llvm.org GIT mirror llvm / release_27 lib / CodeGen / ScheduleDAGInstrs.cpp
release_27

Tree @release_27 (Download .tar.gz)

ScheduleDAGInstrs.cpp @release_27raw · history · blame

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
//===---- ScheduleDAGInstrs.cpp - MachineInstr Rescheduling ---------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements the ScheduleDAGInstrs class, which implements re-scheduling
// of MachineInstrs.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "sched-instrs"
#include "ScheduleDAGInstrs.h"
#include "llvm/Operator.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetSubtarget.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/ADT/SmallSet.h"
using namespace llvm;

ScheduleDAGInstrs::ScheduleDAGInstrs(MachineFunction &mf,
                                     const MachineLoopInfo &mli,
                                     const MachineDominatorTree &mdt)
  : ScheduleDAG(mf), MLI(mli), MDT(mdt), LoopRegs(MLI, MDT) {
  MFI = mf.getFrameInfo();
}

/// Run - perform scheduling.
///
void ScheduleDAGInstrs::Run(MachineBasicBlock *bb,
                            MachineBasicBlock::iterator begin,
                            MachineBasicBlock::iterator end,
                            unsigned endcount) {
  BB = bb;
  Begin = begin;
  InsertPosIndex = endcount;

  ScheduleDAG::Run(bb, end);
}

/// getUnderlyingObjectFromInt - This is the function that does the work of
/// looking through basic ptrtoint+arithmetic+inttoptr sequences.
static const Value *getUnderlyingObjectFromInt(const Value *V) {
  do {
    if (const Operator *U = dyn_cast<Operator>(V)) {
      // If we find a ptrtoint, we can transfer control back to the
      // regular getUnderlyingObjectFromInt.
      if (U->getOpcode() == Instruction::PtrToInt)
        return U->getOperand(0);
      // If we find an add of a constant or a multiplied value, it's
      // likely that the other operand will lead us to the base
      // object. We don't have to worry about the case where the
      // object address is somehow being computed by the multiply,
      // because our callers only care when the result is an
      // identifibale object.
      if (U->getOpcode() != Instruction::Add ||
          (!isa<ConstantInt>(U->getOperand(1)) &&
           Operator::getOpcode(U->getOperand(1)) != Instruction::Mul))
        return V;
      V = U->getOperand(0);
    } else {
      return V;
    }
    assert(V->getType()->isIntegerTy() && "Unexpected operand type!");
  } while (1);
}

/// getUnderlyingObject - This is a wrapper around Value::getUnderlyingObject
/// and adds support for basic ptrtoint+arithmetic+inttoptr sequences.
static const Value *getUnderlyingObject(const Value *V) {
  // First just call Value::getUnderlyingObject to let it do what it does.
  do {
    V = V->getUnderlyingObject();
    // If it found an inttoptr, use special code to continue climing.
    if (Operator::getOpcode(V) != Instruction::IntToPtr)
      break;
    const Value *O = getUnderlyingObjectFromInt(cast<User>(V)->getOperand(0));
    // If that succeeded in finding a pointer, continue the search.
    if (!O->getType()->isPointerTy())
      break;
    V = O;
  } while (1);
  return V;
}

/// getUnderlyingObjectForInstr - If this machine instr has memory reference
/// information and it can be tracked to a normal reference to a known
/// object, return the Value for that object. Otherwise return null.
static const Value *getUnderlyingObjectForInstr(const MachineInstr *MI,
                                                const MachineFrameInfo *MFI,
                                                bool &MayAlias) {
  MayAlias = true;
  if (!MI->hasOneMemOperand() ||
      !(*MI->memoperands_begin())->getValue() ||
      (*MI->memoperands_begin())->isVolatile())
    return 0;

  const Value *V = (*MI->memoperands_begin())->getValue();
  if (!V)
    return 0;

  V = getUnderlyingObject(V);
  if (const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(V)) {
    // For now, ignore PseudoSourceValues which may alias LLVM IR values
    // because the code that uses this function has no way to cope with
    // such aliases.
    if (PSV->isAliased(MFI))
      return 0;
    
    MayAlias = PSV->mayAlias(MFI);
    return V;
  }

  if (isIdentifiedObject(V))
    return V;

  return 0;
}

void ScheduleDAGInstrs::StartBlock(MachineBasicBlock *BB) {
  if (MachineLoop *ML = MLI.getLoopFor(BB))
    if (BB == ML->getLoopLatch()) {
      MachineBasicBlock *Header = ML->getHeader();
      for (MachineBasicBlock::livein_iterator I = Header->livein_begin(),
           E = Header->livein_end(); I != E; ++I)
        LoopLiveInRegs.insert(*I);
      LoopRegs.VisitLoop(ML);
    }
}

void ScheduleDAGInstrs::BuildSchedGraph(AliasAnalysis *AA) {
  // We'll be allocating one SUnit for each instruction, plus one for
  // the region exit node.
  SUnits.reserve(BB->size());

  // We build scheduling units by walking a block's instruction list from bottom
  // to top.

  // Remember where a generic side-effecting instruction is as we procede.
  SUnit *BarrierChain = 0, *AliasChain = 0;

  // Memory references to specific known memory locations are tracked
  // so that they can be given more precise dependencies. We track
  // separately the known memory locations that may alias and those
  // that are known not to alias
  std::map<const Value *, SUnit *> AliasMemDefs, NonAliasMemDefs;
  std::map<const Value *, std::vector<SUnit *> > AliasMemUses, NonAliasMemUses;

  // Check to see if the scheduler cares about latencies.
  bool UnitLatencies = ForceUnitLatencies();

  // Ask the target if address-backscheduling is desirable, and if so how much.
  const TargetSubtarget &ST = TM.getSubtarget<TargetSubtarget>();
  unsigned SpecialAddressLatency = ST.getSpecialAddressLatency();

  // Walk the list of instructions, from bottom moving up.
  for (MachineBasicBlock::iterator MII = InsertPos, MIE = Begin;
       MII != MIE; --MII) {
    MachineInstr *MI = prior(MII);
    const TargetInstrDesc &TID = MI->getDesc();
    assert(!TID.isTerminator() && !MI->isLabel() &&
           "Cannot schedule terminators or labels!");
    // Create the SUnit for this MI.
    SUnit *SU = NewSUnit(MI);

    // Assign the Latency field of SU using target-provided information.
    if (UnitLatencies)
      SU->Latency = 1;
    else
      ComputeLatency(SU);

    // Add register-based dependencies (data, anti, and output).
    for (unsigned j = 0, n = MI->getNumOperands(); j != n; ++j) {
      const MachineOperand &MO = MI->getOperand(j);
      if (!MO.isReg()) continue;
      unsigned Reg = MO.getReg();
      if (Reg == 0) continue;

      assert(TRI->isPhysicalRegister(Reg) && "Virtual register encountered!");
      std::vector<SUnit *> &UseList = Uses[Reg];
      std::vector<SUnit *> &DefList = Defs[Reg];
      // Optionally add output and anti dependencies. For anti
      // dependencies we use a latency of 0 because for a multi-issue
      // target we want to allow the defining instruction to issue
      // in the same cycle as the using instruction.
      // TODO: Using a latency of 1 here for output dependencies assumes
      //       there's no cost for reusing registers.
      SDep::Kind Kind = MO.isUse() ? SDep::Anti : SDep::Output;
      unsigned AOLatency = (Kind == SDep::Anti) ? 0 : 1;
      for (unsigned i = 0, e = DefList.size(); i != e; ++i) {
        SUnit *DefSU = DefList[i];
        if (DefSU != SU &&
            (Kind != SDep::Output || !MO.isDead() ||
             !DefSU->getInstr()->registerDefIsDead(Reg)))
          DefSU->addPred(SDep(SU, Kind, AOLatency, /*Reg=*/Reg));
      }
      for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
        std::vector<SUnit *> &DefList = Defs[*Alias];
        for (unsigned i = 0, e = DefList.size(); i != e; ++i) {
          SUnit *DefSU = DefList[i];
          if (DefSU != SU &&
              (Kind != SDep::Output || !MO.isDead() ||
               !DefSU->getInstr()->registerDefIsDead(*Alias)))
            DefSU->addPred(SDep(SU, Kind, AOLatency, /*Reg=*/ *Alias));
        }
      }

      if (MO.isDef()) {
        // Add any data dependencies.
        unsigned DataLatency = SU->Latency;
        for (unsigned i = 0, e = UseList.size(); i != e; ++i) {
          SUnit *UseSU = UseList[i];
          if (UseSU != SU) {
            unsigned LDataLatency = DataLatency;
            // Optionally add in a special extra latency for nodes that
            // feed addresses.
            // TODO: Do this for register aliases too.
            // TODO: Perhaps we should get rid of
            // SpecialAddressLatency and just move this into
            // adjustSchedDependency for the targets that care about
            // it.
            if (SpecialAddressLatency != 0 && !UnitLatencies) {
              MachineInstr *UseMI = UseSU->getInstr();
              const TargetInstrDesc &UseTID = UseMI->getDesc();
              int RegUseIndex = UseMI->findRegisterUseOperandIdx(Reg);
              assert(RegUseIndex >= 0 && "UseMI doesn's use register!");
              if ((UseTID.mayLoad() || UseTID.mayStore()) &&
                  (unsigned)RegUseIndex < UseTID.getNumOperands() &&
                  UseTID.OpInfo[RegUseIndex].isLookupPtrRegClass())
                LDataLatency += SpecialAddressLatency;
            }
            // Adjust the dependence latency using operand def/use
            // information (if any), and then allow the target to
            // perform its own adjustments.
            const SDep& dep = SDep(SU, SDep::Data, LDataLatency, Reg);
            if (!UnitLatencies) {
              ComputeOperandLatency(SU, UseSU, (SDep &)dep);
              ST.adjustSchedDependency(SU, UseSU, (SDep &)dep);
            }
            UseSU->addPred(dep);
          }
        }
        for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
          std::vector<SUnit *> &UseList = Uses[*Alias];
          for (unsigned i = 0, e = UseList.size(); i != e; ++i) {
            SUnit *UseSU = UseList[i];
            if (UseSU != SU) {
              const SDep& dep = SDep(SU, SDep::Data, DataLatency, *Alias);
              if (!UnitLatencies) {
                ComputeOperandLatency(SU, UseSU, (SDep &)dep);
                ST.adjustSchedDependency(SU, UseSU, (SDep &)dep);
              }
              UseSU->addPred(dep);
            }
          }
        }

        // If a def is going to wrap back around to the top of the loop,
        // backschedule it.
        if (!UnitLatencies && DefList.empty()) {
          LoopDependencies::LoopDeps::iterator I = LoopRegs.Deps.find(Reg);
          if (I != LoopRegs.Deps.end()) {
            const MachineOperand *UseMO = I->second.first;
            unsigned Count = I->second.second;
            const MachineInstr *UseMI = UseMO->getParent();
            unsigned UseMOIdx = UseMO - &UseMI->getOperand(0);
            const TargetInstrDesc &UseTID = UseMI->getDesc();
            // TODO: If we knew the total depth of the region here, we could
            // handle the case where the whole loop is inside the region but
            // is large enough that the isScheduleHigh trick isn't needed.
            if (UseMOIdx < UseTID.getNumOperands()) {
              // Currently, we only support scheduling regions consisting of
              // single basic blocks. Check to see if the instruction is in
              // the same region by checking to see if it has the same parent.
              if (UseMI->getParent() != MI->getParent()) {
                unsigned Latency = SU->Latency;
                if (UseTID.OpInfo[UseMOIdx].isLookupPtrRegClass())
                  Latency += SpecialAddressLatency;
                // This is a wild guess as to the portion of the latency which
                // will be overlapped by work done outside the current
                // scheduling region.
                Latency -= std::min(Latency, Count);
                // Add the artifical edge.
                ExitSU.addPred(SDep(SU, SDep::Order, Latency,
                                    /*Reg=*/0, /*isNormalMemory=*/false,
                                    /*isMustAlias=*/false,
                                    /*isArtificial=*/true));
              } else if (SpecialAddressLatency > 0 &&
                         UseTID.OpInfo[UseMOIdx].isLookupPtrRegClass()) {
                // The entire loop body is within the current scheduling region
                // and the latency of this operation is assumed to be greater
                // than the latency of the loop.
                // TODO: Recursively mark data-edge predecessors as
                //       isScheduleHigh too.
                SU->isScheduleHigh = true;
              }
            }
            LoopRegs.Deps.erase(I);
          }
        }

        UseList.clear();
        if (!MO.isDead())
          DefList.clear();
        DefList.push_back(SU);
      } else {
        UseList.push_back(SU);
      }
    }

    // Add chain dependencies.
    // Chain dependencies used to enforce memory order should have
    // latency of 0 (except for true dependency of Store followed by
    // aliased Load... we estimate that with a single cycle of latency
    // assuming the hardware will bypass)
    // Note that isStoreToStackSlot and isLoadFromStackSLot are not usable
    // after stack slots are lowered to actual addresses.
    // TODO: Use an AliasAnalysis and do real alias-analysis queries, and
    // produce more precise dependence information.
#define STORE_LOAD_LATENCY 1
    unsigned TrueMemOrderLatency = 0;
    if (TID.isCall() || TID.hasUnmodeledSideEffects() ||
        (MI->hasVolatileMemoryRef() && 
         (!TID.mayLoad() || !MI->isInvariantLoad(AA)))) {
      // Be conservative with these and add dependencies on all memory
      // references, even those that are known to not alias.
      for (std::map<const Value *, SUnit *>::iterator I = 
             NonAliasMemDefs.begin(), E = NonAliasMemDefs.end(); I != E; ++I) {
        I->second->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
      }
      for (std::map<const Value *, std::vector<SUnit *> >::iterator I =
             NonAliasMemUses.begin(), E = NonAliasMemUses.end(); I != E; ++I) {
        for (unsigned i = 0, e = I->second.size(); i != e; ++i)
          I->second[i]->addPred(SDep(SU, SDep::Order, TrueMemOrderLatency));
      }
      NonAliasMemDefs.clear();
      NonAliasMemUses.clear();
      // Add SU to the barrier chain.
      if (BarrierChain)
        BarrierChain->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
      BarrierChain = SU;

      // fall-through
    new_alias_chain:
      // Chain all possibly aliasing memory references though SU.
      if (AliasChain)
        AliasChain->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
      AliasChain = SU;
      for (unsigned k = 0, m = PendingLoads.size(); k != m; ++k)
        PendingLoads[k]->addPred(SDep(SU, SDep::Order, TrueMemOrderLatency));
      for (std::map<const Value *, SUnit *>::iterator I = AliasMemDefs.begin(),
           E = AliasMemDefs.end(); I != E; ++I) {
        I->second->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
      }
      for (std::map<const Value *, std::vector<SUnit *> >::iterator I =
           AliasMemUses.begin(), E = AliasMemUses.end(); I != E; ++I) {
        for (unsigned i = 0, e = I->second.size(); i != e; ++i)
          I->second[i]->addPred(SDep(SU, SDep::Order, TrueMemOrderLatency));
      }
      PendingLoads.clear();
      AliasMemDefs.clear();
      AliasMemUses.clear();
    } else if (TID.mayStore()) {
      bool MayAlias = true;
      TrueMemOrderLatency = STORE_LOAD_LATENCY;
      if (const Value *V = getUnderlyingObjectForInstr(MI, MFI, MayAlias)) {
        // A store to a specific PseudoSourceValue. Add precise dependencies.
        // Record the def in MemDefs, first adding a dep if there is
        // an existing def.
        std::map<const Value *, SUnit *>::iterator I = 
          ((MayAlias) ? AliasMemDefs.find(V) : NonAliasMemDefs.find(V));
        std::map<const Value *, SUnit *>::iterator IE = 
          ((MayAlias) ? AliasMemDefs.end() : NonAliasMemDefs.end());
        if (I != IE) {
          I->second->addPred(SDep(SU, SDep::Order, /*Latency=*/0, /*Reg=*/0,
                                  /*isNormalMemory=*/true));
          I->second = SU;
        } else {
          if (MayAlias)
            AliasMemDefs[V] = SU;
          else
            NonAliasMemDefs[V] = SU;
        }
        // Handle the uses in MemUses, if there are any.
        std::map<const Value *, std::vector<SUnit *> >::iterator J =
          ((MayAlias) ? AliasMemUses.find(V) : NonAliasMemUses.find(V));
        std::map<const Value *, std::vector<SUnit *> >::iterator JE =
          ((MayAlias) ? AliasMemUses.end() : NonAliasMemUses.end());
        if (J != JE) {
          for (unsigned i = 0, e = J->second.size(); i != e; ++i)
            J->second[i]->addPred(SDep(SU, SDep::Order, TrueMemOrderLatency,
                                       /*Reg=*/0, /*isNormalMemory=*/true));
          J->second.clear();
        }
        if (MayAlias) {
          // Add dependencies from all the PendingLoads, i.e. loads
          // with no underlying object.
          for (unsigned k = 0, m = PendingLoads.size(); k != m; ++k)
            PendingLoads[k]->addPred(SDep(SU, SDep::Order, TrueMemOrderLatency));
          // Add dependence on alias chain, if needed.
          if (AliasChain)
            AliasChain->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
        }
        // Add dependence on barrier chain, if needed.
        if (BarrierChain)
          BarrierChain->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
      } else {
        // Treat all other stores conservatively.
        goto new_alias_chain;
      }
    } else if (TID.mayLoad()) {
      bool MayAlias = true;
      TrueMemOrderLatency = 0;
      if (MI->isInvariantLoad(AA)) {
        // Invariant load, no chain dependencies needed!
      } else {
        if (const Value *V = 
            getUnderlyingObjectForInstr(MI, MFI, MayAlias)) {
          // A load from a specific PseudoSourceValue. Add precise dependencies.
          std::map<const Value *, SUnit *>::iterator I = 
            ((MayAlias) ? AliasMemDefs.find(V) : NonAliasMemDefs.find(V));
          std::map<const Value *, SUnit *>::iterator IE = 
            ((MayAlias) ? AliasMemDefs.end() : NonAliasMemDefs.end());
          if (I != IE)
            I->second->addPred(SDep(SU, SDep::Order, /*Latency=*/0, /*Reg=*/0,
                                    /*isNormalMemory=*/true));
          if (MayAlias)
            AliasMemUses[V].push_back(SU);
          else 
            NonAliasMemUses[V].push_back(SU);
        } else {
          // A load with no underlying object. Depend on all
          // potentially aliasing stores.
          for (std::map<const Value *, SUnit *>::iterator I = 
                 AliasMemDefs.begin(), E = AliasMemDefs.end(); I != E; ++I)
            I->second->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
          
          PendingLoads.push_back(SU);
          MayAlias = true;
        }
        
        // Add dependencies on alias and barrier chains, if needed.
        if (MayAlias && AliasChain)
          AliasChain->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
        if (BarrierChain)
          BarrierChain->addPred(SDep(SU, SDep::Order, /*Latency=*/0));
      } 
    }
  }

  for (int i = 0, e = TRI->getNumRegs(); i != e; ++i) {
    Defs[i].clear();
    Uses[i].clear();
  }
  PendingLoads.clear();
}

void ScheduleDAGInstrs::FinishBlock() {
  // Nothing to do.
}

void ScheduleDAGInstrs::ComputeLatency(SUnit *SU) {
  const InstrItineraryData &InstrItins = TM.getInstrItineraryData();

  // Compute the latency for the node.
  SU->Latency =
    InstrItins.getStageLatency(SU->getInstr()->getDesc().getSchedClass());

  // Simplistic target-independent heuristic: assume that loads take
  // extra time.
  if (InstrItins.isEmpty())
    if (SU->getInstr()->getDesc().mayLoad())
      SU->Latency += 2;
}

void ScheduleDAGInstrs::ComputeOperandLatency(SUnit *Def, SUnit *Use, 
                                              SDep& dep) const {
  const InstrItineraryData &InstrItins = TM.getInstrItineraryData();
  if (InstrItins.isEmpty())
    return;
  
  // For a data dependency with a known register...
  if ((dep.getKind() != SDep::Data) || (dep.getReg() == 0))
    return;

  const unsigned Reg = dep.getReg();

  // ... find the definition of the register in the defining
  // instruction
  MachineInstr *DefMI = Def->getInstr();
  int DefIdx = DefMI->findRegisterDefOperandIdx(Reg);
  if (DefIdx != -1) {
    int DefCycle = InstrItins.getOperandCycle(DefMI->getDesc().getSchedClass(), DefIdx);
    if (DefCycle >= 0) {
      MachineInstr *UseMI = Use->getInstr();
      const unsigned UseClass = UseMI->getDesc().getSchedClass();

      // For all uses of the register, calculate the maxmimum latency
      int Latency = -1;
      for (unsigned i = 0, e = UseMI->getNumOperands(); i != e; ++i) {
        const MachineOperand &MO = UseMI->getOperand(i);
        if (!MO.isReg() || !MO.isUse())
          continue;
        unsigned MOReg = MO.getReg();
        if (MOReg != Reg)
          continue;

        int UseCycle = InstrItins.getOperandCycle(UseClass, i);
        if (UseCycle >= 0)
          Latency = std::max(Latency, DefCycle - UseCycle + 1);
      }

      // If we found a latency, then replace the existing dependence latency.
      if (Latency >= 0)
        dep.setLatency(Latency);
    }
  }
}

void ScheduleDAGInstrs::dumpNode(const SUnit *SU) const {
  SU->getInstr()->dump();
}

std::string ScheduleDAGInstrs::getGraphNodeLabel(const SUnit *SU) const {
  std::string s;
  raw_string_ostream oss(s);
  if (SU == &EntrySU)
    oss << "<entry>";
  else if (SU == &ExitSU)
    oss << "<exit>";
  else
    SU->getInstr()->print(oss);
  return oss.str();
}

// EmitSchedule - Emit the machine code in scheduled order.
MachineBasicBlock *ScheduleDAGInstrs::
EmitSchedule(DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) {
  // For MachineInstr-based scheduling, we're rescheduling the instructions in
  // the block, so start by removing them from the block.
  while (Begin != InsertPos) {
    MachineBasicBlock::iterator I = Begin;
    ++Begin;
    BB->remove(I);
  }

  // Then re-insert them according to the given schedule.
  for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
    SUnit *SU = Sequence[i];
    if (!SU) {
      // Null SUnit* is a noop.
      EmitNoop();
      continue;
    }

    BB->insert(InsertPos, SU->getInstr());
  }

  // Update the Begin iterator, as the first instruction in the block
  // may have been scheduled later.
  if (!Sequence.empty())
    Begin = Sequence[0]->getInstr();

  return BB;
}