diff options
author | Evan Cheng <evan.cheng@apple.com> | 2009-05-03 18:32:42 +0000 |
---|---|---|
committer | Evan Cheng <evan.cheng@apple.com> | 2009-05-03 18:32:42 +0000 |
commit | c781a243a3d17e7e763515794168d8fa6043f565 (patch) | |
tree | dcaf1b9201cdb221cb2206858891575d2b241bbb /lib | |
parent | 87e3caf81939db20d2359bd38df2ed206040026d (diff) | |
download | external_llvm-c781a243a3d17e7e763515794168d8fa6043f565.zip external_llvm-c781a243a3d17e7e763515794168d8fa6043f565.tar.gz external_llvm-c781a243a3d17e7e763515794168d8fa6043f565.tar.bz2 |
In some rare cases, the register allocator can spill registers but end up not utilizing registers at all. The fundamental problem is linearscan's backtracking can end up freeing more than one allocated registers. However, reloads and restores might be folded into uses / defs and freed registers might not be used at all.
VirtRegMap keeps track of allocations so it knows what's not used. As a horrible hack, the stack coloring can color spill slots with *free* registers. That is, it replace reload and spills with copies from and to the free register. It unfold instructions that load and store the spill slot and replace them with register using variants.
Not yet enabled. This is part 1. More coming.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@70787 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib')
-rw-r--r-- | lib/CodeGen/LiveIntervalAnalysis.cpp | 58 | ||||
-rw-r--r-- | lib/CodeGen/LiveStackAnalysis.cpp | 12 | ||||
-rw-r--r-- | lib/CodeGen/PreAllocSplitting.cpp | 7 | ||||
-rw-r--r-- | lib/CodeGen/RegAllocLinearScan.cpp | 52 | ||||
-rw-r--r-- | lib/CodeGen/RegAllocPBQP.cpp | 16 | ||||
-rw-r--r-- | lib/CodeGen/StackSlotColoring.cpp | 335 | ||||
-rw-r--r-- | lib/CodeGen/VirtRegMap.cpp | 36 | ||||
-rw-r--r-- | lib/CodeGen/VirtRegMap.h | 55 |
8 files changed, 427 insertions, 144 deletions
diff --git a/lib/CodeGen/LiveIntervalAnalysis.cpp b/lib/CodeGen/LiveIntervalAnalysis.cpp index d2927ed..612b9ac 100644 --- a/lib/CodeGen/LiveIntervalAnalysis.cpp +++ b/lib/CodeGen/LiveIntervalAnalysis.cpp @@ -1229,9 +1229,7 @@ rewriteInstructionForSpills(const LiveInterval &li, const VNInfo *VNI, const MachineLoopInfo *loopInfo, unsigned &NewVReg, unsigned ImpUse, bool &HasDef, bool &HasUse, DenseMap<unsigned,unsigned> &MBBVRegsMap, - std::vector<LiveInterval*> &NewLIs, float &SSWeight) { - MachineBasicBlock *MBB = MI->getParent(); - unsigned loopDepth = loopInfo->getLoopDepth(MBB); + std::vector<LiveInterval*> &NewLIs) { bool CanFold = false; RestartInstruction: for (unsigned i = 0; i != MI->getNumOperands(); ++i) { @@ -1312,11 +1310,6 @@ rewriteInstructionForSpills(const LiveInterval &li, const VNInfo *VNI, // the INSERT_SUBREG and both target registers that would overlap. HasUse = false; - // Update stack slot spill weight if we are splitting. - float Weight = getSpillWeight(HasDef, HasUse, loopDepth); - if (!TrySplit) - SSWeight += Weight; - // Create a new virtual register for the spill interval. // Create the new register now so we can map the fold instruction // to the new register so when it is unfolded we get the correct @@ -1348,10 +1341,8 @@ rewriteInstructionForSpills(const LiveInterval &li, const VNInfo *VNI, HasUse = false; HasDef = false; CanFold = false; - if (isNotInMIMap(MI)) { - SSWeight -= Weight; + if (isNotInMIMap(MI)) break; - } goto RestartInstruction; } } else { @@ -1486,7 +1477,7 @@ rewriteInstructionsForSpills(const LiveInterval &li, bool TrySplit, BitVector &RestoreMBBs, DenseMap<unsigned, std::vector<SRInfo> > &RestoreIdxes, DenseMap<unsigned,unsigned> &MBBVRegsMap, - std::vector<LiveInterval*> &NewLIs, float &SSWeight) { + std::vector<LiveInterval*> &NewLIs) { bool AllCanFold = true; unsigned NewVReg = 0; unsigned start = getBaseIndex(I->start); @@ -1588,7 +1579,7 @@ rewriteInstructionsForSpills(const LiveInterval &li, bool TrySplit, index, end, MI, ReMatOrigDefMI, ReMatDefMI, Slot, LdSlot, isLoad, isLoadSS, DefIsReMat, CanDelete, vrm, rc, ReMatIds, loopInfo, NewVReg, - ImpUse, HasDef, HasUse, MBBVRegsMap, NewLIs, SSWeight); + ImpUse, HasDef, HasUse, MBBVRegsMap, NewLIs); if (!HasDef && !HasUse) continue; @@ -1747,7 +1738,7 @@ LiveIntervals::handleSpilledImpDefs(const LiveInterval &li, VirtRegMap &vrm, std::vector<LiveInterval*> LiveIntervals:: addIntervalsForSpillsFast(const LiveInterval &li, const MachineLoopInfo *loopInfo, - VirtRegMap &vrm, float& SSWeight) { + VirtRegMap &vrm) { unsigned slot = vrm.assignVirt2StackSlot(li.reg); std::vector<LiveInterval*> added; @@ -1761,8 +1752,6 @@ addIntervalsForSpillsFast(const LiveInterval &li, const TargetRegisterClass* rc = mri_->getRegClass(li.reg); - SSWeight = 0.0f; - MachineRegisterInfo::reg_iterator RI = mri_->reg_begin(li.reg); while (RI != mri_->reg_end()) { MachineInstr* MI = &*RI; @@ -1825,15 +1814,6 @@ addIntervalsForSpillsFast(const LiveInterval &li, DOUT << "\t\t\t\tadded new interval: "; DEBUG(nI.dump()); DOUT << '\n'; - - unsigned loopDepth = loopInfo->getLoopDepth(MI->getParent()); - if (HasUse) { - if (HasDef) - SSWeight += getSpillWeight(true, true, loopDepth); - else - SSWeight += getSpillWeight(false, true, loopDepth); - } else - SSWeight += getSpillWeight(true, false, loopDepth); } @@ -1846,11 +1826,10 @@ addIntervalsForSpillsFast(const LiveInterval &li, std::vector<LiveInterval*> LiveIntervals:: addIntervalsForSpills(const LiveInterval &li, SmallVectorImpl<LiveInterval*> &SpillIs, - const MachineLoopInfo *loopInfo, VirtRegMap &vrm, - float &SSWeight) { + const MachineLoopInfo *loopInfo, VirtRegMap &vrm) { if (EnableFastSpilling) - return addIntervalsForSpillsFast(li, loopInfo, vrm, SSWeight); + return addIntervalsForSpillsFast(li, loopInfo, vrm); assert(li.weight != HUGE_VALF && "attempt to spill already spilled interval!"); @@ -1859,9 +1838,6 @@ addIntervalsForSpills(const LiveInterval &li, li.print(DOUT, tri_); DOUT << '\n'; - // Spill slot weight. - SSWeight = 0.0f; - // Each bit specify whether a spill is required in the MBB. BitVector SpillMBBs(mf_->getNumBlockIDs()); DenseMap<unsigned, std::vector<SRInfo> > SpillIdxes; @@ -1916,18 +1892,17 @@ addIntervalsForSpills(const LiveInterval &li, Slot, LdSlot, isLoad, isLoadSS, DefIsReMat, false, vrm, rc, ReMatIds, loopInfo, SpillMBBs, SpillIdxes, RestoreMBBs, RestoreIdxes, - MBBVRegsMap, NewLIs, SSWeight); + MBBVRegsMap, NewLIs); } else { rewriteInstructionsForSpills(li, false, I, NULL, 0, Slot, 0, false, false, false, false, vrm, rc, ReMatIds, loopInfo, SpillMBBs, SpillIdxes, RestoreMBBs, RestoreIdxes, - MBBVRegsMap, NewLIs, SSWeight); + MBBVRegsMap, NewLIs); } IsFirstRange = false; } - SSWeight = 0.0f; // Already accounted for when split. handleSpilledImpDefs(li, vrm, rc, NewLIs); return NewLIs; } @@ -2001,7 +1976,7 @@ addIntervalsForSpills(const LiveInterval &li, Slot, LdSlot, isLoad, isLoadSS, DefIsReMat, CanDelete, vrm, rc, ReMatIds, loopInfo, SpillMBBs, SpillIdxes, RestoreMBBs, RestoreIdxes, - MBBVRegsMap, NewLIs, SSWeight); + MBBVRegsMap, NewLIs); } // Insert spills / restores if we are splitting. @@ -2015,8 +1990,6 @@ addIntervalsForSpills(const LiveInterval &li, if (NeedStackSlot) { int Id = SpillMBBs.find_first(); while (Id != -1) { - MachineBasicBlock *MBB = mf_->getBlockNumbered(Id); - unsigned loopDepth = loopInfo->getLoopDepth(MBB); std::vector<SRInfo> &spills = SpillIdxes[Id]; for (unsigned i = 0, e = spills.size(); i != e; ++i) { int index = spills[i].index; @@ -2073,10 +2046,6 @@ addIntervalsForSpills(const LiveInterval &li, if (isKill) AddedKill.insert(&nI); } - - // Update spill slot weight. - if (!isReMat) - SSWeight += getSpillWeight(true, false, loopDepth); } Id = SpillMBBs.find_next(Id); } @@ -2084,9 +2053,6 @@ addIntervalsForSpills(const LiveInterval &li, int Id = RestoreMBBs.find_first(); while (Id != -1) { - MachineBasicBlock *MBB = mf_->getBlockNumbered(Id); - unsigned loopDepth = loopInfo->getLoopDepth(MBB); - std::vector<SRInfo> &restores = RestoreIdxes[Id]; for (unsigned i = 0, e = restores.size(); i != e; ++i) { int index = restores[i].index; @@ -2148,10 +2114,6 @@ addIntervalsForSpills(const LiveInterval &li, nI.removeRange(getLoadIndex(index), getUseIndex(index)+1); else vrm.addRestorePoint(VReg, MI); - - // Update spill slot weight. - if (!isReMat) - SSWeight += getSpillWeight(false, true, loopDepth); } Id = RestoreMBBs.find_next(Id); } diff --git a/lib/CodeGen/LiveStackAnalysis.cpp b/lib/CodeGen/LiveStackAnalysis.cpp index 2baf699..c68a2d9 100644 --- a/lib/CodeGen/LiveStackAnalysis.cpp +++ b/lib/CodeGen/LiveStackAnalysis.cpp @@ -32,7 +32,8 @@ void LiveStacks::getAnalysisUsage(AnalysisUsage &AU) const { void LiveStacks::releaseMemory() { // Release VNInfo memroy regions after all VNInfo objects are dtor'd. VNInfoAllocator.Reset(); - s2iMap.clear(); + S2IMap.clear(); + S2RCMap.clear(); } bool LiveStacks::runOnMachineFunction(MachineFunction &) { @@ -42,10 +43,15 @@ bool LiveStacks::runOnMachineFunction(MachineFunction &) { } /// print - Implement the dump method. -void LiveStacks::print(std::ostream &O, const Module* ) const { +void LiveStacks::print(std::ostream &O, const Module*) const { O << "********** INTERVALS **********\n"; for (const_iterator I = begin(), E = end(); I != E; ++I) { I->second.print(O); - O << "\n"; + int Slot = I->first; + const TargetRegisterClass *RC = getIntervalRegClass(Slot); + if (RC) + O << " [" << RC->getName() << "]\n"; + else + O << " [Unknown]\n"; } } diff --git a/lib/CodeGen/PreAllocSplitting.cpp b/lib/CodeGen/PreAllocSplitting.cpp index c4bda48..97d4728 100644 --- a/lib/CodeGen/PreAllocSplitting.cpp +++ b/lib/CodeGen/PreAllocSplitting.cpp @@ -339,7 +339,7 @@ int PreAllocSplitting::CreateSpillStackSlot(unsigned Reg, } // Create live interval for stack slot. - CurrSLI = &LSs->getOrCreateInterval(SS); + CurrSLI = &LSs->getOrCreateInterval(SS, RC); if (CurrSLI->hasAtLeastOneValue()) CurrSValNo = CurrSLI->getValNumInfo(0); else @@ -926,8 +926,7 @@ MachineInstr* PreAllocSplitting::FoldSpill(unsigned vreg, if (I != IntervalSSMap.end()) { SS = I->second; } else { - SS = MFI->CreateStackObject(RC->getSize(), RC->getAlignment()); - + SS = MFI->CreateStackObject(RC->getSize(), RC->getAlignment()); } MachineInstr* FMI = TII->foldMemoryOperand(*MBB->getParent(), @@ -939,7 +938,7 @@ MachineInstr* PreAllocSplitting::FoldSpill(unsigned vreg, ++NumFolds; IntervalSSMap[vreg] = SS; - CurrSLI = &LSs->getOrCreateInterval(SS); + CurrSLI = &LSs->getOrCreateInterval(SS, RC); if (CurrSLI->hasAtLeastOneValue()) CurrSValNo = CurrSLI->getValNumInfo(0); else diff --git a/lib/CodeGen/RegAllocLinearScan.cpp b/lib/CodeGen/RegAllocLinearScan.cpp index 83c1cbb..b5f581c 100644 --- a/lib/CodeGen/RegAllocLinearScan.cpp +++ b/lib/CodeGen/RegAllocLinearScan.cpp @@ -216,6 +216,18 @@ namespace { } void finalizeRegUses() { +#ifndef NDEBUG + // Verify all the registers are "freed". + bool Error = false; + for (unsigned i = 0, e = tri_->getNumRegs(); i != e; ++i) { + if (regUse_[i] != 0) { + cerr << tri_->getName(i) << " is still in use!\n"; + Error = true; + } + } + if (Error) + abort(); +#endif regUse_.clear(); regUseBackUp_.clear(); } @@ -514,6 +526,13 @@ void RALinScan::linearScan() } DOUT << *vrm_; + + // Look for physical registers that end up not being allocated even though + // register allocator had to spill other registers in its register class. + if (ls_->getNumIntervals() == 0) + return; + if (!vrm_->FindUnusedRegisters(tri_, li_)) + return; } /// processActiveIntervals - expire old intervals and move non-overlapping ones @@ -630,9 +649,9 @@ void RALinScan::updateSpillWeights(std::vector<float> &Weights, // bl should get the same spill weight otherwise it will be choosen // as a spill candidate since spilling bh doesn't make ebx available. for (unsigned i = 0, e = Supers.size(); i != e; ++i) { - for (const unsigned *sr = tri_->getSubRegisters(Supers[i]); *sr; ++sr) - if (!Processed.count(*sr)) - Weights[*sr] += weight; + for (const unsigned *sr = tri_->getSubRegisters(Supers[i]); *sr; ++sr) + if (!Processed.count(*sr)) + Weights[*sr] += weight; } } @@ -658,13 +677,14 @@ static void RevertVectorIteratorsTo(RALinScan::IntervalPtrs &V, unsigned Point){ /// addStackInterval - Create a LiveInterval for stack if the specified live /// interval has been spilled. static void addStackInterval(LiveInterval *cur, LiveStacks *ls_, - LiveIntervals *li_, float &Weight, - VirtRegMap &vrm_) { + LiveIntervals *li_, + MachineRegisterInfo* mri_, VirtRegMap &vrm_) { int SS = vrm_.getStackSlot(cur->reg); if (SS == VirtRegMap::NO_STACK_SLOT) return; - LiveInterval &SI = ls_->getOrCreateInterval(SS); - SI.weight += Weight; + + const TargetRegisterClass *RC = mri_->getRegClass(cur->reg); + LiveInterval &SI = ls_->getOrCreateInterval(SS, RC); VNInfo *VNI; if (SI.hasAtLeastOneValue()) @@ -679,10 +699,10 @@ static void addStackInterval(LiveInterval *cur, LiveStacks *ls_, /// getConflictWeight - Return the number of conflicts between cur /// live interval and defs and uses of Reg weighted by loop depthes. -static float getConflictWeight(LiveInterval *cur, unsigned Reg, - LiveIntervals *li_, - MachineRegisterInfo *mri_, - const MachineLoopInfo *loopInfo) { +static +float getConflictWeight(LiveInterval *cur, unsigned Reg, LiveIntervals *li_, + MachineRegisterInfo *mri_, + const MachineLoopInfo *loopInfo) { float Conflicts = 0; for (MachineRegisterInfo::reg_iterator I = mri_->reg_begin(Reg), E = mri_->reg_end(); I != E; ++I) { @@ -1072,12 +1092,11 @@ void RALinScan::assignRegOrStackSlotAtInterval(LiveInterval* cur) // linearscan. if (cur->weight != HUGE_VALF && cur->weight <= minWeight) { DOUT << "\t\t\tspilling(c): " << *cur << '\n'; - float SSWeight; SmallVector<LiveInterval*, 8> spillIs; std::vector<LiveInterval*> added = - li_->addIntervalsForSpills(*cur, spillIs, loopInfo, *vrm_, SSWeight); + li_->addIntervalsForSpills(*cur, spillIs, loopInfo, *vrm_); std::sort(added.begin(), added.end(), LISorter()); - addStackInterval(cur, ls_, li_, SSWeight, *vrm_); + addStackInterval(cur, ls_, li_, mri_, *vrm_); if (added.empty()) return; // Early exit if all spills were folded. @@ -1149,10 +1168,9 @@ void RALinScan::assignRegOrStackSlotAtInterval(LiveInterval* cur) spillIs.pop_back(); DOUT << "\t\t\tspilling(a): " << *sli << '\n'; earliestStart = std::min(earliestStart, sli->beginNumber()); - float SSWeight; std::vector<LiveInterval*> newIs = - li_->addIntervalsForSpills(*sli, spillIs, loopInfo, *vrm_, SSWeight); - addStackInterval(sli, ls_, li_, SSWeight, *vrm_); + li_->addIntervalsForSpills(*sli, spillIs, loopInfo, *vrm_); + addStackInterval(sli, ls_, li_, mri_, *vrm_); std::copy(newIs.begin(), newIs.end(), std::back_inserter(added)); spilled.insert(sli->reg); } diff --git a/lib/CodeGen/RegAllocPBQP.cpp b/lib/CodeGen/RegAllocPBQP.cpp index 748fae4..8cdf4fa 100644 --- a/lib/CodeGen/RegAllocPBQP.cpp +++ b/lib/CodeGen/RegAllocPBQP.cpp @@ -165,7 +165,7 @@ namespace { //! \brief Adds a stack interval if the given live interval has been //! spilled. Used to support stack slot coloring. - void addStackInterval(const LiveInterval *spilled, float &weight); + void addStackInterval(const LiveInterval *spilled,MachineRegisterInfo* mri); //! \brief Given a solved PBQP problem maps this solution back to a register //! assignment. @@ -637,14 +637,15 @@ pbqp* PBQPRegAlloc::constructPBQPProblem() { return solver; } -void PBQPRegAlloc::addStackInterval(const LiveInterval *spilled, float &weight) { +void PBQPRegAlloc::addStackInterval(const LiveInterval *spilled, + MachineRegisterInfo* mri) { int stackSlot = vrm->getStackSlot(spilled->reg); if (stackSlot == VirtRegMap::NO_STACK_SLOT) return; - LiveInterval &stackInterval = lss->getOrCreateInterval(stackSlot); - stackInterval.weight += weight; + const TargetRegisterClass *RC = mri->getRegClass(spilled->reg); + LiveInterval &stackInterval = lss->getOrCreateInterval(stackSlot, RC); VNInfo *vni; if (stackInterval.getNumValNums() != 0) @@ -688,16 +689,13 @@ bool PBQPRegAlloc::mapPBQPToRegAlloc(pbqp *problem) { // of allocation vregIntervalsToAlloc.erase(&lis->getInterval(virtReg)); - float ssWeight; - // Insert spill ranges for this live range const LiveInterval *spillInterval = node2LI[node]; double oldSpillWeight = spillInterval->weight; SmallVector<LiveInterval*, 8> spillIs; std::vector<LiveInterval*> newSpills = - lis->addIntervalsForSpills(*spillInterval, spillIs, loopInfo, *vrm, - ssWeight); - addStackInterval(spillInterval, ssWeight); + lis->addIntervalsForSpills(*spillInterval, spillIs, loopInfo, *vrm); + addStackInterval(spillInterval, mri); DOUT << "VREG " << virtReg << " -> SPILLED (Cost: " << oldSpillWeight << ", New vregs: "; diff --git a/lib/CodeGen/StackSlotColoring.cpp b/lib/CodeGen/StackSlotColoring.cpp index 4fedc1a..139d12b 100644 --- a/lib/CodeGen/StackSlotColoring.cpp +++ b/lib/CodeGen/StackSlotColoring.cpp @@ -12,9 +12,13 @@ //===----------------------------------------------------------------------===// #define DEBUG_TYPE "stackcoloring" +#include "VirtRegMap.h" #include "llvm/CodeGen/Passes.h" +#include "llvm/CodeGen/LiveIntervalAnalysis.h" #include "llvm/CodeGen/LiveStackAnalysis.h" #include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineLoopInfo.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/PseudoSourceValue.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Compiler.h" @@ -32,21 +36,34 @@ DisableSharing("no-stack-slot-sharing", cl::init(false), cl::Hidden, cl::desc("Suppress slot sharing during stack coloring")); +static cl::opt<bool> +ColorWithRegs("-color-ss-with-regs", + cl::init(false), cl::Hidden, + cl::desc("Color stack slots with free registers")); + + static cl::opt<int> DCELimit("ssc-dce-limit", cl::init(-1), cl::Hidden); -STATISTIC(NumEliminated, "Number of stack slots eliminated due to coloring"); -STATISTIC(NumDeadAccesses, - "Number of trivially dead stack accesses eliminated"); +STATISTIC(NumEliminated, "Number of stack slots eliminated due to coloring"); +STATISTIC(NumDead, "Number of trivially dead stack accesses eliminated"); +STATISTIC(NumRegRepl, "Number of stack slot refs replaced with reg refs"); namespace { class VISIBILITY_HIDDEN StackSlotColoring : public MachineFunctionPass { LiveStacks* LS; + VirtRegMap* VRM; MachineFrameInfo *MFI; + MachineRegisterInfo *MRI; const TargetInstrInfo *TII; + const TargetRegisterInfo *TRI; + const MachineLoopInfo *loopInfo; // SSIntervals - Spill slot intervals. std::vector<LiveInterval*> SSIntervals; + // SSRefs - Keep a list of frame index references for each spill slot. + SmallVector<SmallVector<MachineInstr*, 8>, 16> SSRefs; + // OrigAlignments - Alignments of stack objects before coloring. SmallVector<unsigned, 16> OrigAlignments; @@ -66,7 +83,7 @@ namespace { BitVector UsedColors; // Assignments - Color to intervals mapping. - SmallVector<SmallVector<LiveInterval*,4>,16> Assignments; + SmallVector<SmallVector<LiveInterval*,4>, 16> Assignments; public: static char ID; // Pass identification @@ -74,8 +91,10 @@ namespace { virtual void getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired<LiveStacks>(); - - AU.addPreservedID(MachineLoopInfoID); + AU.addRequired<VirtRegMap>(); + AU.addPreserved<VirtRegMap>(); + AU.addRequired<MachineLoopInfo>(); + AU.addPreserved<MachineLoopInfo>(); AU.addPreservedID(MachineDominatorsID); MachineFunctionPass::getAnalysisUsage(AU); } @@ -86,11 +105,20 @@ namespace { } private: - bool InitializeSlots(); + void InitializeSlots(); + void ScanForSpillSlotRefs(MachineFunction &MF); bool OverlapWithAssignments(LiveInterval *li, int Color) const; int ColorSlot(LiveInterval *li); bool ColorSlots(MachineFunction &MF); - bool removeDeadStores(MachineBasicBlock* MBB); + bool ColorSlotsWithFreeRegs(SmallVector<int, 16> &SlotMapping, + SmallVector<SmallVector<int, 4>, 16> &RevMap, + BitVector &SlotIsReg); + void RewriteInstruction(MachineInstr *MI, int OldFI, int NewFI, + MachineFunction &MF); + void UnfoldAndRewriteInstruction(MachineInstr *MI, int OldFI, + unsigned Reg, MachineFunction &MF); + bool AllMemRefsCanBeUnfolded(int SS); + bool RemoveDeadStores(MachineBasicBlock* MBB); }; } // end anonymous namespace @@ -113,12 +141,39 @@ namespace { }; } +/// ScanForSpillSlotRefs - Scan all the machine instructions for spill slot +/// references and update spill slot weights. +void StackSlotColoring::ScanForSpillSlotRefs(MachineFunction &MF) { + SSRefs.resize(MFI->getObjectIndexEnd()); + + // FIXME: Need the equivalent of MachineRegisterInfo for frameindex operands. + for (MachineFunction::iterator MBBI = MF.begin(), E = MF.end(); + MBBI != E; ++MBBI) { + MachineBasicBlock *MBB = &*MBBI; + unsigned loopDepth = loopInfo->getLoopDepth(MBB); + for (MachineBasicBlock::iterator MII = MBB->begin(), EE = MBB->end(); + MII != EE; ++MII) { + MachineInstr *MI = &*MII; + for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { + MachineOperand &MO = MI->getOperand(i); + if (!MO.isFI()) + continue; + int FI = MO.getIndex(); + if (FI < 0) + continue; + if (!LS->hasInterval(FI)) + continue; + LiveInterval &li = LS->getInterval(FI); + li.weight += LiveIntervals::getSpillWeight(false, true, loopDepth); + SSRefs[FI].push_back(MI); + } + } + } +} + /// InitializeSlots - Process all spill stack slot liveintervals and add them /// to a sorted (by weight) list. -bool StackSlotColoring::InitializeSlots() { - if (LS->getNumIntervals() < 2) - return false; - +void StackSlotColoring::InitializeSlots() { int LastFI = MFI->getObjectIndexEnd(); OrigAlignments.resize(LastFI); OrigSizes.resize(LastFI); @@ -127,8 +182,10 @@ bool StackSlotColoring::InitializeSlots() { Assignments.resize(LastFI); // Gather all spill slots into a list. + DOUT << "Spill slot intervals:\n"; for (LiveStacks::iterator i = LS->begin(), e = LS->end(); i != e; ++i) { LiveInterval &li = i->second; + DEBUG(li.dump()); int FI = li.getStackSlotIndex(); if (MFI->isDeadObjectIndex(FI)) continue; @@ -137,13 +194,13 @@ bool StackSlotColoring::InitializeSlots() { OrigSizes[FI] = MFI->getObjectSize(FI); AllColors.set(FI); } + DOUT << '\n'; // Sort them by weight. std::stable_sort(SSIntervals.begin(), SSIntervals.end(), IntervalSorter()); // Get first "color". NextColor = AllColors.find_first(); - return true; } /// OverlapWithAssignments - Return true if LiveInterval overlaps with any @@ -159,6 +216,83 @@ StackSlotColoring::OverlapWithAssignments(LiveInterval *li, int Color) const { return false; } +/// ColorSlotsWithFreeRegs - If there are any free registers available, try +/// replacing spill slots references with registers instead. +bool +StackSlotColoring::ColorSlotsWithFreeRegs(SmallVector<int, 16> &SlotMapping, + SmallVector<SmallVector<int, 4>, 16> &RevMap, + BitVector &SlotIsReg) { + if (!ColorWithRegs || !VRM->HasUnusedRegisters()) + return false; + + bool Changed = false; + DOUT << "Assigning unused registers to spill slots:\n"; + for (unsigned i = 0, e = SSIntervals.size(); i != e; ++i) { + LiveInterval *li = SSIntervals[i]; + int SS = li->getStackSlotIndex(); + if (!UsedColors[SS]) + continue; + // Get the largest common sub- register class of all the stack slots that + // are colored to this stack slot. + const TargetRegisterClass *RC = 0; + for (unsigned j = 0, ee = RevMap[SS].size(); j != ee; ++j) { + int RSS = RevMap[SS][j]; + const TargetRegisterClass *RRC = LS->getIntervalRegClass(RSS); + if (!RC) + RC = RRC; + else + RC = getCommonSubClass(RC, RRC); + } + + // If it's not colored to another stack slot, try coloring it + // to a "free" register. + if (!RC) + continue; + unsigned Reg = VRM->getFirstUnusedRegister(RC); + if (!Reg) + continue; + bool IsSafe = true; + for (unsigned j = 0, ee = RevMap[SS].size(); j != ee; ++j) { + int RSS = RevMap[SS][j]; + if (!AllMemRefsCanBeUnfolded(RSS)) { + IsSafe = false; + break; + } + } + if (!IsSafe) + // Try color the next spill slot. + continue; + + DOUT << "Assigning fi#" << SS << " to " << TRI->getName(Reg) + << ", which in turn means...\n"; + // Register and its sub-registers are no longer free. + VRM->setRegisterUsed(Reg); + // If reg is a callee-saved register, it will have to be spilled in + // the prologue. + MRI->setPhysRegUsed(Reg); + for (const unsigned *AS = TRI->getAliasSet(Reg); *AS; ++AS) { + VRM->setRegisterUsed(*AS); + MRI->setPhysRegUsed(*AS); + } + // This spill slot is dead after the rewrites + MFI->RemoveStackObject(SS); + + // Remember all these FI references will have to be unfolded. + for (unsigned j = 0, ee = RevMap[SS].size(); j != ee; ++j) { + int RSS = RevMap[SS][j]; + DOUT << " Assigning fi#" << RSS << " to " << TRI->getName(Reg) << '\n'; + SlotMapping[RSS] = Reg; + SlotIsReg.set(RSS); + } + + ++NumEliminated; + Changed = true; + } + DOUT << '\n'; + + return Changed; +} + /// ColorSlot - Assign a "color" (stack slot) to the specified stack slot. /// int StackSlotColoring::ColorSlot(LiveInterval *li) { @@ -207,56 +341,61 @@ int StackSlotColoring::ColorSlot(LiveInterval *li) { /// operands in the function. bool StackSlotColoring::ColorSlots(MachineFunction &MF) { unsigned NumObjs = MFI->getObjectIndexEnd(); - std::vector<int> SlotMapping(NumObjs, -1); + SmallVector<int, 16> SlotMapping(NumObjs, -1); + SmallVector<float, 16> SlotWeights(NumObjs, 0.0); + SmallVector<SmallVector<int, 4>, 16> RevMap(NumObjs); + BitVector SlotIsReg(NumObjs); + BitVector UsedColors(NumObjs); + DOUT << "Color spill slot intervals:\n"; bool Changed = false; for (unsigned i = 0, e = SSIntervals.size(); i != e; ++i) { LiveInterval *li = SSIntervals[i]; int SS = li->getStackSlotIndex(); int NewSS = ColorSlot(li); + assert(NewSS >= 0 && "Stack coloring failed?"); SlotMapping[SS] = NewSS; + RevMap[NewSS].push_back(SS); + SlotWeights[NewSS] += li->weight; + UsedColors.set(NewSS); Changed |= (SS != NewSS); } + DOUT << "\nSpill slots after coloring:\n"; + for (unsigned i = 0, e = SSIntervals.size(); i != e; ++i) { + LiveInterval *li = SSIntervals[i]; + int SS = li->getStackSlotIndex(); + li->weight = SlotWeights[SS]; + } + // Sort them by new weight. + std::stable_sort(SSIntervals.begin(), SSIntervals.end(), IntervalSorter()); + +#ifndef NDEBUG + for (unsigned i = 0, e = SSIntervals.size(); i != e; ++i) + DEBUG(SSIntervals[i]->dump()); + DOUT << '\n'; +#endif + + // Can we "color" a stack slot with a unused register? + Changed |= ColorSlotsWithFreeRegs(SlotMapping, RevMap, SlotIsReg); + if (!Changed) return false; // Rewrite all MO_FrameIndex operands. - // FIXME: Need the equivalent of MachineRegisterInfo for frameindex operands. - for (MachineFunction::iterator MBB = MF.begin(), E = MF.end(); - MBB != E; ++MBB) { - for (MachineBasicBlock::iterator MII = MBB->begin(), EE = MBB->end(); - MII != EE; ++MII) { - MachineInstr &MI = *MII; - for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) { - MachineOperand &MO = MI.getOperand(i); - if (!MO.isFI()) - continue; - int FI = MO.getIndex(); - if (FI < 0) - continue; - int NewFI = SlotMapping[FI]; - if (NewFI == -1) - continue; - MO.setIndex(NewFI); - - // Update the MachineMemOperand for the new memory location. - // FIXME: We need a better method of managing these too. - SmallVector<MachineMemOperand, 2> MMOs(MI.memoperands_begin(), - MI.memoperands_end()); - MI.clearMemOperands(MF); - const Value *OldSV = PseudoSourceValue::getFixedStack(FI); - for (unsigned i = 0, e = MMOs.size(); i != e; ++i) { - if (MMOs[i].getValue() == OldSV) { - MachineMemOperand MMO(PseudoSourceValue::getFixedStack(NewFI), - MMOs[i].getFlags(), MMOs[i].getOffset(), - MMOs[i].getSize(), MMOs[i].getAlignment()); - MI.addMemOperand(MF, MMO); - } else - MI.addMemOperand(MF, MMOs[i]); - } - } - } + for (unsigned SS = 0, SE = SSRefs.size(); SS != SE; ++SS) { + bool isReg = SlotIsReg[SS]; + int NewFI = SlotMapping[SS]; + if (NewFI == -1 || (NewFI == (int)SS && !isReg)) + continue; + + SmallVector<MachineInstr*, 8> &RefMIs = SSRefs[SS]; + for (unsigned i = 0, e = RefMIs.size(); i != e; ++i) + if (isReg) + // Rewrite to use a register instead. + UnfoldAndRewriteInstruction(RefMIs[i], SS, NewFI, MF); + else + RewriteInstruction(RefMIs[i], SS, NewFI, MF); } // Delete unused stack slots. @@ -269,12 +408,77 @@ bool StackSlotColoring::ColorSlots(MachineFunction &MF) { return true; } -/// removeDeadStores - Scan through a basic block and look for loads followed +/// AllMemRefsCanBeUnfolded - Return true if all references of the specified +/// spill slot index can be unfolded. +bool StackSlotColoring::AllMemRefsCanBeUnfolded(int SS) { + SmallVector<MachineInstr*, 8> &RefMIs = SSRefs[SS]; + for (unsigned i = 0, e = RefMIs.size(); i != e; ++i) { + MachineInstr *MI = RefMIs[i]; + if (!TII->getOpcodeAfterMemoryUnfold(MI->getOpcode(), false, false)) + return false; + for (unsigned j = 0, ee = MI->getNumOperands(); j != ee; ++j) { + MachineOperand &MO = MI->getOperand(j); + if (MO.isFI() && MO.getIndex() != SS) + // If it uses another frameindex, we can, currently* unfold it. + return false; + } + } + return true; +} + +/// RewriteInstruction - Rewrite specified instruction by replacing references +/// to old frame index with new one. +void StackSlotColoring::RewriteInstruction(MachineInstr *MI, int OldFI, + int NewFI, MachineFunction &MF) { + for (unsigned i = 0, ee = MI->getNumOperands(); i != ee; ++i) { + MachineOperand &MO = MI->getOperand(i); + if (!MO.isFI()) + continue; + int FI = MO.getIndex(); + if (FI != OldFI) + continue; + MO.setIndex(NewFI); + } + + // Update the MachineMemOperand for the new memory location. + // FIXME: We need a better method of managing these too. + SmallVector<MachineMemOperand, 2> MMOs(MI->memoperands_begin(), + MI->memoperands_end()); + MI->clearMemOperands(MF); + const Value *OldSV = PseudoSourceValue::getFixedStack(OldFI); + for (unsigned i = 0, ee = MMOs.size(); i != ee; ++i) { + if (MMOs[i].getValue() != OldSV) + MI->addMemOperand(MF, MMOs[i]); + else { + MachineMemOperand MMO(PseudoSourceValue::getFixedStack(NewFI), + MMOs[i].getFlags(), MMOs[i].getOffset(), + MMOs[i].getSize(), MMOs[i].getAlignment()); + MI->addMemOperand(MF, MMO); + } + } +} + +/// UnfoldAndRewriteInstruction - Rewrite specified instruction by unfolding +/// folded memory references and replacing those references with register +/// references instead. +void StackSlotColoring::UnfoldAndRewriteInstruction(MachineInstr *MI, int OldFI, + unsigned Reg, + MachineFunction &MF) { + MachineBasicBlock *MBB = MI->getParent(); + SmallVector<MachineInstr*, 4> NewMIs; + bool Success = TII->unfoldMemoryOperand(MF, MI, Reg, false, false, NewMIs); + assert(Success && "Failed to unfold!"); + MBB->insert(MI, NewMIs[0]); + MBB->erase(MI); + ++NumRegRepl; +} + +/// RemoveDeadStores - Scan through a basic block and look for loads followed /// by stores. If they're both using the same stack slot, then the store is /// definitely dead. This could obviously be much more aggressive (consider /// pairs with instructions between them), but such extensions might have a /// considerable compile time impact. -bool StackSlotColoring::removeDeadStores(MachineBasicBlock* MBB) { +bool StackSlotColoring::RemoveDeadStores(MachineBasicBlock* MBB) { // FIXME: This could be much more aggressive, but we need to investigate // the compile time impact of doing so. bool changed = false; @@ -283,7 +487,7 @@ bool StackSlotColoring::removeDeadStores(MachineBasicBlock* MBB) { for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E; ++I) { - if (DCELimit != -1 && (int)NumDeadAccesses >= DCELimit) + if (DCELimit != -1 && (int)NumDead >= DCELimit) break; MachineBasicBlock::iterator NextMI = next(I); @@ -296,11 +500,11 @@ bool StackSlotColoring::removeDeadStores(MachineBasicBlock* MBB) { if (!(StoreReg = TII->isStoreToStackSlot(NextMI, SecondSS))) continue; if (FirstSS != SecondSS || LoadReg != StoreReg || FirstSS == -1) continue; - ++NumDeadAccesses; + ++NumDead; changed = true; if (NextMI->findRegisterUseOperandIdx(LoadReg, true, 0) != -1) { - ++NumDeadAccesses; + ++NumDead; toErase.push_back(I); } @@ -320,15 +524,32 @@ bool StackSlotColoring::runOnMachineFunction(MachineFunction &MF) { DOUT << "********** Stack Slot Coloring **********\n"; MFI = MF.getFrameInfo(); + MRI = &MF.getRegInfo(); TII = MF.getTarget().getInstrInfo(); + TRI = MF.getTarget().getRegisterInfo(); LS = &getAnalysis<LiveStacks>(); + VRM = &getAnalysis<VirtRegMap>(); + loopInfo = &getAnalysis<MachineLoopInfo>(); bool Changed = false; - if (InitializeSlots()) - Changed = ColorSlots(MF); + + unsigned NumSlots = LS->getNumIntervals(); + if (NumSlots < 2) { + if (NumSlots == 0 || !VRM->HasUnusedRegisters()) + // Nothing to do! + return false; + } + + // Gather spill slot references + ScanForSpillSlotRefs(MF); + InitializeSlots(); + Changed = ColorSlots(MF); NextColor = -1; SSIntervals.clear(); + for (unsigned i = 0, e = SSRefs.size(); i != e; ++i) + SSRefs[i].clear(); + SSRefs.clear(); OrigAlignments.clear(); OrigSizes.clear(); AllColors.clear(); @@ -339,7 +560,7 @@ bool StackSlotColoring::runOnMachineFunction(MachineFunction &MF) { if (Changed) { for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) - Changed |= removeDeadStores(I); + Changed |= RemoveDeadStores(I); } return Changed; diff --git a/lib/CodeGen/VirtRegMap.cpp b/lib/CodeGen/VirtRegMap.cpp index cb0f764..f2f6ab0 100644 --- a/lib/CodeGen/VirtRegMap.cpp +++ b/lib/CodeGen/VirtRegMap.cpp @@ -19,6 +19,7 @@ #define DEBUG_TYPE "virtregmap" #include "VirtRegMap.h" #include "llvm/Function.h" +#include "llvm/CodeGen/LiveIntervalAnalysis.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" @@ -201,6 +202,41 @@ void VirtRegMap::RemoveMachineInstrFromMaps(MachineInstr *MI) { EmergencySpillMap.erase(MI); } +/// FindUnusedRegisters - Gather a list of allocatable registers that +/// have not been allocated to any virtual register. +bool VirtRegMap::FindUnusedRegisters(const TargetRegisterInfo *TRI, + LiveIntervals* LIs) { + unsigned NumRegs = TRI->getNumRegs(); + UnusedRegs.reset(); + UnusedRegs.resize(NumRegs); + + BitVector Used(NumRegs); + for (unsigned i = TargetRegisterInfo::FirstVirtualRegister, + e = MF->getRegInfo().getLastVirtReg(); i <= e; ++i) + if (Virt2PhysMap[i] != (unsigned)VirtRegMap::NO_PHYS_REG) + Used.set(Virt2PhysMap[i]); + + BitVector Allocatable = TRI->getAllocatableSet(*MF); + bool AnyUnused = false; + for (unsigned Reg = 1; Reg < NumRegs; ++Reg) { + if (Allocatable[Reg] && !Used[Reg] && !LIs->hasInterval(Reg)) { + bool ReallyUnused = true; + for (const unsigned *AS = TRI->getAliasSet(Reg); *AS; ++AS) { + if (Used[*AS] || LIs->hasInterval(*AS)) { + ReallyUnused = false; + break; + } + } + if (ReallyUnused) { + AnyUnused = true; + UnusedRegs.set(Reg); + } + } + } + + return AnyUnused; +} + void VirtRegMap::print(std::ostream &OS, const Module* M) const { const TargetRegisterInfo* TRI = MF->getTarget().getRegisterInfo(); diff --git a/lib/CodeGen/VirtRegMap.h b/lib/CodeGen/VirtRegMap.h index 2e9c899..91c8322 100644 --- a/lib/CodeGen/VirtRegMap.h +++ b/lib/CodeGen/VirtRegMap.h @@ -20,6 +20,7 @@ #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/Target/TargetRegisterInfo.h" #include "llvm/ADT/BitVector.h" +#include "llvm/ADT/DenseMap.h" #include "llvm/ADT/IndexedMap.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" @@ -27,6 +28,7 @@ #include <map> namespace llvm { + class LiveIntervals; class MachineInstr; class MachineFunction; class TargetInstrInfo; @@ -122,6 +124,9 @@ namespace llvm { /// the register is implicitly defined. BitVector ImplicitDefed; + /// UnusedRegs - A list of physical registers that have not been used. + BitVector UnusedRegs; + VirtRegMap(const VirtRegMap&); // DO NOT IMPLEMENT void operator=(const VirtRegMap&); // DO NOT IMPLEMENT @@ -277,8 +282,10 @@ namespace llvm { /// @brief records the specified MachineInstr as a spill point for virtReg. void addSpillPoint(unsigned virtReg, bool isKill, MachineInstr *Pt) { - if (SpillPt2VirtMap.find(Pt) != SpillPt2VirtMap.end()) - SpillPt2VirtMap[Pt].push_back(std::make_pair(virtReg, isKill)); + std::map<MachineInstr*, std::vector<std::pair<unsigned,bool> > >::iterator + I = SpillPt2VirtMap.find(Pt); + if (I != SpillPt2VirtMap.end()) + I->second.push_back(std::make_pair(virtReg, isKill)); else { std::vector<std::pair<unsigned,bool> > Virts; Virts.push_back(std::make_pair(virtReg, isKill)); @@ -289,7 +296,7 @@ namespace llvm { /// @brief - transfer spill point information from one instruction to /// another. void transferSpillPts(MachineInstr *Old, MachineInstr *New) { - std::map<MachineInstr*,std::vector<std::pair<unsigned,bool> > >::iterator + std::map<MachineInstr*, std::vector<std::pair<unsigned,bool> > >::iterator I = SpillPt2VirtMap.find(Old); if (I == SpillPt2VirtMap.end()) return; @@ -315,8 +322,10 @@ namespace llvm { /// @brief records the specified MachineInstr as a restore point for virtReg. void addRestorePoint(unsigned virtReg, MachineInstr *Pt) { - if (RestorePt2VirtMap.find(Pt) != RestorePt2VirtMap.end()) - RestorePt2VirtMap[Pt].push_back(virtReg); + std::map<MachineInstr*, std::vector<unsigned> >::iterator I = + RestorePt2VirtMap.find(Pt); + if (I != RestorePt2VirtMap.end()) + I->second.push_back(virtReg); else { std::vector<unsigned> Virts; Virts.push_back(virtReg); @@ -327,7 +336,7 @@ namespace llvm { /// @brief - transfer restore point information from one instruction to /// another. void transferRestorePts(MachineInstr *Old, MachineInstr *New) { - std::map<MachineInstr*,std::vector<unsigned> >::iterator I = + std::map<MachineInstr*, std::vector<unsigned> >::iterator I = RestorePt2VirtMap.find(Old); if (I == RestorePt2VirtMap.end()) return; @@ -430,6 +439,40 @@ namespace llvm { /// the folded instruction map and spill point map. void RemoveMachineInstrFromMaps(MachineInstr *MI); + /// FindUnusedRegisters - Gather a list of allocatable registers that + /// have not been allocated to any virtual register. + bool FindUnusedRegisters(const TargetRegisterInfo *TRI, + LiveIntervals* LIs); + + /// HasUnusedRegisters - Return true if there are any allocatable registers + /// that have not been allocated to any virtual register. + bool HasUnusedRegisters() const { + return !UnusedRegs.none(); + } + + /// setRegisterUsed - Remember the physical register is now used. + void setRegisterUsed(unsigned Reg) { + UnusedRegs.reset(Reg); + } + + /// isRegisterUnused - Return true if the physical register has not been + /// used. + bool isRegisterUnused(unsigned Reg) const { + return UnusedRegs[Reg]; + } + + /// getFirstUnusedRegister - Return the first physical register that has not + /// been used. + unsigned getFirstUnusedRegister(const TargetRegisterClass *RC) { + int Reg = UnusedRegs.find_first(); + while (Reg != -1) { + if (RC->contains(Reg)) + return (unsigned)Reg; + Reg = UnusedRegs.find_next(Reg); + } + return 0; + } + void print(std::ostream &OS, const Module* M = 0) const; void print(std::ostream *OS) const { if (OS) print(*OS); } void dump() const; |