cl::desc("Force top-down list scheduling"));
cl::opt<bool> ForceBottomUp("misched-bottomup", cl::Hidden,
cl::desc("Force bottom-up list scheduling"));
+cl::opt<bool>
+DumpCriticalPathLength("misched-dcpl", cl::Hidden,
+ cl::desc("Print critical path length to stdout"));
}
#ifndef NDEBUG
static cl::opt<bool> 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<unsigned> ViewMISchedCutoff("view-misched-cutoff", cl::Hidden,
+ cl::desc("Hide nodes with more predecessor/successor than cutoff"));
+
static cl::opt<unsigned> MISchedCutoff("misched-cutoff", cl::Hidden,
cl::desc("Stop scheduling after N instructions"), cl::init(~0U));
void print(raw_ostream &O, const Module* = nullptr) const override;
protected:
- void scheduleRegions(ScheduleDAGInstrs &Scheduler);
+ void scheduleRegions(ScheduleDAGInstrs &Scheduler, bool FixKillFlags);
};
/// MachineScheduler runs after coalescing and before register allocation.
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()
AU.setPreservesCFG();
AU.addRequiredID(MachineDominatorsID);
AU.addRequired<MachineLoopInfo>();
- AU.addRequired<AliasAnalysis>();
+ AU.addRequired<AAResultsWrapperPass>();
AU.addRequired<TargetPassConfig>();
AU.addRequired<SlotIndexes>();
AU.addPreserved<SlotIndexes>();
DefaultSchedRegistry("default", "Use the target's default scheduler choice.",
useDefaultMachineSched);
+static cl::opt<bool> 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);
/// 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) {
- DEBUG(dbgs() << "Before MISsched:\n"; mf.print(dbgs()));
+ if (EnableMachineSched.getNumOccurrences()) {
+ if (!EnableMachineSched)
+ return false;
+ } else if (!mf.getSubtarget().enableMachineScheduler())
+ return false;
+
+ DEBUG(dbgs() << "Before MISched:\n"; mf.print(dbgs()));
// Initialize the context of the pass.
MF = &mf;
MLI = &getAnalysis<MachineLoopInfo>();
MDT = &getAnalysis<MachineDominatorTree>();
PassConfig = &getAnalysis<TargetPassConfig>();
- AA = &getAnalysis<AliasAnalysis>();
+ AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
LIS = &getAnalysis<LiveIntervals>();
// Instantiate the selected scheduler for this target, function, and
// optimization level.
std::unique_ptr<ScheduleDAGInstrs> Scheduler(createMachineScheduler());
- scheduleRegions(*Scheduler);
+ scheduleRegions(*Scheduler, false);
DEBUG(LIS->dump());
if (VerifyScheduling)
if (skipOptnoneFunction(*mf.getFunction()))
return false;
- const TargetSubtargetInfo &ST =
- mf.getTarget().getSubtarget<TargetSubtargetInfo>();
- if (!ST.enablePostMachineScheduler()) {
+ if (!mf.getSubtarget().enablePostRAScheduler()) {
DEBUG(dbgs() << "Subtarget disables post-MI-sched.\n");
return false;
}
// Instantiate the selected scheduler for this target, function, and
// optimization level.
std::unique_ptr<ScheduleDAGInstrs> Scheduler(createPostMachineScheduler());
- scheduleRegions(*Scheduler);
+ scheduleRegions(*Scheduler, true);
if (VerifyScheduling)
MF->verify(this, "After post machine scheduling.");
static bool isSchedBoundary(MachineBasicBlock::iterator MI,
MachineBasicBlock *MBB,
MachineFunction *MF,
- const TargetInstrInfo *TII,
- bool IsPostRA) {
+ const TargetInstrInfo *TII) {
return MI->isCall() || TII->isSchedulingBoundary(MI, MBB, *MF);
}
/// Main driver for both MachineScheduler and PostMachineScheduler.
-void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler) {
- const TargetInstrInfo *TII = MF->getTarget().getInstrInfo();
- bool IsPostRA = Scheduler.isPostRA();
+void MachineSchedulerBase::scheduleRegions(ScheduleDAGInstrs &Scheduler,
+ bool FixKillFlags) {
+ const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
// Visit all machine basic blocks.
//
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())
// 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)) {
--RegionEnd;
// Count the boundary instruction.
--RemainingInstrs;
// 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))
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)) {
Scheduler.exitRegion();
continue;
}
- DEBUG(dbgs() << "********** " << ((Scheduler.isPostRA()) ? "PostRA " : "")
- << "MI Scheduling **********\n");
+ DEBUG(dbgs() << "********** MI Scheduling **********\n");
DEBUG(dbgs() << MF->getName()
<< ":BB#" << MBB->getNumber() << " " << MBB->getName()
<< "\n From: " << *I << " To: ";
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'.
}
assert(RemainingInstrs == 0 && "Instruction count mismatch!");
Scheduler.finishBlock();
- if (Scheduler.isPostRA()) {
- // FIXME: Ideally, no further passes should rely on kill flags. However,
- // thumb2 size reduction is currently an exception.
- Scheduler.fixupKills(MBB);
- }
+ // FIXME: Ideally, no further passes should rely on kill flags. However,
+ // thumb2 size reduction is currently an exception, so the PostMIScheduler
+ // needs to do this.
+ if (FixKillFlags)
+ Scheduler.fixupKills(&*MBB);
}
Scheduler.finalizeSchedule();
}
// 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
/// 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);
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;
updatePressureDiffs(LiveUses);
}
+ DEBUG(
+ dbgs() << "Top Pressure:\n";
+ dumpRegSetPressure(TopRPTracker.getRegSetPressureAtPos(), TRI);
+ dbgs() << "Bottom Pressure:\n";
+ dumpRegSetPressure(BotRPTracker.getRegSetPressureAtPos(), TRI);
+ );
+
assert(BotRPTracker.getPos() == RegionEnd && "Can't find the region bottom");
// Cache the list of excess pressure sets in this region. This will also track
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");
}
}
}
}
// RegisterPressureTracker guarantees that readsReg is true for LiveUses.
assert(VNI && "No live value at use.");
- for (VReg2UseMap::iterator
- UI = VRegUses.find(Reg); UI != VRegUses.end(); ++UI) {
- SUnit *SU = UI->SU;
- DEBUG(dbgs() << " UpdateRegP: SU(" << SU->NodeNum << ") "
- << *SU->getInstr());
+ for (const VReg2SUnit &V2SU
+ : make_range(VRegUses.find(Reg), VRegUses.end())) {
+ SUnit *SU = V2SU.SU;
// If this use comes before the reaching def, it cannot be a last use, so
// descrease its pressure change.
if (!SU->isScheduled && SU != &ExitSU) {
LiveQueryResult LRQ
= LI.Query(LIS->getInstructionIndex(SU->getInstr()));
- if (LRQ.valueIn() == VNI)
- getPressureDiff(SU).addPressureChange(Reg, true, &MRI);
+ if (LRQ.valueIn() == VNI) {
+ PressureDiff &PDiff = getPressureDiff(SU);
+ PDiff.addPressureChange(Reg, true, &MRI);
+ DEBUG(
+ dbgs() << " UpdateRegP: SU(" << SU->NodeNum << ") "
+ << *SU->getInstr();
+ dbgs() << " to ";
+ PDiff.dump(*TRI);
+ );
+ }
}
}
}
/// 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();
// This may initialize a DFSResult to be used for queue priority.
SchedImpl->initialize(this);
- DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
- SUnits[su].dumpAll(this));
+ DEBUG(
+ for (const SUnit &SU : SUnits) {
+ SU.dumpAll(this);
+ if (ShouldTrackPressure) {
+ dbgs() << " Pressure Diff : ";
+ getPressureDiff(&SU).dump(*TRI);
+ }
+ dbgs() << '\n';
+ }
+ );
if (ViewMISchedDAGs) viewGraph();
// Initialize ready queues now that the DAG and priority data are finalized.
}
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)) {
// Notify the scheduling strategy after updating the DAG.
SchedImpl->schedNode(SU, IsTopNode);
+
+ updateQueues(SU, IsTopNode);
}
assert(CurrentTop == CurrentBottom && "Nonempty unscheduled zone.");
unsigned LiveOutHeight = DefSU->getHeight();
unsigned LiveOutDepth = DefSU->getDepth() + DefSU->Latency;
// Visit all local users of the vreg def.
- for (VReg2UseMap::iterator
- UI = VRegUses.find(Reg); UI != VRegUses.end(); ++UI) {
- if (UI->SU == &ExitSU)
+ for (const VReg2SUnit &V2SU
+ : make_range(VRegUses.find(Reg), VRegUses.end())) {
+ SUnit *SU = V2SU.SU;
+ if (SU == &ExitSU)
continue;
// Only consider uses of the phi.
LiveQueryResult LRQ =
- LI.Query(LIS->getInstructionIndex(UI->SU->getInstr()));
+ LI.Query(LIS->getInstructionIndex(SU->getInstr()));
if (!LRQ.valueIn()->isPHIDef())
continue;
// overestimate in strange cases. This allows cyclic latency to be
// estimated as the minimum slack of the vreg's depth or height.
unsigned CyclicLatency = 0;
- if (LiveOutDepth > UI->SU->getDepth())
- CyclicLatency = LiveOutDepth - UI->SU->getDepth();
+ if (LiveOutDepth > SU->getDepth())
+ CyclicLatency = LiveOutDepth - SU->getDepth();
- unsigned LiveInHeight = UI->SU->getHeight() + DefSU->Latency;
+ unsigned LiveInHeight = SU->getHeight() + DefSU->Latency;
if (LiveInHeight > LiveOutHeight) {
if (LiveInHeight - LiveOutHeight < CyclicLatency)
CyclicLatency = LiveInHeight - LiveOutHeight;
CyclicLatency = 0;
DEBUG(dbgs() << "Cyclic Path: SU(" << DefSU->NodeNum << ") -> SU("
- << UI->SU->NodeNum << ") = " << CyclicLatency << "c\n");
+ << SU->NodeNum << ") = " << CyclicLatency << "c\n");
if (CyclicLatency > MaxCyclicLatency)
MaxCyclicLatency = CyclicLatency;
}
// Update top scheduled pressure.
TopRPTracker.advance();
assert(TopRPTracker.getPos() == CurrentTop && "out of sync");
+ DEBUG(
+ dbgs() << "Top Pressure:\n";
+ dumpRegSetPressure(TopRPTracker.getRegSetPressureAtPos(), TRI);
+ );
+
updateScheduledPressure(SU, TopRPTracker.getPressure().MaxSetPressure);
}
}
SmallVector<unsigned, 8> LiveUses;
BotRPTracker.recede(&LiveUses);
assert(BotRPTracker.getPos() == CurrentBottom && "out of sync");
+ DEBUG(
+ dbgs() << "Bottom Pressure:\n";
+ dumpRegSetPressure(BotRPTracker.getRegSetPressureAtPos(), TRI);
+ );
+
updateScheduledPressure(SU, BotRPTracker.getPressure().MaxSetPressure);
updatePressureDiffs(LiveUses);
}
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)
/// \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
// 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;
}
}
// 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;
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;
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;
}
}
}
}
}
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();
}
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.
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() << " ";
// 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);
}
}
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
RegionPolicy.OnlyBottomUp = true;
// Allow the subtarget to override default policy.
- const TargetSubtargetInfo &ST = TM.getSubtarget<TargetSubtargetInfo>();
- ST.overrideSchedPolicy(RegionPolicy, Begin, End, NumRegionInstrs);
+ MF.getSubtarget().overrideSchedPolicy(RegionPolicy, Begin, End,
+ NumRegionInstrs);
// After subtarget overrides, apply command line options.
if (!EnableRegPressure)
}
}
+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:
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();
const PressureChange &CandP,
GenericSchedulerBase::SchedCandidate &TryCand,
GenericSchedulerBase::SchedCandidate &Cand,
- GenericSchedulerBase::CandReason Reason) {
- int TryRank = TryP.getPSetOrMax();
- int CandRank = CandP.getPSetOrMax();
+ GenericSchedulerBase::CandReason Reason,
+ const TargetRegisterInfo *TRI,
+ const MachineFunction &MF) {
+ unsigned TryPSet = TryP.getPSetOrMax();
+ unsigned CandPSet = CandP.getPSetOrMax();
// If both candidates affect the same set, go with the smallest increase.
- if (TryRank == CandRank) {
+ if (TryPSet == CandPSet) {
return tryLess(TryP.getUnitInc(), CandP.getUnitInc(), TryCand, Cand,
Reason);
}
// 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;
}
+
+ int TryRank = TryP.isValid() ? TRI->getRegPressureSetScore(MF, TryPSet) :
+ std::numeric_limits<int>::max();
+
+ int CandRank = CandP.isValid() ? TRI->getRegPressureSetScore(MF, CandPSet) :
+ std::numeric_limits<int>::max();
+
// If the candidates are decreasing pressure, reverse priority.
if (TryP.getUnitInc() < 0)
std::swap(TryRank, CandRank);
}
}
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");
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))
+ TryCand, Cand, RegExcess, TRI,
+ DAG->MF))
return;
// Avoid increasing the max critical pressure in the scheduled region.
if (DAG->isTrackingPressure() && tryPressure(TryCand.RPDelta.CriticalMax,
Cand.RPDelta.CriticalMax,
- TryCand, Cand, RegCritical))
+ TryCand, Cand, RegCritical, TRI,
+ DAG->MF))
return;
// For loops that are acyclic path limited, aggressively schedule for latency.
// Avoid increasing the max pressure of the entire region.
if (DAG->isTrackingPressure() && tryPressure(TryCand.RPDelta.CurrentMax,
Cand.RPDelta.CurrentMax,
- TryCand, Cand, RegMax))
+ TryCand, Cand, RegMax, TRI,
+ DAG->MF))
return;
// Avoid critical resource consumption and balance the schedule.
// 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;
}
// efficient, but also provides the best heuristics for CriticalPSets.
if (SUnit *SU = Bot.pickOnlyChoice()) {
IsTopNode = false;
- DEBUG(dbgs() << "Pick Bot NOCAND\n");
+ DEBUG(dbgs() << "Pick Bot ONLY1\n");
return SU;
}
if (SUnit *SU = Top.pickOnlyChoice()) {
IsTopNode = true;
- DEBUG(dbgs() << "Pick Top NOCAND\n");
+ DEBUG(dbgs() << "Pick Top ONLY1\n");
return SU;
}
CandPolicy NoPolicy;
if (EnableLoadCluster && DAG->TII->enableClusterLoads())
DAG->addMutation(make_unique<LoadClusterMutation>(DAG->TII, DAG->TRI));
if (EnableMacroFusion)
- DAG->addMutation(make_unique<MacroFusion>(DAG->TII));
+ DAG->addMutation(make_unique<MacroFusion>(*DAG->TII, *DAG->TRI));
return 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 ((*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.
}
static bool isNodeHidden(const SUnit *Node) {
- return (Node->Preds.size() > 10 || Node->Succs.size() > 10);
- }
-
- static bool hasNodeAddressLabel(const SUnit *Node,
- const ScheduleDAG *Graph) {
- return false;
+ if (ViewMISchedCutoff == 0)
+ return false;
+ return (Node->Preds.size() > ViewMISchedCutoff
+ || Node->Succs.size() > ViewMISchedCutoff);
}
/// If you want to override the dot attributes printed for a particular