llvm/lib/CodeGen/PostRASchedulerList.cpp

   1 //===----- SchedulePostRAList.cpp - list scheduler ------------------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This implements a top-down list scheduler, using standard algorithms.
  10 // The basic approach uses a priority queue of available nodes to schedule.
  11 // One at a time, nodes are taken from the priority queue (thus in priority
  12 // order), checked for legality to schedule, and emitted if legal.
  13 //
  14 // Nodes may not be legal to schedule either due to structural hazards (e.g.
  15 // pipeline or resource constraints) or because an input to the instruction has
  16 // not completed execution.
  17 //
  18 //===----------------------------------------------------------------------===//
  19
  20 #include "AggressiveAntiDepBreaker.h"
  21 #include "AntiDepBreaker.h"
  22 #include "CriticalAntiDepBreaker.h"
  23 #include "llvm/ADT/Statistic.h"
  24 #include "llvm/Analysis/AliasAnalysis.h"
  25 #include "llvm/CodeGen/LatencyPriorityQueue.h"
  26 #include "llvm/CodeGen/MachineDominators.h"
  27 #include "llvm/CodeGen/MachineFunctionPass.h"
  28 #include "llvm/CodeGen/MachineLoopInfo.h"
  29 #include "llvm/CodeGen/MachineRegisterInfo.h"
  30 #include "llvm/CodeGen/Passes.h"
  31 #include "llvm/CodeGen/RegisterClassInfo.h"
  32 #include "llvm/CodeGen/ScheduleDAGInstrs.h"
  33 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
  34 #include "llvm/CodeGen/SchedulerRegistry.h"
  35 #include "llvm/CodeGen/TargetInstrInfo.h"
  36 #include "llvm/CodeGen/TargetLowering.h"
  37 #include "llvm/CodeGen/TargetPassConfig.h"
  38 #include "llvm/CodeGen/TargetRegisterInfo.h"
  39 #include "llvm/CodeGen/TargetSubtargetInfo.h"
  40 #include "llvm/Config/llvm-config.h"
  41 #include "llvm/InitializePasses.h"
  42 #include "llvm/Support/CommandLine.h"
  43 #include "llvm/Support/Debug.h"
  44 #include "llvm/Support/ErrorHandling.h"
  45 #include "llvm/Support/raw_ostream.h"
  46 using namespace llvm;
  47
  48 #define DEBUG_TYPE "post-RA-sched"
  49
  50 STATISTIC(NumNoops, "Number of noops inserted");
  51 STATISTIC(NumStalls, "Number of pipeline stalls");
  52 STATISTIC(NumFixedAnti, "Number of fixed anti-dependencies");
  53
  54 // Post-RA scheduling is enabled with
  55 // TargetSubtargetInfo.enablePostRAScheduler(). This flag can be used to
  56 // override the target.
  57 static cl::opt<bool>
  58 EnablePostRAScheduler("post-RA-scheduler",
  59                        cl::desc("Enable scheduling after register allocation"),
  60                        cl::init(false), cl::Hidden);
  61 static cl::opt<std::string>
  62 EnableAntiDepBreaking("break-anti-dependencies",
  63                       cl::desc("Break post-RA scheduling anti-dependencies: "
  64                                "\"critical\", \"all\", or \"none\""),
  65                       cl::init("none"), cl::Hidden);
  66
  67 // If DebugDiv > 0 then only schedule MBB with (ID % DebugDiv) == DebugMod
  68 static cl::opt<int>
  69 DebugDiv("postra-sched-debugdiv",
  70                       cl::desc("Debug control MBBs that are scheduled"),
  71                       cl::init(0), cl::Hidden);
  72 static cl::opt<int>
  73 DebugMod("postra-sched-debugmod",
  74                       cl::desc("Debug control MBBs that are scheduled"),
  75                       cl::init(0), cl::Hidden);
  76
  77 AntiDepBreaker::~AntiDepBreaker() { }
  78
  79 namespace {
  80   class PostRAScheduler : public MachineFunctionPass {
  81     const TargetInstrInfo *TII = nullptr;
  82     RegisterClassInfo RegClassInfo;
  83
  84   public:
  85     static char ID;
  86     PostRAScheduler() : MachineFunctionPass(ID) {}
  87
  88     void getAnalysisUsage(AnalysisUsage &AU) const override {
  89       AU.setPreservesCFG();
  90       AU.addRequired<AAResultsWrapperPass>();
  91       AU.addRequired<TargetPassConfig>();
  92       AU.addRequired<MachineDominatorTree>();
  93       AU.addPreserved<MachineDominatorTree>();
  94       AU.addRequired<MachineLoopInfo>();
  95       AU.addPreserved<MachineLoopInfo>();
  96       MachineFunctionPass::getAnalysisUsage(AU);
  97     }
  98
  99     MachineFunctionProperties getRequiredProperties() const override {
 100       return MachineFunctionProperties().set(
 101           MachineFunctionProperties::Property::NoVRegs);
 102     }
 103
 104     bool runOnMachineFunction(MachineFunction &Fn) override;
 105
 106   private:
 107     bool enablePostRAScheduler(
 108         const TargetSubtargetInfo &ST, CodeGenOpt::Level OptLevel,
 109         TargetSubtargetInfo::AntiDepBreakMode &Mode,
 110         TargetSubtargetInfo::RegClassVector &CriticalPathRCs) const;
 111   };
 112   char PostRAScheduler::ID = 0;
 113
 114   class SchedulePostRATDList : public ScheduleDAGInstrs {
 115     /// AvailableQueue - The priority queue to use for the available SUnits.
 116     ///
 117     LatencyPriorityQueue AvailableQueue;
 118
 119     /// PendingQueue - This contains all of the instructions whose operands have
 120     /// been issued, but their results are not ready yet (due to the latency of
 121     /// the operation).  Once the operands becomes available, the instruction is
 122     /// added to the AvailableQueue.
 123     std::vector<SUnit*> PendingQueue;
 124
 125     /// HazardRec - The hazard recognizer to use.
 126     ScheduleHazardRecognizer *HazardRec;
 127
 128     /// AntiDepBreak - Anti-dependence breaking object, or NULL if none
 129     AntiDepBreaker *AntiDepBreak;
 130
 131     /// AA - AliasAnalysis for making memory reference queries.
 132     AliasAnalysis *AA;
 133
 134     /// The schedule. Null SUnit*'s represent noop instructions.
 135     std::vector<SUnit*> Sequence;
 136
 137     /// Ordered list of DAG postprocessing steps.
 138     std::vector<std::unique_ptr<ScheduleDAGMutation>> Mutations;
 139
 140     /// The index in BB of RegionEnd.
 141     ///
 142     /// This is the instruction number from the top of the current block, not
 143     /// the SlotIndex. It is only used by the AntiDepBreaker.
 144     unsigned EndIndex;
 145
 146   public:
 147     SchedulePostRATDList(
 148         MachineFunction &MF, MachineLoopInfo &MLI, AliasAnalysis *AA,
 149         const RegisterClassInfo &,
 150         TargetSubtargetInfo::AntiDepBreakMode AntiDepMode,
 151         SmallVectorImpl<const TargetRegisterClass *> &CriticalPathRCs);
 152
 153     ~SchedulePostRATDList() override;
 154
 155     /// startBlock - Initialize register live-range state for scheduling in
 156     /// this block.
 157     ///
 158     void startBlock(MachineBasicBlock *BB) override;
 159
 160     // Set the index of RegionEnd within the current BB.
 161     void setEndIndex(unsigned EndIdx) { EndIndex = EndIdx; }
 162
 163     /// Initialize the scheduler state for the next scheduling region.
 164     void enterRegion(MachineBasicBlock *bb,
 165                      MachineBasicBlock::iterator begin,
 166                      MachineBasicBlock::iterator end,
 167                      unsigned regioninstrs) override;
 168
 169     /// Notify that the scheduler has finished scheduling the current region.
 170     void exitRegion() override;
 171
 172     /// Schedule - Schedule the instruction range using list scheduling.
 173     ///
 174     void schedule() override;
 175
 176     void EmitSchedule();
 177
 178     /// Observe - Update liveness information to account for the current
 179     /// instruction, which will not be scheduled.
 180     ///
 181     void Observe(MachineInstr &MI, unsigned Count);
 182
 183     /// finishBlock - Clean up register live-range state.
 184     ///
 185     void finishBlock() override;
 186
 187   private:
 188     /// Apply each ScheduleDAGMutation step in order.
 189     void postprocessDAG();
 190
 191     void ReleaseSucc(SUnit *SU, SDep *SuccEdge);
 192     void ReleaseSuccessors(SUnit *SU);
 193     void ScheduleNodeTopDown(SUnit *SU, unsigned CurCycle);
 194     void ListScheduleTopDown();
 195
 196     void dumpSchedule() const;
 197     void emitNoop(unsigned CurCycle);
 198   };
 199 }
 200
 201 char &llvm::PostRASchedulerID = PostRAScheduler::ID;
 202
 203 INITIALIZE_PASS(PostRAScheduler, DEBUG_TYPE,
 204                 "Post RA top-down list latency scheduler", false, false)
 205
 206 SchedulePostRATDList::SchedulePostRATDList(
 207     MachineFunction &MF, MachineLoopInfo &MLI, AliasAnalysis *AA,
 208     const RegisterClassInfo &RCI,
 209     TargetSubtargetInfo::AntiDepBreakMode AntiDepMode,
 210     SmallVectorImpl<const TargetRegisterClass *> &CriticalPathRCs)
 211     : ScheduleDAGInstrs(MF, &MLI), AA(AA), EndIndex(0) {
 212
 213   const InstrItineraryData *InstrItins =
 214       MF.getSubtarget().getInstrItineraryData();
 215   HazardRec =
 216       MF.getSubtarget().getInstrInfo()->CreateTargetPostRAHazardRecognizer(
 217           InstrItins, this);
 218   MF.getSubtarget().getPostRAMutations(Mutations);
 219
 220   assert((AntiDepMode == TargetSubtargetInfo::ANTIDEP_NONE ||
 221           MRI.tracksLiveness()) &&
 222          "Live-ins must be accurate for anti-dependency breaking");
 223   AntiDepBreak =
 224     ((AntiDepMode == TargetSubtargetInfo::ANTIDEP_ALL) ?
 225      (AntiDepBreaker *)new AggressiveAntiDepBreaker(MF, RCI, CriticalPathRCs) :
 226      ((AntiDepMode == TargetSubtargetInfo::ANTIDEP_CRITICAL) ?
 227       (AntiDepBreaker *)new CriticalAntiDepBreaker(MF, RCI) : nullptr));
 228 }
 229
 230 SchedulePostRATDList::~SchedulePostRATDList() {
 231   delete HazardRec;
 232   delete AntiDepBreak;
 233 }
 234
 235 /// Initialize state associated with the next scheduling region.
 236 void SchedulePostRATDList::enterRegion(MachineBasicBlock *bb,
 237                  MachineBasicBlock::iterator begin,
 238                  MachineBasicBlock::iterator end,
 239                  unsigned regioninstrs) {
 240   ScheduleDAGInstrs::enterRegion(bb, begin, end, regioninstrs);
 241   Sequence.clear();
 242 }
 243
 244 /// Print the schedule before exiting the region.
 245 void SchedulePostRATDList::exitRegion() {
 246   LLVM_DEBUG({
 247     dbgs() << "*** Final schedule ***\n";
 248     dumpSchedule();
 249     dbgs() << '\n';
 250   });
 251   ScheduleDAGInstrs::exitRegion();
 252 }
 253
 254 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 255 /// dumpSchedule - dump the scheduled Sequence.
 256 LLVM_DUMP_METHOD void SchedulePostRATDList::dumpSchedule() const {
 257   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
 258     if (SUnit *SU = Sequence[i])
 259       dumpNode(*SU);
 260     else
 261       dbgs() << "**** NOOP ****\n";
 262   }
 263 }
 264 #endif
 265
 266 bool PostRAScheduler::enablePostRAScheduler(
 267     const TargetSubtargetInfo &ST,
 268     CodeGenOpt::Level OptLevel,
 269     TargetSubtargetInfo::AntiDepBreakMode &Mode,
 270     TargetSubtargetInfo::RegClassVector &CriticalPathRCs) const {
 271   Mode = ST.getAntiDepBreakMode();
 272   ST.getCriticalPathRCs(CriticalPathRCs);
 273
 274   // Check for explicit enable/disable of post-ra scheduling.
 275   if (EnablePostRAScheduler.getPosition() > 0)
 276     return EnablePostRAScheduler;
 277
 278   return ST.enablePostRAScheduler() &&
 279          OptLevel >= ST.getOptLevelToEnablePostRAScheduler();
 280 }
 281
 282 bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
 283   if (skipFunction(Fn.getFunction()))
 284     return false;
 285
 286   TII = Fn.getSubtarget().getInstrInfo();
 287   MachineLoopInfo &MLI = getAnalysis<MachineLoopInfo>();
 288   AliasAnalysis *AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
 289   TargetPassConfig *PassConfig = &getAnalysis<TargetPassConfig>();
 290
 291   RegClassInfo.runOnMachineFunction(Fn);
 292
 293   TargetSubtargetInfo::AntiDepBreakMode AntiDepMode =
 294     TargetSubtargetInfo::ANTIDEP_NONE;
 295   SmallVector<const TargetRegisterClass*, 4> CriticalPathRCs;
 296
 297   // Check that post-RA scheduling is enabled for this target.
 298   // This may upgrade the AntiDepMode.
 299   if (!enablePostRAScheduler(Fn.getSubtarget(), PassConfig->getOptLevel(),
 300                              AntiDepMode, CriticalPathRCs))
 301     return false;
 302
 303   // Check for antidep breaking override...
 304   if (EnableAntiDepBreaking.getPosition() > 0) {
 305     AntiDepMode = (EnableAntiDepBreaking == "all")
 306       ? TargetSubtargetInfo::ANTIDEP_ALL
 307       : ((EnableAntiDepBreaking == "critical")
 308          ? TargetSubtargetInfo::ANTIDEP_CRITICAL
 309          : TargetSubtargetInfo::ANTIDEP_NONE);
 310   }
 311
 312   LLVM_DEBUG(dbgs() << "PostRAScheduler\n");
 313
 314   SchedulePostRATDList Scheduler(Fn, MLI, AA, RegClassInfo, AntiDepMode,
 315                                  CriticalPathRCs);
 316
 317   // Loop over all of the basic blocks
 318   for (auto &MBB : Fn) {
 319 #ifndef NDEBUG
 320     // If DebugDiv > 0 then only schedule MBB with (ID % DebugDiv) == DebugMod
 321     if (DebugDiv > 0) {
 322       static int bbcnt = 0;
 323       if (bbcnt++ % DebugDiv != DebugMod)
 324         continue;
 325       dbgs() << "*** DEBUG scheduling " << Fn.getName() << ":"
 326              << printMBBReference(MBB) << " ***\n";
 327     }
 328 #endif
 329
 330     // Initialize register live-range state for scheduling in this block.
 331     Scheduler.startBlock(&MBB);
 332
 333     // Schedule each sequence of instructions not interrupted by a label
 334     // or anything else that effectively needs to shut down scheduling.
 335     MachineBasicBlock::iterator Current = MBB.end();
 336     unsigned Count = MBB.size(), CurrentCount = Count;
 337     for (MachineBasicBlock::iterator I = Current; I != MBB.begin();) {
 338       MachineInstr &MI = *std::prev(I);
 339       --Count;
 340       // Calls are not scheduling boundaries before register allocation, but
 341       // post-ra we don't gain anything by scheduling across calls since we
 342       // don't need to worry about register pressure.
 343       if (MI.isCall() || TII->isSchedulingBoundary(MI, &MBB, Fn)) {
 344         Scheduler.enterRegion(&MBB, I, Current, CurrentCount - Count);
 345         Scheduler.setEndIndex(CurrentCount);
 346         Scheduler.schedule();
 347         Scheduler.exitRegion();
 348         Scheduler.EmitSchedule();
 349         Current = &MI;
 350         CurrentCount = Count;
 351         Scheduler.Observe(MI, CurrentCount);
 352       }
 353       I = MI;
 354       if (MI.isBundle())
 355         Count -= MI.getBundleSize();
 356     }
 357     assert(Count == 0 && "Instruction count mismatch!");
 358     assert((MBB.begin() == Current || CurrentCount != 0) &&
 359            "Instruction count mismatch!");
 360     Scheduler.enterRegion(&MBB, MBB.begin(), Current, CurrentCount);
 361     Scheduler.setEndIndex(CurrentCount);
 362     Scheduler.schedule();
 363     Scheduler.exitRegion();
 364     Scheduler.EmitSchedule();
 365
 366     // Clean up register live-range state.
 367     Scheduler.finishBlock();
 368
 369     // Update register kills
 370     Scheduler.fixupKills(MBB);
 371   }
 372
 373   return true;
 374 }
 375
 376 /// StartBlock - Initialize register live-range state for scheduling in
 377 /// this block.
 378 ///
 379 void SchedulePostRATDList::startBlock(MachineBasicBlock *BB) {
 380   // Call the superclass.
 381   ScheduleDAGInstrs::startBlock(BB);
 382
 383   // Reset the hazard recognizer and anti-dep breaker.
 384   HazardRec->Reset();
 385   if (AntiDepBreak)
 386     AntiDepBreak->StartBlock(BB);
 387 }
 388
 389 /// Schedule - Schedule the instruction range using list scheduling.
 390 ///
 391 void SchedulePostRATDList::schedule() {
 392   // Build the scheduling graph.
 393   buildSchedGraph(AA);
 394
 395   if (AntiDepBreak) {
 396     unsigned Broken =
 397       AntiDepBreak->BreakAntiDependencies(SUnits, RegionBegin, RegionEnd,
 398                                           EndIndex, DbgValues);
 399
 400     if (Broken != 0) {
 401       // We made changes. Update the dependency graph.
 402       // Theoretically we could update the graph in place:
 403       // When a live range is changed to use a different register, remove
 404       // the def's anti-dependence *and* output-dependence edges due to
 405       // that register, and add new anti-dependence and output-dependence
 406       // edges based on the next live range of the register.
 407       ScheduleDAG::clearDAG();
 408       buildSchedGraph(AA);
 409
 410       NumFixedAnti += Broken;
 411     }
 412   }
 413
 414   postprocessDAG();
 415
 416   LLVM_DEBUG(dbgs() << "********** List Scheduling **********\n");
 417   LLVM_DEBUG(dump());
 418
 419   AvailableQueue.initNodes(SUnits);
 420   ListScheduleTopDown();
 421   AvailableQueue.releaseState();
 422 }
 423
 424 /// Observe - Update liveness information to account for the current
 425 /// instruction, which will not be scheduled.
 426 ///
 427 void SchedulePostRATDList::Observe(MachineInstr &MI, unsigned Count) {
 428   if (AntiDepBreak)
 429     AntiDepBreak->Observe(MI, Count, EndIndex);
 430 }
 431
 432 /// FinishBlock - Clean up register live-range state.
 433 ///
 434 void SchedulePostRATDList::finishBlock() {
 435   if (AntiDepBreak)
 436     AntiDepBreak->FinishBlock();
 437
 438   // Call the superclass.
 439   ScheduleDAGInstrs::finishBlock();
 440 }
 441
 442 /// Apply each ScheduleDAGMutation step in order.
 443 void SchedulePostRATDList::postprocessDAG() {
 444   for (auto &M : Mutations)
 445     M->apply(this);
 446 }
 447
 448 //===----------------------------------------------------------------------===//
 449 //  Top-Down Scheduling
 450 //===----------------------------------------------------------------------===//
 451
 452 /// ReleaseSucc - Decrement the NumPredsLeft count of a successor. Add it to
 453 /// the PendingQueue if the count reaches zero.
 454 void SchedulePostRATDList::ReleaseSucc(SUnit *SU, SDep *SuccEdge) {
 455   SUnit *SuccSU = SuccEdge->getSUnit();
 456
 457   if (SuccEdge->isWeak()) {
 458     --SuccSU->WeakPredsLeft;
 459     return;
 460   }
 461 #ifndef NDEBUG
 462   if (SuccSU->NumPredsLeft == 0) {
 463     dbgs() << "*** Scheduling failed! ***\n";
 464     dumpNode(*SuccSU);
 465     dbgs() << " has been released too many times!\n";
 466     llvm_unreachable(nullptr);
 467   }
 468 #endif
 469   --SuccSU->NumPredsLeft;
 470
 471   // Standard scheduler algorithms will recompute the depth of the successor
 472   // here as such:
 473   //   SuccSU->setDepthToAtLeast(SU->getDepth() + SuccEdge->getLatency());
 474   //
 475   // However, we lazily compute node depth instead. Note that
 476   // ScheduleNodeTopDown has already updated the depth of this node which causes
 477   // all descendents to be marked dirty. Setting the successor depth explicitly
 478   // here would cause depth to be recomputed for all its ancestors. If the
 479   // successor is not yet ready (because of a transitively redundant edge) then
 480   // this causes depth computation to be quadratic in the size of the DAG.
 481
 482   // If all the node's predecessors are scheduled, this node is ready
 483   // to be scheduled. Ignore the special ExitSU node.
 484   if (SuccSU->NumPredsLeft == 0 && SuccSU != &ExitSU)
 485     PendingQueue.push_back(SuccSU);
 486 }
 487
 488 /// ReleaseSuccessors - Call ReleaseSucc on each of SU's successors.
 489 void SchedulePostRATDList::ReleaseSuccessors(SUnit *SU) {
 490   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 491        I != E; ++I) {
 492     ReleaseSucc(SU, &*I);
 493   }
 494 }
 495
 496 /// ScheduleNodeTopDown - Add the node to the schedule. Decrement the pending
 497 /// count of its successors. If a successor pending count is zero, add it to
 498 /// the Available queue.
 499 void SchedulePostRATDList::ScheduleNodeTopDown(SUnit *SU, unsigned CurCycle) {
 500   LLVM_DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
 501   LLVM_DEBUG(dumpNode(*SU));
 502
 503   Sequence.push_back(SU);
 504   assert(CurCycle >= SU->getDepth() &&
 505          "Node scheduled above its depth!");
 506   SU->setDepthToAtLeast(CurCycle);
 507
 508   ReleaseSuccessors(SU);
 509   SU->isScheduled = true;
 510   AvailableQueue.scheduledNode(SU);
 511 }
 512
 513 /// emitNoop - Add a noop to the current instruction sequence.
 514 void SchedulePostRATDList::emitNoop(unsigned CurCycle) {
 515   LLVM_DEBUG(dbgs() << "*** Emitting noop in cycle " << CurCycle << '\n');
 516   HazardRec->EmitNoop();
 517   Sequence.push_back(nullptr);   // NULL here means noop
 518   ++NumNoops;
 519 }
 520
 521 /// ListScheduleTopDown - The main loop of list scheduling for top-down
 522 /// schedulers.
 523 void SchedulePostRATDList::ListScheduleTopDown() {
 524   unsigned CurCycle = 0;
 525
 526   // We're scheduling top-down but we're visiting the regions in
 527   // bottom-up order, so we don't know the hazards at the start of a
 528   // region. So assume no hazards (this should usually be ok as most
 529   // blocks are a single region).
 530   HazardRec->Reset();
 531
 532   // Release any successors of the special Entry node.
 533   ReleaseSuccessors(&EntrySU);
 534
 535   // Add all leaves to Available queue.
 536   for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
 537     // It is available if it has no predecessors.
 538     if (!SUnits[i].NumPredsLeft && !SUnits[i].isAvailable) {
 539       AvailableQueue.push(&SUnits[i]);
 540       SUnits[i].isAvailable = true;
 541     }
 542   }
 543
 544   // In any cycle where we can't schedule any instructions, we must
 545   // stall or emit a noop, depending on the target.
 546   bool CycleHasInsts = false;
 547
 548   // While Available queue is not empty, grab the node with the highest
 549   // priority. If it is not ready put it back.  Schedule the node.
 550   std::vector<SUnit*> NotReady;
 551   Sequence.reserve(SUnits.size());
 552   while (!AvailableQueue.empty() || !PendingQueue.empty()) {
 553     // Check to see if any of the pending instructions are ready to issue.  If
 554     // so, add them to the available queue.
 555     unsigned MinDepth = ~0u;
 556     for (unsigned i = 0, e = PendingQueue.size(); i != e; ++i) {
 557       if (PendingQueue[i]->getDepth() <= CurCycle) {
 558         AvailableQueue.push(PendingQueue[i]);
 559         PendingQueue[i]->isAvailable = true;
 560         PendingQueue[i] = PendingQueue.back();
 561         PendingQueue.pop_back();
 562         --i; --e;
 563       } else if (PendingQueue[i]->getDepth() < MinDepth)
 564         MinDepth = PendingQueue[i]->getDepth();
 565     }
 566
 567     LLVM_DEBUG(dbgs() << "\n*** Examining Available\n";
 568                AvailableQueue.dump(this));
 569
 570     SUnit *FoundSUnit = nullptr, *NotPreferredSUnit = nullptr;
 571     bool HasNoopHazards = false;
 572     while (!AvailableQueue.empty()) {
 573       SUnit *CurSUnit = AvailableQueue.pop();
 574
 575       ScheduleHazardRecognizer::HazardType HT =
 576         HazardRec->getHazardType(CurSUnit, 0/*no stalls*/);
 577       if (HT == ScheduleHazardRecognizer::NoHazard) {
 578         if (HazardRec->ShouldPreferAnother(CurSUnit)) {
 579           if (!NotPreferredSUnit) {
 580             // If this is the first non-preferred node for this cycle, then
 581             // record it and continue searching for a preferred node. If this
 582             // is not the first non-preferred node, then treat it as though
 583             // there had been a hazard.
 584             NotPreferredSUnit = CurSUnit;
 585             continue;
 586           }
 587         } else {
 588           FoundSUnit = CurSUnit;
 589           break;
 590         }
 591       }
 592
 593       // Remember if this is a noop hazard.
 594       HasNoopHazards |= HT == ScheduleHazardRecognizer::NoopHazard;
 595
 596       NotReady.push_back(CurSUnit);
 597     }
 598
 599     // If we have a non-preferred node, push it back onto the available list.
 600     // If we did not find a preferred node, then schedule this first
 601     // non-preferred node.
 602     if (NotPreferredSUnit) {
 603       if (!FoundSUnit) {
 604         LLVM_DEBUG(
 605             dbgs() << "*** Will schedule a non-preferred instruction...\n");
 606         FoundSUnit = NotPreferredSUnit;
 607       } else {
 608         AvailableQueue.push(NotPreferredSUnit);
 609       }
 610
 611       NotPreferredSUnit = nullptr;
 612     }
 613
 614     // Add the nodes that aren't ready back onto the available list.
 615     if (!NotReady.empty()) {
 616       AvailableQueue.push_all(NotReady);
 617       NotReady.clear();
 618     }
 619
 620     // If we found a node to schedule...
 621     if (FoundSUnit) {
 622       // If we need to emit noops prior to this instruction, then do so.
 623       unsigned NumPreNoops = HazardRec->PreEmitNoops(FoundSUnit);
 624       for (unsigned i = 0; i != NumPreNoops; ++i)
 625         emitNoop(CurCycle);
 626
 627       // ... schedule the node...
 628       ScheduleNodeTopDown(FoundSUnit, CurCycle);
 629       HazardRec->EmitInstruction(FoundSUnit);
 630       CycleHasInsts = true;
 631       if (HazardRec->atIssueLimit()) {
 632         LLVM_DEBUG(dbgs() << "*** Max instructions per cycle " << CurCycle
 633                           << '\n');
 634         HazardRec->AdvanceCycle();
 635         ++CurCycle;
 636         CycleHasInsts = false;
 637       }
 638     } else {
 639       if (CycleHasInsts) {
 640         LLVM_DEBUG(dbgs() << "*** Finished cycle " << CurCycle << '\n');
 641         HazardRec->AdvanceCycle();
 642       } else if (!HasNoopHazards) {
 643         // Otherwise, we have a pipeline stall, but no other problem,
 644         // just advance the current cycle and try again.
 645         LLVM_DEBUG(dbgs() << "*** Stall in cycle " << CurCycle << '\n');
 646         HazardRec->AdvanceCycle();
 647         ++NumStalls;
 648       } else {
 649         // Otherwise, we have no instructions to issue and we have instructions
 650         // that will fault if we don't do this right.  This is the case for
 651         // processors without pipeline interlocks and other cases.
 652         emitNoop(CurCycle);
 653       }
 654
 655       ++CurCycle;
 656       CycleHasInsts = false;
 657     }
 658   }
 659
 660 #ifndef NDEBUG
 661   unsigned ScheduledNodes = VerifyScheduledDAG(/*isBottomUp=*/false);
 662   unsigned Noops = 0;
 663   for (unsigned i = 0, e = Sequence.size(); i != e; ++i)
 664     if (!Sequence[i])
 665       ++Noops;
 666   assert(Sequence.size() - Noops == ScheduledNodes &&
 667          "The number of nodes scheduled doesn't match the expected number!");
 668 #endif // NDEBUG
 669 }
 670
 671 // EmitSchedule - Emit the machine code in scheduled order.
 672 void SchedulePostRATDList::EmitSchedule() {
 673   RegionBegin = RegionEnd;
 674
 675   // If first instruction was a DBG_VALUE then put it back.
 676   if (FirstDbgValue)
 677     BB->splice(RegionEnd, BB, FirstDbgValue);
 678
 679   // Then re-insert them according to the given schedule.
 680   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
 681     if (SUnit *SU = Sequence[i])
 682       BB->splice(RegionEnd, BB, SU->getInstr());
 683     else
 684       // Null SUnit* is a noop.
 685       TII->insertNoop(*BB, RegionEnd);
 686
 687     // Update the Begin iterator, as the first instruction in the block
 688     // may have been scheduled later.
 689     if (i == 0)
 690       RegionBegin = std::prev(RegionEnd);
 691   }
 692
 693   // Reinsert any remaining debug_values.
 694   for (std::vector<std::pair<MachineInstr *, MachineInstr *> >::iterator
 695          DI = DbgValues.end(), DE = DbgValues.begin(); DI != DE; --DI) {
 696     std::pair<MachineInstr *, MachineInstr *> P = *std::prev(DI);
 697     MachineInstr *DbgValue = P.first;
 698     MachineBasicBlock::iterator OrigPrivMI = P.second;
 699     BB->splice(++OrigPrivMI, BB, DbgValue);
 700   }
 701   DbgValues.clear();
 702   FirstDbgValue = nullptr;
 703 }