWin64 ? X86::TAILJMPd64_REX : X86::TAILJMPd64);
      MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(Op));
      if (JumpTarget.isGlobal())
        MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
                             JumpTarget.getTargetFlags());
      else {
        assert(JumpTarget.isSymbol());
        MIB.addExternalSymbol(JumpTarget.getSymbolName(),
                              JumpTarget.getTargetFlags());
      }
    } else if (Opcode == X86::TCRETURNmi || Opcode == X86::TCRETURNmi64) {
      unsigned Op = (Opcode == X86::TCRETURNmi)
                        ? X86::TAILJMPm
                        : (IsWin64 ? X86::TAILJMPm64_REX : X86::TAILJMPm64);
      MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(Op));
      for (unsigned i = 0; i != 5; ++i)
        MIB.addOperand(MBBI->getOperand(i));
    } else if (Opcode == X86::TCRETURNri64) {
      BuildMI(MBB, MBBI, DL,
              TII->get(IsWin64 ? X86::TAILJMPr64_REX : X86::TAILJMPr64))
          .addReg(JumpTarget.getReg(), RegState::Kill);
    } else {
      BuildMI(MBB, MBBI, DL, TII->get(X86::TAILJMPr))
          .addReg(JumpTarget.getReg(), RegState::Kill);
    }

    MachineInstr *NewMI = std::prev(MBBI);
    NewMI->copyImplicitOps(*MBBI->getParent()->getParent(), *MBBI);

    // Delete the pseudo instruction TCRETURN.
    MBB.erase(MBBI);

    return true;
  }
  case X86::EH_RETURN:
  case X86::EH_RETURN64: {
    MachineOperand &DestAddr = MBBI->getOperand(0);
    assert(DestAddr.isReg() && "Offset should be in register!");
    const bool Uses64BitFramePtr =
        STI->isTarget64BitLP64() || STI->isTargetNaCl64();
    unsigned StackPtr = TRI->getStackRegister();
    BuildMI(MBB, MBBI, DL,
            TII->get(Uses64BitFramePtr ? X86::MOV64rr : X86::MOV32rr), StackPtr)
        .addReg(DestAddr.getReg());
    // The EH_RETURN pseudo is really removed during the MC Lowering.
    return true;
  }
  case X86::IRET: {
    // Adjust stack to erase error code
    int64_t StackAdj = MBBI->getOperand(0).getImm();
    X86FL->emitSPUpdate(MBB, MBBI, StackAdj, true);
    // Replace pseudo with machine iret
    BuildMI(MBB, MBBI, DL,
            TII->get(STI->is64Bit() ? X86::IRET64 : X86::IRET32));
    MBB.erase(MBBI);
    return true;
  }
  case X86::RET: {
    // Adjust stack to erase error code
    int64_t StackAdj = MBBI->getOperand(0).getImm();
    MachineInstrBuilder MIB;
    if (StackAdj == 0) {
      MIB = BuildMI(MBB, MBBI, DL,
                    TII->get(STI->is64Bit() ? X86::RETQ : X86::RETL));
//.........这里部分代码省略.........

bool PHIElimination::SplitPHIEdges(MachineFunction &MF,
                                   MachineBasicBlock &MBB,
                                   MachineLoopInfo *MLI) {
    if (MBB.empty() || !MBB.front().isPHI() || MBB.isLandingPad())
        return false;   // Quick exit for basic blocks without PHIs.

    const MachineLoop *CurLoop = MLI ? MLI->getLoopFor(&MBB) : 0;
    bool IsLoopHeader = CurLoop && &MBB == CurLoop->getHeader();

    bool Changed = false;
    for (MachineBasicBlock::iterator BBI = MBB.begin(), BBE = MBB.end();
            BBI != BBE && BBI->isPHI(); ++BBI) {
        for (unsigned i = 1, e = BBI->getNumOperands(); i != e; i += 2) {
            unsigned Reg = BBI->getOperand(i).getReg();
            MachineBasicBlock *PreMBB = BBI->getOperand(i+1).getMBB();
            // Is there a critical edge from PreMBB to MBB?
            if (PreMBB->succ_size() == 1)
                continue;

            // Avoid splitting backedges of loops. It would introduce small
            // out-of-line blocks into the loop which is very bad for code placement.
            if (PreMBB == &MBB && !SplitAllCriticalEdges)
                continue;
            const MachineLoop *PreLoop = MLI ? MLI->getLoopFor(PreMBB) : 0;
            if (IsLoopHeader && PreLoop == CurLoop && !SplitAllCriticalEdges)
                continue;

            // LV doesn't consider a phi use live-out, so isLiveOut only returns true
            // when the source register is live-out for some other reason than a phi
            // use. That means the copy we will insert in PreMBB won't be a kill, and
            // there is a risk it may not be coalesced away.
            //
            // If the copy would be a kill, there is no need to split the edge.
            if (!isLiveOutPastPHIs(Reg, PreMBB) && !SplitAllCriticalEdges)
                continue;

            DEBUG(dbgs() << PrintReg(Reg) << " live-out before critical edge BB#"
                  << PreMBB->getNumber() << " -> BB#" << MBB.getNumber()
                  << ": " << *BBI);

            // If Reg is not live-in to MBB, it means it must be live-in to some
            // other PreMBB successor, and we can avoid the interference by splitting
            // the edge.
            //
            // If Reg *is* live-in to MBB, the interference is inevitable and a copy
            // is likely to be left after coalescing. If we are looking at a loop
            // exiting edge, split it so we won't insert code in the loop, otherwise
            // don't bother.
            bool ShouldSplit = !isLiveIn(Reg, &MBB) || SplitAllCriticalEdges;

            // Check for a loop exiting edge.
            if (!ShouldSplit && CurLoop != PreLoop) {
                DEBUG({
                    dbgs() << "Split wouldn't help, maybe avoid loop copies?\n";
                    if (PreLoop) dbgs() << "PreLoop: " << *PreLoop;
                    if (CurLoop) dbgs() << "CurLoop: " << *CurLoop;
                });
                // This edge could be entering a loop, exiting a loop, or it could be
                // both: Jumping directly form one loop to the header of a sibling
                // loop.
                // Split unless this edge is entering CurLoop from an outer loop.
                ShouldSplit = PreLoop && !PreLoop->contains(CurLoop);
            }
            if (!ShouldSplit)
                continue;
            if (!PreMBB->SplitCriticalEdge(&MBB, this)) {
                DEBUG(dbgs() << "Failed to split ciritcal edge.\n");
                continue;
            }
            Changed = true;
            ++NumCriticalEdgesSplit;
        }

void AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
                                        MachineBasicBlock &MBB) const {
  MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>();
  const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo();
  const TargetInstrInfo *TII = Subtarget.getInstrInfo();
  DebugLoc DL;
  bool IsTailCallReturn = false;
  if (MBB.end() != MBBI) {
    DL = MBBI->getDebugLoc();
    unsigned RetOpcode = MBBI->getOpcode();
    IsTailCallReturn = RetOpcode == AArch64::TCRETURNdi ||
      RetOpcode == AArch64::TCRETURNri;
  }
  int NumBytes = MFI->getStackSize();
  const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();

  // All calls are tail calls in GHC calling conv, and functions have no
  // prologue/epilogue.
  if (MF.getFunction()->getCallingConv() == CallingConv::GHC)
    return;

  // Initial and residual are named for consistency with the prologue. Note that
  // in the epilogue, the residual adjustment is executed first.
  uint64_t ArgumentPopSize = 0;
  if (IsTailCallReturn) {
    MachineOperand &StackAdjust = MBBI->getOperand(1);

    // For a tail-call in a callee-pops-arguments environment, some or all of
    // the stack may actually be in use for the call's arguments, this is
    // calculated during LowerCall and consumed here...
    ArgumentPopSize = StackAdjust.getImm();
  } else {
    // ... otherwise the amount to pop is *all* of the argument space,
    // conveniently stored in the MachineFunctionInfo by
    // LowerFormalArguments. This will, of course, be zero for the C calling
    // convention.
    ArgumentPopSize = AFI->getArgumentStackToRestore();
  }

  // The stack frame should be like below,
  //
  //      ----------------------                     ---
  //      |                    |                      |
  //      | BytesInStackArgArea|              CalleeArgStackSize
  //      | (NumReusableBytes) |                (of tail call)
  //      |                    |                     ---
  //      |                    |                      |
  //      ---------------------|        ---           |
  //      |                    |         |            |
  //      |   CalleeSavedReg   |         |            |
  //      | (NumRestores * 8)  |         |            |
  //      |                    |         |            |
  //      ---------------------|         |         NumBytes
  //      |                    |     StackSize  (StackAdjustUp)
  //      |   LocalStackSize   |         |            |
  //      | (covering callee   |         |            |
  //      |       args)        |         |            |
  //      |                    |         |            |
  //      ----------------------        ---          ---
  //
  // So NumBytes = StackSize + BytesInStackArgArea - CalleeArgStackSize
  //             = StackSize + ArgumentPopSize
  //
  // AArch64TargetLowering::LowerCall figures out ArgumentPopSize and keeps
  // it as the 2nd argument of AArch64ISD::TC_RETURN.
  NumBytes += ArgumentPopSize;

  unsigned NumRestores = 0;
  // Move past the restores of the callee-saved registers.
  MachineBasicBlock::iterator LastPopI = MBB.getFirstTerminator();
  const MCPhysReg *CSRegs = RegInfo->getCalleeSavedRegs(&MF);
  MachineBasicBlock::iterator Begin = MBB.begin();
  while (LastPopI != Begin) {
    --LastPopI;
    unsigned Restores = getNumCSRestores(*LastPopI, CSRegs);
    NumRestores += Restores;
    if (Restores == 0) {
      ++LastPopI;
      break;
    }
  }
  NumBytes -= NumRestores * 8;
  assert(NumBytes >= 0 && "Negative stack allocation size!?");

  if (!hasFP(MF)) {
    // If this was a redzone leaf function, we don't need to restore the
    // stack pointer.
    if (!canUseRedZone(MF))
      emitFrameOffset(MBB, LastPopI, DL, AArch64::SP, AArch64::SP, NumBytes,
                      TII);
    return;
  }

  // Restore the original stack pointer.
  // FIXME: Rather than doing the math here, we should instead just use
  // non-post-indexed loads for the restores if we aren't actually going to
  // be able to save any instructions.
  if (NumBytes || MFI->hasVarSizedObjects())
//.........这里部分代码省略.........

/// reMaterializeFor - Attempt to rematerialize before MI instead of reloading.
bool InlineSpiller::reMaterializeFor(LiveInterval &VirtReg,
                                     MachineBasicBlock::iterator MI) {
  SlotIndex UseIdx = LIS.getInstructionIndex(MI).getRegSlot(true);
  VNInfo *ParentVNI = VirtReg.getVNInfoAt(UseIdx.getBaseIndex());

  if (!ParentVNI) {
    DEBUG(dbgs() << "\tadding <undef> flags: ");
    for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
      MachineOperand &MO = MI->getOperand(i);
      if (MO.isReg() && MO.isUse() && MO.getReg() == VirtReg.reg)
        MO.setIsUndef();
    }
    DEBUG(dbgs() << UseIdx << '\t' << *MI);
    return true;
  }

  if (SnippetCopies.count(MI))
    return false;

  // Use an OrigVNI from traceSiblingValue when ParentVNI is a sibling copy.
  LiveRangeEdit::Remat RM(ParentVNI);
  SibValueMap::const_iterator SibI = SibValues.find(ParentVNI);
  if (SibI != SibValues.end())
    RM.OrigMI = SibI->second.DefMI;
  if (!Edit->canRematerializeAt(RM, UseIdx, false)) {
    markValueUsed(&VirtReg, ParentVNI);
    DEBUG(dbgs() << "\tcannot remat for " << UseIdx << '\t' << *MI);
    return false;
  }

  // If the instruction also writes VirtReg.reg, it had better not require the
  // same register for uses and defs.
  SmallVector<std::pair<MachineInstr*, unsigned>, 8> Ops;
  MIBundleOperands::VirtRegInfo RI =
    MIBundleOperands(MI).analyzeVirtReg(VirtReg.reg, &Ops);
  if (RI.Tied) {
    markValueUsed(&VirtReg, ParentVNI);
    DEBUG(dbgs() << "\tcannot remat tied reg: " << UseIdx << '\t' << *MI);
    return false;
  }

  // Before rematerializing into a register for a single instruction, try to
  // fold a load into the instruction. That avoids allocating a new register.
  if (RM.OrigMI->canFoldAsLoad() &&
      foldMemoryOperand(Ops, RM.OrigMI)) {
    Edit->markRematerialized(RM.ParentVNI);
    ++NumFoldedLoads;
    return true;
  }

  // Alocate a new register for the remat.
  unsigned NewVReg = Edit->createFrom(Original);

  // Finally we can rematerialize OrigMI before MI.
  SlotIndex DefIdx = Edit->rematerializeAt(*MI->getParent(), MI, NewVReg, RM,
                                           TRI);
  (void)DefIdx;
  DEBUG(dbgs() << "\tremat:  " << DefIdx << '\t'
               << *LIS.getInstructionFromIndex(DefIdx));

  // Replace operands
  for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
    MachineOperand &MO = MI->getOperand(Ops[i].second);
    if (MO.isReg() && MO.isUse() && MO.getReg() == VirtReg.reg) {
      MO.setReg(NewVReg);
      MO.setIsKill();
    }
  }
  DEBUG(dbgs() << "\t        " << UseIdx << '\t' << *MI << '\n');

  ++NumRemats;
  return true;
}