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/InlineAsm.h"
  17 #include "llvm/CodeGen/SchedulerRegistry.h"
  18 #include "llvm/CodeGen/SelectionDAGISel.h"
  19 #include "llvm/Target/TargetRegisterInfo.h"
  20 #include "llvm/Target/TargetData.h"
  21 #include "llvm/Target/TargetInstrInfo.h"
  22 #include "llvm/Support/Debug.h"
  23 #include "llvm/ADT/SmallSet.h"
  24 #include "llvm/ADT/Statistic.h"
  25 #include "llvm/ADT/STLExtras.h"
  26 #include "llvm/Support/ErrorHandling.h"
  27 #include "llvm/Support/raw_ostream.h"
  28 using namespace llvm;
  29
  30 STATISTIC(NumUnfolds,    "Number of nodes unfolded");
  31 STATISTIC(NumDups,       "Number of duplicated nodes");
  32 STATISTIC(NumPRCopies,   "Number of physical copies");
  33
  34 static RegisterScheduler
  35   fastDAGScheduler("fast", "Fast suboptimal list scheduling",
  36                    createFastDAGScheduler);
  37
  38 namespace {
  39   /// FastPriorityQueue - A degenerate priority queue that considers
  40   /// all nodes to have the same priority.
  41   ///
  42   struct FastPriorityQueue {
  43     SmallVector<SUnit *, 16> Queue;
  44
  45     bool empty() const { return Queue.empty(); }
  46
  47     void push(SUnit *U) {
  48       Queue.push_back(U);
  49     }
  50
  51     SUnit *pop() {
  52       if (empty()) return NULL;
  53       SUnit *V = Queue.back();
  54       Queue.pop_back();
  55       return V;
  56     }
  57   };
  58
  59 //===----------------------------------------------------------------------===//
  60 /// ScheduleDAGFast - The actual "fast" list scheduler implementation.
  61 ///
  62 class ScheduleDAGFast : public ScheduleDAGSDNodes {
  63 private:
  64   /// AvailableQueue - The priority queue to use for the available SUnits.
  65   FastPriorityQueue AvailableQueue;
  66
  67   /// LiveRegDefs - A set of physical registers and their definition
  68   /// that are "live". These nodes must be scheduled before any other nodes that
  69   /// modifies the registers can be scheduled.
  70   unsigned NumLiveRegs;
  71   std::vector<SUnit*> LiveRegDefs;
  72   std::vector<unsigned> LiveRegCycles;
  73
  74 public:
  75   ScheduleDAGFast(MachineFunction &mf)
  76     : ScheduleDAGSDNodes(mf) {}
  77
  78   void Schedule();
  79
  80   /// AddPred - adds a predecessor edge to SUnit SU.
  81   /// This returns true if this is a new predecessor.
  82   void AddPred(SUnit *SU, const SDep &D) {
  83     SU->addPred(D);
  84   }
  85
  86   /// RemovePred - removes a predecessor edge from SUnit SU.
  87   /// This returns true if an edge was removed.
  88   void RemovePred(SUnit *SU, const SDep &D) {
  89     SU->removePred(D);
  90   }
  91
  92 private:
  93   void ReleasePred(SUnit *SU, SDep *PredEdge);
  94   void ReleasePredecessors(SUnit *SU, unsigned CurCycle);
  95   void ScheduleNodeBottomUp(SUnit*, unsigned);
  96   SUnit *CopyAndMoveSuccessors(SUnit*);
  97   void InsertCopiesAndMoveSuccs(SUnit*, unsigned,
  98                                 const TargetRegisterClass*,
  99                                 const TargetRegisterClass*,
 100                                 SmallVector<SUnit*, 2>&);
 101   bool DelayForLiveRegsBottomUp(SUnit*, SmallVector<unsigned, 4>&);
 102   void ListScheduleBottomUp();
 103
 104   /// ForceUnitLatencies - The fast scheduler doesn't care about real latencies.
 105   bool ForceUnitLatencies() const { return true; }
 106 };
 107 }  // end anonymous namespace
 108
 109
 110 /// Schedule - Schedule the DAG using list scheduling.
 111 void ScheduleDAGFast::Schedule() {
 112   DEBUG(dbgs() << "********** List Scheduling **********\n");
 113
 114   NumLiveRegs = 0;
 115   LiveRegDefs.resize(TRI->getNumRegs(), NULL);
 116   LiveRegCycles.resize(TRI->getNumRegs(), 0);
 117
 118   // Build the scheduling graph.
 119   BuildSchedGraph(NULL);
 120
 121   DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
 122           SUnits[su].dumpAll(this));
 123
 124   // Execute the actual scheduling loop.
 125   ListScheduleBottomUp();
 126 }
 127
 128 //===----------------------------------------------------------------------===//
 129 //  Bottom-Up Scheduling
 130 //===----------------------------------------------------------------------===//
 131
 132 /// ReleasePred - Decrement the NumSuccsLeft count of a predecessor. Add it to
 133 /// the AvailableQueue if the count reaches zero. Also update its cycle bound.
 134 void ScheduleDAGFast::ReleasePred(SUnit *SU, SDep *PredEdge) {
 135   SUnit *PredSU = PredEdge->getSUnit();
 136
 137 #ifndef NDEBUG
 138   if (PredSU->NumSuccsLeft == 0) {
 139     dbgs() << "*** Scheduling failed! ***\n";
 140     PredSU->dump(this);
 141     dbgs() << " has been released too many times!\n";
 142     llvm_unreachable(0);
 143   }
 144 #endif
 145   --PredSU->NumSuccsLeft;
 146
 147   // If all the node's successors are scheduled, this node is ready
 148   // to be scheduled. Ignore the special EntrySU node.
 149   if (PredSU->NumSuccsLeft == 0 && PredSU != &EntrySU) {
 150     PredSU->isAvailable = true;
 151     AvailableQueue.push(PredSU);
 152   }
 153 }
 154
 155 void ScheduleDAGFast::ReleasePredecessors(SUnit *SU, unsigned CurCycle) {
 156   // Bottom up: release predecessors
 157   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 158        I != E; ++I) {
 159     ReleasePred(SU, &*I);
 160     if (I->isAssignedRegDep()) {
 161       // This is a physical register dependency and it's impossible or
 162       // expensive to copy the register. Make sure nothing that can
 163       // clobber the register is scheduled between the predecessor and
 164       // this node.
 165       if (!LiveRegDefs[I->getReg()]) {
 166         ++NumLiveRegs;
 167         LiveRegDefs[I->getReg()] = I->getSUnit();
 168         LiveRegCycles[I->getReg()] = CurCycle;
 169       }
 170     }
 171   }
 172 }
 173
 174 /// ScheduleNodeBottomUp - Add the node to the schedule. Decrement the pending
 175 /// count of its predecessors. If a predecessor pending count is zero, add it to
 176 /// the Available queue.
 177 void ScheduleDAGFast::ScheduleNodeBottomUp(SUnit *SU, unsigned CurCycle) {
 178   DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
 179   DEBUG(SU->dump(this));
 180
 181   assert(CurCycle >= SU->getHeight() && "Node scheduled below its height!");
 182   SU->setHeightToAtLeast(CurCycle);
 183   Sequence.push_back(SU);
 184
 185   ReleasePredecessors(SU, CurCycle);
 186
 187   // Release all the implicit physical register defs that are live.
 188   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 189        I != E; ++I) {
 190     if (I->isAssignedRegDep()) {
 191       if (LiveRegCycles[I->getReg()] == I->getSUnit()->getHeight()) {
 192         assert(NumLiveRegs > 0 && "NumLiveRegs is already zero!");
 193         assert(LiveRegDefs[I->getReg()] == SU &&
 194                "Physical register dependency violated?");
 195         --NumLiveRegs;
 196         LiveRegDefs[I->getReg()] = NULL;
 197         LiveRegCycles[I->getReg()] = 0;
 198       }
 199     }
 200   }
 201
 202   SU->isScheduled = true;
 203 }
 204
 205 /// CopyAndMoveSuccessors - Clone the specified node and move its scheduled
 206 /// successors to the newly created node.
 207 SUnit *ScheduleDAGFast::CopyAndMoveSuccessors(SUnit *SU) {
 208   if (SU->getNode()->getGluedNode())
 209     return NULL;
 210
 211   SDNode *N = SU->getNode();
 212   if (!N)
 213     return NULL;
 214
 215   SUnit *NewSU;
 216   bool TryUnfold = false;
 217   for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
 218     EVT VT = N->getValueType(i);
 219     if (VT == MVT::Glue)
 220       return NULL;
 221     else if (VT == MVT::Other)
 222       TryUnfold = true;
 223   }
 224   for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
 225     const SDValue &Op = N->getOperand(i);
 226     EVT VT = Op.getNode()->getValueType(Op.getResNo());
 227     if (VT == MVT::Glue)
 228       return NULL;
 229   }
 230
 231   if (TryUnfold) {
 232     SmallVector<SDNode*, 2> NewNodes;
 233     if (!TII->unfoldMemoryOperand(*DAG, N, NewNodes))
 234       return NULL;
 235
 236     DEBUG(dbgs() << "Unfolding SU # " << SU->NodeNum << "\n");
 237     assert(NewNodes.size() == 2 && "Expected a load folding node!");
 238
 239     N = NewNodes[1];
 240     SDNode *LoadNode = NewNodes[0];
 241     unsigned NumVals = N->getNumValues();
 242     unsigned OldNumVals = SU->getNode()->getNumValues();
 243     for (unsigned i = 0; i != NumVals; ++i)
 244       DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), i), SDValue(N, i));
 245     DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), OldNumVals-1),
 246                                    SDValue(LoadNode, 1));
 247
 248     SUnit *NewSU = NewSUnit(N);
 249     assert(N->getNodeId() == -1 && "Node already inserted!");
 250     N->setNodeId(NewSU->NodeNum);
 251
 252     const TargetInstrDesc &TID = TII->get(N->getMachineOpcode());
 253     for (unsigned i = 0; i != TID.getNumOperands(); ++i) {
 254       if (TID.getOperandConstraint(i, TOI::TIED_TO) != -1) {
 255         NewSU->isTwoAddress = true;
 256         break;
 257       }
 258     }
 259     if (TID.isCommutable())
 260       NewSU->isCommutable = true;
 261
 262     // LoadNode may already exist. This can happen when there is another
 263     // load from the same location and producing the same type of value
 264     // but it has different alignment or volatileness.
 265     bool isNewLoad = true;
 266     SUnit *LoadSU;
 267     if (LoadNode->getNodeId() != -1) {
 268       LoadSU = &SUnits[LoadNode->getNodeId()];
 269       isNewLoad = false;
 270     } else {
 271       LoadSU = NewSUnit(LoadNode);
 272       LoadNode->setNodeId(LoadSU->NodeNum);
 273     }
 274
 275     SDep ChainPred;
 276     SmallVector<SDep, 4> ChainSuccs;
 277     SmallVector<SDep, 4> LoadPreds;
 278     SmallVector<SDep, 4> NodePreds;
 279     SmallVector<SDep, 4> NodeSuccs;
 280     for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 281          I != E; ++I) {
 282       if (I->isCtrl())
 283         ChainPred = *I;
 284       else if (I->getSUnit()->getNode() &&
 285                I->getSUnit()->getNode()->isOperandOf(LoadNode))
 286         LoadPreds.push_back(*I);
 287       else
 288         NodePreds.push_back(*I);
 289     }
 290     for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 291          I != E; ++I) {
 292       if (I->isCtrl())
 293         ChainSuccs.push_back(*I);
 294       else
 295         NodeSuccs.push_back(*I);
 296     }
 297
 298     if (ChainPred.getSUnit()) {
 299       RemovePred(SU, ChainPred);
 300       if (isNewLoad)
 301         AddPred(LoadSU, ChainPred);
 302     }
 303     for (unsigned i = 0, e = LoadPreds.size(); i != e; ++i) {
 304       const SDep &Pred = LoadPreds[i];
 305       RemovePred(SU, Pred);
 306       if (isNewLoad) {
 307         AddPred(LoadSU, Pred);
 308       }
 309     }
 310     for (unsigned i = 0, e = NodePreds.size(); i != e; ++i) {
 311       const SDep &Pred = NodePreds[i];
 312       RemovePred(SU, Pred);
 313       AddPred(NewSU, Pred);
 314     }
 315     for (unsigned i = 0, e = NodeSuccs.size(); i != e; ++i) {
 316       SDep D = NodeSuccs[i];
 317       SUnit *SuccDep = D.getSUnit();
 318       D.setSUnit(SU);
 319       RemovePred(SuccDep, D);
 320       D.setSUnit(NewSU);
 321       AddPred(SuccDep, D);
 322     }
 323     for (unsigned i = 0, e = ChainSuccs.size(); i != e; ++i) {
 324       SDep D = ChainSuccs[i];
 325       SUnit *SuccDep = D.getSUnit();
 326       D.setSUnit(SU);
 327       RemovePred(SuccDep, D);
 328       if (isNewLoad) {
 329         D.setSUnit(LoadSU);
 330         AddPred(SuccDep, D);
 331       }
 332     }
 333     if (isNewLoad) {
 334       AddPred(NewSU, SDep(LoadSU, SDep::Order, LoadSU->Latency));
 335     }
 336
 337     ++NumUnfolds;
 338
 339     if (NewSU->NumSuccsLeft == 0) {
 340       NewSU->isAvailable = true;
 341       return NewSU;
 342     }
 343     SU = NewSU;
 344   }
 345
 346   DEBUG(dbgs() << "Duplicating SU # " << SU->NodeNum << "\n");
 347   NewSU = Clone(SU);
 348
 349   // New SUnit has the exact same predecessors.
 350   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 351        I != E; ++I)
 352     if (!I->isArtificial())
 353       AddPred(NewSU, *I);
 354
 355   // Only copy scheduled successors. Cut them from old node's successor
 356   // list and move them over.
 357   SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
 358   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 359        I != E; ++I) {
 360     if (I->isArtificial())
 361       continue;
 362     SUnit *SuccSU = I->getSUnit();
 363     if (SuccSU->isScheduled) {
 364       SDep D = *I;
 365       D.setSUnit(NewSU);
 366       AddPred(SuccSU, D);
 367       D.setSUnit(SU);
 368       DelDeps.push_back(std::make_pair(SuccSU, D));
 369     }
 370   }
 371   for (unsigned i = 0, e = DelDeps.size(); i != e; ++i)
 372     RemovePred(DelDeps[i].first, DelDeps[i].second);
 373
 374   ++NumDups;
 375   return NewSU;
 376 }
 377
 378 /// InsertCopiesAndMoveSuccs - Insert register copies and move all
 379 /// scheduled successors of the given SUnit to the last copy.
 380 void ScheduleDAGFast::InsertCopiesAndMoveSuccs(SUnit *SU, unsigned Reg,
 381                                               const TargetRegisterClass *DestRC,
 382                                               const TargetRegisterClass *SrcRC,
 383                                                SmallVector<SUnit*, 2> &Copies) {
 384   SUnit *CopyFromSU = NewSUnit(static_cast<SDNode *>(NULL));
 385   CopyFromSU->CopySrcRC = SrcRC;
 386   CopyFromSU->CopyDstRC = DestRC;
 387
 388   SUnit *CopyToSU = NewSUnit(static_cast<SDNode *>(NULL));
 389   CopyToSU->CopySrcRC = DestRC;
 390   CopyToSU->CopyDstRC = SrcRC;
 391
 392   // Only copy scheduled successors. Cut them from old node's successor
 393   // list and move them over.
 394   SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
 395   for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
 396        I != E; ++I) {
 397     if (I->isArtificial())
 398       continue;
 399     SUnit *SuccSU = I->getSUnit();
 400     if (SuccSU->isScheduled) {
 401       SDep D = *I;
 402       D.setSUnit(CopyToSU);
 403       AddPred(SuccSU, D);
 404       DelDeps.push_back(std::make_pair(SuccSU, *I));
 405     }
 406   }
 407   for (unsigned i = 0, e = DelDeps.size(); i != e; ++i) {
 408     RemovePred(DelDeps[i].first, DelDeps[i].second);
 409   }
 410
 411   AddPred(CopyFromSU, SDep(SU, SDep::Data, SU->Latency, Reg));
 412   AddPred(CopyToSU, SDep(CopyFromSU, SDep::Data, CopyFromSU->Latency, 0));
 413
 414   Copies.push_back(CopyFromSU);
 415   Copies.push_back(CopyToSU);
 416
 417   ++NumPRCopies;
 418 }
 419
 420 /// getPhysicalRegisterVT - Returns the ValueType of the physical register
 421 /// definition of the specified node.
 422 /// FIXME: Move to SelectionDAG?
 423 static EVT getPhysicalRegisterVT(SDNode *N, unsigned Reg,
 424                                  const TargetInstrInfo *TII) {
 425   const TargetInstrDesc &TID = TII->get(N->getMachineOpcode());
 426   assert(TID.ImplicitDefs && "Physical reg def must be in implicit def list!");
 427   unsigned NumRes = TID.getNumDefs();
 428   for (const unsigned *ImpDef = TID.getImplicitDefs(); *ImpDef; ++ImpDef) {
 429     if (Reg == *ImpDef)
 430       break;
 431     ++NumRes;
 432   }
 433   return N->getValueType(NumRes);
 434 }
 435
 436 /// CheckForLiveRegDef - Return true and update live register vector if the
 437 /// specified register def of the specified SUnit clobbers any "live" registers.
 438 static bool CheckForLiveRegDef(SUnit *SU, unsigned Reg,
 439                                std::vector<SUnit*> &LiveRegDefs,
 440                                SmallSet<unsigned, 4> &RegAdded,
 441                                SmallVector<unsigned, 4> &LRegs,
 442                                const TargetRegisterInfo *TRI) {
 443   bool Added = false;
 444   if (LiveRegDefs[Reg] && LiveRegDefs[Reg] != SU) {
 445     if (RegAdded.insert(Reg)) {
 446       LRegs.push_back(Reg);
 447       Added = true;
 448     }
 449   }
 450   for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias)
 451     if (LiveRegDefs[*Alias] && LiveRegDefs[*Alias] != SU) {
 452       if (RegAdded.insert(*Alias)) {
 453         LRegs.push_back(*Alias);
 454         Added = true;
 455       }
 456     }
 457   return Added;
 458 }
 459
 460 /// DelayForLiveRegsBottomUp - Returns true if it is necessary to delay
 461 /// scheduling of the given node to satisfy live physical register dependencies.
 462 /// If the specific node is the last one that's available to schedule, do
 463 /// whatever is necessary (i.e. backtracking or cloning) to make it possible.
 464 bool ScheduleDAGFast::DelayForLiveRegsBottomUp(SUnit *SU,
 465                                                SmallVector<unsigned, 4> &LRegs){
 466   if (NumLiveRegs == 0)
 467     return false;
 468
 469   SmallSet<unsigned, 4> RegAdded;
 470   // If this node would clobber any "live" register, then it's not ready.
 471   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 472        I != E; ++I) {
 473     if (I->isAssignedRegDep()) {
 474       CheckForLiveRegDef(I->getSUnit(), I->getReg(), LiveRegDefs,
 475                          RegAdded, LRegs, TRI);
 476     }
 477   }
 478
 479   for (SDNode *Node = SU->getNode(); Node; Node = Node->getGluedNode()) {
 480     if (Node->getOpcode() == ISD::INLINEASM) {
 481       // Inline asm can clobber physical defs.
 482       unsigned NumOps = Node->getNumOperands();
 483       if (Node->getOperand(NumOps-1).getValueType() == MVT::Glue)
 484         --NumOps;  // Ignore the glue operand.
 485
 486       for (unsigned i = InlineAsm::Op_FirstOperand; i != NumOps;) {
 487         unsigned Flags =
 488           cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
 489         unsigned NumVals = InlineAsm::getNumOperandRegisters(Flags);
 490
 491         ++i; // Skip the ID value.
 492         if (InlineAsm::isRegDefKind(Flags) ||
 493             InlineAsm::isRegDefEarlyClobberKind(Flags)) {
 494           // Check for def of register or earlyclobber register.
 495           for (; NumVals; --NumVals, ++i) {
 496             unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
 497             if (TargetRegisterInfo::isPhysicalRegister(Reg))
 498               CheckForLiveRegDef(SU, Reg, LiveRegDefs, RegAdded, LRegs, TRI);
 499           }
 500         } else
 501           i += NumVals;
 502       }
 503       continue;
 504     }
 505     if (!Node->isMachineOpcode())
 506       continue;
 507     const TargetInstrDesc &TID = TII->get(Node->getMachineOpcode());
 508     if (!TID.ImplicitDefs)
 509       continue;
 510     for (const unsigned *Reg = TID.ImplicitDefs; *Reg; ++Reg) {
 511       CheckForLiveRegDef(SU, *Reg, LiveRegDefs, RegAdded, LRegs, TRI);
 512     }
 513   }
 514   return !LRegs.empty();
 515 }
 516
 517
 518 /// ListScheduleBottomUp - The main loop of list scheduling for bottom-up
 519 /// schedulers.
 520 void ScheduleDAGFast::ListScheduleBottomUp() {
 521   unsigned CurCycle = 0;
 522
 523   // Release any predecessors of the special Exit node.
 524   ReleasePredecessors(&ExitSU, CurCycle);
 525
 526   // Add root to Available queue.
 527   if (!SUnits.empty()) {
 528     SUnit *RootSU = &SUnits[DAG->getRoot().getNode()->getNodeId()];
 529     assert(RootSU->Succs.empty() && "Graph root shouldn't have successors!");
 530     RootSU->isAvailable = true;
 531     AvailableQueue.push(RootSU);
 532   }
 533
 534   // While Available queue is not empty, grab the node with the highest
 535   // priority. If it is not ready put it back.  Schedule the node.
 536   SmallVector<SUnit*, 4> NotReady;
 537   DenseMap<SUnit*, SmallVector<unsigned, 4> > LRegsMap;
 538   Sequence.reserve(SUnits.size());
 539   while (!AvailableQueue.empty()) {
 540     bool Delayed = false;
 541     LRegsMap.clear();
 542     SUnit *CurSU = AvailableQueue.pop();
 543     while (CurSU) {
 544       SmallVector<unsigned, 4> LRegs;
 545       if (!DelayForLiveRegsBottomUp(CurSU, LRegs))
 546         break;
 547       Delayed = true;
 548       LRegsMap.insert(std::make_pair(CurSU, LRegs));
 549
 550       CurSU->isPending = true;  // This SU is not in AvailableQueue right now.
 551       NotReady.push_back(CurSU);
 552       CurSU = AvailableQueue.pop();
 553     }
 554
 555     // All candidates are delayed due to live physical reg dependencies.
 556     // Try code duplication or inserting cross class copies
 557     // to resolve it.
 558     if (Delayed && !CurSU) {
 559       if (!CurSU) {
 560         // Try duplicating the nodes that produces these
 561         // "expensive to copy" values to break the dependency. In case even
 562         // that doesn't work, insert cross class copies.
 563         SUnit *TrySU = NotReady[0];
 564         SmallVector<unsigned, 4> &LRegs = LRegsMap[TrySU];
 565         assert(LRegs.size() == 1 && "Can't handle this yet!");
 566         unsigned Reg = LRegs[0];
 567         SUnit *LRDef = LiveRegDefs[Reg];
 568         EVT VT = getPhysicalRegisterVT(LRDef->getNode(), Reg, TII);
 569         const TargetRegisterClass *RC =
 570           TRI->getMinimalPhysRegClass(Reg, VT);
 571         const TargetRegisterClass *DestRC = TRI->getCrossCopyRegClass(RC);
 572
 573         // If cross copy register class is null, then it must be possible copy
 574         // the value directly. Do not try duplicate the def.
 575         SUnit *NewDef = 0;
 576         if (DestRC)
 577           NewDef = CopyAndMoveSuccessors(LRDef);
 578         else
 579           DestRC = RC;
 580         if (!NewDef) {
 581           // Issue copies, these can be expensive cross register class copies.
 582           SmallVector<SUnit*, 2> Copies;
 583           InsertCopiesAndMoveSuccs(LRDef, Reg, DestRC, RC, Copies);
 584           DEBUG(dbgs() << "Adding an edge from SU # " << TrySU->NodeNum
 585                        << " to SU #" << Copies.front()->NodeNum << "\n");
 586           AddPred(TrySU, SDep(Copies.front(), SDep::Order, /*Latency=*/1,
 587                               /*Reg=*/0, /*isNormalMemory=*/false,
 588                               /*isMustAlias=*/false, /*isArtificial=*/true));
 589           NewDef = Copies.back();
 590         }
 591
 592         DEBUG(dbgs() << "Adding an edge from SU # " << NewDef->NodeNum
 593                      << " to SU #" << TrySU->NodeNum << "\n");
 594         LiveRegDefs[Reg] = NewDef;
 595         AddPred(NewDef, SDep(TrySU, SDep::Order, /*Latency=*/1,
 596                              /*Reg=*/0, /*isNormalMemory=*/false,
 597                              /*isMustAlias=*/false, /*isArtificial=*/true));
 598         TrySU->isAvailable = false;
 599         CurSU = NewDef;
 600       }
 601
 602       if (!CurSU) {
 603         llvm_unreachable("Unable to resolve live physical register dependencies!");
 604       }
 605     }
 606
 607     // Add the nodes that aren't ready back onto the available list.
 608     for (unsigned i = 0, e = NotReady.size(); i != e; ++i) {
 609       NotReady[i]->isPending = false;
 610       // May no longer be available due to backtracking.
 611       if (NotReady[i]->isAvailable)
 612         AvailableQueue.push(NotReady[i]);
 613     }
 614     NotReady.clear();
 615
 616     if (CurSU)
 617       ScheduleNodeBottomUp(CurSU, CurCycle);
 618     ++CurCycle;
 619   }
 620
 621   // Reverse the order since it is bottom up.
 622   std::reverse(Sequence.begin(), Sequence.end());
 623
 624 #ifndef NDEBUG
 625   VerifySchedule(/*isBottomUp=*/true);
 626 #endif
 627 }
 628
 629 //===----------------------------------------------------------------------===//
 630 //                         Public Constructor Functions
 631 //===----------------------------------------------------------------------===//
 632
 633 llvm::ScheduleDAGSDNodes *
 634 llvm::createFastDAGScheduler(SelectionDAGISel *IS, CodeGenOpt::Level) {
 635   return new ScheduleDAGFast(*IS->MF);
 636 }