llvm.org GIT mirror llvm / 9e08ee5 lib / CodeGen / SelectionDAG / ScheduleDAGSDNodes.cpp
9e08ee5

Tree @9e08ee5 (Download .tar.gz)

ScheduleDAGSDNodes.cpp @9e08ee5raw · 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
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
//===--- ScheduleDAGSDNodes.cpp - Implement the ScheduleDAGSDNodes class --===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements the ScheduleDAG class, which is a base class used by
// scheduling implementation classes.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "pre-RA-sched"
#include "SDNodeDbgValue.h"
#include "ScheduleDAGSDNodes.h"
#include "InstrEmitter.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetSubtarget.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;

STATISTIC(LoadsClustered, "Number of loads clustered together");

ScheduleDAGSDNodes::ScheduleDAGSDNodes(MachineFunction &mf)
  : ScheduleDAG(mf),
    InstrItins(mf.getTarget().getInstrItineraryData()) {}

/// Run - perform scheduling.
///
void ScheduleDAGSDNodes::Run(SelectionDAG *dag, MachineBasicBlock *bb,
                             MachineBasicBlock::iterator insertPos) {
  DAG = dag;
  ScheduleDAG::Run(bb, insertPos);
}

/// NewSUnit - Creates a new SUnit and return a ptr to it.
///
SUnit *ScheduleDAGSDNodes::NewSUnit(SDNode *N) {
#ifndef NDEBUG
  const SUnit *Addr = 0;
  if (!SUnits.empty())
    Addr = &SUnits[0];
#endif
  SUnits.push_back(SUnit(N, (unsigned)SUnits.size()));
  assert((Addr == 0 || Addr == &SUnits[0]) &&
         "SUnits std::vector reallocated on the fly!");
  SUnits.back().OrigNode = &SUnits.back();
  SUnit *SU = &SUnits.back();
  const TargetLowering &TLI = DAG->getTargetLoweringInfo();
  if (!N ||
      (N->isMachineOpcode() &&
       N->getMachineOpcode() == TargetOpcode::IMPLICIT_DEF))
    SU->SchedulingPref = Sched::None;
  else
    SU->SchedulingPref = TLI.getSchedulingPreference(N);
  return SU;
}

SUnit *ScheduleDAGSDNodes::Clone(SUnit *Old) {
  SUnit *SU = NewSUnit(Old->getNode());
  SU->OrigNode = Old->OrigNode;
  SU->Latency = Old->Latency;
  SU->isTwoAddress = Old->isTwoAddress;
  SU->isCommutable = Old->isCommutable;
  SU->hasPhysRegDefs = Old->hasPhysRegDefs;
  SU->hasPhysRegClobbers = Old->hasPhysRegClobbers;
  SU->SchedulingPref = Old->SchedulingPref;
  Old->isCloned = true;
  return SU;
}

/// CheckForPhysRegDependency - Check if the dependency between def and use of
/// a specified operand is a physical register dependency. If so, returns the
/// register and the cost of copying the register.
static void CheckForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
                                      const TargetRegisterInfo *TRI, 
                                      const TargetInstrInfo *TII,
                                      unsigned &PhysReg, int &Cost) {
  if (Op != 2 || User->getOpcode() != ISD::CopyToReg)
    return;

  unsigned Reg = cast<RegisterSDNode>(User->getOperand(1))->getReg();
  if (TargetRegisterInfo::isVirtualRegister(Reg))
    return;

  unsigned ResNo = User->getOperand(2).getResNo();
  if (Def->isMachineOpcode()) {
    const TargetInstrDesc &II = TII->get(Def->getMachineOpcode());
    if (ResNo >= II.getNumDefs() &&
        II.ImplicitDefs[ResNo - II.getNumDefs()] == Reg) {
      PhysReg = Reg;
      const TargetRegisterClass *RC =
        TRI->getMinimalPhysRegClass(Reg, Def->getValueType(ResNo));
      Cost = RC->getCopyCost();
    }
  }
}

static void AddFlags(SDNode *N, SDValue Flag, bool AddFlag,
                     SelectionDAG *DAG) {
  SmallVector<EVT, 4> VTs;
  SDNode *FlagDestNode = Flag.getNode();

  // Don't add a flag from a node to itself.
  if (FlagDestNode == N) return;

  // Don't add a flag to something which already has a flag.
  if (N->getValueType(N->getNumValues() - 1) == MVT::Flag) return;

  for (unsigned I = 0, E = N->getNumValues(); I != E; ++I)
    VTs.push_back(N->getValueType(I));

  if (AddFlag)
    VTs.push_back(MVT::Flag);

  SmallVector<SDValue, 4> Ops;
  for (unsigned I = 0, E = N->getNumOperands(); I != E; ++I)
    Ops.push_back(N->getOperand(I));

  if (FlagDestNode)
    Ops.push_back(Flag);

  SDVTList VTList = DAG->getVTList(&VTs[0], VTs.size());
  MachineSDNode::mmo_iterator Begin = 0, End = 0;
  MachineSDNode *MN = dyn_cast<MachineSDNode>(N);

  // Store memory references.
  if (MN) {
    Begin = MN->memoperands_begin();
    End = MN->memoperands_end();
  }

  DAG->MorphNodeTo(N, N->getOpcode(), VTList, &Ops[0], Ops.size());

  // Reset the memory references
  if (MN)
    MN->setMemRefs(Begin, End);
}

/// ClusterNeighboringLoads - Force nearby loads together by "flagging" them.
/// This function finds loads of the same base and different offsets. If the
/// offsets are not far apart (target specific), it add MVT::Flag inputs and
/// outputs to ensure they are scheduled together and in order. This
/// optimization may benefit some targets by improving cache locality.
void ScheduleDAGSDNodes::ClusterNeighboringLoads(SDNode *Node) {
  SDNode *Chain = 0;
  unsigned NumOps = Node->getNumOperands();
  if (Node->getOperand(NumOps-1).getValueType() == MVT::Other)
    Chain = Node->getOperand(NumOps-1).getNode();
  if (!Chain)
    return;

  // Look for other loads of the same chain. Find loads that are loading from
  // the same base pointer and different offsets.
  SmallPtrSet<SDNode*, 16> Visited;
  SmallVector<int64_t, 4> Offsets;
  DenseMap<long long, SDNode*> O2SMap;  // Map from offset to SDNode.
  bool Cluster = false;
  SDNode *Base = Node;
  for (SDNode::use_iterator I = Chain->use_begin(), E = Chain->use_end();
       I != E; ++I) {
    SDNode *User = *I;
    if (User == Node || !Visited.insert(User))
      continue;
    int64_t Offset1, Offset2;
    if (!TII->areLoadsFromSameBasePtr(Base, User, Offset1, Offset2) ||
        Offset1 == Offset2)
      // FIXME: Should be ok if they addresses are identical. But earlier
      // optimizations really should have eliminated one of the loads.
      continue;
    if (O2SMap.insert(std::make_pair(Offset1, Base)).second)
      Offsets.push_back(Offset1);
    O2SMap.insert(std::make_pair(Offset2, User));
    Offsets.push_back(Offset2);
    if (Offset2 < Offset1)
      Base = User;
    Cluster = true;
  }

  if (!Cluster)
    return;

  // Sort them in increasing order.
  std::sort(Offsets.begin(), Offsets.end());

  // Check if the loads are close enough.
  SmallVector<SDNode*, 4> Loads;
  unsigned NumLoads = 0;
  int64_t BaseOff = Offsets[0];
  SDNode *BaseLoad = O2SMap[BaseOff];
  Loads.push_back(BaseLoad);
  for (unsigned i = 1, e = Offsets.size(); i != e; ++i) {
    int64_t Offset = Offsets[i];
    SDNode *Load = O2SMap[Offset];
    if (!TII->shouldScheduleLoadsNear(BaseLoad, Load, BaseOff, Offset,NumLoads))
      break; // Stop right here. Ignore loads that are further away.
    Loads.push_back(Load);
    ++NumLoads;
  }

  if (NumLoads == 0)
    return;

  // Cluster loads by adding MVT::Flag outputs and inputs. This also
  // ensure they are scheduled in order of increasing addresses.
  SDNode *Lead = Loads[0];
  AddFlags(Lead, SDValue(0, 0), true, DAG);

  SDValue InFlag = SDValue(Lead, Lead->getNumValues() - 1);
  for (unsigned I = 1, E = Loads.size(); I != E; ++I) {
    bool OutFlag = I < E - 1;
    SDNode *Load = Loads[I];

    AddFlags(Load, InFlag, OutFlag, DAG);

    if (OutFlag)
      InFlag = SDValue(Load, Load->getNumValues() - 1);

    ++LoadsClustered;
  }
}

/// ClusterNodes - Cluster certain nodes which should be scheduled together.
///
void ScheduleDAGSDNodes::ClusterNodes() {
  for (SelectionDAG::allnodes_iterator NI = DAG->allnodes_begin(),
       E = DAG->allnodes_end(); NI != E; ++NI) {
    SDNode *Node = &*NI;
    if (!Node || !Node->isMachineOpcode())
      continue;

    unsigned Opc = Node->getMachineOpcode();
    const TargetInstrDesc &TID = TII->get(Opc);
    if (TID.mayLoad())
      // Cluster loads from "near" addresses into combined SUnits.
      ClusterNeighboringLoads(Node);
  }
}

void ScheduleDAGSDNodes::BuildSchedUnits() {
  // During scheduling, the NodeId field of SDNode is used to map SDNodes
  // to their associated SUnits by holding SUnits table indices. A value
  // of -1 means the SDNode does not yet have an associated SUnit.
  unsigned NumNodes = 0;
  for (SelectionDAG::allnodes_iterator NI = DAG->allnodes_begin(),
       E = DAG->allnodes_end(); NI != E; ++NI) {
    NI->setNodeId(-1);
    ++NumNodes;
  }

  // Reserve entries in the vector for each of the SUnits we are creating.  This
  // ensure that reallocation of the vector won't happen, so SUnit*'s won't get
  // invalidated.
  // FIXME: Multiply by 2 because we may clone nodes during scheduling.
  // This is a temporary workaround.
  SUnits.reserve(NumNodes * 2);
  
  // Add all nodes in depth first order.
  SmallVector<SDNode*, 64> Worklist;
  SmallPtrSet<SDNode*, 64> Visited;
  Worklist.push_back(DAG->getRoot().getNode());
  Visited.insert(DAG->getRoot().getNode());
  
  while (!Worklist.empty()) {
    SDNode *NI = Worklist.pop_back_val();
    
    // Add all operands to the worklist unless they've already been added.
    for (unsigned i = 0, e = NI->getNumOperands(); i != e; ++i)
      if (Visited.insert(NI->getOperand(i).getNode()))
        Worklist.push_back(NI->getOperand(i).getNode());
  
    if (isPassiveNode(NI))  // Leaf node, e.g. a TargetImmediate.
      continue;
    
    // If this node has already been processed, stop now.
    if (NI->getNodeId() != -1) continue;
    
    SUnit *NodeSUnit = NewSUnit(NI);
    
    // See if anything is flagged to this node, if so, add them to flagged
    // nodes.  Nodes can have at most one flag input and one flag output.  Flags
    // are required to be the last operand and result of a node.
    
    // Scan up to find flagged preds.
    SDNode *N = NI;
    while (N->getNumOperands() &&
           N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Flag) {
      N = N->getOperand(N->getNumOperands()-1).getNode();
      assert(N->getNodeId() == -1 && "Node already inserted!");
      N->setNodeId(NodeSUnit->NodeNum);
    }
    
    // Scan down to find any flagged succs.
    N = NI;
    while (N->getValueType(N->getNumValues()-1) == MVT::Flag) {
      SDValue FlagVal(N, N->getNumValues()-1);
      
      // There are either zero or one users of the Flag result.
      bool HasFlagUse = false;
      for (SDNode::use_iterator UI = N->use_begin(), E = N->use_end(); 
           UI != E; ++UI)
        if (FlagVal.isOperandOf(*UI)) {
          HasFlagUse = true;
          assert(N->getNodeId() == -1 && "Node already inserted!");
          N->setNodeId(NodeSUnit->NodeNum);
          N = *UI;
          break;
        }
      if (!HasFlagUse) break;
    }
    
    // If there are flag operands involved, N is now the bottom-most node
    // of the sequence of nodes that are flagged together.
    // Update the SUnit.
    NodeSUnit->setNode(N);
    assert(N->getNodeId() == -1 && "Node already inserted!");
    N->setNodeId(NodeSUnit->NodeNum);

    // Assign the Latency field of NodeSUnit using target-provided information.
    ComputeLatency(NodeSUnit);
  }
}

void ScheduleDAGSDNodes::AddSchedEdges() {
  const TargetSubtarget &ST = TM.getSubtarget<TargetSubtarget>();

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

  // Pass 2: add the preds, succs, etc.
  for (unsigned su = 0, e = SUnits.size(); su != e; ++su) {
    SUnit *SU = &SUnits[su];
    SDNode *MainNode = SU->getNode();
    
    if (MainNode->isMachineOpcode()) {
      unsigned Opc = MainNode->getMachineOpcode();
      const TargetInstrDesc &TID = TII->get(Opc);
      for (unsigned i = 0; i != TID.getNumOperands(); ++i) {
        if (TID.getOperandConstraint(i, TOI::TIED_TO) != -1) {
          SU->isTwoAddress = true;
          break;
        }
      }
      if (TID.isCommutable())
        SU->isCommutable = true;
    }
    
    // Find all predecessors and successors of the group.
    for (SDNode *N = SU->getNode(); N; N = N->getFlaggedNode()) {
      if (N->isMachineOpcode() &&
          TII->get(N->getMachineOpcode()).getImplicitDefs()) {
        SU->hasPhysRegClobbers = true;
        unsigned NumUsed = InstrEmitter::CountResults(N);
        while (NumUsed != 0 && !N->hasAnyUseOfValue(NumUsed - 1))
          --NumUsed;    // Skip over unused values at the end.
        if (NumUsed > TII->get(N->getMachineOpcode()).getNumDefs())
          SU->hasPhysRegDefs = true;
      }
      
      for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
        SDNode *OpN = N->getOperand(i).getNode();
        if (isPassiveNode(OpN)) continue;   // Not scheduled.
        SUnit *OpSU = &SUnits[OpN->getNodeId()];
        assert(OpSU && "Node has no SUnit!");
        if (OpSU == SU) continue;           // In the same group.

        EVT OpVT = N->getOperand(i).getValueType();
        assert(OpVT != MVT::Flag && "Flagged nodes should be in same sunit!");
        bool isChain = OpVT == MVT::Other;

        unsigned PhysReg = 0;
        int Cost = 1;
        // Determine if this is a physical register dependency.
        CheckForPhysRegDependency(OpN, N, i, TRI, TII, PhysReg, Cost);
        assert((PhysReg == 0 || !isChain) &&
               "Chain dependence via physreg data?");
        // FIXME: See ScheduleDAGSDNodes::EmitCopyFromReg. For now, scheduler
        // emits a copy from the physical register to a virtual register unless
        // it requires a cross class copy (cost < 0). That means we are only
        // treating "expensive to copy" register dependency as physical register
        // dependency. This may change in the future though.
        if (Cost >= 0)
          PhysReg = 0;

        // If this is a ctrl dep, latency is 1.
        unsigned OpLatency = isChain ? 1 : OpSU->Latency;
        const SDep &dep = SDep(OpSU, isChain ? SDep::Order : SDep::Data,
                               OpLatency, PhysReg);
        if (!isChain && !UnitLatencies) {
          ComputeOperandLatency(OpN, N, i, const_cast<SDep &>(dep));
          ST.adjustSchedDependency(OpSU, SU, const_cast<SDep &>(dep));
        }

        SU->addPred(dep);
      }
    }
  }
}

/// BuildSchedGraph - Build the SUnit graph from the selection dag that we
/// are input.  This SUnit graph is similar to the SelectionDAG, but
/// excludes nodes that aren't interesting to scheduling, and represents
/// flagged together nodes with a single SUnit.
void ScheduleDAGSDNodes::BuildSchedGraph(AliasAnalysis *AA) {
  // Cluster certain nodes which should be scheduled together.
  ClusterNodes();
  // Populate the SUnits array.
  BuildSchedUnits();
  // Compute all the scheduling dependencies between nodes.
  AddSchedEdges();
}

void ScheduleDAGSDNodes::ComputeLatency(SUnit *SU) {
  // Check to see if the scheduler cares about latencies.
  if (ForceUnitLatencies()) {
    SU->Latency = 1;
    return;
  }

  if (!InstrItins || InstrItins->isEmpty()) {
    SU->Latency = 1;
    return;
  }
  
  // Compute the latency for the node.  We use the sum of the latencies for
  // all nodes flagged together into this SUnit.
  SU->Latency = 0;
  for (SDNode *N = SU->getNode(); N; N = N->getFlaggedNode())
    if (N->isMachineOpcode()) {
      SU->Latency += InstrItins->
        getStageLatency(TII->get(N->getMachineOpcode()).getSchedClass());
    }
}

void ScheduleDAGSDNodes::ComputeOperandLatency(SDNode *Def, SDNode *Use,
                                               unsigned OpIdx, SDep& dep) const{
  // Check to see if the scheduler cares about latencies.
  if (ForceUnitLatencies())
    return;

  if (dep.getKind() != SDep::Data)
    return;

  unsigned DefIdx = Use->getOperand(OpIdx).getResNo();
  int Latency = TII->getOperandLatency(InstrItins, Def, DefIdx, Use, OpIdx);
  if (Latency >= 0)
    dep.setLatency(Latency);
}

void ScheduleDAGSDNodes::dumpNode(const SUnit *SU) const {
  if (!SU->getNode()) {
    dbgs() << "PHYS REG COPY\n";
    return;
  }

  SU->getNode()->dump(DAG);
  dbgs() << "\n";
  SmallVector<SDNode *, 4> FlaggedNodes;
  for (SDNode *N = SU->getNode()->getFlaggedNode(); N; N = N->getFlaggedNode())
    FlaggedNodes.push_back(N);
  while (!FlaggedNodes.empty()) {
    dbgs() << "    ";
    FlaggedNodes.back()->dump(DAG);
    dbgs() << "\n";
    FlaggedNodes.pop_back();
  }
}

namespace {
  struct OrderSorter {
    bool operator()(const std::pair<unsigned, MachineInstr*> &A,
                    const std::pair<unsigned, MachineInstr*> &B) {
      return A.first < B.first;
    }
  };
}

// ProcessSourceNode - Process nodes with source order numbers. These are added
// to a vector which EmitSchedule uses to determine how to insert dbg_value
// instructions in the right order.
static void ProcessSourceNode(SDNode *N, SelectionDAG *DAG,
                           InstrEmitter &Emitter,
                           DenseMap<SDValue, unsigned> &VRBaseMap,
                    SmallVector<std::pair<unsigned, MachineInstr*>, 32> &Orders,
                           SmallSet<unsigned, 8> &Seen) {
  unsigned Order = DAG->GetOrdering(N);
  if (!Order || !Seen.insert(Order))
    return;

  MachineBasicBlock *BB = Emitter.getBlock();
  if (Emitter.getInsertPos() == BB->begin() || BB->back().isPHI()) {
    // Did not insert any instruction.
    Orders.push_back(std::make_pair(Order, (MachineInstr*)0));
    return;
  }

  Orders.push_back(std::make_pair(Order, prior(Emitter.getInsertPos())));
  if (!N->getHasDebugValue())
    return;
  // Opportunistically insert immediate dbg_value uses, i.e. those with source
  // order number right after the N.
  MachineBasicBlock::iterator InsertPos = Emitter.getInsertPos();
  SmallVector<SDDbgValue*,2> &DVs = DAG->GetDbgValues(N);
  for (unsigned i = 0, e = DVs.size(); i != e; ++i) {
    if (DVs[i]->isInvalidated())
      continue;
    unsigned DVOrder = DVs[i]->getOrder();
    if (DVOrder == ++Order) {
      MachineInstr *DbgMI = Emitter.EmitDbgValue(DVs[i], VRBaseMap);
      if (DbgMI) {
        Orders.push_back(std::make_pair(DVOrder, DbgMI));
        BB->insert(InsertPos, DbgMI);
      }
      DVs[i]->setIsInvalidated();
    }
  }
}


/// EmitSchedule - Emit the machine code in scheduled order.
MachineBasicBlock *ScheduleDAGSDNodes::EmitSchedule() {
  InstrEmitter Emitter(BB, InsertPos);
  DenseMap<SDValue, unsigned> VRBaseMap;
  DenseMap<SUnit*, unsigned> CopyVRBaseMap;
  SmallVector<std::pair<unsigned, MachineInstr*>, 32> Orders;
  SmallSet<unsigned, 8> Seen;
  bool HasDbg = DAG->hasDebugValues();

  // If this is the first BB, emit byval parameter dbg_value's.
  if (HasDbg && BB->getParent()->begin() == MachineFunction::iterator(BB)) {
    SDDbgInfo::DbgIterator PDI = DAG->ByvalParmDbgBegin();
    SDDbgInfo::DbgIterator PDE = DAG->ByvalParmDbgEnd();
    for (; PDI != PDE; ++PDI) {
      MachineInstr *DbgMI= Emitter.EmitDbgValue(*PDI, VRBaseMap);
      if (DbgMI)
        BB->insert(InsertPos, DbgMI);
    }
  }

  for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
    SUnit *SU = Sequence[i];
    if (!SU) {
      // Null SUnit* is a noop.
      EmitNoop();
      continue;
    }

    // For pre-regalloc scheduling, create instructions corresponding to the
    // SDNode and any flagged SDNodes and append them to the block.
    if (!SU->getNode()) {
      // Emit a copy.
      EmitPhysRegCopy(SU, CopyVRBaseMap);
      continue;
    }

    SmallVector<SDNode *, 4> FlaggedNodes;
    for (SDNode *N = SU->getNode()->getFlaggedNode(); N;
         N = N->getFlaggedNode())
      FlaggedNodes.push_back(N);
    while (!FlaggedNodes.empty()) {
      SDNode *N = FlaggedNodes.back();
      Emitter.EmitNode(FlaggedNodes.back(), SU->OrigNode != SU, SU->isCloned,
                       VRBaseMap);
      // Remember the source order of the inserted instruction.
      if (HasDbg)
        ProcessSourceNode(N, DAG, Emitter, VRBaseMap, Orders, Seen);
      FlaggedNodes.pop_back();
    }
    Emitter.EmitNode(SU->getNode(), SU->OrigNode != SU, SU->isCloned,
                     VRBaseMap);
    // Remember the source order of the inserted instruction.
    if (HasDbg)
      ProcessSourceNode(SU->getNode(), DAG, Emitter, VRBaseMap, Orders,
                        Seen);
  }

  // Insert all the dbg_values which have not already been inserted in source
  // order sequence.
  if (HasDbg) {
    MachineBasicBlock::iterator BBBegin = BB->getFirstNonPHI();

    // Sort the source order instructions and use the order to insert debug
    // values.
    std::sort(Orders.begin(), Orders.end(), OrderSorter());

    SDDbgInfo::DbgIterator DI = DAG->DbgBegin();
    SDDbgInfo::DbgIterator DE = DAG->DbgEnd();
    // Now emit the rest according to source order.
    unsigned LastOrder = 0;
    for (unsigned i = 0, e = Orders.size(); i != e && DI != DE; ++i) {
      unsigned Order = Orders[i].first;
      MachineInstr *MI = Orders[i].second;
      // Insert all SDDbgValue's whose order(s) are before "Order".
      if (!MI)
        continue;
#ifndef NDEBUG
      unsigned LastDIOrder = 0;
#endif
      for (; DI != DE &&
             (*DI)->getOrder() >= LastOrder && (*DI)->getOrder() < Order; ++DI) {
#ifndef NDEBUG
        assert((*DI)->getOrder() >= LastDIOrder &&
               "SDDbgValue nodes must be in source order!");
        LastDIOrder = (*DI)->getOrder();
#endif
        if ((*DI)->isInvalidated())
          continue;
        MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap);
        if (DbgMI) {
          if (!LastOrder)
            // Insert to start of the BB (after PHIs).
            BB->insert(BBBegin, DbgMI);
          else {
            // Insert at the instruction, which may be in a different
            // block, if the block was split by a custom inserter.
            MachineBasicBlock::iterator Pos = MI;
            MI->getParent()->insert(llvm::next(Pos), DbgMI);
          }
        }
      }
      LastOrder = Order;
    }
    // Add trailing DbgValue's before the terminator. FIXME: May want to add
    // some of them before one or more conditional branches?
    while (DI != DE) {
      MachineBasicBlock *InsertBB = Emitter.getBlock();
      MachineBasicBlock::iterator Pos= Emitter.getBlock()->getFirstTerminator();
      if (!(*DI)->isInvalidated()) {
        MachineInstr *DbgMI= Emitter.EmitDbgValue(*DI, VRBaseMap);
        if (DbgMI)
          InsertBB->insert(Pos, DbgMI);
      }
      ++DI;
    }
  }

  BB = Emitter.getBlock();
  InsertPos = Emitter.getInsertPos();
  return BB;
}