summaryrefslogtreecommitdiff
path: root/libavcodec/vp8.c
blob: 760958ef844f49ed31e071147e1b7c70433069d0 (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
/**
 * VP8 compatible video decoder
 *
 * Copyright (C) 2010 David Conrad
 * Copyright (C) 2010 Ronald S. Bultje
 * Copyright (C) 2010 Jason Garrett-Glaser
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/imgutils.h"
#include "avcodec.h"
#include "vp8.h"
#include "vp8data.h"
#include "rectangle.h"

#if ARCH_ARM
#   include "arm/vp8.h"
#endif

static void vp8_decode_flush(AVCodecContext *avctx)
{
    VP8Context *s = avctx->priv_data;
    int i;

    for (i = 0; i < 4; i++)
        if (s->frames[i].data[0])
            avctx->release_buffer(avctx, &s->frames[i]);
    memset(s->framep, 0, sizeof(s->framep));

    av_freep(&s->macroblocks_base);
    av_freep(&s->filter_strength);
    av_freep(&s->intra4x4_pred_mode_top);
    av_freep(&s->top_nnz);
    av_freep(&s->edge_emu_buffer);
    av_freep(&s->top_border);
    av_freep(&s->segmentation_map);

    s->macroblocks        = NULL;
}

static int update_dimensions(VP8Context *s, int width, int height)
{
    if (av_image_check_size(width, height, 0, s->avctx))
        return AVERROR_INVALIDDATA;

    vp8_decode_flush(s->avctx);

    avcodec_set_dimensions(s->avctx, width, height);

    s->mb_width  = (s->avctx->coded_width +15) / 16;
    s->mb_height = (s->avctx->coded_height+15) / 16;

    s->macroblocks_base        = av_mallocz((s->mb_width+s->mb_height*2+1)*sizeof(*s->macroblocks));
    s->filter_strength         = av_mallocz(s->mb_width*sizeof(*s->filter_strength));
    s->intra4x4_pred_mode_top  = av_mallocz(s->mb_width*4);
    s->top_nnz                 = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
    s->top_border              = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));
    s->segmentation_map        = av_mallocz(s->mb_width*s->mb_height);

    if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_top ||
        !s->top_nnz || !s->top_border || !s->segmentation_map)
        return AVERROR(ENOMEM);

    s->macroblocks        = s->macroblocks_base + 1;

    return 0;
}

static void parse_segment_info(VP8Context *s)
{
    VP56RangeCoder *c = &s->c;
    int i;

    s->segmentation.update_map = vp8_rac_get(c);

    if (vp8_rac_get(c)) { // update segment feature data
        s->segmentation.absolute_vals = vp8_rac_get(c);

        for (i = 0; i < 4; i++)
            s->segmentation.base_quant[i]   = vp8_rac_get_sint(c, 7);

        for (i = 0; i < 4; i++)
            s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6);
    }
    if (s->segmentation.update_map)
        for (i = 0; i < 3; i++)
            s->prob->segmentid[i] = vp8_rac_get(c) ? vp8_rac_get_uint(c, 8) : 255;
}

static void update_lf_deltas(VP8Context *s)
{
    VP56RangeCoder *c = &s->c;
    int i;

    for (i = 0; i < 4; i++)
        s->lf_delta.ref[i]  = vp8_rac_get_sint(c, 6);

    for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++)
        s->lf_delta.mode[i] = vp8_rac_get_sint(c, 6);
}

static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
{
    const uint8_t *sizes = buf;
    int i;

    s->num_coeff_partitions = 1 << vp8_rac_get_uint(&s->c, 2);

    buf      += 3*(s->num_coeff_partitions-1);
    buf_size -= 3*(s->num_coeff_partitions-1);
    if (buf_size < 0)
        return -1;

    for (i = 0; i < s->num_coeff_partitions-1; i++) {
        int size = AV_RL24(sizes + 3*i);
        if (buf_size - size < 0)
            return -1;

        ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, size);
        buf      += size;
        buf_size -= size;
    }
    ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, buf_size);

    return 0;
}

static void get_quants(VP8Context *s)
{
    VP56RangeCoder *c = &s->c;
    int i, base_qi;

    int yac_qi     = vp8_rac_get_uint(c, 7);
    int ydc_delta  = vp8_rac_get_sint(c, 4);
    int y2dc_delta = vp8_rac_get_sint(c, 4);
    int y2ac_delta = vp8_rac_get_sint(c, 4);
    int uvdc_delta = vp8_rac_get_sint(c, 4);
    int uvac_delta = vp8_rac_get_sint(c, 4);

    for (i = 0; i < 4; i++) {
        if (s->segmentation.enabled) {
            base_qi = s->segmentation.base_quant[i];
            if (!s->segmentation.absolute_vals)
                base_qi += yac_qi;
        } else
            base_qi = yac_qi;

        s->qmat[i].luma_qmul[0]    =       vp8_dc_qlookup[av_clip(base_qi + ydc_delta , 0, 127)];
        s->qmat[i].luma_qmul[1]    =       vp8_ac_qlookup[av_clip(base_qi             , 0, 127)];
        s->qmat[i].luma_dc_qmul[0] =   2 * vp8_dc_qlookup[av_clip(base_qi + y2dc_delta, 0, 127)];
        s->qmat[i].luma_dc_qmul[1] = 155 * vp8_ac_qlookup[av_clip(base_qi + y2ac_delta, 0, 127)] / 100;
        s->qmat[i].chroma_qmul[0]  =       vp8_dc_qlookup[av_clip(base_qi + uvdc_delta, 0, 127)];
        s->qmat[i].chroma_qmul[1]  =       vp8_ac_qlookup[av_clip(base_qi + uvac_delta, 0, 127)];

        s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8);
        s->qmat[i].chroma_qmul[0]  = FFMIN(s->qmat[i].chroma_qmul[0], 132);
    }
}

/**
 * Determine which buffers golden and altref should be updated with after this frame.
 * The spec isn't clear here, so I'm going by my understanding of what libvpx does
 *
 * Intra frames update all 3 references
 * Inter frames update VP56_FRAME_PREVIOUS if the update_last flag is set
 * If the update (golden|altref) flag is set, it's updated with the current frame
 *      if update_last is set, and VP56_FRAME_PREVIOUS otherwise.
 * If the flag is not set, the number read means:
 *      0: no update
 *      1: VP56_FRAME_PREVIOUS
 *      2: update golden with altref, or update altref with golden
 */
static VP56Frame ref_to_update(VP8Context *s, int update, VP56Frame ref)
{
    VP56RangeCoder *c = &s->c;

    if (update)
        return VP56_FRAME_CURRENT;

    switch (vp8_rac_get_uint(c, 2)) {
    case 1:
        return VP56_FRAME_PREVIOUS;
    case 2:
        return (ref == VP56_FRAME_GOLDEN) ? VP56_FRAME_GOLDEN2 : VP56_FRAME_GOLDEN;
    }
    return VP56_FRAME_NONE;
}

static void update_refs(VP8Context *s)
{
    VP56RangeCoder *c = &s->c;

    int update_golden = vp8_rac_get(c);
    int update_altref = vp8_rac_get(c);

    s->update_golden = ref_to_update(s, update_golden, VP56_FRAME_GOLDEN);
    s->update_altref = ref_to_update(s, update_altref, VP56_FRAME_GOLDEN2);
}

static int decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
{
    VP56RangeCoder *c = &s->c;
    int header_size, hscale, vscale, i, j, k, l, m, ret;
    int width  = s->avctx->width;
    int height = s->avctx->height;

    s->keyframe  = !(buf[0] & 1);
    s->profile   =  (buf[0]>>1) & 7;
    s->invisible = !(buf[0] & 0x10);
    header_size  = AV_RL24(buf) >> 5;
    buf      += 3;
    buf_size -= 3;

    if (s->profile > 3)
        av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile);

    if (!s->profile)
        memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab));
    else    // profile 1-3 use bilinear, 4+ aren't defined so whatever
        memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab, sizeof(s->put_pixels_tab));

    if (header_size > buf_size - 7*s->keyframe) {
        av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n");
        return AVERROR_INVALIDDATA;
    }

    if (s->keyframe) {
        if (AV_RL24(buf) != 0x2a019d) {
            av_log(s->avctx, AV_LOG_ERROR, "Invalid start code 0x%x\n", AV_RL24(buf));
            return AVERROR_INVALIDDATA;
        }
        width  = AV_RL16(buf+3) & 0x3fff;
        height = AV_RL16(buf+5) & 0x3fff;
        hscale = buf[4] >> 6;
        vscale = buf[6] >> 6;
        buf      += 7;
        buf_size -= 7;

        if (hscale || vscale)
            av_log_missing_feature(s->avctx, "Upscaling", 1);

        s->update_golden = s->update_altref = VP56_FRAME_CURRENT;
        for (i = 0; i < 4; i++)
            for (j = 0; j < 16; j++)
                memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]],
                       sizeof(s->prob->token[i][j]));
        memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter, sizeof(s->prob->pred16x16));
        memcpy(s->prob->pred8x8c , vp8_pred8x8c_prob_inter , sizeof(s->prob->pred8x8c));
        memcpy(s->prob->mvc      , vp8_mv_default_prob     , sizeof(s->prob->mvc));
        memset(&s->segmentation, 0, sizeof(s->segmentation));
    }

    if (!s->macroblocks_base || /* first frame */
        width != s->avctx->width || height != s->avctx->height) {
        if ((ret = update_dimensions(s, width, height) < 0))
            return ret;
    }

    ff_vp56_init_range_decoder(c, buf, header_size);
    buf      += header_size;
    buf_size -= header_size;

    if (s->keyframe) {
        if (vp8_rac_get(c))
            av_log(s->avctx, AV_LOG_WARNING, "Unspecified colorspace\n");
        vp8_rac_get(c); // whether we can skip clamping in dsp functions
    }

    if ((s->segmentation.enabled = vp8_rac_get(c)))
        parse_segment_info(s);
    else
        s->segmentation.update_map = 0; // FIXME: move this to some init function?

    s->filter.simple    = vp8_rac_get(c);
    s->filter.level     = vp8_rac_get_uint(c, 6);
    s->filter.sharpness = vp8_rac_get_uint(c, 3);

    if ((s->lf_delta.enabled = vp8_rac_get(c)))
        if (vp8_rac_get(c))
            update_lf_deltas(s);

    if (setup_partitions(s, buf, buf_size)) {
        av_log(s->avctx, AV_LOG_ERROR, "Invalid partitions\n");
        return AVERROR_INVALIDDATA;
    }

    get_quants(s);

    if (!s->keyframe) {
        update_refs(s);
        s->sign_bias[VP56_FRAME_GOLDEN]               = vp8_rac_get(c);
        s->sign_bias[VP56_FRAME_GOLDEN2 /* altref */] = vp8_rac_get(c);
    }

    // if we aren't saving this frame's probabilities for future frames,
    // make a copy of the current probabilities
    if (!(s->update_probabilities = vp8_rac_get(c)))
        s->prob[1] = s->prob[0];

    s->update_last = s->keyframe || vp8_rac_get(c);

    for (i = 0; i < 4; i++)
        for (j = 0; j < 8; j++)
            for (k = 0; k < 3; k++)
                for (l = 0; l < NUM_DCT_TOKENS-1; l++)
                    if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l])) {
                        int prob = vp8_rac_get_uint(c, 8);
                        for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
                            s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
                    }

    if ((s->mbskip_enabled = vp8_rac_get(c)))
        s->prob->mbskip = vp8_rac_get_uint(c, 8);

    if (!s->keyframe) {
        s->prob->intra  = vp8_rac_get_uint(c, 8);
        s->prob->last   = vp8_rac_get_uint(c, 8);
        s->prob->golden = vp8_rac_get_uint(c, 8);

        if (vp8_rac_get(c))
            for (i = 0; i < 4; i++)
                s->prob->pred16x16[i] = vp8_rac_get_uint(c, 8);
        if (vp8_rac_get(c))
            for (i = 0; i < 3; i++)
                s->prob->pred8x8c[i]  = vp8_rac_get_uint(c, 8);

        // 17.2 MV probability update
        for (i = 0; i < 2; i++)
            for (j = 0; j < 19; j++)
                if (vp56_rac_get_prob_branchy(c, vp8_mv_update_prob[i][j]))
                    s->prob->mvc[i][j] = vp8_rac_get_nn(c);
    }

    return 0;
}

static av_always_inline void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src)
{
    dst->x = av_clip(src->x, s->mv_min.x, s->mv_max.x);
    dst->y = av_clip(src->y, s->mv_min.y, s->mv_max.y);
}

/**
 * Motion vector coding, 17.1.
 */
static int read_mv_component(VP56RangeCoder *c, const uint8_t *p)
{
    int bit, x = 0;

    if (vp56_rac_get_prob_branchy(c, p[0])) {
        int i;

        for (i = 0; i < 3; i++)
            x += vp56_rac_get_prob(c, p[9 + i]) << i;
        for (i = 9; i > 3; i--)
            x += vp56_rac_get_prob(c, p[9 + i]) << i;
        if (!(x & 0xFFF0) || vp56_rac_get_prob(c, p[12]))
            x += 8;
    } else {
        // small_mvtree
        const uint8_t *ps = p+2;
        bit = vp56_rac_get_prob(c, *ps);
        ps += 1 + 3*bit;
        x  += 4*bit;
        bit = vp56_rac_get_prob(c, *ps);
        ps += 1 + bit;
        x  += 2*bit;
        x  += vp56_rac_get_prob(c, *ps);
    }

    return (x && vp56_rac_get_prob(c, p[1])) ? -x : x;
}

static av_always_inline
const uint8_t *get_submv_prob(uint32_t left, uint32_t top)
{
    if (left == top)
        return vp8_submv_prob[4-!!left];
    if (!top)
        return vp8_submv_prob[2];
    return vp8_submv_prob[1-!!left];
}

/**
 * Split motion vector prediction, 16.4.
 * @returns the number of motion vectors parsed (2, 4 or 16)
 */
static av_always_inline
int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb)
{
    int part_idx;
    int n, num;
    VP8Macroblock *top_mb  = &mb[2];
    VP8Macroblock *left_mb = &mb[-1];
    const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning],
                  *mbsplits_top = vp8_mbsplits[top_mb->partitioning],
                  *mbsplits_cur, *firstidx;
    VP56mv *top_mv  = top_mb->bmv;
    VP56mv *left_mv = left_mb->bmv;
    VP56mv *cur_mv  = mb->bmv;

    if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) {
        if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) {
            part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]);
        } else {
            part_idx = VP8_SPLITMVMODE_8x8;
        }
    } else {
        part_idx = VP8_SPLITMVMODE_4x4;
    }

    num = vp8_mbsplit_count[part_idx];
    mbsplits_cur = vp8_mbsplits[part_idx],
    firstidx = vp8_mbfirstidx[part_idx];
    mb->partitioning = part_idx;

    for (n = 0; n < num; n++) {
        int k = firstidx[n];
        uint32_t left, above;
        const uint8_t *submv_prob;

        if (!(k & 3))
            left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]);
        else
            left  = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]);
        if (k <= 3)
            above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]);
        else
            above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]);

        submv_prob = get_submv_prob(left, above);

        if (vp56_rac_get_prob_branchy(c, submv_prob[0])) {
            if (vp56_rac_get_prob_branchy(c, submv_prob[1])) {
                if (vp56_rac_get_prob_branchy(c, submv_prob[2])) {
                    mb->bmv[n].y = mb->mv.y + read_mv_component(c, s->prob->mvc[0]);
                    mb->bmv[n].x = mb->mv.x + read_mv_component(c, s->prob->mvc[1]);
                } else {
                    AV_ZERO32(&mb->bmv[n]);
                }
            } else {
                AV_WN32A(&mb->bmv[n], above);
            }
        } else {
            AV_WN32A(&mb->bmv[n], left);
        }
    }

    return num;
}

static av_always_inline
void decode_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y)
{
    VP8Macroblock *mb_edge[3] = { mb + 2 /* top */,
                                  mb - 1 /* left */,
                                  mb + 1 /* top-left */ };
    enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV };
    enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
    int idx = CNT_ZERO;
    int cur_sign_bias = s->sign_bias[mb->ref_frame];
    int8_t *sign_bias = s->sign_bias;
    VP56mv near_mv[4];
    uint8_t cnt[4] = { 0 };
    VP56RangeCoder *c = &s->c;

    AV_ZERO32(&near_mv[0]);
    AV_ZERO32(&near_mv[1]);

    /* Process MB on top, left and top-left */
    #define MV_EDGE_CHECK(n)\
    {\
        VP8Macroblock *edge = mb_edge[n];\
        int edge_ref = edge->ref_frame;\
        if (edge_ref != VP56_FRAME_CURRENT) {\
            uint32_t mv = AV_RN32A(&edge->mv);\
            if (mv) {\
                if (cur_sign_bias != sign_bias[edge_ref]) {\
                    /* SWAR negate of the values in mv. */\
                    mv = ~mv;\
                    mv = ((mv&0x7fff7fff) + 0x00010001) ^ (mv&0x80008000);\
                }\
                if (!n || mv != AV_RN32A(&near_mv[idx]))\
                    AV_WN32A(&near_mv[++idx], mv);\
                cnt[idx]      += 1 + (n != 2);\
            } else\
                cnt[CNT_ZERO] += 1 + (n != 2);\
        }\
    }

    MV_EDGE_CHECK(0)
    MV_EDGE_CHECK(1)
    MV_EDGE_CHECK(2)

    mb->partitioning = VP8_SPLITMVMODE_NONE;
    if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) {
        mb->mode = VP8_MVMODE_MV;

        /* If we have three distinct MVs, merge first and last if they're the same */
        if (cnt[CNT_SPLITMV] && AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT]))
            cnt[CNT_NEAREST] += 1;

        /* Swap near and nearest if necessary */
        if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) {
            FFSWAP(uint8_t,     cnt[CNT_NEAREST],     cnt[CNT_NEAR]);
            FFSWAP( VP56mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
        }

        if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) {
            if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) {

                /* Choose the best mv out of 0,0 and the nearest mv */
                clamp_mv(s, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
                cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode    == VP8_MVMODE_SPLIT) +
                                    (mb_edge[VP8_EDGE_TOP]->mode     == VP8_MVMODE_SPLIT)) * 2 +
                                    (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);

                if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) {
                    mb->mode = VP8_MVMODE_SPLIT;
                    mb->mv = mb->bmv[decode_splitmvs(s, c, mb) - 1];
                } else {
                    mb->mv.y += read_mv_component(c, s->prob->mvc[0]);
                    mb->mv.x += read_mv_component(c, s->prob->mvc[1]);
                    mb->bmv[0] = mb->mv;
                }
            } else {
                clamp_mv(s, &mb->mv, &near_mv[CNT_NEAR]);
                mb->bmv[0] = mb->mv;
            }
        } else {
            clamp_mv(s, &mb->mv, &near_mv[CNT_NEAREST]);
            mb->bmv[0] = mb->mv;
        }
    } else {
        mb->mode = VP8_MVMODE_ZERO;
        AV_ZERO32(&mb->mv);
        mb->bmv[0] = mb->mv;
    }
}

static av_always_inline
void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c,
                           int mb_x, int keyframe)
{
    uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
    if (keyframe) {
        int x, y;
        uint8_t* const top = s->intra4x4_pred_mode_top + 4 * mb_x;
        uint8_t* const left = s->intra4x4_pred_mode_left;
        for (y = 0; y < 4; y++) {
            for (x = 0; x < 4; x++) {
                const uint8_t *ctx;
                ctx = vp8_pred4x4_prob_intra[top[x]][left[y]];
                *intra4x4 = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx);
                left[y] = top[x] = *intra4x4;
                intra4x4++;
            }
        }
    } else {
        int i;
        for (i = 0; i < 16; i++)
            intra4x4[i] = vp8_rac_get_tree(c, vp8_pred4x4_tree, vp8_pred4x4_prob_inter);
    }
}

static av_always_inline
void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, uint8_t *segment)
{
    VP56RangeCoder *c = &s->c;

    if (s->segmentation.update_map)
        *segment = vp8_rac_get_tree(c, vp8_segmentid_tree, s->prob->segmentid);
    s->segment = *segment;

    mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0;

    if (s->keyframe) {
        mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra, vp8_pred16x16_prob_intra);

        if (mb->mode == MODE_I4x4) {
            decode_intra4x4_modes(s, c, mb_x, 1);
        } else {
            const uint32_t modes = vp8_pred4x4_mode[mb->mode] * 0x01010101u;
            AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
            AV_WN32A(s->intra4x4_pred_mode_left, modes);
        }

        s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, vp8_pred8x8c_prob_intra);
        mb->ref_frame = VP56_FRAME_CURRENT;
    } else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) {
        // inter MB, 16.2
        if (vp56_rac_get_prob_branchy(c, s->prob->last))
            mb->ref_frame = vp56_rac_get_prob(c, s->prob->golden) ?
                VP56_FRAME_GOLDEN2 /* altref */ : VP56_FRAME_GOLDEN;
        else
            mb->ref_frame = VP56_FRAME_PREVIOUS;
        s->ref_count[mb->ref_frame-1]++;

        // motion vectors, 16.3
        decode_mvs(s, mb, mb_x, mb_y);
    } else {
        // intra MB, 16.1
        mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);

        if (mb->mode == MODE_I4x4)
            decode_intra4x4_modes(s, c, mb_x, 0);

        s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, s->prob->pred8x8c);
        mb->ref_frame = VP56_FRAME_CURRENT;
        mb->partitioning = VP8_SPLITMVMODE_NONE;
        AV_ZERO32(&mb->bmv[0]);
    }
}

#ifndef decode_block_coeffs_internal
/**
 * @param c arithmetic bitstream reader context
 * @param block destination for block coefficients
 * @param probs probabilities to use when reading trees from the bitstream
 * @param i initial coeff index, 0 unless a separate DC block is coded
 * @param zero_nhood the initial prediction context for number of surrounding
 *                   all-zero blocks (only left/top, so 0-2)
 * @param qmul array holding the dc/ac dequant factor at position 0/1
 * @return 0 if no coeffs were decoded
 *         otherwise, the index of the last coeff decoded plus one
 */
static int decode_block_coeffs_internal(VP56RangeCoder *c, DCTELEM block[16],
                                        uint8_t probs[16][3][NUM_DCT_TOKENS-1],
                                        int i, uint8_t *token_prob, int16_t qmul[2])
{
    goto skip_eob;
    do {
        int coeff;
        if (!vp56_rac_get_prob_branchy(c, token_prob[0]))   // DCT_EOB
            return i;

skip_eob:
        if (!vp56_rac_get_prob_branchy(c, token_prob[1])) { // DCT_0
            if (++i == 16)
                return i; // invalid input; blocks should end with EOB
            token_prob = probs[i][0];
            goto skip_eob;
        }

        if (!vp56_rac_get_prob_branchy(c, token_prob[2])) { // DCT_1
            coeff = 1;
            token_prob = probs[i+1][1];
        } else {
            if (!vp56_rac_get_prob_branchy(c, token_prob[3])) { // DCT 2,3,4
                coeff = vp56_rac_get_prob_branchy(c, token_prob[4]);
                if (coeff)
                    coeff += vp56_rac_get_prob(c, token_prob[5]);
                coeff += 2;
            } else {
                // DCT_CAT*
                if (!vp56_rac_get_prob_branchy(c, token_prob[6])) {
                    if (!vp56_rac_get_prob_branchy(c, token_prob[7])) { // DCT_CAT1
                        coeff  = 5 + vp56_rac_get_prob(c, vp8_dct_cat1_prob[0]);
                    } else {                                    // DCT_CAT2
                        coeff  = 7;
                        coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[0]) << 1;
                        coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[1]);
                    }
                } else {    // DCT_CAT3 and up
                    int a = vp56_rac_get_prob(c, token_prob[8]);
                    int b = vp56_rac_get_prob(c, token_prob[9+a]);
                    int cat = (a<<1) + b;
                    coeff  = 3 + (8<<cat);
                    coeff += vp8_rac_get_coeff(c, ff_vp8_dct_cat_prob[cat]);
                }
            }
            token_prob = probs[i+1][2];
        }
        block[zigzag_scan[i]] = (vp8_rac_get(c) ? -coeff : coeff) * qmul[!!i];
    } while (++i < 16);

    return i;
}
#endif

static av_always_inline
int decode_block_coeffs(VP56RangeCoder *c, DCTELEM block[16],
                        uint8_t probs[16][3][NUM_DCT_TOKENS-1],
                        int i, int zero_nhood, int16_t qmul[2])
{
    uint8_t *token_prob = probs[i][zero_nhood];
    if (!vp56_rac_get_prob_branchy(c, token_prob[0]))   // DCT_EOB
        return 0;
    return decode_block_coeffs_internal(c, block, probs, i, token_prob, qmul);
}

static av_always_inline
void decode_mb_coeffs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
                      uint8_t t_nnz[9], uint8_t l_nnz[9])
{
    int i, x, y, luma_start = 0, luma_ctx = 3;
    int nnz_pred, nnz, nnz_total = 0;
    int segment = s->segment;
    int block_dc = 0;

    if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
        nnz_pred = t_nnz[8] + l_nnz[8];

        // decode DC values and do hadamard
        nnz = decode_block_coeffs(c, s->block_dc, s->prob->token[1], 0, nnz_pred,
                                  s->qmat[segment].luma_dc_qmul);
        l_nnz[8] = t_nnz[8] = !!nnz;
        if (nnz) {
            nnz_total += nnz;
            block_dc = 1;
            if (nnz == 1)
                s->vp8dsp.vp8_luma_dc_wht_dc(s->block, s->block_dc);
            else
                s->vp8dsp.vp8_luma_dc_wht(s->block, s->block_dc);
        }
        luma_start = 1;
        luma_ctx = 0;
    }

    // luma blocks
    for (y = 0; y < 4; y++)
        for (x = 0; x < 4; x++) {
            nnz_pred = l_nnz[y] + t_nnz[x];
            nnz = decode_block_coeffs(c, s->block[y][x], s->prob->token[luma_ctx], luma_start,
                                      nnz_pred, s->qmat[segment].luma_qmul);
            // nnz+block_dc may be one more than the actual last index, but we don't care
            s->non_zero_count_cache[y][x] = nnz + block_dc;
            t_nnz[x] = l_nnz[y] = !!nnz;
            nnz_total += nnz;
        }

    // chroma blocks
    // TODO: what to do about dimensions? 2nd dim for luma is x,
    // but for chroma it's (y<<1)|x
    for (i = 4; i < 6; i++)
        for (y = 0; y < 2; y++)
            for (x = 0; x < 2; x++) {
                nnz_pred = l_nnz[i+2*y] + t_nnz[i+2*x];
                nnz = decode_block_coeffs(c, s->block[i][(y<<1)+x], s->prob->token[2], 0,
                                          nnz_pred, s->qmat[segment].chroma_qmul);
                s->non_zero_count_cache[i][(y<<1)+x] = nnz;
                t_nnz[i+2*x] = l_nnz[i+2*y] = !!nnz;
                nnz_total += nnz;
            }

    // if there were no coded coeffs despite the macroblock not being marked skip,
    // we MUST not do the inner loop filter and should not do IDCT
    // Since skip isn't used for bitstream prediction, just manually set it.
    if (!nnz_total)
        mb->skip = 1;
}

static av_always_inline
void backup_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr,
                      int linesize, int uvlinesize, int simple)
{
    AV_COPY128(top_border, src_y + 15*linesize);
    if (!simple) {
        AV_COPY64(top_border+16, src_cb + 7*uvlinesize);
        AV_COPY64(top_border+24, src_cr + 7*uvlinesize);
    }
}

static av_always_inline
void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr,
                    int linesize, int uvlinesize, int mb_x, int mb_y, int mb_width,
                    int simple, int xchg)
{
    uint8_t *top_border_m1 = top_border-32;     // for TL prediction
    src_y  -=   linesize;
    src_cb -= uvlinesize;
    src_cr -= uvlinesize;

#define XCHG(a,b,xchg) do {                     \
        if (xchg) AV_SWAP64(b,a);               \
        else      AV_COPY64(b,a);               \
    } while (0)

    XCHG(top_border_m1+8, src_y-8, xchg);
    XCHG(top_border,      src_y,   xchg);
    XCHG(top_border+8,    src_y+8, 1);
    if (mb_x < mb_width-1)
        XCHG(top_border+32, src_y+16, 1);

    // only copy chroma for normal loop filter
    // or to initialize the top row to 127
    if (!simple || !mb_y) {
        XCHG(top_border_m1+16, src_cb-8, xchg);
        XCHG(top_border_m1+24, src_cr-8, xchg);
        XCHG(top_border+16,    src_cb, 1);
        XCHG(top_border+24,    src_cr, 1);
    }
}

static av_always_inline
int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
{
    if (!mb_x) {
        return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8;
    } else {
        return mb_y ? mode : LEFT_DC_PRED8x8;
    }
}

static av_always_inline
int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y)
{
    if (!mb_x) {
        return mb_y ? VERT_PRED8x8 : DC_129_PRED8x8;
    } else {
        return mb_y ? mode : HOR_PRED8x8;
    }
}

static av_always_inline
int check_intra_pred8x8_mode(int mode, int mb_x, int mb_y)
{
    if (mode == DC_PRED8x8) {
        return check_dc_pred8x8_mode(mode, mb_x, mb_y);
    } else {
        return mode;
    }
}

static av_always_inline
int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y)
{
    switch (mode) {
    case DC_PRED8x8:
        return check_dc_pred8x8_mode(mode, mb_x, mb_y);
    case VERT_PRED8x8:
        return !mb_y ? DC_127_PRED8x8 : mode;
    case HOR_PRED8x8:
        return !mb_x ? DC_129_PRED8x8 : mode;
    case PLANE_PRED8x8 /*TM*/:
        return check_tm_pred8x8_mode(mode, mb_x, mb_y);
    }
    return mode;
}

static av_always_inline
int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y)
{
    if (!mb_x) {
        return mb_y ? VERT_VP8_PRED : DC_129_PRED;
    } else {
        return mb_y ? mode : HOR_VP8_PRED;
    }
}

static av_always_inline
int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y, int *copy_buf)
{
    switch (mode) {
    case VERT_PRED:
        if (!mb_x && mb_y) {
            *copy_buf = 1;
            return mode;
        }
        /* fall-through */
    case DIAG_DOWN_LEFT_PRED:
    case VERT_LEFT_PRED:
        return !mb_y ? DC_127_PRED : mode;
    case HOR_PRED:
        if (!mb_y) {
            *copy_buf = 1;
            return mode;
        }
        /* fall-through */
    case HOR_UP_PRED:
        return !mb_x ? DC_129_PRED : mode;
    case TM_VP8_PRED:
        return check_tm_pred4x4_mode(mode, mb_x, mb_y);
    case DC_PRED: // 4x4 DC doesn't use the same "H.264-style" exceptions as 16x16/8x8 DC
    case DIAG_DOWN_RIGHT_PRED:
    case VERT_RIGHT_PRED:
    case HOR_DOWN_PRED:
        if (!mb_y || !mb_x)
            *copy_buf = 1;
        return mode;
    }
    return mode;
}

static av_always_inline
void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
                   int mb_x, int mb_y)
{
    AVCodecContext *avctx = s->avctx;
    int x, y, mode, nnz, tr;

    // for the first row, we need to run xchg_mb_border to init the top edge to 127
    // otherwise, skip it if we aren't going to deblock
    if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y))
        xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2],
                       s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
                       s->filter.simple, 1);

    if (mb->mode < MODE_I4x4) {
        if (avctx->flags & CODEC_FLAG_EMU_EDGE) { // tested
            mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y);
        } else {
            mode = check_intra_pred8x8_mode(mb->mode, mb_x, mb_y);
        }
        s->hpc.pred16x16[mode](dst[0], s->linesize);
    } else {
        uint8_t *ptr = dst[0];
        uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
        uint8_t tr_top[4] = { 127, 127, 127, 127 };

        // all blocks on the right edge of the macroblock use bottom edge
        // the top macroblock for their topright edge
        uint8_t *tr_right = ptr - s->linesize + 16;

        // if we're on the right edge of the frame, said edge is extended
        // from the top macroblock
        if (!(!mb_y && avctx->flags & CODEC_FLAG_EMU_EDGE) &&
            mb_x == s->mb_width-1) {
            tr = tr_right[-1]*0x01010101;
            tr_right = (uint8_t *)&tr;
        }

        if (mb->skip)
            AV_ZERO128(s->non_zero_count_cache);

        for (y = 0; y < 4; y++) {
            uint8_t *topright = ptr + 4 - s->linesize;
            for (x = 0; x < 4; x++) {
                int copy = 0, linesize = s->linesize;
                uint8_t *dst = ptr+4*x;
                DECLARE_ALIGNED(4, uint8_t, copy_dst)[5*8];

                if ((y == 0 || x == 3) && mb_y == 0 && avctx->flags & CODEC_FLAG_EMU_EDGE) {
                    topright = tr_top;
                } else if (x == 3)
                    topright = tr_right;

                if (avctx->flags & CODEC_FLAG_EMU_EDGE) { // mb_x+x or mb_y+y is a hack but works
                    mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x, mb_y + y, &copy);
                    if (copy) {
                        dst = copy_dst + 12;
                        linesize = 8;
                        if (!(mb_y + y)) {
                            copy_dst[3] = 127U;
                            AV_WN32A(copy_dst+4, 127U * 0x01010101U);
                        } else {
                            AV_COPY32(copy_dst+4, ptr+4*x-s->linesize);
                            if (!(mb_x + x)) {
                                copy_dst[3] = 129U;
                            } else {
                                copy_dst[3] = ptr[4*x-s->linesize-1];
                            }
                        }
                        if (!(mb_x + x)) {
                            copy_dst[11] =
                            copy_dst[19] =
                            copy_dst[27] =
                            copy_dst[35] = 129U;
                        } else {
                            copy_dst[11] = ptr[4*x              -1];
                            copy_dst[19] = ptr[4*x+s->linesize  -1];
                            copy_dst[27] = ptr[4*x+s->linesize*2-1];
                            copy_dst[35] = ptr[4*x+s->linesize*3-1];
                        }
                    }
                } else {
                    mode = intra4x4[x];
                }
                s->hpc.pred4x4[mode](dst, topright, linesize);
                if (copy) {
                    AV_COPY32(ptr+4*x              , copy_dst+12);
                    AV_COPY32(ptr+4*x+s->linesize  , copy_dst+20);
                    AV_COPY32(ptr+4*x+s->linesize*2, copy_dst+28);
                    AV_COPY32(ptr+4*x+s->linesize*3, copy_dst+36);
                }

                nnz = s->non_zero_count_cache[y][x];
                if (nnz) {
                    if (nnz == 1)
                        s->vp8dsp.vp8_idct_dc_add(ptr+4*x, s->block[y][x], s->linesize);
                    else
                        s->vp8dsp.vp8_idct_add(ptr+4*x, s->block[y][x], s->linesize);
                }
                topright += 4;
            }

            ptr   += 4*s->linesize;
            intra4x4 += 4;
        }
    }

    if (avctx->flags & CODEC_FLAG_EMU_EDGE) {
        mode = check_intra_pred8x8_mode_emuedge(s->chroma_pred_mode, mb_x, mb_y);
    } else {
        mode = check_intra_pred8x8_mode(s->chroma_pred_mode, mb_x, mb_y);
    }
    s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
    s->hpc.pred8x8[mode](dst[2], s->uvlinesize);

    if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y))
        xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2],
                       s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
                       s->filter.simple, 0);
}

static const uint8_t subpel_idx[3][8] = {
    { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels,
                                // also function pointer index
    { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required
    { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels
};

/**
 * Generic MC function.
 *
 * @param s VP8 decoding context
 * @param luma 1 for luma (Y) planes, 0 for chroma (Cb/Cr) planes
 * @param dst target buffer for block data at block position
 * @param src reference picture buffer at origin (0, 0)
 * @param mv motion vector (relative to block position) to get pixel data from
 * @param x_off horizontal position of block from origin (0, 0)
 * @param y_off vertical position of block from origin (0, 0)
 * @param block_w width of block (16, 8 or 4)
 * @param block_h height of block (always same as block_w)
 * @param width width of src/dst plane data
 * @param height height of src/dst plane data
 * @param linesize size of a single line of plane data, including padding
 * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
 */
static av_always_inline
void vp8_mc_luma(VP8Context *s, uint8_t *dst, uint8_t *src, const VP56mv *mv,
                 int x_off, int y_off, int block_w, int block_h,
                 int width, int height, int linesize,
                 vp8_mc_func mc_func[3][3])
{
    if (AV_RN32A(mv)) {

        int mx = (mv->x << 1)&7, mx_idx = subpel_idx[0][mx];
        int my = (mv->y << 1)&7, my_idx = subpel_idx[0][my];

        x_off += mv->x >> 2;
        y_off += mv->y >> 2;

        // edge emulation
        src += y_off * linesize + x_off;
        if (x_off < mx_idx || x_off >= width  - block_w - subpel_idx[2][mx] ||
            y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
            s->dsp.emulated_edge_mc(s->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize,
                                    block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
                                    x_off - mx_idx, y_off - my_idx, width, height);
            src = s->edge_emu_buffer + mx_idx + linesize * my_idx;
        }
        mc_func[my_idx][mx_idx](dst, linesize, src, linesize, block_h, mx, my);
    } else
        mc_func[0][0](dst, linesize, src + y_off * linesize + x_off, linesize, block_h, 0, 0);
}

static av_always_inline
void vp8_mc_chroma(VP8Context *s, uint8_t *dst1, uint8_t *dst2, uint8_t *src1,
                   uint8_t *src2, const VP56mv *mv, int x_off, int y_off,
                   int block_w, int block_h, int width, int height, int linesize,
                   vp8_mc_func mc_func[3][3])
{
    if (AV_RN32A(mv)) {
        int mx = mv->x&7, mx_idx = subpel_idx[0][mx];
        int my = mv->y&7, my_idx = subpel_idx[0][my];

        x_off += mv->x >> 3;
        y_off += mv->y >> 3;

        // edge emulation
        src1 += y_off * linesize + x_off;
        src2 += y_off * linesize + x_off;
        if (x_off < mx_idx || x_off >= width  - block_w - subpel_idx[2][mx] ||
            y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
            s->dsp.emulated_edge_mc(s->edge_emu_buffer, src1 - my_idx * linesize - mx_idx, linesize,
                                    block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
                                    x_off - mx_idx, y_off - my_idx, width, height);
            src1 = s->edge_emu_buffer + mx_idx + linesize * my_idx;
            mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);

            s->dsp.emulated_edge_mc(s->edge_emu_buffer, src2 - my_idx * linesize - mx_idx, linesize,
                                    block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my],
                                    x_off - mx_idx, y_off - my_idx, width, height);
            src2 = s->edge_emu_buffer + mx_idx + linesize * my_idx;
            mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
        } else {
            mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
            mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
        }
    } else {
        mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0);
        mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0);
    }
}

static av_always_inline
void vp8_mc_part(VP8Context *s, uint8_t *dst[3],
                 AVFrame *ref_frame, int x_off, int y_off,
                 int bx_off, int by_off,
                 int block_w, int block_h,
                 int width, int height, VP56mv *mv)
{
    VP56mv uvmv = *mv;

    /* Y */
    vp8_mc_luma(s, dst[0] + by_off * s->linesize + bx_off,
                ref_frame->data[0], mv, x_off + bx_off, y_off + by_off,
                block_w, block_h, width, height, s->linesize,
                s->put_pixels_tab[block_w == 8]);

    /* U/V */
    if (s->profile == 3) {
        uvmv.x &= ~7;
        uvmv.y &= ~7;
    }
    x_off   >>= 1; y_off   >>= 1;
    bx_off  >>= 1; by_off  >>= 1;
    width   >>= 1; height  >>= 1;
    block_w >>= 1; block_h >>= 1;
    vp8_mc_chroma(s, dst[1] + by_off * s->uvlinesize + bx_off,
                  dst[2] + by_off * s->uvlinesize + bx_off, ref_frame->data[1],
                  ref_frame->data[2], &uvmv, x_off + bx_off, y_off + by_off,
                  block_w, block_h, width, height, s->uvlinesize,
                  s->put_pixels_tab[1 + (block_w == 4)]);
}

/* Fetch pixels for estimated mv 4 macroblocks ahead.
 * Optimized for 64-byte cache lines.  Inspired by ffh264 prefetch_motion. */
static av_always_inline void prefetch_motion(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, int mb_xy, int ref)
{
    /* Don't prefetch refs that haven't been used very often this frame. */
    if (s->ref_count[ref-1] > (mb_xy >> 5)) {
        int x_off = mb_x << 4, y_off = mb_y << 4;
        int mx = (mb->mv.x>>2) + x_off + 8;
        int my = (mb->mv.y>>2) + y_off;
        uint8_t **src= s->framep[ref]->data;
        int off= mx + (my + (mb_x&3)*4)*s->linesize + 64;
        s->dsp.prefetch(src[0]+off, s->linesize, 4);
        off= (mx>>1) + ((my>>1) + (mb_x&7))*s->uvlinesize + 64;
        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
    }
}

/**
 * Apply motion vectors to prediction buffer, chapter 18.
 */
static av_always_inline
void inter_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
                   int mb_x, int mb_y)
{
    int x_off = mb_x << 4, y_off = mb_y << 4;
    int width = 16*s->mb_width, height = 16*s->mb_height;
    AVFrame *ref = s->framep[mb->ref_frame];
    VP56mv *bmv = mb->bmv;

    switch (mb->partitioning) {
    case VP8_SPLITMVMODE_NONE:
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    0, 0, 16, 16, width, height, &mb->mv);
        break;
    case VP8_SPLITMVMODE_4x4: {
        int x, y;
        VP56mv uvmv;

        /* Y */
        for (y = 0; y < 4; y++) {
            for (x = 0; x < 4; x++) {
                vp8_mc_luma(s, dst[0] + 4*y*s->linesize + x*4,
                            ref->data[0], &bmv[4*y + x],
                            4*x + x_off, 4*y + y_off, 4, 4,
                            width, height, s->linesize,
                            s->put_pixels_tab[2]);
            }
        }

        /* U/V */
        x_off >>= 1; y_off >>= 1; width >>= 1; height >>= 1;
        for (y = 0; y < 2; y++) {
            for (x = 0; x < 2; x++) {
                uvmv.x = mb->bmv[ 2*y    * 4 + 2*x  ].x +
                         mb->bmv[ 2*y    * 4 + 2*x+1].x +
                         mb->bmv[(2*y+1) * 4 + 2*x  ].x +
                         mb->bmv[(2*y+1) * 4 + 2*x+1].x;
                uvmv.y = mb->bmv[ 2*y    * 4 + 2*x  ].y +
                         mb->bmv[ 2*y    * 4 + 2*x+1].y +
                         mb->bmv[(2*y+1) * 4 + 2*x  ].y +
                         mb->bmv[(2*y+1) * 4 + 2*x+1].y;
                uvmv.x = (uvmv.x + 2 + (uvmv.x >> (INT_BIT-1))) >> 2;
                uvmv.y = (uvmv.y + 2 + (uvmv.y >> (INT_BIT-1))) >> 2;
                if (s->profile == 3) {
                    uvmv.x &= ~7;
                    uvmv.y &= ~7;
                }
                vp8_mc_chroma(s, dst[1] + 4*y*s->uvlinesize + x*4,
                              dst[2] + 4*y*s->uvlinesize + x*4,
                              ref->data[1], ref->data[2], &uvmv,
                              4*x + x_off, 4*y + y_off, 4, 4,
                              width, height, s->uvlinesize,
                              s->put_pixels_tab[2]);
            }
        }
        break;
    }
    case VP8_SPLITMVMODE_16x8:
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    0, 0, 16, 8, width, height, &bmv[0]);
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    0, 8, 16, 8, width, height, &bmv[1]);
        break;
    case VP8_SPLITMVMODE_8x16:
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    0, 0, 8, 16, width, height, &bmv[0]);
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    8, 0, 8, 16, width, height, &bmv[1]);
        break;
    case VP8_SPLITMVMODE_8x8:
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    0, 0, 8, 8, width, height, &bmv[0]);
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    8, 0, 8, 8, width, height, &bmv[1]);
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    0, 8, 8, 8, width, height, &bmv[2]);
        vp8_mc_part(s, dst, ref, x_off, y_off,
                    8, 8, 8, 8, width, height, &bmv[3]);
        break;
    }
}

static av_always_inline void idct_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb)
{
    int x, y, ch;

    if (mb->mode != MODE_I4x4) {
        uint8_t *y_dst = dst[0];
        for (y = 0; y < 4; y++) {
            uint32_t nnz4 = AV_RL32(s->non_zero_count_cache[y]);
            if (nnz4) {
                if (nnz4&~0x01010101) {
                    for (x = 0; x < 4; x++) {
                        if ((uint8_t)nnz4 == 1)
                            s->vp8dsp.vp8_idct_dc_add(y_dst+4*x, s->block[y][x], s->linesize);
                        else if((uint8_t)nnz4 > 1)
                            s->vp8dsp.vp8_idct_add(y_dst+4*x, s->block[y][x], s->linesize);
                        nnz4 >>= 8;
                        if (!nnz4)
                            break;
                    }
                } else {
                    s->vp8dsp.vp8_idct_dc_add4y(y_dst, s->block[y], s->linesize);
                }
            }
            y_dst += 4*s->linesize;
        }
    }

    for (ch = 0; ch < 2; ch++) {
        uint32_t nnz4 = AV_RL32(s->non_zero_count_cache[4+ch]);
        if (nnz4) {
            uint8_t *ch_dst = dst[1+ch];
            if (nnz4&~0x01010101) {
                for (y = 0; y < 2; y++) {
                    for (x = 0; x < 2; x++) {
                        if ((uint8_t)nnz4 == 1)
                            s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
                        else if((uint8_t)nnz4 > 1)
                            s->vp8dsp.vp8_idct_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
                        nnz4 >>= 8;
                        if (!nnz4)
                            goto chroma_idct_end;
                    }
                    ch_dst += 4*s->uvlinesize;
                }
            } else {
                s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, s->block[4+ch], s->uvlinesize);
            }
        }
chroma_idct_end: ;
    }
}

static av_always_inline void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb, VP8FilterStrength *f )
{
    int interior_limit, filter_level;

    if (s->segmentation.enabled) {
        filter_level = s->segmentation.filter_level[s->segment];
        if (!s->segmentation.absolute_vals)
            filter_level += s->filter.level;
    } else
        filter_level = s->filter.level;

    if (s->lf_delta.enabled) {
        filter_level += s->lf_delta.ref[mb->ref_frame];
        filter_level += s->lf_delta.mode[mb->mode];
    }

/* Like av_clip for inputs 0 and max, where max is equal to (2^n-1) */
#define POW2CLIP(x,max) (((x) & ~max) ? (-(x))>>31 & max : (x));
    filter_level = POW2CLIP(filter_level, 63);

    interior_limit = filter_level;
    if (s->filter.sharpness) {
        interior_limit >>= (s->filter.sharpness + 3) >> 2;
        interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
    }
    interior_limit = FFMAX(interior_limit, 1);

    f->filter_level = filter_level;
    f->inner_limit = interior_limit;
    f->inner_filter = !mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT;
}

static av_always_inline void filter_mb(VP8Context *s, uint8_t *dst[3], VP8FilterStrength *f, int mb_x, int mb_y)
{
    int mbedge_lim, bedge_lim, hev_thresh;
    int filter_level = f->filter_level;
    int inner_limit = f->inner_limit;
    int inner_filter = f->inner_filter;
    int linesize = s->linesize;
    int uvlinesize = s->uvlinesize;
    static const uint8_t hev_thresh_lut[2][64] = {
        { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
          2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
          3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
          3, 3, 3, 3 },
        { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
          1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
          2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
          2, 2, 2, 2 }
    };

    if (!filter_level)
        return;

     bedge_lim = 2*filter_level + inner_limit;
    mbedge_lim = bedge_lim + 4;

    hev_thresh = hev_thresh_lut[s->keyframe][filter_level];

    if (mb_x) {
        s->vp8dsp.vp8_h_loop_filter16y(dst[0],     linesize,
                                       mbedge_lim, inner_limit, hev_thresh);
        s->vp8dsp.vp8_h_loop_filter8uv(dst[1],     dst[2],      uvlinesize,
                                       mbedge_lim, inner_limit, hev_thresh);
    }

    if (inner_filter) {
        s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 4, linesize, bedge_lim,
                                             inner_limit, hev_thresh);
        s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 8, linesize, bedge_lim,
                                             inner_limit, hev_thresh);
        s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+12, linesize, bedge_lim,
                                             inner_limit, hev_thresh);
        s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4,
                                             uvlinesize,  bedge_lim,
                                             inner_limit, hev_thresh);
    }

    if (mb_y) {
        s->vp8dsp.vp8_v_loop_filter16y(dst[0],     linesize,
                                       mbedge_lim, inner_limit, hev_thresh);
        s->vp8dsp.vp8_v_loop_filter8uv(dst[1],     dst[2],      uvlinesize,
                                       mbedge_lim, inner_limit, hev_thresh);
    }

    if (inner_filter) {
        s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 4*linesize,
                                             linesize,    bedge_lim,
                                             inner_limit, hev_thresh);
        s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 8*linesize,
                                             linesize,    bedge_lim,
                                             inner_limit, hev_thresh);
        s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+12*linesize,
                                             linesize,    bedge_lim,
                                             inner_limit, hev_thresh);
        s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize,
                                             dst[2] + 4 * uvlinesize,
                                             uvlinesize,  bedge_lim,
                                             inner_limit, hev_thresh);
    }
}

static av_always_inline void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8FilterStrength *f, int mb_x, int mb_y)
{
    int mbedge_lim, bedge_lim;
    int filter_level = f->filter_level;
    int inner_limit = f->inner_limit;
    int inner_filter = f->inner_filter;
    int linesize = s->linesize;

    if (!filter_level)
        return;

     bedge_lim = 2*filter_level + inner_limit;
    mbedge_lim = bedge_lim + 4;

    if (mb_x)
        s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim);
    if (inner_filter) {
        s->vp8dsp.vp8_h_loop_filter_simple(dst+ 4, linesize, bedge_lim);
        s->vp8dsp.vp8_h_loop_filter_simple(dst+ 8, linesize, bedge_lim);
        s->vp8dsp.vp8_h_loop_filter_simple(dst+12, linesize, bedge_lim);
    }

    if (mb_y)
        s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim);
    if (inner_filter) {
        s->vp8dsp.vp8_v_loop_filter_simple(dst+ 4*linesize, linesize, bedge_lim);
        s->vp8dsp.vp8_v_loop_filter_simple(dst+ 8*linesize, linesize, bedge_lim);
        s->vp8dsp.vp8_v_loop_filter_simple(dst+12*linesize, linesize, bedge_lim);
    }
}

static void filter_mb_row(VP8Context *s, int mb_y)
{
    VP8FilterStrength *f = s->filter_strength;
    uint8_t *dst[3] = {
        s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize,
        s->framep[VP56_FRAME_CURRENT]->data[1] +  8*mb_y*s->uvlinesize,
        s->framep[VP56_FRAME_CURRENT]->data[2] +  8*mb_y*s->uvlinesize
    };
    int mb_x;

    for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
        backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
        filter_mb(s, dst, f++, mb_x, mb_y);
        dst[0] += 16;
        dst[1] += 8;
        dst[2] += 8;
    }
}

static void filter_mb_row_simple(VP8Context *s, int mb_y)
{
    VP8FilterStrength *f = s->filter_strength;
    uint8_t *dst = s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize;
    int mb_x;

    for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
        backup_mb_border(s->top_border[mb_x+1], dst, NULL, NULL, s->linesize, 0, 1);
        filter_mb_simple(s, dst, f++, mb_x, mb_y);
        dst += 16;
    }
}

static int vp8_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
                            AVPacket *avpkt)
{
    VP8Context *s = avctx->priv_data;
    int ret, mb_x, mb_y, i, y, referenced;
    enum AVDiscard skip_thresh;
    AVFrame *av_uninit(curframe);

    if ((ret = decode_frame_header(s, avpkt->data, avpkt->size)) < 0)
        return ret;

    referenced = s->update_last || s->update_golden == VP56_FRAME_CURRENT
                                || s->update_altref == VP56_FRAME_CURRENT;

    skip_thresh = !referenced ? AVDISCARD_NONREF :
                    !s->keyframe ? AVDISCARD_NONKEY : AVDISCARD_ALL;

    if (avctx->skip_frame >= skip_thresh) {
        s->invisible = 1;
        goto skip_decode;
    }
    s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh;

    for (i = 0; i < 4; i++)
        if (&s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
            &s->frames[i] != s->framep[VP56_FRAME_GOLDEN] &&
            &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) {
            curframe = s->framep[VP56_FRAME_CURRENT] = &s->frames[i];
            break;
        }
    if (curframe->data[0])
        avctx->release_buffer(avctx, curframe);

    curframe->key_frame = s->keyframe;
    curframe->pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
    curframe->reference = referenced ? 3 : 0;
    if ((ret = avctx->get_buffer(avctx, curframe))) {
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed!\n");
        return ret;
    }

    // Given that arithmetic probabilities are updated every frame, it's quite likely
    // that the values we have on a random interframe are complete junk if we didn't
    // start decode on a keyframe. So just don't display anything rather than junk.
    if (!s->keyframe && (!s->framep[VP56_FRAME_PREVIOUS] ||
                         !s->framep[VP56_FRAME_GOLDEN] ||
                         !s->framep[VP56_FRAME_GOLDEN2])) {
        av_log(avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n");
        return AVERROR_INVALIDDATA;
    }

    s->linesize   = curframe->linesize[0];
    s->uvlinesize = curframe->linesize[1];

    if (!s->edge_emu_buffer)
        s->edge_emu_buffer = av_malloc(21*s->linesize);

    memset(s->top_nnz, 0, s->mb_width*sizeof(*s->top_nnz));

    /* Zero macroblock structures for top/top-left prediction from outside the frame. */
    memset(s->macroblocks + s->mb_height*2 - 1, 0, (s->mb_width+1)*sizeof(*s->macroblocks));

    // top edge of 127 for intra prediction
    if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
        s->top_border[0][15] = s->top_border[0][23] = 127;
        memset(s->top_border[1]-1, 127, s->mb_width*sizeof(*s->top_border)+1);
    }
    memset(s->ref_count, 0, sizeof(s->ref_count));
    if (s->keyframe)
        memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width*4);

    #define MARGIN (16 << 2)
    s->mv_min.y = -MARGIN;
    s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;

    for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
        VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions-1)];
        VP8Macroblock *mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
        int mb_xy = mb_y*s->mb_width;
        uint8_t *dst[3] = {
            curframe->data[0] + 16*mb_y*s->linesize,
            curframe->data[1] +  8*mb_y*s->uvlinesize,
            curframe->data[2] +  8*mb_y*s->uvlinesize
        };

        memset(mb - 1, 0, sizeof(*mb));   // zero left macroblock
        memset(s->left_nnz, 0, sizeof(s->left_nnz));
        AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101);

        // left edge of 129 for intra prediction
        if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
            for (i = 0; i < 3; i++)
                for (y = 0; y < 16>>!!i; y++)
                    dst[i][y*curframe->linesize[i]-1] = 129;
            if (mb_y == 1) // top left edge is also 129
                s->top_border[0][15] = s->top_border[0][23] = s->top_border[0][31] = 129;
        }

        s->mv_min.x = -MARGIN;
        s->mv_max.x = ((s->mb_width  - 1) << 6) + MARGIN;

        for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
            /* Prefetch the current frame, 4 MBs ahead */
            s->dsp.prefetch(dst[0] + (mb_x&3)*4*s->linesize + 64, s->linesize, 4);
            s->dsp.prefetch(dst[1] + (mb_x&7)*s->uvlinesize + 64, dst[2] - dst[1], 2);

            decode_mb_mode(s, mb, mb_x, mb_y, s->segmentation_map + mb_xy);

            prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);

            if (!mb->skip)
                decode_mb_coeffs(s, c, mb, s->top_nnz[mb_x], s->left_nnz);

            if (mb->mode <= MODE_I4x4)
                intra_predict(s, dst, mb, mb_x, mb_y);
            else
                inter_predict(s, dst, mb, mb_x, mb_y);

            prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN);

            if (!mb->skip) {
                idct_mb(s, dst, mb);
            } else {
                AV_ZERO64(s->left_nnz);
                AV_WN64(s->top_nnz[mb_x], 0);   // array of 9, so unaligned

                // Reset DC block predictors if they would exist if the mb had coefficients
                if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
                    s->left_nnz[8]      = 0;
                    s->top_nnz[mb_x][8] = 0;
                }
            }

            if (s->deblock_filter)
                filter_level_for_mb(s, mb, &s->filter_strength[mb_x]);

            prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2);

            dst[0] += 16;
            dst[1] += 8;
            dst[2] += 8;
            s->mv_min.x -= 64;
            s->mv_max.x -= 64;
        }
        if (s->deblock_filter) {
            if (s->filter.simple)
                filter_mb_row_simple(s, mb_y);
            else
                filter_mb_row(s, mb_y);
        }
        s->mv_min.y -= 64;
        s->mv_max.y -= 64;
    }

skip_decode:
    // if future frames don't use the updated probabilities,
    // reset them to the values we saved
    if (!s->update_probabilities)
        s->prob[0] = s->prob[1];

    // check if golden and altref are swapped
    if (s->update_altref == VP56_FRAME_GOLDEN &&
        s->update_golden == VP56_FRAME_GOLDEN2)
        FFSWAP(AVFrame *, s->framep[VP56_FRAME_GOLDEN], s->framep[VP56_FRAME_GOLDEN2]);
    else {
        if (s->update_altref != VP56_FRAME_NONE)
            s->framep[VP56_FRAME_GOLDEN2] = s->framep[s->update_altref];

        if (s->update_golden != VP56_FRAME_NONE)
            s->framep[VP56_FRAME_GOLDEN] = s->framep[s->update_golden];
    }

    if (s->update_last) // move cur->prev
        s->framep[VP56_FRAME_PREVIOUS] = s->framep[VP56_FRAME_CURRENT];

    // release no longer referenced frames
    for (i = 0; i < 4; i++)
        if (s->frames[i].data[0] &&
            &s->frames[i] != s->framep[VP56_FRAME_CURRENT] &&
            &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
            &s->frames[i] != s->framep[VP56_FRAME_GOLDEN] &&
            &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2])
            avctx->release_buffer(avctx, &s->frames[i]);

    if (!s->invisible) {
        *(AVFrame*)data = *s->framep[VP56_FRAME_CURRENT];
        *data_size = sizeof(AVFrame);
    }

    return avpkt->size;
}

static av_cold int vp8_decode_init(AVCodecContext *avctx)
{
    VP8Context *s = avctx->priv_data;

    s->avctx = avctx;
    avctx->pix_fmt = PIX_FMT_YUV420P;

    dsputil_init(&s->dsp, avctx);
    ff_h264_pred_init(&s->hpc, CODEC_ID_VP8, 8);
    ff_vp8dsp_init(&s->vp8dsp);

    return 0;
}

static av_cold int vp8_decode_free(AVCodecContext *avctx)
{
    vp8_decode_flush(avctx);
    return 0;
}

AVCodec ff_vp8_decoder = {
    "vp8",
    AVMEDIA_TYPE_VIDEO,
    CODEC_ID_VP8,
    sizeof(VP8Context),
    vp8_decode_init,
    NULL,
    vp8_decode_free,
    vp8_decode_frame,
    CODEC_CAP_DR1,
    .flush = vp8_decode_flush,
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP8"),
};