llvm.org GIT mirror llvm / 15c435a lib / CodeGen / RegisterCoalescer.cpp
15c435a

Tree @15c435a (Download .tar.gz)

RegisterCoalescer.cpp @15c435araw · history · blame

   1
   2
   3
   4
   5
   6
   7
   8
   9
  10
  11
  12
  13
  14
  15
  16
  17
  18
  19
  20
  21
  22
  23
  24
  25
  26
  27
  28
  29
  30
  31
  32
  33
  34
  35
  36
  37
  38
  39
  40
  41
  42
  43
  44
  45
  46
  47
  48
  49
  50
  51
  52
  53
  54
  55
  56
  57
  58
  59
  60
  61
  62
  63
  64
  65
  66
  67
  68
  69
  70
  71
  72
  73
  74
  75
  76
  77
  78
  79
  80
  81
  82
  83
  84
  85
  86
  87
  88
  89
  90
  91
  92
  93
  94
  95
  96
  97
  98
  99
 100
 101
 102
 103
 104
 105
 106
 107
 108
 109
 110
 111
 112
 113
 114
 115
 116
 117
 118
 119
 120
 121
 122
 123
 124
 125
 126
 127
 128
 129
 130
 131
 132
 133
 134
 135
 136
 137
 138
 139
 140
 141
 142
 143
 144
 145
 146
 147
 148
 149
 150
 151
 152
 153
 154
 155
 156
 157
 158
 159
 160
 161
 162
 163
 164
 165
 166
 167
 168
 169
 170
 171
 172
 173
 174
 175
 176
 177
 178
 179
 180
 181
 182
 183
 184
 185
 186
 187
 188
 189
 190
 191
 192
 193
 194
 195
 196
 197
 198
 199
 200
 201
 202
 203
 204
 205
 206
 207
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 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
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
//===- RegisterCoalescer.cpp - Generic Register Coalescing Interface -------==//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the generic RegisterCoalescer interface which
// is used as the common interface used by all clients and
// implementations of register coalescing.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "regalloc"
#include "RegisterCoalescer.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/LiveIntervalAnalysis.h"
#include "llvm/CodeGen/LiveRangeEdit.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/CodeGen/RegisterClassInfo.h"
#include "llvm/CodeGen/VirtRegMap.h"
#include "llvm/IR/Value.h"
#include "llvm/Pass.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/TargetInstrInfo.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetSubtargetInfo.h"
#include <algorithm>
#include <cmath>
using namespace llvm;

STATISTIC(numJoins    , "Number of interval joins performed");
STATISTIC(numCrossRCs , "Number of cross class joins performed");
STATISTIC(numCommutes , "Number of instruction commuting performed");
STATISTIC(numExtends  , "Number of copies extended");
STATISTIC(NumReMats   , "Number of instructions re-materialized");
STATISTIC(NumInflated , "Number of register classes inflated");
STATISTIC(NumLaneConflicts, "Number of dead lane conflicts tested");
STATISTIC(NumLaneResolves,  "Number of dead lane conflicts resolved");

static cl::opt<bool>
EnableJoining("join-liveintervals",
              cl::desc("Coalesce copies (default=true)"),
              cl::init(true));

// Temporary flag to test critical edge unsplitting.
static cl::opt<bool>
EnableJoinSplits("join-splitedges",
  cl::desc("Coalesce copies on split edges (default=subtarget)"), cl::Hidden);

// Temporary flag to test global copy optimization.
static cl::opt<cl::boolOrDefault>
EnableGlobalCopies("join-globalcopies",
  cl::desc("Coalesce copies that span blocks (default=subtarget)"),
  cl::init(cl::BOU_UNSET), cl::Hidden);

static cl::opt<bool>
VerifyCoalescing("verify-coalescing",
         cl::desc("Verify machine instrs before and after register coalescing"),
         cl::Hidden);

namespace {
  class RegisterCoalescer : public MachineFunctionPass,
                            private LiveRangeEdit::Delegate {
    MachineFunction* MF;
    MachineRegisterInfo* MRI;
    const TargetMachine* TM;
    const TargetRegisterInfo* TRI;
    const TargetInstrInfo* TII;
    LiveIntervals *LIS;
    const MachineLoopInfo* Loops;
    AliasAnalysis *AA;
    RegisterClassInfo RegClassInfo;

    /// \brief True if the coalescer should aggressively coalesce global copies
    /// in favor of keeping local copies.
    bool JoinGlobalCopies;

    /// \brief True if the coalescer should aggressively coalesce fall-thru
    /// blocks exclusively containing copies.
    bool JoinSplitEdges;

    /// WorkList - Copy instructions yet to be coalesced.
    SmallVector<MachineInstr*, 8> WorkList;
    SmallVector<MachineInstr*, 8> LocalWorkList;

    /// ErasedInstrs - Set of instruction pointers that have been erased, and
    /// that may be present in WorkList.
    SmallPtrSet<MachineInstr*, 8> ErasedInstrs;

    /// Dead instructions that are about to be deleted.
    SmallVector<MachineInstr*, 8> DeadDefs;

    /// Virtual registers to be considered for register class inflation.
    SmallVector<unsigned, 8> InflateRegs;

    /// Recursively eliminate dead defs in DeadDefs.
    void eliminateDeadDefs();

    /// LiveRangeEdit callback.
    void LRE_WillEraseInstruction(MachineInstr *MI) override;

    /// coalesceLocals - coalesce the LocalWorkList.
    void coalesceLocals();

    /// joinAllIntervals - join compatible live intervals
    void joinAllIntervals();

    /// copyCoalesceInMBB - Coalesce copies in the specified MBB, putting
    /// copies that cannot yet be coalesced into WorkList.
    void copyCoalesceInMBB(MachineBasicBlock *MBB);

    /// copyCoalesceWorkList - Try to coalesce all copies in CurrList. Return
    /// true if any progress was made.
    bool copyCoalesceWorkList(MutableArrayRef<MachineInstr*> CurrList);

    /// joinCopy - Attempt to join intervals corresponding to SrcReg/DstReg,
    /// which are the src/dst of the copy instruction CopyMI.  This returns
    /// true if the copy was successfully coalesced away. If it is not
    /// currently possible to coalesce this interval, but it may be possible if
    /// other things get coalesced, then it returns true by reference in
    /// 'Again'.
    bool joinCopy(MachineInstr *TheCopy, bool &Again);

    /// joinIntervals - Attempt to join these two intervals.  On failure, this
    /// returns false.  The output "SrcInt" will not have been modified, so we
    /// can use this information below to update aliases.
    bool joinIntervals(CoalescerPair &CP);

    /// Attempt joining two virtual registers. Return true on success.
    bool joinVirtRegs(CoalescerPair &CP);

    /// Attempt joining with a reserved physreg.
    bool joinReservedPhysReg(CoalescerPair &CP);

    /// adjustCopiesBackFrom - We found a non-trivially-coalescable copy. If
    /// the source value number is defined by a copy from the destination reg
    /// see if we can merge these two destination reg valno# into a single
    /// value number, eliminating a copy.
    bool adjustCopiesBackFrom(const CoalescerPair &CP, MachineInstr *CopyMI);

    /// hasOtherReachingDefs - Return true if there are definitions of IntB
    /// other than BValNo val# that can reach uses of AValno val# of IntA.
    bool hasOtherReachingDefs(LiveInterval &IntA, LiveInterval &IntB,
                              VNInfo *AValNo, VNInfo *BValNo);

    /// removeCopyByCommutingDef - We found a non-trivially-coalescable copy.
    /// If the source value number is defined by a commutable instruction and
    /// its other operand is coalesced to the copy dest register, see if we
    /// can transform the copy into a noop by commuting the definition.
    bool removeCopyByCommutingDef(const CoalescerPair &CP,MachineInstr *CopyMI);

    /// reMaterializeTrivialDef - If the source of a copy is defined by a
    /// trivial computation, replace the copy by rematerialize the definition.
    bool reMaterializeTrivialDef(CoalescerPair &CP, MachineInstr *CopyMI,
                                 bool &IsDefCopy);

    /// canJoinPhys - Return true if a physreg copy should be joined.
    bool canJoinPhys(const CoalescerPair &CP);

    /// updateRegDefsUses - Replace all defs and uses of SrcReg to DstReg and
    /// update the subregister number if it is not zero. If DstReg is a
    /// physical register and the existing subregister number of the def / use
    /// being updated is not zero, make sure to set it to the correct physical
    /// subregister.
    void updateRegDefsUses(unsigned SrcReg, unsigned DstReg, unsigned SubIdx);

    /// eliminateUndefCopy - Handle copies of undef values.
    bool eliminateUndefCopy(MachineInstr *CopyMI, const CoalescerPair &CP);

  public:
    static char ID; // Class identification, replacement for typeinfo
    RegisterCoalescer() : MachineFunctionPass(ID) {
      initializeRegisterCoalescerPass(*PassRegistry::getPassRegistry());
    }

    void getAnalysisUsage(AnalysisUsage &AU) const override;

    void releaseMemory() override;

    /// runOnMachineFunction - pass entry point
    bool runOnMachineFunction(MachineFunction&) override;

    /// print - Implement the dump method.
    void print(raw_ostream &O, const Module* = 0) const override;
  };
} /// end anonymous namespace

char &llvm::RegisterCoalescerID = RegisterCoalescer::ID;

INITIALIZE_PASS_BEGIN(RegisterCoalescer, "simple-register-coalescing",
                      "Simple Register Coalescing", false, false)
INITIALIZE_PASS_DEPENDENCY(LiveIntervals)
INITIALIZE_PASS_DEPENDENCY(SlotIndexes)
INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(RegisterCoalescer, "simple-register-coalescing",
                    "Simple Register Coalescing", false, false)

char RegisterCoalescer::ID = 0;

static bool isMoveInstr(const TargetRegisterInfo &tri, const MachineInstr *MI,
                        unsigned &Src, unsigned &Dst,
                        unsigned &SrcSub, unsigned &DstSub) {
  if (MI->isCopy()) {
    Dst = MI->getOperand(0).getReg();
    DstSub = MI->getOperand(0).getSubReg();
    Src = MI->getOperand(1).getReg();
    SrcSub = MI->getOperand(1).getSubReg();
  } else if (MI->isSubregToReg()) {
    Dst = MI->getOperand(0).getReg();
    DstSub = tri.composeSubRegIndices(MI->getOperand(0).getSubReg(),
                                      MI->getOperand(3).getImm());
    Src = MI->getOperand(2).getReg();
    SrcSub = MI->getOperand(2).getSubReg();
  } else
    return false;
  return true;
}

// Return true if this block should be vacated by the coalescer to eliminate
// branches. The important cases to handle in the coalescer are critical edges
// split during phi elimination which contain only copies. Simple blocks that
// contain non-branches should also be vacated, but this can be handled by an
// earlier pass similar to early if-conversion.
static bool isSplitEdge(const MachineBasicBlock *MBB) {
  if (MBB->pred_size() != 1 || MBB->succ_size() != 1)
    return false;

  for (MachineBasicBlock::const_iterator MII = MBB->begin(), E = MBB->end();
       MII != E; ++MII) {
    if (!MII->isCopyLike() && !MII->isUnconditionalBranch())
      return false;
  }
  return true;
}

bool CoalescerPair::setRegisters(const MachineInstr *MI) {
  SrcReg = DstReg = 0;
  SrcIdx = DstIdx = 0;
  NewRC = 0;
  Flipped = CrossClass = false;

  unsigned Src, Dst, SrcSub, DstSub;
  if (!isMoveInstr(TRI, MI, Src, Dst, SrcSub, DstSub))
    return false;
  Partial = SrcSub || DstSub;

  // If one register is a physreg, it must be Dst.
  if (TargetRegisterInfo::isPhysicalRegister(Src)) {
    if (TargetRegisterInfo::isPhysicalRegister(Dst))
      return false;
    std::swap(Src, Dst);
    std::swap(SrcSub, DstSub);
    Flipped = true;
  }

  const MachineRegisterInfo &MRI = MI->getParent()->getParent()->getRegInfo();

  if (TargetRegisterInfo::isPhysicalRegister(Dst)) {
    // Eliminate DstSub on a physreg.
    if (DstSub) {
      Dst = TRI.getSubReg(Dst, DstSub);
      if (!Dst) return false;
      DstSub = 0;
    }

    // Eliminate SrcSub by picking a corresponding Dst superregister.
    if (SrcSub) {
      Dst = TRI.getMatchingSuperReg(Dst, SrcSub, MRI.getRegClass(Src));
      if (!Dst) return false;
    } else if (!MRI.getRegClass(Src)->contains(Dst)) {
      return false;
    }
  } else {
    // Both registers are virtual.
    const TargetRegisterClass *SrcRC = MRI.getRegClass(Src);
    const TargetRegisterClass *DstRC = MRI.getRegClass(Dst);

    // Both registers have subreg indices.
    if (SrcSub && DstSub) {
      // Copies between different sub-registers are never coalescable.
      if (Src == Dst && SrcSub != DstSub)
        return false;

      NewRC = TRI.getCommonSuperRegClass(SrcRC, SrcSub, DstRC, DstSub,
                                         SrcIdx, DstIdx);
      if (!NewRC)
        return false;
    } else if (DstSub) {
      // SrcReg will be merged with a sub-register of DstReg.
      SrcIdx = DstSub;
      NewRC = TRI.getMatchingSuperRegClass(DstRC, SrcRC, DstSub);
    } else if (SrcSub) {
      // DstReg will be merged with a sub-register of SrcReg.
      DstIdx = SrcSub;
      NewRC = TRI.getMatchingSuperRegClass(SrcRC, DstRC, SrcSub);
    } else {
      // This is a straight copy without sub-registers.
      NewRC = TRI.getCommonSubClass(DstRC, SrcRC);
    }

    // The combined constraint may be impossible to satisfy.
    if (!NewRC)
      return false;

    // Prefer SrcReg to be a sub-register of DstReg.
    // FIXME: Coalescer should support subregs symmetrically.
    if (DstIdx && !SrcIdx) {
      std::swap(Src, Dst);
      std::swap(SrcIdx, DstIdx);
      Flipped = !Flipped;
    }

    CrossClass = NewRC != DstRC || NewRC != SrcRC;
  }
  // Check our invariants
  assert(TargetRegisterInfo::isVirtualRegister(Src) && "Src must be virtual");
  assert(!(TargetRegisterInfo::isPhysicalRegister(Dst) && DstSub) &&
         "Cannot have a physical SubIdx");
  SrcReg = Src;
  DstReg = Dst;
  return true;
}

bool CoalescerPair::flip() {
  if (TargetRegisterInfo::isPhysicalRegister(DstReg))
    return false;
  std::swap(SrcReg, DstReg);
  std::swap(SrcIdx, DstIdx);
  Flipped = !Flipped;
  return true;
}

bool CoalescerPair::isCoalescable(const MachineInstr *MI) const {
  if (!MI)
    return false;
  unsigned Src, Dst, SrcSub, DstSub;
  if (!isMoveInstr(TRI, MI, Src, Dst, SrcSub, DstSub))
    return false;

  // Find the virtual register that is SrcReg.
  if (Dst == SrcReg) {
    std::swap(Src, Dst);
    std::swap(SrcSub, DstSub);
  } else if (Src != SrcReg) {
    return false;
  }

  // Now check that Dst matches DstReg.
  if (TargetRegisterInfo::isPhysicalRegister(DstReg)) {
    if (!TargetRegisterInfo::isPhysicalRegister(Dst))
      return false;
    assert(!DstIdx && !SrcIdx && "Inconsistent CoalescerPair state.");
    // DstSub could be set for a physreg from INSERT_SUBREG.
    if (DstSub)
      Dst = TRI.getSubReg(Dst, DstSub);
    // Full copy of Src.
    if (!SrcSub)
      return DstReg == Dst;
    // This is a partial register copy. Check that the parts match.
    return TRI.getSubReg(DstReg, SrcSub) == Dst;
  } else {
    // DstReg is virtual.
    if (DstReg != Dst)
      return false;
    // Registers match, do the subregisters line up?
    return TRI.composeSubRegIndices(SrcIdx, SrcSub) ==
           TRI.composeSubRegIndices(DstIdx, DstSub);
  }
}

void RegisterCoalescer::getAnalysisUsage(AnalysisUsage &AU) const {
  AU.setPreservesCFG();
  AU.addRequired<AliasAnalysis>();
  AU.addRequired<LiveIntervals>();
  AU.addPreserved<LiveIntervals>();
  AU.addPreserved<SlotIndexes>();
  AU.addRequired<MachineLoopInfo>();
  AU.addPreserved<MachineLoopInfo>();
  AU.addPreservedID(MachineDominatorsID);
  MachineFunctionPass::getAnalysisUsage(AU);
}

void RegisterCoalescer::eliminateDeadDefs() {
  SmallVector<unsigned, 8> NewRegs;
  LiveRangeEdit(0, NewRegs, *MF, *LIS, 0, this).eliminateDeadDefs(DeadDefs);
}

// Callback from eliminateDeadDefs().
void RegisterCoalescer::LRE_WillEraseInstruction(MachineInstr *MI) {
  // MI may be in WorkList. Make sure we don't visit it.
  ErasedInstrs.insert(MI);
}

/// adjustCopiesBackFrom - We found a non-trivially-coalescable copy with IntA
/// being the source and IntB being the dest, thus this defines a value number
/// in IntB.  If the source value number (in IntA) is defined by a copy from B,
/// see if we can merge these two pieces of B into a single value number,
/// eliminating a copy.  For example:
///
///  A3 = B0
///    ...
///  B1 = A3      <- this copy
///
/// In this case, B0 can be extended to where the B1 copy lives, allowing the B1
/// value number to be replaced with B0 (which simplifies the B liveinterval).
///
/// This returns true if an interval was modified.
///
bool RegisterCoalescer::adjustCopiesBackFrom(const CoalescerPair &CP,
                                             MachineInstr *CopyMI) {
  assert(!CP.isPartial() && "This doesn't work for partial copies.");
  assert(!CP.isPhys() && "This doesn't work for physreg copies.");

  LiveInterval &IntA =
    LIS->getInterval(CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg());
  LiveInterval &IntB =
    LIS->getInterval(CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg());
  SlotIndex CopyIdx = LIS->getInstructionIndex(CopyMI).getRegSlot();

  // BValNo is a value number in B that is defined by a copy from A.  'B1' in
  // the example above.
  LiveInterval::iterator BS = IntB.FindSegmentContaining(CopyIdx);
  if (BS == IntB.end()) return false;
  VNInfo *BValNo = BS->valno;

  // Get the location that B is defined at.  Two options: either this value has
  // an unknown definition point or it is defined at CopyIdx.  If unknown, we
  // can't process it.
  if (BValNo->def != CopyIdx) return false;

  // AValNo is the value number in A that defines the copy, A3 in the example.
  SlotIndex CopyUseIdx = CopyIdx.getRegSlot(true);
  LiveInterval::iterator AS = IntA.FindSegmentContaining(CopyUseIdx);
  // The live segment might not exist after fun with physreg coalescing.
  if (AS == IntA.end()) return false;
  VNInfo *AValNo = AS->valno;

  // If AValNo is defined as a copy from IntB, we can potentially process this.
  // Get the instruction that defines this value number.
  MachineInstr *ACopyMI = LIS->getInstructionFromIndex(AValNo->def);
  // Don't allow any partial copies, even if isCoalescable() allows them.
  if (!CP.isCoalescable(ACopyMI) || !ACopyMI->isFullCopy())
    return false;

  // Get the Segment in IntB that this value number starts with.
  LiveInterval::iterator ValS =
    IntB.FindSegmentContaining(AValNo->def.getPrevSlot());
  if (ValS == IntB.end())
    return false;

  // Make sure that the end of the live segment is inside the same block as
  // CopyMI.
  MachineInstr *ValSEndInst =
    LIS->getInstructionFromIndex(ValS->end.getPrevSlot());
  if (!ValSEndInst || ValSEndInst->getParent() != CopyMI->getParent())
    return false;

  // Okay, we now know that ValS ends in the same block that the CopyMI
  // live-range starts.  If there are no intervening live segments between them
  // in IntB, we can merge them.
  if (ValS+1 != BS) return false;

  DEBUG(dbgs() << "Extending: " << PrintReg(IntB.reg, TRI));

  SlotIndex FillerStart = ValS->end, FillerEnd = BS->start;
  // We are about to delete CopyMI, so need to remove it as the 'instruction
  // that defines this value #'. Update the valnum with the new defining
  // instruction #.
  BValNo->def = FillerStart;

  // Okay, we can merge them.  We need to insert a new liverange:
  // [ValS.end, BS.begin) of either value number, then we merge the
  // two value numbers.
  IntB.addSegment(LiveInterval::Segment(FillerStart, FillerEnd, BValNo));

  // Okay, merge "B1" into the same value number as "B0".
  if (BValNo != ValS->valno)
    IntB.MergeValueNumberInto(BValNo, ValS->valno);
  DEBUG(dbgs() << "   result = " << IntB << '\n');

  // If the source instruction was killing the source register before the
  // merge, unset the isKill marker given the live range has been extended.
  int UIdx = ValSEndInst->findRegisterUseOperandIdx(IntB.reg, true);
  if (UIdx != -1) {
    ValSEndInst->getOperand(UIdx).setIsKill(false);
  }

  // Rewrite the copy. If the copy instruction was killing the destination
  // register before the merge, find the last use and trim the live range. That
  // will also add the isKill marker.
  CopyMI->substituteRegister(IntA.reg, IntB.reg, 0, *TRI);
  if (AS->end == CopyIdx)
    LIS->shrinkToUses(&IntA);

  ++numExtends;
  return true;
}

/// hasOtherReachingDefs - Return true if there are definitions of IntB
/// other than BValNo val# that can reach uses of AValno val# of IntA.
bool RegisterCoalescer::hasOtherReachingDefs(LiveInterval &IntA,
                                             LiveInterval &IntB,
                                             VNInfo *AValNo,
                                             VNInfo *BValNo) {
  // If AValNo has PHI kills, conservatively assume that IntB defs can reach
  // the PHI values.
  if (LIS->hasPHIKill(IntA, AValNo))
    return true;

  for (LiveInterval::iterator AI = IntA.begin(), AE = IntA.end();
       AI != AE; ++AI) {
    if (AI->valno != AValNo) continue;
    LiveInterval::iterator BI =
      std::upper_bound(IntB.begin(), IntB.end(), AI->start);
    if (BI != IntB.begin())
      --BI;
    for (; BI != IntB.end() && AI->end >= BI->start; ++BI) {
      if (BI->valno == BValNo)
        continue;
      if (BI->start <= AI->start && BI->end > AI->start)
        return true;
      if (BI->start > AI->start && BI->start < AI->end)
        return true;
    }
  }
  return false;
}

/// removeCopyByCommutingDef - We found a non-trivially-coalescable copy with
/// IntA being the source and IntB being the dest, thus this defines a value
/// number in IntB.  If the source value number (in IntA) is defined by a
/// commutable instruction and its other operand is coalesced to the copy dest
/// register, see if we can transform the copy into a noop by commuting the
/// definition. For example,
///
///  A3 = op A2 B0<kill>
///    ...
///  B1 = A3      <- this copy
///    ...
///     = op A3   <- more uses
///
/// ==>
///
///  B2 = op B0 A2<kill>
///    ...
///  B1 = B2      <- now an identify copy
///    ...
///     = op B2   <- more uses
///
/// This returns true if an interval was modified.
///
bool RegisterCoalescer::removeCopyByCommutingDef(const CoalescerPair &CP,
                                                 MachineInstr *CopyMI) {
  assert (!CP.isPhys());

  SlotIndex CopyIdx = LIS->getInstructionIndex(CopyMI).getRegSlot();

  LiveInterval &IntA =
    LIS->getInterval(CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg());
  LiveInterval &IntB =
    LIS->getInterval(CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg());

  // BValNo is a value number in B that is defined by a copy from A. 'B1' in
  // the example above.
  VNInfo *BValNo = IntB.getVNInfoAt(CopyIdx);
  if (!BValNo || BValNo->def != CopyIdx)
    return false;

  // AValNo is the value number in A that defines the copy, A3 in the example.
  VNInfo *AValNo = IntA.getVNInfoAt(CopyIdx.getRegSlot(true));
  assert(AValNo && "COPY source not live");
  if (AValNo->isPHIDef() || AValNo->isUnused())
    return false;
  MachineInstr *DefMI = LIS->getInstructionFromIndex(AValNo->def);
  if (!DefMI)
    return false;
  if (!DefMI->isCommutable())
    return false;
  // If DefMI is a two-address instruction then commuting it will change the
  // destination register.
  int DefIdx = DefMI->findRegisterDefOperandIdx(IntA.reg);
  assert(DefIdx != -1);
  unsigned UseOpIdx;
  if (!DefMI->isRegTiedToUseOperand(DefIdx, &UseOpIdx))
    return false;
  unsigned Op1, Op2, NewDstIdx;
  if (!TII->findCommutedOpIndices(DefMI, Op1, Op2))
    return false;
  if (Op1 == UseOpIdx)
    NewDstIdx = Op2;
  else if (Op2 == UseOpIdx)
    NewDstIdx = Op1;
  else
    return false;

  MachineOperand &NewDstMO = DefMI->getOperand(NewDstIdx);
  unsigned NewReg = NewDstMO.getReg();
  if (NewReg != IntB.reg || !IntB.Query(AValNo->def).isKill())
    return false;

  // Make sure there are no other definitions of IntB that would reach the
  // uses which the new definition can reach.
  if (hasOtherReachingDefs(IntA, IntB, AValNo, BValNo))
    return false;

  // If some of the uses of IntA.reg is already coalesced away, return false.
  // It's not possible to determine whether it's safe to perform the coalescing.
  for (MachineOperand &MO : MRI->use_nodbg_operands(IntA.reg)) {
    MachineInstr *UseMI = MO.getParent();
    unsigned OpNo = &MO - &UseMI->getOperand(0);
    SlotIndex UseIdx = LIS->getInstructionIndex(UseMI);
    LiveInterval::iterator US = IntA.FindSegmentContaining(UseIdx);
    if (US == IntA.end() || US->valno != AValNo)
      continue;
    // If this use is tied to a def, we can't rewrite the register.
    if (UseMI->isRegTiedToDefOperand(OpNo))
      return false;
  }

  DEBUG(dbgs() << "\tremoveCopyByCommutingDef: " << AValNo->def << '\t'
               << *DefMI);

  // At this point we have decided that it is legal to do this
  // transformation.  Start by commuting the instruction.
  MachineBasicBlock *MBB = DefMI->getParent();
  MachineInstr *NewMI = TII->commuteInstruction(DefMI);
  if (!NewMI)
    return false;
  if (TargetRegisterInfo::isVirtualRegister(IntA.reg) &&
      TargetRegisterInfo::isVirtualRegister(IntB.reg) &&
      !MRI->constrainRegClass(IntB.reg, MRI->getRegClass(IntA.reg)))
    return false;
  if (NewMI != DefMI) {
    LIS->ReplaceMachineInstrInMaps(DefMI, NewMI);
    MachineBasicBlock::iterator Pos = DefMI;
    MBB->insert(Pos, NewMI);
    MBB->erase(DefMI);
  }
  unsigned OpIdx = NewMI->findRegisterUseOperandIdx(IntA.reg, false);
  NewMI->getOperand(OpIdx).setIsKill();

  // If ALR and BLR overlaps and end of BLR extends beyond end of ALR, e.g.
  // A = or A, B
  // ...
  // B = A
  // ...
  // C = A<kill>
  // ...
  //   = B

  // Update uses of IntA of the specific Val# with IntB.
  for (MachineRegisterInfo::use_iterator UI = MRI->use_begin(IntA.reg),
         UE = MRI->use_end(); UI != UE;) {
    MachineOperand &UseMO = *UI;
    MachineInstr *UseMI = UseMO.getParent();
    ++UI;
    if (UseMI->isDebugValue()) {
      // FIXME These don't have an instruction index.  Not clear we have enough
      // info to decide whether to do this replacement or not.  For now do it.
      UseMO.setReg(NewReg);
      continue;
    }
    SlotIndex UseIdx = LIS->getInstructionIndex(UseMI).getRegSlot(true);
    LiveInterval::iterator US = IntA.FindSegmentContaining(UseIdx);
    if (US == IntA.end() || US->valno != AValNo)
      continue;
    // Kill flags are no longer accurate. They are recomputed after RA.
    UseMO.setIsKill(false);
    if (TargetRegisterInfo::isPhysicalRegister(NewReg))
      UseMO.substPhysReg(NewReg, *TRI);
    else
      UseMO.setReg(NewReg);
    if (UseMI == CopyMI)
      continue;
    if (!UseMI->isCopy())
      continue;
    if (UseMI->getOperand(0).getReg() != IntB.reg ||
        UseMI->getOperand(0).getSubReg())
      continue;

    // This copy will become a noop. If it's defining a new val#, merge it into
    // BValNo.
    SlotIndex DefIdx = UseIdx.getRegSlot();
    VNInfo *DVNI = IntB.getVNInfoAt(DefIdx);
    if (!DVNI)
      continue;
    DEBUG(dbgs() << "\t\tnoop: " << DefIdx << '\t' << *UseMI);
    assert(DVNI->def == DefIdx);
    BValNo = IntB.MergeValueNumberInto(BValNo, DVNI);
    ErasedInstrs.insert(UseMI);
    LIS->RemoveMachineInstrFromMaps(UseMI);
    UseMI->eraseFromParent();
  }

  // Extend BValNo by merging in IntA live segments of AValNo. Val# definition
  // is updated.
  VNInfo *ValNo = BValNo;
  ValNo->def = AValNo->def;
  for (LiveInterval::iterator AI = IntA.begin(), AE = IntA.end();
       AI != AE; ++AI) {
    if (AI->valno != AValNo) continue;
    IntB.addSegment(LiveInterval::Segment(AI->start, AI->end, ValNo));
  }
  DEBUG(dbgs() << "\t\textended: " << IntB << '\n');

  IntA.removeValNo(AValNo);
  DEBUG(dbgs() << "\t\ttrimmed:  " << IntA << '\n');
  ++numCommutes;
  return true;
}

/// reMaterializeTrivialDef - If the source of a copy is defined by a trivial
/// computation, replace the copy by rematerialize the definition.
bool RegisterCoalescer::reMaterializeTrivialDef(CoalescerPair &CP,
                                                MachineInstr *CopyMI,
                                                bool &IsDefCopy) {
  IsDefCopy = false;
  unsigned SrcReg = CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg();
  unsigned SrcIdx = CP.isFlipped() ? CP.getDstIdx() : CP.getSrcIdx();
  unsigned DstReg = CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg();
  unsigned DstIdx = CP.isFlipped() ? CP.getSrcIdx() : CP.getDstIdx();
  if (TargetRegisterInfo::isPhysicalRegister(SrcReg))
    return false;

  LiveInterval &SrcInt = LIS->getInterval(SrcReg);
  SlotIndex CopyIdx = LIS->getInstructionIndex(CopyMI);
  VNInfo *ValNo = SrcInt.Query(CopyIdx).valueIn();
  assert(ValNo && "CopyMI input register not live");
  if (ValNo->isPHIDef() || ValNo->isUnused())
    return false;
  MachineInstr *DefMI = LIS->getInstructionFromIndex(ValNo->def);
  if (!DefMI)
    return false;
  if (DefMI->isCopyLike()) {
    IsDefCopy = true;
    return false;
  }
  if (!DefMI->isAsCheapAsAMove())
    return false;
  if (!TII->isTriviallyReMaterializable(DefMI, AA))
    return false;
  bool SawStore = false;
  if (!DefMI->isSafeToMove(TII, AA, SawStore))
    return false;
  const MCInstrDesc &MCID = DefMI->getDesc();
  if (MCID.getNumDefs() != 1)
    return false;
  // Only support subregister destinations when the def is read-undef.
  MachineOperand &DstOperand = CopyMI->getOperand(0);
  unsigned CopyDstReg = DstOperand.getReg();
  if (DstOperand.getSubReg() && !DstOperand.isUndef())
    return false;

  // If both SrcIdx and DstIdx are set, correct rematerialization would widen
  // the register substantially (beyond both source and dest size). This is bad
  // for performance since it can cascade through a function, introducing many
  // extra spills and fills (e.g. ARM can easily end up copying QQQQPR registers
  // around after a few subreg copies).
  if (SrcIdx && DstIdx)
    return false;

  const TargetRegisterClass *DefRC = TII->getRegClass(MCID, 0, TRI, *MF);
  if (!DefMI->isImplicitDef()) {
    if (TargetRegisterInfo::isPhysicalRegister(DstReg)) {
      unsigned NewDstReg = DstReg;

      unsigned NewDstIdx = TRI->composeSubRegIndices(CP.getSrcIdx(),
                                              DefMI->getOperand(0).getSubReg());
      if (NewDstIdx)
        NewDstReg = TRI->getSubReg(DstReg, NewDstIdx);

      // Finally, make sure that the physical subregister that will be
      // constructed later is permitted for the instruction.
      if (!DefRC->contains(NewDstReg))
        return false;
    } else {
      // Theoretically, some stack frame reference could exist. Just make sure
      // it hasn't actually happened.
      assert(TargetRegisterInfo::isVirtualRegister(DstReg) &&
             "Only expect to deal with virtual or physical registers");
    }
  }

  MachineBasicBlock *MBB = CopyMI->getParent();
  MachineBasicBlock::iterator MII =
    std::next(MachineBasicBlock::iterator(CopyMI));
  TII->reMaterialize(*MBB, MII, DstReg, SrcIdx, DefMI, *TRI);
  MachineInstr *NewMI = std::prev(MII);

  LIS->ReplaceMachineInstrInMaps(CopyMI, NewMI);
  CopyMI->eraseFromParent();
  ErasedInstrs.insert(CopyMI);

  // NewMI may have dead implicit defs (E.g. EFLAGS for MOV<bits>r0 on X86).
  // We need to remember these so we can add intervals once we insert
  // NewMI into SlotIndexes.
  SmallVector<unsigned, 4> NewMIImplDefs;
  for (unsigned i = NewMI->getDesc().getNumOperands(),
         e = NewMI->getNumOperands(); i != e; ++i) {
    MachineOperand &MO = NewMI->getOperand(i);
    if (MO.isReg()) {
      assert(MO.isDef() && MO.isImplicit() && MO.isDead() &&
             TargetRegisterInfo::isPhysicalRegister(MO.getReg()));
      NewMIImplDefs.push_back(MO.getReg());
    }
  }

  if (TargetRegisterInfo::isVirtualRegister(DstReg)) {
    const TargetRegisterClass *NewRC = CP.getNewRC();
    unsigned NewIdx = NewMI->getOperand(0).getSubReg();

    if (NewIdx)
      NewRC = TRI->getMatchingSuperRegClass(NewRC, DefRC, NewIdx);
    else
      NewRC = TRI->getCommonSubClass(NewRC, DefRC);

    assert(NewRC && "subreg chosen for remat incompatible with instruction");
    MRI->setRegClass(DstReg, NewRC);

    updateRegDefsUses(DstReg, DstReg, DstIdx);
    NewMI->getOperand(0).setSubReg(NewIdx);
  } else if (NewMI->getOperand(0).getReg() != CopyDstReg) {
    // The New instruction may be defining a sub-register of what's actually
    // been asked for. If so it must implicitly define the whole thing.
    assert(TargetRegisterInfo::isPhysicalRegister(DstReg) &&
           "Only expect virtual or physical registers in remat");
    NewMI->getOperand(0).setIsDead(true);
    NewMI->addOperand(MachineOperand::CreateReg(CopyDstReg,
                                                true  /*IsDef*/,
                                                true  /*IsImp*/,
                                                false /*IsKill*/));
    // Record small dead def live-ranges for all the subregisters
    // of the destination register.
    // Otherwise, variables that live through may miss some
    // interferences, thus creating invalid allocation.
    // E.g., i386 code:
    // vreg1 = somedef ; vreg1 GR8
    // vreg2 = remat ; vreg2 GR32
    // CL = COPY vreg2.sub_8bit
    // = somedef vreg1 ; vreg1 GR8
    // =>
    // vreg1 = somedef ; vreg1 GR8
    // ECX<def, dead> = remat ; CL<imp-def>
    // = somedef vreg1 ; vreg1 GR8
    // vreg1 will see the inteferences with CL but not with CH since
    // no live-ranges would have been created for ECX.
    // Fix that!
    SlotIndex NewMIIdx = LIS->getInstructionIndex(NewMI);
    for (MCRegUnitIterator Units(NewMI->getOperand(0).getReg(), TRI);
         Units.isValid(); ++Units)
      if (LiveRange *LR = LIS->getCachedRegUnit(*Units))
        LR->createDeadDef(NewMIIdx.getRegSlot(), LIS->getVNInfoAllocator());
  }

  if (NewMI->getOperand(0).getSubReg())
    NewMI->getOperand(0).setIsUndef();

  // CopyMI may have implicit operands, transfer them over to the newly
  // rematerialized instruction. And update implicit def interval valnos.
  for (unsigned i = CopyMI->getDesc().getNumOperands(),
         e = CopyMI->getNumOperands(); i != e; ++i) {
    MachineOperand &MO = CopyMI->getOperand(i);
    if (MO.isReg()) {
      assert(MO.isImplicit() && "No explicit operands after implict operands.");
      // Discard VReg implicit defs.
      if (TargetRegisterInfo::isPhysicalRegister(MO.getReg())) {
        NewMI->addOperand(MO);
      }
    }
  }

  SlotIndex NewMIIdx = LIS->getInstructionIndex(NewMI);
  for (unsigned i = 0, e = NewMIImplDefs.size(); i != e; ++i) {
    unsigned Reg = NewMIImplDefs[i];
    for (MCRegUnitIterator Units(Reg, TRI); Units.isValid(); ++Units)
      if (LiveRange *LR = LIS->getCachedRegUnit(*Units))
        LR->createDeadDef(NewMIIdx.getRegSlot(), LIS->getVNInfoAllocator());
  }

  DEBUG(dbgs() << "Remat: " << *NewMI);
  ++NumReMats;

  // The source interval can become smaller because we removed a use.
  LIS->shrinkToUses(&SrcInt, &DeadDefs);
  if (!DeadDefs.empty())
    eliminateDeadDefs();

  return true;
}

/// eliminateUndefCopy - ProcessImpicitDefs may leave some copies of <undef>
/// values, it only removes local variables. When we have a copy like:
///
///   %vreg1 = COPY %vreg2<undef>
///
/// We delete the copy and remove the corresponding value number from %vreg1.
/// Any uses of that value number are marked as <undef>.
bool RegisterCoalescer::eliminateUndefCopy(MachineInstr *CopyMI,
                                           const CoalescerPair &CP) {
  SlotIndex Idx = LIS->getInstructionIndex(CopyMI);
  LiveInterval *SrcInt = &LIS->getInterval(CP.getSrcReg());
  if (SrcInt->liveAt(Idx))
    return false;
  LiveInterval *DstInt = &LIS->getInterval(CP.getDstReg());
  if (DstInt->liveAt(Idx))
    return false;

  // No intervals are live-in to CopyMI - it is undef.
  if (CP.isFlipped())
    DstInt = SrcInt;
  SrcInt = 0;

  VNInfo *DeadVNI = DstInt->getVNInfoAt(Idx.getRegSlot());
  assert(DeadVNI && "No value defined in DstInt");
  DstInt->removeValNo(DeadVNI);

  // Find new undef uses.
  for (MachineOperand &MO : MRI->reg_nodbg_operands(DstInt->reg)) {
    if (MO.isDef() || MO.isUndef())
      continue;
    MachineInstr *MI = MO.getParent();
    SlotIndex Idx = LIS->getInstructionIndex(MI);
    if (DstInt->liveAt(Idx))
      continue;
    MO.setIsUndef(true);
    DEBUG(dbgs() << "\tnew undef: " << Idx << '\t' << *MI);
  }
  return true;
}

/// updateRegDefsUses - Replace all defs and uses of SrcReg to DstReg and
/// update the subregister number if it is not zero. If DstReg is a
/// physical register and the existing subregister number of the def / use
/// being updated is not zero, make sure to set it to the correct physical
/// subregister.
void RegisterCoalescer::updateRegDefsUses(unsigned SrcReg,
                                          unsigned DstReg,
                                          unsigned SubIdx) {
  bool DstIsPhys = TargetRegisterInfo::isPhysicalRegister(DstReg);
  LiveInterval *DstInt = DstIsPhys ? 0 : &LIS->getInterval(DstReg);

  SmallPtrSet<MachineInstr*, 8> Visited;
  for (MachineRegisterInfo::reg_instr_iterator
       I = MRI->reg_instr_begin(SrcReg), E = MRI->reg_instr_end();
       I != E; ) {
    MachineInstr *UseMI = &*(I++);

    // Each instruction can only be rewritten once because sub-register
    // composition is not always idempotent. When SrcReg != DstReg, rewriting
    // the UseMI operands removes them from the SrcReg use-def chain, but when
    // SrcReg is DstReg we could encounter UseMI twice if it has multiple
    // operands mentioning the virtual register.
    if (SrcReg == DstReg && !Visited.insert(UseMI))
      continue;

    SmallVector<unsigned,8> Ops;
    bool Reads, Writes;
    std::tie(Reads, Writes) = UseMI->readsWritesVirtualRegister(SrcReg, &Ops);

    // If SrcReg wasn't read, it may still be the case that DstReg is live-in
    // because SrcReg is a sub-register.
    if (DstInt && !Reads && SubIdx)
      Reads = DstInt->liveAt(LIS->getInstructionIndex(UseMI));

    // Replace SrcReg with DstReg in all UseMI operands.
    for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
      MachineOperand &MO = UseMI->getOperand(Ops[i]);

      // Adjust <undef> flags in case of sub-register joins. We don't want to
      // turn a full def into a read-modify-write sub-register def and vice
      // versa.
      if (SubIdx && MO.isDef())
        MO.setIsUndef(!Reads);

      if (DstIsPhys)
        MO.substPhysReg(DstReg, *TRI);
      else
        MO.substVirtReg(DstReg, SubIdx, *TRI);
    }

    DEBUG({
        dbgs() << "\t\tupdated: ";
        if (!UseMI->isDebugValue())
          dbgs() << LIS->getInstructionIndex(UseMI) << "\t";
        dbgs() << *UseMI;
      });
  }
}

/// canJoinPhys - Return true if a copy involving a physreg should be joined.
bool RegisterCoalescer::canJoinPhys(const CoalescerPair &CP) {
  /// Always join simple intervals that are defined by a single copy from a
  /// reserved register. This doesn't increase register pressure, so it is
  /// always beneficial.
  if (!MRI->isReserved(CP.getDstReg())) {
    DEBUG(dbgs() << "\tCan only merge into reserved registers.\n");
    return false;
  }

  LiveInterval &JoinVInt = LIS->getInterval(CP.getSrcReg());
  if (CP.isFlipped() && JoinVInt.containsOneValue())
    return true;

  DEBUG(dbgs() << "\tCannot join defs into reserved register.\n");
  return false;
}

/// joinCopy - Attempt to join intervals corresponding to SrcReg/DstReg,
/// which are the src/dst of the copy instruction CopyMI.  This returns true
/// if the copy was successfully coalesced away. If it is not currently
/// possible to coalesce this interval, but it may be possible if other
/// things get coalesced, then it returns true by reference in 'Again'.
bool RegisterCoalescer::joinCopy(MachineInstr *CopyMI, bool &Again) {

  Again = false;
  DEBUG(dbgs() << LIS->getInstructionIndex(CopyMI) << '\t' << *CopyMI);

  CoalescerPair CP(*TRI);
  if (!CP.setRegisters(CopyMI)) {
    DEBUG(dbgs() << "\tNot coalescable.\n");
    return false;
  }

  // Dead code elimination. This really should be handled by MachineDCE, but
  // sometimes dead copies slip through, and we can't generate invalid live
  // ranges.
  if (!CP.isPhys() && CopyMI->allDefsAreDead()) {
    DEBUG(dbgs() << "\tCopy is dead.\n");
    DeadDefs.push_back(CopyMI);
    eliminateDeadDefs();
    return true;
  }

  // Eliminate undefs.
  if (!CP.isPhys() && eliminateUndefCopy(CopyMI, CP)) {
    DEBUG(dbgs() << "\tEliminated copy of <undef> value.\n");
    LIS->RemoveMachineInstrFromMaps(CopyMI);
    CopyMI->eraseFromParent();
    return false;  // Not coalescable.
  }

  // Coalesced copies are normally removed immediately, but transformations
  // like removeCopyByCommutingDef() can inadvertently create identity copies.
  // When that happens, just join the values and remove the copy.
  if (CP.getSrcReg() == CP.getDstReg()) {
    LiveInterval &LI = LIS->getInterval(CP.getSrcReg());
    DEBUG(dbgs() << "\tCopy already coalesced: " << LI << '\n');
    LiveQueryResult LRQ = LI.Query(LIS->getInstructionIndex(CopyMI));
    if (VNInfo *DefVNI = LRQ.valueDefined()) {
      VNInfo *ReadVNI = LRQ.valueIn();
      assert(ReadVNI && "No value before copy and no <undef> flag.");
      assert(ReadVNI != DefVNI && "Cannot read and define the same value.");
      LI.MergeValueNumberInto(DefVNI, ReadVNI);
      DEBUG(dbgs() << "\tMerged values:          " << LI << '\n');
    }
    LIS->RemoveMachineInstrFromMaps(CopyMI);
    CopyMI->eraseFromParent();
    return true;
  }

  // Enforce policies.
  if (CP.isPhys()) {
    DEBUG(dbgs() << "\tConsidering merging " << PrintReg(CP.getSrcReg(), TRI)
                 << " with " << PrintReg(CP.getDstReg(), TRI, CP.getSrcIdx())
                 << '\n');
    if (!canJoinPhys(CP)) {
      // Before giving up coalescing, if definition of source is defined by
      // trivial computation, try rematerializing it.
      bool IsDefCopy;
      if (reMaterializeTrivialDef(CP, CopyMI, IsDefCopy))
        return true;
      if (IsDefCopy)
        Again = true;  // May be possible to coalesce later.
      return false;
    }
  } else {
    DEBUG({
      dbgs() << "\tConsidering merging to " << CP.getNewRC()->getName()
             << " with ";
      if (CP.getDstIdx() && CP.getSrcIdx())
        dbgs() << PrintReg(CP.getDstReg()) << " in "
               << TRI->getSubRegIndexName(CP.getDstIdx()) << " and "
               << PrintReg(CP.getSrcReg()) << " in "
               << TRI->getSubRegIndexName(CP.getSrcIdx()) << '\n';
      else
        dbgs() << PrintReg(CP.getSrcReg(), TRI) << " in "
               << PrintReg(CP.getDstReg(), TRI, CP.getSrcIdx()) << '\n';
    });

    // When possible, let DstReg be the larger interval.
    if (!CP.isPartial() && LIS->getInterval(CP.getSrcReg()).size() >
                           LIS->getInterval(CP.getDstReg()).size())
      CP.flip();
  }

  // Okay, attempt to join these two intervals.  On failure, this returns false.
  // Otherwise, if one of the intervals being joined is a physreg, this method
  // always canonicalizes DstInt to be it.  The output "SrcInt" will not have
  // been modified, so we can use this information below to update aliases.
  if (!joinIntervals(CP)) {
    // Coalescing failed.

    // If definition of source is defined by trivial computation, try
    // rematerializing it.
    bool IsDefCopy;
    if (reMaterializeTrivialDef(CP, CopyMI, IsDefCopy))
      return true;

    // If we can eliminate the copy without merging the live segments, do so
    // now.
    if (!CP.isPartial() && !CP.isPhys()) {
      if (adjustCopiesBackFrom(CP, CopyMI) ||
          removeCopyByCommutingDef(CP, CopyMI)) {
        LIS->RemoveMachineInstrFromMaps(CopyMI);
        CopyMI->eraseFromParent();
        DEBUG(dbgs() << "\tTrivial!\n");
        return true;
      }
    }

    // Otherwise, we are unable to join the intervals.
    DEBUG(dbgs() << "\tInterference!\n");
    Again = true;  // May be possible to coalesce later.
    return false;
  }

  // Coalescing to a virtual register that is of a sub-register class of the
  // other. Make sure the resulting register is set to the right register class.
  if (CP.isCrossClass()) {
    ++numCrossRCs;
    MRI->setRegClass(CP.getDstReg(), CP.getNewRC());
  }

  // Removing sub-register copies can ease the register class constraints.
  // Make sure we attempt to inflate the register class of DstReg.
  if (!CP.isPhys() && RegClassInfo.isProperSubClass(CP.getNewRC()))
    InflateRegs.push_back(CP.getDstReg());

  // CopyMI has been erased by joinIntervals at this point. Remove it from
  // ErasedInstrs since copyCoalesceWorkList() won't add a successful join back
  // to the work list. This keeps ErasedInstrs from growing needlessly.
  ErasedInstrs.erase(CopyMI);

  // Rewrite all SrcReg operands to DstReg.
  // Also update DstReg operands to include DstIdx if it is set.
  if (CP.getDstIdx())
    updateRegDefsUses(CP.getDstReg(), CP.getDstReg(), CP.getDstIdx());
  updateRegDefsUses(CP.getSrcReg(), CP.getDstReg(), CP.getSrcIdx());

  // SrcReg is guaranteed to be the register whose live interval that is
  // being merged.
  LIS->removeInterval(CP.getSrcReg());

  // Update regalloc hint.
  TRI->UpdateRegAllocHint(CP.getSrcReg(), CP.getDstReg(), *MF);

  DEBUG({
    dbgs() << "\tJoined. Result = ";
    if (CP.isPhys())
      dbgs() << PrintReg(CP.getDstReg(), TRI);
    else
      dbgs() << LIS->getInterval(CP.getDstReg());
    dbgs() << '\n';
  });

  ++numJoins;
  return true;
}

/// Attempt joining with a reserved physreg.
bool RegisterCoalescer::joinReservedPhysReg(CoalescerPair &CP) {
  assert(CP.isPhys() && "Must be a physreg copy");
  assert(MRI->isReserved(CP.getDstReg()) && "Not a reserved register");
  LiveInterval &RHS = LIS->getInterval(CP.getSrcReg());
  DEBUG(dbgs() << "\t\tRHS = " << RHS << '\n');

  assert(CP.isFlipped() && RHS.containsOneValue() &&
         "Invalid join with reserved register");

  // Optimization for reserved registers like ESP. We can only merge with a
  // reserved physreg if RHS has a single value that is a copy of CP.DstReg().
  // The live range of the reserved register will look like a set of dead defs
  // - we don't properly track the live range of reserved registers.

  // Deny any overlapping intervals.  This depends on all the reserved
  // register live ranges to look like dead defs.
  for (MCRegUnitIterator UI(CP.getDstReg(), TRI); UI.isValid(); ++UI)
    if (RHS.overlaps(LIS->getRegUnit(*UI))) {
      DEBUG(dbgs() << "\t\tInterference: " << PrintRegUnit(*UI, TRI) << '\n');
      return false;
    }

  // Skip any value computations, we are not adding new values to the
  // reserved register.  Also skip merging the live ranges, the reserved
  // register live range doesn't need to be accurate as long as all the
  // defs are there.

  // Delete the identity copy.
  MachineInstr *CopyMI = MRI->getVRegDef(RHS.reg);
  LIS->RemoveMachineInstrFromMaps(CopyMI);
  CopyMI->eraseFromParent();

  // We don't track kills for reserved registers.
  MRI->clearKillFlags(CP.getSrcReg());

  return true;
}

//===----------------------------------------------------------------------===//
//                 Interference checking and interval joining
//===----------------------------------------------------------------------===//
//
// In the easiest case, the two live ranges being joined are disjoint, and
// there is no interference to consider. It is quite common, though, to have
// overlapping live ranges, and we need to check if the interference can be
// resolved.
//
// The live range of a single SSA value forms a sub-tree of the dominator tree.
// This means that two SSA values overlap if and only if the def of one value
// is contained in the live range of the other value. As a special case, the
// overlapping values can be defined at the same index.
//
// The interference from an overlapping def can be resolved in these cases:
//
// 1. Coalescable copies. The value is defined by a copy that would become an
//    identity copy after joining SrcReg and DstReg. The copy instruction will
//    be removed, and the value will be merged with the source value.
//
//    There can be several copies back and forth, causing many values to be
//    merged into one. We compute a list of ultimate values in the joined live
//    range as well as a mappings from the old value numbers.
//
// 2. IMPLICIT_DEF. This instruction is only inserted to ensure all PHI
//    predecessors have a live out value. It doesn't cause real interference,
//    and can be merged into the value it overlaps. Like a coalescable copy, it
//    can be erased after joining.
//
// 3. Copy of external value. The overlapping def may be a copy of a value that
//    is already in the other register. This is like a coalescable copy, but
//    the live range of the source register must be trimmed after erasing the
//    copy instruction:
//
//      %src = COPY %ext
//      %dst = COPY %ext  <-- Remove this COPY, trim the live range of %ext.
//
// 4. Clobbering undefined lanes. Vector registers are sometimes built by
//    defining one lane at a time:
//
//      %dst:ssub0<def,read-undef> = FOO
//      %src = BAR
//      %dst:ssub1<def> = COPY %src
//
//    The live range of %src overlaps the %dst value defined by FOO, but
//    merging %src into %dst:ssub1 is only going to clobber the ssub1 lane
//    which was undef anyway.
//
//    The value mapping is more complicated in this case. The final live range
//    will have different value numbers for both FOO and BAR, but there is no
//    simple mapping from old to new values. It may even be necessary to add
//    new PHI values.
//
// 5. Clobbering dead lanes. A def may clobber a lane of a vector register that
//    is live, but never read. This can happen because we don't compute
//    individual live ranges per lane.
//
//      %dst<def> = FOO
//      %src = BAR
//      %dst:ssub1<def> = COPY %src
//
//    This kind of interference is only resolved locally. If the clobbered
//    lane value escapes the block, the join is aborted.

namespace {
/// Track information about values in a single virtual register about to be
/// joined. Objects of this class are always created in pairs - one for each
/// side of the CoalescerPair.
class JoinVals {
  LiveInterval &LI;

  // Location of this register in the final joined register.
  // Either CP.DstIdx or CP.SrcIdx.
  unsigned SubIdx;

  // Values that will be present in the final live range.
  SmallVectorImpl<VNInfo*> &NewVNInfo;

  const CoalescerPair &CP;
  LiveIntervals *LIS;
  SlotIndexes *Indexes;
  const TargetRegisterInfo *TRI;

  // Value number assignments. Maps value numbers in LI to entries in NewVNInfo.
  // This is suitable for passing to LiveInterval::join().
  SmallVector<int, 8> Assignments;

  // Conflict resolution for overlapping values.
  enum ConflictResolution {
    // No overlap, simply keep this value.
    CR_Keep,

    // Merge this value into OtherVNI and erase the defining instruction.
    // Used for IMPLICIT_DEF, coalescable copies, and copies from external
    // values.
    CR_Erase,

    // Merge this value into OtherVNI but keep the defining instruction.
    // This is for the special case where OtherVNI is defined by the same
    // instruction.
    CR_Merge,

    // Keep this value, and have it replace OtherVNI where possible. This
    // complicates value mapping since OtherVNI maps to two different values
    // before and after this def.
    // Used when clobbering undefined or dead lanes.
    CR_Replace,

    // Unresolved conflict. Visit later when all values have been mapped.
    CR_Unresolved,

    // Unresolvable conflict. Abort the join.
    CR_Impossible
  };

  // Per-value info for LI. The lane bit masks are all relative to the final
  // joined register, so they can be compared directly between SrcReg and
  // DstReg.
  struct Val {
    ConflictResolution Resolution;

    // Lanes written by this def, 0 for unanalyzed values.
    unsigned WriteLanes;

    // Lanes with defined values in this register. Other lanes are undef and
    // safe to clobber.
    unsigned ValidLanes;

    // Value in LI being redefined by this def.
    VNInfo *RedefVNI;

    // Value in the other live range that overlaps this def, if any.
    VNInfo *OtherVNI;

    // Is this value an IMPLICIT_DEF that can be erased?
    //
    // IMPLICIT_DEF values should only exist at the end of a basic block that
    // is a predecessor to a phi-value. These IMPLICIT_DEF instructions can be
    // safely erased if they are overlapping a live value in the other live
    // interval.
    //
    // Weird control flow graphs and incomplete PHI handling in
    // ProcessImplicitDefs can very rarely create IMPLICIT_DEF values with
    // longer live ranges. Such IMPLICIT_DEF values should be treated like
    // normal values.
    bool ErasableImplicitDef;

    // True when the live range of this value will be pruned because of an
    // overlapping CR_Replace value in the other live range.
    bool Pruned;

    // True once Pruned above has been computed.
    bool PrunedComputed;

    Val() : Resolution(CR_Keep), WriteLanes(0), ValidLanes(0),
            RedefVNI(0), OtherVNI(0), ErasableImplicitDef(false),
            Pruned(false), PrunedComputed(false) {}

    bool isAnalyzed() const { return WriteLanes != 0; }
  };

  // One entry per value number in LI.
  SmallVector<Val, 8> Vals;

  unsigned computeWriteLanes(const MachineInstr *DefMI, bool &Redef);
  VNInfo *stripCopies(VNInfo *VNI);
  ConflictResolution analyzeValue(unsigned ValNo, JoinVals &Other);
  void computeAssignment(unsigned ValNo, JoinVals &Other);
  bool taintExtent(unsigned, unsigned, JoinVals&,
                   SmallVectorImpl<std::pair<SlotIndex, unsigned> >&);
  bool usesLanes(MachineInstr *MI, unsigned, unsigned, unsigned);
  bool isPrunedValue(unsigned ValNo, JoinVals &Other);

public:
  JoinVals(LiveInterval &li, unsigned subIdx,
           SmallVectorImpl<VNInfo*> &newVNInfo,
           const CoalescerPair &cp,
           LiveIntervals *lis,
           const TargetRegisterInfo *tri)
    : LI(li), SubIdx(subIdx), NewVNInfo(newVNInfo), CP(cp), LIS(lis),
      Indexes(LIS->getSlotIndexes()), TRI(tri),
      Assignments(LI.getNumValNums(), -1), Vals(LI.getNumValNums())
  {}

  /// Analyze defs in LI and compute a value mapping in NewVNInfo.
  /// Returns false if any conflicts were impossible to resolve.
  bool mapValues(JoinVals &Other);

  /// Try to resolve conflicts that require all values to be mapped.
  /// Returns false if any conflicts were impossible to resolve.
  bool resolveConflicts(JoinVals &Other);

  /// Prune the live range of values in Other.LI where they would conflict with
  /// CR_Replace values in LI. Collect end points for restoring the live range
  /// after joining.
  void pruneValues(JoinVals &Other, SmallVectorImpl<SlotIndex> &EndPoints);

  /// Erase any machine instructions that have been coalesced away.
  /// Add erased instructions to ErasedInstrs.
  /// Add foreign virtual registers to ShrinkRegs if their live range ended at
  /// the erased instrs.
  void eraseInstrs(SmallPtrSet<MachineInstr*, 8> &ErasedInstrs,
                   SmallVectorImpl<unsigned> &ShrinkRegs);

  /// Get the value assignments suitable for passing to LiveInterval::join.
  const int *getAssignments() const { return Assignments.data(); }
};
} // end anonymous namespace

/// Compute the bitmask of lanes actually written by DefMI.
/// Set Redef if there are any partial register definitions that depend on the
/// previous value of the register.
unsigned JoinVals::computeWriteLanes(const MachineInstr *DefMI, bool &Redef) {
  unsigned L = 0;
  for (ConstMIOperands MO(DefMI); MO.isValid(); ++MO) {
    if (!MO->isReg() || MO->getReg() != LI.reg || !MO->isDef())
      continue;
    L |= TRI->getSubRegIndexLaneMask(
           TRI->composeSubRegIndices(SubIdx, MO->getSubReg()));
    if (MO->readsReg())
      Redef = true;
  }
  return L;
}

/// Find the ultimate value that VNI was copied from.
VNInfo *JoinVals::stripCopies(VNInfo *VNI) {
  while (!VNI->isPHIDef()) {
    MachineInstr *MI = Indexes->getInstructionFromIndex(VNI->def);
    assert(MI && "No defining instruction");
    if (!MI->isFullCopy())
      break;
    unsigned Reg = MI->getOperand(1).getReg();
    if (!TargetRegisterInfo::isVirtualRegister(Reg))
      break;
    LiveQueryResult LRQ = LIS->getInterval(Reg).Query(VNI->def);
    if (!LRQ.valueIn())
      break;
    VNI = LRQ.valueIn();
  }
  return VNI;
}

/// Analyze ValNo in this live range, and set all fields of Vals[ValNo].
/// Return a conflict resolution when possible, but leave the hard cases as
/// CR_Unresolved.
/// Recursively calls computeAssignment() on this and Other, guaranteeing that
/// both OtherVNI and RedefVNI have been analyzed and mapped before returning.
/// The recursion always goes upwards in the dominator tree, making loops
/// impossible.
JoinVals::ConflictResolution
JoinVals::analyzeValue(unsigned ValNo, JoinVals &Other) {
  Val &V = Vals[ValNo];
  assert(!V.isAnalyzed() && "Value has already been analyzed!");
  VNInfo *VNI = LI.getValNumInfo(ValNo);
  if (VNI->isUnused()) {
    V.WriteLanes = ~0u;
    return CR_Keep;
  }

  // Get the instruction defining this value, compute the lanes written.
  const MachineInstr *DefMI = 0;
  if (VNI->isPHIDef()) {
    // Conservatively assume that all lanes in a PHI are valid.
    V.ValidLanes = V.WriteLanes = TRI->getSubRegIndexLaneMask(SubIdx);
  } else {
    DefMI = Indexes->getInstructionFromIndex(VNI->def);
    bool Redef = false;
    V.ValidLanes = V.WriteLanes = computeWriteLanes(DefMI, Redef);

    // If this is a read-modify-write instruction, there may be more valid
    // lanes than the ones written by this instruction.
    // This only covers partial redef operands. DefMI may have normal use
    // operands reading the register. They don't contribute valid lanes.
    //
    // This adds ssub1 to the set of valid lanes in %src:
    //
    //   %src:ssub1<def> = FOO
    //
    // This leaves only ssub1 valid, making any other lanes undef:
    //
    //   %src:ssub1<def,read-undef> = FOO %src:ssub2
    //
    // The <read-undef> flag on the def operand means that old lane values are
    // not important.
    if (Redef) {
      V.RedefVNI = LI.Query(VNI->def).valueIn();
      assert(V.RedefVNI && "Instruction is reading nonexistent value");
      computeAssignment(V.RedefVNI->id, Other);
      V.ValidLanes |= Vals[V.RedefVNI->id].ValidLanes;
    }

    // An IMPLICIT_DEF writes undef values.
    if (DefMI->isImplicitDef()) {
      // We normally expect IMPLICIT_DEF values to be live only until the end
      // of their block. If the value is really live longer and gets pruned in
      // another block, this flag is cleared again.
      V.ErasableImplicitDef = true;
      V.ValidLanes &= ~V.WriteLanes;
    }
  }

  // Find the value in Other that overlaps VNI->def, if any.
  LiveQueryResult OtherLRQ = Other.LI.Query(VNI->def);

  // It is possible that both values are defined by the same instruction, or
  // the values are PHIs defined in the same block. When that happens, the two
  // values should be merged into one, but not into any preceding value.
  // The first value defined or visited gets CR_Keep, the other gets CR_Merge.
  if (VNInfo *OtherVNI = OtherLRQ.valueDefined()) {
    assert(SlotIndex::isSameInstr(VNI->def, OtherVNI->def) && "Broken LRQ");

    // One value stays, the other is merged. Keep the earlier one, or the first
    // one we see.
    if (OtherVNI->def < VNI->def)
      Other.computeAssignment(OtherVNI->id, *this);
    else if (VNI->def < OtherVNI->def && OtherLRQ.valueIn()) {
      // This is an early-clobber def overlapping a live-in value in the other
      // register. Not mergeable.
      V.OtherVNI = OtherLRQ.valueIn();
      return CR_Impossible;
    }
    V.OtherVNI = OtherVNI;
    Val &OtherV = Other.Vals[OtherVNI->id];
    // Keep this value, check for conflicts when analyzing OtherVNI.
    if (!OtherV.isAnalyzed())
      return CR_Keep;
    // Both sides have been analyzed now.
    // Allow overlapping PHI values. Any real interference would show up in a
    // predecessor, the PHI itself can't introduce any conflicts.
    if (VNI->isPHIDef())
      return CR_Merge;
    if (V.ValidLanes & OtherV.ValidLanes)
      // Overlapping lanes can't be resolved.
      return CR_Impossible;
    else
      return CR_Merge;
  }

  // No simultaneous def. Is Other live at the def?
  V.OtherVNI = OtherLRQ.valueIn();
  if (!V.OtherVNI)
    // No overlap, no conflict.
    return CR_Keep;

  assert(!SlotIndex::isSameInstr(VNI->def, V.OtherVNI->def) && "Broken LRQ");

  // We have overlapping values, or possibly a kill of Other.
  // Recursively compute assignments up the dominator tree.
  Other.computeAssignment(V.OtherVNI->id, *this);
  Val &OtherV = Other.Vals[V.OtherVNI->id];

  // Check if OtherV is an IMPLICIT_DEF that extends beyond its basic block.
  // This shouldn't normally happen, but ProcessImplicitDefs can leave such
  // IMPLICIT_DEF instructions behind, and there is nothing wrong with it
  // technically.
  //
  // WHen it happens, treat that IMPLICIT_DEF as a normal value, and don't try
  // to erase the IMPLICIT_DEF instruction.
  if (OtherV.ErasableImplicitDef && DefMI &&
      DefMI->getParent() != Indexes->getMBBFromIndex(V.OtherVNI->def)) {
    DEBUG(dbgs() << "IMPLICIT_DEF defined at " << V.OtherVNI->def
                 << " extends into BB#" << DefMI->getParent()->getNumber()
                 << ", keeping it.\n");
    OtherV.ErasableImplicitDef = false;
  }

  // Allow overlapping PHI values. Any real interference would show up in a
  // predecessor, the PHI itself can't introduce any conflicts.
  if (VNI->isPHIDef())
    return CR_Replace;

  // Check for simple erasable conflicts.
  if (DefMI->isImplicitDef())
    return CR_Erase;

  // Include the non-conflict where DefMI is a coalescable copy that kills
  // OtherVNI. We still want the copy erased and value numbers merged.
  if (CP.isCoalescable(DefMI)) {
    // Some of the lanes copied from OtherVNI may be undef, making them undef
    // here too.
    V.ValidLanes &= ~V.WriteLanes | OtherV.ValidLanes;
    return CR_Erase;
  }

  // This may not be a real conflict if DefMI simply kills Other and defines
  // VNI.
  if (OtherLRQ.isKill() && OtherLRQ.endPoint() <= VNI->def)
    return CR_Keep;

  // Handle the case where VNI and OtherVNI can be proven to be identical:
  //
  //   %other = COPY %ext
  //   %this  = COPY %ext <-- Erase this copy
  //
  if (DefMI->isFullCopy() && !CP.isPartial() &&
      stripCopies(VNI) == stripCopies(V.OtherVNI))
    return CR_Erase;

  // If the lanes written by this instruction were all undef in OtherVNI, it is
  // still safe to join the live ranges. This can't be done with a simple value
  // mapping, though - OtherVNI will map to multiple values:
  //
  //   1 %dst:ssub0 = FOO                <-- OtherVNI
  //   2 %src = BAR                      <-- VNI
  //   3 %dst:ssub1 = COPY %src<kill>    <-- Eliminate this copy.
  //   4 BAZ %dst<kill>
  //   5 QUUX %src<kill>
  //
  // Here OtherVNI will map to itself in [1;2), but to VNI in [2;5). CR_Replace
  // handles this complex value mapping.
  if ((V.WriteLanes & OtherV.ValidLanes) == 0)
    return CR_Replace;

  // If the other live range is killed by DefMI and the live ranges are still
  // overlapping, it must be because we're looking at an early clobber def:
  //
  //   %dst<def,early-clobber> = ASM %src<kill>
  //
  // In this case, it is illegal to merge the two live ranges since the early
  // clobber def would clobber %src before it was read.
  if (OtherLRQ.isKill()) {
    // This case where the def doesn't overlap the kill is handled above.
    assert(VNI->def.isEarlyClobber() &&
           "Only early clobber defs can overlap a kill");
    return CR_Impossible;
  }

  // VNI is clobbering live lanes in OtherVNI, but there is still the
  // possibility that no instructions actually read the clobbered lanes.
  // If we're clobbering all the lanes in OtherVNI, at least one must be read.
  // Otherwise Other.LI wouldn't be live here.
  if ((TRI->getSubRegIndexLaneMask(Other.SubIdx) & ~V.WriteLanes) == 0)
    return CR_Impossible;

  // We need to verify that no instructions are reading the clobbered lanes. To
  // save compile time, we'll only check that locally. Don't allow the tainted
  // value to escape the basic block.
  MachineBasicBlock *MBB = Indexes->getMBBFromIndex(VNI->def);
  if (OtherLRQ.endPoint() >= Indexes->getMBBEndIdx(MBB))
    return CR_Impossible;

  // There are still some things that could go wrong besides clobbered lanes
  // being read, for example OtherVNI may be only partially redefined in MBB,
  // and some clobbered lanes could escape the block. Save this analysis for
  // resolveConflicts() when all values have been mapped. We need to know
  // RedefVNI and WriteLanes for any later defs in MBB, and we can't compute
  // that now - the recursive analyzeValue() calls must go upwards in the
  // dominator tree.
  return CR_Unresolved;
}

/// Compute the value assignment for ValNo in LI.
/// This may be called recursively by analyzeValue(), but never for a ValNo on
/// the stack.
void JoinVals::computeAssignment(unsigned ValNo, JoinVals &Other) {
  Val &V = Vals[ValNo];
  if (V.isAnalyzed()) {
    // Recursion should always move up the dominator tree, so ValNo is not
    // supposed to reappear before it has been assigned.
    assert(Assignments[ValNo] != -1 && "Bad recursion?");
    return;
  }
  switch ((V.Resolution = analyzeValue(ValNo, Other))) {
  case CR_Erase:
  case CR_Merge:
    // Merge this ValNo into OtherVNI.
    assert(V.OtherVNI && "OtherVNI not assigned, can't merge.");
    assert(Other.Vals[V.OtherVNI->id].isAnalyzed() && "Missing recursion");
    Assignments[ValNo] = Other.Assignments[V.OtherVNI->id];
    DEBUG(dbgs() << "\t\tmerge " << PrintReg(LI.reg) << ':' << ValNo << '@'
                 << LI.getValNumInfo(ValNo)->def << " into "
                 << PrintReg(Other.LI.reg) << ':' << V.OtherVNI->id << '@'
                 << V.OtherVNI->def << " --> @"
                 << NewVNInfo[Assignments[ValNo]]->def << '\n');
    break;
  case CR_Replace:
  case CR_Unresolved:
    // The other value is going to be pruned if this join is successful.
    assert(V.OtherVNI && "OtherVNI not assigned, can't prune");
    Other.Vals[V.OtherVNI->id].Pruned = true;
    // Fall through.
  default:
    // This value number needs to go in the final joined live range.
    Assignments[ValNo] = NewVNInfo.size();
    NewVNInfo.push_back(LI.getValNumInfo(ValNo));
    break;
  }
}

bool JoinVals::mapValues(JoinVals &Other) {
  for (unsigned i = 0, e = LI.getNumValNums(); i != e; ++i) {
    computeAssignment(i, Other);
    if (Vals[i].Resolution == CR_Impossible) {
      DEBUG(dbgs() << "\t\tinterference at " << PrintReg(LI.reg) << ':' << i
                   << '@' << LI.getValNumInfo(i)->def << '\n');
      return false;
    }
  }
  return true;
}

/// Assuming ValNo is going to clobber some valid lanes in Other.LI, compute
/// the extent of the tainted lanes in the block.
///
/// Multiple values in Other.LI can be affected since partial redefinitions can
/// preserve previously tainted lanes.
///
///   1 %dst = VLOAD           <-- Define all lanes in %dst
///   2 %src = FOO             <-- ValNo to be joined with %dst:ssub0
///   3 %dst:ssub1 = BAR       <-- Partial redef doesn't clear taint in ssub0
///   4 %dst:ssub0 = COPY %src <-- Conflict resolved, ssub0 wasn't read
///
/// For each ValNo in Other that is affected, add an (EndIndex, TaintedLanes)
/// entry to TaintedVals.
///
/// Returns false if the tainted lanes extend beyond the basic block.
bool JoinVals::
taintExtent(unsigned ValNo, unsigned TaintedLanes, JoinVals &Other,
            SmallVectorImpl<std::pair<SlotIndex, unsigned> > &TaintExtent) {
  VNInfo *VNI = LI.getValNumInfo(ValNo);
  MachineBasicBlock *MBB = Indexes->getMBBFromIndex(VNI->def);
  SlotIndex MBBEnd = Indexes->getMBBEndIdx(MBB);

  // Scan Other.LI from VNI.def to MBBEnd.
  LiveInterval::iterator OtherI = Other.LI.find(VNI->def);
  assert(OtherI != Other.LI.end() && "No conflict?");
  do {
    // OtherI is pointing to a tainted value. Abort the join if the tainted
    // lanes escape the block.
    SlotIndex End = OtherI->end;
    if (End >= MBBEnd) {
      DEBUG(dbgs() << "\t\ttaints global " << PrintReg(Other.LI.reg) << ':'
                   << OtherI->valno->id << '@' << OtherI->start << '\n');
      return false;
    }
    DEBUG(dbgs() << "\t\ttaints local " << PrintReg(Other.LI.reg) << ':'
                 << OtherI->valno->id << '@' << OtherI->start
                 << " to " << End << '\n');
    // A dead def is not a problem.
    if (End.isDead())
      break;
    TaintExtent.push_back(std::make_pair(End, TaintedLanes));

    // Check for another def in the MBB.
    if (++OtherI == Other.LI.end() || OtherI->start >= MBBEnd)
      break;

    // Lanes written by the new def are no longer tainted.
    const Val &OV = Other.Vals[OtherI->valno->id];
    TaintedLanes &= ~OV.WriteLanes;
    if (!OV.RedefVNI)
      break;
  } while (TaintedLanes);
  return true;
}

/// Return true if MI uses any of the given Lanes from Reg.
/// This does not include partial redefinitions of Reg.
bool JoinVals::usesLanes(MachineInstr *MI, unsigned Reg, unsigned SubIdx,
                         unsigned Lanes) {
  if (MI->isDebugValue())
    return false;
  for (ConstMIOperands MO(MI); MO.isValid(); ++MO) {
    if (!MO->isReg() || MO->isDef() || MO->getReg() != Reg)
      continue;
    if (!MO->readsReg())
      continue;
    if (Lanes & TRI->getSubRegIndexLaneMask(
                  TRI->composeSubRegIndices(SubIdx, MO->getSubReg())))
      return true;
  }
  return false;
}

bool JoinVals::resolveConflicts(JoinVals &Other) {
  for (unsigned i = 0, e = LI.getNumValNums(); i != e; ++i) {
    Val &V = Vals[i];
    assert (V.Resolution != CR_Impossible && "Unresolvable conflict");
    if (V.Resolution != CR_Unresolved)
      continue;
    DEBUG(dbgs() << "\t\tconflict at " << PrintReg(LI.reg) << ':' << i
                 << '@' << LI.getValNumInfo(i)->def << '\n');
    ++NumLaneConflicts;
    assert(V.OtherVNI && "Inconsistent conflict resolution.");
    VNInfo *VNI = LI.getValNumInfo(i);
    const Val &OtherV = Other.Vals[V.OtherVNI->id];

    // VNI is known to clobber some lanes in OtherVNI. If we go ahead with the
    // join, those lanes will be tainted with a wrong value. Get the extent of
    // the tainted lanes.
    unsigned TaintedLanes = V.WriteLanes & OtherV.ValidLanes;
    SmallVector<std::pair<SlotIndex, unsigned>, 8> TaintExtent;
    if (!taintExtent(i, TaintedLanes, Other, TaintExtent))
      // Tainted lanes would extend beyond the basic block.
      return false;

    assert(!TaintExtent.empty() && "There should be at least one conflict.");

    // Now look at the instructions from VNI->def to TaintExtent (inclusive).
    MachineBasicBlock *MBB = Indexes->getMBBFromIndex(VNI->def);
    MachineBasicBlock::iterator MI = MBB->begin();
    if (!VNI->isPHIDef()) {
      MI = Indexes->getInstructionFromIndex(VNI->def);
      // No need to check the instruction defining VNI for reads.
      ++MI;
    }
    assert(!SlotIndex::isSameInstr(VNI->def, TaintExtent.front().first) &&
           "Interference ends on VNI->def. Should have been handled earlier");
    MachineInstr *LastMI =
      Indexes->getInstructionFromIndex(TaintExtent.front().first);
    assert(LastMI && "Range must end at a proper instruction");
    unsigned TaintNum = 0;
    for(;;) {
      assert(MI != MBB->end() && "Bad LastMI");
      if (usesLanes(MI, Other.LI.reg, Other.SubIdx, TaintedLanes)) {
        DEBUG(dbgs() << "\t\ttainted lanes used by: " << *MI);
        return false;
      }
      // LastMI is the last instruction to use the current value.
      if (&*MI == LastMI) {
        if (++TaintNum == TaintExtent.size())
          break;
        LastMI = Indexes->getInstructionFromIndex(TaintExtent[TaintNum].first);
        assert(LastMI && "Range must end at a proper instruction");
        TaintedLanes = TaintExtent[TaintNum].second;
      }
      ++MI;
    }

    // The tainted lanes are unused.
    V.Resolution = CR_Replace;
    ++NumLaneResolves;
  }
  return true;
}

// Determine if ValNo is a copy of a value number in LI or Other.LI that will
// be pruned:
//
//   %dst = COPY %src
//   %src = COPY %dst  <-- This value to be pruned.
//   %dst = COPY %src  <-- This value is a copy of a pruned value.
//
bool JoinVals::isPrunedValue(unsigned ValNo, JoinVals &Other) {
  Val &V = Vals[ValNo];
  if (V.Pruned || V.PrunedComputed)
    return V.Pruned;

  if (V.Resolution != CR_Erase && V.Resolution != CR_Merge)
    return V.Pruned;

  // Follow copies up the dominator tree and check if any intermediate value
  // has been pruned.
  V.PrunedComputed = true;
  V.Pruned = Other.isPrunedValue(V.OtherVNI->id, *this);
  return V.Pruned;
}

void JoinVals::pruneValues(JoinVals &Other,
                           SmallVectorImpl<SlotIndex> &EndPoints) {
  for (unsigned i = 0, e = LI.getNumValNums(); i != e; ++i) {
    SlotIndex Def = LI.getValNumInfo(i)->def;
    switch (Vals[i].Resolution) {
    case CR_Keep:
      break;
    case CR_Replace: {
      // This value takes precedence over the value in Other.LI.
      LIS->pruneValue(&Other.LI, Def, &EndPoints);
      // Check if we're replacing an IMPLICIT_DEF value. The IMPLICIT_DEF
      // instructions are only inserted to provide a live-out value for PHI
      // predecessors, so the instruction should simply go away once its value
      // has been replaced.
      Val &OtherV = Other.Vals[Vals[i].OtherVNI->id];
      bool EraseImpDef = OtherV.ErasableImplicitDef &&
                         OtherV.Resolution == CR_Keep;
      if (!Def.isBlock()) {
        // Remove <def,read-undef> flags. This def is now a partial redef.
        // Also remove <def,dead> flags since the joined live range will
        // continue past this instruction.
        for (MIOperands MO(Indexes->getInstructionFromIndex(Def));
             MO.isValid(); ++MO)
          if (MO->isReg() && MO->isDef() && MO->getReg() == LI.reg) {
            MO->setIsUndef(EraseImpDef);
            MO->setIsDead(false);
          }
        // This value will reach instructions below, but we need to make sure
        // the live range also reaches the instruction at Def.
        if (!EraseImpDef)
          EndPoints.push_back(Def);
      }
      DEBUG(dbgs() << "\t\tpruned " << PrintReg(Other.LI.reg) << " at " << Def
                   << ": " << Other.LI << '\n');
      break;
    }
    case CR_Erase:
    case CR_Merge:
      if (isPrunedValue(i, Other)) {
        // This value is ultimately a copy of a pruned value in LI or Other.LI.
        // We can no longer trust the value mapping computed by
        // computeAssignment(), the value that was originally copied could have
        // been replaced.
        LIS->pruneValue(&LI, Def, &EndPoints);
        DEBUG(dbgs() << "\t\tpruned all of " << PrintReg(LI.reg) << " at "
                     << Def << ": " << LI << '\n');
      }
      break;
    case CR_Unresolved:
    case CR_Impossible:
      llvm_unreachable("Unresolved conflicts");
    }
  }
}

void JoinVals::eraseInstrs(SmallPtrSet<MachineInstr*, 8> &ErasedInstrs,
                           SmallVectorImpl<unsigned> &ShrinkRegs) {
  for (unsigned i = 0, e = LI.getNumValNums(); i != e; ++i) {
    // Get the def location before markUnused() below invalidates it.
    SlotIndex Def = LI.getValNumInfo(i)->def;
    switch (Vals[i].Resolution) {
    case CR_Keep:
      // If an IMPLICIT_DEF value is pruned, it doesn't serve a purpose any
      // longer. The IMPLICIT_DEF instructions are only inserted by
      // PHIElimination to guarantee that all PHI predecessors have a value.
      if (!Vals[i].ErasableImplicitDef || !Vals[i].Pruned)
        break;
      // Remove value number i from LI. Note that this VNInfo is still present
      // in NewVNInfo, so it will appear as an unused value number in the final
      // joined interval.
      LI.getValNumInfo(i)->markUnused();
      LI.removeValNo(LI.getValNumInfo(i));
      DEBUG(dbgs() << "\t\tremoved " << i << '@' << Def << ": " << LI << '\n');
      // FALL THROUGH.

    case CR_Erase: {
      MachineInstr *MI = Indexes->getInstructionFromIndex(Def);
      assert(MI && "No instruction to erase");
      if (MI->isCopy()) {
        unsigned Reg = MI->getOperand(1).getReg();
        if (TargetRegisterInfo::isVirtualRegister(Reg) &&
            Reg != CP.getSrcReg() && Reg != CP.getDstReg())
          ShrinkRegs.push_back(Reg);
      }
      ErasedInstrs.insert(MI);
      DEBUG(dbgs() << "\t\terased:\t" << Def << '\t' << *MI);
      LIS->RemoveMachineInstrFromMaps(MI);
      MI->eraseFromParent();
      break;
    }
    default:
      break;
    }
  }
}

bool RegisterCoalescer::joinVirtRegs(CoalescerPair &CP) {
  SmallVector<VNInfo*, 16> NewVNInfo;
  LiveInterval &RHS = LIS->getInterval(CP.getSrcReg());
  LiveInterval &LHS = LIS->getInterval(CP.getDstReg());
  JoinVals RHSVals(RHS, CP.getSrcIdx(), NewVNInfo, CP, LIS, TRI);
  JoinVals LHSVals(LHS, CP.getDstIdx(), NewVNInfo, CP, LIS, TRI);

  DEBUG(dbgs() << "\t\tRHS = " << RHS
               << "\n\t\tLHS = " << LHS
               << '\n');

  // First compute NewVNInfo and the simple value mappings.
  // Detect impossible conflicts early.
  if (!LHSVals.mapValues(RHSVals) || !RHSVals.mapValues(LHSVals))
    return false;

  // Some conflicts can only be resolved after all values have been mapped.
  if (!LHSVals.resolveConflicts(RHSVals) || !RHSVals.resolveConflicts(LHSVals))
    return false;

  // All clear, the live ranges can be merged.

  // The merging algorithm in LiveInterval::join() can't handle conflicting
  // value mappings, so we need to remove any live ranges that overlap a
  // CR_Replace resolution. Collect a set of end points that can be used to
  // restore the live range after joining.
  SmallVector<SlotIndex, 8> EndPoints;
  LHSVals.pruneValues(RHSVals, EndPoints);
  RHSVals.pruneValues(LHSVals, EndPoints);

  // Erase COPY and IMPLICIT_DEF instructions. This may cause some external
  // registers to require trimming.
  SmallVector<unsigned, 8> ShrinkRegs;
  LHSVals.eraseInstrs(ErasedInstrs, ShrinkRegs);
  RHSVals.eraseInstrs(ErasedInstrs, ShrinkRegs);
  while (!ShrinkRegs.empty())
    LIS->shrinkToUses(&LIS->getInterval(ShrinkRegs.pop_back_val()));

  // Join RHS into LHS.
  LHS.join(RHS, LHSVals.getAssignments(), RHSVals.getAssignments(), NewVNInfo);

  // Kill flags are going to be wrong if the live ranges were overlapping.
  // Eventually, we should simply clear all kill flags when computing live
  // ranges. They are reinserted after register allocation.
  MRI->clearKillFlags(LHS.reg);
  MRI->clearKillFlags(RHS.reg);

  if (EndPoints.empty())
    return true;

  // Recompute the parts of the live range we had to remove because of
  // CR_Replace conflicts.
  DEBUG(dbgs() << "\t\trestoring liveness to " << EndPoints.size()
               << " points: " << LHS << '\n');
  LIS->extendToIndices(LHS, EndPoints);
  return true;
}

/// joinIntervals - Attempt to join these two intervals.  On failure, this
/// returns false.
bool RegisterCoalescer::joinIntervals(CoalescerPair &CP) {
  return CP.isPhys() ? joinReservedPhysReg(CP) : joinVirtRegs(CP);
}

namespace {
// Information concerning MBB coalescing priority.
struct MBBPriorityInfo {
  MachineBasicBlock *MBB;
  unsigned Depth;
  bool IsSplit;

  MBBPriorityInfo(MachineBasicBlock *mbb, unsigned depth, bool issplit)
    : MBB(mbb), Depth(depth), IsSplit(issplit) {}
};
}

// C-style comparator that sorts first based on the loop depth of the basic
// block (the unsigned), and then on the MBB number.
//
// EnableGlobalCopies assumes that the primary sort key is loop depth.
static int compareMBBPriority(const MBBPriorityInfo *LHS,
                              const MBBPriorityInfo *RHS) {
  // Deeper loops first
  if (LHS->Depth != RHS->Depth)
    return LHS->Depth > RHS->Depth ? -1 : 1;

  // Try to unsplit critical edges next.
  if (LHS->IsSplit != RHS->IsSplit)
    return LHS->IsSplit ? -1 : 1;

  // Prefer blocks that are more connected in the CFG. This takes care of
  // the most difficult copies first while intervals are short.
  unsigned cl = LHS->MBB->pred_size() + LHS->MBB->succ_size();
  unsigned cr = RHS->MBB->pred_size() + RHS->MBB->succ_size();
  if (cl != cr)
    return cl > cr ? -1 : 1;

  // As a last resort, sort by block number.
  return LHS->MBB->getNumber() < RHS->MBB->getNumber() ? -1 : 1;
}

/// \returns true if the given copy uses or defines a local live range.
static bool isLocalCopy(MachineInstr *Copy, const LiveIntervals *LIS) {
  if (!Copy->isCopy())
    return false;

  if (Copy->getOperand(1).isUndef())
    return false;

  unsigned SrcReg = Copy->getOperand(1).getReg();
  unsigned DstReg = Copy->getOperand(0).getReg();
  if (TargetRegisterInfo::isPhysicalRegister(SrcReg)
      || TargetRegisterInfo::isPhysicalRegister(DstReg))
    return false;

  return LIS->intervalIsInOneMBB(LIS->getInterval(SrcReg))
    || LIS->intervalIsInOneMBB(LIS->getInterval(DstReg));
}

// Try joining WorkList copies starting from index From.
// Null out any successful joins.
bool RegisterCoalescer::
copyCoalesceWorkList(MutableArrayRef<MachineInstr*> CurrList) {
  bool Progress = false;
  for (unsigned i = 0, e = CurrList.size(); i != e; ++i) {
    if (!CurrList[i])
      continue;
    // Skip instruction pointers that have already been erased, for example by
    // dead code elimination.
    if (ErasedInstrs.erase(CurrList[i])) {
      CurrList[i] = 0;
      continue;
    }
    bool Again = false;
    bool Success = joinCopy(CurrList[i], Again);
    Progress |= Success;
    if (Success || !Again)
      CurrList[i] = 0;
  }
  return Progress;
}

void
RegisterCoalescer::copyCoalesceInMBB(MachineBasicBlock *MBB) {
  DEBUG(dbgs() << MBB->getName() << ":\n");

  // Collect all copy-like instructions in MBB. Don't start coalescing anything
  // yet, it might invalidate the iterator.
  const unsigned PrevSize = WorkList.size();
  if (JoinGlobalCopies) {
    // Coalesce copies bottom-up to coalesce local defs before local uses. They
    // are not inherently easier to resolve, but slightly preferable until we
    // have local live range splitting. In particular this is required by
    // cmp+jmp macro fusion.
    for (MachineBasicBlock::iterator MII = MBB->begin(), E = MBB->end();
         MII != E; ++MII) {
      if (!MII->isCopyLike())
        continue;
      if (isLocalCopy(&(*MII), LIS))
        LocalWorkList.push_back(&(*MII));
      else
        WorkList.push_back(&(*MII));
    }
  }
  else {
     for (MachineBasicBlock::iterator MII = MBB->begin(), E = MBB->end();
          MII != E; ++MII)
       if (MII->isCopyLike())
         WorkList.push_back(MII);
  }
  // Try coalescing the collected copies immediately, and remove the nulls.
  // This prevents the WorkList from getting too large since most copies are
  // joinable on the first attempt.
  MutableArrayRef<MachineInstr*>
    CurrList(WorkList.begin() + PrevSize, WorkList.end());
  if (copyCoalesceWorkList(CurrList))
    WorkList.erase(std::remove(WorkList.begin() + PrevSize, WorkList.end(),
                               (MachineInstr*)0), WorkList.end());
}

void RegisterCoalescer::coalesceLocals() {
  copyCoalesceWorkList(LocalWorkList);
  for (unsigned j = 0, je = LocalWorkList.size(); j != je; ++j) {
    if (LocalWorkList[j])
      WorkList.push_back(LocalWorkList[j]);
  }
  LocalWorkList.clear();
}

void RegisterCoalescer::joinAllIntervals() {
  DEBUG(dbgs() << "********** JOINING INTERVALS ***********\n");
  assert(WorkList.empty() && LocalWorkList.empty() && "Old data still around.");

  std::vector<MBBPriorityInfo> MBBs;
  MBBs.reserve(MF->size());
  for (MachineFunction::iterator I = MF->begin(), E = MF->end();I != E;++I){
    MachineBasicBlock *MBB = I;
    MBBs.push_back(MBBPriorityInfo(MBB, Loops->getLoopDepth(MBB),
                                   JoinSplitEdges && isSplitEdge(MBB)));
  }
  array_pod_sort(MBBs.begin(), MBBs.end(), compareMBBPriority);

  // Coalesce intervals in MBB priority order.
  unsigned CurrDepth = UINT_MAX;
  for (unsigned i = 0, e = MBBs.size(); i != e; ++i) {
    // Try coalescing the collected local copies for deeper loops.
    if (JoinGlobalCopies && MBBs[i].Depth < CurrDepth) {
      coalesceLocals();
      CurrDepth = MBBs[i].Depth;
    }
    copyCoalesceInMBB(MBBs[i].MBB);
  }
  coalesceLocals();

  // Joining intervals can allow other intervals to be joined.  Iteratively join
  // until we make no progress.
  while (copyCoalesceWorkList(WorkList))
    /* empty */ ;
}

void RegisterCoalescer::releaseMemory() {
  ErasedInstrs.clear();
  WorkList.clear();
  DeadDefs.clear();
  InflateRegs.clear();
}

bool RegisterCoalescer::runOnMachineFunction(MachineFunction &fn) {
  MF = &fn;
  MRI = &fn.getRegInfo();
  TM = &fn.getTarget();
  TRI = TM->getRegisterInfo();
  TII = TM->getInstrInfo();
  LIS = &getAnalysis<LiveIntervals>();
  AA = &getAnalysis<AliasAnalysis>();
  Loops = &getAnalysis<MachineLoopInfo>();

  const TargetSubtargetInfo &ST = TM->getSubtarget<TargetSubtargetInfo>();
  if (EnableGlobalCopies == cl::BOU_UNSET)
    JoinGlobalCopies = ST.useMachineScheduler();
  else
    JoinGlobalCopies = (EnableGlobalCopies == cl::BOU_TRUE);

  // The MachineScheduler does not currently require JoinSplitEdges. This will
  // either be enabled unconditionally or replaced by a more general live range
  // splitting optimization.
  JoinSplitEdges = EnableJoinSplits;

  DEBUG(dbgs() << "********** SIMPLE REGISTER COALESCING **********\n"
               << "********** Function: " << MF->getName() << '\n');

  if (VerifyCoalescing)
    MF->verify(this, "Before register coalescing");

  RegClassInfo.runOnMachineFunction(fn);

  // Join (coalesce) intervals if requested.
  if (EnableJoining)
    joinAllIntervals();

  // After deleting a lot of copies, register classes may be less constrained.
  // Removing sub-register operands may allow GR32_ABCD -> GR32 and DPR_VFP2 ->
  // DPR inflation.
  array_pod_sort(InflateRegs.begin(), InflateRegs.end());
  InflateRegs.erase(std::unique(InflateRegs.begin(), InflateRegs.end()),
                    InflateRegs.end());
  DEBUG(dbgs() << "Trying to inflate " << InflateRegs.size() << " regs.\n");
  for (unsigned i = 0, e = InflateRegs.size(); i != e; ++i) {
    unsigned Reg = InflateRegs[i];
    if (MRI->reg_nodbg_empty(Reg))
      continue;
    if (MRI->recomputeRegClass(Reg, *TM)) {
      DEBUG(dbgs() << PrintReg(Reg) << " inflated to "
                   << MRI->getRegClass(Reg)->getName() << '\n');
      ++NumInflated;
    }
  }

  DEBUG(dump());
  if (VerifyCoalescing)
    MF->verify(this, "After register coalescing");
  return true;
}

/// print - Implement the dump method.
void RegisterCoalescer::print(raw_ostream &O, const Module* m) const {
   LIS->print(O, m);
}