CombinerHelper.cpp 62.8 KB
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
//===-- lib/CodeGen/GlobalISel/GICombinerHelper.cpp -----------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/GlobalISel/CombinerHelper.h"
#include "llvm/CodeGen/GlobalISel/Combiner.h"
#include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
#include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
#include "llvm/CodeGen/GlobalISel/LegalizerInfo.h"
#include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
#include "llvm/CodeGen/GlobalISel/Utils.h"
#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetMachine.h"

#define DEBUG_TYPE "gi-combiner"

using namespace llvm;
using namespace MIPatternMatch;

// Option to allow testing of the combiner while no targets know about indexed
// addressing.
static cl::opt<bool>
    ForceLegalIndexing("force-legal-indexing", cl::Hidden, cl::init(false),
                       cl::desc("Force all indexed operations to be "
                                "legal for the GlobalISel combiner"));


CombinerHelper::CombinerHelper(GISelChangeObserver &Observer,
                               MachineIRBuilder &B, GISelKnownBits *KB,
                               MachineDominatorTree *MDT,
                               const LegalizerInfo *LI)
    : Builder(B), MRI(Builder.getMF().getRegInfo()), Observer(Observer),
      KB(KB), MDT(MDT), LI(LI) {
  (void)this->KB;
}

void CombinerHelper::replaceRegWith(MachineRegisterInfo &MRI, Register FromReg,
                                    Register ToReg) const {
  Observer.changingAllUsesOfReg(MRI, FromReg);

  if (MRI.constrainRegAttrs(ToReg, FromReg))
    MRI.replaceRegWith(FromReg, ToReg);
  else
    Builder.buildCopy(ToReg, FromReg);

  Observer.finishedChangingAllUsesOfReg();
}

void CombinerHelper::replaceRegOpWith(MachineRegisterInfo &MRI,
                                      MachineOperand &FromRegOp,
                                      Register ToReg) const {
  assert(FromRegOp.getParent() && "Expected an operand in an MI");
  Observer.changingInstr(*FromRegOp.getParent());

  FromRegOp.setReg(ToReg);

  Observer.changedInstr(*FromRegOp.getParent());
}

bool CombinerHelper::tryCombineCopy(MachineInstr &MI) {
  if (matchCombineCopy(MI)) {
    applyCombineCopy(MI);
    return true;
  }
  return false;
}
bool CombinerHelper::matchCombineCopy(MachineInstr &MI) {
  if (MI.getOpcode() != TargetOpcode::COPY)
    return false;
  Register DstReg = MI.getOperand(0).getReg();
  Register SrcReg = MI.getOperand(1).getReg();
  return canReplaceReg(DstReg, SrcReg, MRI);
}
void CombinerHelper::applyCombineCopy(MachineInstr &MI) {
  Register DstReg = MI.getOperand(0).getReg();
  Register SrcReg = MI.getOperand(1).getReg();
  MI.eraseFromParent();
  replaceRegWith(MRI, DstReg, SrcReg);
}

bool CombinerHelper::tryCombineConcatVectors(MachineInstr &MI) {
  bool IsUndef = false;
  SmallVector<Register, 4> Ops;
  if (matchCombineConcatVectors(MI, IsUndef, Ops)) {
    applyCombineConcatVectors(MI, IsUndef, Ops);
    return true;
  }
  return false;
}

bool CombinerHelper::matchCombineConcatVectors(MachineInstr &MI, bool &IsUndef,
                                               SmallVectorImpl<Register> &Ops) {
  assert(MI.getOpcode() == TargetOpcode::G_CONCAT_VECTORS &&
         "Invalid instruction");
  IsUndef = true;
  MachineInstr *Undef = nullptr;

  // Walk over all the operands of concat vectors and check if they are
  // build_vector themselves or undef.
  // Then collect their operands in Ops.
  for (const MachineOperand &MO : MI.uses()) {
    Register Reg = MO.getReg();
    MachineInstr *Def = MRI.getVRegDef(Reg);
    assert(Def && "Operand not defined");
    switch (Def->getOpcode()) {
    case TargetOpcode::G_BUILD_VECTOR:
      IsUndef = false;
      // Remember the operands of the build_vector to fold
      // them into the yet-to-build flattened concat vectors.
      for (const MachineOperand &BuildVecMO : Def->uses())
        Ops.push_back(BuildVecMO.getReg());
      break;
    case TargetOpcode::G_IMPLICIT_DEF: {
      LLT OpType = MRI.getType(Reg);
      // Keep one undef value for all the undef operands.
      if (!Undef) {
        Builder.setInsertPt(*MI.getParent(), MI);
        Undef = Builder.buildUndef(OpType.getScalarType());
      }
      assert(MRI.getType(Undef->getOperand(0).getReg()) ==
                 OpType.getScalarType() &&
             "All undefs should have the same type");
      // Break the undef vector in as many scalar elements as needed
      // for the flattening.
      for (unsigned EltIdx = 0, EltEnd = OpType.getNumElements();
           EltIdx != EltEnd; ++EltIdx)
        Ops.push_back(Undef->getOperand(0).getReg());
      break;
    }
    default:
      return false;
    }
  }
  return true;
}
void CombinerHelper::applyCombineConcatVectors(
    MachineInstr &MI, bool IsUndef, const ArrayRef<Register> Ops) {
  // We determined that the concat_vectors can be flatten.
  // Generate the flattened build_vector.
  Register DstReg = MI.getOperand(0).getReg();
  Builder.setInsertPt(*MI.getParent(), MI);
  Register NewDstReg = MRI.cloneVirtualRegister(DstReg);

  // Note: IsUndef is sort of redundant. We could have determine it by
  // checking that at all Ops are undef.  Alternatively, we could have
  // generate a build_vector of undefs and rely on another combine to
  // clean that up.  For now, given we already gather this information
  // in tryCombineConcatVectors, just save compile time and issue the
  // right thing.
  if (IsUndef)
    Builder.buildUndef(NewDstReg);
  else
    Builder.buildBuildVector(NewDstReg, Ops);
  MI.eraseFromParent();
  replaceRegWith(MRI, DstReg, NewDstReg);
}

bool CombinerHelper::tryCombineShuffleVector(MachineInstr &MI) {
  SmallVector<Register, 4> Ops;
  if (matchCombineShuffleVector(MI, Ops)) {
    applyCombineShuffleVector(MI, Ops);
    return true;
  }
  return false;
}

bool CombinerHelper::matchCombineShuffleVector(MachineInstr &MI,
                                               SmallVectorImpl<Register> &Ops) {
  assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR &&
         "Invalid instruction kind");
  LLT DstType = MRI.getType(MI.getOperand(0).getReg());
  Register Src1 = MI.getOperand(1).getReg();
  LLT SrcType = MRI.getType(Src1);
  // As bizarre as it may look, shuffle vector can actually produce
  // scalar! This is because at the IR level a <1 x ty> shuffle
  // vector is perfectly valid.
  unsigned DstNumElts = DstType.isVector() ? DstType.getNumElements() : 1;
  unsigned SrcNumElts = SrcType.isVector() ? SrcType.getNumElements() : 1;

  // If the resulting vector is smaller than the size of the source
  // vectors being concatenated, we won't be able to replace the
  // shuffle vector into a concat_vectors.
  //
  // Note: We may still be able to produce a concat_vectors fed by
  //       extract_vector_elt and so on. It is less clear that would
  //       be better though, so don't bother for now.
  //
  // If the destination is a scalar, the size of the sources doesn't
  // matter. we will lower the shuffle to a plain copy. This will
  // work only if the source and destination have the same size. But
  // that's covered by the next condition.
  //
  // TODO: If the size between the source and destination don't match
  //       we could still emit an extract vector element in that case.
  if (DstNumElts < 2 * SrcNumElts && DstNumElts != 1)
    return false;

  // Check that the shuffle mask can be broken evenly between the
  // different sources.
  if (DstNumElts % SrcNumElts != 0)
    return false;

  // Mask length is a multiple of the source vector length.
  // Check if the shuffle is some kind of concatenation of the input
  // vectors.
  unsigned NumConcat = DstNumElts / SrcNumElts;
  SmallVector<int, 8> ConcatSrcs(NumConcat, -1);
  ArrayRef<int> Mask = MI.getOperand(3).getShuffleMask();
  for (unsigned i = 0; i != DstNumElts; ++i) {
    int Idx = Mask[i];
    // Undef value.
    if (Idx < 0)
      continue;
    // Ensure the indices in each SrcType sized piece are sequential and that
    // the same source is used for the whole piece.
    if ((Idx % SrcNumElts != (i % SrcNumElts)) ||
        (ConcatSrcs[i / SrcNumElts] >= 0 &&
         ConcatSrcs[i / SrcNumElts] != (int)(Idx / SrcNumElts)))
      return false;
    // Remember which source this index came from.
    ConcatSrcs[i / SrcNumElts] = Idx / SrcNumElts;
  }

  // The shuffle is concatenating multiple vectors together.
  // Collect the different operands for that.
  Register UndefReg;
  Register Src2 = MI.getOperand(2).getReg();
  for (auto Src : ConcatSrcs) {
    if (Src < 0) {
      if (!UndefReg) {
        Builder.setInsertPt(*MI.getParent(), MI);
        UndefReg = Builder.buildUndef(SrcType).getReg(0);
      }
      Ops.push_back(UndefReg);
    } else if (Src == 0)
      Ops.push_back(Src1);
    else
      Ops.push_back(Src2);
  }
  return true;
}

void CombinerHelper::applyCombineShuffleVector(MachineInstr &MI,
                                               const ArrayRef<Register> Ops) {
  Register DstReg = MI.getOperand(0).getReg();
  Builder.setInsertPt(*MI.getParent(), MI);
  Register NewDstReg = MRI.cloneVirtualRegister(DstReg);

  if (Ops.size() == 1)
    Builder.buildCopy(NewDstReg, Ops[0]);
  else
    Builder.buildMerge(NewDstReg, Ops);

  MI.eraseFromParent();
  replaceRegWith(MRI, DstReg, NewDstReg);
}

namespace {

/// Select a preference between two uses. CurrentUse is the current preference
/// while *ForCandidate is attributes of the candidate under consideration.
PreferredTuple ChoosePreferredUse(PreferredTuple &CurrentUse,
                                  const LLT TyForCandidate,
                                  unsigned OpcodeForCandidate,
                                  MachineInstr *MIForCandidate) {
  if (!CurrentUse.Ty.isValid()) {
    if (CurrentUse.ExtendOpcode == OpcodeForCandidate ||
        CurrentUse.ExtendOpcode == TargetOpcode::G_ANYEXT)
      return {TyForCandidate, OpcodeForCandidate, MIForCandidate};
    return CurrentUse;
  }

  // We permit the extend to hoist through basic blocks but this is only
  // sensible if the target has extending loads. If you end up lowering back
  // into a load and extend during the legalizer then the end result is
  // hoisting the extend up to the load.

  // Prefer defined extensions to undefined extensions as these are more
  // likely to reduce the number of instructions.
  if (OpcodeForCandidate == TargetOpcode::G_ANYEXT &&
      CurrentUse.ExtendOpcode != TargetOpcode::G_ANYEXT)
    return CurrentUse;
  else if (CurrentUse.ExtendOpcode == TargetOpcode::G_ANYEXT &&
           OpcodeForCandidate != TargetOpcode::G_ANYEXT)
    return {TyForCandidate, OpcodeForCandidate, MIForCandidate};

  // Prefer sign extensions to zero extensions as sign-extensions tend to be
  // more expensive.
  if (CurrentUse.Ty == TyForCandidate) {
    if (CurrentUse.ExtendOpcode == TargetOpcode::G_SEXT &&
        OpcodeForCandidate == TargetOpcode::G_ZEXT)
      return CurrentUse;
    else if (CurrentUse.ExtendOpcode == TargetOpcode::G_ZEXT &&
             OpcodeForCandidate == TargetOpcode::G_SEXT)
      return {TyForCandidate, OpcodeForCandidate, MIForCandidate};
  }

  // This is potentially target specific. We've chosen the largest type
  // because G_TRUNC is usually free. One potential catch with this is that
  // some targets have a reduced number of larger registers than smaller
  // registers and this choice potentially increases the live-range for the
  // larger value.
  if (TyForCandidate.getSizeInBits() > CurrentUse.Ty.getSizeInBits()) {
    return {TyForCandidate, OpcodeForCandidate, MIForCandidate};
  }
  return CurrentUse;
}

/// Find a suitable place to insert some instructions and insert them. This
/// function accounts for special cases like inserting before a PHI node.
/// The current strategy for inserting before PHI's is to duplicate the
/// instructions for each predecessor. However, while that's ok for G_TRUNC
/// on most targets since it generally requires no code, other targets/cases may
/// want to try harder to find a dominating block.
static void InsertInsnsWithoutSideEffectsBeforeUse(
    MachineIRBuilder &Builder, MachineInstr &DefMI, MachineOperand &UseMO,
    std::function<void(MachineBasicBlock *, MachineBasicBlock::iterator,
                       MachineOperand &UseMO)>
        Inserter) {
  MachineInstr &UseMI = *UseMO.getParent();

  MachineBasicBlock *InsertBB = UseMI.getParent();

  // If the use is a PHI then we want the predecessor block instead.
  if (UseMI.isPHI()) {
    MachineOperand *PredBB = std::next(&UseMO);
    InsertBB = PredBB->getMBB();
  }

  // If the block is the same block as the def then we want to insert just after
  // the def instead of at the start of the block.
  if (InsertBB == DefMI.getParent()) {
    MachineBasicBlock::iterator InsertPt = &DefMI;
    Inserter(InsertBB, std::next(InsertPt), UseMO);
    return;
  }

  // Otherwise we want the start of the BB
  Inserter(InsertBB, InsertBB->getFirstNonPHI(), UseMO);
}
} // end anonymous namespace

bool CombinerHelper::tryCombineExtendingLoads(MachineInstr &MI) {
  PreferredTuple Preferred;
  if (matchCombineExtendingLoads(MI, Preferred)) {
    applyCombineExtendingLoads(MI, Preferred);
    return true;
  }
  return false;
}

bool CombinerHelper::matchCombineExtendingLoads(MachineInstr &MI,
                                                PreferredTuple &Preferred) {
  // We match the loads and follow the uses to the extend instead of matching
  // the extends and following the def to the load. This is because the load
  // must remain in the same position for correctness (unless we also add code
  // to find a safe place to sink it) whereas the extend is freely movable.
  // It also prevents us from duplicating the load for the volatile case or just
  // for performance.

  if (MI.getOpcode() != TargetOpcode::G_LOAD &&
      MI.getOpcode() != TargetOpcode::G_SEXTLOAD &&
      MI.getOpcode() != TargetOpcode::G_ZEXTLOAD)
    return false;

  auto &LoadValue = MI.getOperand(0);
  assert(LoadValue.isReg() && "Result wasn't a register?");

  LLT LoadValueTy = MRI.getType(LoadValue.getReg());
  if (!LoadValueTy.isScalar())
    return false;

  // Most architectures are going to legalize <s8 loads into at least a 1 byte
  // load, and the MMOs can only describe memory accesses in multiples of bytes.
  // If we try to perform extload combining on those, we can end up with
  // %a(s8) = extload %ptr (load 1 byte from %ptr)
  // ... which is an illegal extload instruction.
  if (LoadValueTy.getSizeInBits() < 8)
    return false;

  // For non power-of-2 types, they will very likely be legalized into multiple
  // loads. Don't bother trying to match them into extending loads.
  if (!isPowerOf2_32(LoadValueTy.getSizeInBits()))
    return false;

  // Find the preferred type aside from the any-extends (unless it's the only
  // one) and non-extending ops. We'll emit an extending load to that type and
  // and emit a variant of (extend (trunc X)) for the others according to the
  // relative type sizes. At the same time, pick an extend to use based on the
  // extend involved in the chosen type.
  unsigned PreferredOpcode = MI.getOpcode() == TargetOpcode::G_LOAD
                                 ? TargetOpcode::G_ANYEXT
                                 : MI.getOpcode() == TargetOpcode::G_SEXTLOAD
                                       ? TargetOpcode::G_SEXT
                                       : TargetOpcode::G_ZEXT;
  Preferred = {LLT(), PreferredOpcode, nullptr};
  for (auto &UseMI : MRI.use_nodbg_instructions(LoadValue.getReg())) {
    if (UseMI.getOpcode() == TargetOpcode::G_SEXT ||
        UseMI.getOpcode() == TargetOpcode::G_ZEXT ||
        (UseMI.getOpcode() == TargetOpcode::G_ANYEXT)) {
      // Check for legality.
      if (LI) {
        LegalityQuery::MemDesc MMDesc;
        const auto &MMO = **MI.memoperands_begin();
        MMDesc.SizeInBits = MMO.getSizeInBits();
        MMDesc.AlignInBits = MMO.getAlign().value() * 8;
        MMDesc.Ordering = MMO.getOrdering();
        LLT UseTy = MRI.getType(UseMI.getOperand(0).getReg());
        LLT SrcTy = MRI.getType(MI.getOperand(1).getReg());
        if (LI->getAction({MI.getOpcode(), {UseTy, SrcTy}, {MMDesc}}).Action !=
            LegalizeActions::Legal)
          continue;
      }
      Preferred = ChoosePreferredUse(Preferred,
                                     MRI.getType(UseMI.getOperand(0).getReg()),
                                     UseMI.getOpcode(), &UseMI);
    }
  }

  // There were no extends
  if (!Preferred.MI)
    return false;
  // It should be impossible to chose an extend without selecting a different
  // type since by definition the result of an extend is larger.
  assert(Preferred.Ty != LoadValueTy && "Extending to same type?");

  LLVM_DEBUG(dbgs() << "Preferred use is: " << *Preferred.MI);
  return true;
}

void CombinerHelper::applyCombineExtendingLoads(MachineInstr &MI,
                                                PreferredTuple &Preferred) {
  // Rewrite the load to the chosen extending load.
  Register ChosenDstReg = Preferred.MI->getOperand(0).getReg();

  // Inserter to insert a truncate back to the original type at a given point
  // with some basic CSE to limit truncate duplication to one per BB.
  DenseMap<MachineBasicBlock *, MachineInstr *> EmittedInsns;
  auto InsertTruncAt = [&](MachineBasicBlock *InsertIntoBB,
                           MachineBasicBlock::iterator InsertBefore,
                           MachineOperand &UseMO) {
    MachineInstr *PreviouslyEmitted = EmittedInsns.lookup(InsertIntoBB);
    if (PreviouslyEmitted) {
      Observer.changingInstr(*UseMO.getParent());
      UseMO.setReg(PreviouslyEmitted->getOperand(0).getReg());
      Observer.changedInstr(*UseMO.getParent());
      return;
    }

    Builder.setInsertPt(*InsertIntoBB, InsertBefore);
    Register NewDstReg = MRI.cloneVirtualRegister(MI.getOperand(0).getReg());
    MachineInstr *NewMI = Builder.buildTrunc(NewDstReg, ChosenDstReg);
    EmittedInsns[InsertIntoBB] = NewMI;
    replaceRegOpWith(MRI, UseMO, NewDstReg);
  };

  Observer.changingInstr(MI);
  MI.setDesc(
      Builder.getTII().get(Preferred.ExtendOpcode == TargetOpcode::G_SEXT
                               ? TargetOpcode::G_SEXTLOAD
                               : Preferred.ExtendOpcode == TargetOpcode::G_ZEXT
                                     ? TargetOpcode::G_ZEXTLOAD
                                     : TargetOpcode::G_LOAD));

  // Rewrite all the uses to fix up the types.
  auto &LoadValue = MI.getOperand(0);
  SmallVector<MachineOperand *, 4> Uses;
  for (auto &UseMO : MRI.use_operands(LoadValue.getReg()))
    Uses.push_back(&UseMO);

  for (auto *UseMO : Uses) {
    MachineInstr *UseMI = UseMO->getParent();

    // If the extend is compatible with the preferred extend then we should fix
    // up the type and extend so that it uses the preferred use.
    if (UseMI->getOpcode() == Preferred.ExtendOpcode ||
        UseMI->getOpcode() == TargetOpcode::G_ANYEXT) {
      Register UseDstReg = UseMI->getOperand(0).getReg();
      MachineOperand &UseSrcMO = UseMI->getOperand(1);
      const LLT UseDstTy = MRI.getType(UseDstReg);
      if (UseDstReg != ChosenDstReg) {
        if (Preferred.Ty == UseDstTy) {
          // If the use has the same type as the preferred use, then merge
          // the vregs and erase the extend. For example:
          //    %1:_(s8) = G_LOAD ...
          //    %2:_(s32) = G_SEXT %1(s8)
          //    %3:_(s32) = G_ANYEXT %1(s8)
          //    ... = ... %3(s32)
          // rewrites to:
          //    %2:_(s32) = G_SEXTLOAD ...
          //    ... = ... %2(s32)
          replaceRegWith(MRI, UseDstReg, ChosenDstReg);
          Observer.erasingInstr(*UseMO->getParent());
          UseMO->getParent()->eraseFromParent();
        } else if (Preferred.Ty.getSizeInBits() < UseDstTy.getSizeInBits()) {
          // If the preferred size is smaller, then keep the extend but extend
          // from the result of the extending load. For example:
          //    %1:_(s8) = G_LOAD ...
          //    %2:_(s32) = G_SEXT %1(s8)
          //    %3:_(s64) = G_ANYEXT %1(s8)
          //    ... = ... %3(s64)
          /// rewrites to:
          //    %2:_(s32) = G_SEXTLOAD ...
          //    %3:_(s64) = G_ANYEXT %2:_(s32)
          //    ... = ... %3(s64)
          replaceRegOpWith(MRI, UseSrcMO, ChosenDstReg);
        } else {
          // If the preferred size is large, then insert a truncate. For
          // example:
          //    %1:_(s8) = G_LOAD ...
          //    %2:_(s64) = G_SEXT %1(s8)
          //    %3:_(s32) = G_ZEXT %1(s8)
          //    ... = ... %3(s32)
          /// rewrites to:
          //    %2:_(s64) = G_SEXTLOAD ...
          //    %4:_(s8) = G_TRUNC %2:_(s32)
          //    %3:_(s64) = G_ZEXT %2:_(s8)
          //    ... = ... %3(s64)
          InsertInsnsWithoutSideEffectsBeforeUse(Builder, MI, *UseMO,
                                                 InsertTruncAt);
        }
        continue;
      }
      // The use is (one of) the uses of the preferred use we chose earlier.
      // We're going to update the load to def this value later so just erase
      // the old extend.
      Observer.erasingInstr(*UseMO->getParent());
      UseMO->getParent()->eraseFromParent();
      continue;
    }

    // The use isn't an extend. Truncate back to the type we originally loaded.
    // This is free on many targets.
    InsertInsnsWithoutSideEffectsBeforeUse(Builder, MI, *UseMO, InsertTruncAt);
  }

  MI.getOperand(0).setReg(ChosenDstReg);
  Observer.changedInstr(MI);
}

bool CombinerHelper::isPredecessor(const MachineInstr &DefMI,
                                   const MachineInstr &UseMI) {
  assert(!DefMI.isDebugInstr() && !UseMI.isDebugInstr() &&
         "shouldn't consider debug uses");
  assert(DefMI.getParent() == UseMI.getParent());
  if (&DefMI == &UseMI)
    return false;

  // Loop through the basic block until we find one of the instructions.
  MachineBasicBlock::const_iterator I = DefMI.getParent()->begin();
  for (; &*I != &DefMI && &*I != &UseMI; ++I)
    return &*I == &DefMI;

  llvm_unreachable("Block must contain instructions");
}

bool CombinerHelper::dominates(const MachineInstr &DefMI,
                               const MachineInstr &UseMI) {
  assert(!DefMI.isDebugInstr() && !UseMI.isDebugInstr() &&
         "shouldn't consider debug uses");
  if (MDT)
    return MDT->dominates(&DefMI, &UseMI);
  else if (DefMI.getParent() != UseMI.getParent())
    return false;

  return isPredecessor(DefMI, UseMI);
}

bool CombinerHelper::matchSextAlreadyExtended(MachineInstr &MI) {
  assert(MI.getOpcode() == TargetOpcode::G_SEXT_INREG);
  Register SrcReg = MI.getOperand(1).getReg();
  unsigned SrcSignBits = KB->computeNumSignBits(SrcReg);
  unsigned NumSextBits =
      MRI.getType(MI.getOperand(0).getReg()).getScalarSizeInBits() -
      MI.getOperand(2).getImm();
  return SrcSignBits >= NumSextBits;
}

bool CombinerHelper::applySextAlreadyExtended(MachineInstr &MI) {
  assert(MI.getOpcode() == TargetOpcode::G_SEXT_INREG);
  MachineIRBuilder MIB(MI);
  MIB.buildCopy(MI.getOperand(0).getReg(), MI.getOperand(1).getReg());
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::findPostIndexCandidate(MachineInstr &MI, Register &Addr,
                                            Register &Base, Register &Offset) {
  auto &MF = *MI.getParent()->getParent();
  const auto &TLI = *MF.getSubtarget().getTargetLowering();

#ifndef NDEBUG
  unsigned Opcode = MI.getOpcode();
  assert(Opcode == TargetOpcode::G_LOAD || Opcode == TargetOpcode::G_SEXTLOAD ||
         Opcode == TargetOpcode::G_ZEXTLOAD || Opcode == TargetOpcode::G_STORE);
#endif

  Base = MI.getOperand(1).getReg();
  MachineInstr *BaseDef = MRI.getUniqueVRegDef(Base);
  if (BaseDef && BaseDef->getOpcode() == TargetOpcode::G_FRAME_INDEX)
    return false;

  LLVM_DEBUG(dbgs() << "Searching for post-indexing opportunity for: " << MI);

  for (auto &Use : MRI.use_nodbg_instructions(Base)) {
    if (Use.getOpcode() != TargetOpcode::G_PTR_ADD)
      continue;

    Offset = Use.getOperand(2).getReg();
    if (!ForceLegalIndexing &&
        !TLI.isIndexingLegal(MI, Base, Offset, /*IsPre*/ false, MRI)) {
      LLVM_DEBUG(dbgs() << "    Ignoring candidate with illegal addrmode: "
                        << Use);
      continue;
    }

    // Make sure the offset calculation is before the potentially indexed op.
    // FIXME: we really care about dependency here. The offset calculation might
    // be movable.
    MachineInstr *OffsetDef = MRI.getUniqueVRegDef(Offset);
    if (!OffsetDef || !dominates(*OffsetDef, MI)) {
      LLVM_DEBUG(dbgs() << "    Ignoring candidate with offset after mem-op: "
                        << Use);
      continue;
    }

    // FIXME: check whether all uses of Base are load/store with foldable
    // addressing modes. If so, using the normal addr-modes is better than
    // forming an indexed one.

    bool MemOpDominatesAddrUses = true;
    for (auto &PtrAddUse :
         MRI.use_nodbg_instructions(Use.getOperand(0).getReg())) {
      if (!dominates(MI, PtrAddUse)) {
        MemOpDominatesAddrUses = false;
        break;
      }
    }

    if (!MemOpDominatesAddrUses) {
      LLVM_DEBUG(
          dbgs() << "    Ignoring candidate as memop does not dominate uses: "
                 << Use);
      continue;
    }

    LLVM_DEBUG(dbgs() << "    Found match: " << Use);
    Addr = Use.getOperand(0).getReg();
    return true;
  }

  return false;
}

bool CombinerHelper::findPreIndexCandidate(MachineInstr &MI, Register &Addr,
                                           Register &Base, Register &Offset) {
  auto &MF = *MI.getParent()->getParent();
  const auto &TLI = *MF.getSubtarget().getTargetLowering();

#ifndef NDEBUG
  unsigned Opcode = MI.getOpcode();
  assert(Opcode == TargetOpcode::G_LOAD || Opcode == TargetOpcode::G_SEXTLOAD ||
         Opcode == TargetOpcode::G_ZEXTLOAD || Opcode == TargetOpcode::G_STORE);
#endif

  Addr = MI.getOperand(1).getReg();
  MachineInstr *AddrDef = getOpcodeDef(TargetOpcode::G_PTR_ADD, Addr, MRI);
  if (!AddrDef || MRI.hasOneNonDBGUse(Addr))
    return false;

  Base = AddrDef->getOperand(1).getReg();
  Offset = AddrDef->getOperand(2).getReg();

  LLVM_DEBUG(dbgs() << "Found potential pre-indexed load_store: " << MI);

  if (!ForceLegalIndexing &&
      !TLI.isIndexingLegal(MI, Base, Offset, /*IsPre*/ true, MRI)) {
    LLVM_DEBUG(dbgs() << "    Skipping, not legal for target");
    return false;
  }

  MachineInstr *BaseDef = getDefIgnoringCopies(Base, MRI);
  if (BaseDef->getOpcode() == TargetOpcode::G_FRAME_INDEX) {
    LLVM_DEBUG(dbgs() << "    Skipping, frame index would need copy anyway.");
    return false;
  }

  if (MI.getOpcode() == TargetOpcode::G_STORE) {
    // Would require a copy.
    if (Base == MI.getOperand(0).getReg()) {
      LLVM_DEBUG(dbgs() << "    Skipping, storing base so need copy anyway.");
      return false;
    }

    // We're expecting one use of Addr in MI, but it could also be the
    // value stored, which isn't actually dominated by the instruction.
    if (MI.getOperand(0).getReg() == Addr) {
      LLVM_DEBUG(dbgs() << "    Skipping, does not dominate all addr uses");
      return false;
    }
  }

  // FIXME: check whether all uses of the base pointer are constant PtrAdds.
  // That might allow us to end base's liveness here by adjusting the constant.

  for (auto &UseMI : MRI.use_nodbg_instructions(Addr)) {
    if (!dominates(MI, UseMI)) {
      LLVM_DEBUG(dbgs() << "    Skipping, does not dominate all addr uses.");
      return false;
    }
  }

  return true;
}

bool CombinerHelper::tryCombineIndexedLoadStore(MachineInstr &MI) {
  IndexedLoadStoreMatchInfo MatchInfo;
  if (matchCombineIndexedLoadStore(MI, MatchInfo)) {
    applyCombineIndexedLoadStore(MI, MatchInfo);
    return true;
  }
  return false;
}

bool CombinerHelper::matchCombineIndexedLoadStore(MachineInstr &MI, IndexedLoadStoreMatchInfo &MatchInfo) {
  unsigned Opcode = MI.getOpcode();
  if (Opcode != TargetOpcode::G_LOAD && Opcode != TargetOpcode::G_SEXTLOAD &&
      Opcode != TargetOpcode::G_ZEXTLOAD && Opcode != TargetOpcode::G_STORE)
    return false;

  MatchInfo.IsPre = findPreIndexCandidate(MI, MatchInfo.Addr, MatchInfo.Base,
                                          MatchInfo.Offset);
  if (!MatchInfo.IsPre &&
      !findPostIndexCandidate(MI, MatchInfo.Addr, MatchInfo.Base,
                              MatchInfo.Offset))
    return false;

  return true;
}

void CombinerHelper::applyCombineIndexedLoadStore(
    MachineInstr &MI, IndexedLoadStoreMatchInfo &MatchInfo) {
  MachineInstr &AddrDef = *MRI.getUniqueVRegDef(MatchInfo.Addr);
  MachineIRBuilder MIRBuilder(MI);
  unsigned Opcode = MI.getOpcode();
  bool IsStore = Opcode == TargetOpcode::G_STORE;
  unsigned NewOpcode;
  switch (Opcode) {
  case TargetOpcode::G_LOAD:
    NewOpcode = TargetOpcode::G_INDEXED_LOAD;
    break;
  case TargetOpcode::G_SEXTLOAD:
    NewOpcode = TargetOpcode::G_INDEXED_SEXTLOAD;
    break;
  case TargetOpcode::G_ZEXTLOAD:
    NewOpcode = TargetOpcode::G_INDEXED_ZEXTLOAD;
    break;
  case TargetOpcode::G_STORE:
    NewOpcode = TargetOpcode::G_INDEXED_STORE;
    break;
  default:
    llvm_unreachable("Unknown load/store opcode");
  }

  auto MIB = MIRBuilder.buildInstr(NewOpcode);
  if (IsStore) {
    MIB.addDef(MatchInfo.Addr);
    MIB.addUse(MI.getOperand(0).getReg());
  } else {
    MIB.addDef(MI.getOperand(0).getReg());
    MIB.addDef(MatchInfo.Addr);
  }

  MIB.addUse(MatchInfo.Base);
  MIB.addUse(MatchInfo.Offset);
  MIB.addImm(MatchInfo.IsPre);
  MI.eraseFromParent();
  AddrDef.eraseFromParent();

  LLVM_DEBUG(dbgs() << "    Combinined to indexed operation");
}

bool CombinerHelper::matchElideBrByInvertingCond(MachineInstr &MI) {
  if (MI.getOpcode() != TargetOpcode::G_BR)
    return false;

  // Try to match the following:
  // bb1:
  //   %c(s32) = G_ICMP pred, %a, %b
  //   %c1(s1) = G_TRUNC %c(s32)
  //   G_BRCOND %c1, %bb2
  //   G_BR %bb3
  // bb2:
  // ...
  // bb3:

  // The above pattern does not have a fall through to the successor bb2, always
  // resulting in a branch no matter which path is taken. Here we try to find
  // and replace that pattern with conditional branch to bb3 and otherwise
  // fallthrough to bb2.

  MachineBasicBlock *MBB = MI.getParent();
  MachineBasicBlock::iterator BrIt(MI);
  if (BrIt == MBB->begin())
    return false;
  assert(std::next(BrIt) == MBB->end() && "expected G_BR to be a terminator");

  MachineInstr *BrCond = &*std::prev(BrIt);
  if (BrCond->getOpcode() != TargetOpcode::G_BRCOND)
    return false;

  // Check that the next block is the conditional branch target.
  if (!MBB->isLayoutSuccessor(BrCond->getOperand(1).getMBB()))
    return false;

  MachineInstr *CmpMI = MRI.getVRegDef(BrCond->getOperand(0).getReg());
  if (!CmpMI || CmpMI->getOpcode() != TargetOpcode::G_ICMP ||
      !MRI.hasOneNonDBGUse(CmpMI->getOperand(0).getReg()))
    return false;
  return true;
}

bool CombinerHelper::tryElideBrByInvertingCond(MachineInstr &MI) {
  if (!matchElideBrByInvertingCond(MI))
    return false;
  applyElideBrByInvertingCond(MI);
  return true;
}

void CombinerHelper::applyElideBrByInvertingCond(MachineInstr &MI) {
  MachineBasicBlock *BrTarget = MI.getOperand(0).getMBB();
  MachineBasicBlock::iterator BrIt(MI);
  MachineInstr *BrCond = &*std::prev(BrIt);
  MachineInstr *CmpMI = MRI.getVRegDef(BrCond->getOperand(0).getReg());

  CmpInst::Predicate InversePred = CmpInst::getInversePredicate(
      (CmpInst::Predicate)CmpMI->getOperand(1).getPredicate());

  // Invert the G_ICMP condition.
  Observer.changingInstr(*CmpMI);
  CmpMI->getOperand(1).setPredicate(InversePred);
  Observer.changedInstr(*CmpMI);

  // Change the conditional branch target.
  Observer.changingInstr(*BrCond);
  BrCond->getOperand(1).setMBB(BrTarget);
  Observer.changedInstr(*BrCond);
  MI.eraseFromParent();
}

static bool shouldLowerMemFuncForSize(const MachineFunction &MF) {
  // On Darwin, -Os means optimize for size without hurting performance, so
  // only really optimize for size when -Oz (MinSize) is used.
  if (MF.getTarget().getTargetTriple().isOSDarwin())
    return MF.getFunction().hasMinSize();
  return MF.getFunction().hasOptSize();
}

// Returns a list of types to use for memory op lowering in MemOps. A partial
// port of findOptimalMemOpLowering in TargetLowering.
static bool findGISelOptimalMemOpLowering(std::vector<LLT> &MemOps,
                                          unsigned Limit, const MemOp &Op,
                                          unsigned DstAS, unsigned SrcAS,
                                          const AttributeList &FuncAttributes,
                                          const TargetLowering &TLI) {
  if (Op.isMemcpyWithFixedDstAlign() && Op.getSrcAlign() < Op.getDstAlign())
    return false;

  LLT Ty = TLI.getOptimalMemOpLLT(Op, FuncAttributes);

  if (Ty == LLT()) {
    // Use the largest scalar type whose alignment constraints are satisfied.
    // We only need to check DstAlign here as SrcAlign is always greater or
    // equal to DstAlign (or zero).
    Ty = LLT::scalar(64);
    if (Op.isFixedDstAlign())
      while (Op.getDstAlign() < Ty.getSizeInBytes() &&
             !TLI.allowsMisalignedMemoryAccesses(Ty, DstAS, Op.getDstAlign()))
        Ty = LLT::scalar(Ty.getSizeInBytes());
    assert(Ty.getSizeInBits() > 0 && "Could not find valid type");
    // FIXME: check for the largest legal type we can load/store to.
  }

  unsigned NumMemOps = 0;
  uint64_t Size = Op.size();
  while (Size) {
    unsigned TySize = Ty.getSizeInBytes();
    while (TySize > Size) {
      // For now, only use non-vector load / store's for the left-over pieces.
      LLT NewTy = Ty;
      // FIXME: check for mem op safety and legality of the types. Not all of
      // SDAGisms map cleanly to GISel concepts.
      if (NewTy.isVector())
        NewTy = NewTy.getSizeInBits() > 64 ? LLT::scalar(64) : LLT::scalar(32);
      NewTy = LLT::scalar(PowerOf2Floor(NewTy.getSizeInBits() - 1));
      unsigned NewTySize = NewTy.getSizeInBytes();
      assert(NewTySize > 0 && "Could not find appropriate type");

      // If the new LLT cannot cover all of the remaining bits, then consider
      // issuing a (or a pair of) unaligned and overlapping load / store.
      bool Fast;
      // Need to get a VT equivalent for allowMisalignedMemoryAccesses().
      MVT VT = getMVTForLLT(Ty);
      if (NumMemOps && Op.allowOverlap() && NewTySize < Size &&
          TLI.allowsMisalignedMemoryAccesses(
              VT, DstAS, Op.isFixedDstAlign() ? Op.getDstAlign().value() : 0,
              MachineMemOperand::MONone, &Fast) &&
          Fast)
        TySize = Size;
      else {
        Ty = NewTy;
        TySize = NewTySize;
      }
    }

    if (++NumMemOps > Limit)
      return false;

    MemOps.push_back(Ty);
    Size -= TySize;
  }

  return true;
}

static Type *getTypeForLLT(LLT Ty, LLVMContext &C) {
  if (Ty.isVector())
    return FixedVectorType::get(IntegerType::get(C, Ty.getScalarSizeInBits()),
                                Ty.getNumElements());
  return IntegerType::get(C, Ty.getSizeInBits());
}

// Get a vectorized representation of the memset value operand, GISel edition.
static Register getMemsetValue(Register Val, LLT Ty, MachineIRBuilder &MIB) {
  MachineRegisterInfo &MRI = *MIB.getMRI();
  unsigned NumBits = Ty.getScalarSizeInBits();
  auto ValVRegAndVal = getConstantVRegValWithLookThrough(Val, MRI);
  if (!Ty.isVector() && ValVRegAndVal) {
    unsigned KnownVal = ValVRegAndVal->Value;
    APInt Scalar = APInt(8, KnownVal);
    APInt SplatVal = APInt::getSplat(NumBits, Scalar);
    return MIB.buildConstant(Ty, SplatVal).getReg(0);
  }

  // Extend the byte value to the larger type, and then multiply by a magic
  // value 0x010101... in order to replicate it across every byte.
  // Unless it's zero, in which case just emit a larger G_CONSTANT 0.
  if (ValVRegAndVal && ValVRegAndVal->Value == 0) {
    return MIB.buildConstant(Ty, 0).getReg(0);
  }

  LLT ExtType = Ty.getScalarType();
  auto ZExt = MIB.buildZExtOrTrunc(ExtType, Val);
  if (NumBits > 8) {
    APInt Magic = APInt::getSplat(NumBits, APInt(8, 0x01));
    auto MagicMI = MIB.buildConstant(ExtType, Magic);
    Val = MIB.buildMul(ExtType, ZExt, MagicMI).getReg(0);
  }

  // For vector types create a G_BUILD_VECTOR.
  if (Ty.isVector())
    Val = MIB.buildSplatVector(Ty, Val).getReg(0);

  return Val;
}

bool CombinerHelper::optimizeMemset(MachineInstr &MI, Register Dst,
                                    Register Val, unsigned KnownLen,
                                    Align Alignment, bool IsVolatile) {
  auto &MF = *MI.getParent()->getParent();
  const auto &TLI = *MF.getSubtarget().getTargetLowering();
  auto &DL = MF.getDataLayout();
  LLVMContext &C = MF.getFunction().getContext();

  assert(KnownLen != 0 && "Have a zero length memset length!");

  bool DstAlignCanChange = false;
  MachineFrameInfo &MFI = MF.getFrameInfo();
  bool OptSize = shouldLowerMemFuncForSize(MF);

  MachineInstr *FIDef = getOpcodeDef(TargetOpcode::G_FRAME_INDEX, Dst, MRI);
  if (FIDef && !MFI.isFixedObjectIndex(FIDef->getOperand(1).getIndex()))
    DstAlignCanChange = true;

  unsigned Limit = TLI.getMaxStoresPerMemset(OptSize);
  std::vector<LLT> MemOps;

  const auto &DstMMO = **MI.memoperands_begin();
  MachinePointerInfo DstPtrInfo = DstMMO.getPointerInfo();

  auto ValVRegAndVal = getConstantVRegValWithLookThrough(Val, MRI);
  bool IsZeroVal = ValVRegAndVal && ValVRegAndVal->Value == 0;

  if (!findGISelOptimalMemOpLowering(MemOps, Limit,
                                     MemOp::Set(KnownLen, DstAlignCanChange,
                                                Alignment,
                                                /*IsZeroMemset=*/IsZeroVal,
                                                /*IsVolatile=*/IsVolatile),
                                     DstPtrInfo.getAddrSpace(), ~0u,
                                     MF.getFunction().getAttributes(), TLI))
    return false;

  if (DstAlignCanChange) {
    // Get an estimate of the type from the LLT.
    Type *IRTy = getTypeForLLT(MemOps[0], C);
    Align NewAlign = DL.getABITypeAlign(IRTy);
    if (NewAlign > Alignment) {
      Alignment = NewAlign;
      unsigned FI = FIDef->getOperand(1).getIndex();
      // Give the stack frame object a larger alignment if needed.
      if (MFI.getObjectAlign(FI) < Alignment)
        MFI.setObjectAlignment(FI, Alignment);
    }
  }

  MachineIRBuilder MIB(MI);
  // Find the largest store and generate the bit pattern for it.
  LLT LargestTy = MemOps[0];
  for (unsigned i = 1; i < MemOps.size(); i++)
    if (MemOps[i].getSizeInBits() > LargestTy.getSizeInBits())
      LargestTy = MemOps[i];

  // The memset stored value is always defined as an s8, so in order to make it
  // work with larger store types we need to repeat the bit pattern across the
  // wider type.
  Register MemSetValue = getMemsetValue(Val, LargestTy, MIB);

  if (!MemSetValue)
    return false;

  // Generate the stores. For each store type in the list, we generate the
  // matching store of that type to the destination address.
  LLT PtrTy = MRI.getType(Dst);
  unsigned DstOff = 0;
  unsigned Size = KnownLen;
  for (unsigned I = 0; I < MemOps.size(); I++) {
    LLT Ty = MemOps[I];
    unsigned TySize = Ty.getSizeInBytes();
    if (TySize > Size) {
      // Issuing an unaligned load / store pair that overlaps with the previous
      // pair. Adjust the offset accordingly.
      assert(I == MemOps.size() - 1 && I != 0);
      DstOff -= TySize - Size;
    }

    // If this store is smaller than the largest store see whether we can get
    // the smaller value for free with a truncate.
    Register Value = MemSetValue;
    if (Ty.getSizeInBits() < LargestTy.getSizeInBits()) {
      MVT VT = getMVTForLLT(Ty);
      MVT LargestVT = getMVTForLLT(LargestTy);
      if (!LargestTy.isVector() && !Ty.isVector() &&
          TLI.isTruncateFree(LargestVT, VT))
        Value = MIB.buildTrunc(Ty, MemSetValue).getReg(0);
      else
        Value = getMemsetValue(Val, Ty, MIB);
      if (!Value)
        return false;
    }

    auto *StoreMMO =
        MF.getMachineMemOperand(&DstMMO, DstOff, Ty.getSizeInBytes());

    Register Ptr = Dst;
    if (DstOff != 0) {
      auto Offset =
          MIB.buildConstant(LLT::scalar(PtrTy.getSizeInBits()), DstOff);
      Ptr = MIB.buildPtrAdd(PtrTy, Dst, Offset).getReg(0);
    }

    MIB.buildStore(Value, Ptr, *StoreMMO);
    DstOff += Ty.getSizeInBytes();
    Size -= TySize;
  }

  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::optimizeMemcpy(MachineInstr &MI, Register Dst,
                                    Register Src, unsigned KnownLen,
                                    Align DstAlign, Align SrcAlign,
                                    bool IsVolatile) {
  auto &MF = *MI.getParent()->getParent();
  const auto &TLI = *MF.getSubtarget().getTargetLowering();
  auto &DL = MF.getDataLayout();
  LLVMContext &C = MF.getFunction().getContext();

  assert(KnownLen != 0 && "Have a zero length memcpy length!");

  bool DstAlignCanChange = false;
  MachineFrameInfo &MFI = MF.getFrameInfo();
  bool OptSize = shouldLowerMemFuncForSize(MF);
  Align Alignment = commonAlignment(DstAlign, SrcAlign);

  MachineInstr *FIDef = getOpcodeDef(TargetOpcode::G_FRAME_INDEX, Dst, MRI);
  if (FIDef && !MFI.isFixedObjectIndex(FIDef->getOperand(1).getIndex()))
    DstAlignCanChange = true;

  // FIXME: infer better src pointer alignment like SelectionDAG does here.
  // FIXME: also use the equivalent of isMemSrcFromConstant and alwaysinlining
  // if the memcpy is in a tail call position.

  unsigned Limit = TLI.getMaxStoresPerMemcpy(OptSize);
  std::vector<LLT> MemOps;

  const auto &DstMMO = **MI.memoperands_begin();
  const auto &SrcMMO = **std::next(MI.memoperands_begin());
  MachinePointerInfo DstPtrInfo = DstMMO.getPointerInfo();
  MachinePointerInfo SrcPtrInfo = SrcMMO.getPointerInfo();

  if (!findGISelOptimalMemOpLowering(
          MemOps, Limit,
          MemOp::Copy(KnownLen, DstAlignCanChange, Alignment, SrcAlign,
                      IsVolatile),
          DstPtrInfo.getAddrSpace(), SrcPtrInfo.getAddrSpace(),
          MF.getFunction().getAttributes(), TLI))
    return false;

  if (DstAlignCanChange) {
    // Get an estimate of the type from the LLT.
    Type *IRTy = getTypeForLLT(MemOps[0], C);
    Align NewAlign = DL.getABITypeAlign(IRTy);

    // Don't promote to an alignment that would require dynamic stack
    // realignment.
    const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
    if (!TRI->needsStackRealignment(MF))
      while (NewAlign > Alignment && DL.exceedsNaturalStackAlignment(NewAlign))
        NewAlign = NewAlign / 2;

    if (NewAlign > Alignment) {
      Alignment = NewAlign;
      unsigned FI = FIDef->getOperand(1).getIndex();
      // Give the stack frame object a larger alignment if needed.
      if (MFI.getObjectAlign(FI) < Alignment)
        MFI.setObjectAlignment(FI, Alignment);
    }
  }

  LLVM_DEBUG(dbgs() << "Inlining memcpy: " << MI << " into loads & stores\n");

  MachineIRBuilder MIB(MI);
  // Now we need to emit a pair of load and stores for each of the types we've
  // collected. I.e. for each type, generate a load from the source pointer of
  // that type width, and then generate a corresponding store to the dest buffer
  // of that value loaded. This can result in a sequence of loads and stores
  // mixed types, depending on what the target specifies as good types to use.
  unsigned CurrOffset = 0;
  LLT PtrTy = MRI.getType(Src);
  unsigned Size = KnownLen;
  for (auto CopyTy : MemOps) {
    // Issuing an unaligned load / store pair  that overlaps with the previous
    // pair. Adjust the offset accordingly.
    if (CopyTy.getSizeInBytes() > Size)
      CurrOffset -= CopyTy.getSizeInBytes() - Size;

    // Construct MMOs for the accesses.
    auto *LoadMMO =
        MF.getMachineMemOperand(&SrcMMO, CurrOffset, CopyTy.getSizeInBytes());
    auto *StoreMMO =
        MF.getMachineMemOperand(&DstMMO, CurrOffset, CopyTy.getSizeInBytes());

    // Create the load.
    Register LoadPtr = Src;
    Register Offset;
    if (CurrOffset != 0) {
      Offset = MIB.buildConstant(LLT::scalar(PtrTy.getSizeInBits()), CurrOffset)
                   .getReg(0);
      LoadPtr = MIB.buildPtrAdd(PtrTy, Src, Offset).getReg(0);
    }
    auto LdVal = MIB.buildLoad(CopyTy, LoadPtr, *LoadMMO);

    // Create the store.
    Register StorePtr =
        CurrOffset == 0 ? Dst : MIB.buildPtrAdd(PtrTy, Dst, Offset).getReg(0);
    MIB.buildStore(LdVal, StorePtr, *StoreMMO);
    CurrOffset += CopyTy.getSizeInBytes();
    Size -= CopyTy.getSizeInBytes();
  }

  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::optimizeMemmove(MachineInstr &MI, Register Dst,
                                     Register Src, unsigned KnownLen,
                                     Align DstAlign, Align SrcAlign,
                                     bool IsVolatile) {
  auto &MF = *MI.getParent()->getParent();
  const auto &TLI = *MF.getSubtarget().getTargetLowering();
  auto &DL = MF.getDataLayout();
  LLVMContext &C = MF.getFunction().getContext();

  assert(KnownLen != 0 && "Have a zero length memmove length!");

  bool DstAlignCanChange = false;
  MachineFrameInfo &MFI = MF.getFrameInfo();
  bool OptSize = shouldLowerMemFuncForSize(MF);
  Align Alignment = commonAlignment(DstAlign, SrcAlign);

  MachineInstr *FIDef = getOpcodeDef(TargetOpcode::G_FRAME_INDEX, Dst, MRI);
  if (FIDef && !MFI.isFixedObjectIndex(FIDef->getOperand(1).getIndex()))
    DstAlignCanChange = true;

  unsigned Limit = TLI.getMaxStoresPerMemmove(OptSize);
  std::vector<LLT> MemOps;

  const auto &DstMMO = **MI.memoperands_begin();
  const auto &SrcMMO = **std::next(MI.memoperands_begin());
  MachinePointerInfo DstPtrInfo = DstMMO.getPointerInfo();
  MachinePointerInfo SrcPtrInfo = SrcMMO.getPointerInfo();

  // FIXME: SelectionDAG always passes false for 'AllowOverlap', apparently due
  // to a bug in it's findOptimalMemOpLowering implementation. For now do the
  // same thing here.
  if (!findGISelOptimalMemOpLowering(
          MemOps, Limit,
          MemOp::Copy(KnownLen, DstAlignCanChange, Alignment, SrcAlign,
                      /*IsVolatile*/ true),
          DstPtrInfo.getAddrSpace(), SrcPtrInfo.getAddrSpace(),
          MF.getFunction().getAttributes(), TLI))
    return false;

  if (DstAlignCanChange) {
    // Get an estimate of the type from the LLT.
    Type *IRTy = getTypeForLLT(MemOps[0], C);
    Align NewAlign = DL.getABITypeAlign(IRTy);

    // Don't promote to an alignment that would require dynamic stack
    // realignment.
    const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
    if (!TRI->needsStackRealignment(MF))
      while (NewAlign > Alignment && DL.exceedsNaturalStackAlignment(NewAlign))
        NewAlign = NewAlign / 2;

    if (NewAlign > Alignment) {
      Alignment = NewAlign;
      unsigned FI = FIDef->getOperand(1).getIndex();
      // Give the stack frame object a larger alignment if needed.
      if (MFI.getObjectAlign(FI) < Alignment)
        MFI.setObjectAlignment(FI, Alignment);
    }
  }

  LLVM_DEBUG(dbgs() << "Inlining memmove: " << MI << " into loads & stores\n");

  MachineIRBuilder MIB(MI);
  // Memmove requires that we perform the loads first before issuing the stores.
  // Apart from that, this loop is pretty much doing the same thing as the
  // memcpy codegen function.
  unsigned CurrOffset = 0;
  LLT PtrTy = MRI.getType(Src);
  SmallVector<Register, 16> LoadVals;
  for (auto CopyTy : MemOps) {
    // Construct MMO for the load.
    auto *LoadMMO =
        MF.getMachineMemOperand(&SrcMMO, CurrOffset, CopyTy.getSizeInBytes());

    // Create the load.
    Register LoadPtr = Src;
    if (CurrOffset != 0) {
      auto Offset =
          MIB.buildConstant(LLT::scalar(PtrTy.getSizeInBits()), CurrOffset);
      LoadPtr = MIB.buildPtrAdd(PtrTy, Src, Offset).getReg(0);
    }
    LoadVals.push_back(MIB.buildLoad(CopyTy, LoadPtr, *LoadMMO).getReg(0));
    CurrOffset += CopyTy.getSizeInBytes();
  }

  CurrOffset = 0;
  for (unsigned I = 0; I < MemOps.size(); ++I) {
    LLT CopyTy = MemOps[I];
    // Now store the values loaded.
    auto *StoreMMO =
        MF.getMachineMemOperand(&DstMMO, CurrOffset, CopyTy.getSizeInBytes());

    Register StorePtr = Dst;
    if (CurrOffset != 0) {
      auto Offset =
          MIB.buildConstant(LLT::scalar(PtrTy.getSizeInBits()), CurrOffset);
      StorePtr = MIB.buildPtrAdd(PtrTy, Dst, Offset).getReg(0);
    }
    MIB.buildStore(LoadVals[I], StorePtr, *StoreMMO);
    CurrOffset += CopyTy.getSizeInBytes();
  }
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::tryCombineMemCpyFamily(MachineInstr &MI, unsigned MaxLen) {
  // This combine is fairly complex so it's not written with a separate
  // matcher function.
  assert(MI.getOpcode() == TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS);
  Intrinsic::ID ID = (Intrinsic::ID)MI.getIntrinsicID();
  assert((ID == Intrinsic::memcpy || ID == Intrinsic::memmove ||
          ID == Intrinsic::memset) &&
         "Expected a memcpy like intrinsic");

  auto MMOIt = MI.memoperands_begin();
  const MachineMemOperand *MemOp = *MMOIt;
  bool IsVolatile = MemOp->isVolatile();
  // Don't try to optimize volatile.
  if (IsVolatile)
    return false;

  Align DstAlign = MemOp->getBaseAlign();
  Align SrcAlign;
  Register Dst = MI.getOperand(1).getReg();
  Register Src = MI.getOperand(2).getReg();
  Register Len = MI.getOperand(3).getReg();

  if (ID != Intrinsic::memset) {
    assert(MMOIt != MI.memoperands_end() && "Expected a second MMO on MI");
    MemOp = *(++MMOIt);
    SrcAlign = MemOp->getBaseAlign();
  }

  // See if this is a constant length copy
  auto LenVRegAndVal = getConstantVRegValWithLookThrough(Len, MRI);
  if (!LenVRegAndVal)
    return false; // Leave it to the legalizer to lower it to a libcall.
  unsigned KnownLen = LenVRegAndVal->Value;

  if (KnownLen == 0) {
    MI.eraseFromParent();
    return true;
  }

  if (MaxLen && KnownLen > MaxLen)
    return false;

  if (ID == Intrinsic::memcpy)
    return optimizeMemcpy(MI, Dst, Src, KnownLen, DstAlign, SrcAlign, IsVolatile);
  if (ID == Intrinsic::memmove)
    return optimizeMemmove(MI, Dst, Src, KnownLen, DstAlign, SrcAlign, IsVolatile);
  if (ID == Intrinsic::memset)
    return optimizeMemset(MI, Dst, Src, KnownLen, DstAlign, IsVolatile);
  return false;
}

bool CombinerHelper::matchPtrAddImmedChain(MachineInstr &MI,
                                           PtrAddChain &MatchInfo) {
  // We're trying to match the following pattern:
  //   %t1 = G_PTR_ADD %base, G_CONSTANT imm1
  //   %root = G_PTR_ADD %t1, G_CONSTANT imm2
  // -->
  //   %root = G_PTR_ADD %base, G_CONSTANT (imm1 + imm2)

  if (MI.getOpcode() != TargetOpcode::G_PTR_ADD)
    return false;

  Register Add2 = MI.getOperand(1).getReg();
  Register Imm1 = MI.getOperand(2).getReg();
  auto MaybeImmVal = getConstantVRegValWithLookThrough(Imm1, MRI);
  if (!MaybeImmVal)
    return false;

  MachineInstr *Add2Def = MRI.getUniqueVRegDef(Add2);
  if (!Add2Def || Add2Def->getOpcode() != TargetOpcode::G_PTR_ADD)
    return false;

  Register Base = Add2Def->getOperand(1).getReg();
  Register Imm2 = Add2Def->getOperand(2).getReg();
  auto MaybeImm2Val = getConstantVRegValWithLookThrough(Imm2, MRI);
  if (!MaybeImm2Val)
    return false;

  // Pass the combined immediate to the apply function.
  MatchInfo.Imm = MaybeImmVal->Value + MaybeImm2Val->Value;
  MatchInfo.Base = Base;
  return true;
}

bool CombinerHelper::applyPtrAddImmedChain(MachineInstr &MI,
                                           PtrAddChain &MatchInfo) {
  assert(MI.getOpcode() == TargetOpcode::G_PTR_ADD && "Expected G_PTR_ADD");
  MachineIRBuilder MIB(MI);
  LLT OffsetTy = MRI.getType(MI.getOperand(2).getReg());
  auto NewOffset = MIB.buildConstant(OffsetTy, MatchInfo.Imm);
  Observer.changingInstr(MI);
  MI.getOperand(1).setReg(MatchInfo.Base);
  MI.getOperand(2).setReg(NewOffset.getReg(0));
  Observer.changedInstr(MI);
  return true;
}

bool CombinerHelper::matchCombineMulToShl(MachineInstr &MI,
                                          unsigned &ShiftVal) {
  assert(MI.getOpcode() == TargetOpcode::G_MUL && "Expected a G_MUL");
  auto MaybeImmVal =
      getConstantVRegValWithLookThrough(MI.getOperand(2).getReg(), MRI);
  if (!MaybeImmVal || !isPowerOf2_64(MaybeImmVal->Value))
    return false;
  ShiftVal = Log2_64(MaybeImmVal->Value);
  return true;
}

bool CombinerHelper::applyCombineMulToShl(MachineInstr &MI,
                                          unsigned &ShiftVal) {
  assert(MI.getOpcode() == TargetOpcode::G_MUL && "Expected a G_MUL");
  MachineIRBuilder MIB(MI);
  LLT ShiftTy = MRI.getType(MI.getOperand(0).getReg());
  auto ShiftCst = MIB.buildConstant(ShiftTy, ShiftVal);
  Observer.changingInstr(MI);
  MI.setDesc(MIB.getTII().get(TargetOpcode::G_SHL));
  MI.getOperand(2).setReg(ShiftCst.getReg(0));
  Observer.changedInstr(MI);
  return true;
}

bool CombinerHelper::matchCombineShiftToUnmerge(MachineInstr &MI,
                                                unsigned TargetShiftSize,
                                                unsigned &ShiftVal) {
  assert((MI.getOpcode() == TargetOpcode::G_SHL ||
          MI.getOpcode() == TargetOpcode::G_LSHR ||
          MI.getOpcode() == TargetOpcode::G_ASHR) && "Expected a shift");

  LLT Ty = MRI.getType(MI.getOperand(0).getReg());
  if (Ty.isVector()) // TODO:
    return false;

  // Don't narrow further than the requested size.
  unsigned Size = Ty.getSizeInBits();
  if (Size <= TargetShiftSize)
    return false;

  auto MaybeImmVal =
    getConstantVRegValWithLookThrough(MI.getOperand(2).getReg(), MRI);
  if (!MaybeImmVal)
    return false;

  ShiftVal = MaybeImmVal->Value;
  return ShiftVal >= Size / 2 && ShiftVal < Size;
}

bool CombinerHelper::applyCombineShiftToUnmerge(MachineInstr &MI,
                                                const unsigned &ShiftVal) {
  Register DstReg = MI.getOperand(0).getReg();
  Register SrcReg = MI.getOperand(1).getReg();
  LLT Ty = MRI.getType(SrcReg);
  unsigned Size = Ty.getSizeInBits();
  unsigned HalfSize = Size / 2;
  assert(ShiftVal >= HalfSize);

  LLT HalfTy = LLT::scalar(HalfSize);

  Builder.setInstr(MI);
  auto Unmerge = Builder.buildUnmerge(HalfTy, SrcReg);
  unsigned NarrowShiftAmt = ShiftVal - HalfSize;

  if (MI.getOpcode() == TargetOpcode::G_LSHR) {
    Register Narrowed = Unmerge.getReg(1);

    //  dst = G_LSHR s64:x, C for C >= 32
    // =>
    //   lo, hi = G_UNMERGE_VALUES x
    //   dst = G_MERGE_VALUES (G_LSHR hi, C - 32), 0

    if (NarrowShiftAmt != 0) {
      Narrowed = Builder.buildLShr(HalfTy, Narrowed,
        Builder.buildConstant(HalfTy, NarrowShiftAmt)).getReg(0);
    }

    auto Zero = Builder.buildConstant(HalfTy, 0);
    Builder.buildMerge(DstReg, { Narrowed, Zero });
  } else if (MI.getOpcode() == TargetOpcode::G_SHL) {
    Register Narrowed = Unmerge.getReg(0);
    //  dst = G_SHL s64:x, C for C >= 32
    // =>
    //   lo, hi = G_UNMERGE_VALUES x
    //   dst = G_MERGE_VALUES 0, (G_SHL hi, C - 32)
    if (NarrowShiftAmt != 0) {
      Narrowed = Builder.buildShl(HalfTy, Narrowed,
        Builder.buildConstant(HalfTy, NarrowShiftAmt)).getReg(0);
    }

    auto Zero = Builder.buildConstant(HalfTy, 0);
    Builder.buildMerge(DstReg, { Zero, Narrowed });
  } else {
    assert(MI.getOpcode() == TargetOpcode::G_ASHR);
    auto Hi = Builder.buildAShr(
      HalfTy, Unmerge.getReg(1),
      Builder.buildConstant(HalfTy, HalfSize - 1));

    if (ShiftVal == HalfSize) {
      // (G_ASHR i64:x, 32) ->
      //   G_MERGE_VALUES hi_32(x), (G_ASHR hi_32(x), 31)
      Builder.buildMerge(DstReg, { Unmerge.getReg(1), Hi });
    } else if (ShiftVal == Size - 1) {
      // Don't need a second shift.
      // (G_ASHR i64:x, 63) ->
      //   %narrowed = (G_ASHR hi_32(x), 31)
      //   G_MERGE_VALUES %narrowed, %narrowed
      Builder.buildMerge(DstReg, { Hi, Hi });
    } else {
      auto Lo = Builder.buildAShr(
        HalfTy, Unmerge.getReg(1),
        Builder.buildConstant(HalfTy, ShiftVal - HalfSize));

      // (G_ASHR i64:x, C) ->, for C >= 32
      //   G_MERGE_VALUES (G_ASHR hi_32(x), C - 32), (G_ASHR hi_32(x), 31)
      Builder.buildMerge(DstReg, { Lo, Hi });
    }
  }

  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::tryCombineShiftToUnmerge(MachineInstr &MI,
                                              unsigned TargetShiftAmount) {
  unsigned ShiftAmt;
  if (matchCombineShiftToUnmerge(MI, TargetShiftAmount, ShiftAmt)) {
    applyCombineShiftToUnmerge(MI, ShiftAmt);
    return true;
  }

  return false;
}

bool CombinerHelper::matchAnyExplicitUseIsUndef(MachineInstr &MI) {
  return any_of(MI.explicit_uses(), [this](const MachineOperand &MO) {
    return MO.isReg() &&
           getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, MO.getReg(), MRI);
  });
}

bool CombinerHelper::matchAllExplicitUsesAreUndef(MachineInstr &MI) {
  return all_of(MI.explicit_uses(), [this](const MachineOperand &MO) {
    return !MO.isReg() ||
           getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, MO.getReg(), MRI);
  });
}

bool CombinerHelper::matchUndefShuffleVectorMask(MachineInstr &MI) {
  assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR);
  ArrayRef<int> Mask = MI.getOperand(3).getShuffleMask();
  return all_of(Mask, [](int Elt) { return Elt < 0; });
}

bool CombinerHelper::matchUndefStore(MachineInstr &MI) {
  assert(MI.getOpcode() == TargetOpcode::G_STORE);
  return getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, MI.getOperand(0).getReg(),
                      MRI);
}

bool CombinerHelper::eraseInst(MachineInstr &MI) {
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::matchEqualDefs(const MachineOperand &MOP1,
                                    const MachineOperand &MOP2) {
  if (!MOP1.isReg() || !MOP2.isReg())
    return false;
  MachineInstr *I1 = getDefIgnoringCopies(MOP1.getReg(), MRI);
  if (!I1)
    return false;
  MachineInstr *I2 = getDefIgnoringCopies(MOP2.getReg(), MRI);
  if (!I2)
    return false;

  // Handle a case like this:
  //
  // %0:_(s64), %1:_(s64) = G_UNMERGE_VALUES %2:_(<2 x s64>)
  //
  // Even though %0 and %1 are produced by the same instruction they are not
  // the same values.
  if (I1 == I2)
    return MOP1.getReg() == MOP2.getReg();

  // If we have an instruction which loads or stores, we can't guarantee that
  // it is identical.
  //
  // For example, we may have
  //
  // %x1 = G_LOAD %addr (load N from @somewhere)
  // ...
  // call @foo
  // ...
  // %x2 = G_LOAD %addr (load N from @somewhere)
  // ...
  // %or = G_OR %x1, %x2
  //
  // It's possible that @foo will modify whatever lives at the address we're
  // loading from. To be safe, let's just assume that all loads and stores
  // are different (unless we have something which is guaranteed to not
  // change.)
  if (I1->mayLoadOrStore() && !I1->isDereferenceableInvariantLoad(nullptr))
    return false;

  // Check for physical registers on the instructions first to avoid cases
  // like this:
  //
  // %a = COPY $physreg
  // ...
  // SOMETHING implicit-def $physreg
  // ...
  // %b = COPY $physreg
  //
  // These copies are not equivalent.
  if (any_of(I1->uses(), [](const MachineOperand &MO) {
        return MO.isReg() && MO.getReg().isPhysical();
      })) {
    // Check if we have a case like this:
    //
    // %a = COPY $physreg
    // %b = COPY %a
    //
    // In this case, I1 and I2 will both be equal to %a = COPY $physreg.
    // From that, we know that they must have the same value, since they must
    // have come from the same COPY.
    return I1->isIdenticalTo(*I2);
  }

  // We don't have any physical registers, so we don't necessarily need the
  // same vreg defs.
  //
  // On the off-chance that there's some target instruction feeding into the
  // instruction, let's use produceSameValue instead of isIdenticalTo.
  return Builder.getTII().produceSameValue(*I1, *I2, &MRI);
}

bool CombinerHelper::matchConstantOp(const MachineOperand &MOP, int64_t C) {
  if (!MOP.isReg())
    return false;
  // MIPatternMatch doesn't let us look through G_ZEXT etc.
  auto ValAndVReg = getConstantVRegValWithLookThrough(MOP.getReg(), MRI);
  return ValAndVReg && ValAndVReg->Value == C;
}

bool CombinerHelper::replaceSingleDefInstWithOperand(MachineInstr &MI,
                                                     unsigned OpIdx) {
  assert(MI.getNumExplicitDefs() == 1 && "Expected one explicit def?");
  Register OldReg = MI.getOperand(0).getReg();
  Register Replacement = MI.getOperand(OpIdx).getReg();
  assert(canReplaceReg(OldReg, Replacement, MRI) && "Cannot replace register?");
  MI.eraseFromParent();
  replaceRegWith(MRI, OldReg, Replacement);
  return true;
}

bool CombinerHelper::matchSelectSameVal(MachineInstr &MI) {
  assert(MI.getOpcode() == TargetOpcode::G_SELECT);
  // Match (cond ? x : x)
  return matchEqualDefs(MI.getOperand(2), MI.getOperand(3)) &&
         canReplaceReg(MI.getOperand(0).getReg(), MI.getOperand(2).getReg(),
                       MRI);
}

bool CombinerHelper::matchBinOpSameVal(MachineInstr &MI) {
  return matchEqualDefs(MI.getOperand(1), MI.getOperand(2)) &&
         canReplaceReg(MI.getOperand(0).getReg(), MI.getOperand(1).getReg(),
                       MRI);
}

bool CombinerHelper::matchOperandIsZero(MachineInstr &MI, unsigned OpIdx) {
  return matchConstantOp(MI.getOperand(OpIdx), 0) &&
         canReplaceReg(MI.getOperand(0).getReg(), MI.getOperand(OpIdx).getReg(),
                       MRI);
}

bool CombinerHelper::replaceInstWithFConstant(MachineInstr &MI, double C) {
  assert(MI.getNumDefs() == 1 && "Expected only one def?");
  Builder.setInstr(MI);
  Builder.buildFConstant(MI.getOperand(0), C);
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::replaceInstWithConstant(MachineInstr &MI, int64_t C) {
  assert(MI.getNumDefs() == 1 && "Expected only one def?");
  Builder.setInstr(MI);
  Builder.buildConstant(MI.getOperand(0), C);
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::replaceInstWithUndef(MachineInstr &MI) {
  assert(MI.getNumDefs() == 1 && "Expected only one def?");
  Builder.setInstr(MI);
  Builder.buildUndef(MI.getOperand(0));
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::matchSimplifyAddToSub(
    MachineInstr &MI, std::tuple<Register, Register> &MatchInfo) {
  Register LHS = MI.getOperand(1).getReg();
  Register RHS = MI.getOperand(2).getReg();
  Register &NewLHS = std::get<0>(MatchInfo);
  Register &NewRHS = std::get<1>(MatchInfo);

  // Helper lambda to check for opportunities for
  // ((0-A) + B) -> B - A
  // (A + (0-B)) -> A - B
  auto CheckFold = [&](Register &MaybeSub, Register &MaybeNewLHS) {
    int64_t Cst;
    if (!mi_match(MaybeSub, MRI, m_GSub(m_ICst(Cst), m_Reg(NewRHS))) ||
        Cst != 0)
      return false;
    NewLHS = MaybeNewLHS;
    return true;
  };

  return CheckFold(LHS, RHS) || CheckFold(RHS, LHS);
}

bool CombinerHelper::applySimplifyAddToSub(
    MachineInstr &MI, std::tuple<Register, Register> &MatchInfo) {
  Builder.setInstr(MI);
  Register SubLHS, SubRHS;
  std::tie(SubLHS, SubRHS) = MatchInfo;
  Builder.buildSub(MI.getOperand(0).getReg(), SubLHS, SubRHS);
  MI.eraseFromParent();
  return true;
}

bool CombinerHelper::tryCombine(MachineInstr &MI) {
  if (tryCombineCopy(MI))
    return true;
  if (tryCombineExtendingLoads(MI))
    return true;
  if (tryCombineIndexedLoadStore(MI))
    return true;
  return false;
}