X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FCodeGen%2FIfConversion.cpp;h=f38fb88575c4e670b5b127a90673ed807d943682;hp=4214ba124252ebe5d6daa6de93c61e203460dc9e;hb=8eab0e68d3e83a820adcac282e20303c9ed3733f;hpb=791e2e0867d1e6fcb9c6f651da89f57043bc6c4a diff --git a/lib/CodeGen/IfConversion.cpp b/lib/CodeGen/IfConversion.cpp index 4214ba12425..f38fb88575c 100644 --- a/lib/CodeGen/IfConversion.cpp +++ b/lib/CodeGen/IfConversion.cpp @@ -11,28 +11,34 @@ // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "ifcvt" -#include "BranchFolding.h" -#include "llvm/Function.h" #include "llvm/CodeGen/Passes.h" -#include "llvm/CodeGen/MachineModuleInfo.h" +#include "BranchFolding.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/CodeGen/LivePhysRegs.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineBranchProbabilityInfo.h" #include "llvm/CodeGen/MachineFunctionPass.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineModuleInfo.h" #include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/CodeGen/TargetSchedule.h" #include "llvm/MC/MCInstrItineraries.h" -#include "llvm/Target/TargetInstrInfo.h" -#include "llvm/Target/TargetLowering.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetRegisterInfo.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/SmallSet.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/ADT/STLExtras.h" +#include "llvm/Target/TargetInstrInfo.h" +#include "llvm/Target/TargetLowering.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/Target/TargetSubtargetInfo.h" + using namespace llvm; +#define DEBUG_TYPE "ifcvt" + // Hidden options for help debugging. static cl::opt IfCvtFnStart("ifcvt-fn-start", cl::init(-1), cl::Hidden); static cl::opt IfCvtFnStop("ifcvt-fn-stop", cl::init(-1), cl::Hidden); @@ -123,7 +129,8 @@ namespace { IsAnalyzed(false), IsEnqueued(false), IsBrAnalyzable(false), HasFallThrough(false), IsUnpredicable(false), CannotBeCopied(false), ClobbersPred(false), NonPredSize(0), - ExtraCost(0), ExtraCost2(0), BB(0), TrueBB(0), FalseBB(0) {} + ExtraCost(0), ExtraCost2(0), BB(nullptr), TrueBB(nullptr), + FalseBB(nullptr) {} }; /// IfcvtToken - Record information about pending if-conversions to attempt: @@ -150,14 +157,18 @@ namespace { /// BBAnalysis - Results of if-conversion feasibility analysis indexed by /// basic block number. std::vector BBAnalysis; + TargetSchedModel SchedModel; - const TargetLowering *TLI; + const TargetLoweringBase *TLI; const TargetInstrInfo *TII; const TargetRegisterInfo *TRI; - const InstrItineraryData *InstrItins; + const MachineBlockFrequencyInfo *MBFI; const MachineBranchProbabilityInfo *MBPI; MachineRegisterInfo *MRI; + LivePhysRegs Redefs; + LivePhysRegs DontKill; + bool PreRegAlloc; bool MadeChange; int FnNum; @@ -167,12 +178,13 @@ namespace { initializeIfConverterPass(*PassRegistry::getPassRegistry()); } - virtual void getAnalysisUsage(AnalysisUsage &AU) const { + void getAnalysisUsage(AnalysisUsage &AU) const override { + AU.addRequired(); AU.addRequired(); MachineFunctionPass::getAnalysisUsage(AU); } - virtual bool runOnMachineFunction(MachineFunction &MF); + bool runOnMachineFunction(MachineFunction &MF) override; private: bool ReverseBranchCondition(BBInfo &BBI); @@ -198,11 +210,9 @@ namespace { void PredicateBlock(BBInfo &BBI, MachineBasicBlock::iterator E, SmallVectorImpl &Cond, - SmallSet &Redefs, - SmallSet *LaterRedefs = 0); + SmallSet *LaterRedefs = nullptr); void CopyAndPredicateBlock(BBInfo &ToBBI, BBInfo &FromBBI, SmallVectorImpl &Cond, - SmallSet &Redefs, bool IgnoreBr = false); void MergeBlocks(BBInfo &ToBBI, BBInfo &FromBBI, bool AddEdges = true); @@ -225,7 +235,7 @@ namespace { // blockAlwaysFallThrough - Block ends without a terminator. bool blockAlwaysFallThrough(BBInfo &BBI) const { - return BBI.IsBrAnalyzable && BBI.TrueBB == NULL; + return BBI.IsBrAnalyzable && BBI.TrueBB == nullptr; } // IfcvtTokenCmp - Used to sort if-conversion candidates. @@ -262,12 +272,17 @@ INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo) INITIALIZE_PASS_END(IfConverter, "if-converter", "If Converter", false, false) bool IfConverter::runOnMachineFunction(MachineFunction &MF) { - TLI = MF.getTarget().getTargetLowering(); - TII = MF.getTarget().getInstrInfo(); - TRI = MF.getTarget().getRegisterInfo(); + TLI = MF.getSubtarget().getTargetLowering(); + TII = MF.getSubtarget().getInstrInfo(); + TRI = MF.getSubtarget().getRegisterInfo(); + MBFI = &getAnalysis(); MBPI = &getAnalysis(); MRI = &MF.getRegInfo(); - InstrItins = MF.getTarget().getInstrItineraryData(); + + const TargetSubtargetInfo &ST = + MF.getTarget().getSubtarget(); + SchedModel.init(*ST.getSchedModel(), &ST, TII); + if (!TII) return false; PreRegAlloc = MRI->isSSA(); @@ -275,14 +290,13 @@ bool IfConverter::runOnMachineFunction(MachineFunction &MF) { bool BFChange = false; if (!PreRegAlloc) { // Tail merge tend to expose more if-conversion opportunities. - BranchFolder BF(true, false); - BFChange = BF.OptimizeFunction(MF, TII, - MF.getTarget().getRegisterInfo(), + BranchFolder BF(true, false, *MBFI, *MBPI); + BFChange = BF.OptimizeFunction(MF, TII, MF.getSubtarget().getRegisterInfo(), getAnalysisIfAvailable()); } DEBUG(dbgs() << "\nIfcvt: function (" << ++FnNum << ") \'" - << MF.getFunction()->getName() << "\'"); + << MF.getName() << "\'"); if (FnNum < IfCvtFnStart || (IfCvtFnStop != -1 && FnNum > IfCvtFnStop)) { DEBUG(dbgs() << " skipped\n"); @@ -409,9 +423,8 @@ bool IfConverter::runOnMachineFunction(MachineFunction &MF) { BBAnalysis.clear(); if (MadeChange && IfCvtBranchFold) { - BranchFolder BF(false, false); - BF.OptimizeFunction(MF, TII, - MF.getTarget().getRegisterInfo(), + BranchFolder BF(false, false, *MBFI, *MBPI); + BF.OptimizeFunction(MF, TII, MF.getSubtarget().getRegisterInfo(), getAnalysisIfAvailable()); } @@ -429,7 +442,7 @@ static MachineBasicBlock *findFalseBlock(MachineBasicBlock *BB, if (SuccBB != TrueBB) return SuccBB; } - return NULL; + return nullptr; } /// ReverseBranchCondition - Reverse the condition of the end of the block @@ -451,7 +464,7 @@ static inline MachineBasicBlock *getNextBlock(MachineBasicBlock *BB) { MachineFunction::iterator I = BB; MachineFunction::iterator E = BB->getParent()->end(); if (++I == E) - return NULL; + return nullptr; return I; } @@ -542,7 +555,7 @@ bool IfConverter::ValidDiamond(BBInfo &TrueBBI, BBInfo &FalseBBI, FT = getNextBlock(FalseBBI.BB); if (TT != FT) return false; - if (TT == NULL && (TrueBBI.IsBrAnalyzable || FalseBBI.IsBrAnalyzable)) + if (!TT && (TrueBBI.IsBrAnalyzable || FalseBBI.IsBrAnalyzable)) return false; if (TrueBBI.BB->pred_size() > 1 || FalseBBI.BB->pred_size() > 1) return false; @@ -632,11 +645,11 @@ void IfConverter::ScanInstructions(BBInfo &BBI) { bool AlreadyPredicated = !BBI.Predicate.empty(); // First analyze the end of BB branches. - BBI.TrueBB = BBI.FalseBB = NULL; + BBI.TrueBB = BBI.FalseBB = nullptr; BBI.BrCond.clear(); BBI.IsBrAnalyzable = !TII->AnalyzeBranch(*BBI.BB, BBI.TrueBB, BBI.FalseBB, BBI.BrCond); - BBI.HasFallThrough = BBI.IsBrAnalyzable && BBI.FalseBB == NULL; + BBI.HasFallThrough = BBI.IsBrAnalyzable && BBI.FalseBB == nullptr; if (BBI.BrCond.size()) { // No false branch. This BB must end with a conditional branch and a @@ -666,32 +679,28 @@ void IfConverter::ScanInstructions(BBInfo &BBI) { bool isPredicated = TII->isPredicated(I); bool isCondBr = BBI.IsBrAnalyzable && I->isConditionalBranch(); - if (!isCondBr) { - if (!isPredicated) { - BBI.NonPredSize++; - unsigned ExtraPredCost = 0; - unsigned NumCycles = TII->getInstrLatency(InstrItins, &*I, - &ExtraPredCost); - if (NumCycles > 1) - BBI.ExtraCost += NumCycles-1; - BBI.ExtraCost2 += ExtraPredCost; - } else if (!AlreadyPredicated) { - // FIXME: This instruction is already predicated before the - // if-conversion pass. It's probably something like a conditional move. - // Mark this block unpredicable for now. - BBI.IsUnpredicable = true; - return; - } + // A conditional branch is not predicable, but it may be eliminated. + if (isCondBr) + continue; + + if (!isPredicated) { + BBI.NonPredSize++; + unsigned ExtraPredCost = TII->getPredicationCost(&*I); + unsigned NumCycles = SchedModel.computeInstrLatency(&*I, false); + if (NumCycles > 1) + BBI.ExtraCost += NumCycles-1; + BBI.ExtraCost2 += ExtraPredCost; + } else if (!AlreadyPredicated) { + // FIXME: This instruction is already predicated before the + // if-conversion pass. It's probably something like a conditional move. + // Mark this block unpredicable for now. + BBI.IsUnpredicable = true; + return; } if (BBI.ClobbersPred && !isPredicated) { // Predicate modification instruction should end the block (except for // already predicated instructions and end of block branches). - if (isCondBr) { - // A conditional branch is not predicable, but it may be eliminated. - continue; - } - // Predicate may have been modified, the subsequent (currently) // unpredicated instructions cannot be correctly predicated. BBI.IsUnpredicable = true; @@ -720,9 +729,9 @@ bool IfConverter::FeasibilityAnalysis(BBInfo &BBI, if (BBI.IsDone || BBI.IsUnpredicable) return false; - // If it is already predicated, check if its predicate subsumes the new - // predicate. - if (BBI.Predicate.size() && !TII->SubsumesPredicate(BBI.Predicate, Pred)) + // If it is already predicated, check if the new predicate subsumes + // its predicate. + if (BBI.Predicate.size() && !TII->SubsumesPredicate(Pred, BBI.Predicate)) return false; if (BBI.BrCond.size()) { @@ -916,7 +925,7 @@ void IfConverter::AnalyzeBlocks(MachineFunction &MF, /// next block). static bool canFallThroughTo(MachineBasicBlock *BB, MachineBasicBlock *ToBB) { MachineFunction::iterator PI = BB; - MachineFunction::iterator I = llvm::next(PI); + MachineFunction::iterator I = std::next(PI); MachineFunction::iterator TI = ToBB; MachineFunction::iterator E = BB->getParent()->end(); while (I != TI) { @@ -933,9 +942,8 @@ static bool canFallThroughTo(MachineBasicBlock *BB, MachineBasicBlock *ToBB) { /// to determine if it can be if-converted. If predecessor is already enqueued, /// dequeue it! void IfConverter::InvalidatePreds(MachineBasicBlock *BB) { - for (MachineBasicBlock::pred_iterator PI = BB->pred_begin(), - E = BB->pred_end(); PI != E; ++PI) { - BBInfo &PBBI = BBAnalysis[(*PI)->getNumber()]; + for (const auto &Predecessor : BB->predecessors()) { + BBInfo &PBBI = BBAnalysis[Predecessor->getNumber()]; if (PBBI.IsDone || PBBI.BB == BB) continue; PBBI.IsAnalyzed = false; @@ -949,78 +957,68 @@ static void InsertUncondBranch(MachineBasicBlock *BB, MachineBasicBlock *ToBB, const TargetInstrInfo *TII) { DebugLoc dl; // FIXME: this is nowhere SmallVector NoCond; - TII->InsertBranch(*BB, ToBB, NULL, NoCond, dl); + TII->InsertBranch(*BB, ToBB, nullptr, NoCond, dl); } /// RemoveExtraEdges - Remove true / false edges if either / both are no longer /// successors. void IfConverter::RemoveExtraEdges(BBInfo &BBI) { - MachineBasicBlock *TBB = NULL, *FBB = NULL; + MachineBasicBlock *TBB = nullptr, *FBB = nullptr; SmallVector Cond; if (!TII->AnalyzeBranch(*BBI.BB, TBB, FBB, Cond)) BBI.BB->CorrectExtraCFGEdges(TBB, FBB, !Cond.empty()); } -/// InitPredRedefs / UpdatePredRedefs - Defs by predicated instructions are -/// modeled as read + write (sort like two-address instructions). These -/// routines track register liveness and add implicit uses to if-converted -/// instructions to conform to the model. -static void InitPredRedefs(MachineBasicBlock *BB, SmallSet &Redefs, - const TargetRegisterInfo *TRI) { - for (MachineBasicBlock::livein_iterator I = BB->livein_begin(), - E = BB->livein_end(); I != E; ++I) { - unsigned Reg = *I; - Redefs.insert(Reg); - for (MCSubRegIterator SubRegs(Reg, TRI); SubRegs.isValid(); ++SubRegs) - Redefs.insert(*SubRegs); - } -} - -static void UpdatePredRedefs(MachineInstr *MI, SmallSet &Redefs, - const TargetRegisterInfo *TRI, - bool AddImpUse = false) { - SmallVector Defs; - for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { - const MachineOperand &MO = MI->getOperand(i); - if (!MO.isReg()) +/// Behaves like LiveRegUnits::StepForward() but also adds implicit uses to all +/// values defined in MI which are not live/used by MI. +static void UpdatePredRedefs(MachineInstr *MI, LivePhysRegs &Redefs) { + for (ConstMIBundleOperands Ops(MI); Ops.isValid(); ++Ops) { + if (!Ops->isReg() || !Ops->isKill()) continue; - unsigned Reg = MO.getReg(); - if (!Reg) + unsigned Reg = Ops->getReg(); + if (Reg == 0) continue; - if (MO.isDef()) - Defs.push_back(Reg); - else if (MO.isKill()) { - Redefs.erase(Reg); - for (MCSubRegIterator SubRegs(Reg, TRI); SubRegs.isValid(); ++SubRegs) - Redefs.erase(*SubRegs); - } + Redefs.removeReg(Reg); } - for (unsigned i = 0, e = Defs.size(); i != e; ++i) { - unsigned Reg = Defs[i]; - if (Redefs.count(Reg)) { - if (AddImpUse) - // Treat predicated update as read + write. - MI->addOperand(MachineOperand::CreateReg(Reg, false/*IsDef*/, - true/*IsImp*/,false/*IsKill*/, - false/*IsDead*/,true/*IsUndef*/)); - } else { - Redefs.insert(Reg); - for (MCSubRegIterator SubRegs(Reg, TRI); SubRegs.isValid(); ++SubRegs) - Redefs.insert(*SubRegs); - } + for (MIBundleOperands Ops(MI); Ops.isValid(); ++Ops) { + if (!Ops->isReg() || !Ops->isDef()) + continue; + unsigned Reg = Ops->getReg(); + if (Reg == 0 || Redefs.contains(Reg)) + continue; + Redefs.addReg(Reg); + + MachineOperand &Op = *Ops; + MachineInstr *MI = Op.getParent(); + MachineInstrBuilder MIB(*MI->getParent()->getParent(), MI); + MIB.addReg(Reg, RegState::Implicit | RegState::Undef); } } -static void UpdatePredRedefs(MachineBasicBlock::iterator I, - MachineBasicBlock::iterator E, - SmallSet &Redefs, - const TargetRegisterInfo *TRI) { - while (I != E) { - UpdatePredRedefs(I, Redefs, TRI); - ++I; +/** + * Remove kill flags from operands with a registers in the @p DontKill set. + */ +static void RemoveKills(MachineInstr &MI, const LivePhysRegs &DontKill) { + for (MIBundleOperands O(&MI); O.isValid(); ++O) { + if (!O->isReg() || !O->isKill()) + continue; + if (DontKill.contains(O->getReg())) + O->setIsKill(false); } } +/** + * Walks a range of machine instructions and removes kill flags for registers + * in the @p DontKill set. + */ +static void RemoveKills(MachineBasicBlock::iterator I, + MachineBasicBlock::iterator E, + const LivePhysRegs &DontKill, + const MCRegisterInfo &MCRI) { + for ( ; I != E; ++I) + RemoveKills(*I, DontKill); +} + /// IfConvertSimple - If convert a simple (split, no rejoin) sub-CFG. /// bool IfConverter::IfConvertSimple(BBInfo &BBI, IfcvtKind Kind) { @@ -1041,23 +1039,37 @@ bool IfConverter::IfConvertSimple(BBInfo &BBI, IfcvtKind Kind) { return false; } + if (CvtBBI->BB->hasAddressTaken()) + // Conservatively abort if-conversion if BB's address is taken. + return false; + if (Kind == ICSimpleFalse) if (TII->ReverseBranchCondition(Cond)) llvm_unreachable("Unable to reverse branch condition!"); // Initialize liveins to the first BB. These are potentiall redefined by // predicated instructions. - SmallSet Redefs; - InitPredRedefs(CvtBBI->BB, Redefs, TRI); - InitPredRedefs(NextBBI->BB, Redefs, TRI); + Redefs.init(TRI); + Redefs.addLiveIns(CvtBBI->BB); + Redefs.addLiveIns(NextBBI->BB); + + // Compute a set of registers which must not be killed by instructions in + // BB1: This is everything live-in to BB2. + DontKill.init(TRI); + DontKill.addLiveIns(NextBBI->BB); if (CvtBBI->BB->pred_size() > 1) { BBI.NonPredSize -= TII->RemoveBranch(*BBI.BB); // Copy instructions in the true block, predicate them, and add them to // the entry block. - CopyAndPredicateBlock(BBI, *CvtBBI, Cond, Redefs); + CopyAndPredicateBlock(BBI, *CvtBBI, Cond); + + // RemoveExtraEdges won't work if the block has an unanalyzable branch, so + // explicitly remove CvtBBI as a successor. + BBI.BB->removeSuccessor(CvtBBI->BB); } else { - PredicateBlock(*CvtBBI, CvtBBI->BB->end(), Cond, Redefs); + RemoveKills(CvtBBI->BB->begin(), CvtBBI->BB->end(), DontKill, *TRI); + PredicateBlock(*CvtBBI, CvtBBI->BB->end(), Cond); // Merge converted block into entry block. BBI.NonPredSize -= TII->RemoveBranch(*BBI.BB); @@ -1093,6 +1105,28 @@ bool IfConverter::IfConvertSimple(BBInfo &BBI, IfcvtKind Kind) { return true; } +/// Scale down weights to fit into uint32_t. NewTrue is the new weight +/// for successor TrueBB, and NewFalse is the new weight for successor +/// FalseBB. +static void ScaleWeights(uint64_t NewTrue, uint64_t NewFalse, + MachineBasicBlock *MBB, + const MachineBasicBlock *TrueBB, + const MachineBasicBlock *FalseBB, + const MachineBranchProbabilityInfo *MBPI) { + uint64_t NewMax = (NewTrue > NewFalse) ? NewTrue : NewFalse; + uint32_t Scale = (NewMax / UINT32_MAX) + 1; + for (MachineBasicBlock::succ_iterator SI = MBB->succ_begin(), + SE = MBB->succ_end(); + SI != SE; ++SI) { + if (*SI == TrueBB) + MBB->setSuccWeight(SI, (uint32_t)(NewTrue / Scale)); + else if (*SI == FalseBB) + MBB->setSuccWeight(SI, (uint32_t)(NewFalse / Scale)); + else + MBB->setSuccWeight(SI, MBPI->getEdgeWeight(MBB, SI) / Scale); + } +} + /// IfConvertTriangle - If convert a triangle sub-CFG. /// bool IfConverter::IfConvertTriangle(BBInfo &BBI, IfcvtKind Kind) { @@ -1114,6 +1148,10 @@ bool IfConverter::IfConvertTriangle(BBInfo &BBI, IfcvtKind Kind) { return false; } + if (CvtBBI->BB->hasAddressTaken()) + // Conservatively abort if-conversion if BB's address is taken. + return false; + if (Kind == ICTriangleFalse || Kind == ICTriangleFRev) if (TII->ReverseBranchCondition(Cond)) llvm_unreachable("Unable to reverse branch condition!"); @@ -1138,20 +1176,38 @@ bool IfConverter::IfConvertTriangle(BBInfo &BBI, IfcvtKind Kind) { // Initialize liveins to the first BB. These are potentially redefined by // predicated instructions. - SmallSet Redefs; - InitPredRedefs(CvtBBI->BB, Redefs, TRI); - InitPredRedefs(NextBBI->BB, Redefs, TRI); + Redefs.init(TRI); + Redefs.addLiveIns(CvtBBI->BB); + Redefs.addLiveIns(NextBBI->BB); + + DontKill.clear(); + + bool HasEarlyExit = CvtBBI->FalseBB != nullptr; + uint64_t CvtNext = 0, CvtFalse = 0, BBNext = 0, BBCvt = 0, SumWeight = 0; + uint32_t WeightScale = 0; + + if (HasEarlyExit) { + // Get weights before modifying CvtBBI->BB and BBI.BB. + CvtNext = MBPI->getEdgeWeight(CvtBBI->BB, NextBBI->BB); + CvtFalse = MBPI->getEdgeWeight(CvtBBI->BB, CvtBBI->FalseBB); + BBNext = MBPI->getEdgeWeight(BBI.BB, NextBBI->BB); + BBCvt = MBPI->getEdgeWeight(BBI.BB, CvtBBI->BB); + SumWeight = MBPI->getSumForBlock(CvtBBI->BB, WeightScale); + } - bool HasEarlyExit = CvtBBI->FalseBB != NULL; if (CvtBBI->BB->pred_size() > 1) { BBI.NonPredSize -= TII->RemoveBranch(*BBI.BB); // Copy instructions in the true block, predicate them, and add them to // the entry block. - CopyAndPredicateBlock(BBI, *CvtBBI, Cond, Redefs, true); + CopyAndPredicateBlock(BBI, *CvtBBI, Cond, true); + + // RemoveExtraEdges won't work if the block has an unanalyzable branch, so + // explicitly remove CvtBBI as a successor. + BBI.BB->removeSuccessor(CvtBBI->BB); } else { // Predicate the 'true' block after removing its branch. CvtBBI->NonPredSize -= TII->RemoveBranch(*CvtBBI->BB); - PredicateBlock(*CvtBBI, CvtBBI->BB->end(), Cond, Redefs); + PredicateBlock(*CvtBBI, CvtBBI->BB->end(), Cond); // Now merge the entry of the triangle with the true block. BBI.NonPredSize -= TII->RemoveBranch(*BBI.BB); @@ -1164,8 +1220,22 @@ bool IfConverter::IfConvertTriangle(BBInfo &BBI, IfcvtKind Kind) { CvtBBI->BrCond.end()); if (TII->ReverseBranchCondition(RevCond)) llvm_unreachable("Unable to reverse branch condition!"); - TII->InsertBranch(*BBI.BB, CvtBBI->FalseBB, NULL, RevCond, dl); + TII->InsertBranch(*BBI.BB, CvtBBI->FalseBB, nullptr, RevCond, dl); BBI.BB->addSuccessor(CvtBBI->FalseBB); + // Update the edge weight for both CvtBBI->FalseBB and NextBBI. + // New_Weight(BBI.BB, NextBBI->BB) = + // Weight(BBI.BB, NextBBI->BB) * getSumForBlock(CvtBBI->BB) + + // Weight(BBI.BB, CvtBBI->BB) * Weight(CvtBBI->BB, NextBBI->BB) + // New_Weight(BBI.BB, CvtBBI->FalseBB) = + // Weight(BBI.BB, CvtBBI->BB) * Weight(CvtBBI->BB, CvtBBI->FalseBB) + + uint64_t NewNext = BBNext * SumWeight + (BBCvt * CvtNext) / WeightScale; + uint64_t NewFalse = (BBCvt * CvtFalse) / WeightScale; + // We need to scale down all weights of BBI.BB to fit uint32_t. + // Here BBI.BB is connected to CvtBBI->FalseBB and will fall through to + // the next block. + ScaleWeights(NewNext, NewFalse, BBI.BB, getNextBlock(BBI.BB), + CvtBBI->FalseBB, MBPI); } // Merge in the 'false' block if the 'false' block has no other @@ -1178,7 +1248,8 @@ bool IfConverter::IfConvertTriangle(BBInfo &BBI, IfcvtKind Kind) { // block. By not merging them, we make it possible to iteratively // ifcvt the blocks. if (!HasEarlyExit && - NextBBI->BB->pred_size() == 1 && !NextBBI->HasFallThrough) { + NextBBI->BB->pred_size() == 1 && !NextBBI->HasFallThrough && + !NextBBI->BB->hasAddressTaken()) { MergeBlocks(BBI, *NextBBI); FalseBBDead = true; } else { @@ -1228,6 +1299,10 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, return false; } + if (TrueBBI.BB->hasAddressTaken() || FalseBBI.BB->hasAddressTaken()) + // Conservatively abort if-conversion if either BB has its address taken. + return false; + // Put the predicated instructions from the 'true' block before the // instructions from the 'false' block, unless the true block would clobber // the predicate, in which case, do the opposite. @@ -1257,8 +1332,8 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, // Initialize liveins to the first BB. These are potentially redefined by // predicated instructions. - SmallSet Redefs; - InitPredRedefs(BBI1->BB, Redefs, TRI); + Redefs.init(TRI); + Redefs.addLiveIns(BBI1->BB); // Remove the duplicated instructions at the beginnings of both paths. MachineBasicBlock::iterator DI1 = BBI1->BB->begin(); @@ -1285,7 +1360,19 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, --NumDups1; } - UpdatePredRedefs(BBI1->BB->begin(), DI1, Redefs, TRI); + // Compute a set of registers which must not be killed by instructions in BB1: + // This is everything used+live in BB2 after the duplicated instructions. We + // can compute this set by simulating liveness backwards from the end of BB2. + DontKill.init(TRI); + for (MachineBasicBlock::reverse_iterator I = BBI2->BB->rbegin(), + E = MachineBasicBlock::reverse_iterator(DI2); I != E; ++I) { + DontKill.stepBackward(*I); + } + + for (MachineBasicBlock::const_iterator I = BBI1->BB->begin(), E = DI1; I != E; + ++I) { + Redefs.stepForward(*I); + } BBI.BB->splice(BBI.BB->end(), BBI1->BB, BBI1->BB->begin(), DI1); BBI2->BB->erase(BBI2->BB->begin(), DI2); @@ -1303,6 +1390,10 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, } BBI1->BB->erase(DI1, BBI1->BB->end()); + // Kill flags in the true block for registers living into the false block + // must be removed. + RemoveKills(BBI1->BB->begin(), BBI1->BB->end(), DontKill, *TRI); + // Remove 'false' block branch and find the last instruction to predicate. BBI2->NonPredSize -= TII->RemoveBranch(*BBI2->BB); DI2 = BBI2->BB->end(); @@ -1343,8 +1434,8 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, } else if (!RedefsByFalse.count(Reg)) { // These are defined before ctrl flow reach the 'false' instructions. // They cannot be modified by the 'true' instructions. - ExtUses.insert(Reg); - for (MCSubRegIterator SubRegs(Reg, TRI); SubRegs.isValid(); ++SubRegs) + for (MCSubRegIterator SubRegs(Reg, TRI, /*IncludeSelf=*/true); + SubRegs.isValid(); ++SubRegs) ExtUses.insert(*SubRegs); } } @@ -1352,8 +1443,8 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, for (unsigned i = 0, e = Defs.size(); i != e; ++i) { unsigned Reg = Defs[i]; if (!ExtUses.count(Reg)) { - RedefsByFalse.insert(Reg); - for (MCSubRegIterator SubRegs(Reg, TRI); SubRegs.isValid(); ++SubRegs) + for (MCSubRegIterator SubRegs(Reg, TRI, /*IncludeSelf=*/true); + SubRegs.isValid(); ++SubRegs) RedefsByFalse.insert(*SubRegs); } } @@ -1361,14 +1452,14 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, } // Predicate the 'true' block. - PredicateBlock(*BBI1, BBI1->BB->end(), *Cond1, Redefs, &RedefsByFalse); + PredicateBlock(*BBI1, BBI1->BB->end(), *Cond1, &RedefsByFalse); // Predicate the 'false' block. - PredicateBlock(*BBI2, DI2, *Cond2, Redefs); + PredicateBlock(*BBI2, DI2, *Cond2); // Merge the true block into the entry of the diamond. - MergeBlocks(BBI, *BBI1, TailBB == 0); - MergeBlocks(BBI, *BBI2, TailBB == 0); + MergeBlocks(BBI, *BBI1, TailBB == nullptr); + MergeBlocks(BBI, *BBI2, TailBB == nullptr); // If the if-converted block falls through or unconditionally branches into // the tail block, and the tail block does not have other predecessors, then @@ -1376,7 +1467,8 @@ bool IfConverter::IfConvertDiamond(BBInfo &BBI, IfcvtKind Kind, // tail, add a unconditional branch to it. if (TailBB) { BBInfo &TailBBI = BBAnalysis[TailBB->getNumber()]; - bool CanMergeTail = !TailBBI.HasFallThrough; + bool CanMergeTail = !TailBBI.HasFallThrough && + !TailBBI.BB->hasAddressTaken(); // There may still be a fall-through edge from BBI1 or BBI2 to TailBB; // check if there are any other predecessors besides those. unsigned NumPreds = TailBB->pred_size(); @@ -1416,7 +1508,7 @@ static bool MaySpeculate(const MachineInstr *MI, SmallSet &LaterRedefs, const TargetInstrInfo *TII) { bool SawStore = true; - if (!MI->isSafeToMove(TII, 0, SawStore)) + if (!MI->isSafeToMove(TII, nullptr, SawStore)) return false; for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { @@ -1438,10 +1530,9 @@ static bool MaySpeculate(const MachineInstr *MI, void IfConverter::PredicateBlock(BBInfo &BBI, MachineBasicBlock::iterator E, SmallVectorImpl &Cond, - SmallSet &Redefs, SmallSet *LaterRedefs) { bool AnyUnpred = false; - bool MaySpec = LaterRedefs != 0; + bool MaySpec = LaterRedefs != nullptr; for (MachineBasicBlock::iterator I = BBI.BB->begin(); I != E; ++I) { if (I->isDebugValue() || TII->isPredicated(I)) continue; @@ -1459,12 +1550,12 @@ void IfConverter::PredicateBlock(BBInfo &BBI, #ifndef NDEBUG dbgs() << "Unable to predicate " << *I << "!\n"; #endif - llvm_unreachable(0); + llvm_unreachable(nullptr); } // If the predicated instruction now redefines a register as the result of // if-conversion, add an implicit kill. - UpdatePredRedefs(I, Redefs, TRI, true); + UpdatePredRedefs(I, Redefs); } std::copy(Cond.begin(), Cond.end(), std::back_inserter(BBI.Predicate)); @@ -1481,7 +1572,6 @@ void IfConverter::PredicateBlock(BBInfo &BBI, /// the destination block. Skip end of block branches if IgnoreBr is true. void IfConverter::CopyAndPredicateBlock(BBInfo &ToBBI, BBInfo &FromBBI, SmallVectorImpl &Cond, - SmallSet &Redefs, bool IgnoreBr) { MachineFunction &MF = *ToBBI.BB->getParent(); @@ -1494,8 +1584,8 @@ void IfConverter::CopyAndPredicateBlock(BBInfo &ToBBI, BBInfo &FromBBI, MachineInstr *MI = MF.CloneMachineInstr(I); ToBBI.BB->insert(ToBBI.BB->end(), MI); ToBBI.NonPredSize++; - unsigned ExtraPredCost = 0; - unsigned NumCycles = TII->getInstrLatency(InstrItins, &*I, &ExtraPredCost); + unsigned ExtraPredCost = TII->getPredicationCost(&*I); + unsigned NumCycles = SchedModel.computeInstrLatency(&*I, false); if (NumCycles > 1) ToBBI.ExtraCost += NumCycles-1; ToBBI.ExtraCost2 += ExtraPredCost; @@ -1505,20 +1595,24 @@ void IfConverter::CopyAndPredicateBlock(BBInfo &ToBBI, BBInfo &FromBBI, #ifndef NDEBUG dbgs() << "Unable to predicate " << *I << "!\n"; #endif - llvm_unreachable(0); + llvm_unreachable(nullptr); } } // If the predicated instruction now redefines a register as the result of // if-conversion, add an implicit kill. - UpdatePredRedefs(MI, Redefs, TRI, true); + UpdatePredRedefs(MI, Redefs); + + // Some kill flags may not be correct anymore. + if (!DontKill.empty()) + RemoveKills(*MI, DontKill); } if (!IgnoreBr) { std::vector Succs(FromBBI.BB->succ_begin(), FromBBI.BB->succ_end()); MachineBasicBlock *NBB = getNextBlock(FromBBI.BB); - MachineBasicBlock *FallThrough = FromBBI.HasFallThrough ? NBB : NULL; + MachineBasicBlock *FallThrough = FromBBI.HasFallThrough ? NBB : nullptr; for (unsigned i = 0, e = Succs.size(); i != e; ++i) { MachineBasicBlock *Succ = Succs[i]; @@ -1545,13 +1639,16 @@ void IfConverter::CopyAndPredicateBlock(BBInfo &ToBBI, BBInfo &FromBBI, /// i.e., when FromBBI's branch is being moved, add those successor edges to /// ToBBI. void IfConverter::MergeBlocks(BBInfo &ToBBI, BBInfo &FromBBI, bool AddEdges) { + assert(!FromBBI.BB->hasAddressTaken() && + "Removing a BB whose address is taken!"); + ToBBI.BB->splice(ToBBI.BB->end(), FromBBI.BB, FromBBI.BB->begin(), FromBBI.BB->end()); std::vector Succs(FromBBI.BB->succ_begin(), FromBBI.BB->succ_end()); MachineBasicBlock *NBB = getNextBlock(FromBBI.BB); - MachineBasicBlock *FallThrough = FromBBI.HasFallThrough ? NBB : NULL; + MachineBasicBlock *FallThrough = FromBBI.HasFallThrough ? NBB : nullptr; for (unsigned i = 0, e = Succs.size(); i != e; ++i) { MachineBasicBlock *Succ = Succs[i]; @@ -1559,7 +1656,7 @@ void IfConverter::MergeBlocks(BBInfo &ToBBI, BBInfo &FromBBI, bool AddEdges) { if (Succ == FallThrough) continue; FromBBI.BB->removeSuccessor(Succ); - if (AddEdges) + if (AddEdges && !ToBBI.BB->isSuccessor(Succ)) ToBBI.BB->addSuccessor(Succ); }