summaryrefslogtreecommitdiff
path: root/lib/Target/ARM/ARMConstantIslandPass.cpp
blob: 70a25c26a9bda4909db21689374fd4a06a2b206f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
//===-- ARMConstantIslandPass.cpp - ARM constant islands ------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains a pass that splits the constant pool up into 'islands'
// which are scattered through-out the function.  This is required due to the
// limited pc-relative displacements that ARM has.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "arm-cp-islands"
#include "ARM.h"
#include "ARMMachineFunctionInfo.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "Thumb2InstrInfo.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include <algorithm>
using namespace llvm;

STATISTIC(NumCPEs,       "Number of constpool entries");
STATISTIC(NumSplit,      "Number of uncond branches inserted");
STATISTIC(NumCBrFixed,   "Number of cond branches fixed");
STATISTIC(NumUBrFixed,   "Number of uncond branches fixed");
STATISTIC(NumTBs,        "Number of table branches generated");
STATISTIC(NumT2CPShrunk, "Number of Thumb2 constantpool instructions shrunk");
STATISTIC(NumT2BrShrunk, "Number of Thumb2 immediate branches shrunk");
STATISTIC(NumCBZ,        "Number of CBZ / CBNZ formed");
STATISTIC(NumJTMoved,    "Number of jump table destination blocks moved");
STATISTIC(NumJTInserted, "Number of jump table intermediate blocks inserted");


static cl::opt<bool>
AdjustJumpTableBlocks("arm-adjust-jump-tables", cl::Hidden, cl::init(true),
          cl::desc("Adjust basic block layout to better use TB[BH]"));

// FIXME: This option should be removed once it has received sufficient testing.
static cl::opt<bool>
AlignConstantIslands("arm-align-constant-islands", cl::Hidden, cl::init(true),
          cl::desc("Align constant islands in code"));

/// UnknownPadding - Return the worst case padding that could result from
/// unknown offset bits.  This does not include alignment padding caused by
/// known offset bits.
///
/// @param LogAlign log2(alignment)
/// @param KnownBits Number of known low offset bits.
static inline unsigned UnknownPadding(unsigned LogAlign, unsigned KnownBits) {
  if (KnownBits < LogAlign)
    return (1u << LogAlign) - (1u << KnownBits);
  return 0;
}

namespace {
  /// ARMConstantIslands - Due to limited PC-relative displacements, ARM
  /// requires constant pool entries to be scattered among the instructions
  /// inside a function.  To do this, it completely ignores the normal LLVM
  /// constant pool; instead, it places constants wherever it feels like with
  /// special instructions.
  ///
  /// The terminology used in this pass includes:
  ///   Islands - Clumps of constants placed in the function.
  ///   Water   - Potential places where an island could be formed.
  ///   CPE     - A constant pool entry that has been placed somewhere, which
  ///             tracks a list of users.
  class ARMConstantIslands : public MachineFunctionPass {
    /// BasicBlockInfo - Information about the offset and size of a single
    /// basic block.
    struct BasicBlockInfo {
      /// Offset - Distance from the beginning of the function to the beginning
      /// of this basic block.
      ///
      /// Offsets are computed assuming worst case padding before an aligned
      /// block. This means that subtracting basic block offsets always gives a
      /// conservative estimate of the real distance which may be smaller.
      ///
      /// Because worst case padding is used, the computed offset of an aligned
      /// block may not actually be aligned.
      unsigned Offset;

      /// Size - Size of the basic block in bytes.  If the block contains
      /// inline assembly, this is a worst case estimate.
      ///
      /// The size does not include any alignment padding whether from the
      /// beginning of the block, or from an aligned jump table at the end.
      unsigned Size;

      /// KnownBits - The number of low bits in Offset that are known to be
      /// exact.  The remaining bits of Offset are an upper bound.
      uint8_t KnownBits;

      /// Unalign - When non-zero, the block contains instructions (inline asm)
      /// of unknown size.  The real size may be smaller than Size bytes by a
      /// multiple of 1 << Unalign.
      uint8_t Unalign;

      /// PostAlign - When non-zero, the block terminator contains a .align
      /// directive, so the end of the block is aligned to 1 << PostAlign
      /// bytes.
      uint8_t PostAlign;

      BasicBlockInfo() : Offset(0), Size(0), KnownBits(0), Unalign(0),
        PostAlign(0) {}

      /// Compute the number of known offset bits internally to this block.
      /// This number should be used to predict worst case padding when
      /// splitting the block.
      unsigned internalKnownBits() const {
        unsigned Bits = Unalign ? Unalign : KnownBits;
        // If the block size isn't a multiple of the known bits, assume the
        // worst case padding.
        if (Size & ((1u << Bits) - 1))
          Bits = CountTrailingZeros_32(Size);
        return Bits;
      }

      /// Compute the offset immediately following this block.  If LogAlign is
      /// specified, return the offset the successor block will get if it has
      /// this alignment.
      unsigned postOffset(unsigned LogAlign = 0) const {
        unsigned PO = Offset + Size;
        unsigned LA = std::max(unsigned(PostAlign), LogAlign);
        if (!LA)
          return PO;
        // Add alignment padding from the terminator.
        return PO + UnknownPadding(LA, internalKnownBits());
      }

      /// Compute the number of known low bits of postOffset.  If this block
      /// contains inline asm, the number of known bits drops to the
      /// instruction alignment.  An aligned terminator may increase the number
      /// of know bits.
      /// If LogAlign is given, also consider the alignment of the next block.
      unsigned postKnownBits(unsigned LogAlign = 0) const {
        return std::max(std::max(unsigned(PostAlign), LogAlign),
                        internalKnownBits());
      }
    };

    std::vector<BasicBlockInfo> BBInfo;

    /// WaterList - A sorted list of basic blocks where islands could be placed
    /// (i.e. blocks that don't fall through to the following block, due
    /// to a return, unreachable, or unconditional branch).
    std::vector<MachineBasicBlock*> WaterList;

    /// NewWaterList - The subset of WaterList that was created since the
    /// previous iteration by inserting unconditional branches.
    SmallSet<MachineBasicBlock*, 4> NewWaterList;

    typedef std::vector<MachineBasicBlock*>::iterator water_iterator;

    /// CPUser - One user of a constant pool, keeping the machine instruction
    /// pointer, the constant pool being referenced, and the max displacement
    /// allowed from the instruction to the CP.  The HighWaterMark records the
    /// highest basic block where a new CPEntry can be placed.  To ensure this
    /// pass terminates, the CP entries are initially placed at the end of the
    /// function and then move monotonically to lower addresses.  The
    /// exception to this rule is when the current CP entry for a particular
    /// CPUser is out of range, but there is another CP entry for the same
    /// constant value in range.  We want to use the existing in-range CP
    /// entry, but if it later moves out of range, the search for new water
    /// should resume where it left off.  The HighWaterMark is used to record
    /// that point.
    struct CPUser {
      MachineInstr *MI;
      MachineInstr *CPEMI;
      MachineBasicBlock *HighWaterMark;
    private:
      unsigned MaxDisp;
    public:
      bool NegOk;
      bool IsSoImm;
      bool KnownAlignment;
      CPUser(MachineInstr *mi, MachineInstr *cpemi, unsigned maxdisp,
             bool neg, bool soimm)
        : MI(mi), CPEMI(cpemi), MaxDisp(maxdisp), NegOk(neg), IsSoImm(soimm),
          KnownAlignment(false) {
        HighWaterMark = CPEMI->getParent();
      }
      /// getMaxDisp - Returns the maximum displacement supported by MI.
      /// Correct for unknown alignment.
      /// Conservatively subtract 2 bytes to handle weird alignment effects.
      unsigned getMaxDisp() const {
        return (KnownAlignment ? MaxDisp : MaxDisp - 2) - 2;
      }
    };

    /// CPUsers - Keep track of all of the machine instructions that use various
    /// constant pools and their max displacement.
    std::vector<CPUser> CPUsers;

    /// CPEntry - One per constant pool entry, keeping the machine instruction
    /// pointer, the constpool index, and the number of CPUser's which
    /// reference this entry.
    struct CPEntry {
      MachineInstr *CPEMI;
      unsigned CPI;
      unsigned RefCount;
      CPEntry(MachineInstr *cpemi, unsigned cpi, unsigned rc = 0)
        : CPEMI(cpemi), CPI(cpi), RefCount(rc) {}
    };

    /// CPEntries - Keep track of all of the constant pool entry machine
    /// instructions. For each original constpool index (i.e. those that
    /// existed upon entry to this pass), it keeps a vector of entries.
    /// Original elements are cloned as we go along; the clones are
    /// put in the vector of the original element, but have distinct CPIs.
    std::vector<std::vector<CPEntry> > CPEntries;

    /// ImmBranch - One per immediate branch, keeping the machine instruction
    /// pointer, conditional or unconditional, the max displacement,
    /// and (if isCond is true) the corresponding unconditional branch
    /// opcode.
    struct ImmBranch {
      MachineInstr *MI;
      unsigned MaxDisp : 31;
      bool isCond : 1;
      int UncondBr;
      ImmBranch(MachineInstr *mi, unsigned maxdisp, bool cond, int ubr)
        : MI(mi), MaxDisp(maxdisp), isCond(cond), UncondBr(ubr) {}
    };

    /// ImmBranches - Keep track of all the immediate branch instructions.
    ///
    std::vector<ImmBranch> ImmBranches;

    /// PushPopMIs - Keep track of all the Thumb push / pop instructions.
    ///
    SmallVector<MachineInstr*, 4> PushPopMIs;

    /// T2JumpTables - Keep track of all the Thumb2 jumptable instructions.
    SmallVector<MachineInstr*, 4> T2JumpTables;

    /// HasFarJump - True if any far jump instruction has been emitted during
    /// the branch fix up pass.
    bool HasFarJump;

    MachineFunction *MF;
    MachineConstantPool *MCP;
    const ARMBaseInstrInfo *TII;
    const ARMSubtarget *STI;
    ARMFunctionInfo *AFI;
    bool isThumb;
    bool isThumb1;
    bool isThumb2;
  public:
    static char ID;
    ARMConstantIslands() : MachineFunctionPass(ID) {}

    virtual bool runOnMachineFunction(MachineFunction &MF);

    virtual const char *getPassName() const {
      return "ARM constant island placement and branch shortening pass";
    }

  private:
    void doInitialPlacement(std::vector<MachineInstr*> &CPEMIs);
    CPEntry *findConstPoolEntry(unsigned CPI, const MachineInstr *CPEMI);
    unsigned getCPELogAlign(const MachineInstr *CPEMI);
    void scanFunctionJumpTables();
    void initializeFunctionInfo(const std::vector<MachineInstr*> &CPEMIs);
    MachineBasicBlock *splitBlockBeforeInstr(MachineInstr *MI);
    void updateForInsertedWaterBlock(MachineBasicBlock *NewBB);
    void adjustBBOffsetsAfter(MachineBasicBlock *BB);
    bool decrementCPEReferenceCount(unsigned CPI, MachineInstr* CPEMI);
    int findInRangeCPEntry(CPUser& U, unsigned UserOffset);
    bool findAvailableWater(CPUser&U, unsigned UserOffset,
                            water_iterator &WaterIter);
    void createNewWater(unsigned CPUserIndex, unsigned UserOffset,
                        MachineBasicBlock *&NewMBB);
    bool handleConstantPoolUser(unsigned CPUserIndex);
    void removeDeadCPEMI(MachineInstr *CPEMI);
    bool removeUnusedCPEntries();
    bool isCPEntryInRange(MachineInstr *MI, unsigned UserOffset,
                          MachineInstr *CPEMI, unsigned Disp, bool NegOk,
                          bool DoDump = false);
    bool isWaterInRange(unsigned UserOffset, MachineBasicBlock *Water,
                        CPUser &U, unsigned &Growth);
    bool isBBInRange(MachineInstr *MI, MachineBasicBlock *BB, unsigned Disp);
    bool fixupImmediateBr(ImmBranch &Br);
    bool fixupConditionalBr(ImmBranch &Br);
    bool fixupUnconditionalBr(ImmBranch &Br);
    bool undoLRSpillRestore();
    bool mayOptimizeThumb2Instruction(const MachineInstr *MI) const;
    bool optimizeThumb2Instructions();
    bool optimizeThumb2Branches();
    bool reorderThumb2JumpTables();
    bool optimizeThumb2JumpTables();
    MachineBasicBlock *adjustJTTargetBlockForward(MachineBasicBlock *BB,
                                                  MachineBasicBlock *JTBB);

    void computeBlockSize(MachineBasicBlock *MBB);
    unsigned getOffsetOf(MachineInstr *MI) const;
    unsigned getUserOffset(CPUser&) const;
    void dumpBBs();
    void verify();

    bool isOffsetInRange(unsigned UserOffset, unsigned TrialOffset,
                         unsigned Disp, bool NegativeOK, bool IsSoImm = false);
    bool isOffsetInRange(unsigned UserOffset, unsigned TrialOffset,
                         const CPUser &U) {
      return isOffsetInRange(UserOffset, TrialOffset,
                             U.getMaxDisp(), U.NegOk, U.IsSoImm);
    }
  };
  char ARMConstantIslands::ID = 0;
}

/// verify - check BBOffsets, BBSizes, alignment of islands
void ARMConstantIslands::verify() {
#ifndef NDEBUG
  for (MachineFunction::iterator MBBI = MF->begin(), E = MF->end();
       MBBI != E; ++MBBI) {
    MachineBasicBlock *MBB = MBBI;
    unsigned MBBId = MBB->getNumber();
    assert(!MBBId || BBInfo[MBBId - 1].postOffset() <= BBInfo[MBBId].Offset);
  }
  DEBUG(dbgs() << "Verifying " << CPUsers.size() << " CP users.\n");
  for (unsigned i = 0, e = CPUsers.size(); i != e; ++i) {
    CPUser &U = CPUsers[i];
    unsigned UserOffset = getUserOffset(U);
    // Verify offset using the real max displacement without the safety
    // adjustment.
    if (isCPEntryInRange(U.MI, UserOffset, U.CPEMI, U.getMaxDisp()+2, U.NegOk,
                         /* DoDump = */ true)) {
      DEBUG(dbgs() << "OK\n");
      continue;
    }
    DEBUG(dbgs() << "Out of range.\n");
    dumpBBs();
    DEBUG(MF->dump());
    llvm_unreachable("Constant pool entry out of range!");
  }
#endif
}

/// print block size and offset information - debugging
void ARMConstantIslands::dumpBBs() {
  DEBUG({
    for (unsigned J = 0, E = BBInfo.size(); J !=E; ++J) {
      const BasicBlockInfo &BBI = BBInfo[J];
      dbgs() << format("%08x BB#%u\t", BBI.Offset, J)
             << " kb=" << unsigned(BBI.KnownBits)
             << " ua=" << unsigned(BBI.Unalign)
             << " pa=" << unsigned(BBI.PostAlign)
             << format(" size=%#x\n", BBInfo[J].Size);
    }
  });
}

/// createARMConstantIslandPass - returns an instance of the constpool
/// island pass.
FunctionPass *llvm::createARMConstantIslandPass() {
  return new ARMConstantIslands();
}

bool ARMConstantIslands::runOnMachineFunction(MachineFunction &mf) {
  MF = &mf;
  MCP = mf.getConstantPool();

  DEBUG(dbgs() << "***** ARMConstantIslands: "
               << MCP->getConstants().size() << " CP entries, aligned to "
               << MCP->getConstantPoolAlignment() << " bytes *****\n");

  TII = (const ARMBaseInstrInfo*)MF->getTarget().getInstrInfo();
  AFI = MF->getInfo<ARMFunctionInfo>();
  STI = &MF->getTarget().getSubtarget<ARMSubtarget>();

  isThumb = AFI->isThumbFunction();
  isThumb1 = AFI->isThumb1OnlyFunction();
  isThumb2 = AFI->isThumb2Function();

  HasFarJump = false;

  // This pass invalidates liveness information when it splits basic blocks.
  MF->getRegInfo().invalidateLiveness();

  // Renumber all of the machine basic blocks in the function, guaranteeing that
  // the numbers agree with the position of the block in the function.
  MF->RenumberBlocks();

  // Try to reorder and otherwise adjust the block layout to make good use
  // of the TB[BH] instructions.
  bool MadeChange = false;
  if (isThumb2 && AdjustJumpTableBlocks) {
    scanFunctionJumpTables();
    MadeChange |= reorderThumb2JumpTables();
    // Data is out of date, so clear it. It'll be re-computed later.
    T2JumpTables.clear();
    // Blocks may have shifted around. Keep the numbering up to date.
    MF->RenumberBlocks();
  }

  // Thumb1 functions containing constant pools get 4-byte alignment.
  // This is so we can keep exact track of where the alignment padding goes.

  // ARM and Thumb2 functions need to be 4-byte aligned.
  if (!isThumb1)
    MF->ensureAlignment(2);  // 2 = log2(4)

  // Perform the initial placement of the constant pool entries.  To start with,
  // we put them all at the end of the function.
  std::vector<MachineInstr*> CPEMIs;
  if (!MCP->isEmpty())
    doInitialPlacement(CPEMIs);

  /// The next UID to take is the first unused one.
  AFI->initPICLabelUId(CPEMIs.size());

  // Do the initial scan of the function, building up information about the
  // sizes of each block, the location of all the water, and finding all of the
  // constant pool users.
  initializeFunctionInfo(CPEMIs);
  CPEMIs.clear();
  DEBUG(dumpBBs());


  /// Remove dead constant pool entries.
  MadeChange |= removeUnusedCPEntries();

  // Iteratively place constant pool entries and fix up branches until there
  // is no change.
  unsigned NoCPIters = 0, NoBRIters = 0;
  while (true) {
    DEBUG(dbgs() << "Beginning CP iteration #" << NoCPIters << '\n');
    bool CPChange = false;
    for (unsigned i = 0, e = CPUsers.size(); i != e; ++i)
      CPChange |= handleConstantPoolUser(i);
    if (CPChange && ++NoCPIters > 30)
      report_fatal_error("Constant Island pass failed to converge!");
    DEBUG(dumpBBs());

    // Clear NewWaterList now.  If we split a block for branches, it should
    // appear as "new water" for the next iteration of constant pool placement.
    NewWaterList.clear();

    DEBUG(dbgs() << "Beginning BR iteration #" << NoBRIters << '\n');
    bool BRChange = false;
    for (unsigned i = 0, e = ImmBranches.size(); i != e; ++i)
      BRChange |= fixupImmediateBr(ImmBranches[i]);
    if (BRChange && ++NoBRIters > 30)
      report_fatal_error("Branch Fix Up pass failed to converge!");
    DEBUG(dumpBBs());

    if (!CPChange && !BRChange)
      break;
    MadeChange = true;
  }

  // Shrink 32-bit Thumb2 branch, load, and store instructions.
  if (isThumb2 && !STI->prefers32BitThumb())
    MadeChange |= optimizeThumb2Instructions();

  // After a while, this might be made debug-only, but it is not expensive.
  verify();

  // If LR has been forced spilled and no far jump (i.e. BL) has been issued,
  // undo the spill / restore of LR if possible.
  if (isThumb && !HasFarJump && AFI->isLRSpilledForFarJump())
    MadeChange |= undoLRSpillRestore();

  // Save the mapping between original and cloned constpool entries.
  for (unsigned i = 0, e = CPEntries.size(); i != e; ++i) {
    for (unsigned j = 0, je = CPEntries[i].size(); j != je; ++j) {
      const CPEntry & CPE = CPEntries[i][j];
      AFI->recordCPEClone(i, CPE.CPI);
    }
  }

  DEBUG(dbgs() << '\n'; dumpBBs());

  BBInfo.clear();
  WaterList.clear();
  CPUsers.clear();
  CPEntries.clear();
  ImmBranches.clear();
  PushPopMIs.clear();
  T2JumpTables.clear();

  return MadeChange;
}

/// doInitialPlacement - Perform the initial placement of the constant pool
/// entries.  To start with, we put them all at the end of the function.
void
ARMConstantIslands::doInitialPlacement(std::vector<MachineInstr*> &CPEMIs) {
  // Create the basic block to hold the CPE's.
  MachineBasicBlock *BB = MF->CreateMachineBasicBlock();
  MF->push_back(BB);

  // MachineConstantPool measures alignment in bytes. We measure in log2(bytes).
  unsigned MaxAlign = Log2_32(MCP->getConstantPoolAlignment());

  // Mark the basic block as required by the const-pool.
  // If AlignConstantIslands isn't set, use 4-byte alignment for everything.
  BB->setAlignment(AlignConstantIslands ? MaxAlign : 2);

  // The function needs to be as aligned as the basic blocks. The linker may
  // move functions around based on their alignment.
  MF->ensureAlignment(BB->getAlignment());

  // Order the entries in BB by descending alignment.  That ensures correct
  // alignment of all entries as long as BB is sufficiently aligned.  Keep
  // track of the insertion point for each alignment.  We are going to bucket
  // sort the entries as they are created.
  SmallVector<MachineBasicBlock::iterator, 8> InsPoint(MaxAlign + 1, BB->end());

  // Add all of the constants from the constant pool to the end block, use an
  // identity mapping of CPI's to CPE's.
  const std::vector<MachineConstantPoolEntry> &CPs = MCP->getConstants();

  const DataLayout &TD = *MF->getTarget().getDataLayout();
  for (unsigned i = 0, e = CPs.size(); i != e; ++i) {
    unsigned Size = TD.getTypeAllocSize(CPs[i].getType());
    assert(Size >= 4 && "Too small constant pool entry");
    unsigned Align = CPs[i].getAlignment();
    assert(isPowerOf2_32(Align) && "Invalid alignment");
    // Verify that all constant pool entries are a multiple of their alignment.
    // If not, we would have to pad them out so that instructions stay aligned.
    assert((Size % Align) == 0 && "CP Entry not multiple of 4 bytes!");

    // Insert CONSTPOOL_ENTRY before entries with a smaller alignment.
    unsigned LogAlign = Log2_32(Align);
    MachineBasicBlock::iterator InsAt = InsPoint[LogAlign];
    MachineInstr *CPEMI =
      BuildMI(*BB, InsAt, DebugLoc(), TII->get(ARM::CONSTPOOL_ENTRY))
        .addImm(i).addConstantPoolIndex(i).addImm(Size);
    CPEMIs.push_back(CPEMI);

    // Ensure that future entries with higher alignment get inserted before
    // CPEMI. This is bucket sort with iterators.
    for (unsigned a = LogAlign + 1; a <= MaxAlign; ++a)
      if (InsPoint[a] == InsAt)
        InsPoint[a] = CPEMI;

    // Add a new CPEntry, but no corresponding CPUser yet.
    std::vector<CPEntry> CPEs;
    CPEs.push_back(CPEntry(CPEMI, i));
    CPEntries.push_back(CPEs);
    ++NumCPEs;
    DEBUG(dbgs() << "Moved CPI#" << i << " to end of function, size = "
                 << Size << ", align = " << Align <<'\n');
  }
  DEBUG(BB->dump());
}

/// BBHasFallthrough - Return true if the specified basic block can fallthrough
/// into the block immediately after it.
static bool BBHasFallthrough(MachineBasicBlock *MBB) {
  // Get the next machine basic block in the function.
  MachineFunction::iterator MBBI = MBB;
  // Can't fall off end of function.
  if (llvm::next(MBBI) == MBB->getParent()->end())
    return false;

  MachineBasicBlock *NextBB = llvm::next(MBBI);
  for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(),
       E = MBB->succ_end(); I != E; ++I)
    if (*I == NextBB)
      return true;

  return false;
}

/// findConstPoolEntry - Given the constpool index and CONSTPOOL_ENTRY MI,
/// look up the corresponding CPEntry.
ARMConstantIslands::CPEntry
*ARMConstantIslands::findConstPoolEntry(unsigned CPI,
                                        const MachineInstr *CPEMI) {
  std::vector<CPEntry> &CPEs = CPEntries[CPI];
  // Number of entries per constpool index should be small, just do a
  // linear search.
  for (unsigned i = 0, e = CPEs.size(); i != e; ++i) {
    if (CPEs[i].CPEMI == CPEMI)
      return &CPEs[i];
  }
  return NULL;
}

/// getCPELogAlign - Returns the required alignment of the constant pool entry
/// represented by CPEMI.  Alignment is measured in log2(bytes) units.
unsigned ARMConstantIslands::getCPELogAlign(const MachineInstr *CPEMI) {
  assert(CPEMI && CPEMI->getOpcode() == ARM::CONSTPOOL_ENTRY);

  // Everything is 4-byte aligned unless AlignConstantIslands is set.
  if (!AlignConstantIslands)
    return 2;

  unsigned CPI = CPEMI->getOperand(1).getIndex();
  assert(CPI < MCP->getConstants().size() && "Invalid constant pool index.");
  unsigned Align = MCP->getConstants()[CPI].getAlignment();
  assert(isPowerOf2_32(Align) && "Invalid CPE alignment");
  return Log2_32(Align);
}

/// scanFunctionJumpTables - Do a scan of the function, building up
/// information about the sizes of each block and the locations of all
/// the jump tables.
void ARMConstantIslands::scanFunctionJumpTables() {
  for (MachineFunction::iterator MBBI = MF->begin(), E = MF->end();
       MBBI != E; ++MBBI) {
    MachineBasicBlock &MBB = *MBBI;

    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
         I != E; ++I)
      if (I->isBranch() && I->getOpcode() == ARM::t2BR_JT)
        T2JumpTables.push_back(I);
  }
}

/// initializeFunctionInfo - Do the initial scan of the function, building up
/// information about the sizes of each block, the location of all the water,
/// and finding all of the constant pool users.
void ARMConstantIslands::
initializeFunctionInfo(const std::vector<MachineInstr*> &CPEMIs) {
  BBInfo.clear();
  BBInfo.resize(MF->getNumBlockIDs());

  // First thing, compute the size of all basic blocks, and see if the function
  // has any inline assembly in it. If so, we have to be conservative about
  // alignment assumptions, as we don't know for sure the size of any
  // instructions in the inline assembly.
  for (MachineFunction::iterator I = MF->begin(), E = MF->end(); I != E; ++I)
    computeBlockSize(I);

  // The known bits of the entry block offset are determined by the function
  // alignment.
  BBInfo.front().KnownBits = MF->getAlignment();

  // Compute block offsets and known bits.
  adjustBBOffsetsAfter(MF->begin());

  // Now go back through the instructions and build up our data structures.
  for (MachineFunction::iterator MBBI = MF->begin(), E = MF->end();
       MBBI != E; ++MBBI) {
    MachineBasicBlock &MBB = *MBBI;

    // If this block doesn't fall through into the next MBB, then this is
    // 'water' that a constant pool island could be placed.
    if (!BBHasFallthrough(&MBB))
      WaterList.push_back(&MBB);

    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
         I != E; ++I) {
      if (I->isDebugValue())
        continue;

      int Opc = I->getOpcode();
      if (I->isBranch()) {
        bool isCond = false;
        unsigned Bits = 0;
        unsigned Scale = 1;
        int UOpc = Opc;
        switch (Opc) {
        default:
          continue;  // Ignore other JT branches
        case ARM::t2BR_JT:
          T2JumpTables.push_back(I);
          continue;   // Does not get an entry in ImmBranches
        case ARM::Bcc:
          isCond = true;
          UOpc = ARM::B;
          // Fallthrough
        case ARM::B:
          Bits = 24;
          Scale = 4;
          break;
        case ARM::tBcc:
          isCond = true;
          UOpc = ARM::tB;
          Bits = 8;
          Scale = 2;
          break;
        case ARM::tB:
          Bits = 11;
          Scale = 2;
          break;
        case ARM::t2Bcc:
          isCond = true;
          UOpc = ARM::t2B;
          Bits = 20;
          Scale = 2;
          break;
        case ARM::t2B:
          Bits = 24;
          Scale = 2;
          break;
        }

        // Record this immediate branch.
        unsigned MaxOffs = ((1 << (Bits-1))-1) * Scale;
        ImmBranches.push_back(ImmBranch(I, MaxOffs, isCond, UOpc));
      }

      if (Opc == ARM::tPUSH || Opc == ARM::tPOP_RET)
        PushPopMIs.push_back(I);

      if (Opc == ARM::CONSTPOOL_ENTRY)
        continue;

      // Scan the instructions for constant pool operands.
      for (unsigned op = 0, e = I->getNumOperands(); op != e; ++op)
        if (I->getOperand(op).isCPI()) {
          // We found one.  The addressing mode tells us the max displacement
          // from the PC that this instruction permits.

          // Basic size info comes from the TSFlags field.
          unsigned Bits = 0;
          unsigned Scale = 1;
          bool NegOk = false;
          bool IsSoImm = false;

          switch (Opc) {
          default:
            llvm_unreachable("Unknown addressing mode for CP reference!");

          // Taking the address of a CP entry.
          case ARM::LEApcrel:
            // This takes a SoImm, which is 8 bit immediate rotated. We'll
            // pretend the maximum offset is 255 * 4. Since each instruction
            // 4 byte wide, this is always correct. We'll check for other
            // displacements that fits in a SoImm as well.
            Bits = 8;
            Scale = 4;
            NegOk = true;
            IsSoImm = true;
            break;
          case ARM::t2LEApcrel:
            Bits = 12;
            NegOk = true;
            break;
          case ARM::tLEApcrel:
            Bits = 8;
            Scale = 4;
            break;

          case ARM::LDRi12:
          case ARM::LDRcp:
          case ARM::t2LDRpci:
            Bits = 12;  // +-offset_12
            NegOk = true;
            break;

          case ARM::tLDRpci:
            Bits = 8;
            Scale = 4;  // +(offset_8*4)
            break;

          case ARM::VLDRD:
          case ARM::VLDRS:
            Bits = 8;
            Scale = 4;  // +-(offset_8*4)
            NegOk = true;
            break;
          }

          // Remember that this is a user of a CP entry.
          unsigned CPI = I->getOperand(op).getIndex();
          MachineInstr *CPEMI = CPEMIs[CPI];
          unsigned MaxOffs = ((1 << Bits)-1) * Scale;
          CPUsers.push_back(CPUser(I, CPEMI, MaxOffs, NegOk, IsSoImm));

          // Increment corresponding CPEntry reference count.
          CPEntry *CPE = findConstPoolEntry(CPI, CPEMI);
          assert(CPE && "Cannot find a corresponding CPEntry!");
          CPE->RefCount++;

          // Instructions can only use one CP entry, don't bother scanning the
          // rest of the operands.
          break;
        }
    }
  }
}

/// computeBlockSize - Compute the size and some alignment information for MBB.
/// This function updates BBInfo directly.
void ARMConstantIslands::computeBlockSize(MachineBasicBlock *MBB) {
  BasicBlockInfo &BBI = BBInfo[MBB->getNumber()];
  BBI.Size = 0;
  BBI.Unalign = 0;
  BBI.PostAlign = 0;

  for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E;
       ++I) {
    BBI.Size += TII->GetInstSizeInBytes(I);
    // For inline asm, GetInstSizeInBytes returns a conservative estimate.
    // The actual size may be smaller, but still a multiple of the instr size.
    if (I->isInlineAsm())
      BBI.Unalign = isThumb ? 1 : 2;
    // Also consider instructions that may be shrunk later.
    else if (isThumb && mayOptimizeThumb2Instruction(I))
      BBI.Unalign = 1;
  }

  // tBR_JTr contains a .align 2 directive.
  if (!MBB->empty() && MBB->back().getOpcode() == ARM::tBR_JTr) {
    BBI.PostAlign = 2;
    MBB->getParent()->ensureAlignment(2);
  }
}

/// getOffsetOf - Return the current offset of the specified machine instruction
/// from the start of the function.  This offset changes as stuff is moved
/// around inside the function.
unsigned ARMConstantIslands::getOffsetOf(MachineInstr *MI) const {
  MachineBasicBlock *MBB = MI->getParent();

  // The offset is composed of two things: the sum of the sizes of all MBB's
  // before this instruction's block, and the offset from the start of the block
  // it is in.
  unsigned Offset = BBInfo[MBB->getNumber()].Offset;

  // Sum instructions before MI in MBB.
  for (MachineBasicBlock::iterator I = MBB->begin(); &*I != MI; ++I) {
    assert(I != MBB->end() && "Didn't find MI in its own basic block?");
    Offset += TII->GetInstSizeInBytes(I);
  }
  return Offset;
}

/// CompareMBBNumbers - Little predicate function to sort the WaterList by MBB
/// ID.
static bool CompareMBBNumbers(const MachineBasicBlock *LHS,
                              const MachineBasicBlock *RHS) {
  return LHS->getNumber() < RHS->getNumber();
}

/// updateForInsertedWaterBlock - When a block is newly inserted into the
/// machine function, it upsets all of the block numbers.  Renumber the blocks
/// and update the arrays that parallel this numbering.
void ARMConstantIslands::updateForInsertedWaterBlock(MachineBasicBlock *NewBB) {
  // Renumber the MBB's to keep them consecutive.
  NewBB->getParent()->RenumberBlocks(NewBB);

  // Insert an entry into BBInfo to align it properly with the (newly
  // renumbered) block numbers.
  BBInfo.insert(BBInfo.begin() + NewBB->getNumber(), BasicBlockInfo());

  // Next, update WaterList.  Specifically, we need to add NewMBB as having
  // available water after it.
  water_iterator IP =
    std::lower_bound(WaterList.begin(), WaterList.end(), NewBB,
                     CompareMBBNumbers);
  WaterList.insert(IP, NewBB);
}


/// Split the basic block containing MI into two blocks, which are joined by
/// an unconditional branch.  Update data structures and renumber blocks to
/// account for this change and returns the newly created block.
MachineBasicBlock *ARMConstantIslands::splitBlockBeforeInstr(MachineInstr *MI) {
  MachineBasicBlock *OrigBB = MI->getParent();

  // Create a new MBB for the code after the OrigBB.
  MachineBasicBlock *NewBB =
    MF->CreateMachineBasicBlock(OrigBB->getBasicBlock());
  MachineFunction::iterator MBBI = OrigBB; ++MBBI;
  MF->insert(MBBI, NewBB);

  // Splice the instructions starting with MI over to NewBB.
  NewBB->splice(NewBB->end(), OrigBB, MI, OrigBB->end());

  // Add an unconditional branch from OrigBB to NewBB.
  // Note the new unconditional branch is not being recorded.
  // There doesn't seem to be meaningful DebugInfo available; this doesn't
  // correspond to anything in the source.
  unsigned Opc = isThumb ? (isThumb2 ? ARM::t2B : ARM::tB) : ARM::B;
  if (!isThumb)
    BuildMI(OrigBB, DebugLoc(), TII->get(Opc)).addMBB(NewBB);
  else
    BuildMI(OrigBB, DebugLoc(), TII->get(Opc)).addMBB(NewBB)
            .addImm(ARMCC::AL).addReg(0);
  ++NumSplit;

  // Update the CFG.  All succs of OrigBB are now succs of NewBB.
  NewBB->transferSuccessors(OrigBB);

  // OrigBB branches to NewBB.
  OrigBB->addSuccessor(NewBB);

  // Update internal data structures to account for the newly inserted MBB.
  // This is almost the same as updateForInsertedWaterBlock, except that
  // the Water goes after OrigBB, not NewBB.
  MF->RenumberBlocks(NewBB);

  // Insert an entry into BBInfo to align it properly with the (newly
  // renumbered) block numbers.
  BBInfo.insert(BBInfo.begin() + NewBB->getNumber(), BasicBlockInfo());

  // Next, update WaterList.  Specifically, we need to add OrigMBB as having
  // available water after it (but not if it's already there, which happens
  // when splitting before a conditional branch that is followed by an
  // unconditional branch - in that case we want to insert NewBB).
  water_iterator IP =
    std::lower_bound(WaterList.begin(), WaterList.end(), OrigBB,
                     CompareMBBNumbers);
  MachineBasicBlock* WaterBB = *IP;
  if (WaterBB == OrigBB)
    WaterList.insert(llvm::next(IP), NewBB);
  else
    WaterList.insert(IP, OrigBB);
  NewWaterList.insert(OrigBB);

  // Figure out how large the OrigBB is.  As the first half of the original
  // block, it cannot contain a tablejump.  The size includes
  // the new jump we added.  (It should be possible to do this without
  // recounting everything, but it's very confusing, and this is rarely
  // executed.)
  computeBlockSize(OrigBB);

  // Figure out how large the NewMBB is.  As the second half of the original
  // block, it may contain a tablejump.
  computeBlockSize(NewBB);

  // All BBOffsets following these blocks must be modified.
  adjustBBOffsetsAfter(OrigBB);

  return NewBB;
}

/// getUserOffset - Compute the offset of U.MI as seen by the hardware
/// displacement computation.  Update U.KnownAlignment to match its current
/// basic block location.
unsigned ARMConstantIslands::getUserOffset(CPUser &U) const {
  unsigned UserOffset = getOffsetOf(U.MI);
  const BasicBlockInfo &BBI = BBInfo[U.MI->getParent()->getNumber()];
  unsigned KnownBits = BBI.internalKnownBits();

  // The value read from PC is offset from the actual instruction address.
  UserOffset += (isThumb ? 4 : 8);

  // Because of inline assembly, we may not know the alignment (mod 4) of U.MI.
  // Make sure U.getMaxDisp() returns a constrained range.
  U.KnownAlignment = (KnownBits >= 2);

  // On Thumb, offsets==2 mod 4 are rounded down by the hardware for
  // purposes of the displacement computation; compensate for that here.
  // For unknown alignments, getMaxDisp() constrains the range instead.
  if (isThumb && U.KnownAlignment)
    UserOffset &= ~3u;

  return UserOffset;
}

/// isOffsetInRange - Checks whether UserOffset (the location of a constant pool
/// reference) is within MaxDisp of TrialOffset (a proposed location of a
/// constant pool entry).
/// UserOffset is computed by getUserOffset above to include PC adjustments. If
/// the mod 4 alignment of UserOffset is not known, the uncertainty must be
/// subtracted from MaxDisp instead. CPUser::getMaxDisp() does that.
bool ARMConstantIslands::isOffsetInRange(unsigned UserOffset,
                                         unsigned TrialOffset, unsigned MaxDisp,
                                         bool NegativeOK, bool IsSoImm) {
  if (UserOffset <= TrialOffset) {
    // User before the Trial.
    if (TrialOffset - UserOffset <= MaxDisp)
      return true;
    // FIXME: Make use full range of soimm values.
  } else if (NegativeOK) {
    if (UserOffset - TrialOffset <= MaxDisp)
      return true;
    // FIXME: Make use full range of soimm values.
  }
  return false;
}

/// isWaterInRange - Returns true if a CPE placed after the specified
/// Water (a basic block) will be in range for the specific MI.
///
/// Compute how much the function will grow by inserting a CPE after Water.
bool ARMConstantIslands::isWaterInRange(unsigned UserOffset,
                                        MachineBasicBlock* Water, CPUser &U,
                                        unsigned &Growth) {
  unsigned CPELogAlign = getCPELogAlign(U.CPEMI);
  unsigned CPEOffset = BBInfo[Water->getNumber()].postOffset(CPELogAlign);
  unsigned NextBlockOffset, NextBlockAlignment;
  MachineFunction::const_iterator NextBlock = Water;
  if (++NextBlock == MF->end()) {
    NextBlockOffset = BBInfo[Water->getNumber()].postOffset();
    NextBlockAlignment = 0;
  } else {
    NextBlockOffset = BBInfo[NextBlock->getNumber()].Offset;
    NextBlockAlignment = NextBlock->getAlignment();
  }
  unsigned Size = U.CPEMI->getOperand(2).getImm();
  unsigned CPEEnd = CPEOffset + Size;

  // The CPE may be able to hide in the alignment padding before the next
  // block. It may also cause more padding to be required if it is more aligned
  // that the next block.
  if (CPEEnd > NextBlockOffset) {
    Growth = CPEEnd - NextBlockOffset;
    // Compute the padding that would go at the end of the CPE to align the next
    // block.
    Growth += OffsetToAlignment(CPEEnd, 1u << NextBlockAlignment);

    // If the CPE is to be inserted before the instruction, that will raise
    // the offset of the instruction. Also account for unknown alignment padding
    // in blocks between CPE and the user.
    if (CPEOffset < UserOffset)
      UserOffset += Growth + UnknownPadding(MF->getAlignment(), CPELogAlign);
  } else
    // CPE fits in existing padding.
    Growth = 0;

  return isOffsetInRange(UserOffset, CPEOffset, U);
}

/// isCPEntryInRange - Returns true if the distance between specific MI and
/// specific ConstPool entry instruction can fit in MI's displacement field.
bool ARMConstantIslands::isCPEntryInRange(MachineInstr *MI, unsigned UserOffset,
                                      MachineInstr *CPEMI, unsigned MaxDisp,
                                      bool NegOk, bool DoDump) {
  unsigned CPEOffset  = getOffsetOf(CPEMI);

  if (DoDump) {
    DEBUG({
      unsigned Block = MI->getParent()->getNumber();
      const BasicBlockInfo &BBI = BBInfo[Block];
      dbgs() << "User of CPE#" << CPEMI->getOperand(0).getImm()
             << " max delta=" << MaxDisp
             << format(" insn address=%#x", UserOffset)
             << " in BB#" << Block << ": "
             << format("%#x-%x\t", BBI.Offset, BBI.postOffset()) << *MI
             << format("CPE address=%#x offset=%+d: ", CPEOffset,
                       int(CPEOffset-UserOffset));
    });
  }

  return isOffsetInRange(UserOffset, CPEOffset, MaxDisp, NegOk);
}

#ifndef NDEBUG
/// BBIsJumpedOver - Return true of the specified basic block's only predecessor
/// unconditionally branches to its only successor.
static bool BBIsJumpedOver(MachineBasicBlock *MBB) {
  if (MBB->pred_size() != 1 || MBB->succ_size() != 1)
    return false;

  MachineBasicBlock *Succ = *MBB->succ_begin();
  MachineBasicBlock *Pred = *MBB->pred_begin();
  MachineInstr *PredMI = &Pred->back();
  if (PredMI->getOpcode() == ARM::B || PredMI->getOpcode() == ARM::tB
      || PredMI->getOpcode() == ARM::t2B)
    return PredMI->getOperand(0).getMBB() == Succ;
  return false;
}
#endif // NDEBUG

void ARMConstantIslands::adjustBBOffsetsAfter(MachineBasicBlock *BB) {
  unsigned BBNum = BB->getNumber();
  for(unsigned i = BBNum + 1, e = MF->getNumBlockIDs(); i < e; ++i) {
    // Get the offset and known bits at the end of the layout predecessor.
    // Include the alignment of the current block.
    unsigned LogAlign = MF->getBlockNumbered(i)->getAlignment();
    unsigned Offset = BBInfo[i - 1].postOffset(LogAlign);
    unsigned KnownBits = BBInfo[i - 1].postKnownBits(LogAlign);

    // This is where block i begins.  Stop if the offset is already correct,
    // and we have updated 2 blocks.  This is the maximum number of blocks
    // changed before calling this function.
    if (i > BBNum + 2 &&
        BBInfo[i].Offset == Offset &&
        BBInfo[i].KnownBits == KnownBits)
      break;

    BBInfo[i].Offset = Offset;
    BBInfo[i].KnownBits = KnownBits;
  }
}

/// decrementCPEReferenceCount - find the constant pool entry with index CPI
/// and instruction CPEMI, and decrement its refcount.  If the refcount
/// becomes 0 remove the entry and instruction.  Returns true if we removed
/// the entry, false if we didn't.

bool ARMConstantIslands::decrementCPEReferenceCount(unsigned CPI,
                                                    MachineInstr *CPEMI) {
  // Find the old entry. Eliminate it if it is no longer used.
  CPEntry *CPE = findConstPoolEntry(CPI, CPEMI);
  assert(CPE && "Unexpected!");
  if (--CPE->RefCount == 0) {
    removeDeadCPEMI(CPEMI);
    CPE->CPEMI = NULL;
    --NumCPEs;
    return true;
  }
  return false;
}

/// LookForCPEntryInRange - see if the currently referenced CPE is in range;
/// if not, see if an in-range clone of the CPE is in range, and if so,
/// change the data structures so the user references the clone.  Returns:
/// 0 = no existing entry found
/// 1 = entry found, and there were no code insertions or deletions
/// 2 = entry found, and there were code insertions or deletions
int ARMConstantIslands::findInRangeCPEntry(CPUser& U, unsigned UserOffset)
{
  MachineInstr *UserMI = U.MI;
  MachineInstr *CPEMI  = U.CPEMI;

  // Check to see if the CPE is already in-range.
  if (isCPEntryInRange(UserMI, UserOffset, CPEMI, U.getMaxDisp(), U.NegOk,
                       true)) {
    DEBUG(dbgs() << "In range\n");
    return 1;
  }

  // No.  Look for previously created clones of the CPE that are in range.
  unsigned CPI = CPEMI->getOperand(1).getIndex();
  std::vector<CPEntry> &CPEs = CPEntries[CPI];
  for (unsigned i = 0, e = CPEs.size(); i != e; ++i) {
    // We already tried this one
    if (CPEs[i].CPEMI == CPEMI)
      continue;
    // Removing CPEs can leave empty entries, skip
    if (CPEs[i].CPEMI == NULL)
      continue;
    if (isCPEntryInRange(UserMI, UserOffset, CPEs[i].CPEMI, U.getMaxDisp(),
                     U.NegOk)) {
      DEBUG(dbgs() << "Replacing CPE#" << CPI << " with CPE#"
                   << CPEs[i].CPI << "\n");
      // Point the CPUser node to the replacement
      U.CPEMI = CPEs[i].CPEMI;
      // Change the CPI in the instruction operand to refer to the clone.
      for (unsigned j = 0, e = UserMI->getNumOperands(); j != e; ++j)
        if (UserMI->getOperand(j).isCPI()) {
          UserMI->getOperand(j).setIndex(CPEs[i].CPI);
          break;
        }
      // Adjust the refcount of the clone...
      CPEs[i].RefCount++;
      // ...and the original.  If we didn't remove the old entry, none of the
      // addresses changed, so we don't need another pass.
      return decrementCPEReferenceCount(CPI, CPEMI) ? 2 : 1;
    }
  }
  return 0;
}

/// getUnconditionalBrDisp - Returns the maximum displacement that can fit in
/// the specific unconditional branch instruction.
static inline unsigned getUnconditionalBrDisp(int Opc) {
  switch (Opc) {
  case ARM::tB:
    return ((1<<10)-1)*2;
  case ARM::t2B:
    return ((1<<23)-1)*2;
  default:
    break;
  }

  return ((1<<23)-1)*4;
}

/// findAvailableWater - Look for an existing entry in the WaterList in which
/// we can place the CPE referenced from U so it's within range of U's MI.
/// Returns true if found, false if not.  If it returns true, WaterIter
/// is set to the WaterList entry.  For Thumb, prefer water that will not
/// introduce padding to water that will.  To ensure that this pass
/// terminates, the CPE location for a particular CPUser is only allowed to
/// move to a lower address, so search backward from the end of the list and
/// prefer the first water that is in range.
bool ARMConstantIslands::findAvailableWater(CPUser &U, unsigned UserOffset,
                                      water_iterator &WaterIter) {
  if (WaterList.empty())
    return false;

  unsigned BestGrowth = ~0u;
  for (water_iterator IP = prior(WaterList.end()), B = WaterList.begin();;
       --IP) {
    MachineBasicBlock* WaterBB = *IP;
    // Check if water is in range and is either at a lower address than the
    // current "high water mark" or a new water block that was created since
    // the previous iteration by inserting an unconditional branch.  In the
    // latter case, we want to allow resetting the high water mark back to
    // this new water since we haven't seen it before.  Inserting branches
    // should be relatively uncommon and when it does happen, we want to be
    // sure to take advantage of it for all the CPEs near that block, so that
    // we don't insert more branches than necessary.
    unsigned Growth;
    if (isWaterInRange(UserOffset, WaterBB, U, Growth) &&
        (WaterBB->getNumber() < U.HighWaterMark->getNumber() ||
         NewWaterList.count(WaterBB)) && Growth < BestGrowth) {
      // This is the least amount of required padding seen so far.
      BestGrowth = Growth;
      WaterIter = IP;
      DEBUG(dbgs() << "Found water after BB#" << WaterBB->getNumber()
                   << " Growth=" << Growth << '\n');

      // Keep looking unless it is perfect.
      if (BestGrowth == 0)
        return true;
    }
    if (IP == B)
      break;
  }
  return BestGrowth != ~0u;
}

/// createNewWater - No existing WaterList entry will work for
/// CPUsers[CPUserIndex], so create a place to put the CPE.  The end of the
/// block is used if in range, and the conditional branch munged so control
/// flow is correct.  Otherwise the block is split to create a hole with an
/// unconditional branch around it.  In either case NewMBB is set to a
/// block following which the new island can be inserted (the WaterList
/// is not adjusted).
void ARMConstantIslands::createNewWater(unsigned CPUserIndex,
                                        unsigned UserOffset,
                                        MachineBasicBlock *&NewMBB) {
  CPUser &U = CPUsers[CPUserIndex];
  MachineInstr *UserMI = U.MI;
  MachineInstr *CPEMI  = U.CPEMI;
  unsigned CPELogAlign = getCPELogAlign(CPEMI);
  MachineBasicBlock *UserMBB = UserMI->getParent();
  const BasicBlockInfo &UserBBI = BBInfo[UserMBB->getNumber()];

  // If the block does not end in an unconditional branch already, and if the
  // end of the block is within range, make new water there.  (The addition
  // below is for the unconditional branch we will be adding: 4 bytes on ARM +
  // Thumb2, 2 on Thumb1.
  if (BBHasFallthrough(UserMBB)) {
    // Size of branch to insert.
    unsigned Delta = isThumb1 ? 2 : 4;
    // Compute the offset where the CPE will begin.
    unsigned CPEOffset = UserBBI.postOffset(CPELogAlign) + Delta;

    if (isOffsetInRange(UserOffset, CPEOffset, U)) {
      DEBUG(dbgs() << "Split at end of BB#" << UserMBB->getNumber()
            << format(", expected CPE offset %#x\n", CPEOffset));
      NewMBB = llvm::next(MachineFunction::iterator(UserMBB));
      // Add an unconditional branch from UserMBB to fallthrough block.  Record
      // it for branch lengthening; this new branch will not get out of range,
      // but if the preceding conditional branch is out of range, the targets
      // will be exchanged, and the altered branch may be out of range, so the
      // machinery has to know about it.
      int UncondBr = isThumb ? ((isThumb2) ? ARM::t2B : ARM::tB) : ARM::B;
      if (!isThumb)
        BuildMI(UserMBB, DebugLoc(), TII->get(UncondBr)).addMBB(NewMBB);
      else
        BuildMI(UserMBB, DebugLoc(), TII->get(UncondBr)).addMBB(NewMBB)
          .addImm(ARMCC::AL).addReg(0);
      unsigned MaxDisp = getUnconditionalBrDisp(UncondBr);
      ImmBranches.push_back(ImmBranch(&UserMBB->back(),
                                      MaxDisp, false, UncondBr));
      BBInfo[UserMBB->getNumber()].Size += Delta;
      adjustBBOffsetsAfter(UserMBB);
      return;
    }
  }

  // What a big block.  Find a place within the block to split it.  This is a
  // little tricky on Thumb1 since instructions are 2 bytes and constant pool
  // entries are 4 bytes: if instruction I references island CPE, and
  // instruction I+1 references CPE', it will not work well to put CPE as far
  // forward as possible, since then CPE' cannot immediately follow it (that
  // location is 2 bytes farther away from I+1 than CPE was from I) and we'd
  // need to create a new island.  So, we make a first guess, then walk through
  // the instructions between the one currently being looked at and the
  // possible insertion point, and make sure any other instructions that
  // reference CPEs will be able to use the same island area; if not, we back
  // up the insertion point.

  // Try to split the block so it's fully aligned.  Compute the latest split
  // point where we can add a 4-byte branch instruction, and then align to
  // LogAlign which is the largest possible alignment in the function.
  unsigned LogAlign = MF->getAlignment();
  assert(LogAlign >= CPELogAlign && "Over-aligned constant pool entry");
  unsigned KnownBits = UserBBI.internalKnownBits();
  unsigned UPad = UnknownPadding(LogAlign, KnownBits);
  unsigned BaseInsertOffset = UserOffset + U.getMaxDisp() - UPad;
  DEBUG(dbgs() << format("Split in middle of big block before %#x",
                         BaseInsertOffset));

  // The 4 in the following is for the unconditional branch we'll be inserting
  // (allows for long branch on Thumb1).  Alignment of the island is handled
  // inside isOffsetInRange.
  BaseInsertOffset -= 4;

  DEBUG(dbgs() << format(", adjusted to %#x", BaseInsertOffset)
               << " la=" << LogAlign
               << " kb=" << KnownBits
               << " up=" << UPad << '\n');

  // This could point off the end of the block if we've already got constant
  // pool entries following this block; only the last one is in the water list.
  // Back past any possible branches (allow for a conditional and a maximally
  // long unconditional).
  if (BaseInsertOffset + 8 >= UserBBI.postOffset()) {
    BaseInsertOffset = UserBBI.postOffset() - UPad - 8;
    DEBUG(dbgs() << format("Move inside block: %#x\n", BaseInsertOffset));
  }
  unsigned EndInsertOffset = BaseInsertOffset + 4 + UPad +
    CPEMI->getOperand(2).getImm();
  MachineBasicBlock::iterator MI = UserMI;
  ++MI;
  unsigned CPUIndex = CPUserIndex+1;
  unsigned NumCPUsers = CPUsers.size();
  MachineInstr *LastIT = 0;
  for (unsigned Offset = UserOffset+TII->GetInstSizeInBytes(UserMI);
       Offset < BaseInsertOffset;
       Offset += TII->GetInstSizeInBytes(MI),
       MI = llvm::next(MI)) {
    assert(MI != UserMBB->end() && "Fell off end of block");
    if (CPUIndex < NumCPUsers && CPUsers[CPUIndex].MI == MI) {
      CPUser &U = CPUsers[CPUIndex];
      if (!isOffsetInRange(Offset, EndInsertOffset, U)) {
        // Shift intertion point by one unit of alignment so it is within reach.
        BaseInsertOffset -= 1u << LogAlign;
        EndInsertOffset  -= 1u << LogAlign;
      }
      // This is overly conservative, as we don't account for CPEMIs being
      // reused within the block, but it doesn't matter much.  Also assume CPEs
      // are added in order with alignment padding.  We may eventually be able
      // to pack the aligned CPEs better.
      EndInsertOffset += U.CPEMI->getOperand(2).getImm();
      CPUIndex++;
    }

    // Remember the last IT instruction.
    if (MI->getOpcode() == ARM::t2IT)
      LastIT = MI;
  }

  --MI;

  // Avoid splitting an IT block.
  if (LastIT) {
    unsigned PredReg = 0;
    ARMCC::CondCodes CC = getITInstrPredicate(MI, PredReg);
    if (CC != ARMCC::AL)
      MI = LastIT;
  }
  NewMBB = splitBlockBeforeInstr(MI);
}

/// handleConstantPoolUser - Analyze the specified user, checking to see if it
/// is out-of-range.  If so, pick up the constant pool value and move it some
/// place in-range.  Return true if we changed any addresses (thus must run
/// another pass of branch lengthening), false otherwise.
bool ARMConstantIslands::handleConstantPoolUser(unsigned CPUserIndex) {
  CPUser &U = CPUsers[CPUserIndex];
  MachineInstr *UserMI = U.MI;
  MachineInstr *CPEMI  = U.CPEMI;
  unsigned CPI = CPEMI->getOperand(1).getIndex();
  unsigned Size = CPEMI->getOperand(2).getImm();
  // Compute this only once, it's expensive.
  unsigned UserOffset = getUserOffset(U);

  // See if the current entry is within range, or there is a clone of it
  // in range.
  int result = findInRangeCPEntry(U, UserOffset);
  if (result==1) return false;
  else if (result==2) return true;

  // No existing clone of this CPE is within range.
  // We will be generating a new clone.  Get a UID for it.
  unsigned ID = AFI->createPICLabelUId();

  // Look for water where we can place this CPE.
  MachineBasicBlock *NewIsland = MF->CreateMachineBasicBlock();
  MachineBasicBlock *NewMBB;
  water_iterator IP;
  if (findAvailableWater(U, UserOffset, IP)) {
    DEBUG(dbgs() << "Found water in range\n");
    MachineBasicBlock *WaterBB = *IP;

    // If the original WaterList entry was "new water" on this iteration,
    // propagate that to the new island.  This is just keeping NewWaterList
    // updated to match the WaterList, which will be updated below.
    if (NewWaterList.erase(WaterBB))
      NewWaterList.insert(NewIsland);

    // The new CPE goes before the following block (NewMBB).
    NewMBB = llvm::next(MachineFunction::iterator(WaterBB));

  } else {
    // No water found.
    DEBUG(dbgs() << "No water found\n");
    createNewWater(CPUserIndex, UserOffset, NewMBB);

    // splitBlockBeforeInstr adds to WaterList, which is important when it is
    // called while handling branches so that the water will be seen on the
    // next iteration for constant pools, but in this context, we don't want
    // it.  Check for this so it will be removed from the WaterList.
    // Also remove any entry from NewWaterList.
    MachineBasicBlock *WaterBB = prior(MachineFunction::iterator(NewMBB));
    IP = std::find(WaterList.begin(), WaterList.end(), WaterBB);
    if (IP != WaterList.end())
      NewWaterList.erase(WaterBB);

    // We are adding new water.  Update NewWaterList.
    NewWaterList.insert(NewIsland);
  }

  // Remove the original WaterList entry; we want subsequent insertions in
  // this vicinity to go after the one we're about to insert.  This
  // considerably reduces the number of times we have to move the same CPE
  // more than once and is also important to ensure the algorithm terminates.
  if (IP != WaterList.end())
    WaterList.erase(IP);

  // Okay, we know we can put an island before NewMBB now, do it!
  MF->insert(NewMBB, NewIsland);

  // Update internal data structures to account for the newly inserted MBB.
  updateForInsertedWaterBlock(NewIsland);

  // Decrement the old entry, and remove it if refcount becomes 0.
  decrementCPEReferenceCount(CPI, CPEMI);

  // Now that we have an island to add the CPE to, clone the original CPE and
  // add it to the island.
  U.HighWaterMark = NewIsland;
  U.CPEMI = BuildMI(NewIsland, DebugLoc(), TII->get(ARM::CONSTPOOL_ENTRY))
                .addImm(ID).addConstantPoolIndex(CPI).addImm(Size);
  CPEntries[CPI].push_back(CPEntry(U.CPEMI, ID, 1));
  ++NumCPEs;

  // Mark the basic block as aligned as required by the const-pool entry.
  NewIsland->setAlignment(getCPELogAlign(U.CPEMI));

  // Increase the size of the island block to account for the new entry.
  BBInfo[NewIsland->getNumber()].Size += Size;
  adjustBBOffsetsAfter(llvm::prior(MachineFunction::iterator(NewIsland)));

  // Finally, change the CPI in the instruction operand to be ID.
  for (unsigned i = 0, e = UserMI->getNumOperands(); i != e; ++i)
    if (UserMI->getOperand(i).isCPI()) {
      UserMI->getOperand(i).setIndex(ID);
      break;
    }

  DEBUG(dbgs() << "  Moved CPE to #" << ID << " CPI=" << CPI
        << format(" offset=%#x\n", BBInfo[NewIsland->getNumber()].Offset));

  return true;
}

/// removeDeadCPEMI - Remove a dead constant pool entry instruction. Update
/// sizes and offsets of impacted basic blocks.
void ARMConstantIslands::removeDeadCPEMI(MachineInstr *CPEMI) {
  MachineBasicBlock *CPEBB = CPEMI->getParent();
  unsigned Size = CPEMI->getOperand(2).getImm();
  CPEMI->eraseFromParent();
  BBInfo[CPEBB->getNumber()].Size -= Size;
  // All succeeding offsets have the current size value added in, fix this.
  if (CPEBB->empty()) {
    BBInfo[CPEBB->getNumber()].Size = 0;

    // This block no longer needs to be aligned. <rdar://problem/10534709>.
    CPEBB->setAlignment(0);
  } else
    // Entries are sorted by descending alignment, so realign from the front.
    CPEBB->setAlignment(getCPELogAlign(CPEBB->begin()));

  adjustBBOffsetsAfter(CPEBB);
  // An island has only one predecessor BB and one successor BB. Check if
  // this BB's predecessor jumps directly to this BB's successor. This
  // shouldn't happen currently.
  assert(!BBIsJumpedOver(CPEBB) && "How did this happen?");
  // FIXME: remove the empty blocks after all the work is done?
}

/// removeUnusedCPEntries - Remove constant pool entries whose refcounts
/// are zero.
bool ARMConstantIslands::removeUnusedCPEntries() {
  unsigned MadeChange = false;
  for (unsigned i = 0, e = CPEntries.size(); i != e; ++i) {
      std::vector<CPEntry> &CPEs = CPEntries[i];
      for (unsigned j = 0, ee = CPEs.size(); j != ee; ++j) {
        if (CPEs[j].RefCount == 0 && CPEs[j].CPEMI) {
          removeDeadCPEMI(CPEs[j].CPEMI);
          CPEs[j].CPEMI = NULL;
          MadeChange = true;
        }
      }
  }
  return MadeChange;
}

/// isBBInRange - Returns true if the distance between specific MI and
/// specific BB can fit in MI's displacement field.
bool ARMConstantIslands::isBBInRange(MachineInstr *MI,MachineBasicBlock *DestBB,
                                     unsigned MaxDisp) {
  unsigned PCAdj      = isThumb ? 4 : 8;
  unsigned BrOffset   = getOffsetOf(MI) + PCAdj;
  unsigned DestOffset = BBInfo[DestBB->getNumber()].Offset;

  DEBUG(dbgs() << "Branch of destination BB#" << DestBB->getNumber()
               << " from BB#" << MI->getParent()->getNumber()
               << " max delta=" << MaxDisp
               << " from " << getOffsetOf(MI) << " to " << DestOffset
               << " offset " << int(DestOffset-BrOffset) << "\t" << *MI);

  if (BrOffset <= DestOffset) {
    // Branch before the Dest.
    if (DestOffset-BrOffset <= MaxDisp)
      return true;
  } else {
    if (BrOffset-DestOffset <= MaxDisp)
      return true;
  }
  return false;
}

/// fixupImmediateBr - Fix up an immediate branch whose destination is too far
/// away to fit in its displacement field.
bool ARMConstantIslands::fixupImmediateBr(ImmBranch &Br) {
  MachineInstr *MI = Br.MI;
  MachineBasicBlock *DestBB = MI->getOperand(0).getMBB();

  // Check to see if the DestBB is already in-range.
  if (isBBInRange(MI, DestBB, Br.MaxDisp))
    return false;

  if (!Br.isCond)
    return fixupUnconditionalBr(Br);
  return fixupConditionalBr(Br);
}

/// fixupUnconditionalBr - Fix up an unconditional branch whose destination is
/// too far away to fit in its displacement field. If the LR register has been
/// spilled in the epilogue, then we can use BL to implement a far jump.
/// Otherwise, add an intermediate branch instruction to a branch.
bool
ARMConstantIslands::fixupUnconditionalBr(ImmBranch &Br) {
  MachineInstr *MI = Br.MI;
  MachineBasicBlock *MBB = MI->getParent();
  if (!isThumb1)
    llvm_unreachable("fixupUnconditionalBr is Thumb1 only!");

  // Use BL to implement far jump.
  Br.MaxDisp = (1 << 21) * 2;
  MI->setDesc(TII->get(ARM::tBfar));
  BBInfo[MBB->getNumber()].Size += 2;
  adjustBBOffsetsAfter(MBB);
  HasFarJump = true;
  ++NumUBrFixed;

  DEBUG(dbgs() << "  Changed B to long jump " << *MI);

  return true;
}

/// fixupConditionalBr - Fix up a conditional branch whose destination is too
/// far away to fit in its displacement field. It is converted to an inverse
/// conditional branch + an unconditional branch to the destination.
bool
ARMConstantIslands::fixupConditionalBr(ImmBranch &Br) {
  MachineInstr *MI = Br.MI;
  MachineBasicBlock *DestBB = MI->getOperand(0).getMBB();

  // Add an unconditional branch to the destination and invert the branch
  // condition to jump over it:
  // blt L1
  // =>
  // bge L2
  // b   L1
  // L2:
  ARMCC::CondCodes CC = (ARMCC::CondCodes)MI->getOperand(1).getImm();
  CC = ARMCC::getOppositeCondition(CC);
  unsigned CCReg = MI->getOperand(2).getReg();

  // If the branch is at the end of its MBB and that has a fall-through block,
  // direct the updated conditional branch to the fall-through block. Otherwise,
  // split the MBB before the next instruction.
  MachineBasicBlock *MBB = MI->getParent();
  MachineInstr *BMI = &MBB->back();
  bool NeedSplit = (BMI != MI) || !BBHasFallthrough(MBB);

  ++NumCBrFixed;
  if (BMI != MI) {
    if (llvm::next(MachineBasicBlock::iterator(MI)) == prior(MBB->end()) &&
        BMI->getOpcode() == Br.UncondBr) {
      // Last MI in the BB is an unconditional branch. Can we simply invert the
      // condition and swap destinations:
      // beq L1
      // b   L2
      // =>
      // bne L2
      // b   L1
      MachineBasicBlock *NewDest = BMI->getOperand(0).getMBB();
      if (isBBInRange(MI, NewDest, Br.MaxDisp)) {
        DEBUG(dbgs() << "  Invert Bcc condition and swap its destination with "
                     << *BMI);
        BMI->getOperand(0).setMBB(DestBB);
        MI->getOperand(0).setMBB(NewDest);
        MI->getOperand(1).setImm(CC);
        return true;
      }
    }
  }

  if (NeedSplit) {
    splitBlockBeforeInstr(MI);
    // No need for the branch to the next block. We're adding an unconditional
    // branch to the destination.
    int delta = TII->GetInstSizeInBytes(&MBB->back());
    BBInfo[MBB->getNumber()].Size -= delta;
    MBB->back().eraseFromParent();
    // BBInfo[SplitBB].Offset is wrong temporarily, fixed below
  }
  MachineBasicBlock *NextBB = llvm::next(MachineFunction::iterator(MBB));

  DEBUG(dbgs() << "  Insert B to BB#" << DestBB->getNumber()
               << " also invert condition and change dest. to BB#"
               << NextBB->getNumber() << "\n");

  // Insert a new conditional branch and a new unconditional branch.
  // Also update the ImmBranch as well as adding a new entry for the new branch.
  BuildMI(MBB, DebugLoc(), TII->get(MI->getOpcode()))
    .addMBB(NextBB).addImm(CC).addReg(CCReg);
  Br.MI = &MBB->back();
  BBInfo[MBB->getNumber()].Size += TII->GetInstSizeInBytes(&MBB->back());
  if (isThumb)
    BuildMI(MBB, DebugLoc(), TII->get(Br.UncondBr)).addMBB(DestBB)
            .addImm(ARMCC::AL).addReg(0);
  else
    BuildMI(MBB, DebugLoc(), TII->get(Br.UncondBr)).addMBB(DestBB);
  BBInfo[MBB->getNumber()].Size += TII->GetInstSizeInBytes(&MBB->back());
  unsigned MaxDisp = getUnconditionalBrDisp(Br.UncondBr);
  ImmBranches.push_back(ImmBranch(&MBB->back(), MaxDisp, false, Br.UncondBr));

  // Remove the old conditional branch.  It may or may not still be in MBB.
  BBInfo[MI->getParent()->getNumber()].Size -= TII->GetInstSizeInBytes(MI);
  MI->eraseFromParent();
  adjustBBOffsetsAfter(MBB);
  return true;
}

/// undoLRSpillRestore - Remove Thumb push / pop instructions that only spills
/// LR / restores LR to pc. FIXME: This is done here because it's only possible
/// to do this if tBfar is not used.
bool ARMConstantIslands::undoLRSpillRestore() {
  bool MadeChange = false;
  for (unsigned i = 0, e = PushPopMIs.size(); i != e; ++i) {
    MachineInstr *MI = PushPopMIs[i];
    // First two operands are predicates.
    if (MI->getOpcode() == ARM::tPOP_RET &&
        MI->getOperand(2).getReg() == ARM::PC &&
        MI->getNumExplicitOperands() == 3) {
      // Create the new insn and copy the predicate from the old.
      BuildMI(MI->getParent(), MI->getDebugLoc(), TII->get(ARM::tBX_RET))
        .addOperand(MI->getOperand(0))
        .addOperand(MI->getOperand(1));
      MI->eraseFromParent();
      MadeChange = true;
    }
  }
  return MadeChange;
}

// mayOptimizeThumb2Instruction - Returns true if optimizeThumb2Instructions
// below may shrink MI.
bool
ARMConstantIslands::mayOptimizeThumb2Instruction(const MachineInstr *MI) const {
  switch(MI->getOpcode()) {
    // optimizeThumb2Instructions.
    case ARM::t2LEApcrel:
    case ARM::t2LDRpci:
    // optimizeThumb2Branches.
    case ARM::t2B:
    case ARM::t2Bcc:
    case ARM::tBcc:
    // optimizeThumb2JumpTables.
    case ARM::t2BR_JT:
      return true;
  }
  return false;
}

bool ARMConstantIslands::optimizeThumb2Instructions() {
  bool MadeChange = false;

  // Shrink ADR and LDR from constantpool.
  for (unsigned i = 0, e = CPUsers.size(); i != e; ++i) {
    CPUser &U = CPUsers[i];
    unsigned Opcode = U.MI->getOpcode();
    unsigned NewOpc = 0;
    unsigned Scale = 1;
    unsigned Bits = 0;
    switch (Opcode) {
    default: break;
    case ARM::t2LEApcrel:
      if (isARMLowRegister(U.MI->getOperand(0).getReg())) {
        NewOpc = ARM::tLEApcrel;
        Bits = 8;
        Scale = 4;
      }
      break;
    case ARM::t2LDRpci:
      if (isARMLowRegister(U.MI->getOperand(0).getReg())) {
        NewOpc = ARM::tLDRpci;
        Bits = 8;
        Scale = 4;
      }
      break;
    }

    if (!NewOpc)
      continue;

    unsigned UserOffset = getUserOffset(U);
    unsigned MaxOffs = ((1 << Bits) - 1) * Scale;

    // Be conservative with inline asm.
    if (!U.KnownAlignment)
      MaxOffs -= 2;

    // FIXME: Check if offset is multiple of scale if scale is not 4.
    if (isCPEntryInRange(U.MI, UserOffset, U.CPEMI, MaxOffs, false, true)) {
      DEBUG(dbgs() << "Shrink: " << *U.MI);
      U.MI->setDesc(TII->get(NewOpc));
      MachineBasicBlock *MBB = U.MI->getParent();
      BBInfo[MBB->getNumber()].Size -= 2;
      adjustBBOffsetsAfter(MBB);
      ++NumT2CPShrunk;
      MadeChange = true;
    }
  }

  MadeChange |= optimizeThumb2Branches();
  MadeChange |= optimizeThumb2JumpTables();
  return MadeChange;
}

bool ARMConstantIslands::optimizeThumb2Branches() {
  bool MadeChange = false;

  for (unsigned i = 0, e = ImmBranches.size(); i != e; ++i) {
    ImmBranch &Br = ImmBranches[i];
    unsigned Opcode = Br.MI->getOpcode();
    unsigned NewOpc = 0;
    unsigned Scale = 1;
    unsigned Bits = 0;
    switch (Opcode) {
    default: break;
    case ARM::t2B:
      NewOpc = ARM::tB;
      Bits = 11;
      Scale = 2;
      break;
    case ARM::t2Bcc: {
      NewOpc = ARM::tBcc;
      Bits = 8;
      Scale = 2;
      break;
    }
    }
    if (NewOpc) {
      unsigned MaxOffs = ((1 << (Bits-1))-1) * Scale;
      MachineBasicBlock *DestBB = Br.MI->getOperand(0).getMBB();
      if (isBBInRange(Br.MI, DestBB, MaxOffs)) {
        DEBUG(dbgs() << "Shrink branch: " << *Br.MI);
        Br.MI->setDesc(TII->get(NewOpc));
        MachineBasicBlock *MBB = Br.MI->getParent();
        BBInfo[MBB->getNumber()].Size -= 2;
        adjustBBOffsetsAfter(MBB);
        ++NumT2BrShrunk;
        MadeChange = true;
      }
    }

    Opcode = Br.MI->getOpcode();
    if (Opcode != ARM::tBcc)
      continue;

    // If the conditional branch doesn't kill CPSR, then CPSR can be liveout
    // so this transformation is not safe.
    if (!Br.MI->killsRegister(ARM::CPSR))
      continue;

    NewOpc = 0;
    unsigned PredReg = 0;
    ARMCC::CondCodes Pred = getInstrPredicate(Br.MI, PredReg);
    if (Pred == ARMCC::EQ)
      NewOpc = ARM::tCBZ;
    else if (Pred == ARMCC::NE)
      NewOpc = ARM::tCBNZ;
    if (!NewOpc)
      continue;
    MachineBasicBlock *DestBB = Br.MI->getOperand(0).getMBB();
    // Check if the distance is within 126. Subtract starting offset by 2
    // because the cmp will be eliminated.
    unsigned BrOffset = getOffsetOf(Br.MI) + 4 - 2;
    unsigned DestOffset = BBInfo[DestBB->getNumber()].Offset;
    if (BrOffset < DestOffset && (DestOffset - BrOffset) <= 126) {
      MachineBasicBlock::iterator CmpMI = Br.MI;
      if (CmpMI != Br.MI->getParent()->begin()) {
        --CmpMI;
        if (CmpMI->getOpcode() == ARM::tCMPi8) {
          unsigned Reg = CmpMI->getOperand(0).getReg();
          Pred = getInstrPredicate(CmpMI, PredReg);
          if (Pred == ARMCC::AL &&
              CmpMI->getOperand(1).getImm() == 0 &&
              isARMLowRegister(Reg)) {
            MachineBasicBlock *MBB = Br.MI->getParent();
            DEBUG(dbgs() << "Fold: " << *CmpMI << " and: " << *Br.MI);
            MachineInstr *NewBR =
              BuildMI(*MBB, CmpMI, Br.MI->getDebugLoc(), TII->get(NewOpc))
              .addReg(Reg).addMBB(DestBB,Br.MI->getOperand(0).getTargetFlags());
            CmpMI->eraseFromParent();
            Br.MI->eraseFromParent();
            Br.MI = NewBR;
            BBInfo[MBB->getNumber()].Size -= 2;
            adjustBBOffsetsAfter(MBB);
            ++NumCBZ;
            MadeChange = true;
          }
        }
      }
    }
  }

  return MadeChange;
}

/// optimizeThumb2JumpTables - Use tbb / tbh instructions to generate smaller
/// jumptables when it's possible.
bool ARMConstantIslands::optimizeThumb2JumpTables() {
  bool MadeChange = false;

  // FIXME: After the tables are shrunk, can we get rid some of the
  // constantpool tables?
  MachineJumpTableInfo *MJTI = MF->getJumpTableInfo();
  if (MJTI == 0) return false;

  const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables();
  for (unsigned i = 0, e = T2JumpTables.size(); i != e; ++i) {
    MachineInstr *MI = T2JumpTables[i];
    const MCInstrDesc &MCID = MI->getDesc();
    unsigned NumOps = MCID.getNumOperands();
    unsigned JTOpIdx = NumOps - (MI->isPredicable() ? 3 : 2);
    MachineOperand JTOP = MI->getOperand(JTOpIdx);
    unsigned JTI = JTOP.getIndex();
    assert(JTI < JT.size());

    bool ByteOk = true;
    bool HalfWordOk = true;
    unsigned JTOffset = getOffsetOf(MI) + 4;
    const std::vector<MachineBasicBlock*> &JTBBs = JT[JTI].MBBs;
    for (unsigned j = 0, ee = JTBBs.size(); j != ee; ++j) {
      MachineBasicBlock *MBB = JTBBs[j];
      unsigned DstOffset = BBInfo[MBB->getNumber()].Offset;
      // Negative offset is not ok. FIXME: We should change BB layout to make
      // sure all the branches are forward.
      if (ByteOk && (DstOffset - JTOffset) > ((1<<8)-1)*2)
        ByteOk = false;
      unsigned TBHLimit = ((1<<16)-1)*2;
      if (HalfWordOk && (DstOffset - JTOffset) > TBHLimit)
        HalfWordOk = false;
      if (!ByteOk && !HalfWordOk)
        break;
    }

    if (ByteOk || HalfWordOk) {
      MachineBasicBlock *MBB = MI->getParent();
      unsigned BaseReg = MI->getOperand(0).getReg();
      bool BaseRegKill = MI->getOperand(0).isKill();
      if (!BaseRegKill)
        continue;
      unsigned IdxReg = MI->getOperand(1).getReg();
      bool IdxRegKill = MI->getOperand(1).isKill();

      // Scan backwards to find the instruction that defines the base
      // register. Due to post-RA scheduling, we can't count on it
      // immediately preceding the branch instruction.
      MachineBasicBlock::iterator PrevI = MI;
      MachineBasicBlock::iterator B = MBB->begin();
      while (PrevI != B && !PrevI->definesRegister(BaseReg))
        --PrevI;

      // If for some reason we didn't find it, we can't do anything, so
      // just skip this one.
      if (!PrevI->definesRegister(BaseReg))
        continue;

      MachineInstr *AddrMI = PrevI;
      bool OptOk = true;
      // Examine the instruction that calculates the jumptable entry address.
      // Make sure it only defines the base register and kills any uses
      // other than the index register.
      for (unsigned k = 0, eee = AddrMI->getNumOperands(); k != eee; ++k) {
        const MachineOperand &MO = AddrMI->getOperand(k);
        if (!MO.isReg() || !MO.getReg())
          continue;
        if (MO.isDef() && MO.getReg() != BaseReg) {
          OptOk = false;
          break;
        }
        if (MO.isUse() && !MO.isKill() && MO.getReg() != IdxReg) {
          OptOk = false;
          break;
        }
      }
      if (!OptOk)
        continue;

      // Now scan back again to find the tLEApcrel or t2LEApcrelJT instruction
      // that gave us the initial base register definition.
      for (--PrevI; PrevI != B && !PrevI->definesRegister(BaseReg); --PrevI)
        ;

      // The instruction should be a tLEApcrel or t2LEApcrelJT; we want
      // to delete it as well.
      MachineInstr *LeaMI = PrevI;
      if ((LeaMI->getOpcode() != ARM::tLEApcrelJT &&
           LeaMI->getOpcode() != ARM::t2LEApcrelJT) ||
          LeaMI->getOperand(0).getReg() != BaseReg)
        OptOk = false;

      if (!OptOk)
        continue;

      DEBUG(dbgs() << "Shrink JT: " << *MI << "     addr: " << *AddrMI
                   << "      lea: " << *LeaMI);
      unsigned Opc = ByteOk ? ARM::t2TBB_JT : ARM::t2TBH_JT;
      MachineInstr *NewJTMI = BuildMI(MBB, MI->getDebugLoc(), TII->get(Opc))
        .addReg(IdxReg, getKillRegState(IdxRegKill))
        .addJumpTableIndex(JTI, JTOP.getTargetFlags())
        .addImm(MI->getOperand(JTOpIdx+1).getImm());
      DEBUG(dbgs() << "BB#" << MBB->getNumber() << ": " << *NewJTMI);
      // FIXME: Insert an "ALIGN" instruction to ensure the next instruction
      // is 2-byte aligned. For now, asm printer will fix it up.
      unsigned NewSize = TII->GetInstSizeInBytes(NewJTMI);
      unsigned OrigSize = TII->GetInstSizeInBytes(AddrMI);
      OrigSize += TII->GetInstSizeInBytes(LeaMI);
      OrigSize += TII->GetInstSizeInBytes(MI);

      AddrMI->eraseFromParent();
      LeaMI->eraseFromParent();
      MI->eraseFromParent();

      int delta = OrigSize - NewSize;
      BBInfo[MBB->getNumber()].Size -= delta;
      adjustBBOffsetsAfter(MBB);

      ++NumTBs;
      MadeChange = true;
    }
  }

  return MadeChange;
}

/// reorderThumb2JumpTables - Adjust the function's block layout to ensure that
/// jump tables always branch forwards, since that's what tbb and tbh need.
bool ARMConstantIslands::reorderThumb2JumpTables() {
  bool MadeChange = false;

  MachineJumpTableInfo *MJTI = MF->getJumpTableInfo();
  if (MJTI == 0) return false;

  const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables();
  for (unsigned i = 0, e = T2JumpTables.size(); i != e; ++i) {
    MachineInstr *MI = T2JumpTables[i];
    const MCInstrDesc &MCID = MI->getDesc();
    unsigned NumOps = MCID.getNumOperands();
    unsigned JTOpIdx = NumOps - (MI->isPredicable() ? 3 : 2);
    MachineOperand JTOP = MI->getOperand(JTOpIdx);
    unsigned JTI = JTOP.getIndex();
    assert(JTI < JT.size());

    // We prefer if target blocks for the jump table come after the jump
    // instruction so we can use TB[BH]. Loop through the target blocks
    // and try to adjust them such that that's true.
    int JTNumber = MI->getParent()->getNumber();
    const std::vector<MachineBasicBlock*> &JTBBs = JT[JTI].MBBs;
    for (unsigned j = 0, ee = JTBBs.size(); j != ee; ++j) {
      MachineBasicBlock *MBB = JTBBs[j];
      int DTNumber = MBB->getNumber();

      if (DTNumber < JTNumber) {
        // The destination precedes the switch. Try to move the block forward
        // so we have a positive offset.
        MachineBasicBlock *NewBB =
          adjustJTTargetBlockForward(MBB, MI->getParent());
        if (NewBB)
          MJTI->ReplaceMBBInJumpTable(JTI, JTBBs[j], NewBB);
        MadeChange = true;
      }
    }
  }

  return MadeChange;
}

MachineBasicBlock *ARMConstantIslands::
adjustJTTargetBlockForward(MachineBasicBlock *BB, MachineBasicBlock *JTBB) {
  // If the destination block is terminated by an unconditional branch,
  // try to move it; otherwise, create a new block following the jump
  // table that branches back to the actual target. This is a very simple
  // heuristic. FIXME: We can definitely improve it.
  MachineBasicBlock *TBB = 0, *FBB = 0;
  SmallVector<MachineOperand, 4> Cond;
  SmallVector<MachineOperand, 4> CondPrior;
  MachineFunction::iterator BBi = BB;
  MachineFunction::iterator OldPrior = prior(BBi);

  // If the block terminator isn't analyzable, don't try to move the block
  bool B = TII->AnalyzeBranch(*BB, TBB, FBB, Cond);

  // If the block ends in an unconditional branch, move it. The prior block
  // has to have an analyzable terminator for us to move this one. Be paranoid
  // and make sure we're not trying to move the entry block of the function.
  if (!B && Cond.empty() && BB != MF->begin() &&
      !TII->AnalyzeBranch(*OldPrior, TBB, FBB, CondPrior)) {
    BB->moveAfter(JTBB);
    OldPrior->updateTerminator();
    BB->updateTerminator();
    // Update numbering to account for the block being moved.
    MF->RenumberBlocks();
    ++NumJTMoved;
    return NULL;
  }

  // Create a new MBB for the code after the jump BB.
  MachineBasicBlock *NewBB =
    MF->CreateMachineBasicBlock(JTBB->getBasicBlock());
  MachineFunction::iterator MBBI = JTBB; ++MBBI;
  MF->insert(MBBI, NewBB);

  // Add an unconditional branch from NewBB to BB.
  // There doesn't seem to be meaningful DebugInfo available; this doesn't
  // correspond directly to anything in the source.
  assert (isThumb2 && "Adjusting for TB[BH] but not in Thumb2?");
  BuildMI(NewBB, DebugLoc(), TII->get(ARM::t2B)).addMBB(BB)
          .addImm(ARMCC::AL).addReg(0);

  // Update internal data structures to account for the newly inserted MBB.
  MF->RenumberBlocks(NewBB);

  // Update the CFG.
  NewBB->addSuccessor(BB);
  JTBB->removeSuccessor(BB);
  JTBB->addSuccessor(NewBB);

  ++NumJTInserted;
  return NewBB;
}