//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "misched"
-
#include "llvm/CodeGen/MachineScheduler.h"
-#include "llvm/ADT/OwningPtr.h"
#include "llvm/ADT/PriorityQueue.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/LiveIntervalAnalysis.h"
using namespace llvm;
+#define DEBUG_TYPE "misched"
+
namespace llvm {
cl::opt<bool> ForceTopDown("misched-topdown", cl::Hidden,
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));
+
+static cl::opt<std::string> SchedOnlyFunc("misched-only-func", cl::Hidden,
+ cl::desc("Only schedule this function"));
+static cl::opt<unsigned> SchedOnlyBlock("misched-only-block", cl::Hidden,
+ cl::desc("Only schedule this MBB#"));
#else
static bool ViewMISchedDAGs = false;
#endif // NDEBUG
//===----------------------------------------------------------------------===//
MachineSchedContext::MachineSchedContext():
- MF(0), MLI(0), MDT(0), PassConfig(0), AA(0), LIS(0) {
+ MF(nullptr), MLI(nullptr), MDT(nullptr), PassConfig(nullptr), AA(nullptr), LIS(nullptr) {
RegClassInfo = new RegisterClassInfo();
}
public:
MachineSchedulerBase(char &ID): MachineFunctionPass(ID) {}
- virtual void print(raw_ostream &O, const Module* = 0) const;
+ 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.
public:
MachineScheduler();
- virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+ void getAnalysisUsage(AnalysisUsage &AU) const override;
- virtual bool runOnMachineFunction(MachineFunction&);
+ bool runOnMachineFunction(MachineFunction&) override;
static char ID; // Class identification, replacement for typeinfo
public:
PostMachineScheduler();
- virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+ void getAnalysisUsage(AnalysisUsage &AU) const override;
- virtual bool runOnMachineFunction(MachineFunction&);
+ bool runOnMachineFunction(MachineFunction&) override;
static char ID; // Class identification, replacement for typeinfo
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>();
char &llvm::PostMachineSchedulerID = PostMachineScheduler::ID;
INITIALIZE_PASS(PostMachineScheduler, "postmisched",
- "PostRA Machine Instruction Scheduler", false, false);
+ "PostRA Machine Instruction Scheduler", false, false)
PostMachineScheduler::PostMachineScheduler()
: MachineSchedulerBase(ID) {
/// A dummy default scheduler factory indicates whether the scheduler
/// is overridden on the command line.
static ScheduleDAGInstrs *useDefaultMachineSched(MachineSchedContext *C) {
- return 0;
+ return nullptr;
}
/// MachineSchedOpt allows command line selection of the scheduler.
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 *createGenericSched(MachineSchedContext *C);
-static ScheduleDAGInstrs *createRawGenericSched(MachineSchedContext *C);
+static ScheduleDAGInstrs *createGenericSchedLive(MachineSchedContext *C);
+static ScheduleDAGInstrs *createGenericSchedPostRA(MachineSchedContext *C);
/// Decrement this iterator until reaching the top or a non-debug instr.
static MachineBasicBlock::const_iterator
return Scheduler;
// Default to GenericScheduler.
- return createGenericSched(this);
+ return createGenericSchedLive(this);
}
/// Instantiate a ScheduleDAGInstrs for PostRA scheduling that will be owned by
return Scheduler;
// Default to GenericScheduler.
- // return createRawGenericSched(this);
- return NULL;
+ return createGenericSchedPostRA(this);
}
/// Top-level MachineScheduler pass driver.
/// 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.
- OwningPtr<ScheduleDAGInstrs> Scheduler(createMachineScheduler());
- scheduleRegions(*Scheduler);
+ std::unique_ptr<ScheduleDAGInstrs> Scheduler(createMachineScheduler());
+ scheduleRegions(*Scheduler, false);
DEBUG(LIS->dump());
if (VerifyScheduling)
}
bool PostMachineScheduler::runOnMachineFunction(MachineFunction &mf) {
+ if (skipOptnoneFunction(*mf.getFunction()))
+ return false;
+
+ if (!mf.getSubtarget().enablePostRAScheduler()) {
+ DEBUG(dbgs() << "Subtarget disables post-MI-sched.\n");
+ return false;
+ }
DEBUG(dbgs() << "Before post-MI-sched:\n"; mf.print(dbgs()));
// Initialize the context of the pass.
// Instantiate the selected scheduler for this target, function, and
// optimization level.
- OwningPtr<ScheduleDAGInstrs> Scheduler(createPostMachineScheduler());
- scheduleRegions(*Scheduler);
+ std::unique_ptr<ScheduleDAGInstrs> Scheduler(createPostMachineScheduler());
+ scheduleRegions(*Scheduler, true);
if (VerifyScheduling)
MF->verify(this, "After post machine scheduling.");
return true;
}
+/// Return true of the given instruction should not be included in a scheduling
+/// region.
+///
+/// MachineScheduler does not currently support scheduling across calls. To
+/// handle calls, the DAG builder needs to be modified to create register
+/// anti/output dependencies on the registers clobbered by the call's regmask
+/// operand. In PreRA scheduling, the stack pointer adjustment already prevents
+/// scheduling across calls. In PostRA scheduling, we need the isCall to enforce
+/// the boundary, but there would be no benefit to postRA scheduling across
+/// calls this late anyway.
+static bool isSchedBoundary(MachineBasicBlock::iterator MI,
+ MachineBasicBlock *MBB,
+ MachineFunction *MF,
+ 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();
+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())
+ continue;
+ if (SchedOnlyBlock.getNumOccurrences()
+ && (int)SchedOnlyBlock != MBB->getNumber())
+ continue;
+#endif
// Break the block into scheduling regions [I, RegionEnd), and schedule each
// region as soon as it is discovered. RegionEnd points the scheduling
// The Scheduler may insert instructions during either schedule() or
// exitRegion(), even for empty regions. So the local iterators 'I' and
// 'RegionEnd' are invalid across these calls.
- unsigned RemainingInstrs = MBB->size();
+ //
+ // MBB::size() uses instr_iterator to count. Here we need a bundle to count
+ // as a single instruction.
+ unsigned RemainingInstrs = std::distance(MBB->begin(), MBB->end());
for(MachineBasicBlock::iterator RegionEnd = MBB->end();
RegionEnd != MBB->begin(); RegionEnd = Scheduler.begin()) {
// Avoid decrementing RegionEnd for blocks with no terminator.
- if (RegionEnd != MBB->end()
- || TII->isSchedulingBoundary(llvm::prior(RegionEnd), MBB, *MF)) {
+ if (RegionEnd != MBB->end() ||
+ 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 (TII->isSchedulingBoundary(llvm::prior(I), MBB, *MF))
+ 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 == llvm::prior(RegionEnd)) {
+ if (I == RegionEnd || I == std::prev(RegionEnd)) {
// Close the current region. Bundle the terminator if needed.
// This invalidates 'RegionEnd' and 'I'.
Scheduler.exitRegion();
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();
+ // 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
// virtual registers.
// ===----------------------------------------------------------------------===/
+// Provide a vtable anchor.
ScheduleDAGMI::~ScheduleDAGMI() {
- DeleteContainerPointers(Mutations);
- delete SchedImpl;
}
bool ScheduleDAGMI::canAddEdge(SUnit *SuccSU, SUnit *PredSU) {
dbgs() << "*** Scheduling failed! ***\n";
SuccSU->dump(this);
dbgs() << " has been released too many times!\n";
- llvm_unreachable(0);
+ llvm_unreachable(nullptr);
}
#endif
+ // SU->TopReadyCycle was set to CurrCycle when it was scheduled. However,
+ // CurrCycle may have advanced since then.
+ if (SuccSU->TopReadyCycle < SU->TopReadyCycle + SuccEdge->getLatency())
+ SuccSU->TopReadyCycle = SU->TopReadyCycle + SuccEdge->getLatency();
+
--SuccSU->NumPredsLeft;
if (SuccSU->NumPredsLeft == 0 && SuccSU != &ExitSU)
SchedImpl->releaseTopNode(SuccSU);
dbgs() << "*** Scheduling failed! ***\n";
PredSU->dump(this);
dbgs() << " has been released too many times!\n";
- llvm_unreachable(0);
+ llvm_unreachable(nullptr);
}
#endif
+ // SU->BotReadyCycle was set to CurrCycle when it was scheduled. However,
+ // CurrCycle may have advanced since then.
+ if (PredSU->BotReadyCycle < SU->BotReadyCycle + PredEdge->getLatency())
+ PredSU->BotReadyCycle = SU->BotReadyCycle + PredEdge->getLatency();
+
--PredSU->NumSuccsLeft;
if (PredSU->NumSuccsLeft == 0 && PredSU != &EntrySU)
SchedImpl->releaseBottomNode(PredSU);
/// 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;
CurrentBottom = MI;
}
}
- updateQueues(SU, IsTopNode);
-
- // Notify the scheduling strategy after updating the DAG.
+ // Notify the scheduling strategy before updating the DAG.
+ // 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);
+
+ updateQueues(SU, IsTopNode);
}
assert(CurrentTop == CurrentBottom && "Nonempty unscheduled zone.");
/// Identify DAG roots and setup scheduler queues.
void ScheduleDAGMI::initQueues(ArrayRef<SUnit*> TopRoots,
ArrayRef<SUnit*> BotRoots) {
- NextClusterSucc = NULL;
- NextClusterPred = NULL;
+ NextClusterSucc = nullptr;
+ NextClusterPred = nullptr;
// Release all DAG roots for scheduling, not including EntrySU/ExitSU.
//
for (std::vector<std::pair<MachineInstr *, MachineInstr *> >::iterator
DI = DbgValues.end(), DE = DbgValues.begin(); DI != DE; --DI) {
- std::pair<MachineInstr *, MachineInstr *> P = *prior(DI);
+ std::pair<MachineInstr *, MachineInstr *> P = *std::prev(DI);
MachineInstr *DbgValue = P.first;
MachineBasicBlock::iterator OrigPrevMI = P.second;
if (&*RegionBegin == DbgValue)
++RegionBegin;
BB->splice(++OrigPrevMI, BB, DbgValue);
- if (OrigPrevMI == llvm::prior(RegionEnd))
+ if (OrigPrevMI == std::prev(RegionEnd))
RegionEnd = DbgValue;
}
DbgValues.clear();
- FirstDbgValue = NULL;
+ FirstDbgValue = nullptr;
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
ScheduleDAGMI::enterRegion(bb, begin, end, regioninstrs);
// For convenience remember the end of the liveness region.
- LiveRegionEnd =
- (RegionEnd == bb->end()) ? RegionEnd : llvm::next(RegionEnd);
+ LiveRegionEnd = (RegionEnd == bb->end()) ? RegionEnd : std::next(RegionEnd);
SUPressureDiffs.clear();
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);
}
unsigned Offset;
LoadInfo(SUnit *su, unsigned reg, unsigned ofs)
: SU(su), BaseReg(reg), Offset(ofs) {}
+
+ bool operator<(const LoadInfo &RHS) const {
+ return std::tie(BaseReg, Offset) < std::tie(RHS.BaseReg, RHS.Offset);
+ }
};
- static bool LoadInfoLess(const LoadClusterMutation::LoadInfo &LHS,
- const LoadClusterMutation::LoadInfo &RHS);
const TargetInstrInfo *TII;
const TargetRegisterInfo *TRI;
const TargetRegisterInfo *tri)
: TII(tii), TRI(tri) {}
- virtual void apply(ScheduleDAGMI *DAG);
+ void apply(ScheduleDAGMI *DAG) override;
protected:
void clusterNeighboringLoads(ArrayRef<SUnit*> Loads, ScheduleDAGMI *DAG);
};
} // anonymous
-bool LoadClusterMutation::LoadInfoLess(
- const LoadClusterMutation::LoadInfo &LHS,
- const LoadClusterMutation::LoadInfo &RHS) {
- if (LHS.BaseReg != RHS.BaseReg)
- return LHS.BaseReg < RHS.BaseReg;
- return LHS.Offset < RHS.Offset;
-}
-
void LoadClusterMutation::clusterNeighboringLoads(ArrayRef<SUnit*> Loads,
ScheduleDAGMI *DAG) {
SmallVector<LoadClusterMutation::LoadInfo,32> LoadRecords;
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)
return;
- std::sort(LoadRecords.begin(), LoadRecords.end(), LoadInfoLess);
+ std::sort(LoadRecords.begin(), LoadRecords.end());
unsigned ClusterLength = 1;
for (unsigned Idx = 0, End = LoadRecords.size(); Idx < (End - 1); ++Idx) {
if (LoadRecords[Idx].BaseReg != LoadRecords[Idx+1].BaseReg) {
/// \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) {}
- virtual void apply(ScheduleDAGMI *DAG);
+ 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;
}
}
public:
CopyConstrain(const TargetInstrInfo *, const TargetRegisterInfo *) {}
- virtual void apply(ScheduleDAGMI *DAG);
+ void apply(ScheduleDAGMI *DAG) override;
protected:
void 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;
// Check if GlobalLI contains a hole in the vicinity of LocalLI.
if (GlobalSegment != GlobalLI->begin()) {
// Two address defs have no hole.
- if (SlotIndex::isSameInstr(llvm::prior(GlobalSegment)->end,
+ if (SlotIndex::isSameInstr(std::prev(GlobalSegment)->end,
GlobalSegment->start)) {
return;
}
// If the prior global segment may be defined by the same two-address
// instruction that also defines LocalLI, then can't make a hole here.
- if (SlotIndex::isSameInstr(llvm::prior(GlobalSegment)->start,
+ if (SlotIndex::isSameInstr(std::prev(GlobalSegment)->start,
LocalLI->beginIndex())) {
return;
}
// If GlobalLI has a prior segment, it must be live into the EBB. Otherwise
// it would be a disconnected component in the live range.
- assert(llvm::prior(GlobalSegment)->start < LocalLI->beginIndex() &&
+ assert(std::prev(GlobalSegment)->start < LocalLI->beginIndex() &&
"Disconnected LRG within the scheduling region.");
}
MachineInstr *GlobalDef = LIS->getInstructionFromIndex(GlobalSegment->start);
//===----------------------------------------------------------------------===//
// MachineSchedStrategy helpers used by GenericScheduler, GenericPostScheduler
// and possibly other custom schedulers.
-// ===----------------------------------------------------------------------===/
+//===----------------------------------------------------------------------===//
static const unsigned InvalidCycle = ~0U;
// invalid, placeholder HazardRecs.
if (HazardRec && HazardRec->isEnabled()) {
delete HazardRec;
- HazardRec = 0;
+ HazardRec = nullptr;
}
Available.clear();
Pending.clear();
IsResourceLimited = false;
ReservedCycles.clear();
#ifndef NDEBUG
- MaxObservedLatency = 0;
+ // Track the maximum number of stall cycles that could arise either from the
+ // latency of a DAG edge or the number of cycles that a processor resource is
+ // reserved (SchedBoundary::ReservedCycles).
+ MaxObservedStall = 0;
#endif
// Reserve a zero-count for invalid CritResIdx.
ExecutedResCounts.resize(1);
///
/// TODO: Also check whether the SU must start a new group.
bool SchedBoundary::checkHazard(SUnit *SU) {
- if (HazardRec->isEnabled())
- return HazardRec->getHazardType(SU) != ScheduleHazardRecognizer::NoHazard;
-
+ if (HazardRec->isEnabled()
+ && HazardRec->getHazardType(SU) != ScheduleHazardRecognizer::NoHazard) {
+ return true;
+ }
unsigned uops = SchedModel->getNumMicroOps(SU->getInstr());
if ((CurrMOps > 0) && (CurrMOps + uops > SchedModel->getIssueWidth())) {
DEBUG(dbgs() << " SU(" << SU->NodeNum << ") uops="
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;
// Find the unscheduled node in ReadySUs with the highest latency.
unsigned SchedBoundary::
findMaxLatency(ArrayRef<SUnit*> ReadySUs) {
- SUnit *LateSU = 0;
+ SUnit *LateSU = nullptr;
unsigned RemLatency = 0;
for (ArrayRef<SUnit*>::iterator I = ReadySUs.begin(), E = ReadySUs.end();
I != E; ++I) {
}
void SchedBoundary::releaseNode(SUnit *SU, unsigned ReadyCycle) {
+ assert(SU->getInstr() && "Scheduled SUnit must have instr");
+
+#ifndef NDEBUG
+ // 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)
MinReadyCycle = ReadyCycle;
if (SU->isScheduled)
return;
- for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
- I != E; ++I) {
- if (I->isWeak())
- continue;
- unsigned PredReadyCycle = I->getSUnit()->TopReadyCycle;
- unsigned Latency = I->getLatency();
-#ifndef NDEBUG
- MaxObservedLatency = std::max(Latency, MaxObservedLatency);
-#endif
- if (SU->TopReadyCycle < PredReadyCycle + Latency)
- SU->TopReadyCycle = PredReadyCycle + Latency;
- }
releaseNode(SU, SU->TopReadyCycle);
}
if (SU->isScheduled)
return;
- assert(SU->getInstr() && "Scheduled SUnit must have instr");
-
- for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
- I != E; ++I) {
- if (I->isWeak())
- continue;
- unsigned SuccReadyCycle = I->getSUnit()->BotReadyCycle;
- unsigned Latency = I->getLatency();
-#ifndef NDEBUG
- MaxObservedLatency = std::max(Latency, MaxObservedLatency);
-#endif
- if (SU->BotReadyCycle < SuccReadyCycle + Latency)
- SU->BotReadyCycle = SuccReadyCycle + Latency;
- }
releaseNode(SU, SU->BotReadyCycle);
}
PI = SchedModel->getWriteProcResBegin(SC),
PE = SchedModel->getWriteProcResEnd(SC); PI != PE; ++PI) {
unsigned PIdx = PI->ProcResourceIdx;
- if (SchedModel->getProcResource(PIdx)->BufferSize == 0)
- ReservedCycles[PIdx] = isTop() ? NextCycle + PI->Cycles : NextCycle;
+ if (SchedModel->getProcResource(PIdx)->BufferSize == 0) {
+ if (isTop()) {
+ ReservedCycles[PIdx] =
+ std::max(getNextResourceCycle(PIdx, 0), NextCycle + PI->Cycles);
+ }
+ else
+ ReservedCycles[PIdx] = NextCycle;
+ }
}
}
}
}
else {
// After updating ZoneCritResIdx and ExpectedLatency, check if we're
- // resource limited. If a stall occured, bumpCycle does this.
+ // resource limited. If a stall occurred, bumpCycle does this.
unsigned LFactor = SchedModel->getLatencyFactor();
IsResourceLimited =
(int)(getCriticalCount() - (getScheduledLatency() * LFactor))
// bump the cycle to avoid uselessly checking everything in the readyQ.
CurrMOps += IncMOps;
while (CurrMOps >= SchedModel->getIssueWidth()) {
- bumpCycle(++NextCycle);
DEBUG(dbgs() << " *** Max MOps " << CurrMOps
<< " at cycle " << CurrCycle << '\n');
+ bumpCycle(++NextCycle);
}
DEBUG(dumpScheduledState());
}
}
}
for (unsigned i = 0; Available.empty(); ++i) {
- assert(i <= (HazardRec->getMaxLookAhead() + MaxObservedLatency) &&
- "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();
}
if (Available.size() == 1)
return *Available.begin();
- return NULL;
+ return nullptr;
}
#ifndef NDEBUG
#endif
//===----------------------------------------------------------------------===//
-// GenericScheduler - Implementation of the generic MachineSchedStrategy.
+// GenericScheduler - Generic implementation of MachineSchedStrategy.
//===----------------------------------------------------------------------===//
-namespace {
-/// GenericScheduler shrinks the unscheduled zone using heuristics to balance
-/// the schedule.
-class GenericScheduler : public MachineSchedStrategy {
-public:
- /// Represent the type of SchedCandidate found within a single queue.
- /// pickNodeBidirectional depends on these listed by decreasing priority.
- enum CandReason {
- NoCand, PhysRegCopy, RegExcess, RegCritical, Stall, Cluster, Weak, RegMax,
- ResourceReduce, ResourceDemand, BotHeightReduce, BotPathReduce,
- TopDepthReduce, TopPathReduce, NextDefUse, NodeOrder};
-
-#ifndef NDEBUG
- static const char *getReasonStr(GenericScheduler::CandReason Reason);
-#endif
-
- /// Policy for scheduling the next instruction in the candidate's zone.
- struct CandPolicy {
- bool ReduceLatency;
- unsigned ReduceResIdx;
- unsigned DemandResIdx;
+void GenericSchedulerBase::SchedCandidate::
+initResourceDelta(const ScheduleDAGMI *DAG,
+ const TargetSchedModel *SchedModel) {
+ if (!Policy.ReduceResIdx && !Policy.DemandResIdx)
+ return;
- CandPolicy(): ReduceLatency(false), ReduceResIdx(0), DemandResIdx(0) {}
- };
+ const MCSchedClassDesc *SC = DAG->getSchedClass(SU);
+ for (TargetSchedModel::ProcResIter
+ PI = SchedModel->getWriteProcResBegin(SC),
+ PE = SchedModel->getWriteProcResEnd(SC); PI != PE; ++PI) {
+ if (PI->ProcResourceIdx == Policy.ReduceResIdx)
+ ResDelta.CritResources += PI->Cycles;
+ if (PI->ProcResourceIdx == Policy.DemandResIdx)
+ ResDelta.DemandedResources += PI->Cycles;
+ }
+}
- /// Status of an instruction's critical resource consumption.
- struct SchedResourceDelta {
- // Count critical resources in the scheduled region required by SU.
- unsigned CritResources;
+/// Set the CandPolicy given a scheduling zone given the current resources and
+/// latencies inside and outside the zone.
+void GenericSchedulerBase::setPolicy(CandPolicy &Policy,
+ bool IsPostRA,
+ SchedBoundary &CurrZone,
+ SchedBoundary *OtherZone) {
+ // Apply preemptive heuristics based on the total latency and resources
+ // inside and outside this zone. Potential stalls should be considered before
+ // following this policy.
- // Count critical resources from another region consumed by SU.
- unsigned DemandedResources;
+ // Compute remaining latency. We need this both to determine whether the
+ // overall schedule has become latency-limited and whether the instructions
+ // outside this zone are resource or latency limited.
+ //
+ // The "dependent" latency is updated incrementally during scheduling as the
+ // max height/depth of scheduled nodes minus the cycles since it was
+ // scheduled:
+ // DLat = max (N.depth - (CurrCycle - N.ReadyCycle) for N in Zone
+ //
+ // The "independent" latency is the max ready queue depth:
+ // ILat = max N.depth for N in Available|Pending
+ //
+ // RemainingLatency is the greater of independent and dependent latency.
+ unsigned RemLatency = CurrZone.getDependentLatency();
+ RemLatency = std::max(RemLatency,
+ CurrZone.findMaxLatency(CurrZone.Available.elements()));
+ RemLatency = std::max(RemLatency,
+ CurrZone.findMaxLatency(CurrZone.Pending.elements()));
- SchedResourceDelta(): CritResources(0), DemandedResources(0) {}
+ // Compute the critical resource outside the zone.
+ unsigned OtherCritIdx = 0;
+ unsigned OtherCount =
+ OtherZone ? OtherZone->getOtherResourceCount(OtherCritIdx) : 0;
- bool operator==(const SchedResourceDelta &RHS) const {
- return CritResources == RHS.CritResources
- && DemandedResources == RHS.DemandedResources;
- }
- bool operator!=(const SchedResourceDelta &RHS) const {
- return !operator==(RHS);
+ bool OtherResLimited = false;
+ if (SchedModel->hasInstrSchedModel()) {
+ unsigned LFactor = SchedModel->getLatencyFactor();
+ OtherResLimited = (int)(OtherCount - (RemLatency * LFactor)) > (int)LFactor;
+ }
+ // Schedule aggressively for latency in PostRA mode. We don't check for
+ // acyclic latency during PostRA, and highly out-of-order processors will
+ // skip PostRA scheduling.
+ if (!OtherResLimited) {
+ if (IsPostRA || (RemLatency + CurrZone.getCurrCycle() > Rem.CriticalPath)) {
+ Policy.ReduceLatency |= true;
+ DEBUG(dbgs() << " " << CurrZone.Available.getName()
+ << " RemainingLatency " << RemLatency << " + "
+ << CurrZone.getCurrCycle() << "c > CritPath "
+ << Rem.CriticalPath << "\n");
}
- };
-
- /// Store the state used by GenericScheduler heuristics, required for the
- /// lifetime of one invocation of pickNode().
- struct SchedCandidate {
- CandPolicy Policy;
-
- // The best SUnit candidate.
- SUnit *SU;
-
- // The reason for this candidate.
- CandReason Reason;
-
- // Set of reasons that apply to multiple candidates.
- uint32_t RepeatReasonSet;
-
- // Register pressure values for the best candidate.
- RegPressureDelta RPDelta;
-
- // Critical resource consumption of the best candidate.
- SchedResourceDelta ResDelta;
-
- SchedCandidate(const CandPolicy &policy)
- : Policy(policy), SU(NULL), Reason(NoCand), RepeatReasonSet(0) {}
-
- bool isValid() const { return SU; }
+ }
+ // If the same resource is limiting inside and outside the zone, do nothing.
+ if (CurrZone.getZoneCritResIdx() == OtherCritIdx)
+ return;
- // Copy the status of another candidate without changing policy.
- void setBest(SchedCandidate &Best) {
- assert(Best.Reason != NoCand && "uninitialized Sched candidate");
- SU = Best.SU;
- Reason = Best.Reason;
- RPDelta = Best.RPDelta;
- ResDelta = Best.ResDelta;
+ DEBUG(
+ if (CurrZone.isResourceLimited()) {
+ dbgs() << " " << CurrZone.Available.getName() << " ResourceLimited: "
+ << SchedModel->getResourceName(CurrZone.getZoneCritResIdx())
+ << "\n";
}
+ if (OtherResLimited)
+ dbgs() << " RemainingLimit: "
+ << SchedModel->getResourceName(OtherCritIdx) << "\n";
+ if (!CurrZone.isResourceLimited() && !OtherResLimited)
+ dbgs() << " Latency limited both directions.\n");
- bool isRepeat(CandReason R) { return RepeatReasonSet & (1 << R); }
- void setRepeat(CandReason R) { RepeatReasonSet |= (1 << R); }
-
- void initResourceDelta(const ScheduleDAGMILive *DAG,
- const TargetSchedModel *SchedModel);
- };
-
-private:
- const MachineSchedContext *Context;
- ScheduleDAGMILive *DAG;
- const TargetSchedModel *SchedModel;
- const TargetRegisterInfo *TRI;
-
- // State of the top and bottom scheduled instruction boundaries.
- SchedRemainder Rem;
- SchedBoundary Top;
- SchedBoundary Bot;
-
- MachineSchedPolicy RegionPolicy;
-public:
- GenericScheduler(const MachineSchedContext *C):
- Context(C), DAG(0), SchedModel(0), TRI(0),
- Top(SchedBoundary::TopQID, "TopQ"), Bot(SchedBoundary::BotQID, "BotQ") {}
-
- virtual void initPolicy(MachineBasicBlock::iterator Begin,
- MachineBasicBlock::iterator End,
- unsigned NumRegionInstrs);
-
- bool shouldTrackPressure() const { return RegionPolicy.ShouldTrackPressure; }
-
- virtual void initialize(ScheduleDAGMI *dag);
-
- virtual SUnit *pickNode(bool &IsTopNode);
-
- virtual void schedNode(SUnit *SU, bool IsTopNode);
-
- virtual void releaseTopNode(SUnit *SU) { Top.releaseTopNode(SU); }
-
- virtual void releaseBottomNode(SUnit *SU) { Bot.releaseBottomNode(SU); }
-
- virtual void registerRoots();
-
-protected:
- void checkAcyclicLatency();
-
- void setPolicy(CandPolicy &Policy, SchedBoundary &CurrZone,
- SchedBoundary &OtherZone);
-
- void tryCandidate(SchedCandidate &Cand,
- SchedCandidate &TryCand,
- SchedBoundary &Zone,
- const RegPressureTracker &RPTracker,
- RegPressureTracker &TempTracker);
-
- SUnit *pickNodeBidirectional(bool &IsTopNode);
-
- void pickNodeFromQueue(SchedBoundary &Zone,
- const RegPressureTracker &RPTracker,
- SchedCandidate &Candidate);
+ if (CurrZone.isResourceLimited() && !Policy.ReduceResIdx)
+ Policy.ReduceResIdx = CurrZone.getZoneCritResIdx();
- void reschedulePhysRegCopies(SUnit *SU, bool isTop);
+ if (OtherResLimited)
+ Policy.DemandResIdx = OtherCritIdx;
+}
#ifndef NDEBUG
- void traceCandidate(const SchedCandidate &Cand);
-#endif
-};
-} // namespace
-
-void GenericScheduler::initialize(ScheduleDAGMI *dag) {
- assert(dag->hasVRegLiveness() &&
- "(PreRA)GenericScheduler needs vreg liveness");
- DAG = static_cast<ScheduleDAGMILive*>(dag);
- SchedModel = DAG->getSchedModel();
- TRI = DAG->TRI;
-
- Rem.init(DAG, SchedModel);
- Top.init(DAG, SchedModel, &Rem);
- Bot.init(DAG, SchedModel, &Rem);
-
- // Initialize resource counts.
-
- // 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);
- }
- if (!Bot.HazardRec) {
- Bot.HazardRec =
- TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG);
- }
+const char *GenericSchedulerBase::getReasonStr(
+ GenericSchedulerBase::CandReason Reason) {
+ switch (Reason) {
+ case NoCand: return "NOCAND ";
+ case PhysRegCopy: return "PREG-COPY";
+ case RegExcess: return "REG-EXCESS";
+ case RegCritical: return "REG-CRIT ";
+ case Stall: return "STALL ";
+ case Cluster: return "CLUSTER ";
+ case Weak: return "WEAK ";
+ case RegMax: return "REG-MAX ";
+ case ResourceReduce: return "RES-REDUCE";
+ case ResourceDemand: return "RES-DEMAND";
+ case TopDepthReduce: return "TOP-DEPTH ";
+ case TopPathReduce: return "TOP-PATH ";
+ case BotHeightReduce:return "BOT-HEIGHT";
+ case BotPathReduce: return "BOT-PATH ";
+ case NextDefUse: return "DEF-USE ";
+ case NodeOrder: return "ORDER ";
+ };
+ llvm_unreachable("Unknown reason!");
}
-/// Initialize the per-region scheduling policy.
-void GenericScheduler::initPolicy(MachineBasicBlock::iterator Begin,
- MachineBasicBlock::iterator End,
- unsigned NumRegionInstrs) {
- const TargetMachine &TM = Context->MF->getTarget();
-
- // 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
- // schedulable instructions exceeds half the integer register file.
- unsigned NIntRegs = Context->RegClassInfo->getNumAllocatableRegs(
- TM.getTargetLowering()->getRegClassFor(MVT::i32));
-
- RegionPolicy.ShouldTrackPressure = NumRegionInstrs > (NIntRegs / 2);
-
- // For generic targets, we default to bottom-up, because it's simpler and more
- // compile-time optimizations have been implemented in that direction.
- RegionPolicy.OnlyBottomUp = true;
-
- // Allow the subtarget to override default policy.
- const TargetSubtargetInfo &ST = TM.getSubtarget<TargetSubtargetInfo>();
- ST.overrideSchedPolicy(RegionPolicy, Begin, End, NumRegionInstrs);
+void GenericSchedulerBase::traceCandidate(const SchedCandidate &Cand) {
+ PressureChange P;
+ unsigned ResIdx = 0;
+ unsigned Latency = 0;
+ switch (Cand.Reason) {
+ default:
+ break;
+ case RegExcess:
+ P = Cand.RPDelta.Excess;
+ break;
+ case RegCritical:
+ P = Cand.RPDelta.CriticalMax;
+ break;
+ case RegMax:
+ P = Cand.RPDelta.CurrentMax;
+ break;
+ case ResourceReduce:
+ ResIdx = Cand.Policy.ReduceResIdx;
+ break;
+ case ResourceDemand:
+ ResIdx = Cand.Policy.DemandResIdx;
+ break;
+ case TopDepthReduce:
+ Latency = Cand.SU->getDepth();
+ break;
+ case TopPathReduce:
+ Latency = Cand.SU->getHeight();
+ break;
+ case BotHeightReduce:
+ Latency = Cand.SU->getHeight();
+ break;
+ case BotPathReduce:
+ Latency = Cand.SU->getDepth();
+ break;
+ }
+ dbgs() << " Cand SU(" << Cand.SU->NodeNum << ") " << getReasonStr(Cand.Reason);
+ if (P.isValid())
+ dbgs() << " " << TRI->getRegPressureSetName(P.getPSet())
+ << ":" << P.getUnitInc() << " ";
+ else
+ dbgs() << " ";
+ if (ResIdx)
+ dbgs() << " " << SchedModel->getProcResource(ResIdx)->Name << " ";
+ else
+ dbgs() << " ";
+ if (Latency)
+ dbgs() << " " << Latency << " cycles ";
+ else
+ dbgs() << " ";
+ dbgs() << '\n';
+}
+#endif
+
+/// Return true if this heuristic determines order.
+static bool tryLess(int TryVal, int CandVal,
+ GenericSchedulerBase::SchedCandidate &TryCand,
+ GenericSchedulerBase::SchedCandidate &Cand,
+ GenericSchedulerBase::CandReason Reason) {
+ if (TryVal < CandVal) {
+ TryCand.Reason = Reason;
+ return true;
+ }
+ if (TryVal > CandVal) {
+ if (Cand.Reason > Reason)
+ Cand.Reason = Reason;
+ return true;
+ }
+ Cand.setRepeat(Reason);
+ return false;
+}
+
+static bool tryGreater(int TryVal, int CandVal,
+ GenericSchedulerBase::SchedCandidate &TryCand,
+ GenericSchedulerBase::SchedCandidate &Cand,
+ GenericSchedulerBase::CandReason Reason) {
+ if (TryVal > CandVal) {
+ TryCand.Reason = Reason;
+ return true;
+ }
+ if (TryVal < CandVal) {
+ if (Cand.Reason > Reason)
+ Cand.Reason = Reason;
+ return true;
+ }
+ Cand.setRepeat(Reason);
+ return false;
+}
+
+static bool tryLatency(GenericSchedulerBase::SchedCandidate &TryCand,
+ GenericSchedulerBase::SchedCandidate &Cand,
+ SchedBoundary &Zone) {
+ if (Zone.isTop()) {
+ if (Cand.SU->getDepth() > Zone.getScheduledLatency()) {
+ if (tryLess(TryCand.SU->getDepth(), Cand.SU->getDepth(),
+ TryCand, Cand, GenericSchedulerBase::TopDepthReduce))
+ return true;
+ }
+ if (tryGreater(TryCand.SU->getHeight(), Cand.SU->getHeight(),
+ TryCand, Cand, GenericSchedulerBase::TopPathReduce))
+ return true;
+ }
+ else {
+ if (Cand.SU->getHeight() > Zone.getScheduledLatency()) {
+ if (tryLess(TryCand.SU->getHeight(), Cand.SU->getHeight(),
+ TryCand, Cand, GenericSchedulerBase::BotHeightReduce))
+ return true;
+ }
+ if (tryGreater(TryCand.SU->getDepth(), Cand.SU->getDepth(),
+ TryCand, Cand, GenericSchedulerBase::BotPathReduce))
+ return true;
+ }
+ return false;
+}
+
+static void tracePick(const GenericSchedulerBase::SchedCandidate &Cand,
+ bool IsTop) {
+ DEBUG(dbgs() << "Pick " << (IsTop ? "Top " : "Bot ")
+ << GenericSchedulerBase::getReasonStr(Cand.Reason) << '\n');
+}
+
+void GenericScheduler::initialize(ScheduleDAGMI *dag) {
+ assert(dag->hasVRegLiveness() &&
+ "(PreRA)GenericScheduler needs vreg liveness");
+ DAG = static_cast<ScheduleDAGMILive*>(dag);
+ SchedModel = DAG->getSchedModel();
+ TRI = DAG->TRI;
+
+ Rem.init(DAG, SchedModel);
+ Top.init(DAG, SchedModel, &Rem);
+ Bot.init(DAG, SchedModel, &Rem);
+
+ // Initialize resource counts.
+
+ // Initialize the HazardRecognizers. If itineraries don't exist, are empty, or
+ // are disabled, then these HazardRecs will be disabled.
+ const InstrItineraryData *Itin = SchedModel->getInstrItineraries();
+ if (!Top.HazardRec) {
+ Top.HazardRec =
+ DAG->MF.getSubtarget().getInstrInfo()->CreateTargetMIHazardRecognizer(
+ Itin, DAG);
+ }
+ if (!Bot.HazardRec) {
+ Bot.HazardRec =
+ DAG->MF.getSubtarget().getInstrInfo()->CreateTargetMIHazardRecognizer(
+ Itin, DAG);
+ }
+}
+
+/// Initialize the per-region scheduling policy.
+void GenericScheduler::initPolicy(MachineBasicBlock::iterator Begin,
+ MachineBasicBlock::iterator End,
+ unsigned NumRegionInstrs) {
+ 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
+ // schedulable instructions exceeds half the integer register file.
+ RegionPolicy.ShouldTrackPressure = true;
+ for (unsigned VT = MVT::i32; VT > (unsigned)MVT::i1; --VT) {
+ MVT::SimpleValueType LegalIntVT = (MVT::SimpleValueType)VT;
+ if (TLI->isTypeLegal(LegalIntVT)) {
+ unsigned NIntRegs = Context->RegClassInfo->getNumAllocatableRegs(
+ TLI->getRegClassFor(LegalIntVT));
+ RegionPolicy.ShouldTrackPressure = NumRegionInstrs > (NIntRegs / 2);
+ }
+ }
+
+ // For generic targets, we default to bottom-up, because it's simpler and more
+ // compile-time optimizations have been implemented in that direction.
+ RegionPolicy.OnlyBottomUp = true;
+
+ // Allow the subtarget to override default policy.
+ 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();
}
}
-/// Set the CandPolicy given a scheduling zone given the current resources and
-/// latencies inside and outside the zone.
-void GenericScheduler::setPolicy(CandPolicy &Policy, SchedBoundary &CurrZone,
- SchedBoundary &OtherZone) {
- // Apply preemptive heuristics based on the the total latency and resources
- // inside and outside this zone. Potential stalls should be considered before
- // following this policy.
-
- // Compute remaining latency. We need this both to determine whether the
- // overall schedule has become latency-limited and whether the instructions
- // outside this zone are resource or latency limited.
- //
- // The "dependent" latency is updated incrementally during scheduling as the
- // max height/depth of scheduled nodes minus the cycles since it was
- // scheduled:
- // DLat = max (N.depth - (CurrCycle - N.ReadyCycle) for N in Zone
- //
- // The "independent" latency is the max ready queue depth:
- // ILat = max N.depth for N in Available|Pending
- //
- // RemainingLatency is the greater of independent and dependent latency.
- unsigned RemLatency = CurrZone.getDependentLatency();
- RemLatency = std::max(RemLatency,
- CurrZone.findMaxLatency(CurrZone.Available.elements()));
- RemLatency = std::max(RemLatency,
- CurrZone.findMaxLatency(CurrZone.Pending.elements()));
-
- // Compute the critical resource outside the zone.
- unsigned OtherCritIdx;
- unsigned OtherCount = OtherZone.getOtherResourceCount(OtherCritIdx);
-
- bool OtherResLimited = false;
- if (SchedModel->hasInstrSchedModel()) {
- unsigned LFactor = SchedModel->getLatencyFactor();
- OtherResLimited = (int)(OtherCount - (RemLatency * LFactor)) > (int)LFactor;
- }
- if (!OtherResLimited
- && (RemLatency + CurrZone.getCurrCycle() > Rem.CriticalPath)) {
- Policy.ReduceLatency |= true;
- DEBUG(dbgs() << " " << CurrZone.Available.getName() << " RemainingLatency "
- << RemLatency << " + " << CurrZone.getCurrCycle() << "c > CritPath "
- << Rem.CriticalPath << "\n");
- }
- // If the same resource is limiting inside and outside the zone, do nothing.
- if (CurrZone.getZoneCritResIdx() == OtherCritIdx)
- return;
-
- DEBUG(
- if (CurrZone.isResourceLimited()) {
- dbgs() << " " << CurrZone.Available.getName() << " ResourceLimited: "
- << SchedModel->getResourceName(CurrZone.getZoneCritResIdx())
- << "\n";
- }
- if (OtherResLimited)
- dbgs() << " RemainingLimit: "
- << SchedModel->getResourceName(OtherCritIdx) << "\n";
- if (!CurrZone.isResourceLimited() && !OtherResLimited)
- dbgs() << " Latency limited both directions.\n");
-
- if (CurrZone.isResourceLimited() && !Policy.ReduceResIdx)
- Policy.ReduceResIdx = CurrZone.getZoneCritResIdx();
-
- if (OtherResLimited)
- Policy.DemandResIdx = OtherCritIdx;
-}
-
-void GenericScheduler::SchedCandidate::
-initResourceDelta(const ScheduleDAGMILive *DAG,
- const TargetSchedModel *SchedModel) {
- if (!Policy.ReduceResIdx && !Policy.DemandResIdx)
- return;
-
- const MCSchedClassDesc *SC = DAG->getSchedClass(SU);
- for (TargetSchedModel::ProcResIter
- PI = SchedModel->getWriteProcResBegin(SC),
- PE = SchedModel->getWriteProcResEnd(SC); PI != PE; ++PI) {
- if (PI->ProcResourceIdx == Policy.ReduceResIdx)
- ResDelta.CritResources += PI->Cycles;
- if (PI->ProcResourceIdx == Policy.DemandResIdx)
- ResDelta.DemandedResources += PI->Cycles;
- }
-}
-
-/// Return true if this heuristic determines order.
-static bool tryLess(int TryVal, int CandVal,
- GenericScheduler::SchedCandidate &TryCand,
- GenericScheduler::SchedCandidate &Cand,
- GenericScheduler::CandReason Reason) {
- if (TryVal < CandVal) {
- TryCand.Reason = Reason;
- return true;
- }
- if (TryVal > CandVal) {
- if (Cand.Reason > Reason)
- Cand.Reason = Reason;
- return true;
- }
- Cand.setRepeat(Reason);
- return false;
-}
-
-static bool tryGreater(int TryVal, int CandVal,
- GenericScheduler::SchedCandidate &TryCand,
- GenericScheduler::SchedCandidate &Cand,
- GenericScheduler::CandReason Reason) {
- if (TryVal > CandVal) {
- TryCand.Reason = Reason;
- return true;
- }
- if (TryVal < CandVal) {
- if (Cand.Reason > Reason)
- Cand.Reason = Reason;
- return true;
- }
- Cand.setRepeat(Reason);
- return false;
-}
-
static bool tryPressure(const PressureChange &TryP,
const PressureChange &CandP,
- GenericScheduler::SchedCandidate &TryCand,
- GenericScheduler::SchedCandidate &Cand,
- GenericScheduler::CandReason Reason) {
+ GenericSchedulerBase::SchedCandidate &TryCand,
+ GenericSchedulerBase::SchedCandidate &Cand,
+ GenericSchedulerBase::CandReason Reason) {
int TryRank = TryP.getPSetOrMax();
int CandRank = CandP.getPSetOrMax();
// If both candidates affect the same set, go with the smallest increase.
}
// 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.
return 0;
}
-static bool tryLatency(GenericScheduler::SchedCandidate &TryCand,
- GenericScheduler::SchedCandidate &Cand,
- SchedBoundary &Zone) {
- if (Zone.isTop()) {
- if (Cand.SU->getDepth() > Zone.getScheduledLatency()) {
- if (tryLess(TryCand.SU->getDepth(), Cand.SU->getDepth(),
- TryCand, Cand, GenericScheduler::TopDepthReduce))
- return true;
- }
- if (tryGreater(TryCand.SU->getHeight(), Cand.SU->getHeight(),
- TryCand, Cand, GenericScheduler::TopPathReduce))
- return true;
- }
- else {
- if (Cand.SU->getHeight() > Zone.getScheduledLatency()) {
- if (tryLess(TryCand.SU->getHeight(), Cand.SU->getHeight(),
- TryCand, Cand, GenericScheduler::BotHeightReduce))
- return true;
- }
- if (tryGreater(TryCand.SU->getDepth(), Cand.SU->getDepth(),
- TryCand, Cand, GenericScheduler::BotPathReduce))
- return true;
- }
- return false;
-}
-
/// Apply a set of heursitics to a new candidate. Heuristics are currently
/// hierarchical. This may be more efficient than a graduated cost model because
/// we don't need to evaluate all aspects of the model for each node in the
}
}
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))
// 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;
}
}
}
-#ifndef NDEBUG
-const char *GenericScheduler::getReasonStr(
- GenericScheduler::CandReason Reason) {
- switch (Reason) {
- case NoCand: return "NOCAND ";
- case PhysRegCopy: return "PREG-COPY";
- case RegExcess: return "REG-EXCESS";
- case RegCritical: return "REG-CRIT ";
- case Stall: return "STALL ";
- case Cluster: return "CLUSTER ";
- case Weak: return "WEAK ";
- case RegMax: return "REG-MAX ";
- case ResourceReduce: return "RES-REDUCE";
- case ResourceDemand: return "RES-DEMAND";
- case TopDepthReduce: return "TOP-DEPTH ";
- case TopPathReduce: return "TOP-PATH ";
- case BotHeightReduce:return "BOT-HEIGHT";
- case BotPathReduce: return "BOT-PATH ";
- case NextDefUse: return "DEF-USE ";
- case NodeOrder: return "ORDER ";
- };
- llvm_unreachable("Unknown reason!");
-}
-
-void GenericScheduler::traceCandidate(const SchedCandidate &Cand) {
- PressureChange P;
- unsigned ResIdx = 0;
- unsigned Latency = 0;
- switch (Cand.Reason) {
- default:
- break;
- case RegExcess:
- P = Cand.RPDelta.Excess;
- break;
- case RegCritical:
- P = Cand.RPDelta.CriticalMax;
- break;
- case RegMax:
- P = Cand.RPDelta.CurrentMax;
- break;
- case ResourceReduce:
- ResIdx = Cand.Policy.ReduceResIdx;
- break;
- case ResourceDemand:
- ResIdx = Cand.Policy.DemandResIdx;
- break;
- case TopDepthReduce:
- Latency = Cand.SU->getDepth();
- break;
- case TopPathReduce:
- Latency = Cand.SU->getHeight();
- break;
- case BotHeightReduce:
- Latency = Cand.SU->getHeight();
- break;
- case BotPathReduce:
- Latency = Cand.SU->getDepth();
- break;
- }
- dbgs() << " SU(" << Cand.SU->NodeNum << ") " << getReasonStr(Cand.Reason);
- if (P.isValid())
- dbgs() << " " << TRI->getRegPressureSetName(P.getPSet())
- << ":" << P.getUnitInc() << " ";
- else
- dbgs() << " ";
- if (ResIdx)
- dbgs() << " " << SchedModel->getProcResource(ResIdx)->Name << " ";
- else
- dbgs() << " ";
- if (Latency)
- dbgs() << " " << Latency << " cycles ";
- else
- dbgs() << " ";
- dbgs() << '\n';
-}
-#endif
-
/// Pick the best candidate from the queue.
///
/// TODO: getMaxPressureDelta results can be mostly cached for each SUnit during
}
}
-static void tracePick(const GenericScheduler::SchedCandidate &Cand,
- bool IsTop) {
- DEBUG(dbgs() << "Pick " << (IsTop ? "Top " : "Bot ")
- << GenericScheduler::getReasonStr(Cand.Reason) << '\n');
-}
-
/// Pick the best candidate node from either the top or bottom queue.
SUnit *GenericScheduler::pickNodeBidirectional(bool &IsTopNode) {
// Schedule as far as possible in the direction of no choice. This is most
// 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;
SchedCandidate TopCand(NoPolicy);
// Set the bottom-up policy based on the state of the current bottom zone and
// the instructions outside the zone, including the top zone.
- setPolicy(BotCand.Policy, Bot, Top);
+ setPolicy(BotCand.Policy, /*IsPostRA=*/false, Bot, &Top);
// Set the top-down policy based on the state of the current top zone and
// the instructions outside the zone, including the bottom zone.
- setPolicy(TopCand.Policy, Top, Bot);
+ setPolicy(TopCand.Policy, /*IsPostRA=*/false, Top, &Bot);
// Prefer bottom scheduling when heuristics are silent.
pickNodeFromQueue(Bot, DAG->getBotRPTracker(), BotCand);
if (DAG->top() == DAG->bottom()) {
assert(Top.Available.empty() && Top.Pending.empty() &&
Bot.Available.empty() && Bot.Pending.empty() && "ReadyQ garbage");
- return NULL;
+ return nullptr;
}
SUnit *SU;
do {
}
/// Update the scheduler's state after scheduling a node. This is the same node
-/// that was just returned by pickNode(). However, ScheduleDAGMILive needs to update
-/// it's state based on the current cycle before MachineSchedStrategy does.
+/// that was just returned by pickNode(). However, ScheduleDAGMILive needs to
+/// update it's state based on the current cycle before MachineSchedStrategy
+/// does.
///
/// FIXME: Eventually, we may bundle physreg copies rather than rescheduling
/// them here. See comments in biasPhysRegCopy.
}
}
-/// Create a generic scheduler with no DAG mutation passes.
-static ScheduleDAGInstrs *createRawGenericSched(MachineSchedContext *C) {
- return new ScheduleDAGMILive(C, new GenericScheduler(C));
-}
-
/// Create the standard converging machine scheduler. This will be used as the
/// default scheduler if the target does not set a default.
-static ScheduleDAGInstrs *createGenericSched(MachineSchedContext *C) {
- ScheduleDAGMILive *DAG =
- static_cast<ScheduleDAGMILive*>(createRawGenericSched(C));
+static ScheduleDAGInstrs *createGenericSchedLive(MachineSchedContext *C) {
+ ScheduleDAGMILive *DAG = new ScheduleDAGMILive(C, make_unique<GenericScheduler>(C));
// Register DAG post-processors.
//
// FIXME: extend the mutation API to allow earlier mutations to instantiate
// data and pass it to later mutations. Have a single mutation that gathers
// the interesting nodes in one pass.
- DAG->addMutation(new CopyConstrain(DAG->TII, DAG->TRI));
+ DAG->addMutation(make_unique<CopyConstrain>(DAG->TII, DAG->TRI));
if (EnableLoadCluster && DAG->TII->enableClusterLoads())
- DAG->addMutation(new LoadClusterMutation(DAG->TII, DAG->TRI));
+ DAG->addMutation(make_unique<LoadClusterMutation>(DAG->TII, DAG->TRI));
if (EnableMacroFusion)
- DAG->addMutation(new MacroFusion(DAG->TII));
+ DAG->addMutation(make_unique<MacroFusion>(*DAG->TII, *DAG->TRI));
return DAG;
}
+
static MachineSchedRegistry
GenericSchedRegistry("converge", "Standard converging scheduler.",
- createGenericSched);
+ createGenericSchedLive);
+
+//===----------------------------------------------------------------------===//
+// PostGenericScheduler - Generic PostRA implementation of MachineSchedStrategy.
+//===----------------------------------------------------------------------===//
+
+void PostGenericScheduler::initialize(ScheduleDAGMI *Dag) {
+ DAG = Dag;
+ SchedModel = DAG->getSchedModel();
+ TRI = DAG->TRI;
+
+ Rem.init(DAG, SchedModel);
+ Top.init(DAG, SchedModel, &Rem);
+ BotRoots.clear();
+
+ // Initialize the HazardRecognizers. If itineraries don't exist, are empty,
+ // or are disabled, then these HazardRecs will be disabled.
+ const InstrItineraryData *Itin = SchedModel->getInstrItineraries();
+ if (!Top.HazardRec) {
+ Top.HazardRec =
+ DAG->MF.getSubtarget().getInstrInfo()->CreateTargetMIHazardRecognizer(
+ Itin, DAG);
+ }
+}
+
+
+void PostGenericScheduler::registerRoots() {
+ Rem.CriticalPath = DAG->ExitSU.getDepth();
+
+ // Some roots may not feed into ExitSU. Check all of them in case.
+ for (SmallVectorImpl<SUnit*>::const_iterator
+ I = BotRoots.begin(), E = BotRoots.end(); I != E; ++I) {
+ if ((*I)->getDepth() > Rem.CriticalPath)
+ Rem.CriticalPath = (*I)->getDepth();
+ }
+ 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.
+///
+/// \param Cand provides the policy and current best candidate.
+/// \param TryCand refers to the next SUnit candidate, otherwise uninitialized.
+void PostGenericScheduler::tryCandidate(SchedCandidate &Cand,
+ SchedCandidate &TryCand) {
+
+ // Initialize the candidate if needed.
+ if (!Cand.isValid()) {
+ TryCand.Reason = NodeOrder;
+ return;
+ }
+
+ // Prioritize instructions that read unbuffered resources by stall cycles.
+ if (tryLess(Top.getLatencyStallCycles(TryCand.SU),
+ Top.getLatencyStallCycles(Cand.SU), TryCand, Cand, Stall))
+ return;
+
+ // Avoid critical resource consumption and balance the schedule.
+ if (tryLess(TryCand.ResDelta.CritResources, Cand.ResDelta.CritResources,
+ TryCand, Cand, ResourceReduce))
+ return;
+ if (tryGreater(TryCand.ResDelta.DemandedResources,
+ Cand.ResDelta.DemandedResources,
+ TryCand, Cand, ResourceDemand))
+ return;
+
+ // Avoid serializing long latency dependence chains.
+ if (Cand.Policy.ReduceLatency && tryLatency(TryCand, Cand, Top)) {
+ return;
+ }
+
+ // Fall through to original instruction order.
+ if (TryCand.SU->NodeNum < Cand.SU->NodeNum)
+ TryCand.Reason = NodeOrder;
+}
+
+void PostGenericScheduler::pickNodeFromQueue(SchedCandidate &Cand) {
+ ReadyQueue &Q = Top.Available;
+
+ DEBUG(Q.dump());
+
+ for (ReadyQueue::iterator I = Q.begin(), E = Q.end(); I != E; ++I) {
+ SchedCandidate TryCand(Cand.Policy);
+ TryCand.SU = *I;
+ TryCand.initResourceDelta(DAG, SchedModel);
+ tryCandidate(Cand, TryCand);
+ if (TryCand.Reason != NoCand) {
+ Cand.setBest(TryCand);
+ DEBUG(traceCandidate(Cand));
+ }
+ }
+}
+
+/// Pick the next node to schedule.
+SUnit *PostGenericScheduler::pickNode(bool &IsTopNode) {
+ if (DAG->top() == DAG->bottom()) {
+ assert(Top.Available.empty() && Top.Pending.empty() && "ReadyQ garbage");
+ return nullptr;
+ }
+ SUnit *SU;
+ do {
+ SU = Top.pickOnlyChoice();
+ if (!SU) {
+ CandPolicy NoPolicy;
+ SchedCandidate TopCand(NoPolicy);
+ // Set the top-down policy based on the state of the current top zone and
+ // the instructions outside the zone, including the bottom zone.
+ setPolicy(TopCand.Policy, /*IsPostRA=*/true, Top, nullptr);
+ pickNodeFromQueue(TopCand);
+ assert(TopCand.Reason != NoCand && "failed to find a candidate");
+ tracePick(TopCand, true);
+ SU = TopCand.SU;
+ }
+ } while (SU->isScheduled);
+
+ IsTopNode = true;
+ Top.removeReady(SU);
+
+ DEBUG(dbgs() << "Scheduling SU(" << SU->NodeNum << ") " << *SU->getInstr());
+ return SU;
+}
+
+/// Called after ScheduleDAGMI has scheduled an instruction and updated
+/// scheduled/remaining flags in the DAG nodes.
+void PostGenericScheduler::schedNode(SUnit *SU, bool IsTopNode) {
+ SU->TopReadyCycle = std::max(SU->TopReadyCycle, Top.getCurrCycle());
+ Top.bumpNode(SU);
+}
+
+/// Create a generic scheduler with no vreg liveness or DAG mutation passes.
+static ScheduleDAGInstrs *createGenericSchedPostRA(MachineSchedContext *C) {
+ return new ScheduleDAGMI(C, make_unique<PostGenericScheduler>(C), /*IsPostRA=*/true);
+}
//===----------------------------------------------------------------------===//
// ILP Scheduler. Currently for experimental analysis of heuristics.
const BitVector *ScheduledTrees;
bool MaximizeILP;
- ILPOrder(bool MaxILP): DFSResult(0), ScheduledTrees(0), MaximizeILP(MaxILP) {}
+ ILPOrder(bool MaxILP)
+ : DFSResult(nullptr), ScheduledTrees(nullptr), MaximizeILP(MaxILP) {}
/// \brief Apply a less-than relation on node priority.
///
std::vector<SUnit*> ReadyQ;
public:
- ILPScheduler(bool MaximizeILP): DAG(0), Cmp(MaximizeILP) {}
+ ILPScheduler(bool MaximizeILP): DAG(nullptr), Cmp(MaximizeILP) {}
- virtual void initialize(ScheduleDAGMI *dag) {
+ void initialize(ScheduleDAGMI *dag) override {
assert(dag->hasVRegLiveness() && "ILPScheduler needs vreg liveness");
DAG = static_cast<ScheduleDAGMILive*>(dag);
DAG->computeDFSResult();
ReadyQ.clear();
}
- virtual void registerRoots() {
+ void registerRoots() override {
// Restore the heap in ReadyQ with the updated DFS results.
std::make_heap(ReadyQ.begin(), ReadyQ.end(), Cmp);
}
/// -----------------------------------------
/// Callback to select the highest priority node from the ready Q.
- virtual SUnit *pickNode(bool &IsTopNode) {
- if (ReadyQ.empty()) return NULL;
+ SUnit *pickNode(bool &IsTopNode) override {
+ if (ReadyQ.empty()) return nullptr;
std::pop_heap(ReadyQ.begin(), ReadyQ.end(), Cmp);
SUnit *SU = ReadyQ.back();
ReadyQ.pop_back();
}
/// \brief Scheduler callback to notify that a new subtree is scheduled.
- virtual void scheduleTree(unsigned SubtreeID) {
+ void scheduleTree(unsigned SubtreeID) override {
std::make_heap(ReadyQ.begin(), ReadyQ.end(), Cmp);
}
/// Callback after a node is scheduled. Mark a newly scheduled tree, notify
/// DFSResults, and resort the priority Q.
- virtual void schedNode(SUnit *SU, bool IsTopNode) {
+ void schedNode(SUnit *SU, bool IsTopNode) override {
assert(!IsTopNode && "SchedDFSResult needs bottom-up");
}
- virtual void releaseTopNode(SUnit *) { /*only called for top roots*/ }
+ void releaseTopNode(SUnit *) override { /*only called for top roots*/ }
- virtual void releaseBottomNode(SUnit *SU) {
+ void releaseBottomNode(SUnit *SU) override {
ReadyQ.push_back(SU);
std::push_heap(ReadyQ.begin(), ReadyQ.end(), Cmp);
}
} // namespace
static ScheduleDAGInstrs *createILPMaxScheduler(MachineSchedContext *C) {
- return new ScheduleDAGMILive(C, new ILPScheduler(true));
+ return new ScheduleDAGMILive(C, make_unique<ILPScheduler>(true));
}
static ScheduleDAGInstrs *createILPMinScheduler(MachineSchedContext *C) {
- return new ScheduleDAGMILive(C, new ILPScheduler(false));
+ return new ScheduleDAGMILive(C, make_unique<ILPScheduler>(false));
}
static MachineSchedRegistry ILPMaxRegistry(
"ilpmax", "Schedule bottom-up for max ILP", createILPMaxScheduler);
InstructionShuffler(bool alternate, bool topdown)
: IsAlternating(alternate), IsTopDown(topdown) {}
- virtual void initialize(ScheduleDAGMI*) {
+ void initialize(ScheduleDAGMI*) override {
TopQ.clear();
BottomQ.clear();
}
/// Implement MachineSchedStrategy interface.
/// -----------------------------------------
- virtual SUnit *pickNode(bool &IsTopNode) {
+ SUnit *pickNode(bool &IsTopNode) override {
SUnit *SU;
if (IsTopDown) {
do {
- if (TopQ.empty()) return NULL;
+ if (TopQ.empty()) return nullptr;
SU = TopQ.top();
TopQ.pop();
} while (SU->isScheduled);
}
else {
do {
- if (BottomQ.empty()) return NULL;
+ if (BottomQ.empty()) return nullptr;
SU = BottomQ.top();
BottomQ.pop();
} while (SU->isScheduled);
return SU;
}
- virtual void schedNode(SUnit *SU, bool IsTopNode) {}
+ void schedNode(SUnit *SU, bool IsTopNode) override {}
- virtual void releaseTopNode(SUnit *SU) {
+ void releaseTopNode(SUnit *SU) override {
TopQ.push(SU);
}
- virtual void releaseBottomNode(SUnit *SU) {
+ void releaseBottomNode(SUnit *SU) override {
BottomQ.push(SU);
}
};
bool TopDown = !ForceBottomUp;
assert((TopDown || !ForceTopDown) &&
"-misched-topdown incompatible with -misched-bottomup");
- return new ScheduleDAGMILive(C, new InstructionShuffler(Alternate, TopDown));
+ return new ScheduleDAGMILive(C, make_unique<InstructionShuffler>(Alternate, TopDown));
}
static MachineSchedRegistry ShufflerRegistry(
"shuffle", "Shuffle machine instructions alternating directions",
}
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
raw_string_ostream SS(Str);
const ScheduleDAGMI *DAG = static_cast<const ScheduleDAGMI*>(G);
const SchedDFSResult *DFS = DAG->hasVRegLiveness() ?
- static_cast<const ScheduleDAGMILive*>(G)->getDFSResult() : 0;
+ static_cast<const ScheduleDAGMILive*>(G)->getDFSResult() : nullptr;
SS << "SU:" << SU->NodeNum;
if (DFS)
SS << " I:" << DFS->getNumInstrs(SU);
std::string Str("shape=Mrecord");
const ScheduleDAGMI *DAG = static_cast<const ScheduleDAGMI*>(G);
const SchedDFSResult *DFS = DAG->hasVRegLiveness() ?
- static_cast<const ScheduleDAGMILive*>(G)->getDFSResult() : 0;
+ static_cast<const ScheduleDAGMILive*>(G)->getDFSResult() : nullptr;
if (DFS) {
Str += ",style=filled,fillcolor=\"#";
Str += DOT::getColorString(DFS->getSubtreeID(N));
projects / oota-llvm.git / blobdiff