lib/CodeGen/SelectionDAG/ScheduleDAGFast.cpp

   1 //===----- ScheduleDAGFast.cpp - Fast poor list scheduler -----------------===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This implements a fast scheduler.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #define DEBUG_TYPE "pre-RA-sched"
  15 #include "ScheduleDAGSDNodes.h"
  16 #include "llvm/CodeGen/SchedulerRegistry.h"
  17 #include "llvm/CodeGen/SelectionDAGISel.h"
  18 #include "llvm/Target/TargetRegisterInfo.h"
  19 #include "llvm/Target/TargetData.h"
  20 #include "llvm/Target/TargetInstrInfo.h"
  21 #include "llvm/Support/Debug.h"
  22 #include "llvm/Support/Compiler.h"
  23 #include "llvm/ADT/SmallSet.h"
  24 #include "llvm/ADT/Statistic.h"
  25 #include "llvm/ADT/STLExtras.h"
  26 #include "llvm/Support/CommandLine.h"
  27 using namespace llvm;
  28
  29 STATISTIC(NumUnfolds,    "Number of nodes unfolded");
  30 STATISTIC(NumDups,       "Number of duplicated nodes");
  31 STATISTIC(NumPRCopies,   "Number of physical copies");
  32
  33 static RegisterScheduler
  34   fastDAGScheduler("fast", "Fast suboptimal list scheduling",
  35                    createFastDAGScheduler);
  36
  37 namespace {
  38   /// FastPriorityQueue - A degenerate priority queue that considers
  39   /// all nodes to have the same priority.
  40   ///
  41   struct VISIBILITY_HIDDEN FastPriorityQueue {
  42     SmallVector<SUnit *, 16> Queue;
  43
  44     bool empty() const { return Queue.empty(); }
  45
  46     void push(SUnit *U) {
  47       Queue.push_back(U);
  48     }
  49
  50     SUnit *pop() {
  51       if (empty()) return NULL;
  52       SUnit *V = Queue.back();
  53       Queue.pop_back();
  54       return V;
  55     }
  56   };
  57
  58 //===----------------------------------------------------------------------===//
  59 /// ScheduleDAGFast - The actual "fast" list scheduler implementation.
  60 ///
  61 class VISIBILITY_HIDDEN ScheduleDAGFast : public ScheduleDAGSDNodes {
  62 private:
  63   /// AvailableQueue - The priority queue to use for the available SUnits.
  64   FastPriorityQueue AvailableQueue;
  65
  66   /// LiveRegDefs - A set of physical registers and their definition
  67   /// that are "live". These nodes must be scheduled before any other nodes that
  68   /// modifies the registers can be scheduled.
  69   unsigned NumLiveRegs;
  70   std::vector<SUnit*> LiveRegDefs;
  71   std::vector<unsigned> LiveRegCycles;
  72
  73 public:
  74   ScheduleDAGFast(MachineFunction &mf)
  75     : ScheduleDAGSDNodes(mf) {}
  76
  77   void Schedule();
  78
  79   /// AddPred - adds a predecessor edge to SUnit SU.
  80   /// This returns true if this is a new predecessor.
  81   void AddPred(SUnit *SU, const SDep &D) {
  82     SU->addPred(D);
  83   }
  84
  85   /// RemovePred - removes a predecessor edge from SUnit SU.
  86   /// This returns true if an edge was removed.
  87   void RemovePred(SUnit *SU, const SDep &D) {
  88     SU->removePred(D);
  89   }
  90
  91 private:
  92   void ReleasePred(SUnit *SU, SDep *PredEdge);
  93   void ReleasePredecessors(SUnit *SU, unsigned CurCycle);
  94   void ScheduleNodeBottomUp(SUnit*, unsigned);
  95   SUnit *CopyAndMoveSuccessors(SUnit*);
  96   void InsertCopiesAndMoveSuccs(SUnit*, unsigned,
  97                                 const TargetRegisterClass*,
  98                                 const TargetRegisterClass*,
  99                                 SmallVector<SUnit*, 2>&);
 100   bool DelayForLiveRegsBottomUp(SUnit*, SmallVector<unsigned, 4>&);
 101   void ListScheduleBottomUp();
 102
 103   /// ForceUnitLatencies - The fast scheduler doesn't care about real latencies.
 104   bool ForceUnitLatencies() const { return true; }
 105 };
 106 }  // end anonymous namespace
 107
 108
 109 /// Schedule - Schedule the DAG using list scheduling.
 110 void ScheduleDAGFast::Schedule() {
 111   DOUT << "********** List Scheduling **********\n";
 112
 113   NumLiveRegs = 0;
 114   LiveRegDefs.resize(TRI->getNumRegs(), NULL);
 115   LiveRegCycles.resize(TRI->getNumRegs(), 0);
 116
 117   // Build the scheduling graph.
 118   BuildSchedGraph();
 119
 120   DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
 121           SUnits[su].dumpAll(this));
 122
 123   // Execute the actual scheduling loop.
 124   ListScheduleBottomUp();
 125 }
 126
 127 //===----------------------------------------------------------------------===//
 128 //  Bottom-Up Scheduling
 129 //===----------------------------------------------------------------------===//
 130
 131 /// ReleasePred - Decrement the NumSuccsLeft count of a predecessor. Add it to
 132 /// the AvailableQueue if the count reaches zero. Also update its cycle bound.
 133 void ScheduleDAGFast::ReleasePred(SUnit *SU, SDep *PredEdge) {
 134   SUnit *PredSU = PredEdge->getSUnit();
 135   --PredSU->NumSuccsLeft;
 136
 137 #ifndef NDEBUG
 138   if (PredSU->NumSuccsLeft < 0) {
 139     cerr << "*** Scheduling failed! ***\n";
 140     PredSU->dump(this);
 141     cerr << " has been released too many times!\n";
 142     assert(0);
 143   }
 144 #endif
 145
 146   // If all the node's successors are scheduled, this node is ready
 147   // to be scheduled. Ignore the special EntrySU node.
 148   if (PredSU->NumSuccsLeft == 0 && PredSU != &EntrySU) {
 149     PredSU->isAvailable = true;
 150     AvailableQueue.push(PredSU);
 151   }
 152 }
 153
 154 void ScheduleDAGFast::ReleasePredecessors(SUnit *SU, unsigned CurCycle) {
 155   // Bottom up: release predecessors
 156   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 157        I != E; ++I) {
 158     ReleasePred(SU, &*I);
 159     if (I->isAssignedRegDep()) {
 160       // This is a physical register dependency and it's impossible or
 161       // expensive to copy the register. Make sure nothing that can
 162       // clobber the register is scheduled between the predecessor and
 163       // this node.
 164       if (!LiveRegDefs[I->getReg()]) {
 165         ++NumLiveRegs;
 166         LiveRegDefs[I->getReg()] = I->getSUnit();
 167         LiveRegCycles[I->getReg()] = CurCycle;
 168       }
 169     }
 170   }
 171 }
 172
 173 /// ScheduleNodeBottomUp - Add the node to the schedule. Decrement the pending
 174 /// count of its predecessors. If a predecessor pending count is zero, add it to
 175 /// the Available queue.
 176 void ScheduleDAGFast::ScheduleNodeBottomUp(SUnit *SU, unsigned CurCycle) {
 177   DOUT << "*** Scheduling [" << CurCycle << "]: ";
 178   DEBUG(SU->dump(this));
 179
 180   assert(CurCycle >= SU->getHeight() && "Node scheduled below its height!");
 181   SU->setHeightToAtLeast(CurCycle);
 182   Sequence.push_back(SU);
 183
 184   ReleasePredecessors(SU, CurCycle);
 185
 186   // Release all the implicit physical register defs that are live.
 187   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 188        I != E; ++I) {
 189     if (I->isAssignedRegDep()) {
 190       if (LiveRegCycles[I->getReg()] == I->getSUnit()->getHeight()) {
 191         assert(NumLiveRegs > 0 && "NumLiveRegs is already zero!");
 192         assert(LiveRegDefs[I->getReg()] == SU &&
 193                "Physical register dependency violated?");
 194         --NumLiveRegs;
 195         LiveRegDefs[I->getReg()] = NULL;
 196         LiveRegCycles[I->getReg()] = 0;
 197       }
 198     }
 199   }
 200
 201   SU->isScheduled = true;
 202 }
 203
 204 /// CopyAndMoveSuccessors - Clone the specified node and move its scheduled
 205 /// successors to the newly created node.
 206 SUnit *ScheduleDAGFast::CopyAndMoveSuccessors(SUnit *SU) {
 207   if (SU->getNode()->getFlaggedNode())
 208     return NULL;
 209
 210   SDNode *N = SU->getNode();
 211   if (!N)
 212     return NULL;
 213
 214   SUnit *NewSU;
 215   bool TryUnfold = false;
 216   for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
 217     MVT VT = N->getValueType(i);
 218     if (VT == MVT::Flag)
 219       return NULL;
 220     else if (VT == MVT::Other)
 221       TryUnfold = true;
 222   }
 223   for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
 224     const SDValue &Op = N->getOperand(i);
 225     MVT VT = Op.getNode()->getValueType(Op.getResNo());
 226     if (VT == MVT::Flag)
 227       return NULL;
 228   }
 229
 230   if (TryUnfold) {
 231     SmallVector<SDNode*, 2> NewNodes;
 232     if (!TII->unfoldMemoryOperand(*DAG, N, NewNodes))
 233       return NULL;
 234
 235     DOUT << "Unfolding SU # " << SU->NodeNum << "\n";
 236     assert(NewNodes.size() == 2 && "Expected a load folding node!");
 237
 238     N = NewNodes[1];
 239     SDNode *LoadNode = NewNodes[0];
 240     unsigned NumVals = N->getNumValues();
 241     unsigned OldNumVals = SU->getNode()->getNumValues();
 242     for (unsigned i = 0; i != NumVals; ++i)
 243       DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), i), SDValue(N, i));
 244     DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), OldNumVals-1),
 245                                    SDValue(LoadNode, 1));
 246
 247     SUnit *NewSU = NewSUnit(N);
 248     assert(N->getNodeId() == -1 && "Node already inserted!");
 249     N->setNodeId(NewSU->NodeNum);
 250
 251     const TargetInstrDesc &TID = TII->get(N->getMachineOpcode());
 252     for (unsigned i = 0; i != TID.getNumOperands(); ++i) {
 253       if (TID.getOperandConstraint(i, TOI::TIED_TO) != -1) {
 254         NewSU->isTwoAddress = true;
 255         break;
 256       }
 257     }
 258     if (TID.isCommutable())
 259       NewSU->isCommutable = true;
 260
 261     // LoadNode may already exist. This can happen when there is another
 262     // load from the same location and producing the same type of value
 263     // but it has different alignment or volatileness.
 264     bool isNewLoad = true;
 265     SUnit *LoadSU;
 266     if (LoadNode->getNodeId() != -1) {
 267       LoadSU = &SUnits[LoadNode->getNodeId()];
 268       isNewLoad = false;
 269     } else {
 270       LoadSU = NewSUnit(LoadNode);
 271       LoadNode->setNodeId(LoadSU->NodeNum);
 272     }
 273
 274     SDep ChainPred;
 275     SmallVector<SDep, 4> ChainSuccs;
 276     SmallVector<SDep, 4> LoadPreds;
 277     SmallVector<SDep, 4> NodePreds;
 278     SmallVector<SDep, 4> NodeSuccs;
 279     for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 280          I != E; ++I) {
 281       if (I->isCtrl())
 282         ChainPred = *I;
 283       else if (I->getSUnit()->getNode() &&
 284                I->getSUnit()->getNode()->isOperandOf(LoadNode))
 285         LoadPreds.push_back(*I);
 286       else
 287         NodePreds.push_back(*I);
 288     }
 289     for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 290          I != E; ++I) {
 291       if (I->isCtrl())
 292         ChainSuccs.push_back(*I);
 293       else
 294         NodeSuccs.push_back(*I);
 295     }
 296
 297     if (ChainPred.getSUnit()) {
 298       RemovePred(SU, ChainPred);
 299       if (isNewLoad)
 300         AddPred(LoadSU, ChainPred);
 301     }
 302     for (unsigned i = 0, e = LoadPreds.size(); i != e; ++i) {
 303       const SDep &Pred = LoadPreds[i];
 304       RemovePred(SU, Pred);
 305       if (isNewLoad) {
 306         AddPred(LoadSU, Pred);
 307       }
 308     }
 309     for (unsigned i = 0, e = NodePreds.size(); i != e; ++i) {
 310       const SDep &Pred = NodePreds[i];
 311       RemovePred(SU, Pred);
 312       AddPred(NewSU, Pred);
 313     }
 314     for (unsigned i = 0, e = NodeSuccs.size(); i != e; ++i) {
 315       SDep D = NodeSuccs[i];
 316       SUnit *SuccDep = D.getSUnit();
 317       D.setSUnit(SU);
 318       RemovePred(SuccDep, D);
 319       D.setSUnit(NewSU);
 320       AddPred(SuccDep, D);
 321     }
 322     for (unsigned i = 0, e = ChainSuccs.size(); i != e; ++i) {
 323       SDep D = ChainSuccs[i];
 324       SUnit *SuccDep = D.getSUnit();
 325       D.setSUnit(SU);
 326       RemovePred(SuccDep, D);
 327       if (isNewLoad) {
 328         D.setSUnit(LoadSU);
 329         AddPred(SuccDep, D);
 330       }
 331     }
 332     if (isNewLoad) {
 333       AddPred(NewSU, SDep(LoadSU, SDep::Order, LoadSU->Latency));
 334     }
 335
 336     ++NumUnfolds;
 337
 338     if (NewSU->NumSuccsLeft == 0) {
 339       NewSU->isAvailable = true;
 340       return NewSU;
 341     }
 342     SU = NewSU;
 343   }
 344
 345   DOUT << "Duplicating SU # " << SU->NodeNum << "\n";
 346   NewSU = Clone(SU);
 347
 348   // New SUnit has the exact same predecessors.
 349   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 350        I != E; ++I)
 351     if (!I->isArtificial())
 352       AddPred(NewSU, *I);
 353
 354   // Only copy scheduled successors. Cut them from old node's successor
 355   // list and move them over.
 356   SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
 357   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 358        I != E; ++I) {
 359     if (I->isArtificial())
 360       continue;
 361     SUnit *SuccSU = I->getSUnit();
 362     if (SuccSU->isScheduled) {
 363       SDep D = *I;
 364       D.setSUnit(NewSU);
 365       AddPred(SuccSU, D);
 366       D.setSUnit(SU);
 367       DelDeps.push_back(std::make_pair(SuccSU, D));
 368     }
 369   }
 370   for (unsigned i = 0, e = DelDeps.size(); i != e; ++i)
 371     RemovePred(DelDeps[i].first, DelDeps[i].second);
 372
 373   ++NumDups;
 374   return NewSU;
 375 }
 376
 377 /// InsertCopiesAndMoveSuccs - Insert register copies and move all
 378 /// scheduled successors of the given SUnit to the last copy.
 379 void ScheduleDAGFast::InsertCopiesAndMoveSuccs(SUnit *SU, unsigned Reg,
 380                                               const TargetRegisterClass *DestRC,
 381                                               const TargetRegisterClass *SrcRC,
 382                                                SmallVector<SUnit*, 2> &Copies) {
 383   SUnit *CopyFromSU = NewSUnit(static_cast<SDNode *>(NULL));
 384   CopyFromSU->CopySrcRC = SrcRC;
 385   CopyFromSU->CopyDstRC = DestRC;
 386
 387   SUnit *CopyToSU = NewSUnit(static_cast<SDNode *>(NULL));
 388   CopyToSU->CopySrcRC = DestRC;
 389   CopyToSU->CopyDstRC = SrcRC;
 390
 391   // Only copy scheduled successors. Cut them from old node's successor
 392   // list and move them over.
 393   SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
 394   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 395        I != E; ++I) {
 396     if (I->isArtificial())
 397       continue;
 398     SUnit *SuccSU = I->getSUnit();
 399     if (SuccSU->isScheduled) {
 400       SDep D = *I;
 401       D.setSUnit(CopyToSU);
 402       AddPred(SuccSU, D);
 403       DelDeps.push_back(std::make_pair(SuccSU, *I));
 404     }
 405   }
 406   for (unsigned i = 0, e = DelDeps.size(); i != e; ++i) {
 407     RemovePred(DelDeps[i].first, DelDeps[i].second);
 408   }
 409
 410   AddPred(CopyFromSU, SDep(SU, SDep::Data, SU->Latency, Reg));
 411   AddPred(CopyToSU, SDep(CopyFromSU, SDep::Data, CopyFromSU->Latency, 0));
 412
 413   Copies.push_back(CopyFromSU);
 414   Copies.push_back(CopyToSU);
 415
 416   ++NumPRCopies;
 417 }
 418
 419 /// getPhysicalRegisterVT - Returns the ValueType of the physical register
 420 /// definition of the specified node.
 421 /// FIXME: Move to SelectionDAG?
 422 static MVT getPhysicalRegisterVT(SDNode *N, unsigned Reg,
 423                                  const TargetInstrInfo *TII) {
 424   const TargetInstrDesc &TID = TII->get(N->getMachineOpcode());
 425   assert(TID.ImplicitDefs && "Physical reg def must be in implicit def list!");
 426   unsigned NumRes = TID.getNumDefs();
 427   for (const unsigned *ImpDef = TID.getImplicitDefs(); *ImpDef; ++ImpDef) {
 428     if (Reg == *ImpDef)
 429       break;
 430     ++NumRes;
 431   }
 432   return N->getValueType(NumRes);
 433 }
 434
 435 /// DelayForLiveRegsBottomUp - Returns true if it is necessary to delay
 436 /// scheduling of the given node to satisfy live physical register dependencies.
 437 /// If the specific node is the last one that's available to schedule, do
 438 /// whatever is necessary (i.e. backtracking or cloning) to make it possible.
 439 bool ScheduleDAGFast::DelayForLiveRegsBottomUp(SUnit *SU,
 440                                                SmallVector<unsigned, 4> &LRegs){
 441   if (NumLiveRegs == 0)
 442     return false;
 443
 444   SmallSet<unsigned, 4> RegAdded;
 445   // If this node would clobber any "live" register, then it's not ready.
 446   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 447        I != E; ++I) {
 448     if (I->isAssignedRegDep()) {
 449       unsigned Reg = I->getReg();
 450       if (LiveRegDefs[Reg] && LiveRegDefs[Reg] != I->getSUnit()) {
 451         if (RegAdded.insert(Reg))
 452           LRegs.push_back(Reg);
 453       }
 454       for (const unsigned *Alias = TRI->getAliasSet(Reg);
 455            *Alias; ++Alias)
 456         if (LiveRegDefs[*Alias] && LiveRegDefs[*Alias] != I->getSUnit()) {
 457           if (RegAdded.insert(*Alias))
 458             LRegs.push_back(*Alias);
 459         }
 460     }
 461   }
 462
 463   for (SDNode *Node = SU->getNode(); Node; Node = Node->getFlaggedNode()) {
 464     if (!Node->isMachineOpcode())
 465       continue;
 466     const TargetInstrDesc &TID = TII->get(Node->getMachineOpcode());
 467     if (!TID.ImplicitDefs)
 468       continue;
 469     for (const unsigned *Reg = TID.ImplicitDefs; *Reg; ++Reg) {
 470       if (LiveRegDefs[*Reg] && LiveRegDefs[*Reg] != SU) {
 471         if (RegAdded.insert(*Reg))
 472           LRegs.push_back(*Reg);
 473       }
 474       for (const unsigned *Alias = TRI->getAliasSet(*Reg);
 475            *Alias; ++Alias)
 476         if (LiveRegDefs[*Alias] && LiveRegDefs[*Alias] != SU) {
 477           if (RegAdded.insert(*Alias))
 478             LRegs.push_back(*Alias);
 479         }
 480     }
 481   }
 482   return !LRegs.empty();
 483 }
 484
 485
 486 /// ListScheduleBottomUp - The main loop of list scheduling for bottom-up
 487 /// schedulers.
 488 void ScheduleDAGFast::ListScheduleBottomUp() {
 489   unsigned CurCycle = 0;
 490
 491   // Release any predecessors of the special Exit node.
 492   ReleasePredecessors(&ExitSU, CurCycle);
 493
 494   // Add root to Available queue.
 495   if (!SUnits.empty()) {
 496     SUnit *RootSU = &SUnits[DAG->getRoot().getNode()->getNodeId()];
 497     assert(RootSU->Succs.empty() && "Graph root shouldn't have successors!");
 498     RootSU->isAvailable = true;
 499     AvailableQueue.push(RootSU);
 500   }
 501
 502   // While Available queue is not empty, grab the node with the highest
 503   // priority. If it is not ready put it back.  Schedule the node.
 504   SmallVector<SUnit*, 4> NotReady;
 505   DenseMap<SUnit*, SmallVector<unsigned, 4> > LRegsMap;
 506   Sequence.reserve(SUnits.size());
 507   while (!AvailableQueue.empty()) {
 508     bool Delayed = false;
 509     LRegsMap.clear();
 510     SUnit *CurSU = AvailableQueue.pop();
 511     while (CurSU) {
 512       SmallVector<unsigned, 4> LRegs;
 513       if (!DelayForLiveRegsBottomUp(CurSU, LRegs))
 514         break;
 515       Delayed = true;
 516       LRegsMap.insert(std::make_pair(CurSU, LRegs));
 517
 518       CurSU->isPending = true;  // This SU is not in AvailableQueue right now.
 519       NotReady.push_back(CurSU);
 520       CurSU = AvailableQueue.pop();
 521     }
 522
 523     // All candidates are delayed due to live physical reg dependencies.
 524     // Try code duplication or inserting cross class copies
 525     // to resolve it.
 526     if (Delayed && !CurSU) {
 527       if (!CurSU) {
 528         // Try duplicating the nodes that produces these
 529         // "expensive to copy" values to break the dependency. In case even
 530         // that doesn't work, insert cross class copies.
 531         SUnit *TrySU = NotReady[0];
 532         SmallVector<unsigned, 4> &LRegs = LRegsMap[TrySU];
 533         assert(LRegs.size() == 1 && "Can't handle this yet!");
 534         unsigned Reg = LRegs[0];
 535         SUnit *LRDef = LiveRegDefs[Reg];
 536         MVT VT = getPhysicalRegisterVT(LRDef->getNode(), Reg, TII);
 537         const TargetRegisterClass *RC =
 538           TRI->getPhysicalRegisterRegClass(Reg, VT);
 539         const TargetRegisterClass *DestRC = TRI->getCrossCopyRegClass(RC);
 540
 541         // If cross copy register class is null, then it must be possible copy
 542         // the value directly. Do not try duplicate the def.
 543         SUnit *NewDef = 0;
 544         if (DestRC)
 545           NewDef = CopyAndMoveSuccessors(LRDef);
 546         else
 547           DestRC = RC;
 548         if (!NewDef) {
 549           // Issue copies, these can be expensive cross register class copies.
 550           SmallVector<SUnit*, 2> Copies;
 551           InsertCopiesAndMoveSuccs(LRDef, Reg, DestRC, RC, Copies);
 552           DOUT << "Adding an edge from SU # " << TrySU->NodeNum
 553                << " to SU #" << Copies.front()->NodeNum << "\n";
 554           AddPred(TrySU, SDep(Copies.front(), SDep::Order, /*Latency=*/1,
 555                               /*Reg=*/0, /*isNormalMemory=*/false,
 556                               /*isMustAlias=*/false, /*isArtificial=*/true));
 557           NewDef = Copies.back();
 558         }
 559
 560         DOUT << "Adding an edge from SU # " << NewDef->NodeNum
 561              << " to SU #" << TrySU->NodeNum << "\n";
 562         LiveRegDefs[Reg] = NewDef;
 563         AddPred(NewDef, SDep(TrySU, SDep::Order, /*Latency=*/1,
 564                              /*Reg=*/0, /*isNormalMemory=*/false,
 565                              /*isMustAlias=*/false, /*isArtificial=*/true));
 566         TrySU->isAvailable = false;
 567         CurSU = NewDef;
 568       }
 569
 570       if (!CurSU) {
 571         assert(false && "Unable to resolve live physical register dependencies!");
 572         abort();
 573       }
 574     }
 575
 576     // Add the nodes that aren't ready back onto the available list.
 577     for (unsigned i = 0, e = NotReady.size(); i != e; ++i) {
 578       NotReady[i]->isPending = false;
 579       // May no longer be available due to backtracking.
 580       if (NotReady[i]->isAvailable)
 581         AvailableQueue.push(NotReady[i]);
 582     }
 583     NotReady.clear();
 584
 585     if (CurSU)
 586       ScheduleNodeBottomUp(CurSU, CurCycle);
 587     ++CurCycle;
 588   }
 589
 590   // Reverse the order if it is bottom up.
 591   std::reverse(Sequence.begin(), Sequence.end());
 592
 593
 594 #ifndef NDEBUG
 595   // Verify that all SUnits were scheduled.
 596   bool AnyNotSched = false;
 597   unsigned DeadNodes = 0;
 598   unsigned Noops = 0;
 599   for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
 600     if (!SUnits[i].isScheduled) {
 601       if (SUnits[i].NumPreds == 0 && SUnits[i].NumSuccs == 0) {
 602         ++DeadNodes;
 603         continue;
 604       }
 605       if (!AnyNotSched)
 606         cerr << "*** List scheduling failed! ***\n";
 607       SUnits[i].dump(this);
 608       cerr << "has not been scheduled!\n";
 609       AnyNotSched = true;
 610     }
 611     if (SUnits[i].NumSuccsLeft != 0) {
 612       if (!AnyNotSched)
 613         cerr << "*** List scheduling failed! ***\n";
 614       SUnits[i].dump(this);
 615       cerr << "has successors left!\n";
 616       AnyNotSched = true;
 617     }
 618   }
 619   for (unsigned i = 0, e = Sequence.size(); i != e; ++i)
 620     if (!Sequence[i])
 621       ++Noops;
 622   assert(!AnyNotSched);
 623   assert(Sequence.size() + DeadNodes - Noops == SUnits.size() &&
 624          "The number of nodes scheduled doesn't match the expected number!");
 625 #endif
 626 }
 627
 628 //===----------------------------------------------------------------------===//
 629 //                         Public Constructor Functions
 630 //===----------------------------------------------------------------------===//
 631
 632 llvm::ScheduleDAGSDNodes *
 633 llvm::createFastDAGScheduler(SelectionDAGISel *IS, CodeGenOpt::Level) {
 634   return new ScheduleDAGFast(*IS->MF);
 635 }