llvm.org GIT mirror llvm / 8a028a2 lib / Target / X86 / X86FloatingPoint.cpp
8a028a2

Tree @8a028a2 (Download .tar.gz)

X86FloatingPoint.cpp @8a028a2raw · 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
//===-- X86FloatingPoint.cpp - Floating point Reg -> Stack converter ------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the pass which converts floating point instructions from
// pseudo registers into register stack instructions.  This pass uses live
// variable information to indicate where the FPn registers are used and their
// lifetimes.
//
// The x87 hardware tracks liveness of the stack registers, so it is necessary
// to implement exact liveness tracking between basic blocks. The CFG edges are
// partitioned into bundles where the same FP registers must be live in
// identical stack positions. Instructions are inserted at the end of each basic
// block to rearrange the live registers to match the outgoing bundle.
//
// This approach avoids splitting critical edges at the potential cost of more
// live register shuffling instructions when critical edges are present.
//
//===----------------------------------------------------------------------===//

#include "X86.h"
#include "X86InstrInfo.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/EdgeBundles.h"
#include "llvm/CodeGen/LivePhysRegs.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/IR/InlineAsm.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/TargetSubtargetInfo.h"
#include <algorithm>
#include <bitset>
using namespace llvm;

#define DEBUG_TYPE "x86-codegen"

STATISTIC(NumFXCH, "Number of fxch instructions inserted");
STATISTIC(NumFP  , "Number of floating point instructions");

namespace {
  const unsigned ScratchFPReg = 7;

  struct FPS : public MachineFunctionPass {
    static char ID;
    FPS() : MachineFunctionPass(ID) {
      initializeEdgeBundlesPass(*PassRegistry::getPassRegistry());
      // This is really only to keep valgrind quiet.
      // The logic in isLive() is too much for it.
      memset(Stack, 0, sizeof(Stack));
      memset(RegMap, 0, sizeof(RegMap));
    }

    void getAnalysisUsage(AnalysisUsage &AU) const override {
      AU.setPreservesCFG();
      AU.addRequired<EdgeBundles>();
      AU.addPreservedID(MachineLoopInfoID);
      AU.addPreservedID(MachineDominatorsID);
      MachineFunctionPass::getAnalysisUsage(AU);
    }

    bool runOnMachineFunction(MachineFunction &MF) override;

    MachineFunctionProperties getRequiredProperties() const override {
      return MachineFunctionProperties().set(
          MachineFunctionProperties::Property::NoVRegs);
    }

    StringRef getPassName() const override { return "X86 FP Stackifier"; }

  private:
    const TargetInstrInfo *TII; // Machine instruction info.

    // Two CFG edges are related if they leave the same block, or enter the same
    // block. The transitive closure of an edge under this relation is a
    // LiveBundle. It represents a set of CFG edges where the live FP stack
    // registers must be allocated identically in the x87 stack.
    //
    // A LiveBundle is usually all the edges leaving a block, or all the edges
    // entering a block, but it can contain more edges if critical edges are
    // present.
    //
    // The set of live FP registers in a LiveBundle is calculated by bundleCFG,
    // but the exact mapping of FP registers to stack slots is fixed later.
    struct LiveBundle {
      // Bit mask of live FP registers. Bit 0 = FP0, bit 1 = FP1, &c.
      unsigned Mask;

      // Number of pre-assigned live registers in FixStack. This is 0 when the
      // stack order has not yet been fixed.
      unsigned FixCount;

      // Assigned stack order for live-in registers.
      // FixStack[i] == getStackEntry(i) for all i < FixCount.
      unsigned char FixStack[8];

      LiveBundle() : Mask(0), FixCount(0) {}

      // Have the live registers been assigned a stack order yet?
      bool isFixed() const { return !Mask || FixCount; }
    };

    // Numbered LiveBundle structs. LiveBundles[0] is used for all CFG edges
    // with no live FP registers.
    SmallVector<LiveBundle, 8> LiveBundles;

    // The edge bundle analysis provides indices into the LiveBundles vector.
    EdgeBundles *Bundles;

    // Return a bitmask of FP registers in block's live-in list.
    static unsigned calcLiveInMask(MachineBasicBlock *MBB) {
      unsigned Mask = 0;
      for (const auto &LI : MBB->liveins()) {
        if (LI.PhysReg < X86::FP0 || LI.PhysReg > X86::FP6)
          continue;
        Mask |= 1 << (LI.PhysReg - X86::FP0);
      }
      return Mask;
    }

    // Partition all the CFG edges into LiveBundles.
    void bundleCFG(MachineFunction &MF);

    MachineBasicBlock *MBB;     // Current basic block

    // The hardware keeps track of how many FP registers are live, so we have
    // to model that exactly. Usually, each live register corresponds to an
    // FP<n> register, but when dealing with calls, returns, and inline
    // assembly, it is sometimes necessary to have live scratch registers.
    unsigned Stack[8];          // FP<n> Registers in each stack slot...
    unsigned StackTop;          // The current top of the FP stack.

    enum {
      NumFPRegs = 8             // Including scratch pseudo-registers.
    };

    // For each live FP<n> register, point to its Stack[] entry.
    // The first entries correspond to FP0-FP6, the rest are scratch registers
    // used when we need slightly different live registers than what the
    // register allocator thinks.
    unsigned RegMap[NumFPRegs];

    // Set up our stack model to match the incoming registers to MBB.
    void setupBlockStack();

    // Shuffle live registers to match the expectations of successor blocks.
    void finishBlockStack();

#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
    void dumpStack() const {
      dbgs() << "Stack contents:";
      for (unsigned i = 0; i != StackTop; ++i) {
        dbgs() << " FP" << Stack[i];
        assert(RegMap[Stack[i]] == i && "Stack[] doesn't match RegMap[]!");
      }
    }
#endif

    /// getSlot - Return the stack slot number a particular register number is
    /// in.
    unsigned getSlot(unsigned RegNo) const {
      assert(RegNo < NumFPRegs && "Regno out of range!");
      return RegMap[RegNo];
    }

    /// isLive - Is RegNo currently live in the stack?
    bool isLive(unsigned RegNo) const {
      unsigned Slot = getSlot(RegNo);
      return Slot < StackTop && Stack[Slot] == RegNo;
    }

    /// getStackEntry - Return the X86::FP<n> register in register ST(i).
    unsigned getStackEntry(unsigned STi) const {
      if (STi >= StackTop)
        report_fatal_error("Access past stack top!");
      return Stack[StackTop-1-STi];
    }

    /// getSTReg - Return the X86::ST(i) register which contains the specified
    /// FP<RegNo> register.
    unsigned getSTReg(unsigned RegNo) const {
      return StackTop - 1 - getSlot(RegNo) + X86::ST0;
    }

    // pushReg - Push the specified FP<n> register onto the stack.
    void pushReg(unsigned Reg) {
      assert(Reg < NumFPRegs && "Register number out of range!");
      if (StackTop >= 8)
        report_fatal_error("Stack overflow!");
      Stack[StackTop] = Reg;
      RegMap[Reg] = StackTop++;
    }

    // popReg - Pop a register from the stack.
    void popReg() {
      if (StackTop == 0)
        report_fatal_error("Cannot pop empty stack!");
      RegMap[Stack[--StackTop]] = ~0;     // Update state
    }

    bool isAtTop(unsigned RegNo) const { return getSlot(RegNo) == StackTop-1; }
    void moveToTop(unsigned RegNo, MachineBasicBlock::iterator I) {
      DebugLoc dl = I == MBB->end() ? DebugLoc() : I->getDebugLoc();
      if (isAtTop(RegNo)) return;

      unsigned STReg = getSTReg(RegNo);
      unsigned RegOnTop = getStackEntry(0);

      // Swap the slots the regs are in.
      std::swap(RegMap[RegNo], RegMap[RegOnTop]);

      // Swap stack slot contents.
      if (RegMap[RegOnTop] >= StackTop)
        report_fatal_error("Access past stack top!");
      std::swap(Stack[RegMap[RegOnTop]], Stack[StackTop-1]);

      // Emit an fxch to update the runtime processors version of the state.
      BuildMI(*MBB, I, dl, TII->get(X86::XCH_F)).addReg(STReg);
      ++NumFXCH;
    }

    void duplicateToTop(unsigned RegNo, unsigned AsReg,
                        MachineBasicBlock::iterator I) {
      DebugLoc dl = I == MBB->end() ? DebugLoc() : I->getDebugLoc();
      unsigned STReg = getSTReg(RegNo);
      pushReg(AsReg);   // New register on top of stack

      BuildMI(*MBB, I, dl, TII->get(X86::LD_Frr)).addReg(STReg);
    }

    /// popStackAfter - Pop the current value off of the top of the FP stack
    /// after the specified instruction.
    void popStackAfter(MachineBasicBlock::iterator &I);

    /// freeStackSlotAfter - Free the specified register from the register
    /// stack, so that it is no longer in a register.  If the register is
    /// currently at the top of the stack, we just pop the current instruction,
    /// otherwise we store the current top-of-stack into the specified slot,
    /// then pop the top of stack.
    void freeStackSlotAfter(MachineBasicBlock::iterator &I, unsigned Reg);

    /// freeStackSlotBefore - Just the pop, no folding. Return the inserted
    /// instruction.
    MachineBasicBlock::iterator
    freeStackSlotBefore(MachineBasicBlock::iterator I, unsigned FPRegNo);

    /// Adjust the live registers to be the set in Mask.
    void adjustLiveRegs(unsigned Mask, MachineBasicBlock::iterator I);

    /// Shuffle the top FixCount stack entries such that FP reg FixStack[0] is
    /// st(0), FP reg FixStack[1] is st(1) etc.
    void shuffleStackTop(const unsigned char *FixStack, unsigned FixCount,
                         MachineBasicBlock::iterator I);

    bool processBasicBlock(MachineFunction &MF, MachineBasicBlock &MBB);

    void handleCall(MachineBasicBlock::iterator &I);
    void handleReturn(MachineBasicBlock::iterator &I);
    void handleZeroArgFP(MachineBasicBlock::iterator &I);
    void handleOneArgFP(MachineBasicBlock::iterator &I);
    void handleOneArgFPRW(MachineBasicBlock::iterator &I);
    void handleTwoArgFP(MachineBasicBlock::iterator &I);
    void handleCompareFP(MachineBasicBlock::iterator &I);
    void handleCondMovFP(MachineBasicBlock::iterator &I);
    void handleSpecialFP(MachineBasicBlock::iterator &I);

    // Check if a COPY instruction is using FP registers.
    static bool isFPCopy(MachineInstr &MI) {
      unsigned DstReg = MI.getOperand(0).getReg();
      unsigned SrcReg = MI.getOperand(1).getReg();

      return X86::RFP80RegClass.contains(DstReg) ||
        X86::RFP80RegClass.contains(SrcReg);
    }

    void setKillFlags(MachineBasicBlock &MBB) const;
  };
  char FPS::ID = 0;
}

FunctionPass *llvm::createX86FloatingPointStackifierPass() { return new FPS(); }

/// getFPReg - Return the X86::FPx register number for the specified operand.
/// For example, this returns 3 for X86::FP3.
static unsigned getFPReg(const MachineOperand &MO) {
  assert(MO.isReg() && "Expected an FP register!");
  unsigned Reg = MO.getReg();
  assert(Reg >= X86::FP0 && Reg <= X86::FP6 && "Expected FP register!");
  return Reg - X86::FP0;
}

/// runOnMachineFunction - Loop over all of the basic blocks, transforming FP
/// register references into FP stack references.
///
bool FPS::runOnMachineFunction(MachineFunction &MF) {
  // We only need to run this pass if there are any FP registers used in this
  // function.  If it is all integer, there is nothing for us to do!
  bool FPIsUsed = false;

  static_assert(X86::FP6 == X86::FP0+6, "Register enums aren't sorted right!");
  const MachineRegisterInfo &MRI = MF.getRegInfo();
  for (unsigned i = 0; i <= 6; ++i)
    if (!MRI.reg_nodbg_empty(X86::FP0 + i)) {
      FPIsUsed = true;
      break;
    }

  // Early exit.
  if (!FPIsUsed) return false;

  Bundles = &getAnalysis<EdgeBundles>();
  TII = MF.getSubtarget().getInstrInfo();

  // Prepare cross-MBB liveness.
  bundleCFG(MF);

  StackTop = 0;

  // Process the function in depth first order so that we process at least one
  // of the predecessors for every reachable block in the function.
  df_iterator_default_set<MachineBasicBlock*> Processed;
  MachineBasicBlock *Entry = &MF.front();

  LiveBundle &Bundle =
    LiveBundles[Bundles->getBundle(Entry->getNumber(), false)];
  
  // In regcall convention, some FP registers may not be passed through
  // the stack, so they will need to be assigned to the stack first
  if ((Entry->getParent()->getFunction()->getCallingConv() ==
    CallingConv::X86_RegCall) && (Bundle.Mask && !Bundle.FixCount)) {
    // In the register calling convention, up to one FP argument could be 
    // saved in the first FP register.
    // If bundle.mask is non-zero and Bundle.FixCount is zero, it means
    // that the FP registers contain arguments.
    // The actual value is passed in FP0.
    // Here we fix the stack and mark FP0 as pre-assigned register.
    assert((Bundle.Mask & 0xFE) == 0 &&
      "Only FP0 could be passed as an argument");
    Bundle.FixCount = 1;
    Bundle.FixStack[0] = 0;
  }

  bool Changed = false;
  for (MachineBasicBlock *BB : depth_first_ext(Entry, Processed))
    Changed |= processBasicBlock(MF, *BB);

  // Process any unreachable blocks in arbitrary order now.
  if (MF.size() != Processed.size())
    for (MachineBasicBlock &BB : MF)
      if (Processed.insert(&BB).second)
        Changed |= processBasicBlock(MF, BB);

  LiveBundles.clear();

  return Changed;
}

/// bundleCFG - Scan all the basic blocks to determine consistent live-in and
/// live-out sets for the FP registers. Consistent means that the set of
/// registers live-out from a block is identical to the live-in set of all
/// successors. This is not enforced by the normal live-in lists since
/// registers may be implicitly defined, or not used by all successors.
void FPS::bundleCFG(MachineFunction &MF) {
  assert(LiveBundles.empty() && "Stale data in LiveBundles");
  LiveBundles.resize(Bundles->getNumBundles());

  // Gather the actual live-in masks for all MBBs.
  for (MachineBasicBlock &MBB : MF) {
    const unsigned Mask = calcLiveInMask(&MBB);
    if (!Mask)
      continue;
    // Update MBB ingoing bundle mask.
    LiveBundles[Bundles->getBundle(MBB.getNumber(), false)].Mask |= Mask;
  }
}

/// processBasicBlock - Loop over all of the instructions in the basic block,
/// transforming FP instructions into their stack form.
///
bool FPS::processBasicBlock(MachineFunction &MF, MachineBasicBlock &BB) {
  bool Changed = false;
  MBB = &BB;

  setKillFlags(BB);
  setupBlockStack();

  for (MachineBasicBlock::iterator I = BB.begin(); I != BB.end(); ++I) {
    MachineInstr &MI = *I;
    uint64_t Flags = MI.getDesc().TSFlags;

    unsigned FPInstClass = Flags & X86II::FPTypeMask;
    if (MI.isInlineAsm())
      FPInstClass = X86II::SpecialFP;

    if (MI.isCopy() && isFPCopy(MI))
      FPInstClass = X86II::SpecialFP;

    if (MI.isImplicitDef() &&
        X86::RFP80RegClass.contains(MI.getOperand(0).getReg()))
      FPInstClass = X86II::SpecialFP;

    if (MI.isCall())
      FPInstClass = X86II::SpecialFP;

    if (FPInstClass == X86II::NotFP)
      continue;  // Efficiently ignore non-fp insts!

    MachineInstr *PrevMI = nullptr;
    if (I != BB.begin())
      PrevMI = &*std::prev(I);

    ++NumFP;  // Keep track of # of pseudo instrs
    DEBUG(dbgs() << "\nFPInst:\t" << MI);

    // Get dead variables list now because the MI pointer may be deleted as part
    // of processing!
    SmallVector<unsigned, 8> DeadRegs;
    for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
      const MachineOperand &MO = MI.getOperand(i);
      if (MO.isReg() && MO.isDead())
        DeadRegs.push_back(MO.getReg());
    }

    switch (FPInstClass) {
    case X86II::ZeroArgFP:  handleZeroArgFP(I); break;
    case X86II::OneArgFP:   handleOneArgFP(I);  break;  // fstp ST(0)
    case X86II::OneArgFPRW: handleOneArgFPRW(I); break; // ST(0) = fsqrt(ST(0))
    case X86II::TwoArgFP:   handleTwoArgFP(I);  break;
    case X86II::CompareFP:  handleCompareFP(I); break;
    case X86II::CondMovFP:  handleCondMovFP(I); break;
    case X86II::SpecialFP:  handleSpecialFP(I); break;
    default: llvm_unreachable("Unknown FP Type!");
    }

    // Check to see if any of the values defined by this instruction are dead
    // after definition.  If so, pop them.
    for (unsigned i = 0, e = DeadRegs.size(); i != e; ++i) {
      unsigned Reg = DeadRegs[i];
      // Check if Reg is live on the stack. An inline-asm register operand that
      // is in the clobber list and marked dead might not be live on the stack.
      if (Reg >= X86::FP0 && Reg <= X86::FP6 && isLive(Reg-X86::FP0)) {
        DEBUG(dbgs() << "Register FP#" << Reg-X86::FP0 << " is dead!\n");
        freeStackSlotAfter(I, Reg-X86::FP0);
      }
    }

    // Print out all of the instructions expanded to if -debug
    DEBUG({
      MachineBasicBlock::iterator PrevI = PrevMI;
      if (I == PrevI) {
        dbgs() << "Just deleted pseudo instruction\n";
      } else {
        MachineBasicBlock::iterator Start = I;
        // Rewind to first instruction newly inserted.
        while (Start != BB.begin() && std::prev(Start) != PrevI)
          --Start;
        dbgs() << "Inserted instructions:\n\t";
        Start->print(dbgs());
        while (++Start != std::next(I)) {
        }
      }
      dumpStack();
    });
    (void)PrevMI;

    Changed = true;
  }

  finishBlockStack();

  return Changed;
}

/// setupBlockStack - Use the live bundles to set up our model of the stack
/// to match predecessors' live out stack.
void FPS::setupBlockStack() {
  DEBUG(dbgs() << "\nSetting up live-ins for BB#" << MBB->getNumber()
               << " derived from " << MBB->getName() << ".\n");
  StackTop = 0;
  // Get the live-in bundle for MBB.
  const LiveBundle &Bundle =
    LiveBundles[Bundles->getBundle(MBB->getNumber(), false)];

  if (!Bundle.Mask) {
    DEBUG(dbgs() << "Block has no FP live-ins.\n");
    return;
  }

  // Depth-first iteration should ensure that we always have an assigned stack.
  assert(Bundle.isFixed() && "Reached block before any predecessors");

  // Push the fixed live-in registers.
  for (unsigned i = Bundle.FixCount; i > 0; --i) {
    MBB->addLiveIn(X86::ST0+i-1);
    DEBUG(dbgs() << "Live-in st(" << (i-1) << "): %FP"
                 << unsigned(Bundle.FixStack[i-1]) << '\n');
    pushReg(Bundle.FixStack[i-1]);
  }

  // Kill off unwanted live-ins. This can happen with a critical edge.
  // FIXME: We could keep these live registers around as zombies. They may need
  // to be revived at the end of a short block. It might save a few instrs.
  adjustLiveRegs(calcLiveInMask(MBB), MBB->begin());
  DEBUG(MBB->dump());
}

/// finishBlockStack - Revive live-outs that are implicitly defined out of
/// MBB. Shuffle live registers to match the expected fixed stack of any
/// predecessors, and ensure that all predecessors are expecting the same
/// stack.
void FPS::finishBlockStack() {
  // The RET handling below takes care of return blocks for us.
  if (MBB->succ_empty())
    return;

  DEBUG(dbgs() << "Setting up live-outs for BB#" << MBB->getNumber()
               << " derived from " << MBB->getName() << ".\n");

  // Get MBB's live-out bundle.
  unsigned BundleIdx = Bundles->getBundle(MBB->getNumber(), true);
  LiveBundle &Bundle = LiveBundles[BundleIdx];

  // We may need to kill and define some registers to match successors.
  // FIXME: This can probably be combined with the shuffle below.
  MachineBasicBlock::iterator Term = MBB->getFirstTerminator();
  adjustLiveRegs(Bundle.Mask, Term);

  if (!Bundle.Mask) {
    DEBUG(dbgs() << "No live-outs.\n");
    return;
  }

  // Has the stack order been fixed yet?
  DEBUG(dbgs() << "LB#" << BundleIdx << ": ");
  if (Bundle.isFixed()) {
    DEBUG(dbgs() << "Shuffling stack to match.\n");
    shuffleStackTop(Bundle.FixStack, Bundle.FixCount, Term);
  } else {
    // Not fixed yet, we get to choose.
    DEBUG(dbgs() << "Fixing stack order now.\n");
    Bundle.FixCount = StackTop;
    for (unsigned i = 0; i < StackTop; ++i)
      Bundle.FixStack[i] = getStackEntry(i);
  }
}


//===----------------------------------------------------------------------===//
// Efficient Lookup Table Support
//===----------------------------------------------------------------------===//

namespace {
  struct TableEntry {
    uint16_t from;
    uint16_t to;
    bool operator<(const TableEntry &TE) const { return from < TE.from; }
    friend bool operator<(const TableEntry &TE, unsigned V) {
      return TE.from < V;
    }
    friend bool LLVM_ATTRIBUTE_UNUSED operator<(unsigned V,
                                                const TableEntry &TE) {
      return V < TE.from;
    }
  };
}

static int Lookup(ArrayRef<TableEntry> Table, unsigned Opcode) {
  const TableEntry *I = std::lower_bound(Table.begin(), Table.end(), Opcode);
  if (I != Table.end() && I->from == Opcode)
    return I->to;
  return -1;
}

#ifdef NDEBUG
#define ASSERT_SORTED(TABLE)
#else
#define ASSERT_SORTED(TABLE)                                              \
  { static bool TABLE##Checked = false;                                   \
    if (!TABLE##Checked) {                                                \
       assert(std::is_sorted(std::begin(TABLE), std::end(TABLE)) &&       \
              "All lookup tables must be sorted for efficient access!");  \
       TABLE##Checked = true;                                             \
    }                                                                     \
  }
#endif

//===----------------------------------------------------------------------===//
// Register File -> Register Stack Mapping Methods
//===----------------------------------------------------------------------===//

// OpcodeTable - Sorted map of register instructions to their stack version.
// The first element is an register file pseudo instruction, the second is the
// concrete X86 instruction which uses the register stack.
//
static const TableEntry OpcodeTable[] = {
  { X86::ABS_Fp32     , X86::ABS_F     },
  { X86::ABS_Fp64     , X86::ABS_F     },
  { X86::ABS_Fp80     , X86::ABS_F     },
  { X86::ADD_Fp32m    , X86::ADD_F32m  },
  { X86::ADD_Fp64m    , X86::ADD_F64m  },
  { X86::ADD_Fp64m32  , X86::ADD_F32m  },
  { X86::ADD_Fp80m32  , X86::ADD_F32m  },
  { X86::ADD_Fp80m64  , X86::ADD_F64m  },
  { X86::ADD_FpI16m32 , X86::ADD_FI16m },
  { X86::ADD_FpI16m64 , X86::ADD_FI16m },
  { X86::ADD_FpI16m80 , X86::ADD_FI16m },
  { X86::ADD_FpI32m32 , X86::ADD_FI32m },
  { X86::ADD_FpI32m64 , X86::ADD_FI32m },
  { X86::ADD_FpI32m80 , X86::ADD_FI32m },
  { X86::CHS_Fp32     , X86::CHS_F     },
  { X86::CHS_Fp64     , X86::CHS_F     },
  { X86::CHS_Fp80     , X86::CHS_F     },
  { X86::CMOVBE_Fp32  , X86::CMOVBE_F  },
  { X86::CMOVBE_Fp64  , X86::CMOVBE_F  },
  { X86::CMOVBE_Fp80  , X86::CMOVBE_F  },
  { X86::CMOVB_Fp32   , X86::CMOVB_F   },
  { X86::CMOVB_Fp64   , X86::CMOVB_F  },
  { X86::CMOVB_Fp80   , X86::CMOVB_F  },
  { X86::CMOVE_Fp32   , X86::CMOVE_F  },
  { X86::CMOVE_Fp64   , X86::CMOVE_F   },
  { X86::CMOVE_Fp80   , X86::CMOVE_F   },
  { X86::CMOVNBE_Fp32 , X86::CMOVNBE_F },
  { X86::CMOVNBE_Fp64 , X86::CMOVNBE_F },
  { X86::CMOVNBE_Fp80 , X86::CMOVNBE_F },
  { X86::CMOVNB_Fp32  , X86::CMOVNB_F  },
  { X86::CMOVNB_Fp64  , X86::CMOVNB_F  },
  { X86::CMOVNB_Fp80  , X86::CMOVNB_F  },
  { X86::CMOVNE_Fp32  , X86::CMOVNE_F  },
  { X86::CMOVNE_Fp64  , X86::CMOVNE_F  },
  { X86::CMOVNE_Fp80  , X86::CMOVNE_F  },
  { X86::CMOVNP_Fp32  , X86::CMOVNP_F  },
  { X86::CMOVNP_Fp64  , X86::CMOVNP_F  },
  { X86::CMOVNP_Fp80  , X86::CMOVNP_F  },
  { X86::CMOVP_Fp32   , X86::CMOVP_F   },
  { X86::CMOVP_Fp64   , X86::CMOVP_F   },
  { X86::CMOVP_Fp80   , X86::CMOVP_F   },
  { X86::COS_Fp32     , X86::COS_F     },
  { X86::COS_Fp64     , X86::COS_F     },
  { X86::COS_Fp80     , X86::COS_F     },
  { X86::DIVR_Fp32m   , X86::DIVR_F32m },
  { X86::DIVR_Fp64m   , X86::DIVR_F64m },
  { X86::DIVR_Fp64m32 , X86::DIVR_F32m },
  { X86::DIVR_Fp80m32 , X86::DIVR_F32m },
  { X86::DIVR_Fp80m64 , X86::DIVR_F64m },
  { X86::DIVR_FpI16m32, X86::DIVR_FI16m},
  { X86::DIVR_FpI16m64, X86::DIVR_FI16m},
  { X86::DIVR_FpI16m80, X86::DIVR_FI16m},
  { X86::DIVR_FpI32m32, X86::DIVR_FI32m},
  { X86::DIVR_FpI32m64, X86::DIVR_FI32m},
  { X86::DIVR_FpI32m80, X86::DIVR_FI32m},
  { X86::DIV_Fp32m    , X86::DIV_F32m  },
  { X86::DIV_Fp64m    , X86::DIV_F64m  },
  { X86::DIV_Fp64m32  , X86::DIV_F32m  },
  { X86::DIV_Fp80m32  , X86::DIV_F32m  },
  { X86::DIV_Fp80m64  , X86::DIV_F64m  },
  { X86::DIV_FpI16m32 , X86::DIV_FI16m },
  { X86::DIV_FpI16m64 , X86::DIV_FI16m },
  { X86::DIV_FpI16m80 , X86::DIV_FI16m },
  { X86::DIV_FpI32m32 , X86::DIV_FI32m },
  { X86::DIV_FpI32m64 , X86::DIV_FI32m },
  { X86::DIV_FpI32m80 , X86::DIV_FI32m },
  { X86::ILD_Fp16m32  , X86::ILD_F16m  },
  { X86::ILD_Fp16m64  , X86::ILD_F16m  },
  { X86::ILD_Fp16m80  , X86::ILD_F16m  },
  { X86::ILD_Fp32m32  , X86::ILD_F32m  },
  { X86::ILD_Fp32m64  , X86::ILD_F32m  },
  { X86::ILD_Fp32m80  , X86::ILD_F32m  },
  { X86::ILD_Fp64m32  , X86::ILD_F64m  },
  { X86::ILD_Fp64m64  , X86::ILD_F64m  },
  { X86::ILD_Fp64m80  , X86::ILD_F64m  },
  { X86::ISTT_Fp16m32 , X86::ISTT_FP16m},
  { X86::ISTT_Fp16m64 , X86::ISTT_FP16m},
  { X86::ISTT_Fp16m80 , X86::ISTT_FP16m},
  { X86::ISTT_Fp32m32 , X86::ISTT_FP32m},
  { X86::ISTT_Fp32m64 , X86::ISTT_FP32m},
  { X86::ISTT_Fp32m80 , X86::ISTT_FP32m},
  { X86::ISTT_Fp64m32 , X86::ISTT_FP64m},
  { X86::ISTT_Fp64m64 , X86::ISTT_FP64m},
  { X86::ISTT_Fp64m80 , X86::ISTT_FP64m},
  { X86::IST_Fp16m32  , X86::IST_F16m  },
  { X86::IST_Fp16m64  , X86::IST_F16m  },
  { X86::IST_Fp16m80  , X86::IST_F16m  },
  { X86::IST_Fp32m32  , X86::IST_F32m  },
  { X86::IST_Fp32m64  , X86::IST_F32m  },
  { X86::IST_Fp32m80  , X86::IST_F32m  },
  { X86::IST_Fp64m32  , X86::IST_FP64m },
  { X86::IST_Fp64m64  , X86::IST_FP64m },
  { X86::IST_Fp64m80  , X86::IST_FP64m },
  { X86::LD_Fp032     , X86::LD_F0     },
  { X86::LD_Fp064     , X86::LD_F0     },
  { X86::LD_Fp080     , X86::LD_F0     },
  { X86::LD_Fp132     , X86::LD_F1     },
  { X86::LD_Fp164     , X86::LD_F1     },
  { X86::LD_Fp180     , X86::LD_F1     },
  { X86::LD_Fp32m     , X86::LD_F32m   },
  { X86::LD_Fp32m64   , X86::LD_F32m   },
  { X86::LD_Fp32m80   , X86::LD_F32m   },
  { X86::LD_Fp64m     , X86::LD_F64m   },
  { X86::LD_Fp64m80   , X86::LD_F64m   },
  { X86::LD_Fp80m     , X86::LD_F80m   },
  { X86::MUL_Fp32m    , X86::MUL_F32m  },
  { X86::MUL_Fp64m    , X86::MUL_F64m  },
  { X86::MUL_Fp64m32  , X86::MUL_F32m  },
  { X86::MUL_Fp80m32  , X86::MUL_F32m  },
  { X86::MUL_Fp80m64  , X86::MUL_F64m  },
  { X86::MUL_FpI16m32 , X86::MUL_FI16m },
  { X86::MUL_FpI16m64 , X86::MUL_FI16m },
  { X86::MUL_FpI16m80 , X86::MUL_FI16m },
  { X86::MUL_FpI32m32 , X86::MUL_FI32m },
  { X86::MUL_FpI32m64 , X86::MUL_FI32m },
  { X86::MUL_FpI32m80 , X86::MUL_FI32m },
  { X86::SIN_Fp32     , X86::SIN_F     },
  { X86::SIN_Fp64     , X86::SIN_F     },
  { X86::SIN_Fp80     , X86::SIN_F     },
  { X86::SQRT_Fp32    , X86::SQRT_F    },
  { X86::SQRT_Fp64    , X86::SQRT_F    },
  { X86::SQRT_Fp80    , X86::SQRT_F    },
  { X86::ST_Fp32m     , X86::ST_F32m   },
  { X86::ST_Fp64m     , X86::ST_F64m   },
  { X86::ST_Fp64m32   , X86::ST_F32m   },
  { X86::ST_Fp80m32   , X86::ST_F32m   },
  { X86::ST_Fp80m64   , X86::ST_F64m   },
  { X86::ST_FpP80m    , X86::ST_FP80m  },
  { X86::SUBR_Fp32m   , X86::SUBR_F32m },
  { X86::SUBR_Fp64m   , X86::SUBR_F64m },
  { X86::SUBR_Fp64m32 , X86::SUBR_F32m },
  { X86::SUBR_Fp80m32 , X86::SUBR_F32m },
  { X86::SUBR_Fp80m64 , X86::SUBR_F64m },
  { X86::SUBR_FpI16m32, X86::SUBR_FI16m},
  { X86::SUBR_FpI16m64, X86::SUBR_FI16m},
  { X86::SUBR_FpI16m80, X86::SUBR_FI16m},
  { X86::SUBR_FpI32m32, X86::SUBR_FI32m},
  { X86::SUBR_FpI32m64, X86::SUBR_FI32m},
  { X86::SUBR_FpI32m80, X86::SUBR_FI32m},
  { X86::SUB_Fp32m    , X86::SUB_F32m  },
  { X86::SUB_Fp64m    , X86::SUB_F64m  },
  { X86::SUB_Fp64m32  , X86::SUB_F32m  },
  { X86::SUB_Fp80m32  , X86::SUB_F32m  },
  { X86::SUB_Fp80m64  , X86::SUB_F64m  },
  { X86::SUB_FpI16m32 , X86::SUB_FI16m },
  { X86::SUB_FpI16m64 , X86::SUB_FI16m },
  { X86::SUB_FpI16m80 , X86::SUB_FI16m },
  { X86::SUB_FpI32m32 , X86::SUB_FI32m },
  { X86::SUB_FpI32m64 , X86::SUB_FI32m },
  { X86::SUB_FpI32m80 , X86::SUB_FI32m },
  { X86::TST_Fp32     , X86::TST_F     },
  { X86::TST_Fp64     , X86::TST_F     },
  { X86::TST_Fp80     , X86::TST_F     },
  { X86::UCOM_FpIr32  , X86::UCOM_FIr  },
  { X86::UCOM_FpIr64  , X86::UCOM_FIr  },
  { X86::UCOM_FpIr80  , X86::UCOM_FIr  },
  { X86::UCOM_Fpr32   , X86::UCOM_Fr   },
  { X86::UCOM_Fpr64   , X86::UCOM_Fr   },
  { X86::UCOM_Fpr80   , X86::UCOM_Fr   },
};

static unsigned getConcreteOpcode(unsigned Opcode) {
  ASSERT_SORTED(OpcodeTable);
  int Opc = Lookup(OpcodeTable, Opcode);
  assert(Opc != -1 && "FP Stack instruction not in OpcodeTable!");
  return Opc;
}

//===----------------------------------------------------------------------===//
// Helper Methods
//===----------------------------------------------------------------------===//

// PopTable - Sorted map of instructions to their popping version.  The first
// element is an instruction, the second is the version which pops.
//
static const TableEntry PopTable[] = {
  { X86::ADD_FrST0 , X86::ADD_FPrST0  },

  { X86::DIVR_FrST0, X86::DIVR_FPrST0 },
  { X86::DIV_FrST0 , X86::DIV_FPrST0  },

  { X86::IST_F16m  , X86::IST_FP16m   },
  { X86::IST_F32m  , X86::IST_FP32m   },

  { X86::MUL_FrST0 , X86::MUL_FPrST0  },

  { X86::ST_F32m   , X86::ST_FP32m    },
  { X86::ST_F64m   , X86::ST_FP64m    },
  { X86::ST_Frr    , X86::ST_FPrr     },

  { X86::SUBR_FrST0, X86::SUBR_FPrST0 },
  { X86::SUB_FrST0 , X86::SUB_FPrST0  },

  { X86::UCOM_FIr  , X86::UCOM_FIPr   },

  { X86::UCOM_FPr  , X86::UCOM_FPPr   },
  { X86::UCOM_Fr   , X86::UCOM_FPr    },
};

/// popStackAfter - Pop the current value off of the top of the FP stack after
/// the specified instruction.  This attempts to be sneaky and combine the pop
/// into the instruction itself if possible.  The iterator is left pointing to
/// the last instruction, be it a new pop instruction inserted, or the old
/// instruction if it was modified in place.
///
void FPS::popStackAfter(MachineBasicBlock::iterator &I) {
  MachineInstr &MI = *I;
  const DebugLoc &dl = MI.getDebugLoc();
  ASSERT_SORTED(PopTable);

  popReg();

  // Check to see if there is a popping version of this instruction...
  int Opcode = Lookup(PopTable, I->getOpcode());
  if (Opcode != -1) {
    I->setDesc(TII->get(Opcode));
    if (Opcode == X86::UCOM_FPPr)
      I->RemoveOperand(0);
  } else {    // Insert an explicit pop
    I = BuildMI(*MBB, ++I, dl, TII->get(X86::ST_FPrr)).addReg(X86::ST0);
  }
}

/// freeStackSlotAfter - Free the specified register from the register stack, so
/// that it is no longer in a register.  If the register is currently at the top
/// of the stack, we just pop the current instruction, otherwise we store the
/// current top-of-stack into the specified slot, then pop the top of stack.
void FPS::freeStackSlotAfter(MachineBasicBlock::iterator &I, unsigned FPRegNo) {
  if (getStackEntry(0) == FPRegNo) {  // already at the top of stack? easy.
    popStackAfter(I);
    return;
  }

  // Otherwise, store the top of stack into the dead slot, killing the operand
  // without having to add in an explicit xchg then pop.
  //
  I = freeStackSlotBefore(++I, FPRegNo);
}

/// freeStackSlotBefore - Free the specified register without trying any
/// folding.
MachineBasicBlock::iterator
FPS::freeStackSlotBefore(MachineBasicBlock::iterator I, unsigned FPRegNo) {
  unsigned STReg    = getSTReg(FPRegNo);
  unsigned OldSlot  = getSlot(FPRegNo);
  unsigned TopReg   = Stack[StackTop-1];
  Stack[OldSlot]    = TopReg;
  RegMap[TopReg]    = OldSlot;
  RegMap[FPRegNo]   = ~0;
  Stack[--StackTop] = ~0;
  return BuildMI(*MBB, I, DebugLoc(), TII->get(X86::ST_FPrr))
      .addReg(STReg)
      .getInstr();
}

/// adjustLiveRegs - Kill and revive registers such that exactly the FP
/// registers with a bit in Mask are live.
void FPS::adjustLiveRegs(unsigned Mask, MachineBasicBlock::iterator I) {
  unsigned Defs = Mask;
  unsigned Kills = 0;
  for (unsigned i = 0; i < StackTop; ++i) {
    unsigned RegNo = Stack[i];
    if (!(Defs & (1 << RegNo)))
      // This register is live, but we don't want it.
      Kills |= (1 << RegNo);
    else
      // We don't need to imp-def this live register.
      Defs &= ~(1 << RegNo);
  }
  assert((Kills & Defs) == 0 && "Register needs killing and def'ing?");

  // Produce implicit-defs for free by using killed registers.
  while (Kills && Defs) {
    unsigned KReg = countTrailingZeros(Kills);
    unsigned DReg = countTrailingZeros(Defs);
    DEBUG(dbgs() << "Renaming %FP" << KReg << " as imp %FP" << DReg << "\n");
    std::swap(Stack[getSlot(KReg)], Stack[getSlot(DReg)]);
    std::swap(RegMap[KReg], RegMap[DReg]);
    Kills &= ~(1 << KReg);
    Defs &= ~(1 << DReg);
  }

  // Kill registers by popping.
  if (Kills && I != MBB->begin()) {
    MachineBasicBlock::iterator I2 = std::prev(I);
    while (StackTop) {
      unsigned KReg = getStackEntry(0);
      if (!(Kills & (1 << KReg)))
        break;
      DEBUG(dbgs() << "Popping %FP" << KReg << "\n");
      popStackAfter(I2);
      Kills &= ~(1 << KReg);
    }
  }

  // Manually kill the rest.
  while (Kills) {
    unsigned KReg = countTrailingZeros(Kills);
    DEBUG(dbgs() << "Killing %FP" << KReg << "\n");
    freeStackSlotBefore(I, KReg);
    Kills &= ~(1 << KReg);
  }

  // Load zeros for all the imp-defs.
  while(Defs) {
    unsigned DReg = countTrailingZeros(Defs);
    DEBUG(dbgs() << "Defining %FP" << DReg << " as 0\n");
    BuildMI(*MBB, I, DebugLoc(), TII->get(X86::LD_F0));
    pushReg(DReg);
    Defs &= ~(1 << DReg);
  }

  // Now we should have the correct registers live.
  DEBUG(dumpStack());
  assert(StackTop == countPopulation(Mask) && "Live count mismatch");
}

/// shuffleStackTop - emit fxch instructions before I to shuffle the top
/// FixCount entries into the order given by FixStack.
/// FIXME: Is there a better algorithm than insertion sort?
void FPS::shuffleStackTop(const unsigned char *FixStack,
                          unsigned FixCount,
                          MachineBasicBlock::iterator I) {
  // Move items into place, starting from the desired stack bottom.
  while (FixCount--) {
    // Old register at position FixCount.
    unsigned OldReg = getStackEntry(FixCount);
    // Desired register at position FixCount.
    unsigned Reg = FixStack[FixCount];
    if (Reg == OldReg)
      continue;
    // (Reg st0) (OldReg st0) = (Reg OldReg st0)
    moveToTop(Reg, I);
    if (FixCount > 0)
      moveToTop(OldReg, I);
  }
  DEBUG(dumpStack());
}


//===----------------------------------------------------------------------===//
// Instruction transformation implementation
//===----------------------------------------------------------------------===//

void FPS::handleCall(MachineBasicBlock::iterator &I) {
  unsigned STReturns = 0;
  const MachineFunction* MF = I->getParent()->getParent();

  for (const auto &MO : I->operands()) {
    if (!MO.isReg())
      continue;

    unsigned R = MO.getReg() - X86::FP0;

    if (R < 8) {
      if (MF->getFunction()->getCallingConv() != CallingConv::X86_RegCall) {
        assert(MO.isDef() && MO.isImplicit());
      }

      STReturns |= 1 << R;
    }
  }

  unsigned N = countTrailingOnes(STReturns);

  // FP registers used for function return must be consecutive starting at
  // FP0
  assert(STReturns == 0 || (isMask_32(STReturns) && N <= 2));

  // Reset the FP Stack - It is required because of possible leftovers from
  // passed arguments. The caller should assume that the FP stack is 
  // returned empty (unless the callee returns values on FP stack).
  while (StackTop > 0)
    popReg();

  for (unsigned I = 0; I < N; ++I)
    pushReg(N - I - 1);
}

/// If RET has an FP register use operand, pass the first one in ST(0) and
/// the second one in ST(1).
void FPS::handleReturn(MachineBasicBlock::iterator &I) {
  MachineInstr &MI = *I;

  // Find the register operands.
  unsigned FirstFPRegOp = ~0U, SecondFPRegOp = ~0U;
  unsigned LiveMask = 0;

  for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
    MachineOperand &Op = MI.getOperand(i);
    if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6)
      continue;
    // FP Register uses must be kills unless there are two uses of the same
    // register, in which case only one will be a kill.
    assert(Op.isUse() &&
           (Op.isKill() ||                    // Marked kill.
            getFPReg(Op) == FirstFPRegOp ||   // Second instance.
            MI.killsRegister(Op.getReg())) && // Later use is marked kill.
           "Ret only defs operands, and values aren't live beyond it");

    if (FirstFPRegOp == ~0U)
      FirstFPRegOp = getFPReg(Op);
    else {
      assert(SecondFPRegOp == ~0U && "More than two fp operands!");
      SecondFPRegOp = getFPReg(Op);
    }
    LiveMask |= (1 << getFPReg(Op));

    // Remove the operand so that later passes don't see it.
    MI.RemoveOperand(i);
    --i;
    --e;
  }

  // We may have been carrying spurious live-ins, so make sure only the
  // returned registers are left live.
  adjustLiveRegs(LiveMask, MI);
  if (!LiveMask) return;  // Quick check to see if any are possible.

  // There are only four possibilities here:
  // 1) we are returning a single FP value.  In this case, it has to be in
  //    ST(0) already, so just declare success by removing the value from the
  //    FP Stack.
  if (SecondFPRegOp == ~0U) {
    // Assert that the top of stack contains the right FP register.
    assert(StackTop == 1 && FirstFPRegOp == getStackEntry(0) &&
           "Top of stack not the right register for RET!");

    // Ok, everything is good, mark the value as not being on the stack
    // anymore so that our assertion about the stack being empty at end of
    // block doesn't fire.
    StackTop = 0;
    return;
  }

  // Otherwise, we are returning two values:
  // 2) If returning the same value for both, we only have one thing in the FP
  //    stack.  Consider:  RET FP1, FP1
  if (StackTop == 1) {
    assert(FirstFPRegOp == SecondFPRegOp && FirstFPRegOp == getStackEntry(0)&&
           "Stack misconfiguration for RET!");

    // Duplicate the TOS so that we return it twice.  Just pick some other FPx
    // register to hold it.
    unsigned NewReg = ScratchFPReg;
    duplicateToTop(FirstFPRegOp, NewReg, MI);
    FirstFPRegOp = NewReg;
  }

  /// Okay we know we have two different FPx operands now:
  assert(StackTop == 2 && "Must have two values live!");

  /// 3) If SecondFPRegOp is currently in ST(0) and FirstFPRegOp is currently
  ///    in ST(1).  In this case, emit an fxch.
  if (getStackEntry(0) == SecondFPRegOp) {
    assert(getStackEntry(1) == FirstFPRegOp && "Unknown regs live");
    moveToTop(FirstFPRegOp, MI);
  }

  /// 4) Finally, FirstFPRegOp must be in ST(0) and SecondFPRegOp must be in
  /// ST(1).  Just remove both from our understanding of the stack and return.
  assert(getStackEntry(0) == FirstFPRegOp && "Unknown regs live");
  assert(getStackEntry(1) == SecondFPRegOp && "Unknown regs live");
  StackTop = 0;
}

/// handleZeroArgFP - ST(0) = fld0    ST(0) = flds <mem>
///
void FPS::handleZeroArgFP(MachineBasicBlock::iterator &I) {
  MachineInstr &MI = *I;
  unsigned DestReg = getFPReg(MI.getOperand(0));

  // Change from the pseudo instruction to the concrete instruction.
  MI.RemoveOperand(0); // Remove the explicit ST(0) operand
  MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode())));

  // Result gets pushed on the stack.
  pushReg(DestReg);
}

/// handleOneArgFP - fst <mem>, ST(0)
///
void FPS::handleOneArgFP(MachineBasicBlock::iterator &I) {
  MachineInstr &MI = *I;
  unsigned NumOps = MI.getDesc().getNumOperands();
  assert((NumOps == X86::AddrNumOperands + 1 || NumOps == 1) &&
         "Can only handle fst* & ftst instructions!");

  // Is this the last use of the source register?
  unsigned Reg = getFPReg(MI.getOperand(NumOps - 1));
  bool KillsSrc = MI.killsRegister(X86::FP0 + Reg);

  // FISTP64m is strange because there isn't a non-popping versions.
  // If we have one _and_ we don't want to pop the operand, duplicate the value
  // on the stack instead of moving it.  This ensure that popping the value is
  // always ok.
  // Ditto FISTTP16m, FISTTP32m, FISTTP64m, ST_FpP80m.
  //
  if (!KillsSrc && (MI.getOpcode() == X86::IST_Fp64m32 ||
                    MI.getOpcode() == X86::ISTT_Fp16m32 ||
                    MI.getOpcode() == X86::ISTT_Fp32m32 ||
                    MI.getOpcode() == X86::ISTT_Fp64m32 ||
                    MI.getOpcode() == X86::IST_Fp64m64 ||
                    MI.getOpcode() == X86::ISTT_Fp16m64 ||
                    MI.getOpcode() == X86::ISTT_Fp32m64 ||
                    MI.getOpcode() == X86::ISTT_Fp64m64 ||
                    MI.getOpcode() == X86::IST_Fp64m80 ||
                    MI.getOpcode() == X86::ISTT_Fp16m80 ||
                    MI.getOpcode() == X86::ISTT_Fp32m80 ||
                    MI.getOpcode() == X86::ISTT_Fp64m80 ||
                    MI.getOpcode() == X86::ST_FpP80m)) {
    duplicateToTop(Reg, ScratchFPReg, I);
  } else {
    moveToTop(Reg, I);            // Move to the top of the stack...
  }

  // Convert from the pseudo instruction to the concrete instruction.
  MI.RemoveOperand(NumOps - 1); // Remove explicit ST(0) operand
  MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode())));

  if (MI.getOpcode() == X86::IST_FP64m || MI.getOpcode() == X86::ISTT_FP16m ||
      MI.getOpcode() == X86::ISTT_FP32m || MI.getOpcode() == X86::ISTT_FP64m ||
      MI.getOpcode() == X86::ST_FP80m) {
    if (StackTop == 0)
      report_fatal_error("Stack empty??");
    --StackTop;
  } else if (KillsSrc) { // Last use of operand?
    popStackAfter(I);
  }
}


/// handleOneArgFPRW: Handle instructions that read from the top of stack and
/// replace the value with a newly computed value.  These instructions may have
/// non-fp operands after their FP operands.
///
///  Examples:
///     R1 = fchs R2
///     R1 = fadd R2, [mem]
///
void FPS::handleOneArgFPRW(MachineBasicBlock::iterator &I) {
  MachineInstr &MI = *I;
#ifndef NDEBUG
  unsigned NumOps = MI.getDesc().getNumOperands();
  assert(NumOps >= 2 && "FPRW instructions must have 2 ops!!");
#endif

  // Is this the last use of the source register?
  unsigned Reg = getFPReg(MI.getOperand(1));
  bool KillsSrc = MI.killsRegister(X86::FP0 + Reg);

  if (KillsSrc) {
    // If this is the last use of the source register, just make sure it's on
    // the top of the stack.
    moveToTop(Reg, I);
    if (StackTop == 0)
      report_fatal_error("Stack cannot be empty!");
    --StackTop;
    pushReg(getFPReg(MI.getOperand(0)));
  } else {
    // If this is not the last use of the source register, _copy_ it to the top
    // of the stack.
    duplicateToTop(Reg, getFPReg(MI.getOperand(0)), I);
  }

  // Change from the pseudo instruction to the concrete instruction.
  MI.RemoveOperand(1); // Drop the source operand.
  MI.RemoveOperand(0); // Drop the destination operand.
  MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode())));
}


//===----------------------------------------------------------------------===//
// Define tables of various ways to map pseudo instructions
//

// ForwardST0Table - Map: A = B op C  into: ST(0) = ST(0) op ST(i)
static const TableEntry ForwardST0Table[] = {
  { X86::ADD_Fp32  , X86::ADD_FST0r },
  { X86::ADD_Fp64  , X86::ADD_FST0r },
  { X86::ADD_Fp80  , X86::ADD_FST0r },
  { X86::DIV_Fp32  , X86::DIV_FST0r },
  { X86::DIV_Fp64  , X86::DIV_FST0r },
  { X86::DIV_Fp80  , X86::DIV_FST0r },
  { X86::MUL_Fp32  , X86::MUL_FST0r },
  { X86::MUL_Fp64  , X86::MUL_FST0r },
  { X86::MUL_Fp80  , X86::MUL_FST0r },
  { X86::SUB_Fp32  , X86::SUB_FST0r },
  { X86::SUB_Fp64  , X86::SUB_FST0r },
  { X86::SUB_Fp80  , X86::SUB_FST0r },
};

// ReverseST0Table - Map: A = B op C  into: ST(0) = ST(i) op ST(0)
static const TableEntry ReverseST0Table[] = {
  { X86::ADD_Fp32  , X86::ADD_FST0r  },   // commutative
  { X86::ADD_Fp64  , X86::ADD_FST0r  },   // commutative
  { X86::ADD_Fp80  , X86::ADD_FST0r  },   // commutative
  { X86::DIV_Fp32  , X86::DIVR_FST0r },
  { X86::DIV_Fp64  , X86::DIVR_FST0r },
  { X86::DIV_Fp80  , X86::DIVR_FST0r },
  { X86::MUL_Fp32  , X86::MUL_FST0r  },   // commutative
  { X86::MUL_Fp64  , X86::MUL_FST0r  },   // commutative
  { X86::MUL_Fp80  , X86::MUL_FST0r  },   // commutative
  { X86::SUB_Fp32  , X86::SUBR_FST0r },
  { X86::SUB_Fp64  , X86::SUBR_FST0r },
  { X86::SUB_Fp80  , X86::SUBR_FST0r },
};

// ForwardSTiTable - Map: A = B op C  into: ST(i) = ST(0) op ST(i)
static const TableEntry ForwardSTiTable[] = {
  { X86::ADD_Fp32  , X86::ADD_FrST0  },   // commutative
  { X86::ADD_Fp64  , X86::ADD_FrST0  },   // commutative
  { X86::ADD_Fp80  , X86::ADD_FrST0  },   // commutative
  { X86::DIV_Fp32  , X86::DIVR_FrST0 },
  { X86::DIV_Fp64  , X86::DIVR_FrST0 },
  { X86::DIV_Fp80  , X86::DIVR_FrST0 },
  { X86::MUL_Fp32  , X86::MUL_FrST0  },   // commutative
  { X86::MUL_Fp64  , X86::MUL_FrST0  },   // commutative
  { X86::MUL_Fp80  , X86::MUL_FrST0  },   // commutative
  { X86::SUB_Fp32  , X86::SUBR_FrST0 },
  { X86::SUB_Fp64  , X86::SUBR_FrST0 },
  { X86::SUB_Fp80  , X86::SUBR_FrST0 },
};

// ReverseSTiTable - Map: A = B op C  into: ST(i) = ST(i) op ST(0)
static const TableEntry ReverseSTiTable[] = {
  { X86::ADD_Fp32  , X86::ADD_FrST0 },
  { X86::ADD_Fp64  , X86::ADD_FrST0 },
  { X86::ADD_Fp80  , X86::ADD_FrST0 },
  { X86::DIV_Fp32  , X86::DIV_FrST0 },
  { X86::DIV_Fp64  , X86::DIV_FrST0 },
  { X86::DIV_Fp80  , X86::DIV_FrST0 },
  { X86::MUL_Fp32  , X86::MUL_FrST0 },
  { X86::MUL_Fp64  , X86::MUL_FrST0 },
  { X86::MUL_Fp80  , X86::MUL_FrST0 },
  { X86::SUB_Fp32  , X86::SUB_FrST0 },
  { X86::SUB_Fp64  , X86::SUB_FrST0 },
  { X86::SUB_Fp80  , X86::SUB_FrST0 },
};


/// handleTwoArgFP - Handle instructions like FADD and friends which are virtual
/// instructions which need to be simplified and possibly transformed.
///
/// Result: ST(0) = fsub  ST(0), ST(i)
///         ST(i) = fsub  ST(0), ST(i)
///         ST(0) = fsubr ST(0), ST(i)
///         ST(i) = fsubr ST(0), ST(i)
///
void FPS::handleTwoArgFP(MachineBasicBlock::iterator &I) {
  ASSERT_SORTED(ForwardST0Table); ASSERT_SORTED(ReverseST0Table);
  ASSERT_SORTED(ForwardSTiTable); ASSERT_SORTED(ReverseSTiTable);
  MachineInstr &MI = *I;

  unsigned NumOperands = MI.getDesc().getNumOperands();
  assert(NumOperands == 3 && "Illegal TwoArgFP instruction!");
  unsigned Dest = getFPReg(MI.getOperand(0));
  unsigned Op0 = getFPReg(MI.getOperand(NumOperands - 2));
  unsigned Op1 = getFPReg(MI.getOperand(NumOperands - 1));
  bool KillsOp0 = MI.killsRegister(X86::FP0 + Op0);
  bool KillsOp1 = MI.killsRegister(X86::FP0 + Op1);
  DebugLoc dl = MI.getDebugLoc();

  unsigned TOS = getStackEntry(0);

  // One of our operands must be on the top of the stack.  If neither is yet, we
  // need to move one.
  if (Op0 != TOS && Op1 != TOS) {   // No operand at TOS?
    // We can choose to move either operand to the top of the stack.  If one of
    // the operands is killed by this instruction, we want that one so that we
    // can update right on top of the old version.
    if (KillsOp0) {
      moveToTop(Op0, I);         // Move dead operand to TOS.
      TOS = Op0;
    } else if (KillsOp1) {
      moveToTop(Op1, I);
      TOS = Op1;
    } else {
      // All of the operands are live after this instruction executes, so we
      // cannot update on top of any operand.  Because of this, we must
      // duplicate one of the stack elements to the top.  It doesn't matter
      // which one we pick.
      //
      duplicateToTop(Op0, Dest, I);
      Op0 = TOS = Dest;
      KillsOp0 = true;
    }
  } else if (!KillsOp0 && !KillsOp1) {
    // If we DO have one of our operands at the top of the stack, but we don't
    // have a dead operand, we must duplicate one of the operands to a new slot
    // on the stack.
    duplicateToTop(Op0, Dest, I);
    Op0 = TOS = Dest;
    KillsOp0 = true;
  }

  // Now we know that one of our operands is on the top of the stack, and at
  // least one of our operands is killed by this instruction.
  assert((TOS == Op0 || TOS == Op1) && (KillsOp0 || KillsOp1) &&
         "Stack conditions not set up right!");

  // We decide which form to use based on what is on the top of the stack, and
  // which operand is killed by this instruction.
  ArrayRef<TableEntry> InstTable;
  bool isForward = TOS == Op0;
  bool updateST0 = (TOS == Op0 && !KillsOp1) || (TOS == Op1 && !KillsOp0);
  if (updateST0) {
    if (isForward)
      InstTable = ForwardST0Table;
    else
      InstTable = ReverseST0Table;
  } else {
    if (isForward)
      InstTable = ForwardSTiTable;
    else
      InstTable = ReverseSTiTable;
  }

  int Opcode = Lookup(InstTable, MI.getOpcode());
  assert(Opcode != -1 && "Unknown TwoArgFP pseudo instruction!");

  // NotTOS - The register which is not on the top of stack...
  unsigned NotTOS = (TOS == Op0) ? Op1 : Op0;

  // Replace the old instruction with a new instruction
  MBB->remove(&*I++);
  I = BuildMI(*MBB, I, dl, TII->get(Opcode)).addReg(getSTReg(NotTOS));

  // If both operands are killed, pop one off of the stack in addition to
  // overwriting the other one.
  if (KillsOp0 && KillsOp1 && Op0 != Op1) {
    assert(!updateST0 && "Should have updated other operand!");
    popStackAfter(I);   // Pop the top of stack
  }

  // Update stack information so that we know the destination register is now on
  // the stack.
  unsigned UpdatedSlot = getSlot(updateST0 ? TOS : NotTOS);
  assert(UpdatedSlot < StackTop && Dest < 7);
  Stack[UpdatedSlot]   = Dest;
  RegMap[Dest]         = UpdatedSlot;
  MBB->getParent()->DeleteMachineInstr(&MI); // Remove the old instruction
}

/// handleCompareFP - Handle FUCOM and FUCOMI instructions, which have two FP
/// register arguments and no explicit destinations.
///
void FPS::handleCompareFP(MachineBasicBlock::iterator &I) {
  ASSERT_SORTED(ForwardST0Table); ASSERT_SORTED(ReverseST0Table);
  ASSERT_SORTED(ForwardSTiTable); ASSERT_SORTED(ReverseSTiTable);
  MachineInstr &MI = *I;

  unsigned NumOperands = MI.getDesc().getNumOperands();
  assert(NumOperands == 2 && "Illegal FUCOM* instruction!");
  unsigned Op0 = getFPReg(MI.getOperand(NumOperands - 2));
  unsigned Op1 = getFPReg(MI.getOperand(NumOperands - 1));
  bool KillsOp0 = MI.killsRegister(X86::FP0 + Op0);
  bool KillsOp1 = MI.killsRegister(X86::FP0 + Op1);

  // Make sure the first operand is on the top of stack, the other one can be
  // anywhere.
  moveToTop(Op0, I);

  // Change from the pseudo instruction to the concrete instruction.
  MI.getOperand(0).setReg(getSTReg(Op1));
  MI.RemoveOperand(1);
  MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode())));

  // If any of the operands are killed by this instruction, free them.
  if (KillsOp0) freeStackSlotAfter(I, Op0);
  if (KillsOp1 && Op0 != Op1) freeStackSlotAfter(I, Op1);
}

/// handleCondMovFP - Handle two address conditional move instructions.  These
/// instructions move a st(i) register to st(0) iff a condition is true.  These
/// instructions require that the first operand is at the top of the stack, but
/// otherwise don't modify the stack at all.
void FPS::handleCondMovFP(MachineBasicBlock::iterator &I) {
  MachineInstr &MI = *I;

  unsigned Op0 = getFPReg(MI.getOperand(0));
  unsigned Op1 = getFPReg(MI.getOperand(2));
  bool KillsOp1 = MI.killsRegister(X86::FP0 + Op1);

  // The first operand *must* be on the top of the stack.
  moveToTop(Op0, I);

  // Change the second operand to the stack register that the operand is in.
  // Change from the pseudo instruction to the concrete instruction.
  MI.RemoveOperand(0);
  MI.RemoveOperand(1);
  MI.getOperand(0).setReg(getSTReg(Op1));
  MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode())));

  // If we kill the second operand, make sure to pop it from the stack.
  if (Op0 != Op1 && KillsOp1) {
    // Get this value off of the register stack.
    freeStackSlotAfter(I, Op1);
  }
}


/// handleSpecialFP - Handle special instructions which behave unlike other
/// floating point instructions.  This is primarily intended for use by pseudo
/// instructions.
///
void FPS::handleSpecialFP(MachineBasicBlock::iterator &Inst) {
  MachineInstr &MI = *Inst;

  if (MI.isCall()) {
    handleCall(Inst);
    return;
  }

  if (MI.isReturn()) {
    handleReturn(Inst);
    return;
  }

  switch (MI.getOpcode()) {
  default: llvm_unreachable("Unknown SpecialFP instruction!");
  case TargetOpcode::COPY: {
    // We handle three kinds of copies: FP <- FP, FP <- ST, and ST <- FP.
    const MachineOperand &MO1 = MI.getOperand(1);
    const MachineOperand &MO0 = MI.getOperand(0);
    bool KillsSrc = MI.killsRegister(MO1.getReg());

    // FP <- FP copy.
    unsigned DstFP = getFPReg(MO0);
    unsigned SrcFP = getFPReg(MO1);
    assert(isLive(SrcFP) && "Cannot copy dead register");
    if (KillsSrc) {
      // If the input operand is killed, we can just change the owner of the
      // incoming stack slot into the result.
      unsigned Slot = getSlot(SrcFP);
      Stack[Slot] = DstFP;
      RegMap[DstFP] = Slot;
    } else {
      // For COPY we just duplicate the specified value to a new stack slot.
      // This could be made better, but would require substantial changes.
      duplicateToTop(SrcFP, DstFP, Inst);
    }
    break;
  }

  case TargetOpcode::IMPLICIT_DEF: {
    // All FP registers must be explicitly defined, so load a 0 instead.
    unsigned Reg = MI.getOperand(0).getReg() - X86::FP0;
    DEBUG(dbgs() << "Emitting LD_F0 for implicit FP" << Reg << '\n');
    BuildMI(*MBB, Inst, MI.getDebugLoc(), TII->get(X86::LD_F0));
    pushReg(Reg);
    break;
  }

  case TargetOpcode::INLINEASM: {
    // The inline asm MachineInstr currently only *uses* FP registers for the
    // 'f' constraint.  These should be turned into the current ST(x) register
    // in the machine instr.
    //
    // There are special rules for x87 inline assembly. The compiler must know
    // exactly how many registers are popped and pushed implicitly by the asm.
    // Otherwise it is not possible to restore the stack state after the inline
    // asm.
    //
    // There are 3 kinds of input operands:
    //
    // 1. Popped inputs. These must appear at the stack top in ST0-STn. A
    //    popped input operand must be in a fixed stack slot, and it is either
    //    tied to an output operand, or in the clobber list. The MI has ST use
    //    and def operands for these inputs.
    //
    // 2. Fixed inputs. These inputs appear in fixed stack slots, but are
    //    preserved by the inline asm. The fixed stack slots must be STn-STm
    //    following the popped inputs. A fixed input operand cannot be tied to
    //    an output or appear in the clobber list. The MI has ST use operands
    //    and no defs for these inputs.
    //
    // 3. Preserved inputs. These inputs use the "f" constraint which is
    //    represented as an FP register. The inline asm won't change these
    //    stack slots.
    //
    // Outputs must be in ST registers, FP outputs are not allowed. Clobbered
    // registers do not count as output operands. The inline asm changes the
    // stack as if it popped all the popped inputs and then pushed all the
    // output operands.

    // Scan the assembly for ST registers used, defined and clobbered. We can
    // only tell clobbers from defs by looking at the asm descriptor.
    unsigned STUses = 0, STDefs = 0, STClobbers = 0, STDeadDefs = 0;
    unsigned NumOps = 0;
    SmallSet<unsigned, 1> FRegIdx;
    unsigned RCID;

    for (unsigned i = InlineAsm::MIOp_FirstOperand, e = MI.getNumOperands();
         i != e && MI.getOperand(i).isImm(); i += 1 + NumOps) {
      unsigned Flags = MI.getOperand(i).getImm();

      NumOps = InlineAsm::getNumOperandRegisters(Flags);
      if (NumOps != 1)
        continue;
      const MachineOperand &MO = MI.getOperand(i + 1);
      if (!MO.isReg())
        continue;
      unsigned STReg = MO.getReg() - X86::FP0;
      if (STReg >= 8)
        continue;

      // If the flag has a register class constraint, this must be an operand
      // with constraint "f". Record its index and continue.
      if (InlineAsm::hasRegClassConstraint(Flags, RCID)) {
        FRegIdx.insert(i + 1);
        continue;
      }

      switch (InlineAsm::getKind(Flags)) {
      case InlineAsm::Kind_RegUse:
        STUses |= (1u << STReg);
        break;
      case InlineAsm::Kind_RegDef:
      case InlineAsm::Kind_RegDefEarlyClobber:
        STDefs |= (1u << STReg);
        if (MO.isDead())
          STDeadDefs |= (1u << STReg);
        break;
      case InlineAsm::Kind_Clobber:
        STClobbers |= (1u << STReg);
        break;
      default:
        break;
      }
    }

    if (STUses && !isMask_32(STUses))
      MI.emitError("fixed input regs must be last on the x87 stack");
    unsigned NumSTUses = countTrailingOnes(STUses);

    // Defs must be contiguous from the stack top. ST0-STn.
    if (STDefs && !isMask_32(STDefs)) {
      MI.emitError("output regs must be last on the x87 stack");
      STDefs = NextPowerOf2(STDefs) - 1;
    }
    unsigned NumSTDefs = countTrailingOnes(STDefs);

    // So must the clobbered stack slots. ST0-STm, m >= n.
    if (STClobbers && !isMask_32(STDefs | STClobbers))
      MI.emitError("clobbers must be last on the x87 stack");

    // Popped inputs are the ones that are also clobbered or defined.
    unsigned STPopped = STUses & (STDefs | STClobbers);
    if (STPopped && !isMask_32(STPopped))
      MI.emitError("implicitly popped regs must be last on the x87 stack");
    unsigned NumSTPopped = countTrailingOnes(STPopped);

    DEBUG(dbgs() << "Asm uses " << NumSTUses << " fixed regs, pops "
                 << NumSTPopped << ", and defines " << NumSTDefs << " regs.\n");

#ifndef NDEBUG
    // If any input operand uses constraint "f", all output register
    // constraints must be early-clobber defs.
    for (unsigned I = 0, E = MI.getNumOperands(); I < E; ++I)
      if (FRegIdx.count(I)) {
        assert((1 << getFPReg(MI.getOperand(I)) & STDefs) == 0 &&
               "Operands with constraint \"f\" cannot overlap with defs");
      }
#endif

    // Collect all FP registers (register operands with constraints "t", "u",
    // and "f") to kill afer the instruction.
    unsigned FPKills = ((1u << NumFPRegs) - 1) & ~0xff;
    for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
      MachineOperand &Op = MI.getOperand(i);
      if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6)
        continue;
      unsigned FPReg = getFPReg(Op);

      // If we kill this operand, make sure to pop it from the stack after the
      // asm.  We just remember it for now, and pop them all off at the end in
      // a batch.
      if (Op.isUse() && Op.isKill())
        FPKills |= 1U << FPReg;
    }

    // Do not include registers that are implicitly popped by defs/clobbers.
    FPKills &= ~(STDefs | STClobbers);

    // Now we can rearrange the live registers to match what was requested.
    unsigned char STUsesArray[8];

    for (unsigned I = 0; I < NumSTUses; ++I)
      STUsesArray[I] = I;

    shuffleStackTop(STUsesArray, NumSTUses, Inst);
    DEBUG({dbgs() << "Before asm: "; dumpStack();});

    // With the stack layout fixed, rewrite the FP registers.
    for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
      MachineOperand &Op = MI.getOperand(i);
      if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6)
        continue;

      unsigned FPReg = getFPReg(Op);

      if (FRegIdx.count(i))
        // Operand with constraint "f".
        Op.setReg(getSTReg(FPReg));
      else
        // Operand with a single register class constraint ("t" or "u").
        Op.setReg(X86::ST0 + FPReg);
    }

    // Simulate the inline asm popping its inputs and pushing its outputs.
    StackTop -= NumSTPopped;

    for (unsigned i = 0; i < NumSTDefs; ++i)
      pushReg(NumSTDefs - i - 1);

    // If this asm kills any FP registers (is the last use of them) we must
    // explicitly emit pop instructions for them.  Do this now after the asm has
    // executed so that the ST(x) numbers are not off (which would happen if we
    // did this inline with operand rewriting).
    //
    // Note: this might be a non-optimal pop sequence.  We might be able to do
    // better by trying to pop in stack order or something.
    while (FPKills) {
      unsigned FPReg = countTrailingZeros(FPKills);
      if (isLive(FPReg))
        freeStackSlotAfter(Inst, FPReg);
      FPKills &= ~(1U << FPReg);
    }

    // Don't delete the inline asm!
    return;
  }
  }

  Inst = MBB->erase(Inst);  // Remove the pseudo instruction

  // We want to leave I pointing to the previous instruction, but what if we
  // just erased the first instruction?
  if (Inst == MBB->begin()) {
    DEBUG(dbgs() << "Inserting dummy KILL\n");
    Inst = BuildMI(*MBB, Inst, DebugLoc(), TII->get(TargetOpcode::KILL));
  } else
    --Inst;
}

void FPS::setKillFlags(MachineBasicBlock &MBB) const {
  const TargetRegisterInfo *TRI =
      MBB.getParent()->getSubtarget().getRegisterInfo();
  LivePhysRegs LPR(TRI);

  LPR.addLiveOuts(MBB);

  for (MachineBasicBlock::reverse_iterator I = MBB.rbegin(), E = MBB.rend();
       I != E; ++I) {
    if (I->isDebugValue())
      continue;

    std::bitset<8> Defs;
    SmallVector<MachineOperand *, 2> Uses;
    MachineInstr &MI = *I;

    for (auto &MO : I->operands()) {
      if (!MO.isReg())
        continue;

      unsigned Reg = MO.getReg() - X86::FP0;

      if (Reg >= 8)
        continue;

      if (MO.isDef()) {
        Defs.set(Reg);
        if (!LPR.contains(MO.getReg()))
          MO.setIsDead();
      } else
        Uses.push_back(&MO);
    }

    for (auto *MO : Uses)
      if (Defs.test(getFPReg(*MO)) || !LPR.contains(MO->getReg()))
        MO->setIsKill();

    LPR.stepBackward(MI);
  }
}