X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FMachineScheduler.cpp;h=ee2dbc86bf1f760d1a543abe71b5a9b51069dc99;hb=324bf0ddcc8acfaba6c07415335a978ba3f0a57f;hp=cbe937a06f02efc98b5f905f2284153b930a1f67;hpb=808823cf77076a4a77937a02b0c51b315f722b7b;p=oota-llvm.git diff --git a/lib/CodeGen/MachineScheduler.cpp b/lib/CodeGen/MachineScheduler.cpp index cbe937a06f0..ee2dbc86bf1 100644 --- a/lib/CodeGen/MachineScheduler.cpp +++ b/lib/CodeGen/MachineScheduler.cpp @@ -40,12 +40,20 @@ cl::opt ForceTopDown("misched-topdown", cl::Hidden, cl::desc("Force top-down list scheduling")); cl::opt ForceBottomUp("misched-bottomup", cl::Hidden, cl::desc("Force bottom-up list scheduling")); +cl::opt +DumpCriticalPathLength("misched-dcpl", cl::Hidden, + cl::desc("Print critical path length to stdout")); } #ifndef NDEBUG static cl::opt ViewMISchedDAGs("view-misched-dags", cl::Hidden, cl::desc("Pop up a window to show MISched dags after they are processed")); +/// In some situations a few uninteresting nodes depend on nearly all other +/// nodes in the graph, provide a cutoff to hide them. +static cl::opt ViewMISchedCutoff("view-misched-cutoff", cl::Hidden, + cl::desc("Hide nodes with more predecessor/successor than cutoff")); + static cl::opt MISchedCutoff("misched-cutoff", cl::Hidden, cl::desc("Stop scheduling after N instructions"), cl::init(~0U)); @@ -141,12 +149,12 @@ char MachineScheduler::ID = 0; char &llvm::MachineSchedulerID = MachineScheduler::ID; -INITIALIZE_PASS_BEGIN(MachineScheduler, "misched", +INITIALIZE_PASS_BEGIN(MachineScheduler, "machine-scheduler", "Machine Instruction Scheduler", false, false) -INITIALIZE_AG_DEPENDENCY(AliasAnalysis) +INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass) INITIALIZE_PASS_DEPENDENCY(SlotIndexes) INITIALIZE_PASS_DEPENDENCY(LiveIntervals) -INITIALIZE_PASS_END(MachineScheduler, "misched", +INITIALIZE_PASS_END(MachineScheduler, "machine-scheduler", "Machine Instruction Scheduler", false, false) MachineScheduler::MachineScheduler() @@ -158,7 +166,7 @@ void MachineScheduler::getAnalysisUsage(AnalysisUsage &AU) const { AU.setPreservesCFG(); AU.addRequiredID(MachineDominatorsID); AU.addRequired(); - AU.addRequired(); + AU.addRequired(); AU.addRequired(); AU.addRequired(); AU.addPreserved(); @@ -206,6 +214,11 @@ static MachineSchedRegistry DefaultSchedRegistry("default", "Use the target's default scheduler choice.", useDefaultMachineSched); +static cl::opt EnableMachineSched( + "enable-misched", + cl::desc("Enable the machine instruction scheduling pass."), cl::init(true), + cl::Hidden); + /// Forward declare the standard machine scheduler. This will be used as the /// default scheduler if the target does not set a default. static ScheduleDAGInstrs *createGenericSchedLive(MachineSchedContext *C); @@ -301,6 +314,12 @@ ScheduleDAGInstrs *PostMachineScheduler::createPostMachineScheduler() { /// design would be to split blocks at scheduling boundaries, but LLVM has a /// general bias against block splitting purely for implementation simplicity. bool MachineScheduler::runOnMachineFunction(MachineFunction &mf) { + if (EnableMachineSched.getNumOccurrences()) { + if (!EnableMachineSched) + return false; + } else if (!mf.getSubtarget().enableMachineScheduler()) + return false; + DEBUG(dbgs() << "Before MISsched:\n"; mf.print(dbgs())); // Initialize the context of the pass. @@ -308,7 +327,7 @@ bool MachineScheduler::runOnMachineFunction(MachineFunction &mf) { MLI = &getAnalysis(); MDT = &getAnalysis(); PassConfig = &getAnalysis(); - AA = &getAnalysis(); + AA = &getAnalysis().getAAResults(); LIS = &getAnalysis(); @@ -333,9 +352,7 @@ bool PostMachineScheduler::runOnMachineFunction(MachineFunction &mf) { if (skipOptnoneFunction(*mf.getFunction())) return false; - const TargetSubtargetInfo &ST = - mf.getTarget().getSubtarget(); - if (!ST.enablePostMachineScheduler()) { + if (!mf.getSubtarget().enablePostRAScheduler()) { DEBUG(dbgs() << "Subtarget disables post-MI-sched.\n"); return false; } @@ -378,7 +395,7 @@ static bool isSchedBoundary(MachineBasicBlock::iterator MI, /// Main driver for both MachineScheduler and PostMachineScheduler. void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) { - const TargetInstrInfo *TII = MF->getTarget().getInstrInfo(); + const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo(); bool IsPostRA = Scheduler.isPostRA(); // Visit all machine basic blocks. @@ -388,7 +405,7 @@ void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) { for (MachineFunction::iterator MBB = MF->begin(), MBBEnd = MF->end(); MBB != MBBEnd; ++MBB) { - Scheduler.startBlock(MBB); + Scheduler.startBlock(&*MBB); #ifndef NDEBUG if (SchedOnlyFunc.getNumOccurrences() && SchedOnlyFunc != MF->getName()) @@ -417,7 +434,7 @@ void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) { // Avoid decrementing RegionEnd for blocks with no terminator. if (RegionEnd != MBB->end() || - isSchedBoundary(std::prev(RegionEnd), MBB, MF, TII, IsPostRA)) { + isSchedBoundary(&*std::prev(RegionEnd), &*MBB, MF, TII, IsPostRA)) { --RegionEnd; // Count the boundary instruction. --RemainingInstrs; @@ -427,13 +444,15 @@ void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) { // instruction stream until we find the nearest boundary. unsigned NumRegionInstrs = 0; MachineBasicBlock::iterator I = RegionEnd; - for(;I != MBB->begin(); --I, --RemainingInstrs, ++NumRegionInstrs) { - if (isSchedBoundary(std::prev(I), MBB, MF, TII, IsPostRA)) + for(;I != MBB->begin(); --I, --RemainingInstrs) { + if (isSchedBoundary(&*std::prev(I), &*MBB, MF, TII, IsPostRA)) break; + if (!I->isDebugValue()) + ++NumRegionInstrs; } // Notify the scheduler of the region, even if we may skip scheduling // it. Perhaps it still needs to be bundled. - Scheduler.enterRegion(MBB, I, RegionEnd, NumRegionInstrs); + Scheduler.enterRegion(&*MBB, I, RegionEnd, NumRegionInstrs); // Skip empty scheduling regions (0 or 1 schedulable instructions). if (I == RegionEnd || I == std::prev(RegionEnd)) { @@ -451,6 +470,11 @@ void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) { else dbgs() << "End"; dbgs() << " RegionInstrs: " << NumRegionInstrs << " Remaining: " << RemainingInstrs << "\n"); + if (DumpCriticalPathLength) { + errs() << MF->getName(); + errs() << ":BB# " << MBB->getNumber(); + errs() << " " << MBB->getName() << " \n"; + } // Schedule a region: possibly reorder instructions. // This invalidates 'RegionEnd' and 'I'. @@ -468,7 +492,7 @@ void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) { if (Scheduler.isPostRA()) { // FIXME: Ideally, no further passes should rely on kill flags. However, // thumb2 size reduction is currently an exception. - Scheduler.fixupKills(MBB); + Scheduler.fixupKills(&*MBB); } } Scheduler.finalizeSchedule(); @@ -478,14 +502,13 @@ void MachineSchedulerBase::print(raw_ostream &O, const Module* m) const { // unimplemented } -#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) +LLVM_DUMP_METHOD void ReadyQueue::dump() { - dbgs() << Name << ": "; + dbgs() << "Queue " << Name << ": "; for (unsigned i = 0, e = Queue.size(); i < e; ++i) dbgs() << Queue[i]->NodeNum << " "; dbgs() << "\n"; } -#endif //===----------------------------------------------------------------------===// // ScheduleDAGMI - Basic machine instruction scheduling. This is @@ -642,6 +665,9 @@ bool ScheduleDAGMI::checkSchedLimit() { /// does not consider liveness or register pressure. It is useful for PostRA /// scheduling and potentially other custom schedulers. void ScheduleDAGMI::schedule() { + DEBUG(dbgs() << "ScheduleDAGMI::schedule starting\n"); + DEBUG(SchedImpl->dumpPolicy()); + // Build the DAG. buildSchedGraph(AA); @@ -664,7 +690,11 @@ void ScheduleDAGMI::schedule() { initQueues(TopRoots, BotRoots); bool IsTopNode = false; - while (SUnit *SU = SchedImpl->pickNode(IsTopNode)) { + while (true) { + DEBUG(dbgs() << "** ScheduleDAGMI::schedule picking next node\n"); + SUnit *SU = SchedImpl->pickNode(IsTopNode); + if (!SU) break; + assert(!SU->isScheduled && "Node already scheduled"); if (!checkSchedLimit()) break; @@ -691,7 +721,7 @@ void ScheduleDAGMI::schedule() { } } // Notify the scheduling strategy before updating the DAG. - // This sets the scheduled nodes ReadyCycle to CurrCycle. When updateQueues + // This sets the scheduled node's ReadyCycle to CurrCycle. When updateQueues // runs, it can then use the accurate ReadyCycle time to determine whether // newly released nodes can move to the readyQ. SchedImpl->schedNode(SU, IsTopNode); @@ -925,8 +955,9 @@ updateScheduledPressure(const SUnit *SU, unsigned Limit = RegClassInfo->getRegPressureSetLimit(ID); if (NewMaxPressure[ID] >= Limit - 2) { DEBUG(dbgs() << " " << TRI->getRegPressureSetName(ID) << ": " - << NewMaxPressure[ID] << " > " << Limit << "(+ " - << BotRPTracker.getLiveThru()[ID] << " livethru)\n"); + << NewMaxPressure[ID] + << ((NewMaxPressure[ID] > Limit) ? " > " : " <= ") << Limit + << "(+ " << BotRPTracker.getLiveThru()[ID] << " livethru)\n"); } } } @@ -979,12 +1010,14 @@ void ScheduleDAGMILive::updatePressureDiffs(ArrayRef LiveUses) { /// only includes instructions that have DAG nodes, not scheduling boundaries. /// /// This is a skeletal driver, with all the functionality pushed into helpers, -/// so that it can be easilly extended by experimental schedulers. Generally, +/// so that it can be easily extended by experimental schedulers. Generally, /// implementing MachineSchedStrategy should be sufficient to implement a new /// scheduling algorithm. However, if a scheduler further subclasses /// ScheduleDAGMILive then it will want to override this virtual method in order /// to update any specialized state. void ScheduleDAGMILive::schedule() { + DEBUG(dbgs() << "ScheduleDAGMILive::schedule starting\n"); + DEBUG(SchedImpl->dumpPolicy()); buildDAGWithRegPressure(); Topo.InitDAGTopologicalSorting(); @@ -1011,15 +1044,17 @@ void ScheduleDAGMILive::schedule() { } bool IsTopNode = false; - while (SUnit *SU = SchedImpl->pickNode(IsTopNode)) { + while (true) { + DEBUG(dbgs() << "** ScheduleDAGMILive::schedule picking next node\n"); + SUnit *SU = SchedImpl->pickNode(IsTopNode); + if (!SU) break; + assert(!SU->isScheduled && "Node already scheduled"); if (!checkSchedLimit()) break; scheduleMI(SU, IsTopNode); - updateQueues(SU, IsTopNode); - if (DFSResult) { unsigned SubtreeID = DFSResult->getSubtreeID(SU); if (!ScheduledTrees.test(SubtreeID)) { @@ -1031,6 +1066,8 @@ void ScheduleDAGMILive::schedule() { // Notify the scheduling strategy after updating the DAG. SchedImpl->schedNode(SU, IsTopNode); + + updateQueues(SU, IsTopNode); } assert(CurrentTop == CurrentBottom && "Nonempty unscheduled zone."); @@ -1252,7 +1289,7 @@ void LoadClusterMutation::clusterNeighboringLoads(ArrayRef Loads, SUnit *SU = Loads[Idx]; unsigned BaseReg; unsigned Offset; - if (TII->getLdStBaseRegImmOfs(SU->getInstr(), BaseReg, Offset, TRI)) + if (TII->getMemOpBaseRegImmOfs(SU->getInstr(), BaseReg, Offset, TRI)) LoadRecords.push_back(LoadInfo(SU, BaseReg, Offset)); } if (LoadRecords.size() < 2) @@ -1330,25 +1367,49 @@ namespace { /// \brief Post-process the DAG to create cluster edges between instructions /// that may be fused by the processor into a single operation. class MacroFusion : public ScheduleDAGMutation { - const TargetInstrInfo *TII; + const TargetInstrInfo &TII; + const TargetRegisterInfo &TRI; public: - MacroFusion(const TargetInstrInfo *tii): TII(tii) {} + MacroFusion(const TargetInstrInfo &TII, const TargetRegisterInfo &TRI) + : TII(TII), TRI(TRI) {} void apply(ScheduleDAGMI *DAG) override; }; } // anonymous +/// Returns true if \p MI reads a register written by \p Other. +static bool HasDataDep(const TargetRegisterInfo &TRI, const MachineInstr &MI, + const MachineInstr &Other) { + for (const MachineOperand &MO : MI.uses()) { + if (!MO.isReg() || !MO.readsReg()) + continue; + + unsigned Reg = MO.getReg(); + if (Other.modifiesRegister(Reg, &TRI)) + return true; + } + return false; +} + /// \brief Callback from DAG postProcessing to create cluster edges to encourage /// fused operations. void MacroFusion::apply(ScheduleDAGMI *DAG) { // For now, assume targets can only fuse with the branch. - MachineInstr *Branch = DAG->ExitSU.getInstr(); + SUnit &ExitSU = DAG->ExitSU; + MachineInstr *Branch = ExitSU.getInstr(); if (!Branch) return; - for (unsigned Idx = DAG->SUnits.size(); Idx > 0;) { - SUnit *SU = &DAG->SUnits[--Idx]; - if (!TII->shouldScheduleAdjacent(SU->getInstr(), Branch)) + for (SUnit &SU : DAG->SUnits) { + // SUnits with successors can't be schedule in front of the ExitSU. + if (!SU.Succs.empty()) + continue; + // We only care if the node writes to a register that the branch reads. + MachineInstr *Pred = SU.getInstr(); + if (!HasDataDep(TRI, *Branch, *Pred)) + continue; + + if (!TII.shouldScheduleAdjacent(Pred, Branch)) continue; // Create a single weak edge from SU to ExitSU. The only effect is to cause @@ -1357,11 +1418,11 @@ void MacroFusion::apply(ScheduleDAGMI *DAG) { // scheduling cannot prioritize ExitSU anyway. To defer top-down scheduling // of SU, we could create an artificial edge from the deepest root, but it // hasn't been needed yet. - bool Success = DAG->addEdge(&DAG->ExitSU, SDep(SU, SDep::Cluster)); + bool Success = DAG->addEdge(&ExitSU, SDep(&SU, SDep::Cluster)); (void)Success; assert(Success && "No DAG nodes should be reachable from ExitSU"); - DEBUG(dbgs() << "Macro Fuse SU(" << SU->NodeNum << ")\n"); + DEBUG(dbgs() << "Macro Fuse SU(" << SU.NodeNum << ")\n"); break; } } @@ -1425,12 +1486,15 @@ void CopyConstrain::constrainLocalCopy(SUnit *CopySU, ScheduleDAGMILive *DAG) { // Check if either the dest or source is local. If it's live across a back // edge, it's not local. Note that if both vregs are live across the back // edge, we cannot successfully contrain the copy without cyclic scheduling. - unsigned LocalReg = DstReg; - unsigned GlobalReg = SrcReg; + // If both the copy's source and dest are local live intervals, then we + // should treat the dest as the global for the purpose of adding + // constraints. This adds edges from source's other uses to the copy. + unsigned LocalReg = SrcReg; + unsigned GlobalReg = DstReg; LiveInterval *LocalLI = &LIS->getInterval(LocalReg); if (!LocalLI->isLocal(RegionBeginIdx, RegionEndIdx)) { - LocalReg = SrcReg; - GlobalReg = DstReg; + LocalReg = DstReg; + GlobalReg = SrcReg; LocalLI = &LIS->getInterval(LocalReg); if (!LocalLI->isLocal(RegionBeginIdx, RegionEndIdx)) return; @@ -1687,8 +1751,16 @@ bool SchedBoundary::checkHazard(SUnit *SU) { for (TargetSchedModel::ProcResIter PI = SchedModel->getWriteProcResBegin(SC), PE = SchedModel->getWriteProcResEnd(SC); PI != PE; ++PI) { - if (getNextResourceCycle(PI->ProcResourceIdx, PI->Cycles) > CurrCycle) + unsigned NRCycle = getNextResourceCycle(PI->ProcResourceIdx, PI->Cycles); + if (NRCycle > CurrCycle) { +#ifndef NDEBUG + MaxObservedStall = std::max(PI->Cycles, MaxObservedStall); +#endif + DEBUG(dbgs() << " SU(" << SU->NodeNum << ") " + << SchedModel->getResourceName(PI->ProcResourceIdx) + << "=" << NRCycle << "c\n"); return true; + } } } return false; @@ -1747,7 +1819,11 @@ void SchedBoundary::releaseNode(SUnit *SU, unsigned ReadyCycle) { assert(SU->getInstr() && "Scheduled SUnit must have instr"); #ifndef NDEBUG - MaxObservedStall = std::max(ReadyCycle - CurrCycle, MaxObservedStall); + // ReadyCycle was been bumped up to the CurrCycle when this node was + // scheduled, but CurrCycle may have been eagerly advanced immediately after + // scheduling, so may now be greater than ReadyCycle. + if (ReadyCycle > CurrCycle) + MaxObservedStall = std::max(ReadyCycle - CurrCycle, MaxObservedStall); #endif if (ReadyCycle < MinReadyCycle) @@ -1942,10 +2018,12 @@ void SchedBoundary::bumpNode(SUnit *SU) { PE = SchedModel->getWriteProcResEnd(SC); PI != PE; ++PI) { unsigned PIdx = PI->ProcResourceIdx; if (SchedModel->getProcResource(PIdx)->BufferSize == 0) { - ReservedCycles[PIdx] = isTop() ? NextCycle + PI->Cycles : NextCycle; -#ifndef NDEBUG - MaxObservedStall = std::max(PI->Cycles, MaxObservedStall); -#endif + if (isTop()) { + ReservedCycles[PIdx] = + std::max(getNextResourceCycle(PIdx, 0), NextCycle + PI->Cycles); + } + else + ReservedCycles[PIdx] = NextCycle; } } } @@ -2048,8 +2126,10 @@ SUnit *SchedBoundary::pickOnlyChoice() { } } for (unsigned i = 0; Available.empty(); ++i) { - assert(i <= (HazardRec->getMaxLookAhead() + MaxObservedStall) && - "permanent hazard"); (void)i; +// FIXME: Re-enable assert once PR20057 is resolved. +// assert(i <= (HazardRec->getMaxLookAhead() + MaxObservedStall) && +// "permanent hazard"); + (void)i; bumpCycle(CurrCycle + 1); releasePending(); } @@ -2112,7 +2192,7 @@ void GenericSchedulerBase::setPolicy(CandPolicy &Policy, bool IsPostRA, SchedBoundary &CurrZone, SchedBoundary *OtherZone) { - // Apply preemptive heuristics based on the the total latency and resources + // Apply preemptive heuristics based on the total latency and resources // inside and outside this zone. Potential stalls should be considered before // following this policy. @@ -2239,7 +2319,7 @@ void GenericSchedulerBase::traceCandidate(const SchedCandidate &Cand) { Latency = Cand.SU->getDepth(); break; } - dbgs() << " SU(" << Cand.SU->NodeNum << ") " << getReasonStr(Cand.Reason); + dbgs() << " Cand SU(" << Cand.SU->NodeNum << ") " << getReasonStr(Cand.Reason); if (P.isValid()) dbgs() << " " << TRI->getRegPressureSetName(P.getPSet()) << ":" << P.getUnitInc() << " "; @@ -2340,14 +2420,15 @@ void GenericScheduler::initialize(ScheduleDAGMI *dag) { // Initialize the HazardRecognizers. If itineraries don't exist, are empty, or // are disabled, then these HazardRecs will be disabled. const InstrItineraryData *Itin = SchedModel->getInstrItineraries(); - const TargetMachine &TM = DAG->MF.getTarget(); if (!Top.HazardRec) { Top.HazardRec = - TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG); + DAG->MF.getSubtarget().getInstrInfo()->CreateTargetMIHazardRecognizer( + Itin, DAG); } if (!Bot.HazardRec) { Bot.HazardRec = - TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG); + DAG->MF.getSubtarget().getInstrInfo()->CreateTargetMIHazardRecognizer( + Itin, DAG); } } @@ -2355,8 +2436,8 @@ void GenericScheduler::initialize(ScheduleDAGMI *dag) { void GenericScheduler::initPolicy(MachineBasicBlock::iterator Begin, MachineBasicBlock::iterator End, unsigned NumRegionInstrs) { - const TargetMachine &TM = Context->MF->getTarget(); - const TargetLowering *TLI = TM.getTargetLowering(); + const MachineFunction &MF = *Begin->getParent()->getParent(); + const TargetLowering *TLI = MF.getSubtarget().getTargetLowering(); // Avoid setting up the register pressure tracker for small regions to save // compile time. As a rough heuristic, only track pressure when the number of @@ -2376,8 +2457,8 @@ void GenericScheduler::initPolicy(MachineBasicBlock::iterator Begin, RegionPolicy.OnlyBottomUp = true; // Allow the subtarget to override default policy. - const TargetSubtargetInfo &ST = TM.getSubtarget(); - ST.overrideSchedPolicy(RegionPolicy, Begin, End, NumRegionInstrs); + MF.getSubtarget().overrideSchedPolicy(RegionPolicy, Begin, End, + NumRegionInstrs); // After subtarget overrides, apply command line options. if (!EnableRegPressure) @@ -2399,6 +2480,14 @@ void GenericScheduler::initPolicy(MachineBasicBlock::iterator Begin, } } +void GenericScheduler::dumpPolicy() { + dbgs() << "GenericScheduler RegionPolicy: " + << " ShouldTrackPressure=" << RegionPolicy.ShouldTrackPressure + << " OnlyTopDown=" << RegionPolicy.OnlyTopDown + << " OnlyBottomUp=" << RegionPolicy.OnlyBottomUp + << "\n"; +} + /// Set IsAcyclicLatencyLimited if the acyclic path is longer than the cyclic /// critical path by more cycles than it takes to drain the instruction buffer. /// We estimate an upper bounds on in-flight instructions as: @@ -2445,7 +2534,10 @@ void GenericScheduler::registerRoots() { if ((*I)->getDepth() > Rem.CriticalPath) Rem.CriticalPath = (*I)->getDepth(); } - DEBUG(dbgs() << "Critical Path: " << Rem.CriticalPath << '\n'); + DEBUG(dbgs() << "Critical Path(GS-RR ): " << Rem.CriticalPath << '\n'); + if (DumpCriticalPathLength) { + errs() << "Critical Path(GS-RR ): " << Rem.CriticalPath << " \n"; + } if (EnableCyclicPath) { Rem.CyclicCritPath = DAG->computeCyclicCriticalPath(); @@ -2467,8 +2559,8 @@ static bool tryPressure(const PressureChange &TryP, } // If one candidate decreases and the other increases, go with it. // Invalid candidates have UnitInc==0. - if (tryLess(TryP.getUnitInc() < 0, CandP.getUnitInc() < 0, TryCand, Cand, - Reason)) { + if (tryGreater(TryP.getUnitInc() < 0, CandP.getUnitInc() < 0, TryCand, Cand, + Reason)) { return true; } // If the candidates are decreasing pressure, reverse priority. @@ -2555,7 +2647,7 @@ void GenericScheduler::tryCandidate(SchedCandidate &Cand, } } DEBUG(if (TryCand.RPDelta.Excess.isValid()) - dbgs() << " SU(" << TryCand.SU->NodeNum << ") " + dbgs() << " Try SU(" << TryCand.SU->NodeNum << ") " << TRI->getRegPressureSetName(TryCand.RPDelta.Excess.getPSet()) << ":" << TryCand.RPDelta.Excess.getUnitInc() << "\n"); @@ -2570,8 +2662,7 @@ void GenericScheduler::tryCandidate(SchedCandidate &Cand, TryCand, Cand, PhysRegCopy)) return; - // Avoid exceeding the target's limit. If signed PSetID is negative, it is - // invalid; convert it to INT_MAX to give it lowest priority. + // Avoid exceeding the target's limit. if (DAG->isTrackingPressure() && tryPressure(TryCand.RPDelta.Excess, Cand.RPDelta.Excess, TryCand, Cand, RegExcess)) @@ -2631,8 +2722,8 @@ void GenericScheduler::tryCandidate(SchedCandidate &Cand, // Avoid serializing long latency dependence chains. // For acyclic path limited loops, latency was already checked above. - if (Cand.Policy.ReduceLatency && !Rem.IsAcyclicLatencyLimited - && tryLatency(TryCand, Cand, Zone)) { + if (!RegionPolicy.DisableLatencyHeuristic && Cand.Policy.ReduceLatency && + !Rem.IsAcyclicLatencyLimited && tryLatency(TryCand, Cand, Zone)) { return; } @@ -2846,7 +2937,7 @@ static ScheduleDAGInstrs *createGenericSchedLive(MachineSchedContext *C) { if (EnableLoadCluster && DAG->TII->enableClusterLoads()) DAG->addMutation(make_unique(DAG->TII, DAG->TRI)); if (EnableMacroFusion) - DAG->addMutation(make_unique(DAG->TII)); + DAG->addMutation(make_unique(*DAG->TII, *DAG->TRI)); return DAG; } @@ -2870,10 +2961,10 @@ void PostGenericScheduler::initialize(ScheduleDAGMI *Dag) { // Initialize the HazardRecognizers. If itineraries don't exist, are empty, // or are disabled, then these HazardRecs will be disabled. const InstrItineraryData *Itin = SchedModel->getInstrItineraries(); - const TargetMachine &TM = DAG->MF.getTarget(); if (!Top.HazardRec) { Top.HazardRec = - TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG); + DAG->MF.getSubtarget().getInstrInfo()->CreateTargetMIHazardRecognizer( + Itin, DAG); } } @@ -2887,7 +2978,10 @@ void PostGenericScheduler::registerRoots() { if ((*I)->getDepth() > Rem.CriticalPath) Rem.CriticalPath = (*I)->getDepth(); } - DEBUG(dbgs() << "Critical Path: " << Rem.CriticalPath << '\n'); + DEBUG(dbgs() << "Critical Path: (PGS-RR) " << Rem.CriticalPath << '\n'); + if (DumpCriticalPathLength) { + errs() << "Critical Path(PGS-RR ): " << Rem.CriticalPath << " \n"; + } } /// Apply a set of heursitics to a new candidate for PostRA scheduling. @@ -3210,7 +3304,10 @@ struct DOTGraphTraits : public DefaultDOTGraphTraits { } static bool isNodeHidden(const SUnit *Node) { - return (Node->Preds.size() > 10 || Node->Succs.size() > 10); + if (ViewMISchedCutoff == 0) + return false; + return (Node->Preds.size() > ViewMISchedCutoff + || Node->Succs.size() > ViewMISchedCutoff); } static bool hasNodeAddressLabel(const SUnit *Node,