LiveIntervalAnalysis.cpp

//===-- LiveIntervalAnalysis.cpp - Live Interval Analysis -----------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the LiveInterval analysis pass which is used
// by the Linear Scan Register allocator. This pass linearizes the
// basic blocks of the function in DFS order and uses the
// LiveVariables pass to conservatively compute live intervals for
// each virtual and physical register.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "liveintervals"
#include "llvm/CodeGen/LiveIntervalAnalysis.h"
#include "VirtRegMap.h"
#include "llvm/Value.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/LiveVariables.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include <algorithm>
#include <limits>
#include <cmath>
using namespace llvm;

// Hidden options for help debugging.
static cl::opt<bool> DisableReMat("disable-rematerialization", 
                                  cl::init(false), cl::Hidden);

static cl::opt<bool> EnableAggressiveRemat("aggressive-remat", cl::Hidden);

static cl::opt<bool> EnableFastSpilling("fast-spill",
                                        cl::init(false), cl::Hidden);

static cl::opt<bool> EarlyCoalescing("early-coalescing", cl::init(false));

static cl::opt<int> CoalescingLimit("early-coalescing-limit",
                                    cl::init(-1), cl::Hidden);

STATISTIC(numIntervals , "Number of original intervals");
STATISTIC(numFolds     , "Number of loads/stores folded into instructions");
STATISTIC(numSplits    , "Number of intervals split");
STATISTIC(numCoalescing, "Number of early coalescing performed");

char LiveIntervals::ID = 0;
static RegisterPass<LiveIntervals> X("liveintervals", "Live Interval Analysis");

void LiveIntervals::getAnalysisUsage(AnalysisUsage &AU) const {
  AU.setPreservesCFG();
  AU.addRequired<AliasAnalysis>();
  AU.addPreserved<AliasAnalysis>();
  AU.addPreserved<LiveVariables>();
  AU.addRequired<LiveVariables>();
  AU.addPreservedID(MachineLoopInfoID);
  AU.addPreservedID(MachineDominatorsID);
  
  if (!StrongPHIElim) {
    AU.addPreservedID(PHIEliminationID);
    AU.addRequiredID(PHIEliminationID);
  }
  
  AU.addRequiredID(TwoAddressInstructionPassID);
  MachineFunctionPass::getAnalysisUsage(AU);
}

void LiveIntervals::releaseMemory() {
  // Free the live intervals themselves.
  for (DenseMap<unsigned, LiveInterval*>::iterator I = r2iMap_.begin(),
       E = r2iMap_.end(); I != E; ++I)
    delete I->second;
  
  MBB2IdxMap.clear();
  Idx2MBBMap.clear();
  mi2iMap_.clear();
  i2miMap_.clear();
  r2iMap_.clear();
  terminatorGaps.clear();
  phiJoinCopies.clear();

  // Release VNInfo memroy regions after all VNInfo objects are dtor'd.
  VNInfoAllocator.Reset();
  while (!CloneMIs.empty()) {
    MachineInstr *MI = CloneMIs.back();
    CloneMIs.pop_back();
    mf_->DeleteMachineInstr(MI);
  }
}

static bool CanTurnIntoImplicitDef(MachineInstr *MI, unsigned Reg,
                                   unsigned OpIdx, const TargetInstrInfo *tii_){
  unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
  if (tii_->isMoveInstr(*MI, SrcReg, DstReg, SrcSubReg, DstSubReg) &&
      Reg == SrcReg)
    return true;

  if (OpIdx == 2 && MI->getOpcode() == TargetInstrInfo::SUBREG_TO_REG)
    return true;
  if (OpIdx == 1 && MI->getOpcode() == TargetInstrInfo::EXTRACT_SUBREG)
    return true;
  return false;
}

/// processImplicitDefs - Process IMPLICIT_DEF instructions and make sure
/// there is one implicit_def for each use. Add isUndef marker to
/// implicit_def defs and their uses.
void LiveIntervals::processImplicitDefs() {
  SmallSet<unsigned, 8> ImpDefRegs;
  SmallVector<MachineInstr*, 8> ImpDefMIs;
  MachineBasicBlock *Entry = mf_->begin();
  SmallPtrSet<MachineBasicBlock*,16> Visited;
  for (df_ext_iterator<MachineBasicBlock*, SmallPtrSet<MachineBasicBlock*,16> >
         DFI = df_ext_begin(Entry, Visited), E = df_ext_end(Entry, Visited);
       DFI != E; ++DFI) {
    MachineBasicBlock *MBB = *DFI;
    for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end();
         I != E; ) {
      MachineInstr *MI = &*I;
      ++I;
      if (MI->getOpcode() == TargetInstrInfo::IMPLICIT_DEF) {
        unsigned Reg = MI->getOperand(0).getReg();
        ImpDefRegs.insert(Reg);
        if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
          for (const unsigned *SS = tri_->getSubRegisters(Reg); *SS; ++SS)
            ImpDefRegs.insert(*SS);
        }
        ImpDefMIs.push_back(MI);
        continue;
      }

      if (MI->getOpcode() == TargetInstrInfo::INSERT_SUBREG) {
        MachineOperand &MO = MI->getOperand(2);
        if (ImpDefRegs.count(MO.getReg())) {
          // %reg1032<def> = INSERT_SUBREG %reg1032, undef, 2
          // This is an identity copy, eliminate it now.
          if (MO.isKill()) {
            LiveVariables::VarInfo& vi = lv_->getVarInfo(MO.getReg());
            vi.removeKill(MI);
          }
          MI->eraseFromParent();
          continue;
        }
      }

      bool ChangedToImpDef = false;
      for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
        MachineOperand& MO = MI->getOperand(i);
        if (!MO.isReg() || !MO.isUse() || MO.isUndef())
          continue;
        unsigned Reg = MO.getReg();
        if (!Reg)
          continue;
        if (!ImpDefRegs.count(Reg))
          continue;
        // Use is a copy, just turn it into an implicit_def.
        if (CanTurnIntoImplicitDef(MI, Reg, i, tii_)) {
          bool isKill = MO.isKill();
          MI->setDesc(tii_->get(TargetInstrInfo::IMPLICIT_DEF));
          for (int j = MI->getNumOperands() - 1, ee = 0; j > ee; --j)
            MI->RemoveOperand(j);
          if (isKill) {
            ImpDefRegs.erase(Reg);
            LiveVariables::VarInfo& vi = lv_->getVarInfo(Reg);
            vi.removeKill(MI);
          }
          ChangedToImpDef = true;
          break;
        }

        MO.setIsUndef();
        if (MO.isKill() || MI->isRegTiedToDefOperand(i)) {
          // Make sure other uses of 
          for (unsigned j = i+1; j != e; ++j) {
            MachineOperand &MOJ = MI->getOperand(j);
            if (MOJ.isReg() && MOJ.isUse() && MOJ.getReg() == Reg)
              MOJ.setIsUndef();
          }
          ImpDefRegs.erase(Reg);
        }
      }

      if (ChangedToImpDef) {
        // Backtrack to process this new implicit_def.
        --I;
      } else {
        for (unsigned i = 0; i != MI->getNumOperands(); ++i) {
          MachineOperand& MO = MI->getOperand(i);
          if (!MO.isReg() || !MO.isDef())
            continue;
          ImpDefRegs.erase(MO.getReg());
        }
      }
    }

    // Any outstanding liveout implicit_def's?
    for (unsigned i = 0, e = ImpDefMIs.size(); i != e; ++i) {
      MachineInstr *MI = ImpDefMIs[i];
      unsigned Reg = MI->getOperand(0).getReg();
      if (TargetRegisterInfo::isPhysicalRegister(Reg) ||
          !ImpDefRegs.count(Reg)) {
        // Delete all "local" implicit_def's. That include those which define
        // physical registers since they cannot be liveout.
        MI->eraseFromParent();
        continue;
      }

      // If there are multiple defs of the same register and at least one
      // is not an implicit_def, do not insert implicit_def's before the
      // uses.
      bool Skip = false;
      for (MachineRegisterInfo::def_iterator DI = mri_->def_begin(Reg),
             DE = mri_->def_end(); DI != DE; ++DI) {
        if (DI->getOpcode() != TargetInstrInfo::IMPLICIT_DEF) {
          Skip = true;
          break;
        }
      }
      if (Skip)
        continue;

      // The only implicit_def which we want to keep are those that are live
      // out of its block.
      MI->eraseFromParent();

      for (MachineRegisterInfo::use_iterator UI = mri_->use_begin(Reg),
             UE = mri_->use_end(); UI != UE; ) {
        MachineOperand &RMO = UI.getOperand();
        MachineInstr *RMI = &*UI;
        ++UI;
        MachineBasicBlock *RMBB = RMI->getParent();
        if (RMBB == MBB)
          continue;

        // Turn a copy use into an implicit_def.
        unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
        if (tii_->isMoveInstr(*RMI, SrcReg, DstReg, SrcSubReg, DstSubReg) &&
            Reg == SrcReg) {
          RMI->setDesc(tii_->get(TargetInstrInfo::IMPLICIT_DEF));
          for (int j = RMI->getNumOperands() - 1, ee = 0; j > ee; --j)
            RMI->RemoveOperand(j);
          continue;
        }

        const TargetRegisterClass* RC = mri_->getRegClass(Reg);
        unsigned NewVReg = mri_->createVirtualRegister(RC);
        RMO.setReg(NewVReg);
        RMO.setIsUndef();
        RMO.setIsKill();
      }
    }
    ImpDefRegs.clear();
    ImpDefMIs.clear();
  }
}


void LiveIntervals::computeNumbering() {
  Index2MiMap OldI2MI = i2miMap_;
  std::vector<IdxMBBPair> OldI2MBB = Idx2MBBMap;
  
  Idx2MBBMap.clear();
  MBB2IdxMap.clear();
  mi2iMap_.clear();
  i2miMap_.clear();
  terminatorGaps.clear();
  phiJoinCopies.clear();
  
  FunctionSize = 0;
  
  // Number MachineInstrs and MachineBasicBlocks.
  // Initialize MBB indexes to a sentinal.
  MBB2IdxMap.resize(mf_->getNumBlockIDs(),
                    std::make_pair(LiveIndex(),MachineInstrIndex()));
  
  LiveIndex MIIndex;
  for (MachineFunction::iterator MBB = mf_->begin(), E = mf_->end();
       MBB != E; ++MBB) {
    LiveIndex StartIdx = MIIndex;

    // Insert an empty slot at the beginning of each block.
    MIIndex = getNextIndex(MIIndex);
    i2miMap_.push_back(0);

    for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end();
         I != E; ++I) {
      
      if (I == MBB->getFirstTerminator()) {
        // Leave a gap for before terminators, this is where we will point
        // PHI kills.
        LiveIndex tGap(true, MIIndex);
        bool inserted =
          terminatorGaps.insert(std::make_pair(&*MBB, tGap)).second;
        assert(inserted && 
               "Multiple 'first' terminators encountered during numbering.");
        inserted = inserted; // Avoid compiler warning if assertions turned off.
        i2miMap_.push_back(0);

        MIIndex = getNextIndex(MIIndex);
      }

      bool inserted = mi2iMap_.insert(std::make_pair(I, MIIndex)).second;
      assert(inserted && "multiple MachineInstr -> index mappings");
      inserted = true;
      i2miMap_.push_back(I);
      MIIndex = getNextIndex(MIIndex);
      FunctionSize++;
      
      // Insert max(1, numdefs) empty slots after every instruction.
      unsigned Slots = I->getDesc().getNumDefs();
      if (Slots == 0)
        Slots = 1;
      while (Slots--) {
        MIIndex = getNextIndex(MIIndex);
        i2miMap_.push_back(0);
      }

    }
  
    if (MBB->getFirstTerminator() == MBB->end()) {
      // Leave a gap for before terminators, this is where we will point
      // PHI kills.
      LiveIndex tGap(true, MIIndex);
      bool inserted =
        terminatorGaps.insert(std::make_pair(&*MBB, tGap)).second;
      assert(inserted && 
             "Multiple 'first' terminators encountered during numbering.");
      inserted = inserted; // Avoid compiler warning if assertions turned off.
      i2miMap_.push_back(0);
 
      MIIndex = getNextIndex(MIIndex);
    }
    
    // Set the MBB2IdxMap entry for this MBB.
    MBB2IdxMap[MBB->getNumber()] = std::make_pair(StartIdx, getPrevSlot(MIIndex));
    Idx2MBBMap.push_back(std::make_pair(StartIdx, MBB));
  }

  std::sort(Idx2MBBMap.begin(), Idx2MBBMap.end(), Idx2MBBCompare());
  
  if (!OldI2MI.empty())
    for (iterator OI = begin(), OE = end(); OI != OE; ++OI) {
      for (LiveInterval::iterator LI = OI->second->begin(),
           LE = OI->second->end(); LI != LE; ++LI) {
        
        // Remap the start index of the live range to the corresponding new
        // number, or our best guess at what it _should_ correspond to if the
        // original instruction has been erased.  This is either the following
        // instruction or its predecessor.
        unsigned index = LI->start.getVecIndex();
        LiveIndex::Slot offset = LI->start.getSlot();
        if (LI->start.isLoad()) {
          std::vector<IdxMBBPair>::const_iterator I =
                  std::lower_bound(OldI2MBB.begin(), OldI2MBB.end(), LI->start);
          // Take the pair containing the index
          std::vector<IdxMBBPair>::const_iterator J =
                    (I == OldI2MBB.end() && OldI2MBB.size()>0) ? (I-1): I;
          
          LI->start = getMBBStartIdx(J->second);
        } else {
          LI->start = LiveIndex(
            LiveIndex(mi2iMap_[OldI2MI[index]]), 
                              (LiveIndex::Slot)offset);
        }
        
        // Remap the ending index in the same way that we remapped the start,
        // except for the final step where we always map to the immediately
        // following instruction.
        index = (getPrevSlot(LI->end)).getVecIndex();
        offset  = LI->end.getSlot();
        if (LI->end.isLoad()) {
          // VReg dies at end of block.
          std::vector<IdxMBBPair>::const_iterator I =
                  std::lower_bound(OldI2MBB.begin(), OldI2MBB.end(), LI->end);
          --I;
          
          LI->end = getNextSlot(getMBBEndIdx(I->second));
        } else {
          unsigned idx = index;
          while (index < OldI2MI.size() && !OldI2MI[index]) ++index;
          
          if (index != OldI2MI.size())
            LI->end =
              LiveIndex(mi2iMap_[OldI2MI[index]],
                (idx == index ? offset : LiveIndex::LOAD));
          else
            LI->end =
              LiveIndex(MachineInstrIndex::NUM * i2miMap_.size());
        }
      }
      
      for (LiveInterval::vni_iterator VNI = OI->second->vni_begin(),
           VNE = OI->second->vni_end(); VNI != VNE; ++VNI) { 
        VNInfo* vni = *VNI;
        
        // Remap the VNInfo def index, which works the same as the
        // start indices above. VN's with special sentinel defs
        // don't need to be remapped.
        if (vni->isDefAccurate() && !vni->isUnused()) {
          unsigned index = vni->def.getVecIndex();
          LiveIndex::Slot offset = vni->def.getSlot();
          if (vni->def.isLoad()) {
            std::vector<IdxMBBPair>::const_iterator I =
                  std::lower_bound(OldI2MBB.begin(), OldI2MBB.end(), vni->def);
            // Take the pair containing the index
            std::vector<IdxMBBPair>::const_iterator J =
                    (I == OldI2MBB.end() && OldI2MBB.size()>0) ? (I-1): I;
          
            vni->def = getMBBStartIdx(J->second);
          } else {
            vni->def = LiveIndex(mi2iMap_[OldI2MI[index]], offset);
          }
        }
        
        // Remap the VNInfo kill indices, which works the same as
        // the end indices above.
        for (size_t i = 0; i < vni->kills.size(); ++i) {
          unsigned index = getPrevSlot(vni->kills[i]).getVecIndex();
          LiveIndex::Slot offset = vni->kills[i].getSlot();

          if (vni->kills[i].isLoad()) {
            assert("Value killed at a load slot.");
            /*std::vector<IdxMBBPair>::const_iterator I =
             std::lower_bound(OldI2MBB.begin(), OldI2MBB.end(), vni->kills[i]);
            --I;

            vni->kills[i] = getMBBEndIdx(I->second);*/
          } else {
            if (vni->kills[i].isPHIIndex()) {
              std::vector<IdxMBBPair>::const_iterator I =
                std::lower_bound(OldI2MBB.begin(), OldI2MBB.end(), vni->kills[i]);
              --I;
              vni->kills[i] = terminatorGaps[I->second];  
            } else {
              assert(OldI2MI[index] != 0 &&
                     "Kill refers to instruction not present in index maps.");
              vni->kills[i] = LiveIndex(mi2iMap_[OldI2MI[index]], offset);
            }
           
            /*
            unsigned idx = index;
            while (index < OldI2MI.size() && !OldI2MI[index]) ++index;
            
            if (index != OldI2MI.size())
              vni->kills[i] = mi2iMap_[OldI2MI[index]] + 
                              (idx == index ? offset : 0);
            else
              vni->kills[i] = InstrSlots::NUM * i2miMap_.size();
            */
          }
        }
      }
    }
}

void LiveIntervals::scaleNumbering(int factor) {
  // Need to
  //  * scale MBB begin and end points
  //  * scale all ranges.
  //  * Update VNI structures.
  //  * Scale instruction numberings 

  // Scale the MBB indices.
  Idx2MBBMap.clear();
  for (MachineFunction::iterator MBB = mf_->begin(), MBBE = mf_->end();
       MBB != MBBE; ++MBB) {
    std::pair<LiveIndex, MachineInstrIndex> &mbbIndices = MBB2IdxMap[MBB->getNumber()];
    mbbIndices.first = mbbIndices.first.scale(factor);
    mbbIndices.second = mbbIndices.second.scale(factor);
    Idx2MBBMap.push_back(std::make_pair(mbbIndices.first, MBB)); 
  }
  std::sort(Idx2MBBMap.begin(), Idx2MBBMap.end(), Idx2MBBCompare());

  // Scale terminator gaps.
  for (DenseMap<MachineBasicBlock*, LiveIndex>::iterator
       TGI = terminatorGaps.begin(), TGE = terminatorGaps.end();
       TGI != TGE; ++TGI) {
    terminatorGaps[TGI->first] = TGI->second.scale(factor);
  }

  // Scale the intervals.
  for (iterator LI = begin(), LE = end(); LI != LE; ++LI) {
    LI->second->scaleNumbering(factor);
  }

  // Scale MachineInstrs.
  Mi2IndexMap oldmi2iMap = mi2iMap_;
  LiveIndex highestSlot;
  for (Mi2IndexMap::iterator MI = oldmi2iMap.begin(), ME = oldmi2iMap.end();
       MI != ME; ++MI) {
    LiveIndex newSlot = MI->second.scale(factor);
    mi2iMap_[MI->first] = newSlot;
    highestSlot = std::max(highestSlot, newSlot); 
  }

  unsigned highestVIndex = highestSlot.getVecIndex();
  i2miMap_.clear();
  i2miMap_.resize(highestVIndex + 1);
  for (Mi2IndexMap::iterator MI = mi2iMap_.begin(), ME = mi2iMap_.end();
       MI != ME; ++MI) {
    i2miMap_[MI->second.getVecIndex()] = const_cast<MachineInstr *>(MI->first);
  }

}


/// runOnMachineFunction - Register allocate the whole function
///
bool LiveIntervals::runOnMachineFunction(MachineFunction &fn) {
  mf_ = &fn;
  mri_ = &mf_->getRegInfo();
  tm_ = &fn.getTarget();
  tri_ = tm_->getRegisterInfo();
  tii_ = tm_->getInstrInfo();
  aa_ = &getAnalysis<AliasAnalysis>();
  lv_ = &getAnalysis<LiveVariables>();
  allocatableRegs_ = tri_->getAllocatableSet(fn);

  processImplicitDefs();
  computeNumbering();
  computeIntervals();
  performEarlyCoalescing();

  numIntervals += getNumIntervals();

  DEBUG(dump());
  return true;
}

/// print - Implement the dump method.
void LiveIntervals::print(raw_ostream &OS, const Module* ) const {
  OS << "********** INTERVALS **********\n";
  for (const_iterator I = begin(), E = end(); I != E; ++I) {
    I->second->print(OS, tri_);
    OS << "\n";
  }

  printInstrs(OS);
}

void LiveIntervals::printInstrs(raw_ostream &OS) const {
  OS << "********** MACHINEINSTRS **********\n";

  for (MachineFunction::iterator mbbi = mf_->begin(), mbbe = mf_->end();
       mbbi != mbbe; ++mbbi) {
    OS << ((Value*)mbbi->getBasicBlock())->getName() << ":\n";
    for (MachineBasicBlock::iterator mii = mbbi->begin(),
           mie = mbbi->end(); mii != mie; ++mii) {
      OS << getInstructionIndex(mii) << '\t' << *mii;
    }
  }
}

void LiveIntervals::dumpInstrs() const {
  printInstrs(errs());
}

/// conflictsWithPhysRegDef - Returns true if the specified register
/// is defined during the duration of the specified interval.
bool LiveIntervals::conflictsWithPhysRegDef(const LiveInterval &li,
                                            VirtRegMap &vrm, unsigned reg) {
  for (LiveInterval::Ranges::const_iterator
         I = li.ranges.begin(), E = li.ranges.end(); I != E; ++I) {
    for (LiveIndex index = getBaseIndex(I->start),
           end = getNextIndex(getBaseIndex(getPrevSlot(I->end))); index != end;
         index = getNextIndex(index)) {
      // skip deleted instructions
      while (index != end && !getInstructionFromIndex(index))
        index = getNextIndex(index);
      if (index == end) break;

      MachineInstr *MI = getInstructionFromIndex(index);
      unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
      if (tii_->isMoveInstr(*MI, SrcReg, DstReg, SrcSubReg, DstSubReg))
        if (SrcReg == li.reg || DstReg == li.reg)
          continue;
      for (unsigned i = 0; i != MI->getNumOperands(); ++i) {
        MachineOperand& mop = MI->getOperand(i);
        if (!mop.isReg())
          continue;
        unsigned PhysReg = mop.getReg();
        if (PhysReg == 0 || PhysReg == li.reg)
          continue;
        if (TargetRegisterInfo::isVirtualRegister(PhysReg)) {
          if (!vrm.hasPhys(PhysReg))
            continue;
          PhysReg = vrm.getPhys(PhysReg);
        }
        if (PhysReg && tri_->regsOverlap(PhysReg, reg))
          return true;
      }
    }
  }

  return false;
}

/// conflictsWithPhysRegRef - Similar to conflictsWithPhysRegRef except
/// it can check use as well.
bool LiveIntervals::conflictsWithPhysRegRef(LiveInterval &li,
                                            unsigned Reg, bool CheckUse,
                                  SmallPtrSet<MachineInstr*,32> &JoinedCopies) {
  for (LiveInterval::Ranges::const_iterator
         I = li.ranges.begin(), E = li.ranges.end(); I != E; ++I) {
    for (LiveIndex index = getBaseIndex(I->start),
           end = getNextIndex(getBaseIndex(getPrevSlot(I->end))); index != end;
         index = getNextIndex(index)) {
      // Skip deleted instructions.
      MachineInstr *MI = 0;
      while (index != end) {
        MI = getInstructionFromIndex(index);
        if (MI)
          break;
        index = getNextIndex(index);
      }
      if (index == end) break;

      if (JoinedCopies.count(MI))
        continue;
      for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
        MachineOperand& MO = MI->getOperand(i);
        if (!MO.isReg())
          continue;
        if (MO.isUse() && !CheckUse)
          continue;
        unsigned PhysReg = MO.getReg();
        if (PhysReg == 0 || TargetRegisterInfo::isVirtualRegister(PhysReg))
          continue;
        if (tri_->isSubRegister(Reg, PhysReg))
          return true;
      }
    }
  }

  return false;
}

#ifndef NDEBUG
static void printRegName(unsigned reg, const TargetRegisterInfo* tri_) {
  if (TargetRegisterInfo::isPhysicalRegister(reg))
    errs() << tri_->getName(reg);
  else
    errs() << "%reg" << reg;
}
#endif

void LiveIntervals::handleVirtualRegisterDef(MachineBasicBlock *mbb,
                                             MachineBasicBlock::iterator mi,
                                             LiveIndex MIIdx,
                                             MachineOperand& MO,
                                             unsigned MOIdx,
                                             LiveInterval &interval) {
  DEBUG({
      errs() << "\t\tregister: ";
      printRegName(interval.reg, tri_);
    });

  // Virtual registers may be defined multiple times (due to phi
  // elimination and 2-addr elimination).  Much of what we do only has to be
  // done once for the vreg.  We use an empty interval to detect the first
  // time we see a vreg.
  LiveVariables::VarInfo& vi = lv_->getVarInfo(interval.reg);
  if (interval.empty()) {
    // Get the Idx of the defining instructions.
    LiveIndex defIndex = getDefIndex(MIIdx);
    // Earlyclobbers move back one, so that they overlap the live range
    // of inputs.
    if (MO.isEarlyClobber())
      defIndex = getUseIndex(MIIdx);
    VNInfo *ValNo;
    MachineInstr *CopyMI = NULL;
    unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
    if (mi->getOpcode() == TargetInstrInfo::EXTRACT_SUBREG ||
        mi->getOpcode() == TargetInstrInfo::INSERT_SUBREG ||
        mi->getOpcode() == TargetInstrInfo::SUBREG_TO_REG ||
        tii_->isMoveInstr(*mi, SrcReg, DstReg, SrcSubReg, DstSubReg))
      CopyMI = mi;
    // Earlyclobbers move back one.
    ValNo = interval.getNextValue(defIndex, CopyMI, true, VNInfoAllocator);

    assert(ValNo->id == 0 && "First value in interval is not 0?");

    // Loop over all of the blocks that the vreg is defined in.  There are
    // two cases we have to handle here.  The most common case is a vreg
    // whose lifetime is contained within a basic block.  In this case there
    // will be a single kill, in MBB, which comes after the definition.
    if (vi.Kills.size() == 1 && vi.Kills[0]->getParent() == mbb) {
      // FIXME: what about dead vars?
      LiveIndex killIdx;
      if (vi.Kills[0] != mi)
        killIdx = getNextSlot(getUseIndex(getInstructionIndex(vi.Kills[0])));
      else if (MO.isEarlyClobber())
        // Earlyclobbers that die in this instruction move up one extra, to
        // compensate for having the starting point moved back one.  This
        // gets them to overlap the live range of other outputs.
        killIdx = getNextSlot(getNextSlot(defIndex));
      else
        killIdx = getNextSlot(defIndex);

      // If the kill happens after the definition, we have an intra-block
      // live range.
      if (killIdx > defIndex) {
        assert(vi.AliveBlocks.empty() &&
               "Shouldn't be alive across any blocks!");
        LiveRange LR(defIndex, killIdx, ValNo);
        interval.addRange(LR);
        DEBUG(errs() << " +" << LR << "\n");
        ValNo->addKill(killIdx);
        return;
      }
    }

    // The other case we handle is when a virtual register lives to the end
    // of the defining block, potentially live across some blocks, then is
    // live into some number of blocks, but gets killed.  Start by adding a
    // range that goes from this definition to the end of the defining block.
    LiveRange NewLR(defIndex, getNextSlot(getMBBEndIdx(mbb)), ValNo);
    DEBUG(errs() << " +" << NewLR);
    interval.addRange(NewLR);

    // Iterate over all of the blocks that the variable is completely
    // live in, adding [insrtIndex(begin), instrIndex(end)+4) to the
    // live interval.
    for (SparseBitVector<>::iterator I = vi.AliveBlocks.begin(), 
             E = vi.AliveBlocks.end(); I != E; ++I) {
      LiveRange LR(getMBBStartIdx(*I),
                   getNextSlot(getMBBEndIdx(*I)),  // MBB ends at -1.
                   ValNo);
      interval.addRange(LR);
      DEBUG(errs() << " +" << LR);
    }

    // Finally, this virtual register is live from the start of any killing
    // block to the 'use' slot of the killing instruction.
    for (unsigned i = 0, e = vi.Kills.size(); i != e; ++i) {
      MachineInstr *Kill = vi.Kills[i];
      LiveIndex killIdx =
        getNextSlot(getUseIndex(getInstructionIndex(Kill)));
      LiveRange LR(getMBBStartIdx(Kill->getParent()), killIdx, ValNo);
      interval.addRange(LR);
      ValNo->addKill(killIdx);
      DEBUG(errs() << " +" << LR);
    }

  } else {
    // If this is the second time we see a virtual register definition, it
    // must be due to phi elimination or two addr elimination.  If this is
    // the result of two address elimination, then the vreg is one of the
    // def-and-use register operand.
    if (mi->isRegTiedToUseOperand(MOIdx)) {
      // If this is a two-address definition, then we have already processed
      // the live range.  The only problem is that we didn't realize there
      // are actually two values in the live interval.  Because of this we
      // need to take the LiveRegion that defines this register and split it
      // into two values.
      assert(interval.containsOneValue());
      LiveIndex DefIndex = getDefIndex(interval.getValNumInfo(0)->def);
      LiveIndex RedefIndex = getDefIndex(MIIdx);
      if (MO.isEarlyClobber())
        RedefIndex = getUseIndex(MIIdx);

      const LiveRange *OldLR =
        interval.getLiveRangeContaining(getPrevSlot(RedefIndex));
      VNInfo *OldValNo = OldLR->valno;

      // Delete the initial value, which should be short and continuous,
      // because the 2-addr copy must be in the same MBB as the redef.
      interval.removeRange(DefIndex, RedefIndex);

      // Two-address vregs should always only be redefined once.  This means
      // that at this point, there should be exactly one value number in it.
      assert(interval.containsOneValue() && "Unexpected 2-addr liveint!");

      // The new value number (#1) is defined by the instruction we claimed
      // defined value #0.
      VNInfo *ValNo = interval.getNextValue(OldValNo->def, OldValNo->getCopy(),
                                            false, // update at *
                                            VNInfoAllocator);
      ValNo->setFlags(OldValNo->getFlags()); // * <- updating here

      // Value#0 is now defined by the 2-addr instruction.
      OldValNo->def  = RedefIndex;
      OldValNo->setCopy(0);
      if (MO.isEarlyClobber())
        OldValNo->setHasRedefByEC(true);
      
      // Add the new live interval which replaces the range for the input copy.
      LiveRange LR(DefIndex, RedefIndex, ValNo);
      DEBUG(errs() << " replace range with " << LR);
      interval.addRange(LR);
      ValNo->addKill(RedefIndex);

      // If this redefinition is dead, we need to add a dummy unit live
      // range covering the def slot.
      if (MO.isDead())
        interval.addRange(
          LiveRange(RedefIndex, MO.isEarlyClobber() ?
                                getNextSlot(getNextSlot(RedefIndex)) :
                                getNextSlot(RedefIndex), OldValNo));

      DEBUG({
          errs() << " RESULT: ";
          interval.print(errs(), tri_);
        });
    } else {
      // Otherwise, this must be because of phi elimination.  If this is the
      // first redefinition of the vreg that we have seen, go back and change
      // the live range in the PHI block to be a different value number.
      if (interval.containsOneValue()) {
        // Remove the old range that we now know has an incorrect number.
        VNInfo *VNI = interval.getValNumInfo(0);
        MachineInstr *Killer = vi.Kills[0];
        phiJoinCopies.push_back(Killer);
        LiveIndex Start = getMBBStartIdx(Killer->getParent());
        LiveIndex End =
          getNextSlot(getUseIndex(getInstructionIndex(Killer)));
        DEBUG({
            errs() << " Removing [" << Start << "," << End << "] from: ";
            interval.print(errs(), tri_);
            errs() << "\n";
          });
        interval.removeRange(Start, End);        
        assert(interval.ranges.size() == 1 &&
               "Newly discovered PHI interval has >1 ranges.");
        MachineBasicBlock *killMBB = getMBBFromIndex(interval.endIndex());
        VNI->addKill(terminatorGaps[killMBB]);
        VNI->setHasPHIKill(true);
        DEBUG({
            errs() << " RESULT: ";
            interval.print(errs(), tri_);
          });

        // Replace the interval with one of a NEW value number.  Note that this
        // value number isn't actually defined by an instruction, weird huh? :)
        LiveRange LR(Start, End,
          interval.getNextValue(LiveIndex(mbb->getNumber()),
                                0, false, VNInfoAllocator));
        LR.valno->setIsPHIDef(true);
        DEBUG(errs() << " replace range with " << LR);
        interval.addRange(LR);
        LR.valno->addKill(End);
        DEBUG({
            errs() << " RESULT: ";
            interval.print(errs(), tri_);
          });
      }

      // In the case of PHI elimination, each variable definition is only
      // live until the end of the block.  We've already taken care of the
      // rest of the live range.
      LiveIndex defIndex = getDefIndex(MIIdx);
      if (MO.isEarlyClobber())
        defIndex = getUseIndex(MIIdx);

      VNInfo *ValNo;
      MachineInstr *CopyMI = NULL;
      unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
      if (mi->getOpcode() == TargetInstrInfo::EXTRACT_SUBREG ||
          mi->getOpcode() == TargetInstrInfo::INSERT_SUBREG ||
          mi->getOpcode() == TargetInstrInfo::SUBREG_TO_REG ||
          tii_->isMoveInstr(*mi, SrcReg, DstReg, SrcSubReg, DstSubReg))
        CopyMI = mi;
      ValNo = interval.getNextValue(defIndex, CopyMI, true, VNInfoAllocator);
      
      LiveIndex killIndex = getNextSlot(getMBBEndIdx(mbb));
      LiveRange LR(defIndex, killIndex, ValNo);
      interval.addRange(LR);
      ValNo->addKill(terminatorGaps[mbb]);
      ValNo->setHasPHIKill(true);
      DEBUG(errs() << " +" << LR);
    }
  }

  DEBUG(errs() << '\n');
}

void LiveIntervals::handlePhysicalRegisterDef(MachineBasicBlock *MBB,
                                              MachineBasicBlock::iterator mi,
                                              LiveIndex MIIdx,
                                              MachineOperand& MO,
                                              LiveInterval &interval,
                                              MachineInstr *CopyMI) {
  // A physical register cannot be live across basic block, so its
  // lifetime must end somewhere in its defining basic block.
  DEBUG({
      errs() << "\t\tregister: ";
      printRegName(interval.reg, tri_);
    });

  LiveIndex baseIndex = MIIdx;
  LiveIndex start = getDefIndex(baseIndex);
  // Earlyclobbers move back one.
  if (MO.isEarlyClobber())
    start = getUseIndex(MIIdx);
  LiveIndex end = start;

  // If it is not used after definition, it is considered dead at
  // the instruction defining it. Hence its interval is:
  // [defSlot(def), defSlot(def)+1)
  // For earlyclobbers, the defSlot was pushed back one; the extra
  // advance below compensates.
  if (MO.isDead()) {
    DEBUG(errs() << " dead");
    if (MO.isEarlyClobber())
      end = getNextSlot(getNextSlot(start));
    else
      end = getNextSlot(start);
    goto exit;
  }

  // If it is not dead on definition, it must be killed by a
  // subsequent instruction. Hence its interval is:
  // [defSlot(def), useSlot(kill)+1)
  baseIndex = getNextIndex(baseIndex);
  while (++mi != MBB->end()) {
    while (baseIndex.getVecIndex() < i2miMap_.size() &&
           getInstructionFromIndex(baseIndex) == 0)
      baseIndex = getNextIndex(baseIndex);
    if (mi->killsRegister(interval.reg, tri_)) {
      DEBUG(errs() << " killed");
      end = getNextSlot(getUseIndex(baseIndex));
      goto exit;
    } else {
      int DefIdx = mi->findRegisterDefOperandIdx(interval.reg, false, tri_);
      if (DefIdx != -1) {
        if (mi->isRegTiedToUseOperand(DefIdx)) {
          // Two-address instruction.
          end = getDefIndex(baseIndex);
          if (mi->getOperand(DefIdx).isEarlyClobber())
            end = getUseIndex(baseIndex);
        } else {
          // Another instruction redefines the register before it is ever read.
          // Then the register is essentially dead at the instruction that defines
          // it. Hence its interval is:
          // [defSlot(def), defSlot(def)+1)
          DEBUG(errs() << " dead");
          end = getNextSlot(start);
        }
        goto exit;
      }
    }
    
    baseIndex = getNextIndex(baseIndex);
  }
  
  // The only case we should have a dead physreg here without a killing or
  // instruction where we know it's dead is if it is live-in to the function
  // and never used. Another possible case is the implicit use of the
  // physical register has been deleted by two-address pass.
  end = getNextSlot(start);

exit:
  assert(start < end && "did not find end of interval?");

  // Already exists? Extend old live interval.
  LiveInterval::iterator OldLR = interval.FindLiveRangeContaining(start);
  bool Extend = OldLR != interval.end();
  VNInfo *ValNo = Extend
    ? OldLR->valno : interval.getNextValue(start, CopyMI, true, VNInfoAllocator);
  if (MO.isEarlyClobber() && Extend)
    ValNo->setHasRedefByEC(true);
  LiveRange LR(start, end, ValNo);
  interval.addRange(LR);
  LR.valno->addKill(end);
  DEBUG(errs() << " +" << LR << '\n');
}

void LiveIntervals::handleRegisterDef(MachineBasicBlock *MBB,
                                      MachineBasicBlock::iterator MI,
                                      LiveIndex MIIdx,
                                      MachineOperand& MO,
                                      unsigned MOIdx) {
  if (TargetRegisterInfo::isVirtualRegister(MO.getReg()))
    handleVirtualRegisterDef(MBB, MI, MIIdx, MO, MOIdx,
                             getOrCreateInterval(MO.getReg()));
  else if (allocatableRegs_[MO.getReg()]) {
    MachineInstr *CopyMI = NULL;
    unsigned SrcReg, DstReg, SrcSubReg, DstSubReg;
    if (MI->getOpcode() == TargetInstrInfo::EXTRACT_SUBREG ||