Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ e6e32bdc

History | View | Annotate | Download (85.2 KB)

1
/*
2
 * Copyright (C) 2003-2004 the ffmpeg project
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

    
21
/**
22
 * @file libavcodec/vp3.c
23
 * On2 VP3 Video Decoder
24
 *
25
 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26
 * For more information about the VP3 coding process, visit:
27
 *   http://wiki.multimedia.cx/index.php?title=On2_VP3
28
 *
29
 * Theora decoder by Alex Beregszaszi
30
 */
31

    
32
#include <stdio.h>
33
#include <stdlib.h>
34
#include <string.h>
35

    
36
#include "avcodec.h"
37
#include "dsputil.h"
38
#include "get_bits.h"
39

    
40
#include "vp3data.h"
41
#include "xiph.h"
42

    
43
#define FRAGMENT_PIXELS 8
44

    
45
static av_cold int vp3_decode_end(AVCodecContext *avctx);
46

    
47
typedef struct Coeff {
48
    struct Coeff *next;
49
    DCTELEM coeff;
50
    uint8_t index;
51
} Coeff;
52

    
53
//FIXME split things out into their own arrays
54
typedef struct Vp3Fragment {
55
    Coeff *next_coeff;
56
    /* address of first pixel taking into account which plane the fragment
57
     * lives on as well as the plane stride */
58
    int first_pixel;
59
    /* this is the macroblock that the fragment belongs to */
60
    uint16_t macroblock;
61
    uint8_t coding_method;
62
    int8_t motion_x;
63
    int8_t motion_y;
64
    uint8_t qpi;
65
} Vp3Fragment;
66

    
67
#define SB_NOT_CODED        0
68
#define SB_PARTIALLY_CODED  1
69
#define SB_FULLY_CODED      2
70

    
71
#define MODE_INTER_NO_MV      0
72
#define MODE_INTRA            1
73
#define MODE_INTER_PLUS_MV    2
74
#define MODE_INTER_LAST_MV    3
75
#define MODE_INTER_PRIOR_LAST 4
76
#define MODE_USING_GOLDEN     5
77
#define MODE_GOLDEN_MV        6
78
#define MODE_INTER_FOURMV     7
79
#define CODING_MODE_COUNT     8
80

    
81
/* special internal mode */
82
#define MODE_COPY             8
83

    
84
/* There are 6 preset schemes, plus a free-form scheme */
85
static const int ModeAlphabet[6][CODING_MODE_COUNT] =
86
{
87
    /* scheme 1: Last motion vector dominates */
88
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
89
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
90
         MODE_INTRA,            MODE_USING_GOLDEN,
91
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
92

    
93
    /* scheme 2 */
94
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
95
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
96
         MODE_INTRA,            MODE_USING_GOLDEN,
97
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
98

    
99
    /* scheme 3 */
100
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
101
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
102
         MODE_INTRA,            MODE_USING_GOLDEN,
103
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
104

    
105
    /* scheme 4 */
106
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
107
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
108
         MODE_INTRA,            MODE_USING_GOLDEN,
109
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
110

    
111
    /* scheme 5: No motion vector dominates */
112
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
113
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
114
         MODE_INTRA,            MODE_USING_GOLDEN,
115
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
116

    
117
    /* scheme 6 */
118
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
119
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
120
         MODE_INTER_PLUS_MV,    MODE_INTRA,
121
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
122

    
123
};
124

    
125
#define MIN_DEQUANT_VAL 2
126

    
127
typedef struct Vp3DecodeContext {
128
    AVCodecContext *avctx;
129
    int theora, theora_tables;
130
    int version;
131
    int width, height;
132
    AVFrame golden_frame;
133
    AVFrame last_frame;
134
    AVFrame current_frame;
135
    int keyframe;
136
    DSPContext dsp;
137
    int flipped_image;
138

    
139
    int qps[3];
140
    int nqps;
141
    int last_qps[3];
142

    
143
    int superblock_count;
144
    int y_superblock_width;
145
    int y_superblock_height;
146
    int c_superblock_width;
147
    int c_superblock_height;
148
    int u_superblock_start;
149
    int v_superblock_start;
150
    unsigned char *superblock_coding;
151

    
152
    int macroblock_count;
153
    int macroblock_width;
154
    int macroblock_height;
155

    
156
    int fragment_count;
157
    int fragment_width;
158
    int fragment_height;
159

    
160
    Vp3Fragment *all_fragments;
161
    uint8_t *coeff_counts;
162
    Coeff *coeffs;
163
    Coeff *next_coeff;
164
    int fragment_start[3];
165

    
166
    ScanTable scantable;
167

    
168
    /* tables */
169
    uint16_t coded_dc_scale_factor[64];
170
    uint32_t coded_ac_scale_factor[64];
171
    uint8_t base_matrix[384][64];
172
    uint8_t qr_count[2][3];
173
    uint8_t qr_size [2][3][64];
174
    uint16_t qr_base[2][3][64];
175

    
176
    /* this is a list of indexes into the all_fragments array indicating
177
     * which of the fragments are coded */
178
    int *coded_fragment_list;
179
    int coded_fragment_list_index;
180
    int pixel_addresses_initialized;
181

    
182
    /* track which fragments have already been decoded; called 'fast'
183
     * because this data structure avoids having to iterate through every
184
     * fragment in coded_fragment_list; once a fragment has been fully
185
     * decoded, it is removed from this list */
186
    int *fast_fragment_list;
187
    int fragment_list_y_head;
188
    int fragment_list_c_head;
189

    
190
    VLC dc_vlc[16];
191
    VLC ac_vlc_1[16];
192
    VLC ac_vlc_2[16];
193
    VLC ac_vlc_3[16];
194
    VLC ac_vlc_4[16];
195

    
196
    VLC superblock_run_length_vlc;
197
    VLC fragment_run_length_vlc;
198
    VLC mode_code_vlc;
199
    VLC motion_vector_vlc;
200

    
201
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
202
     * index into them */
203
    DECLARE_ALIGNED_16(int16_t, qmat[3][2][3][64]);     //<qmat[qpi][is_inter][plane]
204

    
205
    /* This table contains superblock_count * 16 entries. Each set of 16
206
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
207
     * An entry will be -1 to indicate that no entry corresponds to that
208
     * index. */
209
    int *superblock_fragments;
210

    
211
    /* This table contains superblock_count * 4 entries. Each set of 4
212
     * numbers corresponds to the macroblock indexes 0..3 of the superblock.
213
     * An entry will be -1 to indicate that no entry corresponds to that
214
     * index. */
215
    int *superblock_macroblocks;
216

    
217
    /* This table contains macroblock_count * 6 entries. Each set of 6
218
     * numbers corresponds to the fragment indexes 0..5 which comprise
219
     * the macroblock (4 Y fragments and 2 C fragments). */
220
    int *macroblock_fragments;
221
    /* This is an array that indicates how a particular macroblock
222
     * is coded. */
223
    unsigned char *macroblock_coding;
224

    
225
    int first_coded_y_fragment;
226
    int first_coded_c_fragment;
227
    int last_coded_y_fragment;
228
    int last_coded_c_fragment;
229

    
230
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
231
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
232

    
233
    /* Huffman decode */
234
    int hti;
235
    unsigned int hbits;
236
    int entries;
237
    int huff_code_size;
238
    uint16_t huffman_table[80][32][2];
239

    
240
    uint8_t filter_limit_values[64];
241
    DECLARE_ALIGNED_8(int, bounding_values_array[256+2]);
242
} Vp3DecodeContext;
243

    
244
/************************************************************************
245
 * VP3 specific functions
246
 ************************************************************************/
247

    
248
/*
249
 * This function sets up all of the various blocks mappings:
250
 * superblocks <-> fragments, macroblocks <-> fragments,
251
 * superblocks <-> macroblocks
252
 *
253
 * Returns 0 is successful; returns 1 if *anything* went wrong.
254
 */
255
static int init_block_mapping(Vp3DecodeContext *s)
256
{
257
    int i, j;
258
    signed int hilbert_walk_mb[4];
259

    
260
    int current_fragment = 0;
261
    int current_width = 0;
262
    int current_height = 0;
263
    int right_edge = 0;
264
    int bottom_edge = 0;
265
    int superblock_row_inc = 0;
266
    int mapping_index = 0;
267

    
268
    int current_macroblock;
269
    int c_fragment;
270

    
271
    static const signed char travel_width[16] = {
272
         1,  1,  0, -1,
273
         0,  0,  1,  0,
274
         1,  0,  1,  0,
275
         0, -1,  0,  1
276
    };
277

    
278
    static const signed char travel_height[16] = {
279
         0,  0,  1,  0,
280
         1,  1,  0, -1,
281
         0,  1,  0, -1,
282
        -1,  0, -1,  0
283
    };
284

    
285
    static const signed char travel_width_mb[4] = {
286
         1,  0,  1,  0
287
    };
288

    
289
    static const signed char travel_height_mb[4] = {
290
         0,  1,  0, -1
291
    };
292

    
293
    hilbert_walk_mb[0] = 1;
294
    hilbert_walk_mb[1] = s->macroblock_width;
295
    hilbert_walk_mb[2] = 1;
296
    hilbert_walk_mb[3] = -s->macroblock_width;
297

    
298
    /* iterate through each superblock (all planes) and map the fragments */
299
    for (i = 0; i < s->superblock_count; i++) {
300
        /* time to re-assign the limits? */
301
        if (i == 0) {
302

    
303
            /* start of Y superblocks */
304
            right_edge = s->fragment_width;
305
            bottom_edge = s->fragment_height;
306
            current_width = -1;
307
            current_height = 0;
308
            superblock_row_inc = 3 * s->fragment_width -
309
                (s->y_superblock_width * 4 - s->fragment_width);
310

    
311
            /* the first operation for this variable is to advance by 1 */
312
            current_fragment = -1;
313

    
314
        } else if (i == s->u_superblock_start) {
315

    
316
            /* start of U superblocks */
317
            right_edge = s->fragment_width / 2;
318
            bottom_edge = s->fragment_height / 2;
319
            current_width = -1;
320
            current_height = 0;
321
            superblock_row_inc = 3 * (s->fragment_width / 2) -
322
                (s->c_superblock_width * 4 - s->fragment_width / 2);
323

    
324
            /* the first operation for this variable is to advance by 1 */
325
            current_fragment = s->fragment_start[1] - 1;
326

    
327
        } else if (i == s->v_superblock_start) {
328

    
329
            /* start of V superblocks */
330
            right_edge = s->fragment_width / 2;
331
            bottom_edge = s->fragment_height / 2;
332
            current_width = -1;
333
            current_height = 0;
334
            superblock_row_inc = 3 * (s->fragment_width / 2) -
335
                (s->c_superblock_width * 4 - s->fragment_width / 2);
336

    
337
            /* the first operation for this variable is to advance by 1 */
338
            current_fragment = s->fragment_start[2] - 1;
339

    
340
        }
341

    
342
        if (current_width >= right_edge - 1) {
343
            /* reset width and move to next superblock row */
344
            current_width = -1;
345
            current_height += 4;
346

    
347
            /* fragment is now at the start of a new superblock row */
348
            current_fragment += superblock_row_inc;
349
        }
350

    
351
        /* iterate through all 16 fragments in a superblock */
352
        for (j = 0; j < 16; j++) {
353
            current_fragment += travel_width[j] + right_edge * travel_height[j];
354
            current_width += travel_width[j];
355
            current_height += travel_height[j];
356

    
357
            /* check if the fragment is in bounds */
358
            if ((current_width < right_edge) &&
359
                (current_height < bottom_edge)) {
360
                s->superblock_fragments[mapping_index] = current_fragment;
361
            } else {
362
                s->superblock_fragments[mapping_index] = -1;
363
            }
364

    
365
            mapping_index++;
366
        }
367
    }
368

    
369
    /* initialize the superblock <-> macroblock mapping; iterate through
370
     * all of the Y plane superblocks to build this mapping */
371
    right_edge = s->macroblock_width;
372
    bottom_edge = s->macroblock_height;
373
    current_width = -1;
374
    current_height = 0;
375
    superblock_row_inc = s->macroblock_width -
376
        (s->y_superblock_width * 2 - s->macroblock_width);
377
    mapping_index = 0;
378
    current_macroblock = -1;
379
    for (i = 0; i < s->u_superblock_start; i++) {
380

    
381
        if (current_width >= right_edge - 1) {
382
            /* reset width and move to next superblock row */
383
            current_width = -1;
384
            current_height += 2;
385

    
386
            /* macroblock is now at the start of a new superblock row */
387
            current_macroblock += superblock_row_inc;
388
        }
389

    
390
        /* iterate through each potential macroblock in the superblock */
391
        for (j = 0; j < 4; j++) {
392
            current_macroblock += hilbert_walk_mb[j];
393
            current_width += travel_width_mb[j];
394
            current_height += travel_height_mb[j];
395

    
396
            /* check if the macroblock is in bounds */
397
            if ((current_width < right_edge) &&
398
                (current_height < bottom_edge)) {
399
                s->superblock_macroblocks[mapping_index] = current_macroblock;
400
            } else {
401
                s->superblock_macroblocks[mapping_index] = -1;
402
            }
403

    
404
            mapping_index++;
405
        }
406
    }
407

    
408
    /* initialize the macroblock <-> fragment mapping */
409
    current_fragment = 0;
410
    current_macroblock = 0;
411
    mapping_index = 0;
412
    for (i = 0; i < s->fragment_height; i += 2) {
413

    
414
        for (j = 0; j < s->fragment_width; j += 2) {
415

    
416
            s->all_fragments[current_fragment].macroblock = current_macroblock;
417
            s->macroblock_fragments[mapping_index++] = current_fragment;
418

    
419
            if (j + 1 < s->fragment_width) {
420
                s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
421
                s->macroblock_fragments[mapping_index++] = current_fragment + 1;
422
            } else
423
                s->macroblock_fragments[mapping_index++] = -1;
424

    
425
            if (i + 1 < s->fragment_height) {
426
                s->all_fragments[current_fragment + s->fragment_width].macroblock =
427
                    current_macroblock;
428
                s->macroblock_fragments[mapping_index++] =
429
                    current_fragment + s->fragment_width;
430
            } else
431
                s->macroblock_fragments[mapping_index++] = -1;
432

    
433
            if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
434
                s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
435
                    current_macroblock;
436
                s->macroblock_fragments[mapping_index++] =
437
                    current_fragment + s->fragment_width + 1;
438
            } else
439
                s->macroblock_fragments[mapping_index++] = -1;
440

    
441
            /* C planes */
442
            c_fragment = s->fragment_start[1] +
443
                (i * s->fragment_width / 4) + (j / 2);
444
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
445
            s->macroblock_fragments[mapping_index++] = c_fragment;
446

    
447
            c_fragment = s->fragment_start[2] +
448
                (i * s->fragment_width / 4) + (j / 2);
449
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
450
            s->macroblock_fragments[mapping_index++] = c_fragment;
451

    
452
            if (j + 2 <= s->fragment_width)
453
                current_fragment += 2;
454
            else
455
                current_fragment++;
456
            current_macroblock++;
457
        }
458

    
459
        current_fragment += s->fragment_width;
460
    }
461

    
462
    return 0;  /* successful path out */
463
}
464

    
465
/*
466
 * This function wipes out all of the fragment data.
467
 */
468
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
469
{
470
    int i;
471

    
472
    /* zero out all of the fragment information */
473
    s->coded_fragment_list_index = 0;
474
    for (i = 0; i < s->fragment_count; i++) {
475
        s->coeff_counts[i] = 0;
476
        s->all_fragments[i].motion_x = 127;
477
        s->all_fragments[i].motion_y = 127;
478
        s->all_fragments[i].next_coeff= NULL;
479
        s->all_fragments[i].qpi = 0;
480
        s->coeffs[i].index=
481
        s->coeffs[i].coeff=0;
482
        s->coeffs[i].next= NULL;
483
    }
484
}
485

    
486
/*
487
 * This function sets up the dequantization tables used for a particular
488
 * frame.
489
 */
490
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
491
{
492
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
493
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
494
    int i, plane, inter, qri, bmi, bmj, qistart;
495

    
496
    for(inter=0; inter<2; inter++){
497
        for(plane=0; plane<3; plane++){
498
            int sum=0;
499
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
500
                sum+= s->qr_size[inter][plane][qri];
501
                if(s->qps[qpi] <= sum)
502
                    break;
503
            }
504
            qistart= sum - s->qr_size[inter][plane][qri];
505
            bmi= s->qr_base[inter][plane][qri  ];
506
            bmj= s->qr_base[inter][plane][qri+1];
507
            for(i=0; i<64; i++){
508
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
509
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
510
                            + s->qr_size[inter][plane][qri])
511
                           / (2*s->qr_size[inter][plane][qri]);
512

    
513
                int qmin= 8<<(inter + !i);
514
                int qscale= i ? ac_scale_factor : dc_scale_factor;
515

    
516
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
517
            }
518
            // all DC coefficients use the same quant so as not to interfere with DC prediction
519
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
520
        }
521
    }
522

    
523
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
524
}
525

    
526
/*
527
 * This function initializes the loop filter boundary limits if the frame's
528
 * quality index is different from the previous frame's.
529
 *
530
 * The filter_limit_values may not be larger than 127.
531
 */
532
static void init_loop_filter(Vp3DecodeContext *s)
533
{
534
    int *bounding_values= s->bounding_values_array+127;
535
    int filter_limit;
536
    int x;
537
    int value;
538

    
539
    filter_limit = s->filter_limit_values[s->qps[0]];
540

    
541
    /* set up the bounding values */
542
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
543
    for (x = 0; x < filter_limit; x++) {
544
        bounding_values[-x] = -x;
545
        bounding_values[x] = x;
546
    }
547
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
548
        bounding_values[ x] =  value;
549
        bounding_values[-x] = -value;
550
    }
551
    if (value)
552
        bounding_values[128] = value;
553
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
554
}
555

    
556
/*
557
 * This function unpacks all of the superblock/macroblock/fragment coding
558
 * information from the bitstream.
559
 */
560
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
561
{
562
    int bit = 0;
563
    int current_superblock = 0;
564
    int current_run = 0;
565
    int decode_fully_flags = 0;
566
    int decode_partial_blocks = 0;
567
    int first_c_fragment_seen;
568

    
569
    int i, j;
570
    int current_fragment;
571

    
572
    if (s->keyframe) {
573
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
574

    
575
    } else {
576

    
577
        /* unpack the list of partially-coded superblocks */
578
        bit = get_bits1(gb);
579
        /* toggle the bit because as soon as the first run length is
580
         * fetched the bit will be toggled again */
581
        bit ^= 1;
582
        while (current_superblock < s->superblock_count) {
583
            if (current_run-- == 0) {
584
                bit ^= 1;
585
                current_run = get_vlc2(gb,
586
                    s->superblock_run_length_vlc.table, 6, 2);
587
                if (current_run == 33)
588
                    current_run += get_bits(gb, 12);
589

    
590
                /* if any of the superblocks are not partially coded, flag
591
                 * a boolean to decode the list of fully-coded superblocks */
592
                if (bit == 0) {
593
                    decode_fully_flags = 1;
594
                } else {
595

    
596
                    /* make a note of the fact that there are partially coded
597
                     * superblocks */
598
                    decode_partial_blocks = 1;
599
                }
600
            }
601
            s->superblock_coding[current_superblock++] = bit;
602
        }
603

    
604
        /* unpack the list of fully coded superblocks if any of the blocks were
605
         * not marked as partially coded in the previous step */
606
        if (decode_fully_flags) {
607

    
608
            current_superblock = 0;
609
            current_run = 0;
610
            bit = get_bits1(gb);
611
            /* toggle the bit because as soon as the first run length is
612
             * fetched the bit will be toggled again */
613
            bit ^= 1;
614
            while (current_superblock < s->superblock_count) {
615

    
616
                /* skip any superblocks already marked as partially coded */
617
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
618

    
619
                    if (current_run-- == 0) {
620
                        bit ^= 1;
621
                        current_run = get_vlc2(gb,
622
                            s->superblock_run_length_vlc.table, 6, 2);
623
                        if (current_run == 33)
624
                            current_run += get_bits(gb, 12);
625
                    }
626
                    s->superblock_coding[current_superblock] = 2*bit;
627
                }
628
                current_superblock++;
629
            }
630
        }
631

    
632
        /* if there were partial blocks, initialize bitstream for
633
         * unpacking fragment codings */
634
        if (decode_partial_blocks) {
635

    
636
            current_run = 0;
637
            bit = get_bits1(gb);
638
            /* toggle the bit because as soon as the first run length is
639
             * fetched the bit will be toggled again */
640
            bit ^= 1;
641
        }
642
    }
643

    
644
    /* figure out which fragments are coded; iterate through each
645
     * superblock (all planes) */
646
    s->coded_fragment_list_index = 0;
647
    s->next_coeff= s->coeffs + s->fragment_count;
648
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
649
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
650
    first_c_fragment_seen = 0;
651
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
652
    for (i = 0; i < s->superblock_count; i++) {
653

    
654
        /* iterate through all 16 fragments in a superblock */
655
        for (j = 0; j < 16; j++) {
656

    
657
            /* if the fragment is in bounds, check its coding status */
658
            current_fragment = s->superblock_fragments[i * 16 + j];
659
            if (current_fragment >= s->fragment_count) {
660
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
661
                    current_fragment, s->fragment_count);
662
                return 1;
663
            }
664
            if (current_fragment != -1) {
665
                if (s->superblock_coding[i] == SB_NOT_CODED) {
666

    
667
                    /* copy all the fragments from the prior frame */
668
                    s->all_fragments[current_fragment].coding_method =
669
                        MODE_COPY;
670

    
671
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
672

    
673
                    /* fragment may or may not be coded; this is the case
674
                     * that cares about the fragment coding runs */
675
                    if (current_run-- == 0) {
676
                        bit ^= 1;
677
                        current_run = get_vlc2(gb,
678
                            s->fragment_run_length_vlc.table, 5, 2);
679
                    }
680

    
681
                    if (bit) {
682
                        /* default mode; actual mode will be decoded in
683
                         * the next phase */
684
                        s->all_fragments[current_fragment].coding_method =
685
                            MODE_INTER_NO_MV;
686
                        s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
687
                        s->coded_fragment_list[s->coded_fragment_list_index] =
688
                            current_fragment;
689
                        if ((current_fragment >= s->fragment_start[1]) &&
690
                            (s->last_coded_y_fragment == -1) &&
691
                            (!first_c_fragment_seen)) {
692
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
693
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
694
                            first_c_fragment_seen = 1;
695
                        }
696
                        s->coded_fragment_list_index++;
697
                        s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
698
                    } else {
699
                        /* not coded; copy this fragment from the prior frame */
700
                        s->all_fragments[current_fragment].coding_method =
701
                            MODE_COPY;
702
                    }
703

    
704
                } else {
705

    
706
                    /* fragments are fully coded in this superblock; actual
707
                     * coding will be determined in next step */
708
                    s->all_fragments[current_fragment].coding_method =
709
                        MODE_INTER_NO_MV;
710
                    s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
711
                    s->coded_fragment_list[s->coded_fragment_list_index] =
712
                        current_fragment;
713
                    if ((current_fragment >= s->fragment_start[1]) &&
714
                        (s->last_coded_y_fragment == -1) &&
715
                        (!first_c_fragment_seen)) {
716
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
717
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
718
                        first_c_fragment_seen = 1;
719
                    }
720
                    s->coded_fragment_list_index++;
721
                    s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
722
                }
723
            }
724
        }
725
    }
726

    
727
    if (!first_c_fragment_seen)
728
        /* only Y fragments coded in this frame */
729
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
730
    else
731
        /* end the list of coded C fragments */
732
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
733

    
734
    for (i = 0; i < s->fragment_count - 1; i++) {
735
        s->fast_fragment_list[i] = i + 1;
736
    }
737
    s->fast_fragment_list[s->fragment_count - 1] = -1;
738

    
739
    if (s->last_coded_y_fragment == -1)
740
        s->fragment_list_y_head = -1;
741
    else {
742
        s->fragment_list_y_head = s->first_coded_y_fragment;
743
        s->fast_fragment_list[s->last_coded_y_fragment] = -1;
744
    }
745

    
746
    if (s->last_coded_c_fragment == -1)
747
        s->fragment_list_c_head = -1;
748
    else {
749
        s->fragment_list_c_head = s->first_coded_c_fragment;
750
        s->fast_fragment_list[s->last_coded_c_fragment] = -1;
751
    }
752

    
753
    return 0;
754
}
755

    
756
/*
757
 * This function unpacks all the coding mode data for individual macroblocks
758
 * from the bitstream.
759
 */
760
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
761
{
762
    int i, j, k;
763
    int scheme;
764
    int current_macroblock;
765
    int current_fragment;
766
    int coding_mode;
767
    int custom_mode_alphabet[CODING_MODE_COUNT];
768

    
769
    if (s->keyframe) {
770
        for (i = 0; i < s->fragment_count; i++)
771
            s->all_fragments[i].coding_method = MODE_INTRA;
772

    
773
    } else {
774

    
775
        /* fetch the mode coding scheme for this frame */
776
        scheme = get_bits(gb, 3);
777

    
778
        /* is it a custom coding scheme? */
779
        if (scheme == 0) {
780
            for (i = 0; i < 8; i++)
781
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
782
            for (i = 0; i < 8; i++)
783
                custom_mode_alphabet[get_bits(gb, 3)] = i;
784
        }
785

    
786
        /* iterate through all of the macroblocks that contain 1 or more
787
         * coded fragments */
788
        for (i = 0; i < s->u_superblock_start; i++) {
789

    
790
            for (j = 0; j < 4; j++) {
791
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
792
                if ((current_macroblock == -1) ||
793
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
794
                    continue;
795
                if (current_macroblock >= s->macroblock_count) {
796
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
797
                        current_macroblock, s->macroblock_count);
798
                    return 1;
799
                }
800

    
801
                /* mode 7 means get 3 bits for each coding mode */
802
                if (scheme == 7)
803
                    coding_mode = get_bits(gb, 3);
804
                else if(scheme == 0)
805
                    coding_mode = custom_mode_alphabet
806
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
807
                else
808
                    coding_mode = ModeAlphabet[scheme-1]
809
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
810

    
811
                s->macroblock_coding[current_macroblock] = coding_mode;
812
                for (k = 0; k < 6; k++) {
813
                    current_fragment =
814
                        s->macroblock_fragments[current_macroblock * 6 + k];
815
                    if (current_fragment == -1)
816
                        continue;
817
                    if (current_fragment >= s->fragment_count) {
818
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
819
                            current_fragment, s->fragment_count);
820
                        return 1;
821
                    }
822
                    if (s->all_fragments[current_fragment].coding_method !=
823
                        MODE_COPY)
824
                        s->all_fragments[current_fragment].coding_method =
825
                            coding_mode;
826
                }
827
            }
828
        }
829
    }
830

    
831
    return 0;
832
}
833

    
834
/*
835
 * This function unpacks all the motion vectors for the individual
836
 * macroblocks from the bitstream.
837
 */
838
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
839
{
840
    int i, j, k, l;
841
    int coding_mode;
842
    int motion_x[6];
843
    int motion_y[6];
844
    int last_motion_x = 0;
845
    int last_motion_y = 0;
846
    int prior_last_motion_x = 0;
847
    int prior_last_motion_y = 0;
848
    int current_macroblock;
849
    int current_fragment;
850

    
851
    if (s->keyframe)
852
        return 0;
853

    
854
    memset(motion_x, 0, 6 * sizeof(int));
855
    memset(motion_y, 0, 6 * sizeof(int));
856

    
857
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
858
    coding_mode = get_bits1(gb);
859

    
860
    /* iterate through all of the macroblocks that contain 1 or more
861
     * coded fragments */
862
    for (i = 0; i < s->u_superblock_start; i++) {
863

    
864
        for (j = 0; j < 4; j++) {
865
            current_macroblock = s->superblock_macroblocks[i * 4 + j];
866
            if ((current_macroblock == -1) ||
867
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
868
                continue;
869
            if (current_macroblock >= s->macroblock_count) {
870
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
871
                    current_macroblock, s->macroblock_count);
872
                return 1;
873
            }
874

    
875
            current_fragment = s->macroblock_fragments[current_macroblock * 6];
876
            if (current_fragment >= s->fragment_count) {
877
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
878
                    current_fragment, s->fragment_count);
879
                return 1;
880
            }
881
            switch (s->macroblock_coding[current_macroblock]) {
882

    
883
            case MODE_INTER_PLUS_MV:
884
            case MODE_GOLDEN_MV:
885
                /* all 6 fragments use the same motion vector */
886
                if (coding_mode == 0) {
887
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
888
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
889
                } else {
890
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
891
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
892
                }
893

    
894
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
895
                if (s->macroblock_coding[current_macroblock] ==
896
                    MODE_INTER_PLUS_MV) {
897
                    prior_last_motion_x = last_motion_x;
898
                    prior_last_motion_y = last_motion_y;
899
                    last_motion_x = motion_x[0];
900
                    last_motion_y = motion_y[0];
901
                }
902
                break;
903

    
904
            case MODE_INTER_FOURMV:
905
                /* vector maintenance */
906
                prior_last_motion_x = last_motion_x;
907
                prior_last_motion_y = last_motion_y;
908

    
909
                /* fetch 4 vectors from the bitstream, one for each
910
                 * Y fragment, then average for the C fragment vectors */
911
                motion_x[4] = motion_y[4] = 0;
912
                for (k = 0; k < 4; k++) {
913
                    for (l = 0; l < s->coded_fragment_list_index; l++)
914
                        if (s->coded_fragment_list[l] == s->macroblock_fragments[6*current_macroblock + k])
915
                            break;
916
                    if (l < s->coded_fragment_list_index) {
917
                        if (coding_mode == 0) {
918
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
919
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
920
                        } else {
921
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
922
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
923
                        }
924
                        last_motion_x = motion_x[k];
925
                        last_motion_y = motion_y[k];
926
                    } else {
927
                        motion_x[k] = 0;
928
                        motion_y[k] = 0;
929
                    }
930
                    motion_x[4] += motion_x[k];
931
                    motion_y[4] += motion_y[k];
932
                }
933

    
934
                motion_x[5]=
935
                motion_x[4]= RSHIFT(motion_x[4], 2);
936
                motion_y[5]=
937
                motion_y[4]= RSHIFT(motion_y[4], 2);
938
                break;
939

    
940
            case MODE_INTER_LAST_MV:
941
                /* all 6 fragments use the last motion vector */
942
                motion_x[0] = last_motion_x;
943
                motion_y[0] = last_motion_y;
944

    
945
                /* no vector maintenance (last vector remains the
946
                 * last vector) */
947
                break;
948

    
949
            case MODE_INTER_PRIOR_LAST:
950
                /* all 6 fragments use the motion vector prior to the
951
                 * last motion vector */
952
                motion_x[0] = prior_last_motion_x;
953
                motion_y[0] = prior_last_motion_y;
954

    
955
                /* vector maintenance */
956
                prior_last_motion_x = last_motion_x;
957
                prior_last_motion_y = last_motion_y;
958
                last_motion_x = motion_x[0];
959
                last_motion_y = motion_y[0];
960
                break;
961

    
962
            default:
963
                /* covers intra, inter without MV, golden without MV */
964
                motion_x[0] = 0;
965
                motion_y[0] = 0;
966

    
967
                /* no vector maintenance */
968
                break;
969
            }
970

    
971
            /* assign the motion vectors to the correct fragments */
972
            for (k = 0; k < 6; k++) {
973
                current_fragment =
974
                    s->macroblock_fragments[current_macroblock * 6 + k];
975
                if (current_fragment == -1)
976
                    continue;
977
                if (current_fragment >= s->fragment_count) {
978
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
979
                        current_fragment, s->fragment_count);
980
                    return 1;
981
                }
982
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
983
                s->all_fragments[current_fragment].motion_x = motion_x[k];
984
                s->all_fragments[current_fragment].motion_y = motion_y[k];
985
                } else {
986
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
987
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
988
                }
989
            }
990
        }
991
    }
992

    
993
    return 0;
994
}
995

    
996
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
997
{
998
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
999
    int num_blocks = s->coded_fragment_list_index;
1000

    
1001
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
1002
        i = blocks_decoded = num_blocks_at_qpi = 0;
1003

    
1004
        bit = get_bits1(gb);
1005

    
1006
        do {
1007
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
1008
            if (run_length == 34)
1009
                run_length += get_bits(gb, 12);
1010
            blocks_decoded += run_length;
1011

    
1012
            if (!bit)
1013
                num_blocks_at_qpi += run_length;
1014

    
1015
            for (j = 0; j < run_length; i++) {
1016
                if (i >= s->coded_fragment_list_index)
1017
                    return -1;
1018

    
1019
                if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
1020
                    s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
1021
                    j++;
1022
                }
1023
            }
1024

    
1025
            if (run_length == 4129)
1026
                bit = get_bits1(gb);
1027
            else
1028
                bit ^= 1;
1029
        } while (blocks_decoded < num_blocks);
1030

    
1031
        num_blocks -= num_blocks_at_qpi;
1032
    }
1033

    
1034
    return 0;
1035
}
1036

    
1037
/*
1038
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1039
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1040
 * data. This function unpacks all the VLCs for either the Y plane or both
1041
 * C planes, and is called for DC coefficients or different AC coefficient
1042
 * levels (since different coefficient types require different VLC tables.
1043
 *
1044
 * This function returns a residual eob run. E.g, if a particular token gave
1045
 * instructions to EOB the next 5 fragments and there were only 2 fragments
1046
 * left in the current fragment range, 3 would be returned so that it could
1047
 * be passed into the next call to this same function.
1048
 */
1049
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1050
                        VLC *table, int coeff_index,
1051
                        int y_plane,
1052
                        int eob_run)
1053
{
1054
    int i;
1055
    int token;
1056
    int zero_run = 0;
1057
    DCTELEM coeff = 0;
1058
    Vp3Fragment *fragment;
1059
    int bits_to_get;
1060
    int next_fragment;
1061
    int previous_fragment;
1062
    int fragment_num;
1063
    int *list_head;
1064

    
1065
    /* local references to structure members to avoid repeated deferences */
1066
    uint8_t *perm= s->scantable.permutated;
1067
    int *coded_fragment_list = s->coded_fragment_list;
1068
    Vp3Fragment *all_fragments = s->all_fragments;
1069
    uint8_t *coeff_counts = s->coeff_counts;
1070
    VLC_TYPE (*vlc_table)[2] = table->table;
1071
    int *fast_fragment_list = s->fast_fragment_list;
1072

    
1073
    if (y_plane) {
1074
        next_fragment = s->fragment_list_y_head;
1075
        list_head = &s->fragment_list_y_head;
1076
    } else {
1077
        next_fragment = s->fragment_list_c_head;
1078
        list_head = &s->fragment_list_c_head;
1079
    }
1080

    
1081
    i = next_fragment;
1082
    previous_fragment = -1;  /* this indicates that the previous fragment is actually the list head */
1083
    while (i != -1) {
1084
        fragment_num = coded_fragment_list[i];
1085

    
1086
        if (coeff_counts[fragment_num] > coeff_index) {
1087
            previous_fragment = i;
1088
            i = fast_fragment_list[i];
1089
            continue;
1090
        }
1091
        fragment = &all_fragments[fragment_num];
1092

    
1093
        if (!eob_run) {
1094
            /* decode a VLC into a token */
1095
            token = get_vlc2(gb, vlc_table, 5, 3);
1096
            /* use the token to get a zero run, a coefficient, and an eob run */
1097
            if (token <= 6) {
1098
                eob_run = eob_run_base[token];
1099
                if (eob_run_get_bits[token])
1100
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
1101
                coeff = zero_run = 0;
1102
            } else {
1103
                bits_to_get = coeff_get_bits[token];
1104
                if (bits_to_get)
1105
                    bits_to_get = get_bits(gb, bits_to_get);
1106
                coeff = coeff_tables[token][bits_to_get];
1107

    
1108
                zero_run = zero_run_base[token];
1109
                if (zero_run_get_bits[token])
1110
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
1111
            }
1112
        }
1113

    
1114
        if (!eob_run) {
1115
            coeff_counts[fragment_num] += zero_run;
1116
            if (coeff_counts[fragment_num] < 64){
1117
                fragment->next_coeff->coeff= coeff;
1118
                fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1119
                fragment->next_coeff->next= s->next_coeff;
1120
                s->next_coeff->next=NULL;
1121
                fragment->next_coeff= s->next_coeff++;
1122
            }
1123
            /* previous fragment is now this fragment */
1124
            previous_fragment = i;
1125
        } else {
1126
            coeff_counts[fragment_num] |= 128;
1127
            eob_run--;
1128
            /* remove this fragment from the list */
1129
            if (previous_fragment != -1)
1130
                fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1131
            else
1132
                *list_head = fast_fragment_list[i];
1133
            /* previous fragment remains unchanged */
1134
        }
1135

    
1136
        i = fast_fragment_list[i];
1137
    }
1138

    
1139
    return eob_run;
1140
}
1141

    
1142
static void reverse_dc_prediction(Vp3DecodeContext *s,
1143
                                  int first_fragment,
1144
                                  int fragment_width,
1145
                                  int fragment_height);
1146
/*
1147
 * This function unpacks all of the DCT coefficient data from the
1148
 * bitstream.
1149
 */
1150
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1151
{
1152
    int i;
1153
    int dc_y_table;
1154
    int dc_c_table;
1155
    int ac_y_table;
1156
    int ac_c_table;
1157
    int residual_eob_run = 0;
1158
    VLC *y_tables[64];
1159
    VLC *c_tables[64];
1160

    
1161
    /* fetch the DC table indexes */
1162
    dc_y_table = get_bits(gb, 4);
1163
    dc_c_table = get_bits(gb, 4);
1164

    
1165
    /* unpack the Y plane DC coefficients */
1166
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1167
        1, residual_eob_run);
1168

    
1169
    /* reverse prediction of the Y-plane DC coefficients */
1170
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1171

    
1172
    /* unpack the C plane DC coefficients */
1173
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1174
        0, residual_eob_run);
1175

    
1176
    /* reverse prediction of the C-plane DC coefficients */
1177
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1178
    {
1179
        reverse_dc_prediction(s, s->fragment_start[1],
1180
            s->fragment_width / 2, s->fragment_height / 2);
1181
        reverse_dc_prediction(s, s->fragment_start[2],
1182
            s->fragment_width / 2, s->fragment_height / 2);
1183
    }
1184

    
1185
    /* fetch the AC table indexes */
1186
    ac_y_table = get_bits(gb, 4);
1187
    ac_c_table = get_bits(gb, 4);
1188

    
1189
    /* build tables of AC VLC tables */
1190
    for (i = 1; i <= 5; i++) {
1191
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1192
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1193
    }
1194
    for (i = 6; i <= 14; i++) {
1195
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1196
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1197
    }
1198
    for (i = 15; i <= 27; i++) {
1199
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1200
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1201
    }
1202
    for (i = 28; i <= 63; i++) {
1203
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1204
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1205
    }
1206

    
1207
    /* decode all AC coefficents */
1208
    for (i = 1; i <= 63; i++) {
1209
        if (s->fragment_list_y_head != -1)
1210
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1211
                1, residual_eob_run);
1212

    
1213
        if (s->fragment_list_c_head != -1)
1214
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1215
                0, residual_eob_run);
1216
    }
1217

    
1218
    return 0;
1219
}
1220

    
1221
/*
1222
 * This function reverses the DC prediction for each coded fragment in
1223
 * the frame. Much of this function is adapted directly from the original
1224
 * VP3 source code.
1225
 */
1226
#define COMPATIBLE_FRAME(x) \
1227
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1228
#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1229

    
1230
static void reverse_dc_prediction(Vp3DecodeContext *s,
1231
                                  int first_fragment,
1232
                                  int fragment_width,
1233
                                  int fragment_height)
1234
{
1235

    
1236
#define PUL 8
1237
#define PU 4
1238
#define PUR 2
1239
#define PL 1
1240

    
1241
    int x, y;
1242
    int i = first_fragment;
1243

    
1244
    int predicted_dc;
1245

    
1246
    /* DC values for the left, up-left, up, and up-right fragments */
1247
    int vl, vul, vu, vur;
1248

    
1249
    /* indexes for the left, up-left, up, and up-right fragments */
1250
    int l, ul, u, ur;
1251

    
1252
    /*
1253
     * The 6 fields mean:
1254
     *   0: up-left multiplier
1255
     *   1: up multiplier
1256
     *   2: up-right multiplier
1257
     *   3: left multiplier
1258
     */
1259
    static const int predictor_transform[16][4] = {
1260
        {  0,  0,  0,  0},
1261
        {  0,  0,  0,128},        // PL
1262
        {  0,  0,128,  0},        // PUR
1263
        {  0,  0, 53, 75},        // PUR|PL
1264
        {  0,128,  0,  0},        // PU
1265
        {  0, 64,  0, 64},        // PU|PL
1266
        {  0,128,  0,  0},        // PU|PUR
1267
        {  0,  0, 53, 75},        // PU|PUR|PL
1268
        {128,  0,  0,  0},        // PUL
1269
        {  0,  0,  0,128},        // PUL|PL
1270
        { 64,  0, 64,  0},        // PUL|PUR
1271
        {  0,  0, 53, 75},        // PUL|PUR|PL
1272
        {  0,128,  0,  0},        // PUL|PU
1273
       {-104,116,  0,116},        // PUL|PU|PL
1274
        { 24, 80, 24,  0},        // PUL|PU|PUR
1275
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1276
    };
1277

    
1278
    /* This table shows which types of blocks can use other blocks for
1279
     * prediction. For example, INTRA is the only mode in this table to
1280
     * have a frame number of 0. That means INTRA blocks can only predict
1281
     * from other INTRA blocks. There are 2 golden frame coding types;
1282
     * blocks encoding in these modes can only predict from other blocks
1283
     * that were encoded with these 1 of these 2 modes. */
1284
    static const unsigned char compatible_frame[9] = {
1285
        1,    /* MODE_INTER_NO_MV */
1286
        0,    /* MODE_INTRA */
1287
        1,    /* MODE_INTER_PLUS_MV */
1288
        1,    /* MODE_INTER_LAST_MV */
1289
        1,    /* MODE_INTER_PRIOR_MV */
1290
        2,    /* MODE_USING_GOLDEN */
1291
        2,    /* MODE_GOLDEN_MV */
1292
        1,    /* MODE_INTER_FOUR_MV */
1293
        3     /* MODE_COPY */
1294
    };
1295
    int current_frame_type;
1296

    
1297
    /* there is a last DC predictor for each of the 3 frame types */
1298
    short last_dc[3];
1299

    
1300
    int transform = 0;
1301

    
1302
    vul = vu = vur = vl = 0;
1303
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1304

    
1305
    /* for each fragment row... */
1306
    for (y = 0; y < fragment_height; y++) {
1307

    
1308
        /* for each fragment in a row... */
1309
        for (x = 0; x < fragment_width; x++, i++) {
1310

    
1311
            /* reverse prediction if this block was coded */
1312
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1313

    
1314
                current_frame_type =
1315
                    compatible_frame[s->all_fragments[i].coding_method];
1316

    
1317
                transform= 0;
1318
                if(x){
1319
                    l= i-1;
1320
                    vl = DC_COEFF(l);
1321
                    if(COMPATIBLE_FRAME(l))
1322
                        transform |= PL;
1323
                }
1324
                if(y){
1325
                    u= i-fragment_width;
1326
                    vu = DC_COEFF(u);
1327
                    if(COMPATIBLE_FRAME(u))
1328
                        transform |= PU;
1329
                    if(x){
1330
                        ul= i-fragment_width-1;
1331
                        vul = DC_COEFF(ul);
1332
                        if(COMPATIBLE_FRAME(ul))
1333
                            transform |= PUL;
1334
                    }
1335
                    if(x + 1 < fragment_width){
1336
                        ur= i-fragment_width+1;
1337
                        vur = DC_COEFF(ur);
1338
                        if(COMPATIBLE_FRAME(ur))
1339
                            transform |= PUR;
1340
                    }
1341
                }
1342

    
1343
                if (transform == 0) {
1344

    
1345
                    /* if there were no fragments to predict from, use last
1346
                     * DC saved */
1347
                    predicted_dc = last_dc[current_frame_type];
1348
                } else {
1349

    
1350
                    /* apply the appropriate predictor transform */
1351
                    predicted_dc =
1352
                        (predictor_transform[transform][0] * vul) +
1353
                        (predictor_transform[transform][1] * vu) +
1354
                        (predictor_transform[transform][2] * vur) +
1355
                        (predictor_transform[transform][3] * vl);
1356

    
1357
                    predicted_dc /= 128;
1358

    
1359
                    /* check for outranging on the [ul u l] and
1360
                     * [ul u ur l] predictors */
1361
                    if ((transform == 15) || (transform == 13)) {
1362
                        if (FFABS(predicted_dc - vu) > 128)
1363
                            predicted_dc = vu;
1364
                        else if (FFABS(predicted_dc - vl) > 128)
1365
                            predicted_dc = vl;
1366
                        else if (FFABS(predicted_dc - vul) > 128)
1367
                            predicted_dc = vul;
1368
                    }
1369
                }
1370

    
1371
                /* at long last, apply the predictor */
1372
                if(s->coeffs[i].index){
1373
                    *s->next_coeff= s->coeffs[i];
1374
                    s->coeffs[i].index=0;
1375
                    s->coeffs[i].coeff=0;
1376
                    s->coeffs[i].next= s->next_coeff++;
1377
                }
1378
                s->coeffs[i].coeff += predicted_dc;
1379
                /* save the DC */
1380
                last_dc[current_frame_type] = DC_COEFF(i);
1381
                if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1382
                    s->coeff_counts[i]= 129;
1383
//                    s->all_fragments[i].next_coeff= s->next_coeff;
1384
                    s->coeffs[i].next= s->next_coeff;
1385
                    (s->next_coeff++)->next=NULL;
1386
                }
1387
            }
1388
        }
1389
    }
1390
}
1391

    
1392
/*
1393
 * Perform the final rendering for a particular slice of data.
1394
 * The slice number ranges from 0..(macroblock_height - 1).
1395
 */
1396
static void render_slice(Vp3DecodeContext *s, int slice)
1397
{
1398
    int x;
1399
    int16_t *dequantizer;
1400
    DECLARE_ALIGNED_16(DCTELEM, block[64]);
1401
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1402
    int motion_halfpel_index;
1403
    uint8_t *motion_source;
1404
    int plane;
1405
    int current_macroblock_entry = slice * s->macroblock_width * 6;
1406

    
1407
    if (slice >= s->macroblock_height)
1408
        return;
1409

    
1410
    for (plane = 0; plane < 3; plane++) {
1411
        uint8_t *output_plane = s->current_frame.data    [plane];
1412
        uint8_t *  last_plane = s->   last_frame.data    [plane];
1413
        uint8_t *golden_plane = s-> golden_frame.data    [plane];
1414
        int stride            = s->current_frame.linesize[plane];
1415
        int plane_width       = s->width  >> !!plane;
1416
        int plane_height      = s->height >> !!plane;
1417
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
1418
        int slice_height = y + (FRAGMENT_PIXELS << !plane);
1419
        int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1420

    
1421
        if (!s->flipped_image) stride = -stride;
1422

    
1423

    
1424
        if(FFABS(stride) > 2048)
1425
            return; //various tables are fixed size
1426

    
1427
        /* for each fragment row in the slice (both of them)... */
1428
        for (; y < slice_height; y += 8) {
1429

    
1430
            /* for each fragment in a row... */
1431
            for (x = 0; x < plane_width; x += 8, i++) {
1432

    
1433
                if ((i < 0) || (i >= s->fragment_count)) {
1434
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
1435
                    return;
1436
                }
1437

    
1438
                /* transform if this block was coded */
1439
                if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1440
                    !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1441

    
1442
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1443
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1444
                        motion_source= golden_plane;
1445
                    else
1446
                        motion_source= last_plane;
1447

    
1448
                    motion_source += s->all_fragments[i].first_pixel;
1449
                    motion_halfpel_index = 0;
1450

    
1451
                    /* sort out the motion vector if this fragment is coded
1452
                     * using a motion vector method */
1453
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1454
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1455
                        int src_x, src_y;
1456
                        motion_x = s->all_fragments[i].motion_x;
1457
                        motion_y = s->all_fragments[i].motion_y;
1458
                        if(plane){
1459
                            motion_x= (motion_x>>1) | (motion_x&1);
1460
                            motion_y= (motion_y>>1) | (motion_y&1);
1461
                        }
1462

    
1463
                        src_x= (motion_x>>1) + x;
1464
                        src_y= (motion_y>>1) + y;
1465
                        if ((motion_x == 127) || (motion_y == 127))
1466
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1467

    
1468
                        motion_halfpel_index = motion_x & 0x01;
1469
                        motion_source += (motion_x >> 1);
1470

    
1471
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1472
                        motion_source += ((motion_y >> 1) * stride);
1473

    
1474
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1475
                            uint8_t *temp= s->edge_emu_buffer;
1476
                            if(stride<0) temp -= 9*stride;
1477
                            else temp += 9*stride;
1478

    
1479
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1480
                            motion_source= temp;
1481
                        }
1482
                    }
1483

    
1484

    
1485
                    /* first, take care of copying a block from either the
1486
                     * previous or the golden frame */
1487
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1488
                        /* Note, it is possible to implement all MC cases with
1489
                           put_no_rnd_pixels_l2 which would look more like the
1490
                           VP3 source but this would be slower as
1491
                           put_no_rnd_pixels_tab is better optimzed */
1492
                        if(motion_halfpel_index != 3){
1493
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1494
                                output_plane + s->all_fragments[i].first_pixel,
1495
                                motion_source, stride, 8);
1496
                        }else{
1497
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1498
                            s->dsp.put_no_rnd_pixels_l2[1](
1499
                                output_plane + s->all_fragments[i].first_pixel,
1500
                                motion_source - d,
1501
                                motion_source + stride + 1 + d,
1502
                                stride, 8);
1503
                        }
1504
                        dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1505
                    }else{
1506
                        dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1507
                    }
1508

    
1509
                    /* dequantize the DCT coefficients */
1510
                    if(s->avctx->idct_algo==FF_IDCT_VP3){
1511
                        Coeff *coeff= s->coeffs + i;
1512
                        s->dsp.clear_block(block);
1513
                        while(coeff->next){
1514
                            block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1515
                            coeff= coeff->next;
1516
                        }
1517
                    }else{
1518
                        Coeff *coeff= s->coeffs + i;
1519
                        s->dsp.clear_block(block);
1520
                        while(coeff->next){
1521
                            block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1522
                            coeff= coeff->next;
1523
                        }
1524
                    }
1525

    
1526
                    /* invert DCT and place (or add) in final output */
1527

    
1528
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1529
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1530
                            block[0] += 128<<3;
1531
                        s->dsp.idct_put(
1532
                            output_plane + s->all_fragments[i].first_pixel,
1533
                            stride,
1534
                            block);
1535
                    } else {
1536
                        s->dsp.idct_add(
1537
                            output_plane + s->all_fragments[i].first_pixel,
1538
                            stride,
1539
                            block);
1540
                    }
1541
                } else {
1542

    
1543
                    /* copy directly from the previous frame */
1544
                    s->dsp.put_pixels_tab[1][0](
1545
                        output_plane + s->all_fragments[i].first_pixel,
1546
                        last_plane + s->all_fragments[i].first_pixel,
1547
                        stride, 8);
1548

    
1549
                }
1550
#if 0
1551
                /* perform the left edge filter if:
1552
                 *   - the fragment is not on the left column
1553
                 *   - the fragment is coded in this frame
1554
                 *   - the fragment is not coded in this frame but the left
1555
                 *     fragment is coded in this frame (this is done instead
1556
                 *     of a right edge filter when rendering the left fragment
1557
                 *     since this fragment is not available yet) */
1558
                if ((x > 0) &&
1559
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1560
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1561
                      (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1562
                    horizontal_filter(
1563
                        output_plane + s->all_fragments[i].first_pixel + 7*stride,
1564
                        -stride, s->bounding_values_array + 127);
1565
                }
1566

1567
                /* perform the top edge filter if:
1568
                 *   - the fragment is not on the top row
1569
                 *   - the fragment is coded in this frame
1570
                 *   - the fragment is not coded in this frame but the above
1571
                 *     fragment is coded in this frame (this is done instead
1572
                 *     of a bottom edge filter when rendering the above
1573
                 *     fragment since this fragment is not available yet) */
1574
                if ((y > 0) &&
1575
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1576
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1577
                      (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1578
                    vertical_filter(
1579
                        output_plane + s->all_fragments[i].first_pixel - stride,
1580
                        -stride, s->bounding_values_array + 127);
1581
                }
1582
#endif
1583
            }
1584
        }
1585
    }
1586

    
1587
     /* this looks like a good place for slice dispatch... */
1588
     /* algorithm:
1589
      *   if (slice == s->macroblock_height - 1)
1590
      *     dispatch (both last slice & 2nd-to-last slice);
1591
      *   else if (slice > 0)
1592
      *     dispatch (slice - 1);
1593
      */
1594

    
1595
    emms_c();
1596
}
1597

    
1598
static void apply_loop_filter(Vp3DecodeContext *s)
1599
{
1600
    int plane;
1601
    int x, y;
1602
    int *bounding_values= s->bounding_values_array+127;
1603

    
1604
#if 0
1605
    int bounding_values_array[256];
1606
    int filter_limit;
1607

1608
    /* find the right loop limit value */
1609
    for (x = 63; x >= 0; x--) {
1610
        if (vp31_ac_scale_factor[x] >= s->quality_index)
1611
            break;
1612
    }
1613
    filter_limit = vp31_filter_limit_values[s->quality_index];
1614

1615
    /* set up the bounding values */
1616
    memset(bounding_values_array, 0, 256 * sizeof(int));
1617
    for (x = 0; x < filter_limit; x++) {
1618
        bounding_values[-x - filter_limit] = -filter_limit + x;
1619
        bounding_values[-x] = -x;
1620
        bounding_values[x] = x;
1621
        bounding_values[x + filter_limit] = filter_limit - x;
1622
    }
1623
#endif
1624

    
1625
    for (plane = 0; plane < 3; plane++) {
1626
        int width           = s->fragment_width  >> !!plane;
1627
        int height          = s->fragment_height >> !!plane;
1628
        int fragment        = s->fragment_start        [plane];
1629
        int stride          = s->current_frame.linesize[plane];
1630
        uint8_t *plane_data = s->current_frame.data    [plane];
1631
        if (!s->flipped_image) stride = -stride;
1632

    
1633
        for (y = 0; y < height; y++) {
1634

    
1635
            for (x = 0; x < width; x++) {
1636
                /* This code basically just deblocks on the edges of coded blocks.
1637
                 * However, it has to be much more complicated because of the
1638
                 * braindamaged deblock ordering used in VP3/Theora. Order matters
1639
                 * because some pixels get filtered twice. */
1640
                if( s->all_fragments[fragment].coding_method != MODE_COPY )
1641
                {
1642
                    /* do not perform left edge filter for left columns frags */
1643
                    if (x > 0) {
1644
                        s->dsp.vp3_h_loop_filter(
1645
                            plane_data + s->all_fragments[fragment].first_pixel,
1646
                            stride, bounding_values);
1647
                    }
1648

    
1649
                    /* do not perform top edge filter for top row fragments */
1650
                    if (y > 0) {
1651
                        s->dsp.vp3_v_loop_filter(
1652
                            plane_data + s->all_fragments[fragment].first_pixel,
1653
                            stride, bounding_values);
1654
                    }
1655

    
1656
                    /* do not perform right edge filter for right column
1657
                     * fragments or if right fragment neighbor is also coded
1658
                     * in this frame (it will be filtered in next iteration) */
1659
                    if ((x < width - 1) &&
1660
                        (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1661
                        s->dsp.vp3_h_loop_filter(
1662
                            plane_data + s->all_fragments[fragment + 1].first_pixel,
1663
                            stride, bounding_values);
1664
                    }
1665

    
1666
                    /* do not perform bottom edge filter for bottom row
1667
                     * fragments or if bottom fragment neighbor is also coded
1668
                     * in this frame (it will be filtered in the next row) */
1669
                    if ((y < height - 1) &&
1670
                        (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1671
                        s->dsp.vp3_v_loop_filter(
1672
                            plane_data + s->all_fragments[fragment + width].first_pixel,
1673
                            stride, bounding_values);
1674
                    }
1675
                }
1676

    
1677
                fragment++;
1678
            }
1679
        }
1680
    }
1681
}
1682

    
1683
/*
1684
 * This function computes the first pixel addresses for each fragment.
1685
 * This function needs to be invoked after the first frame is allocated
1686
 * so that it has access to the plane strides.
1687
 */
1688
static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1689
{
1690
#define Y_INITIAL(chroma_shift)  s->flipped_image ? 1  : s->fragment_height >> chroma_shift
1691
#define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
1692

    
1693
    int i, x, y;
1694
    const int y_inc = s->flipped_image ? 1 : -1;
1695

    
1696
    /* figure out the first pixel addresses for each of the fragments */
1697
    /* Y plane */
1698
    i = 0;
1699
    for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) {
1700
        for (x = 0; x < s->fragment_width; x++) {
1701
            s->all_fragments[i++].first_pixel =
1702
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1703
                    s->golden_frame.linesize[0] +
1704
                    x * FRAGMENT_PIXELS;
1705
        }
1706
    }
1707

    
1708
    /* U plane */
1709
    i = s->fragment_start[1];
1710
    for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
1711
        for (x = 0; x < s->fragment_width / 2; x++) {
1712
            s->all_fragments[i++].first_pixel =
1713
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1714
                    s->golden_frame.linesize[1] +
1715
                    x * FRAGMENT_PIXELS;
1716
        }
1717
    }
1718

    
1719
    /* V plane */
1720
    i = s->fragment_start[2];
1721
    for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
1722
        for (x = 0; x < s->fragment_width / 2; x++) {
1723
            s->all_fragments[i++].first_pixel =
1724
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1725
                    s->golden_frame.linesize[2] +
1726
                    x * FRAGMENT_PIXELS;
1727
        }
1728
    }
1729
}
1730

    
1731
/*
1732
 * This is the ffmpeg/libavcodec API init function.
1733
 */
1734
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1735
{
1736
    Vp3DecodeContext *s = avctx->priv_data;
1737
    int i, inter, plane;
1738
    int c_width;
1739
    int c_height;
1740
    int y_superblock_count;
1741
    int c_superblock_count;
1742

    
1743
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1744
        s->version = 0;
1745
    else
1746
        s->version = 1;
1747

    
1748
    s->avctx = avctx;
1749
    s->width = FFALIGN(avctx->width, 16);
1750
    s->height = FFALIGN(avctx->height, 16);
1751
    avctx->pix_fmt = PIX_FMT_YUV420P;
1752
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1753
    if(avctx->idct_algo==FF_IDCT_AUTO)
1754
        avctx->idct_algo=FF_IDCT_VP3;
1755
    dsputil_init(&s->dsp, avctx);
1756

    
1757
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1758

    
1759
    /* initialize to an impossible value which will force a recalculation
1760
     * in the first frame decode */
1761
    for (i = 0; i < 3; i++)
1762
        s->qps[i] = -1;
1763

    
1764
    s->y_superblock_width = (s->width + 31) / 32;
1765
    s->y_superblock_height = (s->height + 31) / 32;
1766
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1767

    
1768
    /* work out the dimensions for the C planes */
1769
    c_width = s->width / 2;
1770
    c_height = s->height / 2;
1771
    s->c_superblock_width = (c_width + 31) / 32;
1772
    s->c_superblock_height = (c_height + 31) / 32;
1773
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1774

    
1775
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1776
    s->u_superblock_start = y_superblock_count;
1777
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1778
    s->superblock_coding = av_malloc(s->superblock_count);
1779

    
1780
    s->macroblock_width = (s->width + 15) / 16;
1781
    s->macroblock_height = (s->height + 15) / 16;
1782
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1783

    
1784
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1785
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1786

    
1787
    /* fragment count covers all 8x8 blocks for all 3 planes */
1788
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1789
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1790
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1791

    
1792
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1793
    s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1794
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1795
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1796
    s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1797
    s->pixel_addresses_initialized = 0;
1798
    if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1799
        !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1800
        vp3_decode_end(avctx);
1801
        return -1;
1802
    }
1803

    
1804
    if (!s->theora_tables)
1805
    {
1806
        for (i = 0; i < 64; i++) {
1807
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1808
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1809
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1810
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1811
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1812
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1813
        }
1814

    
1815
        for(inter=0; inter<2; inter++){
1816
            for(plane=0; plane<3; plane++){
1817
                s->qr_count[inter][plane]= 1;
1818
                s->qr_size [inter][plane][0]= 63;
1819
                s->qr_base [inter][plane][0]=
1820
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1821
            }
1822
        }
1823

    
1824
        /* init VLC tables */
1825
        for (i = 0; i < 16; i++) {
1826

    
1827
            /* DC histograms */
1828
            init_vlc(&s->dc_vlc[i], 5, 32,
1829
                &dc_bias[i][0][1], 4, 2,
1830
                &dc_bias[i][0][0], 4, 2, 0);
1831

    
1832
            /* group 1 AC histograms */
1833
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1834
                &ac_bias_0[i][0][1], 4, 2,
1835
                &ac_bias_0[i][0][0], 4, 2, 0);
1836

    
1837
            /* group 2 AC histograms */
1838
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1839
                &ac_bias_1[i][0][1], 4, 2,
1840
                &ac_bias_1[i][0][0], 4, 2, 0);
1841

    
1842
            /* group 3 AC histograms */
1843
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1844
                &ac_bias_2[i][0][1], 4, 2,
1845
                &ac_bias_2[i][0][0], 4, 2, 0);
1846

    
1847
            /* group 4 AC histograms */
1848
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1849
                &ac_bias_3[i][0][1], 4, 2,
1850
                &ac_bias_3[i][0][0], 4, 2, 0);
1851
        }
1852
    } else {
1853
        for (i = 0; i < 16; i++) {
1854

    
1855
            /* DC histograms */
1856
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1857
                &s->huffman_table[i][0][1], 4, 2,
1858
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1859
                goto vlc_fail;
1860

    
1861
            /* group 1 AC histograms */
1862
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1863
                &s->huffman_table[i+16][0][1], 4, 2,
1864
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1865
                goto vlc_fail;
1866

    
1867
            /* group 2 AC histograms */
1868
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1869
                &s->huffman_table[i+16*2][0][1], 4, 2,
1870
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1871
                goto vlc_fail;
1872

    
1873
            /* group 3 AC histograms */
1874
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1875
                &s->huffman_table[i+16*3][0][1], 4, 2,
1876
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1877
                goto vlc_fail;
1878

    
1879
            /* group 4 AC histograms */
1880
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1881
                &s->huffman_table[i+16*4][0][1], 4, 2,
1882
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1883
                goto vlc_fail;
1884
        }
1885
    }
1886

    
1887
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1888
        &superblock_run_length_vlc_table[0][1], 4, 2,
1889
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1890

    
1891
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1892
        &fragment_run_length_vlc_table[0][1], 4, 2,
1893
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1894

    
1895
    init_vlc(&s->mode_code_vlc, 3, 8,
1896
        &mode_code_vlc_table[0][1], 2, 1,
1897
        &mode_code_vlc_table[0][0], 2, 1, 0);
1898

    
1899
    init_vlc(&s->motion_vector_vlc, 6, 63,
1900
        &motion_vector_vlc_table[0][1], 2, 1,
1901
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1902

    
1903
    /* work out the block mapping tables */
1904
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1905
    s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
1906
    s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
1907
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1908
    if (!s->superblock_fragments || !s->superblock_macroblocks ||
1909
        !s->macroblock_fragments || !s->macroblock_coding) {
1910
        vp3_decode_end(avctx);
1911
        return -1;
1912
    }
1913
    init_block_mapping(s);
1914

    
1915
    for (i = 0; i < 3; i++) {
1916
        s->current_frame.data[i] = NULL;
1917
        s->last_frame.data[i] = NULL;
1918
        s->golden_frame.data[i] = NULL;
1919
    }
1920

    
1921
    return 0;
1922

    
1923
vlc_fail:
1924
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1925
    return -1;
1926
}
1927

    
1928
/*
1929
 * This is the ffmpeg/libavcodec API frame decode function.
1930
 */
1931
static int vp3_decode_frame(AVCodecContext *avctx,
1932
                            void *data, int *data_size,
1933
                            AVPacket *avpkt)
1934
{
1935
    const uint8_t *buf = avpkt->data;
1936
    int buf_size = avpkt->size;
1937
    Vp3DecodeContext *s = avctx->priv_data;
1938
    GetBitContext gb;
1939
    static int counter = 0;
1940
    int i;
1941

    
1942
    init_get_bits(&gb, buf, buf_size * 8);
1943

    
1944
    if (s->theora && get_bits1(&gb))
1945
    {
1946
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1947
        return -1;
1948
    }
1949

    
1950
    s->keyframe = !get_bits1(&gb);
1951
    if (!s->theora)
1952
        skip_bits(&gb, 1);
1953
    for (i = 0; i < 3; i++)
1954
        s->last_qps[i] = s->qps[i];
1955

    
1956
    s->nqps=0;
1957
    do{
1958
        s->qps[s->nqps++]= get_bits(&gb, 6);
1959
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1960
    for (i = s->nqps; i < 3; i++)
1961
        s->qps[i] = -1;
1962

    
1963
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1964
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1965
            s->keyframe?"key":"", counter, s->qps[0]);
1966
    counter++;
1967

    
1968
    if (s->qps[0] != s->last_qps[0])
1969
        init_loop_filter(s);
1970

    
1971
    for (i = 0; i < s->nqps; i++)
1972
        // reinit all dequantizers if the first one changed, because
1973
        // the DC of the first quantizer must be used for all matrices
1974
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1975
            init_dequantizer(s, i);
1976

    
1977
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1978
        return buf_size;
1979

    
1980
    if (s->keyframe) {
1981
        if (!s->theora)
1982
        {
1983
            skip_bits(&gb, 4); /* width code */
1984
            skip_bits(&gb, 4); /* height code */
1985
            if (s->version)
1986
            {
1987
                s->version = get_bits(&gb, 5);
1988
                if (counter == 1)
1989
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1990
            }
1991
        }
1992
        if (s->version || s->theora)
1993
        {
1994
                if (get_bits1(&gb))
1995
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1996
            skip_bits(&gb, 2); /* reserved? */
1997
        }
1998

    
1999
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2000
            if (s->golden_frame.data[0])
2001
                avctx->release_buffer(avctx, &s->golden_frame);
2002
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2003
        } else {
2004
            if (s->golden_frame.data[0])
2005
                avctx->release_buffer(avctx, &s->golden_frame);
2006
            if (s->last_frame.data[0])
2007
                avctx->release_buffer(avctx, &s->last_frame);
2008
        }
2009

    
2010
        s->golden_frame.reference = 3;
2011
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2012
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2013
            return -1;
2014
        }
2015

    
2016
        /* golden frame is also the current frame */
2017
        s->current_frame= s->golden_frame;
2018

    
2019
        /* time to figure out pixel addresses? */
2020
        if (!s->pixel_addresses_initialized)
2021
        {
2022
            vp3_calculate_pixel_addresses(s);
2023
            s->pixel_addresses_initialized = 1;
2024
        }
2025
    } else {
2026
        /* allocate a new current frame */
2027
        s->current_frame.reference = 3;
2028
        if (!s->pixel_addresses_initialized) {
2029
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2030
            return -1;
2031
        }
2032
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2033
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2034
            return -1;
2035
        }
2036
    }
2037

    
2038
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2039
    s->current_frame.qstride= 0;
2040

    
2041
    init_frame(s, &gb);
2042

    
2043
    if (unpack_superblocks(s, &gb)){
2044
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2045
        return -1;
2046
    }
2047
    if (unpack_modes(s, &gb)){
2048
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2049
        return -1;
2050
    }
2051
    if (unpack_vectors(s, &gb)){
2052
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2053
        return -1;
2054
    }
2055
    if (unpack_block_qpis(s, &gb)){
2056
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
2057
        return -1;
2058
    }
2059
    if (unpack_dct_coeffs(s, &gb)){
2060
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2061
        return -1;
2062
    }
2063

    
2064
    for (i = 0; i < s->macroblock_height; i++)
2065
        render_slice(s, i);
2066

    
2067
    apply_loop_filter(s);
2068

    
2069
    *data_size=sizeof(AVFrame);
2070
    *(AVFrame*)data= s->current_frame;
2071

    
2072
    /* release the last frame, if it is allocated and if it is not the
2073
     * golden frame */
2074
    if ((s->last_frame.data[0]) &&
2075
        (s->last_frame.data[0] != s->golden_frame.data[0]))
2076
        avctx->release_buffer(avctx, &s->last_frame);
2077

    
2078
    /* shuffle frames (last = current) */
2079
    s->last_frame= s->current_frame;
2080
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2081

    
2082
    return buf_size;
2083
}
2084

    
2085
/*
2086
 * This is the ffmpeg/libavcodec API module cleanup function.
2087
 */
2088
static av_cold int vp3_decode_end(AVCodecContext *avctx)
2089
{
2090
    Vp3DecodeContext *s = avctx->priv_data;
2091
    int i;
2092

    
2093
    av_free(s->superblock_coding);
2094
    av_free(s->all_fragments);
2095
    av_free(s->coeff_counts);
2096
    av_free(s->coeffs);
2097
    av_free(s->coded_fragment_list);
2098
    av_free(s->fast_fragment_list);
2099
    av_free(s->superblock_fragments);
2100
    av_free(s->superblock_macroblocks);
2101
    av_free(s->macroblock_fragments);
2102
    av_free(s->macroblock_coding);
2103

    
2104
    for (i = 0; i < 16; i++) {
2105
        free_vlc(&s->dc_vlc[i]);
2106
        free_vlc(&s->ac_vlc_1[i]);
2107
        free_vlc(&s->ac_vlc_2[i]);
2108
        free_vlc(&s->ac_vlc_3[i]);
2109
        free_vlc(&s->ac_vlc_4[i]);
2110
    }
2111

    
2112
    free_vlc(&s->superblock_run_length_vlc);
2113
    free_vlc(&s->fragment_run_length_vlc);
2114
    free_vlc(&s->mode_code_vlc);
2115
    free_vlc(&s->motion_vector_vlc);
2116

    
2117
    /* release all frames */
2118
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2119
        avctx->release_buffer(avctx, &s->golden_frame);
2120
    if (s->last_frame.data[0])
2121
        avctx->release_buffer(avctx, &s->last_frame);
2122
    /* no need to release the current_frame since it will always be pointing
2123
     * to the same frame as either the golden or last frame */
2124

    
2125
    return 0;
2126
}
2127

    
2128
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2129
{
2130
    Vp3DecodeContext *s = avctx->priv_data;
2131

    
2132
    if (get_bits1(gb)) {
2133
        int token;
2134
        if (s->entries >= 32) { /* overflow */
2135
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2136
            return -1;
2137
        }
2138
        token = get_bits(gb, 5);
2139
        //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
2140
        s->huffman_table[s->hti][token][0] = s->hbits;
2141
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
2142
        s->entries++;
2143
    }
2144
    else {
2145
        if (s->huff_code_size >= 32) {/* overflow */
2146
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2147
            return -1;
2148
        }
2149
        s->huff_code_size++;
2150
        s->hbits <<= 1;
2151
        if (read_huffman_tree(avctx, gb))
2152
            return -1;
2153
        s->hbits |= 1;
2154
        if (read_huffman_tree(avctx, gb))
2155
            return -1;
2156
        s->hbits >>= 1;
2157
        s->huff_code_size--;
2158
    }
2159
    return 0;
2160
}
2161

    
2162
#if CONFIG_THEORA_DECODER
2163
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2164
{
2165
    Vp3DecodeContext *s = avctx->priv_data;
2166
    int visible_width, visible_height;
2167

    
2168
    s->theora = get_bits_long(gb, 24);
2169
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2170

    
2171
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2172
    /* but previous versions have the image flipped relative to vp3 */
2173
    if (s->theora < 0x030200)
2174
    {
2175
        s->flipped_image = 1;
2176
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2177
    }
2178

    
2179
    visible_width  = s->width  = get_bits(gb, 16) << 4;
2180
    visible_height = s->height = get_bits(gb, 16) << 4;
2181

    
2182
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2183
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2184
        s->width= s->height= 0;
2185
        return -1;
2186
    }
2187

    
2188
    if (s->theora >= 0x030400)
2189
    {
2190
        skip_bits(gb, 32); /* total number of superblocks in a frame */
2191
        // fixme, the next field is 36bits long
2192
        skip_bits(gb, 32); /* total number of blocks in a frame */
2193
        skip_bits(gb, 4); /* total number of blocks in a frame */
2194
        skip_bits(gb, 32); /* total number of macroblocks in a frame */
2195
    }
2196

    
2197
    if (s->theora >= 0x030200) {
2198
        visible_width  = get_bits_long(gb, 24);
2199
        visible_height = get_bits_long(gb, 24);
2200

    
2201
        skip_bits(gb, 8); /* offset x */
2202
        skip_bits(gb, 8); /* offset y */
2203
    }
2204

    
2205
    skip_bits(gb, 32); /* fps numerator */
2206
    skip_bits(gb, 32); /* fps denumerator */
2207
    skip_bits(gb, 24); /* aspect numerator */
2208
    skip_bits(gb, 24); /* aspect denumerator */
2209

    
2210
    if (s->theora < 0x030200)
2211
        skip_bits(gb, 5); /* keyframe frequency force */
2212
    skip_bits(gb, 8); /* colorspace */
2213
    if (s->theora >= 0x030400)
2214
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2215
    skip_bits(gb, 24); /* bitrate */
2216

    
2217
    skip_bits(gb, 6); /* quality hint */
2218

    
2219
    if (s->theora >= 0x030200)
2220
    {
2221
        skip_bits(gb, 5); /* keyframe frequency force */
2222

    
2223
        if (s->theora < 0x030400)
2224
            skip_bits(gb, 5); /* spare bits */
2225
    }
2226

    
2227
//    align_get_bits(gb);
2228

    
2229
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2230
        && visible_height <= s->height && visible_height > s->height-16)
2231
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2232
    else
2233
        avcodec_set_dimensions(avctx, s->width, s->height);
2234

    
2235
    return 0;
2236
}
2237

    
2238
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2239
{
2240
    Vp3DecodeContext *s = avctx->priv_data;
2241
    int i, n, matrices, inter, plane;
2242

    
2243
    if (s->theora >= 0x030200) {
2244
        n = get_bits(gb, 3);
2245
        /* loop filter limit values table */
2246
        for (i = 0; i < 64; i++) {
2247
            s->filter_limit_values[i] = get_bits(gb, n);
2248
            if (s->filter_limit_values[i] > 127) {
2249
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2250
                s->filter_limit_values[i] = 127;
2251
            }
2252
        }
2253
    }
2254

    
2255
    if (s->theora >= 0x030200)
2256
        n = get_bits(gb, 4) + 1;
2257
    else
2258
        n = 16;
2259
    /* quality threshold table */
2260
    for (i = 0; i < 64; i++)
2261
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2262

    
2263
    if (s->theora >= 0x030200)
2264
        n = get_bits(gb, 4) + 1;
2265
    else
2266
        n = 16;
2267
    /* dc scale factor table */
2268
    for (i = 0; i < 64; i++)
2269
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2270

    
2271
    if (s->theora >= 0x030200)
2272
        matrices = get_bits(gb, 9) + 1;
2273
    else
2274
        matrices = 3;
2275

    
2276
    if(matrices > 384){
2277
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2278
        return -1;
2279
    }
2280

    
2281
    for(n=0; n<matrices; n++){
2282
        for (i = 0; i < 64; i++)
2283
            s->base_matrix[n][i]= get_bits(gb, 8);
2284
    }
2285

    
2286
    for (inter = 0; inter <= 1; inter++) {
2287
        for (plane = 0; plane <= 2; plane++) {
2288
            int newqr= 1;
2289
            if (inter || plane > 0)
2290
                newqr = get_bits1(gb);
2291
            if (!newqr) {
2292
                int qtj, plj;
2293
                if(inter && get_bits1(gb)){
2294
                    qtj = 0;
2295
                    plj = plane;
2296
                }else{
2297
                    qtj= (3*inter + plane - 1) / 3;
2298
                    plj= (plane + 2) % 3;
2299
                }
2300
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2301
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2302
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2303
            } else {
2304
                int qri= 0;
2305
                int qi = 0;
2306

    
2307
                for(;;){
2308
                    i= get_bits(gb, av_log2(matrices-1)+1);
2309
                    if(i>= matrices){
2310
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2311
                        return -1;
2312
                    }
2313
                    s->qr_base[inter][plane][qri]= i;
2314
                    if(qi >= 63)
2315
                        break;
2316
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2317
                    s->qr_size[inter][plane][qri++]= i;
2318
                    qi += i;
2319
                }
2320

    
2321
                if (qi > 63) {
2322
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2323
                    return -1;
2324
                }
2325
                s->qr_count[inter][plane]= qri;
2326
            }
2327
        }
2328
    }
2329

    
2330
    /* Huffman tables */
2331
    for (s->hti = 0; s->hti < 80; s->hti++) {
2332
        s->entries = 0;
2333
        s->huff_code_size = 1;
2334
        if (!get_bits1(gb)) {
2335
            s->hbits = 0;
2336
            if(read_huffman_tree(avctx, gb))
2337
                return -1;
2338
            s->hbits = 1;
2339
            if(read_huffman_tree(avctx, gb))
2340
                return -1;
2341
        }
2342
    }
2343

    
2344
    s->theora_tables = 1;
2345

    
2346
    return 0;
2347
}
2348

    
2349
static av_cold int theora_decode_init(AVCodecContext *avctx)
2350
{
2351
    Vp3DecodeContext *s = avctx->priv_data;
2352
    GetBitContext gb;
2353
    int ptype;
2354
    uint8_t *header_start[3];
2355
    int header_len[3];
2356
    int i;
2357

    
2358
    s->theora = 1;
2359

    
2360
    if (!avctx->extradata_size)
2361
    {
2362
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2363
        return -1;
2364
    }
2365

    
2366
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2367
                              42, header_start, header_len) < 0) {
2368
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2369
        return -1;
2370
    }
2371

    
2372
  for(i=0;i<3;i++) {
2373
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2374

    
2375
    ptype = get_bits(&gb, 8);
2376

    
2377
     if (!(ptype & 0x80))
2378
     {
2379
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2380
//        return -1;
2381
     }
2382

    
2383
    // FIXME: Check for this as well.
2384
    skip_bits_long(&gb, 6*8); /* "theora" */
2385

    
2386
    switch(ptype)
2387
    {
2388
        case 0x80:
2389
            theora_decode_header(avctx, &gb);
2390
                break;
2391
        case 0x81:
2392
// FIXME: is this needed? it breaks sometimes
2393
//            theora_decode_comments(avctx, gb);
2394
            break;
2395
        case 0x82:
2396
            if (theora_decode_tables(avctx, &gb))
2397
                return -1;
2398
            break;
2399
        default:
2400
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2401
            break;
2402
    }
2403
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2404
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2405
    if (s->theora < 0x030200)
2406
        break;
2407
  }
2408

    
2409
    return vp3_decode_init(avctx);
2410
}
2411

    
2412
AVCodec theora_decoder = {
2413
    "theora",
2414
    CODEC_TYPE_VIDEO,
2415
    CODEC_ID_THEORA,
2416
    sizeof(Vp3DecodeContext),
2417
    theora_decode_init,
2418
    NULL,
2419
    vp3_decode_end,
2420
    vp3_decode_frame,
2421
    CODEC_CAP_DR1,
2422
    NULL,
2423
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2424
};
2425
#endif
2426

    
2427
AVCodec vp3_decoder = {
2428
    "vp3",
2429
    CODEC_TYPE_VIDEO,
2430
    CODEC_ID_VP3,
2431
    sizeof(Vp3DecodeContext),
2432
    vp3_decode_init,
2433
    NULL,
2434
    vp3_decode_end,
2435
    vp3_decode_frame,
2436
    CODEC_CAP_DR1,
2437
    NULL,
2438
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2439
};