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