aboutsummaryrefslogtreecommitdiffstats
path: root/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
blob: 1ad7919962b3048a04747ad08d5d4e15046b4913 (plain)
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
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
//===----- ScheduleDAGRRList.cpp - Reg pressure reduction list scheduler --===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements bottom-up and top-down register pressure reduction list
// schedulers, using standard algorithms.  The basic approach uses a priority
// queue of available nodes to schedule.  One at a time, nodes are taken from
// the priority queue (thus in priority order), checked for legality to
// schedule, and emitted if legal.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "pre-RA-sched"
#include "ScheduleDAGSDNodes.h"
#include "llvm/CodeGen/SchedulerRegistry.h"
#include "llvm/CodeGen/SelectionDAGISel.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/ADT/PriorityQueue.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <climits>
using namespace llvm;

STATISTIC(NumBacktracks, "Number of times scheduler backtracked");
STATISTIC(NumUnfolds,    "Number of nodes unfolded");
STATISTIC(NumDups,       "Number of duplicated nodes");
STATISTIC(NumPRCopies,   "Number of physical register copies");

static RegisterScheduler
  burrListDAGScheduler("list-burr",
                       "Bottom-up register reduction list scheduling",
                       createBURRListDAGScheduler);
static RegisterScheduler
  tdrListrDAGScheduler("list-tdrr",
                       "Top-down register reduction list scheduling",
                       createTDRRListDAGScheduler);

namespace {
//===----------------------------------------------------------------------===//
/// ScheduleDAGRRList - The actual register reduction list scheduler
/// implementation.  This supports both top-down and bottom-up scheduling.
///
class ScheduleDAGRRList : public ScheduleDAGSDNodes {
private:
  /// isBottomUp - This is true if the scheduling problem is bottom-up, false if
  /// it is top-down.
  bool isBottomUp;

  /// AvailableQueue - The priority queue to use for the available SUnits.
  SchedulingPriorityQueue *AvailableQueue;

  /// LiveRegDefs - A set of physical registers and their definition
  /// that are "live". These nodes must be scheduled before any other nodes that
  /// modifies the registers can be scheduled.
  unsigned NumLiveRegs;
  std::vector<SUnit*> LiveRegDefs;
  std::vector<unsigned> LiveRegCycles;

  /// Topo - A topological ordering for SUnits which permits fast IsReachable
  /// and similar queries.
  ScheduleDAGTopologicalSort Topo;

public:
  ScheduleDAGRRList(MachineFunction &mf,
                    bool isbottomup,
                    SchedulingPriorityQueue *availqueue)
    : ScheduleDAGSDNodes(mf), isBottomUp(isbottomup),
      AvailableQueue(availqueue), Topo(SUnits) {
    }

  ~ScheduleDAGRRList() {
    delete AvailableQueue;
  }

  void Schedule();

  /// IsReachable - Checks if SU is reachable from TargetSU.
  bool IsReachable(const SUnit *SU, const SUnit *TargetSU) {
    return Topo.IsReachable(SU, TargetSU);
  }

  /// WillCreateCycle - Returns true if adding an edge from SU to TargetSU will
  /// create a cycle.
  bool WillCreateCycle(SUnit *SU, SUnit *TargetSU) {
    return Topo.WillCreateCycle(SU, TargetSU);
  }

  /// AddPred - adds a predecessor edge to SUnit SU.
  /// This returns true if this is a new predecessor.
  /// Updates the topological ordering if required.
  void AddPred(SUnit *SU, const SDep &D) {
    Topo.AddPred(SU, D.getSUnit());
    SU->addPred(D);
  }

  /// RemovePred - removes a predecessor edge from SUnit SU.
  /// This returns true if an edge was removed.
  /// Updates the topological ordering if required.
  void RemovePred(SUnit *SU, const SDep &D) {
    Topo.RemovePred(SU, D.getSUnit());
    SU->removePred(D);
  }

private:
  void ReleasePred(SUnit *SU, const SDep *PredEdge);
  void ReleasePredecessors(SUnit *SU, unsigned CurCycle);
  void ReleaseSucc(SUnit *SU, const SDep *SuccEdge);
  void ReleaseSuccessors(SUnit *SU);
  void CapturePred(SDep *PredEdge);
  void ScheduleNodeBottomUp(SUnit*, unsigned);
  void ScheduleNodeTopDown(SUnit*, unsigned);
  void UnscheduleNodeBottomUp(SUnit*);
  void BacktrackBottomUp(SUnit*, unsigned, unsigned&);
  SUnit *CopyAndMoveSuccessors(SUnit*);
  void InsertCopiesAndMoveSuccs(SUnit*, unsigned,
                                const TargetRegisterClass*,
                                const TargetRegisterClass*,
                                SmallVector<SUnit*, 2>&);
  bool DelayForLiveRegsBottomUp(SUnit*, SmallVector<unsigned, 4>&);
  void ListScheduleTopDown();
  void ListScheduleBottomUp();


  /// CreateNewSUnit - Creates a new SUnit and returns a pointer to it.
  /// Updates the topological ordering if required.
  SUnit *CreateNewSUnit(SDNode *N) {
    unsigned NumSUnits = SUnits.size();
    SUnit *NewNode = NewSUnit(N);
    // Update the topological ordering.
    if (NewNode->NodeNum >= NumSUnits)
      Topo.InitDAGTopologicalSorting();
    return NewNode;
  }

  /// CreateClone - Creates a new SUnit from an existing one.
  /// Updates the topological ordering if required.
  SUnit *CreateClone(SUnit *N) {
    unsigned NumSUnits = SUnits.size();
    SUnit *NewNode = Clone(N);
    // Update the topological ordering.
    if (NewNode->NodeNum >= NumSUnits)
      Topo.InitDAGTopologicalSorting();
    return NewNode;
  }

  /// ForceUnitLatencies - Return true, since register-pressure-reducing
  /// scheduling doesn't need actual latency information.
  bool ForceUnitLatencies() const { return true; }
};
}  // end anonymous namespace


/// Schedule - Schedule the DAG using list scheduling.
void ScheduleDAGRRList::Schedule() {
  DEBUG(dbgs() << "********** List Scheduling **********\n");

  NumLiveRegs = 0;
  LiveRegDefs.resize(TRI->getNumRegs(), NULL);  
  LiveRegCycles.resize(TRI->getNumRegs(), 0);

  // Build the scheduling graph.
  BuildSchedGraph(NULL);

  DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
          SUnits[su].dumpAll(this));
  Topo.InitDAGTopologicalSorting();

  AvailableQueue->initNodes(SUnits);
  
  // Execute the actual scheduling loop Top-Down or Bottom-Up as appropriate.
  if (isBottomUp)
    ListScheduleBottomUp();
  else
    ListScheduleTopDown();
  
  AvailableQueue->releaseState();
}

//===----------------------------------------------------------------------===//
//  Bottom-Up Scheduling
//===----------------------------------------------------------------------===//

/// ReleasePred - Decrement the NumSuccsLeft count of a predecessor. Add it to
/// the AvailableQueue if the count reaches zero. Also update its cycle bound.
void ScheduleDAGRRList::ReleasePred(SUnit *SU, const SDep *PredEdge) {
  SUnit *PredSU = PredEdge->getSUnit();

#ifndef NDEBUG
  if (PredSU->NumSuccsLeft == 0) {
    dbgs() << "*** Scheduling failed! ***\n";
    PredSU->dump(this);
    dbgs() << " has been released too many times!\n";
    llvm_unreachable(0);
  }
#endif
  --PredSU->NumSuccsLeft;

  // If all the node's successors are scheduled, this node is ready
  // to be scheduled. Ignore the special EntrySU node.
  if (PredSU->NumSuccsLeft == 0 && PredSU != &EntrySU) {
    PredSU->isAvailable = true;
    AvailableQueue->push(PredSU);
  }
}

void ScheduleDAGRRList::ReleasePredecessors(SUnit *SU, unsigned CurCycle) {
  // Bottom up: release predecessors
  for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
       I != E; ++I) {
    ReleasePred(SU, &*I);
    if (I->isAssignedRegDep()) {
      // This is a physical register dependency and it's impossible or
      // expensive to copy the register. Make sure nothing that can 
      // clobber the register is scheduled between the predecessor and
      // this node.
      if (!LiveRegDefs[I->getReg()]) {
        ++NumLiveRegs;
        LiveRegDefs[I->getReg()] = I->getSUnit();
        LiveRegCycles[I->getReg()] = CurCycle;
      }
    }
  }
}

/// ScheduleNodeBottomUp - Add the node to the schedule. Decrement the pending
/// count of its predecessors. If a predecessor pending count is zero, add it to
/// the Available queue.
void ScheduleDAGRRList::ScheduleNodeBottomUp(SUnit *SU, unsigned CurCycle) {
  DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
  DEBUG(SU->dump(this));

  assert(CurCycle >= SU->getHeight() && "Node scheduled below its height!");
  SU->setHeightToAtLeast(CurCycle);
  Sequence.push_back(SU);

  ReleasePredecessors(SU, CurCycle);

  // Release all the implicit physical register defs that are live.
  for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    if (I->isAssignedRegDep()) {
      if (LiveRegCycles[I->getReg()] == I->getSUnit()->getHeight()) {
        assert(NumLiveRegs > 0 && "NumLiveRegs is already zero!");
        assert(LiveRegDefs[I->getReg()] == SU &&
               "Physical register dependency violated?");
        --NumLiveRegs;
        LiveRegDefs[I->getReg()] = NULL;
        LiveRegCycles[I->getReg()] = 0;
      }
    }
  }

  SU->isScheduled = true;
  AvailableQueue->ScheduledNode(SU);
}

/// CapturePred - This does the opposite of ReleasePred. Since SU is being
/// unscheduled, incrcease the succ left count of its predecessors. Remove
/// them from AvailableQueue if necessary.
void ScheduleDAGRRList::CapturePred(SDep *PredEdge) {  
  SUnit *PredSU = PredEdge->getSUnit();
  if (PredSU->isAvailable) {
    PredSU->isAvailable = false;
    if (!PredSU->isPending)
      AvailableQueue->remove(PredSU);
  }

  assert(PredSU->NumSuccsLeft < UINT_MAX && "NumSuccsLeft will overflow!");
  ++PredSU->NumSuccsLeft;
}

/// UnscheduleNodeBottomUp - Remove the node from the schedule, update its and
/// its predecessor states to reflect the change.
void ScheduleDAGRRList::UnscheduleNodeBottomUp(SUnit *SU) {
  DEBUG(dbgs() << "*** Unscheduling [" << SU->getHeight() << "]: ");
  DEBUG(SU->dump(this));

  AvailableQueue->UnscheduledNode(SU);

  for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
       I != E; ++I) {
    CapturePred(&*I);
    if (I->isAssignedRegDep() && SU->getHeight() == LiveRegCycles[I->getReg()]) {
      assert(NumLiveRegs > 0 && "NumLiveRegs is already zero!");
      assert(LiveRegDefs[I->getReg()] == I->getSUnit() &&
             "Physical register dependency violated?");
      --NumLiveRegs;
      LiveRegDefs[I->getReg()] = NULL;
      LiveRegCycles[I->getReg()] = 0;
    }
  }

  for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    if (I->isAssignedRegDep()) {
      if (!LiveRegDefs[I->getReg()]) {
        LiveRegDefs[I->getReg()] = SU;
        ++NumLiveRegs;
      }
      if (I->getSUnit()->getHeight() < LiveRegCycles[I->getReg()])
        LiveRegCycles[I->getReg()] = I->getSUnit()->getHeight();
    }
  }

  SU->setHeightDirty();
  SU->isScheduled = false;
  SU->isAvailable = true;
  AvailableQueue->push(SU);
}

/// BacktrackBottomUp - Backtrack scheduling to a previous cycle specified in
/// BTCycle in order to schedule a specific node.
void ScheduleDAGRRList::BacktrackBottomUp(SUnit *SU, unsigned BtCycle,
                                          unsigned &CurCycle) {
  SUnit *OldSU = NULL;
  while (CurCycle > BtCycle) {
    OldSU = Sequence.back();
    Sequence.pop_back();
    if (SU->isSucc(OldSU))
      // Don't try to remove SU from AvailableQueue.
      SU->isAvailable = false;
    UnscheduleNodeBottomUp(OldSU);
    --CurCycle;
  }

  assert(!SU->isSucc(OldSU) && "Something is wrong!");

  ++NumBacktracks;
}

/// CopyAndMoveSuccessors - Clone the specified node and move its scheduled
/// successors to the newly created node.
SUnit *ScheduleDAGRRList::CopyAndMoveSuccessors(SUnit *SU) {
  if (SU->getNode()->getFlaggedNode())
    return NULL;

  SDNode *N = SU->getNode();
  if (!N)
    return NULL;

  SUnit *NewSU;
  bool TryUnfold = false;
  for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
    EVT VT = N->getValueType(i);
    if (VT == MVT::Flag)
      return NULL;
    else if (VT == MVT::Other)
      TryUnfold = true;
  }
  for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
    const SDValue &Op = N->getOperand(i);
    EVT VT = Op.getNode()->getValueType(Op.getResNo());
    if (VT == MVT::Flag)
      return NULL;
  }

  if (TryUnfold) {
    SmallVector<SDNode*, 2> NewNodes;
    if (!TII->unfoldMemoryOperand(*DAG, N, NewNodes))
      return NULL;

    DEBUG(dbgs() << "Unfolding SU # " << SU->NodeNum << "\n");
    assert(NewNodes.size() == 2 && "Expected a load folding node!");

    N = NewNodes[1];
    SDNode *LoadNode = NewNodes[0];
    unsigned NumVals = N->getNumValues();
    unsigned OldNumVals = SU->getNode()->getNumValues();
    for (unsigned i = 0; i != NumVals; ++i)
      DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), i), SDValue(N, i));
    DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), OldNumVals-1),
                                   SDValue(LoadNode, 1));

    // LoadNode may already exist. This can happen when there is another
    // load from the same location and producing the same type of value
    // but it has different alignment or volatileness.
    bool isNewLoad = true;
    SUnit *LoadSU;
    if (LoadNode->getNodeId() != -1) {
      LoadSU = &SUnits[LoadNode->getNodeId()];
      isNewLoad = false;
    } else {
      LoadSU = CreateNewSUnit(LoadNode);
      LoadNode->setNodeId(LoadSU->NodeNum);
      ComputeLatency(LoadSU);
    }

    SUnit *NewSU = CreateNewSUnit(N);
    assert(N->getNodeId() == -1 && "Node already inserted!");
    N->setNodeId(NewSU->NodeNum);
      
    const TargetInstrDesc &TID = TII->get(N->getMachineOpcode());
    for (unsigned i = 0; i != TID.getNumOperands(); ++i) {
      if (TID.getOperandConstraint(i, TOI::TIED_TO) != -1) {
        NewSU->isTwoAddress = true;
        break;
      }
    }
    if (TID.isCommutable())
      NewSU->isCommutable = true;
    ComputeLatency(NewSU);

    // Record all the edges to and from the old SU, by category.
    SmallVector<SDep, 4> ChainPreds;
    SmallVector<SDep, 4> ChainSuccs;
    SmallVector<SDep, 4> LoadPreds;
    SmallVector<SDep, 4> NodePreds;
    SmallVector<SDep, 4> NodeSuccs;
    for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
         I != E; ++I) {
      if (I->isCtrl())
        ChainPreds.push_back(*I);
      else if (I->getSUnit()->getNode() &&
               I->getSUnit()->getNode()->isOperandOf(LoadNode))
        LoadPreds.push_back(*I);
      else
        NodePreds.push_back(*I);
    }
    for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
         I != E; ++I) {
      if (I->isCtrl())
        ChainSuccs.push_back(*I);
      else
        NodeSuccs.push_back(*I);
    }

    // Now assign edges to the newly-created nodes.
    for (unsigned i = 0, e = ChainPreds.size(); i != e; ++i) {
      const SDep &Pred = ChainPreds[i];
      RemovePred(SU, Pred);
      if (isNewLoad)
        AddPred(LoadSU, Pred);
    }
    for (unsigned i = 0, e = LoadPreds.size(); i != e; ++i) {
      const SDep &Pred = LoadPreds[i];
      RemovePred(SU, Pred);
      if (isNewLoad)
        AddPred(LoadSU, Pred);
    }
    for (unsigned i = 0, e = NodePreds.size(); i != e; ++i) {
      const SDep &Pred = NodePreds[i];
      RemovePred(SU, Pred);
      AddPred(NewSU, Pred);
    }
    for (unsigned i = 0, e = NodeSuccs.size(); i != e; ++i) {
      SDep D = NodeSuccs[i];
      SUnit *SuccDep = D.getSUnit();
      D.setSUnit(SU);
      RemovePred(SuccDep, D);
      D.setSUnit(NewSU);
      AddPred(SuccDep, D);
    }
    for (unsigned i = 0, e = ChainSuccs.size(); i != e; ++i) {
      SDep D = ChainSuccs[i];
      SUnit *SuccDep = D.getSUnit();
      D.setSUnit(SU);
      RemovePred(SuccDep, D);
      if (isNewLoad) {
        D.setSUnit(LoadSU);
        AddPred(SuccDep, D);
      }
    } 

    // Add a data dependency to reflect that NewSU reads the value defined
    // by LoadSU.
    AddPred(NewSU, SDep(LoadSU, SDep::Data, LoadSU->Latency));

    if (isNewLoad)
      AvailableQueue->addNode(LoadSU);
    AvailableQueue->addNode(NewSU);

    ++NumUnfolds;

    if (NewSU->NumSuccsLeft == 0) {
      NewSU->isAvailable = true;
      return NewSU;
    }
    SU = NewSU;
  }

  DEBUG(dbgs() << "Duplicating SU # " << SU->NodeNum << "\n");
  NewSU = CreateClone(SU);

  // New SUnit has the exact same predecessors.
  for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
       I != E; ++I)
    if (!I->isArtificial())
      AddPred(NewSU, *I);

  // Only copy scheduled successors. Cut them from old node's successor
  // list and move them over.
  SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
  for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    if (I->isArtificial())
      continue;
    SUnit *SuccSU = I->getSUnit();
    if (SuccSU->isScheduled) {
      SDep D = *I;
      D.setSUnit(NewSU);
      AddPred(SuccSU, D);
      D.setSUnit(SU);
      DelDeps.push_back(std::make_pair(SuccSU, D));
    }
  }
  for (unsigned i = 0, e = DelDeps.size(); i != e; ++i)
    RemovePred(DelDeps[i].first, DelDeps[i].second);

  AvailableQueue->updateNode(SU);
  AvailableQueue->addNode(NewSU);

  ++NumDups;
  return NewSU;
}

/// InsertCopiesAndMoveSuccs - Insert register copies and move all
/// scheduled successors of the given SUnit to the last copy.
void ScheduleDAGRRList::InsertCopiesAndMoveSuccs(SUnit *SU, unsigned Reg,
                                               const TargetRegisterClass *DestRC,
                                               const TargetRegisterClass *SrcRC,
                                               SmallVector<SUnit*, 2> &Copies) {
  SUnit *CopyFromSU = CreateNewSUnit(NULL);
  CopyFromSU->CopySrcRC = SrcRC;
  CopyFromSU->CopyDstRC = DestRC;

  SUnit *CopyToSU = CreateNewSUnit(NULL);
  CopyToSU->CopySrcRC = DestRC;
  CopyToSU->CopyDstRC = SrcRC;

  // Only copy scheduled successors. Cut them from old node's successor
  // list and move them over.
  SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
  for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    if (I->isArtificial())
      continue;
    SUnit *SuccSU = I->getSUnit();
    if (SuccSU->isScheduled) {
      SDep D = *I;
      D.setSUnit(CopyToSU);
      AddPred(SuccSU, D);
      DelDeps.push_back(std::make_pair(SuccSU, *I));
    }
  }
  for (unsigned i = 0, e = DelDeps.size(); i != e; ++i)
    RemovePred(DelDeps[i].first, DelDeps[i].second);

  AddPred(CopyFromSU, SDep(SU, SDep::Data, SU->Latency, Reg));
  AddPred(CopyToSU, SDep(CopyFromSU, SDep::Data, CopyFromSU->Latency, 0));

  AvailableQueue->updateNode(SU);
  AvailableQueue->addNode(CopyFromSU);
  AvailableQueue->addNode(CopyToSU);
  Copies.push_back(CopyFromSU);
  Copies.push_back(CopyToSU);

  ++NumPRCopies;
}

/// getPhysicalRegisterVT - Returns the ValueType of the physical register
/// definition of the specified node.
/// FIXME: Move to SelectionDAG?
static EVT getPhysicalRegisterVT(SDNode *N, unsigned Reg,
                                 const TargetInstrInfo *TII) {
  const TargetInstrDesc &TID = TII->get(N->getMachineOpcode());
  assert(TID.ImplicitDefs && "Physical reg def must be in implicit def list!");
  unsigned NumRes = TID.getNumDefs();
  for (const unsigned *ImpDef = TID.getImplicitDefs(); *ImpDef; ++ImpDef) {
    if (Reg == *ImpDef)
      break;
    ++NumRes;
  }
  return N->getValueType(NumRes);
}

/// CheckForLiveRegDef - Return true and update live register vector if the
/// specified register def of the specified SUnit clobbers any "live" registers.
static bool CheckForLiveRegDef(SUnit *SU, unsigned Reg,
                               std::vector<SUnit*> &LiveRegDefs,
                               SmallSet<unsigned, 4> &RegAdded,
                               SmallVector<unsigned, 4> &LRegs,
                               const TargetRegisterInfo *TRI) {
  bool Added = false;
  if (LiveRegDefs[Reg] && LiveRegDefs[Reg] != SU) {
    if (RegAdded.insert(Reg)) {
      LRegs.push_back(Reg);
      Added = true;
    }
  }
  for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias)
    if (LiveRegDefs[*Alias] && LiveRegDefs[*Alias] != SU) {
      if (RegAdded.insert(*Alias)) {
        LRegs.push_back(*Alias);
        Added = true;
      }
    }
  return Added;
}

/// DelayForLiveRegsBottomUp - Returns true if it is necessary to delay
/// scheduling of the given node to satisfy live physical register dependencies.
/// If the specific node is the last one that's available to schedule, do
/// whatever is necessary (i.e. backtracking or cloning) to make it possible.
bool ScheduleDAGRRList::DelayForLiveRegsBottomUp(SUnit *SU,
                                                 SmallVector<unsigned, 4> &LRegs){
  if (NumLiveRegs == 0)
    return false;

  SmallSet<unsigned, 4> RegAdded;
  // If this node would clobber any "live" register, then it's not ready.
  for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
       I != E; ++I) {
    if (I->isAssignedRegDep())
      CheckForLiveRegDef(I->getSUnit(), I->getReg(), LiveRegDefs,
                         RegAdded, LRegs, TRI);
  }

  for (SDNode *Node = SU->getNode(); Node; Node = Node->getFlaggedNode()) {
    if (Node->getOpcode() == ISD::INLINEASM) {
      // Inline asm can clobber physical defs.
      unsigned NumOps = Node->getNumOperands();
      if (Node->getOperand(NumOps-1).getValueType() == MVT::Flag)
        --NumOps;  // Ignore the flag operand.

      for (unsigned i = 2; i != NumOps;) {
        unsigned Flags =
          cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
        unsigned NumVals = (Flags & 0xffff) >> 3;

        ++i; // Skip the ID value.
        if ((Flags & 7) == 2 || (Flags & 7) == 6) {
          // Check for def of register or earlyclobber register.
          for (; NumVals; --NumVals, ++i) {
            unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
            if (TargetRegisterInfo::isPhysicalRegister(Reg))
              CheckForLiveRegDef(SU, Reg, LiveRegDefs, RegAdded, LRegs, TRI);
          }
        } else
          i += NumVals;
      }
      continue;
    }

    if (!Node->isMachineOpcode())
      continue;
    const TargetInstrDesc &TID = TII->get(Node->getMachineOpcode());
    if (!TID.ImplicitDefs)
      continue;
    for (const unsigned *Reg = TID.ImplicitDefs; *Reg; ++Reg)
      CheckForLiveRegDef(SU, *Reg, LiveRegDefs, RegAdded, LRegs, TRI);
  }
  return !LRegs.empty();
}


/// ListScheduleBottomUp - The main loop of list scheduling for bottom-up
/// schedulers.
void ScheduleDAGRRList::ListScheduleBottomUp() {
  unsigned CurCycle = 0;

  // Release any predecessors of the special Exit node.
  ReleasePredecessors(&ExitSU, CurCycle);

  // Add root to Available queue.
  if (!SUnits.empty()) {
    SUnit *RootSU = &SUnits[DAG->getRoot().getNode()->getNodeId()];
    assert(RootSU->Succs.empty() && "Graph root shouldn't have successors!");
    RootSU->isAvailable = true;
    AvailableQueue->push(RootSU);
  }

  // While Available queue is not empty, grab the node with the highest
  // priority. If it is not ready put it back.  Schedule the node.
  SmallVector<SUnit*, 4> NotReady;
  DenseMap<SUnit*, SmallVector<unsigned, 4> > LRegsMap;
  Sequence.reserve(SUnits.size());
  while (!AvailableQueue->empty()) {
    bool Delayed = false;
    LRegsMap.clear();
    SUnit *CurSU = AvailableQueue->pop();
    while (CurSU) {
      SmallVector<unsigned, 4> LRegs;
      if (!DelayForLiveRegsBottomUp(CurSU, LRegs))
        break;
      Delayed = true;
      LRegsMap.insert(std::make_pair(CurSU, LRegs));

      CurSU->isPending = true;  // This SU is not in AvailableQueue right now.
      NotReady.push_back(CurSU);
      CurSU = AvailableQueue->pop();
    }

    // All candidates are delayed due to live physical reg dependencies.
    // Try backtracking, code duplication, or inserting cross class copies
    // to resolve it.
    if (Delayed && !CurSU) {
      for (unsigned i = 0, e = NotReady.size(); i != e; ++i) {
        SUnit *TrySU = NotReady[i];
        SmallVector<unsigned, 4> &LRegs = LRegsMap[TrySU];

        // Try unscheduling up to the point where it's safe to schedule
        // this node.
        unsigned LiveCycle = CurCycle;
        for (unsigned j = 0, ee = LRegs.size(); j != ee; ++j) {
          unsigned Reg = LRegs[j];
          unsigned LCycle = LiveRegCycles[Reg];
          LiveCycle = std::min(LiveCycle, LCycle);
        }
        SUnit *OldSU = Sequence[LiveCycle];
        if (!WillCreateCycle(TrySU, OldSU))  {
          BacktrackBottomUp(TrySU, LiveCycle, CurCycle);
          // Force the current node to be scheduled before the node that
          // requires the physical reg dep.
          if (OldSU->isAvailable) {
            OldSU->isAvailable = false;
            AvailableQueue->remove(OldSU);
          }
          AddPred(TrySU, SDep(OldSU, SDep::Order, /*Latency=*/1,
                              /*Reg=*/0, /*isNormalMemory=*/false,
                              /*isMustAlias=*/false, /*isArtificial=*/true));
          // If one or more successors has been unscheduled, then the current
          // node is no longer avaialable. Schedule a successor that's now
          // available instead.
          if (!TrySU->isAvailable)
            CurSU = AvailableQueue->pop();
          else {
            CurSU = TrySU;
            TrySU->isPending = false;
            NotReady.erase(NotReady.begin()+i);
          }
          break;
        }
      }

      if (!CurSU) {
        // Can't backtrack. If it's too expensive to copy the value, then try
        // duplicate the nodes that produces these "too expensive to copy"
        // values to break the dependency. In case even that doesn't work,
        // insert cross class copies.
        // If it's not too expensive, i.e. cost != -1, issue copies.
        SUnit *TrySU = NotReady[0];
        SmallVector<unsigned, 4> &LRegs = LRegsMap[TrySU];
        assert(LRegs.size() == 1 && "Can't handle this yet!");
        unsigned Reg = LRegs[0];
        SUnit *LRDef = LiveRegDefs[Reg];
        EVT VT = getPhysicalRegisterVT(LRDef->getNode(), Reg, TII);
        const TargetRegisterClass *RC =
          TRI->getPhysicalRegisterRegClass(Reg, VT);
        const TargetRegisterClass *DestRC = TRI->getCrossCopyRegClass(RC);

        // If cross copy register class is null, then it must be possible copy
        // the value directly. Do not try duplicate the def.
        SUnit *NewDef = 0;
        if (DestRC)
          NewDef = CopyAndMoveSuccessors(LRDef);
        else
          DestRC = RC;
        if (!NewDef) {
          // Issue copies, these can be expensive cross register class copies.
          SmallVector<SUnit*, 2> Copies;
          InsertCopiesAndMoveSuccs(LRDef, Reg, DestRC, RC, Copies);
          DEBUG(dbgs() << "Adding an edge from SU #" << TrySU->NodeNum
                       << " to SU #" << Copies.front()->NodeNum << "\n");
          AddPred(TrySU, SDep(Copies.front(), SDep::Order, /*Latency=*/1,
                              /*Reg=*/0, /*isNormalMemory=*/false,
                              /*isMustAlias=*/false,
                              /*isArtificial=*/true));
          NewDef = Copies.back();
        }

        DEBUG(dbgs() << "Adding an edge from SU #" << NewDef->NodeNum
                     << " to SU #" << TrySU->NodeNum << "\n");
        LiveRegDefs[Reg] = NewDef;
        AddPred(NewDef, SDep(TrySU, SDep::Order, /*Latency=*/1,
                             /*Reg=*/0, /*isNormalMemory=*/false,
                             /*isMustAlias=*/false,
                             /*isArtificial=*/true));
        TrySU->isAvailable = false;
        CurSU = NewDef;
      }

      assert(CurSU && "Unable to resolve live physical register dependencies!");
    }

    // Add the nodes that aren't ready back onto the available list.
    for (unsigned i = 0, e = NotReady.size(); i != e; ++i) {
      NotReady[i]->isPending = false;
      // May no longer be available due to backtracking.
      if (NotReady[i]->isAvailable)
        AvailableQueue->push(NotReady[i]);
    }
    NotReady.clear();

    if (CurSU)
      ScheduleNodeBottomUp(CurSU, CurCycle);
    ++CurCycle;
  }

  // Reverse the order if it is bottom up.
  std::reverse(Sequence.begin(), Sequence.end());
  
#ifndef NDEBUG
  VerifySchedule(isBottomUp);
#endif
}

//===----------------------------------------------------------------------===//
//  Top-Down Scheduling
//===----------------------------------------------------------------------===//

/// ReleaseSucc - Decrement the NumPredsLeft count of a successor. Add it to
/// the AvailableQueue if the count reaches zero. Also update its cycle bound.
void ScheduleDAGRRList::ReleaseSucc(SUnit *SU, const SDep *SuccEdge) {
  SUnit *SuccSU = SuccEdge->getSUnit();

#ifndef NDEBUG
  if (SuccSU->NumPredsLeft == 0) {
    dbgs() << "*** Scheduling failed! ***\n";
    SuccSU->dump(this);
    dbgs() << " has been released too many times!\n";
    llvm_unreachable(0);
  }
#endif
  --SuccSU->NumPredsLeft;

  // If all the node's predecessors are scheduled, this node is ready
  // to be scheduled. Ignore the special ExitSU node.
  if (SuccSU->NumPredsLeft == 0 && SuccSU != &ExitSU) {
    SuccSU->isAvailable = true;
    AvailableQueue->push(SuccSU);
  }
}

void ScheduleDAGRRList::ReleaseSuccessors(SUnit *SU) {
  // Top down: release successors
  for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    assert(!I->isAssignedRegDep() &&
           "The list-tdrr scheduler doesn't yet support physreg dependencies!");

    ReleaseSucc(SU, &*I);
  }
}

/// ScheduleNodeTopDown - Add the node to the schedule. Decrement the pending
/// count of its successors. If a successor pending count is zero, add it to
/// the Available queue.
void ScheduleDAGRRList::ScheduleNodeTopDown(SUnit *SU, unsigned CurCycle) {
  DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
  DEBUG(SU->dump(this));

  assert(CurCycle >= SU->getDepth() && "Node scheduled above its depth!");
  SU->setDepthToAtLeast(CurCycle);
  Sequence.push_back(SU);

  ReleaseSuccessors(SU);
  SU->isScheduled = true;
  AvailableQueue->ScheduledNode(SU);
}

/// ListScheduleTopDown - The main loop of list scheduling for top-down
/// schedulers.
void ScheduleDAGRRList::ListScheduleTopDown() {
  unsigned CurCycle = 0;

  // Release any successors of the special Entry node.
  ReleaseSuccessors(&EntrySU);

  // All leaves to Available queue.
  for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
    // It is available if it has no predecessors.
    if (SUnits[i].Preds.empty()) {
      AvailableQueue->push(&SUnits[i]);
      SUnits[i].isAvailable = true;
    }
  }
  
  // While Available queue is not empty, grab the node with the highest
  // priority. If it is not ready put it back.  Schedule the node.
  Sequence.reserve(SUnits.size());
  while (!AvailableQueue->empty()) {
    SUnit *CurSU = AvailableQueue->pop();
    
    if (CurSU)
      ScheduleNodeTopDown(CurSU, CurCycle);
    ++CurCycle;
  }
  
#ifndef NDEBUG
  VerifySchedule(isBottomUp);
#endif
}


//===----------------------------------------------------------------------===//
//                RegReductionPriorityQueue Implementation
//===----------------------------------------------------------------------===//
//
// This is a SchedulingPriorityQueue that schedules using Sethi Ullman numbers
// to reduce register pressure.
// 
namespace {
  template<class SF>
  class RegReductionPriorityQueue;
  
  /// Sorting functions for the Available queue.
  struct bu_ls_rr_sort : public std::binary_function<SUnit*, SUnit*, bool> {
    RegReductionPriorityQueue<bu_ls_rr_sort> *SPQ;
    bu_ls_rr_sort(RegReductionPriorityQueue<bu_ls_rr_sort> *spq) : SPQ(spq) {}
    bu_ls_rr_sort(const bu_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {}
    
    bool operator()(const SUnit* left, const SUnit* right) const;
  };

  struct td_ls_rr_sort : public std::binary_function<SUnit*, SUnit*, bool> {
    RegReductionPriorityQueue<td_ls_rr_sort> *SPQ;
    td_ls_rr_sort(RegReductionPriorityQueue<td_ls_rr_sort> *spq) : SPQ(spq) {}
    td_ls_rr_sort(const td_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {}
    
    bool operator()(const SUnit* left, const SUnit* right) const;
  };
}  // end anonymous namespace

/// CalcNodeSethiUllmanNumber - Compute Sethi Ullman number.
/// Smaller number is the higher priority.
static unsigned
CalcNodeSethiUllmanNumber(const SUnit *SU, std::vector<unsigned> &SUNumbers) {
  unsigned &SethiUllmanNumber = SUNumbers[SU->NodeNum];
  if (SethiUllmanNumber != 0)
    return SethiUllmanNumber;

  unsigned Extra = 0;
  for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
       I != E; ++I) {
    if (I->isCtrl()) continue;  // ignore chain preds
    SUnit *PredSU = I->getSUnit();
    unsigned PredSethiUllman = CalcNodeSethiUllmanNumber(PredSU, SUNumbers);
    if (PredSethiUllman > SethiUllmanNumber) {
      SethiUllmanNumber = PredSethiUllman;
      Extra = 0;
    } else if (PredSethiUllman == SethiUllmanNumber)
      ++Extra;
  }

  SethiUllmanNumber += Extra;

  if (SethiUllmanNumber == 0)
    SethiUllmanNumber = 1;
  
  return SethiUllmanNumber;
}

namespace {
  template<class SF>
  class RegReductionPriorityQueue : public SchedulingPriorityQueue {
    PriorityQueue<SUnit*, std::vector<SUnit*>, SF> Queue;
    unsigned currentQueueId;

  protected:
    // SUnits - The SUnits for the current graph.
    std::vector<SUnit> *SUnits;
    
    const TargetInstrInfo *TII;
    const TargetRegisterInfo *TRI;
    ScheduleDAGRRList *scheduleDAG;

    // SethiUllmanNumbers - The SethiUllman number for each node.
    std::vector<unsigned> SethiUllmanNumbers;

  public:
    RegReductionPriorityQueue(const TargetInstrInfo *tii,
                              const TargetRegisterInfo *tri) :
    Queue(SF(this)), currentQueueId(0),
    TII(tii), TRI(tri), scheduleDAG(NULL) {}
    
    void initNodes(std::vector<SUnit> &sunits) {
      SUnits = &sunits;
      // Add pseudo dependency edges for two-address nodes.
      AddPseudoTwoAddrDeps();
      // Reroute edges to nodes with multiple uses.
      PrescheduleNodesWithMultipleUses();
      // Calculate node priorities.
      CalculateSethiUllmanNumbers();
    }

    void addNode(const SUnit *SU) {
      unsigned SUSize = SethiUllmanNumbers.size();
      if (SUnits->size() > SUSize)
        SethiUllmanNumbers.resize(SUSize*2, 0);
      CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers);
    }

    void updateNode(const SUnit *SU) {
      SethiUllmanNumbers[SU->NodeNum] = 0;
      CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers);
    }

    void releaseState() {
      SUnits = 0;
      SethiUllmanNumbers.clear();
    }

    unsigned getNodePriority(const SUnit *SU) const {
      assert(SU->NodeNum < SethiUllmanNumbers.size());
      unsigned Opc = SU->getNode() ? SU->getNode()->getOpcode() : 0;
      if (Opc == ISD::TokenFactor || Opc == ISD::CopyToReg)
        // CopyToReg should be close to its uses to facilitate coalescing and
        // avoid spilling.
        return 0;
      if (Opc == TargetInstrInfo::EXTRACT_SUBREG ||
          Opc == TargetInstrInfo::SUBREG_TO_REG ||
          Opc == TargetInstrInfo::INSERT_SUBREG)
        // EXTRACT_SUBREG, INSERT_SUBREG, and SUBREG_TO_REG nodes should be
        // close to their uses to facilitate coalescing.
        return 0;
      if (SU->NumSuccs == 0 && SU->NumPreds != 0)
        // If SU does not have a register use, i.e. it doesn't produce a value
        // that would be consumed (e.g. store), then it terminates a chain of
        // computation.  Give it a large SethiUllman number so it will be
        // scheduled right before its predecessors that it doesn't lengthen
        // their live ranges.
        return 0xffff;
      if (SU->NumPreds == 0 && SU->NumSuccs != 0)
        // If SU does not have a register def, schedule it close to its uses
        // because it does not lengthen any live ranges.
        return 0;
      return SethiUllmanNumbers[SU->NodeNum];
    }

    unsigned getNodeOrdering(const SUnit *SU) const {
      return scheduleDAG->DAG->GetOrdering(SU->getNode());
    }
    
    unsigned size() const { return Queue.size(); }

    bool empty() const { return Queue.empty(); }
    
    void push(SUnit *U) {
      assert(!U->NodeQueueId && "Node in the queue already");
      U->NodeQueueId = ++currentQueueId;
      Queue.push(U);
    }

    void push_all(const std::vector<SUnit *> &Nodes) {
      for (unsigned i = 0, e = Nodes.size(); i != e; ++i)
        push(Nodes[i]);
    }
    
    SUnit *pop() {
      if (empty()) return NULL;
      SUnit *V = Queue.top();
      Queue.pop();
      V->NodeQueueId = 0;
      return V;
    }

    void remove(SUnit *SU) {
      assert(!Queue.empty() && "Queue is empty!");
      assert(SU->NodeQueueId != 0 && "Not in queue!");
      Queue.erase_one(SU);
      SU->NodeQueueId = 0;
    }

    void setScheduleDAG(ScheduleDAGRRList *scheduleDag) { 
      scheduleDAG = scheduleDag; 
    }

  protected:
    bool canClobber(const SUnit *SU, const SUnit *Op);
    void AddPseudoTwoAddrDeps();
    void PrescheduleNodesWithMultipleUses();
    void CalculateSethiUllmanNumbers();
  };

  typedef RegReductionPriorityQueue<bu_ls_rr_sort>
    BURegReductionPriorityQueue;

  typedef RegReductionPriorityQueue<td_ls_rr_sort>
    TDRegReductionPriorityQueue;
}

/// closestSucc - Returns the scheduled cycle of the successor which is
/// closest to the current cycle.
static unsigned closestSucc(const SUnit *SU) {
  unsigned MaxHeight = 0;
  for (SUnit::const_succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    if (I->isCtrl()) continue;  // ignore chain succs
    unsigned Height = I->getSUnit()->getHeight();
    // If there are bunch of CopyToRegs stacked up, they should be considered
    // to be at the same position.
    if (I->getSUnit()->getNode() &&
        I->getSUnit()->getNode()->getOpcode() == ISD::CopyToReg)
      Height = closestSucc(I->getSUnit())+1;
    if (Height > MaxHeight)
      MaxHeight = Height;
  }
  return MaxHeight;
}

/// calcMaxScratches - Returns an cost estimate of the worse case requirement
/// for scratch registers, i.e. number of data dependencies.
static unsigned calcMaxScratches(const SUnit *SU) {
  unsigned Scratches = 0;
  for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
       I != E; ++I) {
    if (I->isCtrl()) continue;  // ignore chain preds
    Scratches++;
  }
  return Scratches;
}

// Bottom up
bool bu_ls_rr_sort::operator()(const SUnit *left, const SUnit *right) const {
  unsigned LOrder = SPQ->getNodeOrdering(left);
  unsigned ROrder = SPQ->getNodeOrdering(right);

  // Prefer an ordering where the lower the non-zero order number, the higher
  // the preference.
  if ((LOrder || ROrder) && LOrder != ROrder)
    return LOrder != 0 && (LOrder < ROrder || ROrder == 0);

  unsigned LPriority = SPQ->getNodePriority(left);
  unsigned RPriority = SPQ->getNodePriority(right);
  if (LPriority != RPriority)
    return LPriority > RPriority;

  // Try schedule def + use closer when Sethi-Ullman numbers are the same.
  // e.g.
  // t1 = op t2, c1
  // t3 = op t4, c2
  //
  // and the following instructions are both ready.
  // t2 = op c3
  // t4 = op c4
  //
  // Then schedule t2 = op first.
  // i.e.
  // t4 = op c4
  // t2 = op c3
  // t1 = op t2, c1
  // t3 = op t4, c2
  //
  // This creates more short live intervals.
  unsigned LDist = closestSucc(left);
  unsigned RDist = closestSucc(right);
  if (LDist != RDist)
    return LDist < RDist;

  // How many registers becomes live when the node is scheduled.
  unsigned LScratch = calcMaxScratches(left);
  unsigned RScratch = calcMaxScratches(right);
  if (LScratch != RScratch)
    return LScratch > RScratch;

  if (left->getHeight() != right->getHeight())
    return left->getHeight() > right->getHeight();
  
  if (left->getDepth() != right->getDepth())
    return left->getDepth() < right->getDepth();

  assert(left->NodeQueueId && right->NodeQueueId && 
         "NodeQueueId cannot be zero");
  return (left->NodeQueueId > right->NodeQueueId);
}

template<class SF>
bool
RegReductionPriorityQueue<SF>::canClobber(const SUnit *SU, const SUnit *Op) {
  if (SU->isTwoAddress) {
    unsigned Opc = SU->getNode()->getMachineOpcode();
    const TargetInstrDesc &TID = TII->get(Opc);
    unsigned NumRes = TID.getNumDefs();
    unsigned NumOps = TID.getNumOperands() - NumRes;
    for (unsigned i = 0; i != NumOps; ++i) {
      if (TID.getOperandConstraint(i+NumRes, TOI::TIED_TO) != -1) {
        SDNode *DU = SU->getNode()->getOperand(i).getNode();
        if (DU->getNodeId() != -1 &&
            Op->OrigNode == &(*SUnits)[DU->getNodeId()])
          return true;
      }
    }
  }
  return false;
}


/// hasCopyToRegUse - Return true if SU has a value successor that is a
/// CopyToReg node.
static bool hasCopyToRegUse(const SUnit *SU) {
  for (SUnit::const_succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    if (I->isCtrl()) continue;
    const SUnit *SuccSU = I->getSUnit();
    if (SuccSU->getNode() && SuccSU->getNode()->getOpcode() == ISD::CopyToReg)
      return true;
  }
  return false;
}

/// canClobberPhysRegDefs - True if SU would clobber one of SuccSU's
/// physical register defs.
static bool canClobberPhysRegDefs(const SUnit *SuccSU, const SUnit *SU,
                                  const TargetInstrInfo *TII,
                                  const TargetRegisterInfo *TRI) {
  SDNode *N = SuccSU->getNode();
  unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs();
  const unsigned *ImpDefs = TII->get(N->getMachineOpcode()).getImplicitDefs();
  assert(ImpDefs && "Caller should check hasPhysRegDefs");
  for (const SDNode *SUNode = SU->getNode(); SUNode;
       SUNode = SUNode->getFlaggedNode()) {
    if (!SUNode->isMachineOpcode())
      continue;
    const unsigned *SUImpDefs =
      TII->get(SUNode->getMachineOpcode()).getImplicitDefs();
    if (!SUImpDefs)
      return false;
    for (unsigned i = NumDefs, e = N->getNumValues(); i != e; ++i) {
      EVT VT = N->getValueType(i);
      if (VT == MVT::Flag || VT == MVT::Other)
        continue;
      if (!N->hasAnyUseOfValue(i))
        continue;
      unsigned Reg = ImpDefs[i - NumDefs];
      for (;*SUImpDefs; ++SUImpDefs) {
        unsigned SUReg = *SUImpDefs;
        if (TRI->regsOverlap(Reg, SUReg))
          return true;
      }
    }
  }
  return false;
}

/// PrescheduleNodesWithMultipleUses - Nodes with multiple uses
/// are not handled well by the general register pressure reduction
/// heuristics. When presented with code like this:
///
///      N
///    / |
///   /  |
///  U  store
///  |
/// ...
///
/// the heuristics tend to push the store up, but since the
/// operand of the store has another use (U), this would increase
/// the length of that other use (the U->N edge).
///
/// This function transforms code like the above to route U's
/// dependence through the store when possible, like this:
///
///      N
///      ||
///      ||
///     store
///       |
///       U
///       |
///      ...
///
/// This results in the store being scheduled immediately
/// after N, which shortens the U->N live range, reducing
/// register pressure.
///
template<class SF>
void RegReductionPriorityQueue<SF>::PrescheduleNodesWithMultipleUses() {
  // Visit all the nodes in topological order, working top-down.
  for (unsigned i = 0, e = SUnits->size(); i != e; ++i) {
    SUnit *SU = &(*SUnits)[i];
    // For now, only look at nodes with no data successors, such as stores.
    // These are especially important, due to the heuristics in
    // getNodePriority for nodes with no data successors.
    if (SU->NumSuccs != 0)
      continue;
    // For now, only look at nodes with exactly one data predecessor.
    if (SU->NumPreds != 1)
      continue;
    // Avoid prescheduling copies to virtual registers, which don't behave
    // like other nodes from the perspective of scheduling heuristics.
    if (SDNode *N = SU->getNode())
      if (N->getOpcode() == ISD::CopyToReg &&
          TargetRegisterInfo::isVirtualRegister
            (cast<RegisterSDNode>(N->getOperand(1))->getReg()))
        continue;

    // Locate the single data predecessor.
    SUnit *PredSU = 0;
    for (SUnit::const_pred_iterator II = SU->Preds.begin(),
         EE = SU->Preds.end(); II != EE; ++II)
      if (!II->isCtrl()) {
        PredSU = II->getSUnit();
        break;
      }
    assert(PredSU);

    // Don't rewrite edges that carry physregs, because that requires additional
    // support infrastructure.
    if (PredSU->hasPhysRegDefs)
      continue;
    // Short-circuit the case where SU is PredSU's only data successor.
    if (PredSU->NumSuccs == 1)
      continue;
    // Avoid prescheduling to copies from virtual registers, which don't behave
    // like other nodes from the perspective of scheduling // heuristics.
    if (SDNode *N = SU->getNode())
      if (N->getOpcode() == ISD::CopyFromReg &&
          TargetRegisterInfo::isVirtualRegister
            (cast<RegisterSDNode>(N->getOperand(1))->getReg()))
        continue;

    // Perform checks on the successors of PredSU.
    for (SUnit::const_succ_iterator II = PredSU->Succs.begin(),
         EE = PredSU->Succs.end(); II != EE; ++II) {
      SUnit *PredSuccSU = II->getSUnit();
      if (PredSuccSU == SU) continue;
      // If PredSU has another successor with no data successors, for
      // now don't attempt to choose either over the other.
      if (PredSuccSU->NumSuccs == 0)
        goto outer_loop_continue;
      // Don't break physical register dependencies.
      if (SU->hasPhysRegClobbers && PredSuccSU->hasPhysRegDefs)
        if (canClobberPhysRegDefs(PredSuccSU, SU, TII, TRI))
          goto outer_loop_continue;
      // Don't introduce graph cycles.
      if (scheduleDAG->IsReachable(SU, PredSuccSU))
        goto outer_loop_continue;
    }

    // Ok, the transformation is safe and the heuristics suggest it is
    // profitable. Update the graph.
    DEBUG(dbgs() << "Prescheduling SU # " << SU->NodeNum
                 << " next to PredSU # " << PredSU->NodeNum
                 << " to guide scheduling in the presence of multiple uses\n");
    for (unsigned i = 0; i != PredSU->Succs.size(); ++i) {
      SDep Edge = PredSU->Succs[i];
      assert(!Edge.isAssignedRegDep());
      SUnit *SuccSU = Edge.getSUnit();
      if (SuccSU != SU) {
        Edge.setSUnit(PredSU);
        scheduleDAG->RemovePred(SuccSU, Edge);
        scheduleDAG->AddPred(SU, Edge);
        Edge.setSUnit(SU);
        scheduleDAG->AddPred(SuccSU, Edge);
        --i;
      }
    }
  outer_loop_continue:;
  }
}

/// AddPseudoTwoAddrDeps - If two nodes share an operand and one of them uses
/// it as a def&use operand. Add a pseudo control edge from it to the other
/// node (if it won't create a cycle) so the two-address one will be scheduled
/// first (lower in the schedule). If both nodes are two-address, favor the
/// one that has a CopyToReg use (more likely to be a loop induction update).
/// If both are two-address, but one is commutable while the other is not
/// commutable, favor the one that's not commutable.
template<class SF>
void RegReductionPriorityQueue<SF>::AddPseudoTwoAddrDeps() {
  for (unsigned i = 0, e = SUnits->size(); i != e; ++i) {
    SUnit *SU = &(*SUnits)[i];
    if (!SU->isTwoAddress)
      continue;

    SDNode *Node = SU->getNode();
    if (!Node || !Node->isMachineOpcode() || SU->getNode()->getFlaggedNode())
      continue;

    unsigned Opc = Node->getMachineOpcode();
    const TargetInstrDesc &TID = TII->get(Opc);
    unsigned NumRes = TID.getNumDefs();
    unsigned NumOps = TID.getNumOperands() - NumRes;
    for (unsigned j = 0; j != NumOps; ++j) {
      if (TID.getOperandConstraint(j+NumRes, TOI::TIED_TO) == -1)
        continue;
      SDNode *DU = SU->getNode()->getOperand(j).getNode();
      if (DU->getNodeId() == -1)
        continue;
      const SUnit *DUSU = &(*SUnits)[DU->getNodeId()];
      if (!DUSU) continue;
      for (SUnit::const_succ_iterator I = DUSU->Succs.begin(),
           E = DUSU->Succs.end(); I != E; ++I) {
        if (I->isCtrl()) continue;
        SUnit *SuccSU = I->getSUnit();
        if (SuccSU == SU)
          continue;
        // Be conservative. Ignore if nodes aren't at roughly the same
        // depth and height.
        if (SuccSU->getHeight() < SU->getHeight() &&
            (SU->getHeight() - SuccSU->getHeight()) > 1)
          continue;
        // Skip past COPY_TO_REGCLASS nodes, so that the pseudo edge
        // constrains whatever is using the copy, instead of the copy
        // itself. In the case that the copy is coalesced, this
        // preserves the intent of the pseudo two-address heurietics.
        while (SuccSU->Succs.size() == 1 &&
               SuccSU->getNode()->isMachineOpcode() &&
               SuccSU->getNode()->getMachineOpcode() ==
                 TargetInstrInfo::COPY_TO_REGCLASS)
          SuccSU = SuccSU->Succs.front().getSUnit();
        // Don't constrain non-instruction nodes.
        if (!SuccSU->getNode() || !SuccSU->getNode()->isMachineOpcode())
          continue;
        // Don't constrain nodes with physical register defs if the
        // predecessor can clobber them.
        if (SuccSU->hasPhysRegDefs && SU->hasPhysRegClobbers) {
          if (canClobberPhysRegDefs(SuccSU, SU, TII, TRI))
            continue;
        }
        // Don't constrain EXTRACT_SUBREG, INSERT_SUBREG, and SUBREG_TO_REG;
        // these may be coalesced away. We want them close to their uses.
        unsigned SuccOpc = SuccSU->getNode()->getMachineOpcode();
        if (SuccOpc == TargetInstrInfo::EXTRACT_SUBREG ||
            SuccOpc == TargetInstrInfo::INSERT_SUBREG ||
            SuccOpc == TargetInstrInfo::SUBREG_TO_REG)
          continue;
        if ((!canClobber(SuccSU, DUSU) ||
             (hasCopyToRegUse(SU) && !hasCopyToRegUse(SuccSU)) ||
             (!SU->isCommutable && SuccSU->isCommutable)) &&
            !scheduleDAG->IsReachable(SuccSU, SU)) {
          DEBUG(dbgs() << "Adding a pseudo-two-addr edge from SU # "
                       << SU->NodeNum << " to SU #" << SuccSU->NodeNum << "\n");
          scheduleDAG->AddPred(SU, SDep(SuccSU, SDep::Order, /*Latency=*/0,
                                        /*Reg=*/0, /*isNormalMemory=*/false,
                                        /*isMustAlias=*/false,
                                        /*isArtificial=*/true));
        }
      }
    }
  }
}

/// CalculateSethiUllmanNumbers - Calculate Sethi-Ullman numbers of all
/// scheduling units.
template<class SF>
void RegReductionPriorityQueue<SF>::CalculateSethiUllmanNumbers() {
  SethiUllmanNumbers.assign(SUnits->size(), 0);
  
  for (unsigned i = 0, e = SUnits->size(); i != e; ++i)
    CalcNodeSethiUllmanNumber(&(*SUnits)[i], SethiUllmanNumbers);
}

/// LimitedSumOfUnscheduledPredsOfSuccs - Compute the sum of the unscheduled
/// predecessors of the successors of the SUnit SU. Stop when the provided
/// limit is exceeded.
static unsigned LimitedSumOfUnscheduledPredsOfSuccs(const SUnit *SU, 
                                                    unsigned Limit) {
  unsigned Sum = 0;
  for (SUnit::const_succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
       I != E; ++I) {
    const SUnit *SuccSU = I->getSUnit();
    for (SUnit::const_pred_iterator II = SuccSU->Preds.begin(),
         EE = SuccSU->Preds.end(); II != EE; ++II) {
      SUnit *PredSU = II->getSUnit();
      if (!PredSU->isScheduled)
        if (++Sum > Limit)
          return Sum;
    }
  }
  return Sum;
}


// Top down
bool td_ls_rr_sort::operator()(const SUnit *left, const SUnit *right) const {
  unsigned LPriority = SPQ->getNodePriority(left);
  unsigned RPriority = SPQ->getNodePriority(right);
  bool LIsTarget = left->getNode() && left->getNode()->isMachineOpcode();
  bool RIsTarget = right->getNode() && right->getNode()->isMachineOpcode();
  bool LIsFloater = LIsTarget && left->NumPreds == 0;
  bool RIsFloater = RIsTarget && right->NumPreds == 0;
  unsigned LBonus = (LimitedSumOfUnscheduledPredsOfSuccs(left,1) == 1) ? 2 : 0;
  unsigned RBonus = (LimitedSumOfUnscheduledPredsOfSuccs(right,1) == 1) ? 2 : 0;

  if (left->NumSuccs == 0 && right->NumSuccs != 0)
    return false;
  else if (left->NumSuccs != 0 && right->NumSuccs == 0)
    return true;

  if (LIsFloater)
    LBonus -= 2;
  if (RIsFloater)
    RBonus -= 2;
  if (left->NumSuccs == 1)
    LBonus += 2;
  if (right->NumSuccs == 1)
    RBonus += 2;

  if (LPriority+LBonus != RPriority+RBonus)
    return LPriority+LBonus < RPriority+RBonus;

  if (left->getDepth() != right->getDepth())
    return left->getDepth() < right->getDepth();

  if (left->NumSuccsLeft != right->NumSuccsLeft)
    return left->NumSuccsLeft > right->NumSuccsLeft;

  assert(left->NodeQueueId && right->NodeQueueId && 
         "NodeQueueId cannot be zero");
  return (left->NodeQueueId > right->NodeQueueId);
}

//===----------------------------------------------------------------------===//
//                         Public Constructor Functions
//===----------------------------------------------------------------------===//

llvm::ScheduleDAGSDNodes *
llvm::createBURRListDAGScheduler(SelectionDAGISel *IS, CodeGenOpt::Level) {
  const TargetMachine &TM = IS->TM;
  const TargetInstrInfo *TII = TM.getInstrInfo();
  const TargetRegisterInfo *TRI = TM.getRegisterInfo();
  
  BURegReductionPriorityQueue *PQ = new BURegReductionPriorityQueue(TII, TRI);

  ScheduleDAGRRList *SD =
    new ScheduleDAGRRList(*IS->MF, true, PQ);
  PQ->setScheduleDAG(SD);
  return SD;  
}

llvm::ScheduleDAGSDNodes *
llvm::createTDRRListDAGScheduler(SelectionDAGISel *IS, CodeGenOpt::Level) {
  const TargetMachine &TM = IS->TM;
  const TargetInstrInfo *TII = TM.getInstrInfo();
  const TargetRegisterInfo *TRI = TM.getRegisterInfo();
  
  TDRegReductionPriorityQueue *PQ = new TDRegReductionPriorityQueue(TII, TRI);

  ScheduleDAGRRList *SD =
    new ScheduleDAGRRList(*IS->MF, false, PQ);
  PQ->setScheduleDAG(SD);
  return SD;
}