summaryrefslogtreecommitdiff
path: root/xdelta3-djw.h
blob: d6e28811b27127658b55dec0294352c690df7a56 (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
/* xdelta 3 - delta compression tools and library
 * Copyright (C) 2002, 2006, 2007.  Joshua P. MacDonald
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

/* TODO: This code needs a thorough round of commenting.  There is
 * some slop in the declaration of arrays, which are maybe one element
 * larger than they need to be and comments would help clear it up. */

#ifndef _XDELTA3_DJW_H_
#define _XDELTA3_DJW_H_

/* The following people deserve much credit for the algorithms and
 * techniques contained in this file:

 Julian Seward
 Bzip2 sources, implementation of the multi-table Huffman technique.

 Jean-loup Gailly and Mark Adler and L. Peter Deutsch
 Zlib source code, RFC 1951

 Daniel S. Hirschberg and Debra A. LeLewer
 "Efficient Decoding of Prefix Codes"
 Communications of the ACM, April 1990 33(4).

 David J. Wheeler
 Program bred3.c, bexp3 and accompanying documents bred3.ps, huff.ps.
 This contains the idea behind the multi-table Huffman and 1-2 coding
 techniques.
 ftp://ftp.cl.cam.ac.uk/users/djw3/

*/

/* OPT: during the multi-table iteration, pick the worst-overall
 * performing table and replace it with exactly the frequencies of the
 * worst-overall performing sector or N-worst performing sectors. */

/* REF: See xdfs-0.222 and xdfs-0.226 for some old experiments with
 * the Bzip prefix coding strategy.  xdfs-0.256 contains the last of
 * the other-format tests, including RFC1950 and the RFC1950+MTF
 * tests. */

#define DJW_MAX_CODELEN      20U /* Maximum length of an alphabet code. */

/* Code lengths are themselves code-length encoded, so the total number of
 * codes is: [RUN_0, RUN_1, 1-DJW_MAX_CODELEN] */
#define DJW_TOTAL_CODES      (DJW_MAX_CODELEN+2)

#define RUN_0                0U /* Symbols used in MTF+1/2 coding. */
#define RUN_1                1U

/* Number of code lengths always encoded (djw_encode_basic array) */
#define DJW_BASIC_CODES      5U  
#define DJW_RUN_CODES        2U  /* Number of run codes */

/* Offset of extra codes */
#define DJW_EXTRA_12OFFSET   (DJW_BASIC_CODES + DJW_RUN_CODES)

/* Number of optionally encoded code lengths (djw_encode_extra array) */
#define DJW_EXTRA_CODES      15U

/* Number of bits to code [0-DJW_EXTRA_CODES] */
#define DJW_EXTRA_CODE_BITS  4U  

#define DJW_MAX_GROUPS       8U  /* Max number of group coding tables */
#define DJW_GROUP_BITS       3U  /* Number of bits to code [1-DJW_MAX_GROUPS] */

#define DJW_SECTORSZ_MULT     5U  /* Multiplier for encoded sectorsz */
#define DJW_SECTORSZ_BITS     5U  /* Number of bits to code group size */
#define DJW_SECTORSZ_MAX      ((1U << DJW_SECTORSZ_BITS) * DJW_SECTORSZ_MULT)

/* Maximum number of iterations to find group tables. */
#define DJW_MAX_ITER         6U
/* Minimum number of bits an iteration must reduce coding by. */
#define DJW_MIN_IMPROVEMENT  20U 

/* Maximum code length of a prefix code length */
#define DJW_MAX_CLCLEN       15U

/* Number of bits to code [0-DJW_MAX_CLCLEN] */
#define DJW_CLCLEN_BITS      4U  

#define DJW_MAX_GBCLEN       7U  /* Maximum code length of a group selector */

/* Number of bits to code [0-DJW_MAX_GBCLEN]
 * TODO: Actually, should never have zero code lengths here, or else a group
 * went unused.  Write a test for this: if a group goes unused, eliminate
 * it? */
#define DJW_GBCLEN_BITS      3U

/* It has to save at least this many bits... */
#define EFFICIENCY_BITS      16U

typedef struct _djw_stream   djw_stream;
typedef struct _djw_heapen   djw_heapen;
typedef struct _djw_prefix   djw_prefix;
typedef uint32_t             djw_weight;

struct _djw_heapen
{
  uint32_t depth;
  uint32_t freq;
  uint32_t parent;
};

struct _djw_prefix
{
  usize_t   scount;
  uint8_t *symbol;
  usize_t   mcount;
  uint8_t *mtfsym;
  uint8_t *repcnt;
};

struct _djw_stream
{
  int unused;
};

/* Each Huffman table consists of 256 "code length" (CLEN) codes,
 * which are themselves Huffman coded after eliminating repeats and
 * move-to-front coding.  The prefix consists of all the CLEN codes in
 * djw_encode_basic plus a 4-bit value stating how many of the
 * djw_encode_extra codes are actually coded (the rest are presumed
 * zero, or unused CLEN codes).
 *
 * These values of these two arrays were arrived at by studying the
 * distribution of min and max clen over a collection of DATA, INST,
 * and ADDR inputs.  The goal is to specify the order of
 * djw_extra_codes that is most likely to minimize the number of extra
 * codes that must be encoded.
 *
 * Results: 158896 sections were counted by compressing files (window
 * size 512K) listed with: `find / -type f ( -user jmacd -o -perm +444
 * )`
 *
 * The distribution of CLEN codes for each efficient invocation of the
 * secondary compressor (taking the best number of groups/sector size)
 * was recorded.  Then we look at the distribution of min and max clen
 * values, counting the number of times the value C_low is less than
 * the min and C_high is greater than the max.  Values >= C_high and
 * <= C_low will not have their lengths coded.  The results are sorted
 * and the least likely 15 are placed into the djw_encode_extra[]
 * array in order.  These values are used as the initial MTF ordering.

 clow[1] = 155119
 clow[2] = 140325
 clow[3] = 84072
 ---
 clow[4] = 7225
 clow[5] = 1093
 clow[6] = 215
 ---
 chigh[4] = 1
 chigh[5] = 30
 chigh[6] = 218
 chigh[7] = 2060
 chigh[8] = 13271
 ---
 chigh[9] = 39463
 chigh[10] = 77360
 chigh[11] = 118298
 chigh[12] = 141360
 chigh[13] = 154086
 chigh[14] = 157967
 chigh[15] = 158603
 chigh[16] = 158864
 chigh[17] = 158893
 chigh[18] = 158895
 chigh[19] = 158896
 chigh[20] = 158896

*/

static const uint8_t djw_encode_12extra[DJW_EXTRA_CODES] =
  {
    9, 10, 3, 11, 2, 12, 13, 1, 14, 15, 16, 17, 18, 19, 20,
  };

static const uint8_t djw_encode_12basic[DJW_BASIC_CODES] =
  {
    4, 5, 6, 7, 8,
  };

/*********************************************************************/
/*                              DECLS                                */
/*********************************************************************/

static djw_stream*     djw_alloc           (xd3_stream *stream);
static int             djw_init            (xd3_stream *stream, 
					    djw_stream *h,
					    int is_encode);
static void            djw_destroy         (xd3_stream *stream,
					    djw_stream *h);

#if XD3_ENCODER
static int             xd3_encode_huff     (xd3_stream   *stream,
					    djw_stream  *sec_stream,
					    xd3_output   *input,
					    xd3_output   *output,
					    xd3_sec_cfg  *cfg);
#endif

static int             xd3_decode_huff     (xd3_stream     *stream,
					    djw_stream    *sec_stream,
					    const uint8_t **input,
					    const uint8_t  *const input_end,
					    uint8_t       **output,
					    const uint8_t  *const output_end);

/*********************************************************************/
/*                             HUFFMAN                               */
/*********************************************************************/

static djw_stream*
djw_alloc (xd3_stream *stream)
{
  return (djw_stream*) xd3_alloc (stream, sizeof (djw_stream), 1);
}

static int
djw_init (xd3_stream *stream, djw_stream *h, int is_encode)
{
  /* Fields are initialized prior to use. */
  return 0;
}

static void
djw_destroy (xd3_stream *stream,
	     djw_stream *h)
{
  xd3_free (stream, h);
}


/*********************************************************************/
/*                               HEAP                                */
/*********************************************************************/

static inline int
heap_less (const djw_heapen *a, const djw_heapen *b)
{
  return a->freq   < b->freq ||
    (a->freq  == b->freq &&
     a->depth  < b->depth);
}

static inline void
heap_insert (usize_t *heap, const djw_heapen *ents, usize_t p, const usize_t e)
{
  /* Insert ents[e] into next slot heap[p] */
  usize_t pp = p/2; /* P's parent */

  while (heap_less (& ents[e], & ents[heap[pp]]))
    {
      heap[p] = heap[pp];
      p  = pp;
      pp = p/2;
    }

  heap[p] = e;
}

static inline djw_heapen*
heap_extract (usize_t *heap, const djw_heapen *ents, usize_t heap_last)
{
  usize_t smallest = heap[1];
  usize_t p, pc, t;

  /* Caller decrements heap_last, so heap_last+1 is the replacement elt. */
  heap[1] = heap[heap_last+1];

  /* Re-heapify */
  for (p = 1; ; p = pc)
    {
      pc = p*2;

      /* Reached bottom of heap */
      if (pc > heap_last) { break; }

      /* See if second child is smaller. */
      if (pc < heap_last && heap_less (& ents[heap[pc+1]], & ents[heap[pc]]))
	{
	  pc += 1;
	}

      /* If pc is not smaller than p, heap property re-established. */
      if (! heap_less (& ents[heap[pc]], & ents[heap[p]])) { break; }

      t = heap[pc];
      heap[pc] = heap[p];
      heap[p] = t;
    }

  return (djw_heapen*) & ents[smallest];
}

#if XD3_DEBUG
static void
heap_check (usize_t *heap, djw_heapen *ents, usize_t heap_last)
{
  usize_t i;
  for (i = 1; i <= heap_last; i += 1)
    {
      /* Heap property: child not less than parent */
      XD3_ASSERT (! heap_less (& ents[heap[i]], & ents[heap[i/2]]));

      IF_DEBUG2 (DP(RINT "heap[%d] = %u\n", i, ents[heap[i]].freq));
    }
}
#endif

/*********************************************************************/
/*                             MTF, 1/2                              */
/*********************************************************************/

static inline usize_t
djw_update_mtf (uint8_t *mtf, usize_t mtf_i)
{
  int k;
  usize_t sym = mtf[mtf_i];

  for (k = mtf_i; k != 0; k -= 1) { mtf[k] = mtf[k-1]; }

  mtf[0] = sym;
  return sym;
}

static inline void
djw_update_1_2 (int *mtf_run, usize_t *mtf_i,
		uint8_t *mtfsym, djw_weight *freq)
{
  int code;
  
  do
    {
      /* Offset by 1, since any number of RUN_ symbols implies run>0... */
      *mtf_run -= 1;

      code = (*mtf_run & 1) ? RUN_1 : RUN_0;

      mtfsym[(*mtf_i)++] = code;
      freq[code] += 1;
      *mtf_run >>= 1;
    }
  while (*mtf_run >= 1);

  *mtf_run = 0;
}

static void
djw_init_clen_mtf_1_2 (uint8_t *clmtf)
{
  usize_t i, cl_i = 0;

  clmtf[cl_i++] = 0;
  for (i = 0; i < DJW_BASIC_CODES; i += 1)
    {
      clmtf[cl_i++] = djw_encode_12basic[i];
    }
  for (i = 0; i < DJW_EXTRA_CODES; i += 1)
    {
      clmtf[cl_i++] = djw_encode_12extra[i];
    }
}

/*********************************************************************/
/*                           PREFIX CODES                            */
/*********************************************************************/
#if XD3_ENCODER
static usize_t
djw_build_prefix (const djw_weight *freq, uint8_t *clen, usize_t asize, usize_t maxlen)
{
  /* Heap with 0th entry unused, prefix tree with up to ALPHABET_SIZE-1
   * internal nodes, never more than ALPHABET_SIZE entries actually in the
   * heap (minimum weight subtrees during prefix construction).  First
   * ALPHABET_SIZE entries are the actual symbols, next ALPHABET_SIZE-1 are
   * internal nodes. */
  djw_heapen ents[ALPHABET_SIZE * 2];
  usize_t heap[ALPHABET_SIZE + 1];

  usize_t heap_last; /* Index of the last _valid_ heap entry. */
  usize_t ents_size; /* Number of entries, including 0th fake entry */
  usize_t  overflow;  /* Number of code lengths that overflow */
  uint32_t total_bits;
  usize_t i;

  IF_DEBUG (uint32_t first_bits = 0);

  /* Insert real symbol frequences. */
  for (i = 0; i < asize; i += 1)
    {
      ents[i+1].freq = freq[i];
      IF_DEBUG2 (DP(RINT "ents[%d] = freq[%d] = %d\n",
			i+1, i, freq[i]));
    }

 again:

  /* The loop is re-entered each time an overflow occurs.  Re-initialize... */
  heap_last = 0;
  ents_size = 1;
  overflow  = 0;
  total_bits = 0;

  /* 0th entry terminates the while loop in heap_insert (it's the parent of
   * the smallest element, always less-than) */
  heap[0] = 0;
  ents[0].depth = 0;
  ents[0].freq  = 0;

  /* Initial heap. */
  for (i = 0; i < asize; i += 1, ents_size += 1)
    {
      ents[ents_size].depth  = 0;
      ents[ents_size].parent = 0;

      if (ents[ents_size].freq != 0)
	{
	  heap_insert (heap, ents, ++heap_last, ents_size);
	}
    }

  IF_DEBUG (heap_check (heap, ents, heap_last));

  /* Must be at least one symbol, or else we can't get here. */
  XD3_ASSERT (heap_last != 0);

  /* If there is only one symbol, fake a second to prevent zero-length
   * codes. */
  if (heap_last == 1)
    {
      /* Pick either the first or last symbol. */
      int s = freq[0] ? asize-1 : 0;
      ents[s+1].freq = 1;
      goto again;
    }

  /* Build prefix tree. */
  while (heap_last > 1)
    {
      djw_heapen *h1 = heap_extract (heap, ents, --heap_last);
      djw_heapen *h2 = heap_extract (heap, ents, --heap_last);

      ents[ents_size].freq   = h1->freq + h2->freq;
      ents[ents_size].depth  = 1 + max (h1->depth, h2->depth);
      ents[ents_size].parent = 0;

      h1->parent = h2->parent = ents_size;

      heap_insert (heap, ents, ++heap_last, ents_size++);
    }

  IF_DEBUG (heap_check (heap, ents, heap_last));

  /* Now compute prefix code lengths, counting parents. */
  for (i = 1; i < asize+1; i += 1)
    {
      usize_t b = 0;

      if (ents[i].freq != 0)
	{
	  usize_t p = i;

	  while ((p = ents[p].parent) != 0) { b += 1; }

	  if (b > maxlen) { overflow = 1; }

	  total_bits += b * freq[i-1];
	}

      /* clen is 0-origin, unlike ents. */
      IF_DEBUG2 (DP(RINT "clen[%d] = %d\n", i-1, b));
      clen[i-1] = b;
    }

  IF_DEBUG (if (first_bits == 0) first_bits = total_bits);

  if (! overflow)
    {
      IF_DEBUG2 (if (first_bits != total_bits)
      {
	DP(RINT "code length overflow changed %u bits\n",
	   (usize_t)(total_bits - first_bits));
      });
      return total_bits;
    }

  /* OPT: There is a non-looping way to fix overflow shown in zlib, but this
   * is easier (for now), as done in bzip2. */
  for (i = 1; i < asize+1; i += 1)
    {
      ents[i].freq = ents[i].freq / 2 + 1;
    }

  goto again;
}

static void
djw_build_codes (usize_t *codes, const uint8_t *clen, usize_t asize, usize_t abs_max)
{
  usize_t i, l;
  usize_t min_clen = DJW_MAX_CODELEN;
  usize_t max_clen = 0;
  usize_t code = 0;

  /* Find the min and max code length */
  for (i = 0; i < asize; i += 1)
    {
      if (clen[i] > 0 && clen[i] < min_clen)
	{
	  min_clen = clen[i];
	}

      max_clen = max (max_clen, (usize_t) clen[i]);
    }

  XD3_ASSERT (max_clen <= abs_max);

  /* Generate a code for each symbol with the appropriate length. */
  for (l = min_clen; l <= max_clen; l += 1)
    {
      for (i = 0; i < asize; i += 1)
	{
	  if (clen[i] == l)
	    {
	      codes[i] = code++;
	    } 
	}

      code <<= 1;
    }

  IF_DEBUG2 ({
      for (i = 0; i < asize; i += 1)
	{
	  DP(RINT "code[%d] = %u\n", i, codes[i]);
	}
    });
}

/*********************************************************************/
/*			      MOVE-TO-FRONT                          */
/*********************************************************************/
static void
djw_compute_mtf_1_2 (djw_prefix  *prefix,
		     uint8_t     *mtf,
		     djw_weight  *freq_out,
		     usize_t      nsym)
{
  size_t i, j, k;
  usize_t sym;
  usize_t size = prefix->scount;
  usize_t mtf_i = 0;
  int mtf_run = 0;

  /* This +2 is for the RUN_0, RUN_1 codes */
  memset (freq_out, 0, sizeof (freq_out[0]) * (nsym+2));

  for (i = 0; i < size; )
    {
      /* OPT: Bzip optimizes this algorithm a little by effectively checking
       * j==0 before the MTF update. */
      sym = prefix->symbol[i++];

      for (j = 0; mtf[j] != sym; j += 1) { }

      XD3_ASSERT (j <= nsym);

      for (k = j; k >= 1; k -= 1) { mtf[k] = mtf[k-1]; }

      mtf[0] = sym;

      if (j == 0)
	{
	  mtf_run += 1;
	  continue;
	}

      if (mtf_run > 0)
	{
	  djw_update_1_2 (& mtf_run, & mtf_i, prefix->mtfsym, freq_out);
	}

      /* Non-zero symbols are offset by RUN_1 */
      prefix->mtfsym[mtf_i++] = (uint8_t)(j+RUN_1);
      freq_out[j+RUN_1] += 1;
    }

  if (mtf_run > 0)
    {
      djw_update_1_2 (& mtf_run, & mtf_i, prefix->mtfsym, freq_out);
    }

  prefix->mcount = mtf_i;
}

/* Counts character frequencies of the input buffer, returns the size. */
static usize_t
djw_count_freqs (djw_weight *freq, xd3_output *input)
{
  xd3_output *in;
  usize_t size = 0;

  memset (freq, 0, sizeof (freq[0]) * ALPHABET_SIZE);

  for (in = input; in; in = in->next_page)
    {
      const uint8_t *p     = in->base;
      const uint8_t *p_max = p + in->next;

      size += in->next;

      do
	{
	  ++freq[*p];
	}
      while (++p < p_max);
    }

  IF_DEBUG2 ({int i;
  DP(RINT "freqs: ");
  for (i = 0; i < ALPHABET_SIZE; i += 1)
    {
      DP(RINT "%u ", freq[i]);
    }
  DP(RINT "\n");});

  return size;
}

static void
djw_compute_multi_prefix (usize_t     groups,
			  uint8_t     clen[DJW_MAX_GROUPS][ALPHABET_SIZE],
			  djw_prefix *prefix)
{
  usize_t gp, i;
      
  prefix->scount = ALPHABET_SIZE;
  memcpy (prefix->symbol, clen[0], ALPHABET_SIZE);

  for (gp = 1; gp < groups; gp += 1)
    {
      for (i = 0; i < ALPHABET_SIZE; i += 1)
	{
	  if (clen[gp][i] == 0)
	    {
	      continue;
	    }

	  prefix->symbol[prefix->scount++] = clen[gp][i];
	}
    }
}

static void
djw_compute_prefix_1_2 (djw_prefix *prefix, djw_weight *freq)
{
  /* This +1 is for the 0 code-length. */
  uint8_t clmtf[DJW_MAX_CODELEN+1];

  djw_init_clen_mtf_1_2 (clmtf);

  djw_compute_mtf_1_2 (prefix, clmtf, freq, DJW_MAX_CODELEN);
}

static int
djw_encode_prefix (xd3_stream   *stream,
		   xd3_output  **output,
		   bit_state    *bstate,
		   djw_prefix   *prefix)
{
  int ret;
  size_t i;
  usize_t num_to_encode;
  djw_weight clfreq[DJW_TOTAL_CODES];
  uint8_t    clclen[DJW_TOTAL_CODES];
  usize_t    clcode[DJW_TOTAL_CODES];

  /* Move-to-front encode prefix symbols, count frequencies */
  djw_compute_prefix_1_2 (prefix, clfreq);

  /* Compute codes */
  djw_build_prefix (clfreq, clclen, DJW_TOTAL_CODES, DJW_MAX_CLCLEN);
  djw_build_codes  (clcode, clclen, DJW_TOTAL_CODES, DJW_MAX_CLCLEN);

  /* Compute number of extra codes beyond basic ones for this template. */
  num_to_encode = DJW_TOTAL_CODES;
  while (num_to_encode > DJW_EXTRA_12OFFSET && clclen[num_to_encode-1] == 0)
    {
      num_to_encode -= 1;
    }
  XD3_ASSERT (num_to_encode - DJW_EXTRA_12OFFSET < (1 << DJW_EXTRA_CODE_BITS));

  /* Encode: # of extra codes */
  if ((ret = xd3_encode_bits (stream, output, bstate, DJW_EXTRA_CODE_BITS,
			      num_to_encode - DJW_EXTRA_12OFFSET)))
    {
      return ret;
    }

  /* Encode: MTF code lengths */
  for (i = 0; i < num_to_encode; i += 1)
    {
      if ((ret = xd3_encode_bits (stream, output, bstate,
				  DJW_CLCLEN_BITS, clclen[i])))
	{
	  return ret;
	}
    }

  /* Encode: CLEN code lengths */
  for (i = 0; i < prefix->mcount; i += 1)
    {
      usize_t mtf_sym = prefix->mtfsym[i];
      usize_t bits    = clclen[mtf_sym];
      usize_t code    = clcode[mtf_sym];

      if ((ret = xd3_encode_bits (stream, output, bstate, bits, code)))
	{
	  return ret;
	}
    }

  return 0;
}

static void
djw_compute_selector_1_2 (djw_prefix *prefix,
			  usize_t     groups,
			  djw_weight *gbest_freq)
{
  uint8_t grmtf[DJW_MAX_GROUPS];
  usize_t i;

  for (i = 0; i < groups; i += 1) { grmtf[i] = i; }

  djw_compute_mtf_1_2 (prefix, grmtf, gbest_freq, groups);
}

static int
xd3_encode_howmany_groups (xd3_stream *stream,
			   xd3_sec_cfg *cfg,
			   usize_t input_size,
			   usize_t *ret_groups,
			   usize_t *ret_sector_size)
{
  usize_t cfg_groups = 0;
  usize_t cfg_sector_size = 0;
  usize_t sugg_groups = 0;
  usize_t sugg_sector_size = 0;

  if (cfg->ngroups != 0)
    {
      if (cfg->ngroups > DJW_MAX_GROUPS)
	{
	  stream->msg = "invalid secondary encoder group number";
	  return XD3_INTERNAL;
	}

      cfg_groups = cfg->ngroups;
    }

  if (cfg->sector_size != 0)
    {
      if (cfg->sector_size < DJW_SECTORSZ_MULT ||
	  cfg->sector_size > DJW_SECTORSZ_MAX ||
	  (cfg->sector_size % DJW_SECTORSZ_MULT) != 0)
	{
	  stream->msg = "invalid secondary encoder sector size";
	  return XD3_INTERNAL;
	}

      cfg_sector_size = cfg->sector_size;
    }

  if (cfg_groups == 0 || cfg_sector_size == 0)
    {
      /* These values were found empirically using xdelta3-tune around version
       * xdfs-0.256. */
      switch (cfg->data_type)
	{
	case DATA_SECTION:
	  if      (input_size < 1000)   { sugg_groups = 1; sugg_sector_size = 0; }
	  else if (input_size < 4000)   { sugg_groups = 2; sugg_sector_size = 10; }
	  else if (input_size < 7000)   { sugg_groups = 3; sugg_sector_size = 10; }
	  else if (input_size < 10000)  { sugg_groups = 4; sugg_sector_size = 10; }
	  else if (input_size < 25000)  { sugg_groups = 5; sugg_sector_size = 10; }
	  else if (input_size < 50000)  { sugg_groups = 7; sugg_sector_size = 20; }
	  else if (input_size < 100000) { sugg_groups = 8; sugg_sector_size = 30; }
	  else                          { sugg_groups = 8; sugg_sector_size = 70; }
	  break;
	case INST_SECTION:
	  if      (input_size < 7000)   { sugg_groups = 1; sugg_sector_size = 0; }
	  else if (input_size < 10000)  { sugg_groups = 2; sugg_sector_size = 50; }
	  else if (input_size < 25000)  { sugg_groups = 3; sugg_sector_size = 50; }
	  else if (input_size < 50000)  { sugg_groups = 6; sugg_sector_size = 40; }
	  else if (input_size < 100000) { sugg_groups = 8; sugg_sector_size = 40; }
	  else                          { sugg_groups = 8; sugg_sector_size = 40; }
	  break;
	case ADDR_SECTION:
	  if      (input_size < 9000)   { sugg_groups = 1; sugg_sector_size = 0; }
	  else if (input_size < 25000)  { sugg_groups = 2; sugg_sector_size = 130; }
	  else if (input_size < 50000)  { sugg_groups = 3; sugg_sector_size = 130; }
	  else if (input_size < 100000) { sugg_groups = 5; sugg_sector_size = 130; }
	  else                          { sugg_groups = 7; sugg_sector_size = 130; }
	  break;
	}

      if (cfg_groups == 0)
	{
	  cfg_groups = sugg_groups;
	}

      if (cfg_sector_size == 0)
	{
	  cfg_sector_size = sugg_sector_size;
	}
    }

  if (cfg_groups != 1 && cfg_sector_size == 0)
    {
      switch (cfg->data_type)
	{
	case DATA_SECTION:
	  cfg_sector_size = 20;
	  break;
	case INST_SECTION:
	  cfg_sector_size = 50;
	  break;
	case ADDR_SECTION:
	  cfg_sector_size = 130;
	  break;
	}
    }

  (*ret_groups)     = cfg_groups;
  (*ret_sector_size) = cfg_sector_size;

  XD3_ASSERT (cfg_groups > 0 && cfg_groups <= DJW_MAX_GROUPS);
  XD3_ASSERT (cfg_groups == 1 ||
	      (cfg_sector_size >= DJW_SECTORSZ_MULT &&
	       cfg_sector_size <= DJW_SECTORSZ_MAX));

  return 0;
}

static int
xd3_encode_huff (xd3_stream   *stream,
		 djw_stream   *h,
		 xd3_output   *input,
		 xd3_output   *output,
		 xd3_sec_cfg  *cfg)
{
  int         ret;
  usize_t     groups, sector_size;
  bit_state   bstate = BIT_STATE_ENCODE_INIT;
  xd3_output *in;
  int         output_bits;
  usize_t     input_bits;
  usize_t     input_bytes;
  usize_t     initial_offset = output->next;
  djw_weight  real_freq[ALPHABET_SIZE];
  uint8_t    *gbest = NULL;
  uint8_t    *gbest_mtf = NULL;

  input_bytes = djw_count_freqs (real_freq, input);
  input_bits  = input_bytes * 8;

  XD3_ASSERT (input_bytes > 0);

  if ((ret = xd3_encode_howmany_groups (stream, cfg, input_bytes,
					& groups, & sector_size)))
    {
      return ret;
    }

  if (0)
    {
    regroup:
      /* Sometimes we dynamically decide there are too many groups.  Arrive
       * here. */
      output->next = initial_offset;
      xd3_bit_state_encode_init (& bstate);
    }

  /* Encode: # of groups (3 bits) */
  if ((ret = xd3_encode_bits (stream, & output, & bstate,
			      DJW_GROUP_BITS, groups-1))) { goto failure; }

  if (groups == 1)
    {
      /* Single Huffman group. */
      usize_t    code[ALPHABET_SIZE]; /* Codes */
      uint8_t    clen[ALPHABET_SIZE];
      uint8_t    prefix_mtfsym[ALPHABET_SIZE];
      djw_prefix prefix;

      output_bits =
	djw_build_prefix (real_freq, clen, ALPHABET_SIZE, DJW_MAX_CODELEN);
      djw_build_codes (code, clen, ALPHABET_SIZE, DJW_MAX_CODELEN);

      if (output_bits + EFFICIENCY_BITS >= input_bits && ! cfg->inefficient)
	{
	  goto nosecond;
	}

      /* Encode: prefix */
      prefix.mtfsym = prefix_mtfsym;
      prefix.symbol = clen;
      prefix.scount = ALPHABET_SIZE;

      if ((ret = djw_encode_prefix (stream, & output, & bstate, & prefix)))
	{
	  goto failure;
	}

      if (output_bits + (8 * output->next) + EFFICIENCY_BITS >=
	  input_bits && ! cfg->inefficient)
	{
	  goto nosecond;
	}

      /* Encode: data */
      for (in = input; in; in = in->next_page)
	{
	  const uint8_t *p     = in->base;
	  const uint8_t *p_max = p + in->next;

	  do
	    {
	      usize_t sym  = *p++;
	      usize_t bits = clen[sym];

	      IF_DEBUG (output_bits -= bits);

	      if ((ret = xd3_encode_bits (stream, & output,
					  & bstate, bits, code[sym])))
		{
		  goto failure;
		}
	    }
	  while (p < p_max);
	}

      XD3_ASSERT (output_bits == 0);
    }
  else
    {
      /* DJW Huffman */
      djw_weight evolve_freq[DJW_MAX_GROUPS][ALPHABET_SIZE];
      uint8_t evolve_clen[DJW_MAX_GROUPS][ALPHABET_SIZE];
      djw_weight left = input_bytes;
      usize_t gp;
      usize_t niter = 0;
      usize_t select_bits;
      usize_t sym1 = 0, sym2 = 0, s;
      usize_t gcost[DJW_MAX_GROUPS];
      usize_t gbest_code[DJW_MAX_GROUPS+2];
      uint8_t gbest_clen[DJW_MAX_GROUPS+2];
      usize_t  gbest_max = 1 + (input_bytes - 1) / sector_size;
      usize_t best_bits = 0;
      usize_t  gbest_no;
      usize_t  gpcnt;
      const uint8_t *p;
      IF_DEBUG2 (usize_t gcount[DJW_MAX_GROUPS]);

      /* Encode: sector size (5 bits) */
      if ((ret = xd3_encode_bits (stream, & output, & bstate,
				  DJW_SECTORSZ_BITS,
				  (sector_size/DJW_SECTORSZ_MULT)-1)))
	{
	  goto failure;
	}

      /* Dynamic allocation. */
      if (gbest == NULL)
	{
	  if ((gbest = (uint8_t*) xd3_alloc (stream, gbest_max, 1)) == NULL)
	    {
	      ret = ENOMEM;
	      goto failure;
	    }
	}

      if (gbest_mtf == NULL)
	{
	  if ((gbest_mtf = (uint8_t*) xd3_alloc (stream, gbest_max, 1)) == NULL)
	    {
	      ret = ENOMEM;
	      goto failure;
	    }
	}

      /* OPT: Some of the inner loops can be optimized, as shown in bzip2 */

      /* Generate initial code length tables. */
      for (gp = 0; gp < groups; gp += 1)
	{
	  djw_weight sum  = 0;
	  djw_weight goal = left / (groups - gp);

	  IF_DEBUG2 (usize_t nz = 0);

	  /* Due to the single-code granularity of this distribution, it may
	   * be that we can't generate a distribution for each group.  In that
	   * case subtract one group and try again.  If (inefficient), we're
	   * testing group behavior, so don't mess things up. */
	  if (goal == 0 && !cfg->inefficient)
	    {
	      IF_DEBUG2 (DP(RINT "too many groups (%u), dropping one\n",
			    groups));
	      groups -= 1;
	      goto regroup;
	    }

	  /* Sum == goal is possible when (cfg->inefficient)... */
	  while (sum < goal)
	    {
	      XD3_ASSERT (sym2 < ALPHABET_SIZE);
	      IF_DEBUG2 (nz += real_freq[sym2] != 0);
	      sum += real_freq[sym2++];
	    }

	  IF_DEBUG2(DP(RINT "group %u has symbols %u..%u (%u non-zero) "
		       "(%u/%u = %.3f)\n",
		       gp, sym1, sym2, nz, sum,
		       input_bytes, sum / (double)input_bytes););

	  for (s = 0; s < ALPHABET_SIZE; s += 1)
	    {
	      evolve_clen[gp][s] = (s >= sym1 && s <= sym2) ? 1 : 16;
	    }

	  left -= sum;
	  sym1  = sym2+1;
	}

    repeat:

      niter += 1;
      gbest_no = 0;
      memset (evolve_freq, 0, sizeof (evolve_freq[0]) * groups);
      IF_DEBUG2 (memset (gcount, 0, sizeof (gcount[0]) * groups));

      /* For each input page (loop is irregular to allow non-pow2-size group
       * size. */
      in = input;
      p  = in->base;

      /* For each group-size sector. */
      do
	{
	  const uint8_t *p0  = p;
	  xd3_output    *in0 = in;
	  usize_t best   = 0;
	  usize_t winner = 0;

	  /* Select best group for each sector, update evolve_freq. */
	  memset (gcost, 0, sizeof (gcost[0]) * groups);

	  /* For each byte in sector. */
	  for (gpcnt = 0; gpcnt < sector_size; gpcnt += 1)
	    {
	      /* For each group. */
	      for (gp = 0; gp < groups; gp += 1)
		{
		  gcost[gp] += evolve_clen[gp][*p];
		}

	      /* Check end-of-input-page. */
#             define GP_PAGE()                \
	      if ((usize_t)(++p - in->base) == in->next) \
		{                             \
		  in = in->next_page;         \
		  if (in == NULL) { break; }  \
		  p  = in->base;              \
		}

	      GP_PAGE ();
	    }

	  /* Find min cost group for this sector */
	  best = USIZE_T_MAX;
	  for (gp = 0; gp < groups; gp += 1)
	    {
	      if (gcost[gp] < best) 
		{ 
		  best = gcost[gp]; 
		  winner = gp; 
		}
	    }

	  XD3_ASSERT(gbest_no < gbest_max);
	  gbest[gbest_no++] = winner;
	  IF_DEBUG2 (gcount[winner] += 1);

	  p  = p0;
	  in = in0;

	  /* Update group frequencies. */
	  for (gpcnt = 0; gpcnt < sector_size; gpcnt += 1)
	    {
	      evolve_freq[winner][*p] += 1;

	      GP_PAGE ();
	    }
	}
      while (in != NULL);

      XD3_ASSERT (gbest_no == gbest_max);

      /* Recompute code lengths. */
      output_bits = 0;
      for (gp = 0; gp < groups; gp += 1)
	{
	  int i;
	  uint8_t evolve_zero[ALPHABET_SIZE];
	  int any_zeros = 0;

	  memset (evolve_zero, 0, sizeof (evolve_zero));

	  /* Cannot allow a zero clen when the real frequency is non-zero.
	   * Note: this means we are going to encode a fairly long code for
	   * these unused entries.  An improvement would be to implement a
	   * NOTUSED code for when these are actually zero, but this requires
	   * another data structure (evolve_zero) since we don't know when
	   * evolve_freq[i] == 0...  Briefly tested, looked worse. */
	  for (i = 0; i < ALPHABET_SIZE; i += 1)
	    {
	      if (evolve_freq[gp][i] == 0 && real_freq[i] != 0)
		{
		  evolve_freq[gp][i] = 1;
		  evolve_zero[i] = 1;
		  any_zeros = 1;
		}
	    }

	  output_bits += djw_build_prefix (evolve_freq[gp], evolve_clen[gp],
					   ALPHABET_SIZE, DJW_MAX_CODELEN);

	  /* The above faking of frequencies does not matter for the last
	   * iteration, but we don't know when that is yet.  However, it also
	   * breaks the output_bits computation.  Necessary for accuracy, and
	   * for the (output_bits==0) assert after all bits are output. */
	  if (any_zeros)
	    {
	      IF_DEBUG2 (usize_t save_total = output_bits);

	      for (i = 0; i < ALPHABET_SIZE; i += 1)
		{
		  if (evolve_zero[i]) { output_bits -= evolve_clen[gp][i]; }
		}

	      IF_DEBUG2 (DP(RINT "evolve_zero reduced %u bits in group %u\n",
			    save_total - output_bits, gp));
	    }
	}

      IF_DEBUG2(
	DP(RINT "pass %u total bits: %u group uses: ", niter, output_bits);
	for (gp = 0; gp < groups; gp += 1) { DP(RINT "%u ", gcount[gp]); }
	DP(RINT "\n");
	);

      /* End iteration. */

      IF_DEBUG2 (if (niter > 1 && best_bits < output_bits) {
	DP(RINT "iteration lost %u bits\n", output_bits - best_bits); });

      if (niter == 1 || (niter < DJW_MAX_ITER &&
			 (best_bits - output_bits) >= DJW_MIN_IMPROVEMENT))
	{
	  best_bits = output_bits;
	  goto repeat;
	}

      /* Efficiency check. */
      if (output_bits + EFFICIENCY_BITS >= input_bits && ! cfg->inefficient)
	{
	  goto nosecond;
	}

      IF_DEBUG2 (DP(RINT "djw compression: %u -> %0.3f\n",
		    input_bytes, output_bits / 8.0));

      /* Encode: prefix */
      {
	uint8_t     prefix_symbol[DJW_MAX_GROUPS * ALPHABET_SIZE];
	uint8_t     prefix_mtfsym[DJW_MAX_GROUPS * ALPHABET_SIZE];
	uint8_t     prefix_repcnt[DJW_MAX_GROUPS * ALPHABET_SIZE];
	djw_prefix prefix;

	prefix.symbol = prefix_symbol;
	prefix.mtfsym = prefix_mtfsym;
	prefix.repcnt = prefix_repcnt;

	djw_compute_multi_prefix (groups, evolve_clen, & prefix);
	if ((ret = djw_encode_prefix (stream, & output, & bstate, & prefix)))
	  {
	    goto failure;
	  }
      }

      /* Encode: selector frequencies */
      {
	/* DJW_MAX_GROUPS +2 is for RUN_0, RUN_1 symbols. */
	djw_weight gbest_freq[DJW_MAX_GROUPS+2];
	djw_prefix gbest_prefix;
	usize_t i;

	gbest_prefix.scount = gbest_no;
	gbest_prefix.symbol = gbest;
	gbest_prefix.mtfsym = gbest_mtf;

	djw_compute_selector_1_2 (& gbest_prefix, groups, gbest_freq);

	select_bits =
	  djw_build_prefix (gbest_freq, gbest_clen, groups+1, DJW_MAX_GBCLEN);
	djw_build_codes  (gbest_code, gbest_clen, groups+1, DJW_MAX_GBCLEN);

	for (i = 0; i < groups+1; i += 1)
	  {
	    if ((ret = xd3_encode_bits (stream, & output, & bstate,
					DJW_GBCLEN_BITS, gbest_clen[i])))
	      {
		goto failure;
	      }
	  }

	for (i = 0; i < gbest_prefix.mcount; i += 1)
	  {
	    usize_t gp_mtf      = gbest_mtf[i];
	    usize_t gp_sel_bits = gbest_clen[gp_mtf];
	    usize_t gp_sel_code = gbest_code[gp_mtf];

	    XD3_ASSERT (gp_mtf < groups+1);

	    if ((ret = xd3_encode_bits (stream, & output, & bstate,
					gp_sel_bits, gp_sel_code)))
	      {
		goto failure;
	      }

	    IF_DEBUG (select_bits -= gp_sel_bits);
	  }

	XD3_ASSERT (select_bits == 0);
      }

      /* Efficiency check. */
      if (output_bits + select_bits + (8 * output->next) +
	  EFFICIENCY_BITS >= input_bits && ! cfg->inefficient)
	{
	  goto nosecond;
	}

      /* Encode: data */
      {
	usize_t evolve_code[DJW_MAX_GROUPS][ALPHABET_SIZE];
	usize_t sector = 0;

	/* Build code tables for each group. */
	for (gp = 0; gp < groups; gp += 1)
	  {
	    djw_build_codes (evolve_code[gp], evolve_clen[gp],
			     ALPHABET_SIZE, DJW_MAX_CODELEN);
	  }

	/* Now loop over the input. */
	in = input;
	p  = in->base;

	do
	  {
	    /* For each sector. */
	    usize_t   gp_best  = gbest[sector];
	    usize_t *gp_codes = evolve_code[gp_best];
	    uint8_t *gp_clens = evolve_clen[gp_best];

	    XD3_ASSERT (sector < gbest_no);

	    sector += 1;

	    /* Encode the sector data. */
	    for (gpcnt = 0; gpcnt < sector_size; gpcnt += 1)
	      {
		usize_t sym  = *p;
		usize_t bits = gp_clens[sym];
		usize_t code = gp_codes[sym];

		IF_DEBUG (output_bits -= bits);

		if ((ret = xd3_encode_bits (stream, & output, & bstate,
					    bits, code)))
		  {
		    goto failure;
		  }

		GP_PAGE ();
	      }
	  }
	while (in != NULL);

	XD3_ASSERT (select_bits == 0);
	XD3_ASSERT (output_bits == 0);
      }
    }

  ret = xd3_flush_bits (stream, & output, & bstate);

  if (0)
    {
    nosecond:
      stream->msg = "secondary compression was inefficient";
      ret = XD3_NOSECOND;
    }

 failure:

  xd3_free (stream, gbest);
  xd3_free (stream, gbest_mtf);
  return ret;
}
#endif /* XD3_ENCODER */

/*********************************************************************/
/*                              DECODE                               */
/*********************************************************************/

static void
djw_build_decoder (xd3_stream    *stream,
		   usize_t        asize,
		   usize_t        abs_max,
		   const uint8_t *clen,
		   uint8_t       *inorder,
		   usize_t       *base,
		   usize_t       *limit,
		   usize_t       *min_clenp,
		   usize_t       *max_clenp)
{
  usize_t i, l;
  const uint8_t *ci;
  usize_t nr_clen [DJW_TOTAL_CODES];
  usize_t tmp_base[DJW_TOTAL_CODES];
  usize_t min_clen;
  usize_t max_clen;

  /* Assumption: the two temporary arrays are large enough to hold abs_max. */
  XD3_ASSERT (abs_max <= DJW_MAX_CODELEN);

  /* This looks something like the start of zlib's inftrees.c */
  memset (nr_clen, 0, sizeof (nr_clen[0]) * (abs_max+1));

  /* Count number of each code length */
  i  = asize;
  ci = clen;
  do
    {
      /* Caller _must_ check that values are in-range.  Most of the time the
       * caller decodes a specific number of bits, which imply the max value,
       * and the other time the caller decodes a huffman value, which must be
       * in-range.  Therefore, its an assertion and this function cannot
       * otherwise fail. */
      XD3_ASSERT (*ci <= abs_max);

      nr_clen[*ci++]++;
    }
  while (--i != 0);

  /* Compute min, max. */
  for (i = 1; i <= abs_max; i += 1) { if (nr_clen[i]) { break; } }
  min_clen = i;
  for (i = abs_max; i != 0; i -= 1) { if (nr_clen[i]) { break; } }
  max_clen = i;

  /* Fill the BASE, LIMIT table. */
  tmp_base[min_clen] = 0;
  base[min_clen]     = 0;
  limit[min_clen]    = nr_clen[min_clen] - 1;
  for (i = min_clen + 1; i <= max_clen; i += 1)
    {
      usize_t last_limit = ((limit[i-1] + 1) << 1);
      tmp_base[i] = tmp_base[i-1] + nr_clen[i-1];
      limit[i]    = last_limit + nr_clen[i] - 1;
      base[i]     = last_limit - tmp_base[i];
    }

  /* Fill the inorder array, canonically ordered codes. */
  ci = clen;
  for (i = 0; i < asize; i += 1)
    {
      if ((l = *ci++) != 0)
	{
	  inorder[tmp_base[l]++] = i;
	}
    }

  *min_clenp = min_clen;
  *max_clenp = max_clen;
}

static inline int
djw_decode_symbol (xd3_stream     *stream,
		   bit_state      *bstate,
		   const uint8_t **input,
		   const uint8_t  *input_end,
		   const uint8_t  *inorder,
		   const usize_t  *base,
		   const usize_t  *limit,
		   usize_t         min_clen,
		   usize_t         max_clen,
		   usize_t         *sym,
		   usize_t          max_sym)
{
  usize_t code = 0;
  usize_t bits = 0;

  /* OPT: Supposedly a small lookup table improves speed here... */

  /* Code outline is similar to xd3_decode_bits... */
  if (bstate->cur_mask == 0x100) { goto next_byte; }

  for (;;)
    {
      do
	{
	  if (bits == max_clen) { goto corrupt; }

	  bits += 1;
	  code  = (code << 1);

	  if (bstate->cur_byte & bstate->cur_mask) { code |= 1; }

	  bstate->cur_mask <<= 1;

	  if (bits >= min_clen && code <= limit[bits]) { goto done; }
	}
      while (bstate->cur_mask != 0x100);

    next_byte:

      if (*input == input_end)
	{
	  stream->msg = "secondary decoder end of input";
	  return XD3_INTERNAL;
	}

      bstate->cur_byte = *(*input)++;
      bstate->cur_mask = 1;
    }

 done:

  if (base[bits] <= code)
    {
      usize_t offset = code - base[bits];

      if (offset <= max_sym)
	{
	  IF_DEBUG2 (DP(RINT "(j) %u ", code));
	  *sym = inorder[offset];
	  return 0;
	}
    }

 corrupt:
  stream->msg = "secondary decoder invalid code";
  return XD3_INTERNAL;
}

static int
djw_decode_clclen (xd3_stream     *stream,
		   bit_state      *bstate,
		   const uint8_t **input,
		   const uint8_t  *input_end,
		   uint8_t        *cl_inorder,
		   usize_t        *cl_base,
		   usize_t        *cl_limit,
		   usize_t        *cl_minlen,
		   usize_t        *cl_maxlen,
		   uint8_t        *cl_mtf)
{
  int ret;
  uint8_t cl_clen[DJW_TOTAL_CODES];
  usize_t num_codes, value;
  usize_t i;

  /* How many extra code lengths to encode. */
  if ((ret = xd3_decode_bits (stream, bstate, input,
			      input_end, DJW_EXTRA_CODE_BITS, & num_codes)))
    {
      return ret;
    }

  num_codes += DJW_EXTRA_12OFFSET;

  /* Read num_codes. */
  for (i = 0; i < num_codes; i += 1)
    {
      if ((ret = xd3_decode_bits (stream, bstate, input,
				  input_end, DJW_CLCLEN_BITS, & value)))
	{
	  return ret;
	}

      cl_clen[i] = value;
    }

  /* Set the rest to zero. */
  for (; i < DJW_TOTAL_CODES; i += 1) { cl_clen[i] = 0; }

  /* No need to check for in-range clen values, because: */
  XD3_ASSERT (1 << DJW_CLCLEN_BITS == DJW_MAX_CLCLEN + 1);

  /* Build the code-length decoder. */
  djw_build_decoder (stream, DJW_TOTAL_CODES, DJW_MAX_CLCLEN,
		     cl_clen, cl_inorder, cl_base,
		     cl_limit, cl_minlen, cl_maxlen);

  /* Initialize the MTF state. */
  djw_init_clen_mtf_1_2 (cl_mtf);

  return 0;
}

static inline int
djw_decode_1_2 (xd3_stream     *stream,
		bit_state      *bstate,
		const uint8_t **input,
		const uint8_t  *input_end,
		const uint8_t  *inorder,
		const usize_t  *base,
		const usize_t  *limit,
		const usize_t  *minlen,
		const usize_t  *maxlen,
		uint8_t        *mtfvals,
		usize_t         elts,
		usize_t         skip_offset,
		uint8_t        *values)
{
  usize_t n = 0, rep = 0, mtf = 0, s = 0;
  int ret;
  
  while (n < elts)
    {
      /* Special case inside generic code: CLEN only: If not the first group,
       * we already know the zero frequencies. */
      if (skip_offset != 0 && n >= skip_offset && values[n-skip_offset] == 0)
	{
	  values[n++] = 0;
	  continue;
	}

      /* Repeat last symbol. */
      if (rep != 0)
	{
	  values[n++] = mtfvals[0];
	  rep -= 1;
	  continue;
	}

      /* Symbol following last repeat code. */
      if (mtf != 0)
	{
	  usize_t sym = djw_update_mtf (mtfvals, mtf);
	  values[n++] = sym;
	  mtf = 0;
	  continue;
	}

      /* Decode next symbol/repeat code. */
      if ((ret = djw_decode_symbol (stream, bstate, input, input_end,
				    inorder, base, limit, *minlen, *maxlen,
				    & mtf, DJW_TOTAL_CODES))) { return ret; }

      if (mtf <= RUN_1)
	{
	  /* Repetition. */
	  rep = ((mtf + 1) << s);
	  mtf = 0;
	  s += 1;
	}
      else
	{
	  /* Remove the RUN_1 MTF offset. */
	  mtf -= 1;
	  s = 0;
	}
    }

  /* If (rep != 0) there were too many codes received. */
  if (rep != 0)
    {
      stream->msg = "secondary decoder invalid repeat code";
      return XD3_INTERNAL;
    }
  
  return 0;
}

static inline int
djw_decode_prefix (xd3_stream     *stream,
		   bit_state      *bstate,
		   const uint8_t **input,
		   const uint8_t  *input_end,
		   const uint8_t  *cl_inorder,
		   const usize_t  *cl_base,
		   const usize_t  *cl_limit,
		   const usize_t  *cl_minlen,
		   const usize_t  *cl_maxlen,
		   uint8_t        *cl_mtf,
		   usize_t         groups,
		   uint8_t        *clen)
{
  return djw_decode_1_2 (stream, bstate, input, input_end,
			 cl_inorder, cl_base, cl_limit,
			 cl_minlen, cl_maxlen, cl_mtf,
			 ALPHABET_SIZE * groups, ALPHABET_SIZE, clen);
}

static int
xd3_decode_huff (xd3_stream     *stream,
		 djw_stream    *h,
		 const uint8_t **input_pos,
		 const uint8_t  *const input_end,
		 uint8_t       **output_pos,
		 const uint8_t  *const output_end)
{
  const uint8_t *input = *input_pos;
  uint8_t  *output = *output_pos;
  bit_state bstate = BIT_STATE_DECODE_INIT;
  uint8_t  *sel_group = NULL;
  usize_t    groups, gp;
  usize_t    output_bytes = (usize_t)(output_end - output);
  usize_t    sector_size;
  usize_t    sectors;
  int ret;

  /* Invalid input. */
  if (output_bytes == 0)
    {
      stream->msg = "secondary decoder invalid input";
      return XD3_INTERNAL;
    }

  /* Decode: number of groups */
  if ((ret = xd3_decode_bits (stream, & bstate, & input,
			      input_end, DJW_GROUP_BITS, & groups)))
    {
      goto fail;
    }

  groups += 1;

  if (groups > 1)
    {
      /* Decode: group size */
      if ((ret = xd3_decode_bits (stream, & bstate, & input,
				  input_end, DJW_SECTORSZ_BITS,
				  & sector_size))) { goto fail; }
      
      sector_size = (sector_size + 1) * DJW_SECTORSZ_MULT;
    }
  else
    {
      /* Default for groups == 1 */
      sector_size = output_bytes;
    }

  sectors = 1 + (output_bytes - 1) / sector_size;

  /* TODO: In the case of groups==1, lots of extra stack space gets used here.
   * Could dynamically allocate this memory, which would help with excess
   * parameter passing, too.  Passing too many parameters in this file,
   * simplify it! */

  /* Outer scope: per-group symbol decoder tables. */
  {
    uint8_t inorder[DJW_MAX_GROUPS][ALPHABET_SIZE];
    usize_t base   [DJW_MAX_GROUPS][DJW_TOTAL_CODES];
    usize_t limit  [DJW_MAX_GROUPS][DJW_TOTAL_CODES];
    usize_t minlen [DJW_MAX_GROUPS];
    usize_t maxlen [DJW_MAX_GROUPS];

    /* Nested scope: code length decoder tables. */
    {
      uint8_t clen      [DJW_MAX_GROUPS][ALPHABET_SIZE];
      uint8_t cl_inorder[DJW_TOTAL_CODES];
      usize_t cl_base   [DJW_MAX_CLCLEN+2];
      usize_t cl_limit  [DJW_MAX_CLCLEN+2];
      uint8_t cl_mtf    [DJW_TOTAL_CODES];
      usize_t cl_minlen;
      usize_t cl_maxlen;

      /* Compute the code length decoder. */
      if ((ret = djw_decode_clclen (stream, & bstate, & input, input_end,
				    cl_inorder, cl_base, cl_limit, & cl_minlen,
				    & cl_maxlen, cl_mtf))) { goto fail; }

      /* Now decode each group decoder. */
      if ((ret = djw_decode_prefix (stream, & bstate, & input, input_end,
				    cl_inorder, cl_base, cl_limit,
				    & cl_minlen, & cl_maxlen, cl_mtf,
				    groups, clen[0]))) { goto fail; }

      /* Prepare the actual decoding tables. */
      for (gp = 0; gp < groups; gp += 1)
	{
	  djw_build_decoder (stream, ALPHABET_SIZE, DJW_MAX_CODELEN,
			     clen[gp], inorder[gp], base[gp], limit[gp],
			     & minlen[gp], & maxlen[gp]);
	}
    }

    /* Decode: selector clens. */
    {
      uint8_t sel_inorder[DJW_MAX_GROUPS+2];
      usize_t sel_base   [DJW_MAX_GBCLEN+2];
      usize_t sel_limit  [DJW_MAX_GBCLEN+2];
      uint8_t sel_mtf    [DJW_MAX_GROUPS+2];
      usize_t sel_minlen;
      usize_t sel_maxlen;

      /* Setup group selection. */
      if (groups > 1)
	{
	  uint8_t sel_clen[DJW_MAX_GROUPS+1];

	  for (gp = 0; gp < groups+1; gp += 1)
	    {
	      usize_t value;

	      if ((ret = xd3_decode_bits (stream, & bstate, & input,
					  input_end, DJW_GBCLEN_BITS,
					  & value))) { goto fail; }

	      sel_clen[gp] = value;
	      sel_mtf[gp]  = gp;
	    }

	  if ((sel_group = (uint8_t*) xd3_alloc (stream, sectors, 1)) == NULL)
	    {
	      ret = ENOMEM;
	      goto fail;
	    }

	  djw_build_decoder (stream, groups+1, DJW_MAX_GBCLEN, sel_clen,
			     sel_inorder, sel_base, sel_limit,
			     & sel_minlen, & sel_maxlen);

	  if ((ret = djw_decode_1_2 (stream, & bstate, & input, input_end,
				     sel_inorder, sel_base,
				     sel_limit, & sel_minlen,
				     & sel_maxlen, sel_mtf,
				     sectors, 0, sel_group))) { goto fail; }
	}

      /* Now decode each sector. */
      {
	/* Initialize for (groups==1) case. */
	uint8_t *gp_inorder = inorder[0]; 
	usize_t *gp_base    = base[0];
	usize_t *gp_limit   = limit[0];
	usize_t  gp_minlen  = minlen[0];
	usize_t  gp_maxlen  = maxlen[0];
	usize_t c;

	for (c = 0; c < sectors; c += 1)
	  {
	    usize_t n;

	    if (groups >= 2)
	      {
		gp = sel_group[c];

		XD3_ASSERT (gp < groups);

		gp_inorder = inorder[gp];
		gp_base    = base[gp];
		gp_limit   = limit[gp];
		gp_minlen  = minlen[gp];
		gp_maxlen  = maxlen[gp];
	      }

	    XD3_ASSERT (output_end - output > 0);
	    
	    /* Decode next sector. */
	    n = min (sector_size, (usize_t) (output_end - output));

	    do
	      {
		usize_t sym;

		if ((ret = djw_decode_symbol (stream, & bstate,
					      & input, input_end,
					      gp_inorder, gp_base,
					      gp_limit, gp_minlen, gp_maxlen,
					      & sym, ALPHABET_SIZE)))
		  {
		    goto fail;
		  }

		*output++ = sym;
	      }
	    while (--n);
	  }
      }
    }
  }

  IF_REGRESSION (if ((ret = xd3_test_clean_bits (stream, & bstate)))
		   { goto fail; });
  XD3_ASSERT (ret == 0);

 fail:
  xd3_free (stream, sel_group);

  (*input_pos) = input;
  (*output_pos) = output;
  return ret;
}

#endif