//===-- LiveIntervalAnalysis.cpp - Live Interval Analysis -----------------===// // // The LLVM Compiler Infrastructure // // This file was developed by the LLVM research group and 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 "LiveIntervalAnalysis.h" #include "llvm/Value.h" #include "llvm/Analysis/LoopInfo.h" #include "llvm/CodeGen/LiveVariables.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineInstr.h" #include "llvm/CodeGen/Passes.h" #include "llvm/CodeGen/SSARegMap.h" #include "llvm/Target/MRegisterInfo.h" #include "llvm/Target/TargetInstrInfo.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/ADT/Statistic.h" #include "llvm/ADT/STLExtras.h" #include "VirtRegMap.h" #include using namespace llvm; namespace { RegisterAnalysis X("liveintervals", "Live Interval Analysis"); Statistic<> numIntervals ("liveintervals", "Number of original intervals"); Statistic<> numIntervalsAfter ("liveintervals", "Number of intervals after coalescing"); Statistic<> numJoins ("liveintervals", "Number of interval joins performed"); Statistic<> numPeep ("liveintervals", "Number of identity moves eliminated after coalescing"); Statistic<> numFolded ("liveintervals", "Number of loads/stores folded into instructions"); cl::opt EnableJoining("join-liveintervals", cl::desc("Join compatible live intervals"), cl::init(true)); }; void LiveIntervals::getAnalysisUsage(AnalysisUsage &AU) const { AU.addPreserved(); AU.addRequired(); AU.addPreservedID(PHIEliminationID); AU.addRequiredID(PHIEliminationID); AU.addRequiredID(TwoAddressInstructionPassID); AU.addRequired(); MachineFunctionPass::getAnalysisUsage(AU); } void LiveIntervals::releaseMemory() { mi2iMap_.clear(); i2miMap_.clear(); r2iMap_.clear(); r2rMap_.clear(); } /// runOnMachineFunction - Register allocate the whole function /// bool LiveIntervals::runOnMachineFunction(MachineFunction &fn) { mf_ = &fn; tm_ = &fn.getTarget(); mri_ = tm_->getRegisterInfo(); lv_ = &getAnalysis(); allocatableRegs_ = mri_->getAllocatableSet(fn); // number MachineInstrs unsigned miIndex = 0; for (MachineFunction::iterator mbb = mf_->begin(), mbbEnd = mf_->end(); mbb != mbbEnd; ++mbb) for (MachineBasicBlock::iterator mi = mbb->begin(), miEnd = mbb->end(); mi != miEnd; ++mi) { bool inserted = mi2iMap_.insert(std::make_pair(mi, miIndex)).second; assert(inserted && "multiple MachineInstr -> index mappings"); i2miMap_.push_back(mi); miIndex += InstrSlots::NUM; } computeIntervals(); numIntervals += getNumIntervals(); #if 1 DEBUG(std::cerr << "********** INTERVALS **********\n"); DEBUG(for (iterator I = begin(), E = end(); I != E; ++I) std::cerr << I->second << "\n"); #endif // join intervals if requested if (EnableJoining) joinIntervals(); numIntervalsAfter += getNumIntervals(); // perform a final pass over the instructions and compute spill // weights, coalesce virtual registers and remove identity moves const LoopInfo& loopInfo = getAnalysis(); const TargetInstrInfo& tii = *tm_->getInstrInfo(); for (MachineFunction::iterator mbbi = mf_->begin(), mbbe = mf_->end(); mbbi != mbbe; ++mbbi) { MachineBasicBlock* mbb = mbbi; unsigned loopDepth = loopInfo.getLoopDepth(mbb->getBasicBlock()); for (MachineBasicBlock::iterator mii = mbb->begin(), mie = mbb->end(); mii != mie; ) { // if the move will be an identity move delete it unsigned srcReg, dstReg, RegRep; if (tii.isMoveInstr(*mii, srcReg, dstReg) && (RegRep = rep(srcReg)) == rep(dstReg)) { // remove from def list LiveInterval &interval = getOrCreateInterval(RegRep); // remove index -> MachineInstr and // MachineInstr -> index mappings Mi2IndexMap::iterator mi2i = mi2iMap_.find(mii); if (mi2i != mi2iMap_.end()) { i2miMap_[mi2i->second/InstrSlots::NUM] = 0; mi2iMap_.erase(mi2i); } mii = mbbi->erase(mii); ++numPeep; } else { for (unsigned i = 0; i < mii->getNumOperands(); ++i) { const MachineOperand& mop = mii->getOperand(i); if (mop.isRegister() && mop.getReg() && MRegisterInfo::isVirtualRegister(mop.getReg())) { // replace register with representative register unsigned reg = rep(mop.getReg()); mii->SetMachineOperandReg(i, reg); LiveInterval &RegInt = getInterval(reg); RegInt.weight += (mop.isUse() + mop.isDef()) * pow(10.0F, loopDepth); } } ++mii; } } } DEBUG(std::cerr << "********** INTERVALS **********\n"); DEBUG (for (iterator I = begin(), E = end(); I != E; ++I) std::cerr << I->second << "\n"); DEBUG(std::cerr << "********** MACHINEINSTRS **********\n"); DEBUG( for (MachineFunction::iterator mbbi = mf_->begin(), mbbe = mf_->end(); mbbi != mbbe; ++mbbi) { std::cerr << ((Value*)mbbi->getBasicBlock())->getName() << ":\n"; for (MachineBasicBlock::iterator mii = mbbi->begin(), mie = mbbi->end(); mii != mie; ++mii) { std::cerr << getInstructionIndex(mii) << '\t'; mii->print(std::cerr, tm_); } }); return true; } std::vector LiveIntervals::addIntervalsForSpills( const LiveInterval& li, VirtRegMap& vrm, int slot) { // since this is called after the analysis is done we don't know if // LiveVariables is available lv_ = getAnalysisToUpdate(); std::vector added; assert(li.weight != HUGE_VAL && "attempt to spill already spilled interval!"); DEBUG(std::cerr << "\t\t\t\tadding intervals for spills for interval: " << li << '\n'); const TargetRegisterClass* rc = mf_->getSSARegMap()->getRegClass(li.reg); for (LiveInterval::Ranges::const_iterator i = li.ranges.begin(), e = li.ranges.end(); i != e; ++i) { unsigned index = getBaseIndex(i->start); unsigned end = getBaseIndex(i->end-1) + InstrSlots::NUM; for (; index != end; index += InstrSlots::NUM) { // skip deleted instructions while (index != end && !getInstructionFromIndex(index)) index += InstrSlots::NUM; if (index == end) break; MachineBasicBlock::iterator mi = getInstructionFromIndex(index); for_operand: for (unsigned i = 0; i != mi->getNumOperands(); ++i) { MachineOperand& mop = mi->getOperand(i); if (mop.isRegister() && mop.getReg() == li.reg) { if (MachineInstr* fmi = mri_->foldMemoryOperand(mi, i, slot)) { if (lv_) lv_->instructionChanged(mi, fmi); vrm.virtFolded(li.reg, mi, fmi); mi2iMap_.erase(mi); i2miMap_[index/InstrSlots::NUM] = fmi; mi2iMap_[fmi] = index; MachineBasicBlock& mbb = *mi->getParent(); mi = mbb.insert(mbb.erase(mi), fmi); ++numFolded; goto for_operand; } else { // This is tricky. We need to add information in // the interval about the spill code so we have to // use our extra load/store slots. // // If we have a use we are going to have a load so // we start the interval from the load slot // onwards. Otherwise we start from the def slot. unsigned start = (mop.isUse() ? getLoadIndex(index) : getDefIndex(index)); // If we have a def we are going to have a store // right after it so we end the interval after the // use of the next instruction. Otherwise we end // after the use of this instruction. unsigned end = 1 + (mop.isDef() ? getStoreIndex(index) : getUseIndex(index)); // create a new register for this spill unsigned nReg = mf_->getSSARegMap()->createVirtualRegister(rc); mi->SetMachineOperandReg(i, nReg); vrm.grow(); vrm.assignVirt2StackSlot(nReg, slot); LiveInterval& nI = getOrCreateInterval(nReg); assert(nI.empty()); // the spill weight is now infinity as it // cannot be spilled again nI.weight = HUGE_VAL; LiveRange LR(start, end, nI.getNextValue()); DEBUG(std::cerr << " +" << LR); nI.addRange(LR); added.push_back(&nI); // update live variables if it is available if (lv_) lv_->addVirtualRegisterKilled(nReg, mi); DEBUG(std::cerr << "\t\t\t\tadded new interval: " << nI << '\n'); } } } } } return added; } void LiveIntervals::printRegName(unsigned reg) const { if (MRegisterInfo::isPhysicalRegister(reg)) std::cerr << mri_->getName(reg); else std::cerr << "%reg" << reg; } void LiveIntervals::handleVirtualRegisterDef(MachineBasicBlock* mbb, MachineBasicBlock::iterator mi, LiveInterval& interval) { DEBUG(std::cerr << "\t\tregister: "; printRegName(interval.reg)); LiveVariables::VarInfo& vi = lv_->getVarInfo(interval.reg); // 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. if (interval.empty()) { // Get the Idx of the defining instructions. unsigned defIndex = getDefIndex(getInstructionIndex(mi)); unsigned ValNum = interval.getNextValue(); assert(ValNum == 0 && "First value in interval is not 0?"); ValNum = 0; // Clue in the optimizer. // 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? unsigned killIdx; if (vi.Kills[0] != mi) killIdx = getUseIndex(getInstructionIndex(vi.Kills[0]))+1; else killIdx = defIndex+1; // 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, ValNum); interval.addRange(LR); DEBUG(std::cerr << " +" << LR << "\n"); 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, getInstructionIndex(&mbb->back()) + InstrSlots::NUM, ValNum); DEBUG(std::cerr << " +" << 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 (unsigned i = 0, e = vi.AliveBlocks.size(); i != e; ++i) { if (vi.AliveBlocks[i]) { MachineBasicBlock* mbb = mf_->getBlockNumbered(i); if (!mbb->empty()) { LiveRange LR(getInstructionIndex(&mbb->front()), getInstructionIndex(&mbb->back()) + InstrSlots::NUM, ValNum); interval.addRange(LR); DEBUG(std::cerr << " +" << 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]; LiveRange LR(getInstructionIndex(Kill->getParent()->begin()), getUseIndex(getInstructionIndex(Kill))+1, ValNum); interval.addRange(LR); DEBUG(std::cerr << " +" << 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 the first // operand, and is a def-and-use. if (mi->getOperand(0).isRegister() && mi->getOperand(0).getReg() == interval.reg && mi->getOperand(0).isDef() && mi->getOperand(0).isUse()) { // 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. unsigned DefIndex = getDefIndex(getInstructionIndex(vi.DefInst)); unsigned RedefIndex = getDefIndex(getInstructionIndex(mi)); // Delete the initial value, which should be short and continuous, // becuase the 2-addr copy must be in the same MBB as the redef. interval.removeRange(DefIndex, RedefIndex); LiveRange LR(DefIndex, RedefIndex, interval.getNextValue()); DEBUG(std::cerr << " replace range with " << LR); interval.addRange(LR); // If this redefinition is dead, we need to add a dummy unit live // range covering the def slot. for (LiveVariables::killed_iterator KI = lv_->dead_begin(mi), E = lv_->dead_end(mi); KI != E; ++KI) if (KI->second == interval.reg) { interval.addRange(LiveRange(RedefIndex, RedefIndex+1, 0)); break; } DEBUG(std::cerr << "RESULT: " << interval); } 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()) { assert(vi.Kills.size() == 1 && "PHI elimination vreg should have one kill, the PHI itself!"); // Remove the old range that we now know has an incorrect number. MachineInstr *Killer = vi.Kills[0]; unsigned Start = getInstructionIndex(Killer->getParent()->begin()); unsigned End = getUseIndex(getInstructionIndex(Killer))+1; DEBUG(std::cerr << "Removing [" << Start << "," << End << "] from: " << interval << "\n"); interval.removeRange(Start, End); DEBUG(std::cerr << "RESULT: " << interval); // Replace the interval with one of a NEW value number. LiveRange LR(Start, End, interval.getNextValue()); DEBUG(std::cerr << " replace range with " << LR); interval.addRange(LR); DEBUG(std::cerr << "RESULT: " << interval); } // 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. unsigned defIndex = getDefIndex(getInstructionIndex(mi)); LiveRange LR(defIndex, getInstructionIndex(&mbb->back()) + InstrSlots::NUM, interval.getNextValue()); interval.addRange(LR); DEBUG(std::cerr << " +" << LR); } } DEBUG(std::cerr << '\n'); } void LiveIntervals::handlePhysicalRegisterDef(MachineBasicBlock *MBB, MachineBasicBlock::iterator mi, LiveInterval& interval) { // A physical register cannot be live across basic block, so its // lifetime must end somewhere in its defining basic block. DEBUG(std::cerr << "\t\tregister: "; printRegName(interval.reg)); typedef LiveVariables::killed_iterator KillIter; unsigned baseIndex = getInstructionIndex(mi); unsigned start = getDefIndex(baseIndex); unsigned 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 (KillIter ki = lv_->dead_begin(mi), ke = lv_->dead_end(mi); ki != ke; ++ki) { if (interval.reg == ki->second) { DEBUG(std::cerr << " dead"); end = getDefIndex(start) + 1; 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) while (true) { ++mi; assert(mi != MBB->end() && "physreg was not killed in defining block!"); baseIndex += InstrSlots::NUM; for (KillIter ki = lv_->killed_begin(mi), ke = lv_->killed_end(mi); ki != ke; ++ki) { if (interval.reg == ki->second) { DEBUG(std::cerr << " killed"); end = getUseIndex(baseIndex) + 1; goto exit; } } } exit: assert(start < end && "did not find end of interval?"); LiveRange LR(start, end, interval.getNextValue()); interval.addRange(LR); DEBUG(std::cerr << " +" << LR << '\n'); } void LiveIntervals::handleRegisterDef(MachineBasicBlock *MBB, MachineBasicBlock::iterator MI, unsigned reg) { if (MRegisterInfo::isVirtualRegister(reg)) handleVirtualRegisterDef(MBB, MI, getOrCreateInterval(reg)); else if (allocatableRegs_[reg]) { handlePhysicalRegisterDef(MBB, MI, getOrCreateInterval(reg)); for (const unsigned* AS = mri_->getAliasSet(reg); *AS; ++AS) handlePhysicalRegisterDef(MBB, MI, getOrCreateInterval(*AS)); } } /// computeIntervals - computes the live intervals for virtual /// registers. for some ordering of the machine instructions [1,N] a /// live interval is an interval [i, j) where 1 <= i <= j < N for /// which a variable is live void LiveIntervals::computeIntervals() { DEBUG(std::cerr << "********** COMPUTING LIVE INTERVALS **********\n"); DEBUG(std::cerr << "********** Function: " << ((Value*)mf_->getFunction())->getName() << '\n'); for (MachineFunction::iterator I = mf_->begin(), E = mf_->end(); I != E; ++I) { MachineBasicBlock* mbb = I; DEBUG(std::cerr << ((Value*)mbb->getBasicBlock())->getName() << ":\n"); for (MachineBasicBlock::iterator mi = mbb->begin(), miEnd = mbb->end(); mi != miEnd; ++mi) { const TargetInstrDescriptor& tid = tm_->getInstrInfo()->get(mi->getOpcode()); DEBUG(std::cerr << getInstructionIndex(mi) << "\t"; mi->print(std::cerr, tm_)); // handle implicit defs for (const unsigned* id = tid.ImplicitDefs; *id; ++id) handleRegisterDef(mbb, mi, *id); // handle explicit defs for (int i = mi->getNumOperands() - 1; i >= 0; --i) { MachineOperand& mop = mi->getOperand(i); // handle register defs - build intervals if (mop.isRegister() && mop.getReg() && mop.isDef()) handleRegisterDef(mbb, mi, mop.getReg()); } } } } void LiveIntervals::joinIntervalsInMachineBB(MachineBasicBlock *MBB) { DEBUG(std::cerr << ((Value*)MBB->getBasicBlock())->getName() << ":\n"); const TargetInstrInfo &TII = *tm_->getInstrInfo(); for (MachineBasicBlock::iterator mi = MBB->begin(), mie = MBB->end(); mi != mie; ++mi) { DEBUG(std::cerr << getInstructionIndex(mi) << '\t' << *mi); // we only join virtual registers with allocatable // physical registers since we do not have liveness information // on not allocatable physical registers unsigned regA, regB; if (TII.isMoveInstr(*mi, regA, regB) && (MRegisterInfo::isVirtualRegister(regA) || allocatableRegs_[regA]) && (MRegisterInfo::isVirtualRegister(regB) || allocatableRegs_[regB])) { // Get representative registers. regA = rep(regA); regB = rep(regB); // If they are already joined we continue. if (regA == regB) continue; // If they are both physical registers, we cannot join them. if (MRegisterInfo::isPhysicalRegister(regA) && MRegisterInfo::isPhysicalRegister(regB)) continue; // If they are not of the same register class, we cannot join them. if (differingRegisterClasses(regA, regB)) continue; LiveInterval &IntA = getInterval(regA); LiveInterval &IntB = getInterval(regB); assert(IntA.reg == regA && IntB.reg == regB && "Register mapping is horribly broken!"); DEBUG(std::cerr << "\t\tInspecting " << IntA << " and " << IntB << ": "); // If two intervals contain a single value and are joined by a copy, it // does not matter if the intervals overlap, they can always be joined. bool TriviallyJoinable = IntA.containsOneValue() && IntB.containsOneValue(); unsigned MIDefIdx = getDefIndex(getInstructionIndex(mi)); if ((TriviallyJoinable || IntB.joinable(IntA, MIDefIdx)) && !overlapsAliases(&IntA, &IntB)) { IntB.join(IntA, MIDefIdx); if (!MRegisterInfo::isPhysicalRegister(regA)) { r2iMap_.erase(regA); r2rMap_[regA] = regB; } else { // Otherwise merge the data structures the other way so we don't lose // the physreg information. r2rMap_[regB] = regA; IntB.reg = regA; IntA.swap(IntB); r2iMap_.erase(regB); } DEBUG(std::cerr << "Joined. Result = " << IntB << "\n"); ++numJoins; } else { DEBUG(std::cerr << "Interference!\n"); } } } } namespace { // DepthMBBCompare - Comparison predicate that sort first based on the loop // depth of the basic block (the unsigned), and then on the MBB number. struct DepthMBBCompare { typedef std::pair DepthMBBPair; bool operator()(const DepthMBBPair &LHS, const DepthMBBPair &RHS) const { if (LHS.first > RHS.first) return true; // Deeper loops first return LHS.first == RHS.first && LHS.second->getNumber() < RHS.second->getNumber(); } }; } void LiveIntervals::joinIntervals() { DEBUG(std::cerr << "********** JOINING INTERVALS ***********\n"); const LoopInfo &LI = getAnalysis(); if (LI.begin() == LI.end()) { // If there are no loops in the function, join intervals in function order. for (MachineFunction::iterator I = mf_->begin(), E = mf_->end(); I != E; ++I) joinIntervalsInMachineBB(I); } else { // Otherwise, join intervals in inner loops before other intervals. // Unfortunately we can't just iterate over loop hierarchy here because // there may be more MBB's than BB's. Collect MBB's for sorting. std::vector > MBBs; for (MachineFunction::iterator I = mf_->begin(), E = mf_->end(); I != E; ++I) MBBs.push_back(std::make_pair(LI.getLoopDepth(I->getBasicBlock()), I)); // Sort by loop depth. std::sort(MBBs.begin(), MBBs.end(), DepthMBBCompare()); // Finally, join intervals in loop nest order. for (unsigned i = 0, e = MBBs.size(); i != e; ++i) joinIntervalsInMachineBB(MBBs[i].second); } DEBUG(std::cerr << "*** Register mapping ***\n"); DEBUG(for (std::map::iterator I = r2rMap_.begin(), E = r2rMap_.end(); I != E; ++I) std::cerr << " reg " << I->first << " -> reg " << I->second << "\n";); } /// Return true if the two specified registers belong to different register /// classes. The registers may be either phys or virt regs. bool LiveIntervals::differingRegisterClasses(unsigned RegA, unsigned RegB) const { const TargetRegisterClass *RegClass; // Get the register classes for the first reg. if (MRegisterInfo::isVirtualRegister(RegA)) RegClass = mf_->getSSARegMap()->getRegClass(RegA); else RegClass = mri_->getRegClass(RegA); // Compare against the regclass for the second reg. if (MRegisterInfo::isVirtualRegister(RegB)) return RegClass != mf_->getSSARegMap()->getRegClass(RegB); else return !RegClass->contains(RegB); } bool LiveIntervals::overlapsAliases(const LiveInterval *LHS, const LiveInterval *RHS) const { if (!MRegisterInfo::isPhysicalRegister(LHS->reg)) { if (!MRegisterInfo::isPhysicalRegister(RHS->reg)) return false; // vreg-vreg merge has no aliases! std::swap(LHS, RHS); } assert(MRegisterInfo::isPhysicalRegister(LHS->reg) && MRegisterInfo::isVirtualRegister(RHS->reg) && "first interval must describe a physical register"); for (const unsigned *AS = mri_->getAliasSet(LHS->reg); *AS; ++AS) if (RHS->overlaps(getInterval(*AS))) return true; return false; } LiveInterval LiveIntervals::createInterval(unsigned reg) { float Weight = MRegisterInfo::isPhysicalRegister(reg) ? HUGE_VAL :0.0F; return LiveInterval(reg, Weight); }