aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/llvm16/lib/CodeGen/TwoAddressInstructionPass.cpp
blob: 8cb3667aea28a1140464e4be9976a00dc1e2d65f (plain) (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
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
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
//===- TwoAddressInstructionPass.cpp - Two-Address instruction pass -------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements the TwoAddress instruction pass which is used
// by most register allocators. Two-Address instructions are rewritten
// from:
//
//     A = B op C
//
// to:
//
//     A = B
//     A op= C
//
// Note that if a register allocator chooses to use this pass, that it
// has to be capable of handling the non-SSA nature of these rewritten
// virtual registers.
//
// It is also worth noting that the duplicate operand of the two
// address instruction is removed.
//
//===----------------------------------------------------------------------===//

#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/iterator_range.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/LiveInterval.h"
#include "llvm/CodeGen/LiveIntervals.h"
#include "llvm/CodeGen/LiveVariables.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/CodeGen/SlotIndexes.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/MC/MCInstrDesc.h"
#include "llvm/Pass.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include <cassert>
#include <iterator>
#include <utility>

using namespace llvm;

#define DEBUG_TYPE "twoaddressinstruction"

STATISTIC(NumTwoAddressInstrs, "Number of two-address instructions");
STATISTIC(NumCommuted        , "Number of instructions commuted to coalesce");
STATISTIC(NumAggrCommuted    , "Number of instructions aggressively commuted");
STATISTIC(NumConvertedTo3Addr, "Number of instructions promoted to 3-address");
STATISTIC(NumReSchedUps,       "Number of instructions re-scheduled up");
STATISTIC(NumReSchedDowns,     "Number of instructions re-scheduled down");

// Temporary flag to disable rescheduling.
static cl::opt<bool>
EnableRescheduling("twoaddr-reschedule",
                   cl::desc("Coalesce copies by rescheduling (default=true)"),
                   cl::init(true), cl::Hidden);

// Limit the number of dataflow edges to traverse when evaluating the benefit
// of commuting operands.
static cl::opt<unsigned> MaxDataFlowEdge(
    "dataflow-edge-limit", cl::Hidden, cl::init(3),
    cl::desc("Maximum number of dataflow edges to traverse when evaluating "
             "the benefit of commuting operands"));

namespace {

class TwoAddressInstructionPass : public MachineFunctionPass {
  MachineFunction *MF;
  const TargetInstrInfo *TII;
  const TargetRegisterInfo *TRI;
  const InstrItineraryData *InstrItins;
  MachineRegisterInfo *MRI;
  LiveVariables *LV;
  LiveIntervals *LIS;
  AliasAnalysis *AA;
  CodeGenOpt::Level OptLevel;

  // The current basic block being processed.
  MachineBasicBlock *MBB;

  // Keep track the distance of a MI from the start of the current basic block.
  DenseMap<MachineInstr*, unsigned> DistanceMap;

  // Set of already processed instructions in the current block.
  SmallPtrSet<MachineInstr*, 8> Processed;

  // A map from virtual registers to physical registers which are likely targets
  // to be coalesced to due to copies from physical registers to virtual
  // registers. e.g. v1024 = move r0.
  DenseMap<Register, Register> SrcRegMap;

  // A map from virtual registers to physical registers which are likely targets
  // to be coalesced to due to copies to physical registers from virtual
  // registers. e.g. r1 = move v1024.
  DenseMap<Register, Register> DstRegMap;

  void removeClobberedSrcRegMap(MachineInstr *MI);

  bool isRevCopyChain(Register FromReg, Register ToReg, int Maxlen);

  bool noUseAfterLastDef(Register Reg, unsigned Dist, unsigned &LastDef);

  bool isProfitableToCommute(Register RegA, Register RegB, Register RegC,
                             MachineInstr *MI, unsigned Dist);

  bool commuteInstruction(MachineInstr *MI, unsigned DstIdx,
                          unsigned RegBIdx, unsigned RegCIdx, unsigned Dist);

  bool isProfitableToConv3Addr(Register RegA, Register RegB);

  bool convertInstTo3Addr(MachineBasicBlock::iterator &mi,
                          MachineBasicBlock::iterator &nmi, Register RegA,
                          Register RegB, unsigned &Dist);

  bool isDefTooClose(Register Reg, unsigned Dist, MachineInstr *MI);

  bool rescheduleMIBelowKill(MachineBasicBlock::iterator &mi,
                             MachineBasicBlock::iterator &nmi, Register Reg);
  bool rescheduleKillAboveMI(MachineBasicBlock::iterator &mi,
                             MachineBasicBlock::iterator &nmi, Register Reg);

  bool tryInstructionTransform(MachineBasicBlock::iterator &mi,
                               MachineBasicBlock::iterator &nmi,
                               unsigned SrcIdx, unsigned DstIdx,
                               unsigned &Dist, bool shouldOnlyCommute);

  bool tryInstructionCommute(MachineInstr *MI,
                             unsigned DstOpIdx,
                             unsigned BaseOpIdx,
                             bool BaseOpKilled,
                             unsigned Dist);
  void scanUses(Register DstReg);

  void processCopy(MachineInstr *MI);

  using TiedPairList = SmallVector<std::pair<unsigned, unsigned>, 4>;
  using TiedOperandMap = SmallDenseMap<unsigned, TiedPairList>;

  bool collectTiedOperands(MachineInstr *MI, TiedOperandMap&);
  void processTiedPairs(MachineInstr *MI, TiedPairList&, unsigned &Dist);
  void eliminateRegSequence(MachineBasicBlock::iterator&);
  bool processStatepoint(MachineInstr *MI, TiedOperandMap &TiedOperands);

public:
  static char ID; // Pass identification, replacement for typeid

  TwoAddressInstructionPass() : MachineFunctionPass(ID) {
    initializeTwoAddressInstructionPassPass(*PassRegistry::getPassRegistry());
  }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.setPreservesCFG();
    AU.addUsedIfAvailable<AAResultsWrapperPass>();
    AU.addUsedIfAvailable<LiveVariables>();
    AU.addPreserved<LiveVariables>();
    AU.addPreserved<SlotIndexes>();
    AU.addPreserved<LiveIntervals>();
    AU.addPreservedID(MachineLoopInfoID);
    AU.addPreservedID(MachineDominatorsID);
    MachineFunctionPass::getAnalysisUsage(AU);
  }

  /// Pass entry point.
  bool runOnMachineFunction(MachineFunction&) override;
};

} // end anonymous namespace

char TwoAddressInstructionPass::ID = 0;

char &llvm::TwoAddressInstructionPassID = TwoAddressInstructionPass::ID;

INITIALIZE_PASS_BEGIN(TwoAddressInstructionPass, DEBUG_TYPE,
                "Two-Address instruction pass", false, false)
INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
INITIALIZE_PASS_END(TwoAddressInstructionPass, DEBUG_TYPE,
                "Two-Address instruction pass", false, false)

static bool isPlainlyKilled(MachineInstr *MI, Register Reg, LiveIntervals *LIS);

/// Return the MachineInstr* if it is the single def of the Reg in current BB.
static MachineInstr *getSingleDef(Register Reg, MachineBasicBlock *BB,
                                  const MachineRegisterInfo *MRI) {
  MachineInstr *Ret = nullptr;
  for (MachineInstr &DefMI : MRI->def_instructions(Reg)) {
    if (DefMI.getParent() != BB || DefMI.isDebugValue())
      continue;
    if (!Ret)
      Ret = &DefMI;
    else if (Ret != &DefMI)
      return nullptr;
  }
  return Ret;
}

/// Check if there is a reversed copy chain from FromReg to ToReg:
/// %Tmp1 = copy %Tmp2;
/// %FromReg = copy %Tmp1;
/// %ToReg = add %FromReg ...
/// %Tmp2 = copy %ToReg;
/// MaxLen specifies the maximum length of the copy chain the func
/// can walk through.
bool TwoAddressInstructionPass::isRevCopyChain(Register FromReg, Register ToReg,
                                               int Maxlen) {
  Register TmpReg = FromReg;
  for (int i = 0; i < Maxlen; i++) {
    MachineInstr *Def = getSingleDef(TmpReg, MBB, MRI);
    if (!Def || !Def->isCopy())
      return false;

    TmpReg = Def->getOperand(1).getReg();

    if (TmpReg == ToReg)
      return true;
  }
  return false;
}

/// Return true if there are no intervening uses between the last instruction
/// in the MBB that defines the specified register and the two-address
/// instruction which is being processed. It also returns the last def location
/// by reference.
bool TwoAddressInstructionPass::noUseAfterLastDef(Register Reg, unsigned Dist,
                                                  unsigned &LastDef) {
  LastDef = 0;
  unsigned LastUse = Dist;
  for (MachineOperand &MO : MRI->reg_operands(Reg)) {
    MachineInstr *MI = MO.getParent();
    if (MI->getParent() != MBB || MI->isDebugValue())
      continue;
    DenseMap<MachineInstr*, unsigned>::iterator DI = DistanceMap.find(MI);
    if (DI == DistanceMap.end())
      continue;
    if (MO.isUse() && DI->second < LastUse)
      LastUse = DI->second;
    if (MO.isDef() && DI->second > LastDef)
      LastDef = DI->second;
  }

  return !(LastUse > LastDef && LastUse < Dist);
}

/// Return true if the specified MI is a copy instruction or an extract_subreg
/// instruction. It also returns the source and destination registers and
/// whether they are physical registers by reference.
static bool isCopyToReg(MachineInstr &MI, const TargetInstrInfo *TII,
                        Register &SrcReg, Register &DstReg, bool &IsSrcPhys,
                        bool &IsDstPhys) {
  SrcReg = 0;
  DstReg = 0;
  if (MI.isCopy()) {
    DstReg = MI.getOperand(0).getReg();
    SrcReg = MI.getOperand(1).getReg();
  } else if (MI.isInsertSubreg() || MI.isSubregToReg()) {
    DstReg = MI.getOperand(0).getReg();
    SrcReg = MI.getOperand(2).getReg();
  } else {
    return false;
  }

  IsSrcPhys = SrcReg.isPhysical();
  IsDstPhys = DstReg.isPhysical();
  return true;
}

/// Test if the given register value, which is used by the
/// given instruction, is killed by the given instruction.
static bool isPlainlyKilled(MachineInstr *MI, Register Reg,
                            LiveIntervals *LIS) {
  if (LIS && Reg.isVirtual() && !LIS->isNotInMIMap(*MI)) {
    // FIXME: Sometimes tryInstructionTransform() will add instructions and
    // test whether they can be folded before keeping them. In this case it
    // sets a kill before recursively calling tryInstructionTransform() again.
    // If there is no interval available, we assume that this instruction is
    // one of those. A kill flag is manually inserted on the operand so the
    // check below will handle it.
    LiveInterval &LI = LIS->getInterval(Reg);
    // This is to match the kill flag version where undefs don't have kill
    // flags.
    if (!LI.hasAtLeastOneValue())
      return false;

    SlotIndex useIdx = LIS->getInstructionIndex(*MI);
    LiveInterval::const_iterator I = LI.find(useIdx);
    assert(I != LI.end() && "Reg must be live-in to use.");
    return !I->end.isBlock() && SlotIndex::isSameInstr(I->end, useIdx);
  }

  return MI->killsRegister(Reg);
}

/// Test if the given register value, which is used by the given
/// instruction, is killed by the given instruction. This looks through
/// coalescable copies to see if the original value is potentially not killed.
///
/// For example, in this code:
///
///   %reg1034 = copy %reg1024
///   %reg1035 = copy killed %reg1025
///   %reg1036 = add killed %reg1034, killed %reg1035
///
/// %reg1034 is not considered to be killed, since it is copied from a
/// register which is not killed. Treating it as not killed lets the
/// normal heuristics commute the (two-address) add, which lets
/// coalescing eliminate the extra copy.
///
/// If allowFalsePositives is true then likely kills are treated as kills even
/// if it can't be proven that they are kills.
static bool isKilled(MachineInstr &MI, Register Reg,
                     const MachineRegisterInfo *MRI, const TargetInstrInfo *TII,
                     LiveIntervals *LIS, bool allowFalsePositives) {
  MachineInstr *DefMI = &MI;
  while (true) {
    // All uses of physical registers are likely to be kills.
    if (Reg.isPhysical() && (allowFalsePositives || MRI->hasOneUse(Reg)))
      return true;
    if (!isPlainlyKilled(DefMI, Reg, LIS))
      return false;
    if (Reg.isPhysical())
      return true;
    MachineRegisterInfo::def_iterator Begin = MRI->def_begin(Reg);
    // If there are multiple defs, we can't do a simple analysis, so just
    // go with what the kill flag says.
    if (std::next(Begin) != MRI->def_end())
      return true;
    DefMI = Begin->getParent();
    bool IsSrcPhys, IsDstPhys;
    Register SrcReg, DstReg;
    // If the def is something other than a copy, then it isn't going to
    // be coalesced, so follow the kill flag.
    if (!isCopyToReg(*DefMI, TII, SrcReg, DstReg, IsSrcPhys, IsDstPhys))
      return true;
    Reg = SrcReg;
  }
}

/// Return true if the specified MI uses the specified register as a two-address
/// use. If so, return the destination register by reference.
static bool isTwoAddrUse(MachineInstr &MI, Register Reg, Register &DstReg) {
  for (unsigned i = 0, NumOps = MI.getNumOperands(); i != NumOps; ++i) {
    const MachineOperand &MO = MI.getOperand(i);
    if (!MO.isReg() || !MO.isUse() || MO.getReg() != Reg)
      continue;
    unsigned ti;
    if (MI.isRegTiedToDefOperand(i, &ti)) {
      DstReg = MI.getOperand(ti).getReg();
      return true;
    }
  }
  return false;
}

/// Given a register, if all its uses are in the same basic block, return the
/// last use instruction if it's a copy or a two-address use.
static MachineInstr *
findOnlyInterestingUse(Register Reg, MachineBasicBlock *MBB,
                       MachineRegisterInfo *MRI, const TargetInstrInfo *TII,
                       bool &IsCopy, Register &DstReg, bool &IsDstPhys,
                       LiveIntervals *LIS) {
  MachineOperand *UseOp = nullptr;
  for (MachineOperand &MO : MRI->use_nodbg_operands(Reg)) {
    MachineInstr *MI = MO.getParent();
    if (MI->getParent() != MBB)
      return nullptr;
    if (isPlainlyKilled(MI, Reg, LIS))
      UseOp = &MO;
  }
  if (!UseOp)
    return nullptr;
  MachineInstr &UseMI = *UseOp->getParent();

  Register SrcReg;
  bool IsSrcPhys;
  if (isCopyToReg(UseMI, TII, SrcReg, DstReg, IsSrcPhys, IsDstPhys)) {
    IsCopy = true;
    return &UseMI;
  }
  IsDstPhys = false;
  if (isTwoAddrUse(UseMI, Reg, DstReg)) {
    IsDstPhys = DstReg.isPhysical();
    return &UseMI;
  }
  if (UseMI.isCommutable()) {
    unsigned Src1 = TargetInstrInfo::CommuteAnyOperandIndex;
    unsigned Src2 = UseMI.getOperandNo(UseOp);
    if (TII->findCommutedOpIndices(UseMI, Src1, Src2)) {
      MachineOperand &MO = UseMI.getOperand(Src1);
      if (MO.isReg() && MO.isUse() &&
          isTwoAddrUse(UseMI, MO.getReg(), DstReg)) {
        IsDstPhys = DstReg.isPhysical();
        return &UseMI;
      }
    }
  }
  return nullptr;
}

/// Return the physical register the specified virtual register might be mapped
/// to.
static MCRegister getMappedReg(Register Reg,
                               DenseMap<Register, Register> &RegMap) {
  while (Reg.isVirtual()) {
    DenseMap<Register, Register>::iterator SI = RegMap.find(Reg);
    if (SI == RegMap.end())
      return 0;
    Reg = SI->second;
  }
  if (Reg.isPhysical())
    return Reg;
  return 0;
}

/// Return true if the two registers are equal or aliased.
static bool regsAreCompatible(Register RegA, Register RegB,
                              const TargetRegisterInfo *TRI) {
  if (RegA == RegB)
    return true;
  if (!RegA || !RegB)
    return false;
  return TRI->regsOverlap(RegA, RegB);
}

/// From RegMap remove entries mapped to a physical register which overlaps MO.
static void removeMapRegEntry(const MachineOperand &MO,
                              DenseMap<Register, Register> &RegMap,
                              const TargetRegisterInfo *TRI) {
  assert(
      (MO.isReg() || MO.isRegMask()) &&
      "removeMapRegEntry must be called with a register or regmask operand.");

  SmallVector<Register, 2> Srcs;
  for (auto SI : RegMap) {
    Register ToReg = SI.second;
    if (ToReg.isVirtual())
      continue;

    if (MO.isReg()) {
      Register Reg = MO.getReg();
      if (TRI->regsOverlap(ToReg, Reg))
        Srcs.push_back(SI.first);
    } else if (MO.clobbersPhysReg(ToReg))
      Srcs.push_back(SI.first);
  }

  for (auto SrcReg : Srcs)
    RegMap.erase(SrcReg);
}

/// If a physical register is clobbered, old entries mapped to it should be
/// deleted. For example
///
///     %2:gr64 = COPY killed $rdx
///     MUL64r %3:gr64, implicit-def $rax, implicit-def $rdx
///
/// After the MUL instruction, $rdx contains different value than in the COPY
/// instruction. So %2 should not map to $rdx after MUL.
void TwoAddressInstructionPass::removeClobberedSrcRegMap(MachineInstr *MI) {
  if (MI->isCopy()) {
    // If a virtual register is copied to its mapped physical register, it
    // doesn't change the potential coalescing between them, so we don't remove
    // entries mapped to the physical register. For example
    //
    // %100 = COPY $r8
    //     ...
    // $r8  = COPY %100
    //
    // The first copy constructs SrcRegMap[%100] = $r8, the second copy doesn't
    // destroy the content of $r8, and should not impact SrcRegMap.
    Register Dst = MI->getOperand(0).getReg();
    if (!Dst || Dst.isVirtual())
      return;

    Register Src = MI->getOperand(1).getReg();
    if (regsAreCompatible(Dst, getMappedReg(Src, SrcRegMap), TRI))
      return;
  }

  for (const MachineOperand &MO : MI->operands()) {
    if (MO.isRegMask()) {
      removeMapRegEntry(MO, SrcRegMap, TRI);
      continue;
    }
    if (!MO.isReg() || !MO.isDef())
      continue;
    Register Reg = MO.getReg();
    if (!Reg || Reg.isVirtual())
      continue;
    removeMapRegEntry(MO, SrcRegMap, TRI);
  }
}

// Returns true if Reg is equal or aliased to at least one register in Set.
static bool regOverlapsSet(const SmallVectorImpl<Register> &Set, Register Reg,
                           const TargetRegisterInfo *TRI) {
  for (unsigned R : Set)
    if (TRI->regsOverlap(R, Reg))
      return true;

  return false;
}

/// Return true if it's potentially profitable to commute the two-address
/// instruction that's being processed.
bool TwoAddressInstructionPass::isProfitableToCommute(Register RegA,
                                                      Register RegB,
                                                      Register RegC,
                                                      MachineInstr *MI,
                                                      unsigned Dist) {
  if (OptLevel == CodeGenOpt::None)
    return false;

  // Determine if it's profitable to commute this two address instruction. In
  // general, we want no uses between this instruction and the definition of
  // the two-address register.
  // e.g.
  // %reg1028 = EXTRACT_SUBREG killed %reg1027, 1
  // %reg1029 = COPY %reg1028
  // %reg1029 = SHR8ri %reg1029, 7, implicit dead %eflags
  // insert => %reg1030 = COPY %reg1028
  // %reg1030 = ADD8rr killed %reg1028, killed %reg1029, implicit dead %eflags
  // In this case, it might not be possible to coalesce the second COPY
  // instruction if the first one is coalesced. So it would be profitable to
  // commute it:
  // %reg1028 = EXTRACT_SUBREG killed %reg1027, 1
  // %reg1029 = COPY %reg1028
  // %reg1029 = SHR8ri %reg1029, 7, implicit dead %eflags
  // insert => %reg1030 = COPY %reg1029
  // %reg1030 = ADD8rr killed %reg1029, killed %reg1028, implicit dead %eflags

  if (!isPlainlyKilled(MI, RegC, LIS))
    return false;

  // Ok, we have something like:
  // %reg1030 = ADD8rr killed %reg1028, killed %reg1029, implicit dead %eflags
  // let's see if it's worth commuting it.

  // Look for situations like this:
  // %reg1024 = MOV r1
  // %reg1025 = MOV r0
  // %reg1026 = ADD %reg1024, %reg1025
  // r0            = MOV %reg1026
  // Commute the ADD to hopefully eliminate an otherwise unavoidable copy.
  MCRegister ToRegA = getMappedReg(RegA, DstRegMap);
  if (ToRegA) {
    MCRegister FromRegB = getMappedReg(RegB, SrcRegMap);
    MCRegister FromRegC = getMappedReg(RegC, SrcRegMap);
    bool CompB = FromRegB && regsAreCompatible(FromRegB, ToRegA, TRI);
    bool CompC = FromRegC && regsAreCompatible(FromRegC, ToRegA, TRI);

    // Compute if any of the following are true:
    // -RegB is not tied to a register and RegC is compatible with RegA.
    // -RegB is tied to the wrong physical register, but RegC is.
    // -RegB is tied to the wrong physical register, and RegC isn't tied.
    if ((!FromRegB && CompC) || (FromRegB && !CompB && (!FromRegC || CompC)))
      return true;
    // Don't compute if any of the following are true:
    // -RegC is not tied to a register and RegB is compatible with RegA.
    // -RegC is tied to the wrong physical register, but RegB is.
    // -RegC is tied to the wrong physical register, and RegB isn't tied.
    if ((!FromRegC && CompB) || (FromRegC && !CompC && (!FromRegB || CompB)))
      return false;
  }

  // If there is a use of RegC between its last def (could be livein) and this
  // instruction, then bail.
  unsigned LastDefC = 0;
  if (!noUseAfterLastDef(RegC, Dist, LastDefC))
    return false;

  // If there is a use of RegB between its last def (could be livein) and this
  // instruction, then go ahead and make this transformation.
  unsigned LastDefB = 0;
  if (!noUseAfterLastDef(RegB, Dist, LastDefB))
    return true;

  // Look for situation like this:
  // %reg101 = MOV %reg100
  // %reg102 = ...
  // %reg103 = ADD %reg102, %reg101
  // ... = %reg103 ...
  // %reg100 = MOV %reg103
  // If there is a reversed copy chain from reg101 to reg103, commute the ADD
  // to eliminate an otherwise unavoidable copy.
  // FIXME:
  // We can extend the logic further: If an pair of operands in an insn has
  // been merged, the insn could be regarded as a virtual copy, and the virtual
  // copy could also be used to construct a copy chain.
  // To more generally minimize register copies, ideally the logic of two addr
  // instruction pass should be integrated with register allocation pass where
  // interference graph is available.
  if (isRevCopyChain(RegC, RegA, MaxDataFlowEdge))
    return true;

  if (isRevCopyChain(RegB, RegA, MaxDataFlowEdge))
    return false;

  // Look for other target specific commute preference.
  bool Commute;
  if (TII->hasCommutePreference(*MI, Commute))
    return Commute;

  // Since there are no intervening uses for both registers, then commute
  // if the def of RegC is closer. Its live interval is shorter.
  return LastDefB && LastDefC && LastDefC > LastDefB;
}

/// Commute a two-address instruction and update the basic block, distance map,
/// and live variables if needed. Return true if it is successful.
bool TwoAddressInstructionPass::commuteInstruction(MachineInstr *MI,
                                                   unsigned DstIdx,
                                                   unsigned RegBIdx,
                                                   unsigned RegCIdx,
                                                   unsigned Dist) {
  Register RegC = MI->getOperand(RegCIdx).getReg();
  LLVM_DEBUG(dbgs() << "2addr: COMMUTING  : " << *MI);
  MachineInstr *NewMI = TII->commuteInstruction(*MI, false, RegBIdx, RegCIdx);

  if (NewMI == nullptr) {
    LLVM_DEBUG(dbgs() << "2addr: COMMUTING FAILED!\n");
    return false;
  }

  LLVM_DEBUG(dbgs() << "2addr: COMMUTED TO: " << *NewMI);
  assert(NewMI == MI &&
         "TargetInstrInfo::commuteInstruction() should not return a new "
         "instruction unless it was requested.");

  // Update source register map.
  MCRegister FromRegC = getMappedReg(RegC, SrcRegMap);
  if (FromRegC) {
    Register RegA = MI->getOperand(DstIdx).getReg();
    SrcRegMap[RegA] = FromRegC;
  }

  return true;
}

/// Return true if it is profitable to convert the given 2-address instruction
/// to a 3-address one.
bool TwoAddressInstructionPass::isProfitableToConv3Addr(Register RegA,
                                                        Register RegB) {
  // Look for situations like this:
  // %reg1024 = MOV r1
  // %reg1025 = MOV r0
  // %reg1026 = ADD %reg1024, %reg1025
  // r2            = MOV %reg1026
  // Turn ADD into a 3-address instruction to avoid a copy.
  MCRegister FromRegB = getMappedReg(RegB, SrcRegMap);
  if (!FromRegB)
    return false;
  MCRegister ToRegA = getMappedReg(RegA, DstRegMap);
  return (ToRegA && !regsAreCompatible(FromRegB, ToRegA, TRI));
}

/// Convert the specified two-address instruction into a three address one.
/// Return true if this transformation was successful.
bool TwoAddressInstructionPass::convertInstTo3Addr(
    MachineBasicBlock::iterator &mi, MachineBasicBlock::iterator &nmi,
    Register RegA, Register RegB, unsigned &Dist) {
  MachineInstrSpan MIS(mi, MBB);
  MachineInstr *NewMI = TII->convertToThreeAddress(*mi, LV, LIS);
  if (!NewMI)
    return false;

  LLVM_DEBUG(dbgs() << "2addr: CONVERTING 2-ADDR: " << *mi);
  LLVM_DEBUG(dbgs() << "2addr:         TO 3-ADDR: " << *NewMI);

  // If the old instruction is debug value tracked, an update is required.
  if (auto OldInstrNum = mi->peekDebugInstrNum()) {
    assert(mi->getNumExplicitDefs() == 1);
    assert(NewMI->getNumExplicitDefs() == 1);

    // Find the old and new def location.
    auto OldIt = mi->defs().begin();
    auto NewIt = NewMI->defs().begin();
    unsigned OldIdx = mi->getOperandNo(OldIt);
    unsigned NewIdx = NewMI->getOperandNo(NewIt);

    // Record that one def has been replaced by the other.
    unsigned NewInstrNum = NewMI->getDebugInstrNum();
    MF->makeDebugValueSubstitution(std::make_pair(OldInstrNum, OldIdx),
                                   std::make_pair(NewInstrNum, NewIdx));
  }

  MBB->erase(mi); // Nuke the old inst.

  for (MachineInstr &MI : MIS)
    DistanceMap.insert(std::make_pair(&MI, Dist++));
  Dist--;
  mi = NewMI;
  nmi = std::next(mi);

  // Update source and destination register maps.
  SrcRegMap.erase(RegA);
  DstRegMap.erase(RegB);
  return true;
}

/// Scan forward recursively for only uses, update maps if the use is a copy or
/// a two-address instruction.
void TwoAddressInstructionPass::scanUses(Register DstReg) {
  SmallVector<Register, 4> VirtRegPairs;
  bool IsDstPhys;
  bool IsCopy = false;
  Register NewReg;
  Register Reg = DstReg;
  while (MachineInstr *UseMI = findOnlyInterestingUse(Reg, MBB, MRI, TII,IsCopy,
                                                      NewReg, IsDstPhys, LIS)) {
    if (IsCopy && !Processed.insert(UseMI).second)
      break;

    DenseMap<MachineInstr*, unsigned>::iterator DI = DistanceMap.find(UseMI);
    if (DI != DistanceMap.end())
      // Earlier in the same MBB.Reached via a back edge.
      break;

    if (IsDstPhys) {
      VirtRegPairs.push_back(NewReg);
      break;
    }
    SrcRegMap[NewReg] = Reg;
    VirtRegPairs.push_back(NewReg);
    Reg = NewReg;
  }

  if (!VirtRegPairs.empty()) {
    unsigned ToReg = VirtRegPairs.back();
    VirtRegPairs.pop_back();
    while (!VirtRegPairs.empty()) {
      unsigned FromReg = VirtRegPairs.pop_back_val();
      bool isNew = DstRegMap.insert(std::make_pair(FromReg, ToReg)).second;
      if (!isNew)
        assert(DstRegMap[FromReg] == ToReg &&"Can't map to two dst registers!");
      ToReg = FromReg;
    }
    bool isNew = DstRegMap.insert(std::make_pair(DstReg, ToReg)).second;
    if (!isNew)
      assert(DstRegMap[DstReg] == ToReg && "Can't map to two dst registers!");
  }
}

/// If the specified instruction is not yet processed, process it if it's a
/// copy. For a copy instruction, we find the physical registers the
/// source and destination registers might be mapped to. These are kept in
/// point-to maps used to determine future optimizations. e.g.
/// v1024 = mov r0
/// v1025 = mov r1
/// v1026 = add v1024, v1025
/// r1    = mov r1026
/// If 'add' is a two-address instruction, v1024, v1026 are both potentially
/// coalesced to r0 (from the input side). v1025 is mapped to r1. v1026 is
/// potentially joined with r1 on the output side. It's worthwhile to commute
/// 'add' to eliminate a copy.
void TwoAddressInstructionPass::processCopy(MachineInstr *MI) {
  if (Processed.count(MI))
    return;

  bool IsSrcPhys, IsDstPhys;
  Register SrcReg, DstReg;
  if (!isCopyToReg(*MI, TII, SrcReg, DstReg, IsSrcPhys, IsDstPhys))
    return;

  if (IsDstPhys && !IsSrcPhys) {
    DstRegMap.insert(std::make_pair(SrcReg, DstReg));
  } else if (!IsDstPhys && IsSrcPhys) {
    bool isNew = SrcRegMap.insert(std::make_pair(DstReg, SrcReg)).second;
    if (!isNew)
      assert(SrcRegMap[DstReg] == SrcReg &&
             "Can't map to two src physical registers!");

    scanUses(DstReg);
  }

  Processed.insert(MI);
}

/// If there is one more local instruction that reads 'Reg' and it kills 'Reg,
/// consider moving the instruction below the kill instruction in order to
/// eliminate the need for the copy.
bool TwoAddressInstructionPass::rescheduleMIBelowKill(
    MachineBasicBlock::iterator &mi, MachineBasicBlock::iterator &nmi,
    Register Reg) {
  // Bail immediately if we don't have LV or LIS available. We use them to find
  // kills efficiently.
  if (!LV && !LIS)
    return false;

  MachineInstr *MI = &*mi;
  DenseMap<MachineInstr*, unsigned>::iterator DI = DistanceMap.find(MI);
  if (DI == DistanceMap.end())
    // Must be created from unfolded load. Don't waste time trying this.
    return false;

  MachineInstr *KillMI = nullptr;
  if (LIS) {
    LiveInterval &LI = LIS->getInterval(Reg);
    assert(LI.end() != LI.begin() &&
           "Reg should not have empty live interval.");

    SlotIndex MBBEndIdx = LIS->getMBBEndIdx(MBB).getPrevSlot();
    LiveInterval::const_iterator I = LI.find(MBBEndIdx);
    if (I != LI.end() && I->start < MBBEndIdx)
      return false;

    --I;
    KillMI = LIS->getInstructionFromIndex(I->end);
  } else {
    KillMI = LV->getVarInfo(Reg).findKill(MBB);
  }
  if (!KillMI || MI == KillMI || KillMI->isCopy() || KillMI->isCopyLike())
    // Don't mess with copies, they may be coalesced later.
    return false;

  if (KillMI->hasUnmodeledSideEffects() || KillMI->isCall() ||
      KillMI->isBranch() || KillMI->isTerminator())
    // Don't move pass calls, etc.
    return false;

  Register DstReg;
  if (isTwoAddrUse(*KillMI, Reg, DstReg))
    return false;

  bool SeenStore = true;
  if (!MI->isSafeToMove(AA, SeenStore))
    return false;

  if (TII->getInstrLatency(InstrItins, *MI) > 1)
    // FIXME: Needs more sophisticated heuristics.
    return false;

  SmallVector<Register, 2> Uses;
  SmallVector<Register, 2> Kills;
  SmallVector<Register, 2> Defs;
  for (const MachineOperand &MO : MI->operands()) {
    if (!MO.isReg())
      continue;
    Register MOReg = MO.getReg();
    if (!MOReg)
      continue;
    if (MO.isDef())
      Defs.push_back(MOReg);
    else {
      Uses.push_back(MOReg);
      if (MOReg != Reg && (MO.isKill() ||
                           (LIS && isPlainlyKilled(MI, MOReg, LIS))))
        Kills.push_back(MOReg);
    }
  }

  // Move the copies connected to MI down as well.
  MachineBasicBlock::iterator Begin = MI;
  MachineBasicBlock::iterator AfterMI = std::next(Begin);
  MachineBasicBlock::iterator End = AfterMI;
  while (End != MBB->end()) {
    End = skipDebugInstructionsForward(End, MBB->end());
    if (End->isCopy() && regOverlapsSet(Defs, End->getOperand(1).getReg(), TRI))
      Defs.push_back(End->getOperand(0).getReg());
    else
      break;
    ++End;
  }

  // Check if the reschedule will not break dependencies.
  unsigned NumVisited = 0;
  MachineBasicBlock::iterator KillPos = KillMI;
  ++KillPos;
  for (MachineInstr &OtherMI : make_range(End, KillPos)) {
    // Debug or pseudo instructions cannot be counted against the limit.
    if (OtherMI.isDebugOrPseudoInstr())
      continue;
    if (NumVisited > 10)  // FIXME: Arbitrary limit to reduce compile time cost.
      return false;
    ++NumVisited;
    if (OtherMI.hasUnmodeledSideEffects() || OtherMI.isCall() ||
        OtherMI.isBranch() || OtherMI.isTerminator())
      // Don't move pass calls, etc.
      return false;
    for (const MachineOperand &MO : OtherMI.operands()) {
      if (!MO.isReg())
        continue;
      Register MOReg = MO.getReg();
      if (!MOReg)
        continue;
      if (MO.isDef()) {
        if (regOverlapsSet(Uses, MOReg, TRI))
          // Physical register use would be clobbered.
          return false;
        if (!MO.isDead() && regOverlapsSet(Defs, MOReg, TRI))
          // May clobber a physical register def.
          // FIXME: This may be too conservative. It's ok if the instruction
          // is sunken completely below the use.
          return false;
      } else {
        if (regOverlapsSet(Defs, MOReg, TRI))
          return false;
        bool isKill =
            MO.isKill() || (LIS && isPlainlyKilled(&OtherMI, MOReg, LIS));
        if (MOReg != Reg && ((isKill && regOverlapsSet(Uses, MOReg, TRI)) ||
                             regOverlapsSet(Kills, MOReg, TRI)))
          // Don't want to extend other live ranges and update kills.
          return false;
        if (MOReg == Reg && !isKill)
          // We can't schedule across a use of the register in question.
          return false;
        // Ensure that if this is register in question, its the kill we expect.
        assert((MOReg != Reg || &OtherMI == KillMI) &&
               "Found multiple kills of a register in a basic block");
      }
    }
  }

  // Move debug info as well.
  while (Begin != MBB->begin() && std::prev(Begin)->isDebugInstr())
    --Begin;

  nmi = End;
  MachineBasicBlock::iterator InsertPos = KillPos;
  if (LIS) {
    // We have to move the copies (and any interleaved debug instructions)
    // first so that the MBB is still well-formed when calling handleMove().
    for (MachineBasicBlock::iterator MBBI = AfterMI; MBBI != End;) {
      auto CopyMI = MBBI++;
      MBB->splice(InsertPos, MBB, CopyMI);
      if (!CopyMI->isDebugOrPseudoInstr())
        LIS->handleMove(*CopyMI);
      InsertPos = CopyMI;
    }
    End = std::next(MachineBasicBlock::iterator(MI));
  }

  // Copies following MI may have been moved as well.
  MBB->splice(InsertPos, MBB, Begin, End);
  DistanceMap.erase(DI);

  // Update live variables
  if (LIS) {
    LIS->handleMove(*MI);
  } else {
    LV->removeVirtualRegisterKilled(Reg, *KillMI);
    LV->addVirtualRegisterKilled(Reg, *MI);
  }

  LLVM_DEBUG(dbgs() << "\trescheduled below kill: " << *KillMI);
  return true;
}

/// Return true if the re-scheduling will put the given instruction too close
/// to the defs of its register dependencies.
bool TwoAddressInstructionPass::isDefTooClose(Register Reg, unsigned Dist,
                                              MachineInstr *MI) {
  for (MachineInstr &DefMI : MRI->def_instructions(Reg)) {
    if (DefMI.getParent() != MBB || DefMI.isCopy() || DefMI.isCopyLike())
      continue;
    if (&DefMI == MI)
      return true; // MI is defining something KillMI uses
    DenseMap<MachineInstr*, unsigned>::iterator DDI = DistanceMap.find(&DefMI);
    if (DDI == DistanceMap.end())
      return true;  // Below MI
    unsigned DefDist = DDI->second;
    assert(Dist > DefDist && "Visited def already?");
    if (TII->getInstrLatency(InstrItins, DefMI) > (Dist - DefDist))
      return true;
  }
  return false;
}

/// If there is one more local instruction that reads 'Reg' and it kills 'Reg,
/// consider moving the kill instruction above the current two-address
/// instruction in order to eliminate the need for the copy.
bool TwoAddressInstructionPass::rescheduleKillAboveMI(
    MachineBasicBlock::iterator &mi, MachineBasicBlock::iterator &nmi,
    Register Reg) {
  // Bail immediately if we don't have LV or LIS available. We use them to find
  // kills efficiently.
  if (!LV && !LIS)
    return false;

  MachineInstr *MI = &*mi;
  DenseMap<MachineInstr*, unsigned>::iterator DI = DistanceMap.find(MI);
  if (DI == DistanceMap.end())
    // Must be created from unfolded load. Don't waste time trying this.
    return false;

  MachineInstr *KillMI = nullptr;
  if (LIS) {
    LiveInterval &LI = LIS->getInterval(Reg);
    assert(LI.end() != LI.begin() &&
           "Reg should not have empty live interval.");

    SlotIndex MBBEndIdx = LIS->getMBBEndIdx(MBB).getPrevSlot();
    LiveInterval::const_iterator I = LI.find(MBBEndIdx);
    if (I != LI.end() && I->start < MBBEndIdx)
      return false;

    --I;
    KillMI = LIS->getInstructionFromIndex(I->end);
  } else {
    KillMI = LV->getVarInfo(Reg).findKill(MBB);
  }
  if (!KillMI || MI == KillMI || KillMI->isCopy() || KillMI->isCopyLike())
    // Don't mess with copies, they may be coalesced later.
    return false;

  Register DstReg;
  if (isTwoAddrUse(*KillMI, Reg, DstReg))
    return false;

  bool SeenStore = true;
  if (!KillMI->isSafeToMove(AA, SeenStore))
    return false;

  SmallVector<Register, 2> Uses;
  SmallVector<Register, 2> Kills;
  SmallVector<Register, 2> Defs;
  SmallVector<Register, 2> LiveDefs;
  for (const MachineOperand &MO : KillMI->operands()) {
    if (!MO.isReg())
      continue;
    Register MOReg = MO.getReg();
    if (MO.isUse()) {
      if (!MOReg)
        continue;
      if (isDefTooClose(MOReg, DI->second, MI))
        return false;
      bool isKill = MO.isKill() || (LIS && isPlainlyKilled(KillMI, MOReg, LIS));
      if (MOReg == Reg && !isKill)
        return false;
      Uses.push_back(MOReg);
      if (isKill && MOReg != Reg)
        Kills.push_back(MOReg);
    } else if (MOReg.isPhysical()) {
      Defs.push_back(MOReg);
      if (!MO.isDead())
        LiveDefs.push_back(MOReg);
    }
  }

  // Check if the reschedule will not break depedencies.
  unsigned NumVisited = 0;
  for (MachineInstr &OtherMI :
       make_range(mi, MachineBasicBlock::iterator(KillMI))) {
    // Debug or pseudo instructions cannot be counted against the limit.
    if (OtherMI.isDebugOrPseudoInstr())
      continue;
    if (NumVisited > 10)  // FIXME: Arbitrary limit to reduce compile time cost.
      return false;
    ++NumVisited;
    if (OtherMI.hasUnmodeledSideEffects() || OtherMI.isCall() ||
        OtherMI.isBranch() || OtherMI.isTerminator())
      // Don't move pass calls, etc.
      return false;
    SmallVector<Register, 2> OtherDefs;
    for (const MachineOperand &MO : OtherMI.operands()) {
      if (!MO.isReg())
        continue;
      Register MOReg = MO.getReg();
      if (!MOReg)
        continue;
      if (MO.isUse()) {
        if (regOverlapsSet(Defs, MOReg, TRI))
          // Moving KillMI can clobber the physical register if the def has
          // not been seen.
          return false;
        if (regOverlapsSet(Kills, MOReg, TRI))
          // Don't want to extend other live ranges and update kills.
          return false;
        if (&OtherMI != MI && MOReg == Reg &&
            !(MO.isKill() || (LIS && isPlainlyKilled(&OtherMI, MOReg, LIS))))
          // We can't schedule across a use of the register in question.
          return false;
      } else {
        OtherDefs.push_back(MOReg);
      }
    }

    for (unsigned i = 0, e = OtherDefs.size(); i != e; ++i) {
      Register MOReg = OtherDefs[i];
      if (regOverlapsSet(Uses, MOReg, TRI))
        return false;
      if (MOReg.isPhysical() && regOverlapsSet(LiveDefs, MOReg, TRI))
        return false;
      // Physical register def is seen.
      llvm::erase_value(Defs, MOReg);
    }
  }

  // Move the old kill above MI, don't forget to move debug info as well.
  MachineBasicBlock::iterator InsertPos = mi;
  while (InsertPos != MBB->begin() && std::prev(InsertPos)->isDebugInstr())
    --InsertPos;
  MachineBasicBlock::iterator From = KillMI;
  MachineBasicBlock::iterator To = std::next(From);
  while (std::prev(From)->isDebugInstr())
    --From;
  MBB->splice(InsertPos, MBB, From, To);

  nmi = std::prev(InsertPos); // Backtrack so we process the moved instr.
  DistanceMap.erase(DI);

  // Update live variables
  if (LIS) {
    LIS->handleMove(*KillMI);
  } else {
    LV->removeVirtualRegisterKilled(Reg, *KillMI);
    LV->addVirtualRegisterKilled(Reg, *MI);
  }

  LLVM_DEBUG(dbgs() << "\trescheduled kill: " << *KillMI);
  return true;
}

/// Tries to commute the operand 'BaseOpIdx' and some other operand in the
/// given machine instruction to improve opportunities for coalescing and
/// elimination of a register to register copy.
///
/// 'DstOpIdx' specifies the index of MI def operand.
/// 'BaseOpKilled' specifies if the register associated with 'BaseOpIdx'
/// operand is killed by the given instruction.
/// The 'Dist' arguments provides the distance of MI from the start of the
/// current basic block and it is used to determine if it is profitable
/// to commute operands in the instruction.
///
/// Returns true if the transformation happened. Otherwise, returns false.
bool TwoAddressInstructionPass::tryInstructionCommute(MachineInstr *MI,
                                                      unsigned DstOpIdx,
                                                      unsigned BaseOpIdx,
                                                      bool BaseOpKilled,
                                                      unsigned Dist) {
  if (!MI->isCommutable())
    return false;

  bool MadeChange = false;
  Register DstOpReg = MI->getOperand(DstOpIdx).getReg();
  Register BaseOpReg = MI->getOperand(BaseOpIdx).getReg();
  unsigned OpsNum = MI->getDesc().getNumOperands();
  unsigned OtherOpIdx = MI->getDesc().getNumDefs();
  for (; OtherOpIdx < OpsNum; OtherOpIdx++) {
    // The call of findCommutedOpIndices below only checks if BaseOpIdx
    // and OtherOpIdx are commutable, it does not really search for
    // other commutable operands and does not change the values of passed
    // variables.
    if (OtherOpIdx == BaseOpIdx || !MI->getOperand(OtherOpIdx).isReg() ||
        !TII->findCommutedOpIndices(*MI, BaseOpIdx, OtherOpIdx))
      continue;

    Register OtherOpReg = MI->getOperand(OtherOpIdx).getReg();
    bool AggressiveCommute = false;

    // If OtherOp dies but BaseOp does not, swap the OtherOp and BaseOp
    // operands. This makes the live ranges of DstOp and OtherOp joinable.
    bool OtherOpKilled = isKilled(*MI, OtherOpReg, MRI, TII, LIS, false);
    bool DoCommute = !BaseOpKilled && OtherOpKilled;

    if (!DoCommute &&
        isProfitableToCommute(DstOpReg, BaseOpReg, OtherOpReg, MI, Dist)) {
      DoCommute = true;
      AggressiveCommute = true;
    }

    // If it's profitable to commute, try to do so.
    if (DoCommute && commuteInstruction(MI, DstOpIdx, BaseOpIdx, OtherOpIdx,
                                        Dist)) {
      MadeChange = true;
      ++NumCommuted;
      if (AggressiveCommute)
        ++NumAggrCommuted;

      // There might be more than two commutable operands, update BaseOp and
      // continue scanning.
      // FIXME: This assumes that the new instruction's operands are in the
      // same positions and were simply swapped.
      BaseOpReg = OtherOpReg;
      BaseOpKilled = OtherOpKilled;
      // Resamples OpsNum in case the number of operands was reduced. This
      // happens with X86.
      OpsNum = MI->getDesc().getNumOperands();
    }
  }
  return MadeChange;
}

/// For the case where an instruction has a single pair of tied register
/// operands, attempt some transformations that may either eliminate the tied
/// operands or improve the opportunities for coalescing away the register copy.
/// Returns true if no copy needs to be inserted to untie mi's operands
/// (either because they were untied, or because mi was rescheduled, and will
/// be visited again later). If the shouldOnlyCommute flag is true, only
/// instruction commutation is attempted.
bool TwoAddressInstructionPass::
tryInstructionTransform(MachineBasicBlock::iterator &mi,
                        MachineBasicBlock::iterator &nmi,
                        unsigned SrcIdx, unsigned DstIdx,
                        unsigned &Dist, bool shouldOnlyCommute) {
  if (OptLevel == CodeGenOpt::None)
    return false;

  MachineInstr &MI = *mi;
  Register regA = MI.getOperand(DstIdx).getReg();
  Register regB = MI.getOperand(SrcIdx).getReg();

  assert(regB.isVirtual() && "cannot make instruction into two-address form");
  bool regBKilled = isKilled(MI, regB, MRI, TII, LIS, true);

  if (regA.isVirtual())
    scanUses(regA);

  bool Commuted = tryInstructionCommute(&MI, DstIdx, SrcIdx, regBKilled, Dist);

  // If the instruction is convertible to 3 Addr, instead
  // of returning try 3 Addr transformation aggressively and
  // use this variable to check later. Because it might be better.
  // For example, we can just use `leal (%rsi,%rdi), %eax` and `ret`
  // instead of the following code.
  //   addl     %esi, %edi
  //   movl     %edi, %eax
  //   ret
  if (Commuted && !MI.isConvertibleTo3Addr())
    return false;

  if (shouldOnlyCommute)
    return false;

  // If there is one more use of regB later in the same MBB, consider
  // re-schedule this MI below it.
  if (!Commuted && EnableRescheduling && rescheduleMIBelowKill(mi, nmi, regB)) {
    ++NumReSchedDowns;
    return true;
  }

  // If we commuted, regB may have changed so we should re-sample it to avoid
  // confusing the three address conversion below.
  if (Commuted) {
    regB = MI.getOperand(SrcIdx).getReg();
    regBKilled = isKilled(MI, regB, MRI, TII, LIS, true);
  }

  if (MI.isConvertibleTo3Addr()) {
    // This instruction is potentially convertible to a true
    // three-address instruction.  Check if it is profitable.
    if (!regBKilled || isProfitableToConv3Addr(regA, regB)) {
      // Try to convert it.
      if (convertInstTo3Addr(mi, nmi, regA, regB, Dist)) {
        ++NumConvertedTo3Addr;
        return true; // Done with this instruction.
      }
    }
  }

  // Return if it is commuted but 3 addr conversion is failed.
  if (Commuted)
    return false;

  // If there is one more use of regB later in the same MBB, consider
  // re-schedule it before this MI if it's legal.
  if (EnableRescheduling && rescheduleKillAboveMI(mi, nmi, regB)) {
    ++NumReSchedUps;
    return true;
  }

  // If this is an instruction with a load folded into it, try unfolding
  // the load, e.g. avoid this:
  //   movq %rdx, %rcx
  //   addq (%rax), %rcx
  // in favor of this:
  //   movq (%rax), %rcx
  //   addq %rdx, %rcx
  // because it's preferable to schedule a load than a register copy.
  if (MI.mayLoad() && !regBKilled) {
    // Determine if a load can be unfolded.
    unsigned LoadRegIndex;
    unsigned NewOpc =
      TII->getOpcodeAfterMemoryUnfold(MI.getOpcode(),
                                      /*UnfoldLoad=*/true,
                                      /*UnfoldStore=*/false,
                                      &LoadRegIndex);
    if (NewOpc != 0) {
      const MCInstrDesc &UnfoldMCID = TII->get(NewOpc);
      if (UnfoldMCID.getNumDefs() == 1) {
        // Unfold the load.
        LLVM_DEBUG(dbgs() << "2addr:   UNFOLDING: " << MI);
        const TargetRegisterClass *RC =
          TRI->getAllocatableClass(
            TII->getRegClass(UnfoldMCID, LoadRegIndex, TRI, *MF));
        Register Reg = MRI->createVirtualRegister(RC);
        SmallVector<MachineInstr *, 2> NewMIs;
        if (!TII->unfoldMemoryOperand(*MF, MI, Reg,
                                      /*UnfoldLoad=*/true,
                                      /*UnfoldStore=*/false, NewMIs)) {
          LLVM_DEBUG(dbgs() << "2addr: ABANDONING UNFOLD\n");
          return false;
        }
        assert(NewMIs.size() == 2 &&
               "Unfolded a load into multiple instructions!");
        // The load was previously folded, so this is the only use.
        NewMIs[1]->addRegisterKilled(Reg, TRI);

        // Tentatively insert the instructions into the block so that they
        // look "normal" to the transformation logic.
        MBB->insert(mi, NewMIs[0]);
        MBB->insert(mi, NewMIs[1]);
        DistanceMap.insert(std::make_pair(NewMIs[0], Dist++));
        DistanceMap.insert(std::make_pair(NewMIs[1], Dist));

        LLVM_DEBUG(dbgs() << "2addr:    NEW LOAD: " << *NewMIs[0]
                          << "2addr:    NEW INST: " << *NewMIs[1]);

        // Transform the instruction, now that it no longer has a load.
        unsigned NewDstIdx = NewMIs[1]->findRegisterDefOperandIdx(regA);
        unsigned NewSrcIdx = NewMIs[1]->findRegisterUseOperandIdx(regB);
        MachineBasicBlock::iterator NewMI = NewMIs[1];
        bool TransformResult =
          tryInstructionTransform(NewMI, mi, NewSrcIdx, NewDstIdx, Dist, true);
        (void)TransformResult;
        assert(!TransformResult &&
               "tryInstructionTransform() should return false.");
        if (NewMIs[1]->getOperand(NewSrcIdx).isKill()) {
          // Success, or at least we made an improvement. Keep the unfolded
          // instructions and discard the original.
          if (LV) {
            for (const MachineOperand &MO : MI.operands()) {
              if (MO.isReg() && MO.getReg().isVirtual()) {
                if (MO.isUse()) {
                  if (MO.isKill()) {
                    if (NewMIs[0]->killsRegister(MO.getReg()))
                      LV->replaceKillInstruction(MO.getReg(), MI, *NewMIs[0]);
                    else {
                      assert(NewMIs[1]->killsRegister(MO.getReg()) &&
                             "Kill missing after load unfold!");
                      LV->replaceKillInstruction(MO.getReg(), MI, *NewMIs[1]);
                    }
                  }
                } else if (LV->removeVirtualRegisterDead(MO.getReg(), MI)) {
                  if (NewMIs[1]->registerDefIsDead(MO.getReg()))
                    LV->addVirtualRegisterDead(MO.getReg(), *NewMIs[1]);
                  else {
                    assert(NewMIs[0]->registerDefIsDead(MO.getReg()) &&
                           "Dead flag missing after load unfold!");
                    LV->addVirtualRegisterDead(MO.getReg(), *NewMIs[0]);
                  }
                }
              }
            }
            LV->addVirtualRegisterKilled(Reg, *NewMIs[1]);
          }

          SmallVector<Register, 4> OrigRegs;
          if (LIS) {
            for (const MachineOperand &MO : MI.operands()) {
              if (MO.isReg())
                OrigRegs.push_back(MO.getReg());
            }

            LIS->RemoveMachineInstrFromMaps(MI);
          }

          MI.eraseFromParent();
          DistanceMap.erase(&MI);

          // Update LiveIntervals.
          if (LIS) {
            MachineBasicBlock::iterator Begin(NewMIs[0]);
            MachineBasicBlock::iterator End(NewMIs[1]);
            LIS->repairIntervalsInRange(MBB, Begin, End, OrigRegs);
          }

          mi = NewMIs[1];
        } else {
          // Transforming didn't eliminate the tie and didn't lead to an
          // improvement. Clean up the unfolded instructions and keep the
          // original.
          LLVM_DEBUG(dbgs() << "2addr: ABANDONING UNFOLD\n");
          NewMIs[0]->eraseFromParent();
          NewMIs[1]->eraseFromParent();
          DistanceMap.erase(NewMIs[0]);
          DistanceMap.erase(NewMIs[1]);
          Dist--;
        }
      }
    }
  }

  return false;
}

// Collect tied operands of MI that need to be handled.
// Rewrite trivial cases immediately.
// Return true if any tied operands where found, including the trivial ones.
bool TwoAddressInstructionPass::
collectTiedOperands(MachineInstr *MI, TiedOperandMap &TiedOperands) {
  bool AnyOps = false;
  unsigned NumOps = MI->getNumOperands();

  for (unsigned SrcIdx = 0; SrcIdx < NumOps; ++SrcIdx) {
    unsigned DstIdx = 0;
    if (!MI->isRegTiedToDefOperand(SrcIdx, &DstIdx))
      continue;
    AnyOps = true;
    MachineOperand &SrcMO = MI->getOperand(SrcIdx);
    MachineOperand &DstMO = MI->getOperand(DstIdx);
    Register SrcReg = SrcMO.getReg();
    Register DstReg = DstMO.getReg();
    // Tied constraint already satisfied?
    if (SrcReg == DstReg)
      continue;

    assert(SrcReg && SrcMO.isUse() && "two address instruction invalid");

    // Deal with undef uses immediately - simply rewrite the src operand.
    if (SrcMO.isUndef() && !DstMO.getSubReg()) {
      // Constrain the DstReg register class if required.
      if (DstReg.isVirtual()) {
        const TargetRegisterClass *RC = MRI->getRegClass(SrcReg);
        MRI->constrainRegClass(DstReg, RC);
      }
      SrcMO.setReg(DstReg);
      SrcMO.setSubReg(0);
      LLVM_DEBUG(dbgs() << "\t\trewrite undef:\t" << *MI);
      continue;
    }
    TiedOperands[SrcReg].push_back(std::make_pair(SrcIdx, DstIdx));
  }
  return AnyOps;
}

// Process a list of tied MI operands that all use the same source register.
// The tied pairs are of the form (SrcIdx, DstIdx).
void
TwoAddressInstructionPass::processTiedPairs(MachineInstr *MI,
                                            TiedPairList &TiedPairs,
                                            unsigned &Dist) {
  bool IsEarlyClobber = llvm::any_of(TiedPairs, [MI](auto const &TP) {
    return MI->getOperand(TP.second).isEarlyClobber();
  });

  bool RemovedKillFlag = false;
  bool AllUsesCopied = true;
  unsigned LastCopiedReg = 0;
  SlotIndex LastCopyIdx;
  Register RegB = 0;
  unsigned SubRegB = 0;
  for (auto &TP : TiedPairs) {
    unsigned SrcIdx = TP.first;
    unsigned DstIdx = TP.second;

    const MachineOperand &DstMO = MI->getOperand(DstIdx);
    Register RegA = DstMO.getReg();

    // Grab RegB from the instruction because it may have changed if the
    // instruction was commuted.
    RegB = MI->getOperand(SrcIdx).getReg();
    SubRegB = MI->getOperand(SrcIdx).getSubReg();

    if (RegA == RegB) {
      // The register is tied to multiple destinations (or else we would
      // not have continued this far), but this use of the register
      // already matches the tied destination.  Leave it.
      AllUsesCopied = false;
      continue;
    }
    LastCopiedReg = RegA;

    assert(RegB.isVirtual() && "cannot make instruction into two-address form");

#ifndef NDEBUG
    // First, verify that we don't have a use of "a" in the instruction
    // (a = b + a for example) because our transformation will not
    // work. This should never occur because we are in SSA form.
    for (unsigned i = 0; i != MI->getNumOperands(); ++i)
      assert(i == DstIdx ||
             !MI->getOperand(i).isReg() ||
             MI->getOperand(i).getReg() != RegA);
#endif

    // Emit a copy.
    MachineInstrBuilder MIB = BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
                                      TII->get(TargetOpcode::COPY), RegA);
    // If this operand is folding a truncation, the truncation now moves to the
    // copy so that the register classes remain valid for the operands.
    MIB.addReg(RegB, 0, SubRegB);
    const TargetRegisterClass *RC = MRI->getRegClass(RegB);
    if (SubRegB) {
      if (RegA.isVirtual()) {
        assert(TRI->getMatchingSuperRegClass(RC, MRI->getRegClass(RegA),
                                             SubRegB) &&
               "tied subregister must be a truncation");
        // The superreg class will not be used to constrain the subreg class.
        RC = nullptr;
      } else {
        assert(TRI->getMatchingSuperReg(RegA, SubRegB, MRI->getRegClass(RegB))
               && "tied subregister must be a truncation");
      }
    }

    // Update DistanceMap.
    MachineBasicBlock::iterator PrevMI = MI;
    --PrevMI;
    DistanceMap.insert(std::make_pair(&*PrevMI, Dist));
    DistanceMap[MI] = ++Dist;

    if (LIS) {
      LastCopyIdx = LIS->InsertMachineInstrInMaps(*PrevMI).getRegSlot();

      SlotIndex endIdx =
          LIS->getInstructionIndex(*MI).getRegSlot(IsEarlyClobber);
      if (RegA.isVirtual()) {
        LiveInterval &LI = LIS->getInterval(RegA);
        VNInfo *VNI = LI.getNextValue(LastCopyIdx, LIS->getVNInfoAllocator());
        LI.addSegment(LiveRange::Segment(LastCopyIdx, endIdx, VNI));
        for (auto &S : LI.subranges()) {
          VNI = S.getNextValue(LastCopyIdx, LIS->getVNInfoAllocator());
          S.addSegment(LiveRange::Segment(LastCopyIdx, endIdx, VNI));
        }
      } else {
        for (MCRegUnitIterator Unit(RegA, TRI); Unit.isValid(); ++Unit) {
          if (LiveRange *LR = LIS->getCachedRegUnit(*Unit)) {
            VNInfo *VNI =
                LR->getNextValue(LastCopyIdx, LIS->getVNInfoAllocator());
            LR->addSegment(LiveRange::Segment(LastCopyIdx, endIdx, VNI));
          }
        }
      }
    }

    LLVM_DEBUG(dbgs() << "\t\tprepend:\t" << *MIB);

    MachineOperand &MO = MI->getOperand(SrcIdx);
    assert(MO.isReg() && MO.getReg() == RegB && MO.isUse() &&
           "inconsistent operand info for 2-reg pass");
    if (MO.isKill()) {
      MO.setIsKill(false);
      RemovedKillFlag = true;
    }

    // Make sure regA is a legal regclass for the SrcIdx operand.
    if (RegA.isVirtual() && RegB.isVirtual())
      MRI->constrainRegClass(RegA, RC);
    MO.setReg(RegA);
    // The getMatchingSuper asserts guarantee that the register class projected
    // by SubRegB is compatible with RegA with no subregister. So regardless of
    // whether the dest oper writes a subreg, the source oper should not.
    MO.setSubReg(0);
  }

  if (AllUsesCopied) {
    LaneBitmask RemainingUses = LaneBitmask::getNone();
    // Replace other (un-tied) uses of regB with LastCopiedReg.
    for (MachineOperand &MO : MI->operands()) {
      if (MO.isReg() && MO.getReg() == RegB && MO.isUse()) {
        if (MO.getSubReg() == SubRegB && !IsEarlyClobber) {
          if (MO.isKill()) {
            MO.setIsKill(false);
            RemovedKillFlag = true;
          }
          MO.setReg(LastCopiedReg);
          MO.setSubReg(0);
        } else {
          RemainingUses |= TRI->getSubRegIndexLaneMask(MO.getSubReg());
        }
      }
    }

    // Update live variables for regB.
    if (RemovedKillFlag && RemainingUses.none() && LV &&
        LV->getVarInfo(RegB).removeKill(*MI)) {
      MachineBasicBlock::iterator PrevMI = MI;
      --PrevMI;
      LV->addVirtualRegisterKilled(RegB, *PrevMI);
    }

    if (RemovedKillFlag && RemainingUses.none())
      SrcRegMap[LastCopiedReg] = RegB;

    // Update LiveIntervals.
    if (LIS) {
      SlotIndex UseIdx = LIS->getInstructionIndex(*MI);
      auto Shrink = [=](LiveRange &LR, LaneBitmask LaneMask) {
        LiveRange::Segment *S = LR.getSegmentContaining(LastCopyIdx);
        if (!S)
          return true;
        if ((LaneMask & RemainingUses).any())
          return false;
        if (S->end.getBaseIndex() != UseIdx)
          return false;
        S->end = LastCopyIdx;
        return true;
      };

      LiveInterval &LI = LIS->getInterval(RegB);
      bool ShrinkLI = true;
      for (auto &S : LI.subranges())
        ShrinkLI &= Shrink(S, S.LaneMask);
      if (ShrinkLI)
        Shrink(LI, LaneBitmask::getAll());
    }
  } else if (RemovedKillFlag) {
    // Some tied uses of regB matched their destination registers, so
    // regB is still used in this instruction, but a kill flag was
    // removed from a different tied use of regB, so now we need to add
    // a kill flag to one of the remaining uses of regB.
    for (MachineOperand &MO : MI->operands()) {
      if (MO.isReg() && MO.getReg() == RegB && MO.isUse()) {
        MO.setIsKill(true);
        break;
      }
    }
  }
}

// For every tied operand pair this function transforms statepoint from
//    RegA = STATEPOINT ... RegB(tied-def N)
// to
//    RegB = STATEPOINT ... RegB(tied-def N)
// and replaces all uses of RegA with RegB.
// No extra COPY instruction is necessary because tied use is killed at
// STATEPOINT.
bool TwoAddressInstructionPass::processStatepoint(
    MachineInstr *MI, TiedOperandMap &TiedOperands) {

  bool NeedCopy = false;
  for (auto &TO : TiedOperands) {
    Register RegB = TO.first;
    if (TO.second.size() != 1) {
      NeedCopy = true;
      continue;
    }

    unsigned SrcIdx = TO.second[0].first;
    unsigned DstIdx = TO.second[0].second;

    MachineOperand &DstMO = MI->getOperand(DstIdx);
    Register RegA = DstMO.getReg();

    assert(RegB == MI->getOperand(SrcIdx).getReg());

    if (RegA == RegB)
      continue;

    // CodeGenPrepare can sink pointer compare past statepoint, which
    // breaks assumption that statepoint kills tied-use register when
    // in SSA form (see note in IR/SafepointIRVerifier.cpp). Fall back
    // to generic tied register handling to avoid assertion failures.
    // TODO: Recompute LIS/LV information for new range here.
    if (LIS) {
      const auto &UseLI = LIS->getInterval(RegB);
      const auto &DefLI = LIS->getInterval(RegA);
      if (DefLI.overlaps(UseLI)) {
        LLVM_DEBUG(dbgs() << "LIS: " << printReg(RegB, TRI, 0)
                          << " UseLI overlaps with DefLI\n");
        NeedCopy = true;
        continue;
      }
    } else if (LV && LV->getVarInfo(RegB).findKill(MI->getParent()) != MI) {
      // Note that MachineOperand::isKill does not work here, because it
      // is set only on first register use in instruction and for statepoint
      // tied-use register will usually be found in preceeding deopt bundle.
      LLVM_DEBUG(dbgs() << "LV: " << printReg(RegB, TRI, 0)
                        << " not killed by statepoint\n");
      NeedCopy = true;
      continue;
    }

    if (!MRI->constrainRegClass(RegB, MRI->getRegClass(RegA))) {
      LLVM_DEBUG(dbgs() << "MRI: couldn't constrain" << printReg(RegB, TRI, 0)
                        << " to register class of " << printReg(RegA, TRI, 0)
                        << '\n');
      NeedCopy = true;
      continue;
    }
    MRI->replaceRegWith(RegA, RegB);

    if (LIS) {
      VNInfo::Allocator &A = LIS->getVNInfoAllocator();
      LiveInterval &LI = LIS->getInterval(RegB);
      LiveInterval &Other = LIS->getInterval(RegA);
      SmallVector<VNInfo *> NewVNIs;
      for (const VNInfo *VNI : Other.valnos) {
        assert(VNI->id == NewVNIs.size() && "assumed");
        NewVNIs.push_back(LI.createValueCopy(VNI, A));
      }
      for (auto &S : Other) {
        VNInfo *VNI = NewVNIs[S.valno->id];
        LiveRange::Segment NewSeg(S.start, S.end, VNI);
        LI.addSegment(NewSeg);
      }
      LIS->removeInterval(RegA);
    }

    if (LV) {
      if (MI->getOperand(SrcIdx).isKill())
        LV->removeVirtualRegisterKilled(RegB, *MI);
      LiveVariables::VarInfo &SrcInfo = LV->getVarInfo(RegB);
      LiveVariables::VarInfo &DstInfo = LV->getVarInfo(RegA);
      SrcInfo.AliveBlocks |= DstInfo.AliveBlocks;
      DstInfo.AliveBlocks.clear();
      for (auto *KillMI : DstInfo.Kills)
        LV->addVirtualRegisterKilled(RegB, *KillMI, false);
    }
  }
  return !NeedCopy;
}

/// Reduce two-address instructions to two operands.
bool TwoAddressInstructionPass::runOnMachineFunction(MachineFunction &Func) {
  MF = &Func;
  const TargetMachine &TM = MF->getTarget();
  MRI = &MF->getRegInfo();
  TII = MF->getSubtarget().getInstrInfo();
  TRI = MF->getSubtarget().getRegisterInfo();
  InstrItins = MF->getSubtarget().getInstrItineraryData();
  LV = getAnalysisIfAvailable<LiveVariables>();
  LIS = getAnalysisIfAvailable<LiveIntervals>();
  if (auto *AAPass = getAnalysisIfAvailable<AAResultsWrapperPass>())
    AA = &AAPass->getAAResults();
  else
    AA = nullptr;
  OptLevel = TM.getOptLevel();
  // Disable optimizations if requested. We cannot skip the whole pass as some
  // fixups are necessary for correctness.
  if (skipFunction(Func.getFunction()))
    OptLevel = CodeGenOpt::None;

  bool MadeChange = false;

  LLVM_DEBUG(dbgs() << "********** REWRITING TWO-ADDR INSTRS **********\n");
  LLVM_DEBUG(dbgs() << "********** Function: " << MF->getName() << '\n');

  // This pass takes the function out of SSA form.
  MRI->leaveSSA();

  // This pass will rewrite the tied-def to meet the RegConstraint.
  MF->getProperties()
      .set(MachineFunctionProperties::Property::TiedOpsRewritten);

  TiedOperandMap TiedOperands;
  for (MachineBasicBlock &MBBI : *MF) {
    MBB = &MBBI;
    unsigned Dist = 0;
    DistanceMap.clear();
    SrcRegMap.clear();
    DstRegMap.clear();
    Processed.clear();
    for (MachineBasicBlock::iterator mi = MBB->begin(), me = MBB->end();
         mi != me; ) {
      MachineBasicBlock::iterator nmi = std::next(mi);
      // Skip debug instructions.
      if (mi->isDebugInstr()) {
        mi = nmi;
        continue;
      }

      // Expand REG_SEQUENCE instructions. This will position mi at the first
      // expanded instruction.
      if (mi->isRegSequence())
        eliminateRegSequence(mi);

      DistanceMap.insert(std::make_pair(&*mi, ++Dist));

      processCopy(&*mi);

      // First scan through all the tied register uses in this instruction
      // and record a list of pairs of tied operands for each register.
      if (!collectTiedOperands(&*mi, TiedOperands)) {
        removeClobberedSrcRegMap(&*mi);
        mi = nmi;
        continue;
      }

      ++NumTwoAddressInstrs;
      MadeChange = true;
      LLVM_DEBUG(dbgs() << '\t' << *mi);

      // If the instruction has a single pair of tied operands, try some
      // transformations that may either eliminate the tied operands or
      // improve the opportunities for coalescing away the register copy.
      if (TiedOperands.size() == 1) {
        SmallVectorImpl<std::pair<unsigned, unsigned>> &TiedPairs
          = TiedOperands.begin()->second;
        if (TiedPairs.size() == 1) {
          unsigned SrcIdx = TiedPairs[0].first;
          unsigned DstIdx = TiedPairs[0].second;
          Register SrcReg = mi->getOperand(SrcIdx).getReg();
          Register DstReg = mi->getOperand(DstIdx).getReg();
          if (SrcReg != DstReg &&
              tryInstructionTransform(mi, nmi, SrcIdx, DstIdx, Dist, false)) {
            // The tied operands have been eliminated or shifted further down
            // the block to ease elimination. Continue processing with 'nmi'.
            TiedOperands.clear();
            removeClobberedSrcRegMap(&*mi);
            mi = nmi;
            continue;
          }
        }
      }

      if (mi->getOpcode() == TargetOpcode::STATEPOINT &&
          processStatepoint(&*mi, TiedOperands)) {
        TiedOperands.clear();
        LLVM_DEBUG(dbgs() << "\t\trewrite to:\t" << *mi);
        mi = nmi;
        continue;
      }

      // Now iterate over the information collected above.
      for (auto &TO : TiedOperands) {
        processTiedPairs(&*mi, TO.second, Dist);
        LLVM_DEBUG(dbgs() << "\t\trewrite to:\t" << *mi);
      }

      // Rewrite INSERT_SUBREG as COPY now that we no longer need SSA form.
      if (mi->isInsertSubreg()) {
        // From %reg = INSERT_SUBREG %reg, %subreg, subidx
        // To   %reg:subidx = COPY %subreg
        unsigned SubIdx = mi->getOperand(3).getImm();
        mi->removeOperand(3);
        assert(mi->getOperand(0).getSubReg() == 0 && "Unexpected subreg idx");
        mi->getOperand(0).setSubReg(SubIdx);
        mi->getOperand(0).setIsUndef(mi->getOperand(1).isUndef());
        mi->removeOperand(1);
        mi->setDesc(TII->get(TargetOpcode::COPY));
        LLVM_DEBUG(dbgs() << "\t\tconvert to:\t" << *mi);

        // Update LiveIntervals.
        if (LIS) {
          Register Reg = mi->getOperand(0).getReg();
          LiveInterval &LI = LIS->getInterval(Reg);
          if (LI.hasSubRanges()) {
            // The COPY no longer defines subregs of %reg except for
            // %reg.subidx.
            LaneBitmask LaneMask =
                TRI->getSubRegIndexLaneMask(mi->getOperand(0).getSubReg());
            SlotIndex Idx = LIS->getInstructionIndex(*mi);
            for (auto &S : LI.subranges()) {
              if ((S.LaneMask & LaneMask).none()) {
                LiveRange::iterator UseSeg = S.FindSegmentContaining(Idx);
                LiveRange::iterator DefSeg = std::next(UseSeg);
                S.MergeValueNumberInto(DefSeg->valno, UseSeg->valno);
              }
            }

            // The COPY no longer has a use of %reg.
            LIS->shrinkToUses(&LI);
          } else {
            // The live interval for Reg did not have subranges but now it needs
            // them because we have introduced a subreg def. Recompute it.
            LIS->removeInterval(Reg);
            LIS->createAndComputeVirtRegInterval(Reg);
          }
        }
      }

      // Clear TiedOperands here instead of at the top of the loop
      // since most instructions do not have tied operands.
      TiedOperands.clear();
      removeClobberedSrcRegMap(&*mi);
      mi = nmi;
    }
  }

  return MadeChange;
}

/// Eliminate a REG_SEQUENCE instruction as part of the de-ssa process.
///
/// The instruction is turned into a sequence of sub-register copies:
///
///   %dst = REG_SEQUENCE %v1, ssub0, %v2, ssub1
///
/// Becomes:
///
///   undef %dst:ssub0 = COPY %v1
///   %dst:ssub1 = COPY %v2
void TwoAddressInstructionPass::
eliminateRegSequence(MachineBasicBlock::iterator &MBBI) {
  MachineInstr &MI = *MBBI;
  Register DstReg = MI.getOperand(0).getReg();

  SmallVector<Register, 4> OrigRegs;
  if (LIS) {
    OrigRegs.push_back(MI.getOperand(0).getReg());
    for (unsigned i = 1, e = MI.getNumOperands(); i < e; i += 2)
      OrigRegs.push_back(MI.getOperand(i).getReg());
  }

  bool DefEmitted = false;
  for (unsigned i = 1, e = MI.getNumOperands(); i < e; i += 2) {
    MachineOperand &UseMO = MI.getOperand(i);
    Register SrcReg = UseMO.getReg();
    unsigned SubIdx = MI.getOperand(i+1).getImm();
    // Nothing needs to be inserted for undef operands.
    if (UseMO.isUndef())
      continue;

    // Defer any kill flag to the last operand using SrcReg. Otherwise, we
    // might insert a COPY that uses SrcReg after is was killed.
    bool isKill = UseMO.isKill();
    if (isKill)
      for (unsigned j = i + 2; j < e; j += 2)
        if (MI.getOperand(j).getReg() == SrcReg) {
          MI.getOperand(j).setIsKill();
          UseMO.setIsKill(false);
          isKill = false;
          break;
        }

    // Insert the sub-register copy.
    MachineInstr *CopyMI = BuildMI(*MI.getParent(), MI, MI.getDebugLoc(),
                                   TII->get(TargetOpcode::COPY))
                               .addReg(DstReg, RegState::Define, SubIdx)
                               .add(UseMO);

    // The first def needs an undef flag because there is no live register
    // before it.
    if (!DefEmitted) {
      CopyMI->getOperand(0).setIsUndef(true);
      // Return an iterator pointing to the first inserted instr.
      MBBI = CopyMI;
    }
    DefEmitted = true;

    // Update LiveVariables' kill info.
    if (LV && isKill && !SrcReg.isPhysical())
      LV->replaceKillInstruction(SrcReg, MI, *CopyMI);

    LLVM_DEBUG(dbgs() << "Inserted: " << *CopyMI);
  }

  MachineBasicBlock::iterator EndMBBI =
      std::next(MachineBasicBlock::iterator(MI));

  if (!DefEmitted) {
    LLVM_DEBUG(dbgs() << "Turned: " << MI << " into an IMPLICIT_DEF");
    MI.setDesc(TII->get(TargetOpcode::IMPLICIT_DEF));
    for (int j = MI.getNumOperands() - 1, ee = 0; j > ee; --j)
      MI.removeOperand(j);
  } else {
    if (LIS)
      LIS->RemoveMachineInstrFromMaps(MI);

    LLVM_DEBUG(dbgs() << "Eliminated: " << MI);
    MI.eraseFromParent();
  }

  // Udpate LiveIntervals.
  if (LIS)
    LIS->repairIntervalsInRange(MBB, MBBI, EndMBBI, OrigRegs);
}