llvm/lib/CodeGen/ScheduleDAG.cpp

   1 //===- ScheduleDAG.cpp - Implement the ScheduleDAG class ------------------===//
   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 /// \file Implements the ScheduleDAG class, which is a base class used by
  10 /// scheduling implementation classes.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "llvm/CodeGen/ScheduleDAG.h"
  15 #include "llvm/ADT/STLExtras.h"
  16 #include "llvm/ADT/SmallVector.h"
  17 #include "llvm/ADT/Statistic.h"
  18 #include "llvm/CodeGen/MachineFunction.h"
  19 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
  20 #include "llvm/CodeGen/SelectionDAGNodes.h"
  21 #include "llvm/CodeGen/TargetInstrInfo.h"
  22 #include "llvm/CodeGen/TargetRegisterInfo.h"
  23 #include "llvm/CodeGen/TargetSubtargetInfo.h"
  24 #include "llvm/Config/llvm-config.h"
  25 #include "llvm/Support/CommandLine.h"
  26 #include "llvm/Support/Compiler.h"
  27 #include "llvm/Support/Debug.h"
  28 #include "llvm/Support/raw_ostream.h"
  29 #include <algorithm>
  30 #include <cassert>
  31 #include <iterator>
  32 #include <limits>
  33 #include <utility>
  34 #include <vector>
  35
  36 using namespace llvm;
  37
  38 #define DEBUG_TYPE "pre-RA-sched"
  39
  40 STATISTIC(NumNewPredsAdded, "Number of times a  single predecessor was added");
  41 STATISTIC(NumTopoInits,
  42           "Number of times the topological order has been recomputed");
  43
  44 #ifndef NDEBUG
  45 static cl::opt<bool> StressSchedOpt(
  46   "stress-sched", cl::Hidden, cl::init(false),
  47   cl::desc("Stress test instruction scheduling"));
  48 #endif
  49
  50 void SchedulingPriorityQueue::anchor() {}
  51
  52 ScheduleDAG::ScheduleDAG(MachineFunction &mf)
  53     : TM(mf.getTarget()), TII(mf.getSubtarget().getInstrInfo()),
  54       TRI(mf.getSubtarget().getRegisterInfo()), MF(mf),
  55       MRI(mf.getRegInfo()) {
  56 #ifndef NDEBUG
  57   StressSched = StressSchedOpt;
  58 #endif
  59 }
  60
  61 ScheduleDAG::~ScheduleDAG() = default;
  62
  63 void ScheduleDAG::clearDAG() {
  64   SUnits.clear();
  65   EntrySU = SUnit();
  66   ExitSU = SUnit();
  67 }
  68
  69 const MCInstrDesc *ScheduleDAG::getNodeDesc(const SDNode *Node) const {
  70   if (!Node || !Node->isMachineOpcode()) return nullptr;
  71   return &TII->get(Node->getMachineOpcode());
  72 }
  73
  74 LLVM_DUMP_METHOD void SDep::dump(const TargetRegisterInfo *TRI) const {
  75   switch (getKind()) {
  76   case Data:   dbgs() << "Data"; break;
  77   case Anti:   dbgs() << "Anti"; break;
  78   case Output: dbgs() << "Out "; break;
  79   case Order:  dbgs() << "Ord "; break;
  80   }
  81
  82   switch (getKind()) {
  83   case Data:
  84     dbgs() << " Latency=" << getLatency();
  85     if (TRI && isAssignedRegDep())
  86       dbgs() << " Reg=" << printReg(getReg(), TRI);
  87     break;
  88   case Anti:
  89   case Output:
  90     dbgs() << " Latency=" << getLatency();
  91     break;
  92   case Order:
  93     dbgs() << " Latency=" << getLatency();
  94     switch(Contents.OrdKind) {
  95     case Barrier:      dbgs() << " Barrier"; break;
  96     case MayAliasMem:
  97     case MustAliasMem: dbgs() << " Memory"; break;
  98     case Artificial:   dbgs() << " Artificial"; break;
  99     case Weak:         dbgs() << " Weak"; break;
 100     case Cluster:      dbgs() << " Cluster"; break;
 101     }
 102     break;
 103   }
 104 }
 105
 106 bool SUnit::addPred(const SDep &D, bool Required) {
 107   // If this node already has this dependence, don't add a redundant one.
 108   for (SDep &PredDep : Preds) {
 109     // Zero-latency weak edges may be added purely for heuristic ordering. Don't
 110     // add them if another kind of edge already exists.
 111     if (!Required && PredDep.getSUnit() == D.getSUnit())
 112       return false;
 113     if (PredDep.overlaps(D)) {
 114       // Extend the latency if needed. Equivalent to
 115       // removePred(PredDep) + addPred(D).
 116       if (PredDep.getLatency() < D.getLatency()) {
 117         SUnit *PredSU = PredDep.getSUnit();
 118         // Find the corresponding successor in N.
 119         SDep ForwardD = PredDep;
 120         ForwardD.setSUnit(this);
 121         for (SDep &SuccDep : PredSU->Succs) {
 122           if (SuccDep == ForwardD) {
 123             SuccDep.setLatency(D.getLatency());
 124             break;
 125           }
 126         }
 127         PredDep.setLatency(D.getLatency());
 128       }
 129       return false;
 130     }
 131   }
 132   // Now add a corresponding succ to N.
 133   SDep P = D;
 134   P.setSUnit(this);
 135   SUnit *N = D.getSUnit();
 136   // Update the bookkeeping.
 137   if (D.getKind() == SDep::Data) {
 138     assert(NumPreds < std::numeric_limits<unsigned>::max() &&
 139            "NumPreds will overflow!");
 140     assert(N->NumSuccs < std::numeric_limits<unsigned>::max() &&
 141            "NumSuccs will overflow!");
 142     ++NumPreds;
 143     ++N->NumSuccs;
 144   }
 145   if (!N->isScheduled) {
 146     if (D.isWeak()) {
 147       ++WeakPredsLeft;
 148     }
 149     else {
 150       assert(NumPredsLeft < std::numeric_limits<unsigned>::max() &&
 151              "NumPredsLeft will overflow!");
 152       ++NumPredsLeft;
 153     }
 154   }
 155   if (!isScheduled) {
 156     if (D.isWeak()) {
 157       ++N->WeakSuccsLeft;
 158     }
 159     else {
 160       assert(N->NumSuccsLeft < std::numeric_limits<unsigned>::max() &&
 161              "NumSuccsLeft will overflow!");
 162       ++N->NumSuccsLeft;
 163     }
 164   }
 165   Preds.push_back(D);
 166   N->Succs.push_back(P);
 167   if (P.getLatency() != 0) {
 168     this->setDepthDirty();
 169     N->setHeightDirty();
 170   }
 171   return true;
 172 }
 173
 174 void SUnit::removePred(const SDep &D) {
 175   // Find the matching predecessor.
 176   SmallVectorImpl<SDep>::iterator I = llvm::find(Preds, D);
 177   if (I == Preds.end())
 178     return;
 179   // Find the corresponding successor in N.
 180   SDep P = D;
 181   P.setSUnit(this);
 182   SUnit *N = D.getSUnit();
 183   SmallVectorImpl<SDep>::iterator Succ = llvm::find(N->Succs, P);
 184   assert(Succ != N->Succs.end() && "Mismatching preds / succs lists!");
 185   // Update the bookkeeping.
 186   if (P.getKind() == SDep::Data) {
 187     assert(NumPreds > 0 && "NumPreds will underflow!");
 188     assert(N->NumSuccs > 0 && "NumSuccs will underflow!");
 189     --NumPreds;
 190     --N->NumSuccs;
 191   }
 192   if (!N->isScheduled) {
 193     if (D.isWeak()) {
 194       assert(WeakPredsLeft > 0 && "WeakPredsLeft will underflow!");
 195       --WeakPredsLeft;
 196     } else {
 197       assert(NumPredsLeft > 0 && "NumPredsLeft will underflow!");
 198       --NumPredsLeft;
 199     }
 200   }
 201   if (!isScheduled) {
 202     if (D.isWeak()) {
 203       assert(N->WeakSuccsLeft > 0 && "WeakSuccsLeft will underflow!");
 204       --N->WeakSuccsLeft;
 205     } else {
 206       assert(N->NumSuccsLeft > 0 && "NumSuccsLeft will underflow!");
 207       --N->NumSuccsLeft;
 208     }
 209   }
 210   N->Succs.erase(Succ);
 211   Preds.erase(I);
 212   if (P.getLatency() != 0) {
 213     this->setDepthDirty();
 214     N->setHeightDirty();
 215   }
 216 }
 217
 218 void SUnit::setDepthDirty() {
 219   if (!isDepthCurrent) return;
 220   SmallVector<SUnit*, 8> WorkList;
 221   WorkList.push_back(this);
 222   do {
 223     SUnit *SU = WorkList.pop_back_val();
 224     SU->isDepthCurrent = false;
 225     for (SDep &SuccDep : SU->Succs) {
 226       SUnit *SuccSU = SuccDep.getSUnit();
 227       if (SuccSU->isDepthCurrent)
 228         WorkList.push_back(SuccSU);
 229     }
 230   } while (!WorkList.empty());
 231 }
 232
 233 void SUnit::setHeightDirty() {
 234   if (!isHeightCurrent) return;
 235   SmallVector<SUnit*, 8> WorkList;
 236   WorkList.push_back(this);
 237   do {
 238     SUnit *SU = WorkList.pop_back_val();
 239     SU->isHeightCurrent = false;
 240     for (SDep &PredDep : SU->Preds) {
 241       SUnit *PredSU = PredDep.getSUnit();
 242       if (PredSU->isHeightCurrent)
 243         WorkList.push_back(PredSU);
 244     }
 245   } while (!WorkList.empty());
 246 }
 247
 248 void SUnit::setDepthToAtLeast(unsigned NewDepth) {
 249   if (NewDepth <= getDepth())
 250     return;
 251   setDepthDirty();
 252   Depth = NewDepth;
 253   isDepthCurrent = true;
 254 }
 255
 256 void SUnit::setHeightToAtLeast(unsigned NewHeight) {
 257   if (NewHeight <= getHeight())
 258     return;
 259   setHeightDirty();
 260   Height = NewHeight;
 261   isHeightCurrent = true;
 262 }
 263
 264 /// Calculates the maximal path from the node to the exit.
 265 void SUnit::ComputeDepth() {
 266   SmallVector<SUnit*, 8> WorkList;
 267   WorkList.push_back(this);
 268   do {
 269     SUnit *Cur = WorkList.back();
 270
 271     bool Done = true;
 272     unsigned MaxPredDepth = 0;
 273     for (const SDep &PredDep : Cur->Preds) {
 274       SUnit *PredSU = PredDep.getSUnit();
 275       if (PredSU->isDepthCurrent)
 276         MaxPredDepth = std::max(MaxPredDepth,
 277                                 PredSU->Depth + PredDep.getLatency());
 278       else {
 279         Done = false;
 280         WorkList.push_back(PredSU);
 281       }
 282     }
 283
 284     if (Done) {
 285       WorkList.pop_back();
 286       if (MaxPredDepth != Cur->Depth) {
 287         Cur->setDepthDirty();
 288         Cur->Depth = MaxPredDepth;
 289       }
 290       Cur->isDepthCurrent = true;
 291     }
 292   } while (!WorkList.empty());
 293 }
 294
 295 /// Calculates the maximal path from the node to the entry.
 296 void SUnit::ComputeHeight() {
 297   SmallVector<SUnit*, 8> WorkList;
 298   WorkList.push_back(this);
 299   do {
 300     SUnit *Cur = WorkList.back();
 301
 302     bool Done = true;
 303     unsigned MaxSuccHeight = 0;
 304     for (const SDep &SuccDep : Cur->Succs) {
 305       SUnit *SuccSU = SuccDep.getSUnit();
 306       if (SuccSU->isHeightCurrent)
 307         MaxSuccHeight = std::max(MaxSuccHeight,
 308                                  SuccSU->Height + SuccDep.getLatency());
 309       else {
 310         Done = false;
 311         WorkList.push_back(SuccSU);
 312       }
 313     }
 314
 315     if (Done) {
 316       WorkList.pop_back();
 317       if (MaxSuccHeight != Cur->Height) {
 318         Cur->setHeightDirty();
 319         Cur->Height = MaxSuccHeight;
 320       }
 321       Cur->isHeightCurrent = true;
 322     }
 323   } while (!WorkList.empty());
 324 }
 325
 326 void SUnit::biasCriticalPath() {
 327   if (NumPreds < 2)
 328     return;
 329
 330   SUnit::pred_iterator BestI = Preds.begin();
 331   unsigned MaxDepth = BestI->getSUnit()->getDepth();
 332   for (SUnit::pred_iterator I = std::next(BestI), E = Preds.end(); I != E;
 333        ++I) {
 334     if (I->getKind() == SDep::Data && I->getSUnit()->getDepth() > MaxDepth)
 335       BestI = I;
 336   }
 337   if (BestI != Preds.begin())
 338     std::swap(*Preds.begin(), *BestI);
 339 }
 340
 341 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 342 LLVM_DUMP_METHOD void SUnit::dumpAttributes() const {
 343   dbgs() << "  # preds left       : " << NumPredsLeft << "\n";
 344   dbgs() << "  # succs left       : " << NumSuccsLeft << "\n";
 345   if (WeakPredsLeft)
 346     dbgs() << "  # weak preds left  : " << WeakPredsLeft << "\n";
 347   if (WeakSuccsLeft)
 348     dbgs() << "  # weak succs left  : " << WeakSuccsLeft << "\n";
 349   dbgs() << "  # rdefs left       : " << NumRegDefsLeft << "\n";
 350   dbgs() << "  Latency            : " << Latency << "\n";
 351   dbgs() << "  Depth              : " << getDepth() << "\n";
 352   dbgs() << "  Height             : " << getHeight() << "\n";
 353 }
 354
 355 LLVM_DUMP_METHOD void ScheduleDAG::dumpNodeName(const SUnit &SU) const {
 356   if (&SU == &EntrySU)
 357     dbgs() << "EntrySU";
 358   else if (&SU == &ExitSU)
 359     dbgs() << "ExitSU";
 360   else
 361     dbgs() << "SU(" << SU.NodeNum << ")";
 362 }
 363
 364 LLVM_DUMP_METHOD void ScheduleDAG::dumpNodeAll(const SUnit &SU) const {
 365   dumpNode(SU);
 366   SU.dumpAttributes();
 367   if (SU.Preds.size() > 0) {
 368     dbgs() << "  Predecessors:\n";
 369     for (const SDep &Dep : SU.Preds) {
 370       dbgs() << "    ";
 371       dumpNodeName(*Dep.getSUnit());
 372       dbgs() << ": ";
 373       Dep.dump(TRI);
 374       dbgs() << '\n';
 375     }
 376   }
 377   if (SU.Succs.size() > 0) {
 378     dbgs() << "  Successors:\n";
 379     for (const SDep &Dep : SU.Succs) {
 380       dbgs() << "    ";
 381       dumpNodeName(*Dep.getSUnit());
 382       dbgs() << ": ";
 383       Dep.dump(TRI);
 384       dbgs() << '\n';
 385     }
 386   }
 387 }
 388 #endif
 389
 390 #ifndef NDEBUG
 391 unsigned ScheduleDAG::VerifyScheduledDAG(bool isBottomUp) {
 392   bool AnyNotSched = false;
 393   unsigned DeadNodes = 0;
 394   for (const SUnit &SUnit : SUnits) {
 395     if (!SUnit.isScheduled) {
 396       if (SUnit.NumPreds == 0 && SUnit.NumSuccs == 0) {
 397         ++DeadNodes;
 398         continue;
 399       }
 400       if (!AnyNotSched)
 401         dbgs() << "*** Scheduling failed! ***\n";
 402       dumpNode(SUnit);
 403       dbgs() << "has not been scheduled!\n";
 404       AnyNotSched = true;
 405     }
 406     if (SUnit.isScheduled &&
 407         (isBottomUp ? SUnit.getHeight() : SUnit.getDepth()) >
 408           unsigned(std::numeric_limits<int>::max())) {
 409       if (!AnyNotSched)
 410         dbgs() << "*** Scheduling failed! ***\n";
 411       dumpNode(SUnit);
 412       dbgs() << "has an unexpected "
 413            << (isBottomUp ? "Height" : "Depth") << " value!\n";
 414       AnyNotSched = true;
 415     }
 416     if (isBottomUp) {
 417       if (SUnit.NumSuccsLeft != 0) {
 418         if (!AnyNotSched)
 419           dbgs() << "*** Scheduling failed! ***\n";
 420         dumpNode(SUnit);
 421         dbgs() << "has successors left!\n";
 422         AnyNotSched = true;
 423       }
 424     } else {
 425       if (SUnit.NumPredsLeft != 0) {
 426         if (!AnyNotSched)
 427           dbgs() << "*** Scheduling failed! ***\n";
 428         dumpNode(SUnit);
 429         dbgs() << "has predecessors left!\n";
 430         AnyNotSched = true;
 431       }
 432     }
 433   }
 434   assert(!AnyNotSched);
 435   return SUnits.size() - DeadNodes;
 436 }
 437 #endif
 438
 439 void ScheduleDAGTopologicalSort::InitDAGTopologicalSorting() {
 440   // The idea of the algorithm is taken from
 441   // "Online algorithms for managing the topological order of
 442   // a directed acyclic graph" by David J. Pearce and Paul H.J. Kelly
 443   // This is the MNR algorithm, which was first introduced by
 444   // A. Marchetti-Spaccamela, U. Nanni and H. Rohnert in
 445   // "Maintaining a topological order under edge insertions".
 446   //
 447   // Short description of the algorithm:
 448   //
 449   // Topological ordering, ord, of a DAG maps each node to a topological
 450   // index so that for all edges X->Y it is the case that ord(X) < ord(Y).
 451   //
 452   // This means that if there is a path from the node X to the node Z,
 453   // then ord(X) < ord(Z).
 454   //
 455   // This property can be used to check for reachability of nodes:
 456   // if Z is reachable from X, then an insertion of the edge Z->X would
 457   // create a cycle.
 458   //
 459   // The algorithm first computes a topological ordering for the DAG by
 460   // initializing the Index2Node and Node2Index arrays and then tries to keep
 461   // the ordering up-to-date after edge insertions by reordering the DAG.
 462   //
 463   // On insertion of the edge X->Y, the algorithm first marks by calling DFS
 464   // the nodes reachable from Y, and then shifts them using Shift to lie
 465   // immediately after X in Index2Node.
 466
 467   // Cancel pending updates, mark as valid.
 468   Dirty = false;
 469   Updates.clear();
 470
 471   unsigned DAGSize = SUnits.size();
 472   std::vector<SUnit*> WorkList;
 473   WorkList.reserve(DAGSize);
 474
 475   Index2Node.resize(DAGSize);
 476   Node2Index.resize(DAGSize);
 477
 478   // Initialize the data structures.
 479   if (ExitSU)
 480     WorkList.push_back(ExitSU);
 481   for (SUnit &SU : SUnits) {
 482     int NodeNum = SU.NodeNum;
 483     unsigned Degree = SU.Succs.size();
 484     // Temporarily use the Node2Index array as scratch space for degree counts.
 485     Node2Index[NodeNum] = Degree;
 486
 487     // Is it a node without dependencies?
 488     if (Degree == 0) {
 489       assert(SU.Succs.empty() && "SUnit should have no successors");
 490       // Collect leaf nodes.
 491       WorkList.push_back(&SU);
 492     }
 493   }
 494
 495   int Id = DAGSize;
 496   while (!WorkList.empty()) {
 497     SUnit *SU = WorkList.back();
 498     WorkList.pop_back();
 499     if (SU->NodeNum < DAGSize)
 500       Allocate(SU->NodeNum, --Id);
 501     for (const SDep &PredDep : SU->Preds) {
 502       SUnit *SU = PredDep.getSUnit();
 503       if (SU->NodeNum < DAGSize && !--Node2Index[SU->NodeNum])
 504         // If all dependencies of the node are processed already,
 505         // then the node can be computed now.
 506         WorkList.push_back(SU);
 507     }
 508   }
 509
 510   Visited.resize(DAGSize);
 511   NumTopoInits++;
 512
 513 #ifndef NDEBUG
 514   // Check correctness of the ordering
 515   for (SUnit &SU : SUnits)  {
 516     for (const SDep &PD : SU.Preds) {
 517       assert(Node2Index[SU.NodeNum] > Node2Index[PD.getSUnit()->NodeNum] &&
 518       "Wrong topological sorting");
 519     }
 520   }
 521 #endif
 522 }
 523
 524 void ScheduleDAGTopologicalSort::FixOrder() {
 525   // Recompute from scratch after new nodes have been added.
 526   if (Dirty) {
 527     InitDAGTopologicalSorting();
 528     return;
 529   }
 530
 531   // Otherwise apply updates one-by-one.
 532   for (auto &U : Updates)
 533     AddPred(U.first, U.second);
 534   Updates.clear();
 535 }
 536
 537 void ScheduleDAGTopologicalSort::AddPredQueued(SUnit *Y, SUnit *X) {
 538   // Recomputing the order from scratch is likely more efficient than applying
 539   // updates one-by-one for too many updates. The current cut-off is arbitrarily
 540   // chosen.
 541   Dirty = Dirty || Updates.size() > 10;
 542
 543   if (Dirty)
 544     return;
 545
 546   Updates.emplace_back(Y, X);
 547 }
 548
 549 void ScheduleDAGTopologicalSort::AddPred(SUnit *Y, SUnit *X) {
 550   int UpperBound, LowerBound;
 551   LowerBound = Node2Index[Y->NodeNum];
 552   UpperBound = Node2Index[X->NodeNum];
 553   bool HasLoop = false;
 554   // Is Ord(X) < Ord(Y) ?
 555   if (LowerBound < UpperBound) {
 556     // Update the topological order.
 557     Visited.reset();
 558     DFS(Y, UpperBound, HasLoop);
 559     assert(!HasLoop && "Inserted edge creates a loop!");
 560     // Recompute topological indexes.
 561     Shift(Visited, LowerBound, UpperBound);
 562   }
 563
 564   NumNewPredsAdded++;
 565 }
 566
 567 void ScheduleDAGTopologicalSort::RemovePred(SUnit *M, SUnit *N) {
 568   // InitDAGTopologicalSorting();
 569 }
 570
 571 void ScheduleDAGTopologicalSort::DFS(const SUnit *SU, int UpperBound,
 572                                      bool &HasLoop) {
 573   std::vector<const SUnit*> WorkList;
 574   WorkList.reserve(SUnits.size());
 575
 576   WorkList.push_back(SU);
 577   do {
 578     SU = WorkList.back();
 579     WorkList.pop_back();
 580     Visited.set(SU->NodeNum);
 581     for (const SDep &SuccDep : llvm::reverse(SU->Succs)) {
 582       unsigned s = SuccDep.getSUnit()->NodeNum;
 583       // Edges to non-SUnits are allowed but ignored (e.g. ExitSU).
 584       if (s >= Node2Index.size())
 585         continue;
 586       if (Node2Index[s] == UpperBound) {
 587         HasLoop = true;
 588         return;
 589       }
 590       // Visit successors if not already and in affected region.
 591       if (!Visited.test(s) && Node2Index[s] < UpperBound) {
 592         WorkList.push_back(SuccDep.getSUnit());
 593       }
 594     }
 595   } while (!WorkList.empty());
 596 }
 597
 598 std::vector<int> ScheduleDAGTopologicalSort::GetSubGraph(const SUnit &StartSU,
 599                                                          const SUnit &TargetSU,
 600                                                          bool &Success) {
 601   std::vector<const SUnit*> WorkList;
 602   int LowerBound = Node2Index[StartSU.NodeNum];
 603   int UpperBound = Node2Index[TargetSU.NodeNum];
 604   bool Found = false;
 605   BitVector VisitedBack;
 606   std::vector<int> Nodes;
 607
 608   if (LowerBound > UpperBound) {
 609     Success = false;
 610     return Nodes;
 611   }
 612
 613   WorkList.reserve(SUnits.size());
 614   Visited.reset();
 615
 616   // Starting from StartSU, visit all successors up
 617   // to UpperBound.
 618   WorkList.push_back(&StartSU);
 619   do {
 620     const SUnit *SU = WorkList.back();
 621     WorkList.pop_back();
 622     for (const SDep &SD : llvm::reverse(SU->Succs)) {
 623       const SUnit *Succ = SD.getSUnit();
 624       unsigned s = Succ->NodeNum;
 625       // Edges to non-SUnits are allowed but ignored (e.g. ExitSU).
 626       if (Succ->isBoundaryNode())
 627         continue;
 628       if (Node2Index[s] == UpperBound) {
 629         Found = true;
 630         continue;
 631       }
 632       // Visit successors if not already and in affected region.
 633       if (!Visited.test(s) && Node2Index[s] < UpperBound) {
 634         Visited.set(s);
 635         WorkList.push_back(Succ);
 636       }
 637     }
 638   } while (!WorkList.empty());
 639
 640   if (!Found) {
 641     Success = false;
 642     return Nodes;
 643   }
 644
 645   WorkList.clear();
 646   VisitedBack.resize(SUnits.size());
 647   Found = false;
 648
 649   // Starting from TargetSU, visit all predecessors up
 650   // to LowerBound. SUs that are visited by the two
 651   // passes are added to Nodes.
 652   WorkList.push_back(&TargetSU);
 653   do {
 654     const SUnit *SU = WorkList.back();
 655     WorkList.pop_back();
 656     for (const SDep &SD : llvm::reverse(SU->Preds)) {
 657       const SUnit *Pred = SD.getSUnit();
 658       unsigned s = Pred->NodeNum;
 659       // Edges to non-SUnits are allowed but ignored (e.g. EntrySU).
 660       if (Pred->isBoundaryNode())
 661         continue;
 662       if (Node2Index[s] == LowerBound) {
 663         Found = true;
 664         continue;
 665       }
 666       if (!VisitedBack.test(s) && Visited.test(s)) {
 667         VisitedBack.set(s);
 668         WorkList.push_back(Pred);
 669         Nodes.push_back(s);
 670       }
 671     }
 672   } while (!WorkList.empty());
 673
 674   assert(Found && "Error in SUnit Graph!");
 675   Success = true;
 676   return Nodes;
 677 }
 678
 679 void ScheduleDAGTopologicalSort::Shift(BitVector& Visited, int LowerBound,
 680                                        int UpperBound) {
 681   std::vector<int> L;
 682   int shift = 0;
 683   int i;
 684
 685   for (i = LowerBound; i <= UpperBound; ++i) {
 686     // w is node at topological index i.
 687     int w = Index2Node[i];
 688     if (Visited.test(w)) {
 689       // Unmark.
 690       Visited.reset(w);
 691       L.push_back(w);
 692       shift = shift + 1;
 693     } else {
 694       Allocate(w, i - shift);
 695     }
 696   }
 697
 698   for (unsigned LI : L) {
 699     Allocate(LI, i - shift);
 700     i = i + 1;
 701   }
 702 }
 703
 704 bool ScheduleDAGTopologicalSort::WillCreateCycle(SUnit *TargetSU, SUnit *SU) {
 705   FixOrder();
 706   // Is SU reachable from TargetSU via successor edges?
 707   if (IsReachable(SU, TargetSU))
 708     return true;
 709   for (const SDep &PredDep : TargetSU->Preds)
 710     if (PredDep.isAssignedRegDep() &&
 711         IsReachable(SU, PredDep.getSUnit()))
 712       return true;
 713   return false;
 714 }
 715
 716 void ScheduleDAGTopologicalSort::AddSUnitWithoutPredecessors(const SUnit *SU) {
 717   assert(SU->NodeNum == Index2Node.size() && "Node cannot be added at the end");
 718   assert(SU->NumPreds == 0 && "Can only add SU's with no predecessors");
 719   Node2Index.push_back(Index2Node.size());
 720   Index2Node.push_back(SU->NodeNum);
 721   Visited.resize(Node2Index.size());
 722 }
 723
 724 bool ScheduleDAGTopologicalSort::IsReachable(const SUnit *SU,
 725                                              const SUnit *TargetSU) {
 726   assert(TargetSU != nullptr && "Invalid target SUnit");
 727   assert(SU != nullptr && "Invalid SUnit");
 728   FixOrder();
 729   // If insertion of the edge SU->TargetSU would create a cycle
 730   // then there is a path from TargetSU to SU.
 731   int UpperBound, LowerBound;
 732   LowerBound = Node2Index[TargetSU->NodeNum];
 733   UpperBound = Node2Index[SU->NodeNum];
 734   bool HasLoop = false;
 735   // Is Ord(TargetSU) < Ord(SU) ?
 736   if (LowerBound < UpperBound) {
 737     Visited.reset();
 738     // There may be a path from TargetSU to SU. Check for it.
 739     DFS(TargetSU, UpperBound, HasLoop);
 740   }
 741   return HasLoop;
 742 }
 743
 744 void ScheduleDAGTopologicalSort::Allocate(int n, int index) {
 745   Node2Index[n] = index;
 746   Index2Node[index] = n;
 747 }
 748
 749 ScheduleDAGTopologicalSort::
 750 ScheduleDAGTopologicalSort(std::vector<SUnit> &sunits, SUnit *exitsu)
 751   : SUnits(sunits), ExitSU(exitsu) {}
 752
 753 ScheduleHazardRecognizer::~ScheduleHazardRecognizer() = default;