Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ ecb51b25

History | View | Annotate | Download (77.5 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
    uint8_t coding_method;
57
    int8_t motion_x;
58
    int8_t motion_y;
59
    uint8_t qpi;
60
} Vp3Fragment;
61

    
62
#define SB_NOT_CODED        0
63
#define SB_PARTIALLY_CODED  1
64
#define SB_FULLY_CODED      2
65

    
66
// This is the maximum length of a single long bit run that can be encoded
67
// for superblock coding or block qps. Theora special-cases this to read a
68
// bit instead of flipping the current bit to allow for runs longer than 4129.
69
#define MAXIMUM_LONG_BIT_RUN 4129
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
    int last_slice_end;
139

    
140
    int qps[3];
141
    int nqps;
142
    int last_qps[3];
143

    
144
    int superblock_count;
145
    int y_superblock_width;
146
    int y_superblock_height;
147
    int c_superblock_width;
148
    int c_superblock_height;
149
    int u_superblock_start;
150
    int v_superblock_start;
151
    unsigned char *superblock_coding;
152

    
153
    int macroblock_count;
154
    int macroblock_width;
155
    int macroblock_height;
156

    
157
    int fragment_count;
158
    int fragment_width;
159
    int fragment_height;
160

    
161
    Vp3Fragment *all_fragments;
162
    uint8_t *coeff_counts;
163
    Coeff *coeffs;
164
    Coeff *next_coeff;
165
    int fragment_start[3];
166
    int data_offset[3];
167

    
168
    ScanTable scantable;
169

    
170
    /* tables */
171
    uint16_t coded_dc_scale_factor[64];
172
    uint32_t coded_ac_scale_factor[64];
173
    uint8_t base_matrix[384][64];
174
    uint8_t qr_count[2][3];
175
    uint8_t qr_size [2][3][64];
176
    uint16_t qr_base[2][3][64];
177

    
178
    /* this is a list of indexes into the all_fragments array indicating
179
     * which of the fragments are coded */
180
    int *coded_fragment_list;
181
    int coded_fragment_list_index;
182

    
183
    /* track which fragments have already been decoded; called 'fast'
184
     * because this data structure avoids having to iterate through every
185
     * fragment in coded_fragment_list; once a fragment has been fully
186
     * decoded, it is removed from this list */
187
    int *fast_fragment_list;
188
    int fragment_list_y_head;
189
    int fragment_list_c_head;
190

    
191
    VLC dc_vlc[16];
192
    VLC ac_vlc_1[16];
193
    VLC ac_vlc_2[16];
194
    VLC ac_vlc_3[16];
195
    VLC ac_vlc_4[16];
196

    
197
    VLC superblock_run_length_vlc;
198
    VLC fragment_run_length_vlc;
199
    VLC mode_code_vlc;
200
    VLC motion_vector_vlc;
201

    
202
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
203
     * index into them */
204
    DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
205

    
206
    /* This table contains superblock_count * 16 entries. Each set of 16
207
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
208
     * An entry will be -1 to indicate that no entry corresponds to that
209
     * index. */
210
    int *superblock_fragments;
211

    
212
    /* This is an array that indicates how a particular macroblock
213
     * is coded. */
214
    unsigned char *macroblock_coding;
215

    
216
    int first_coded_y_fragment;
217
    int first_coded_c_fragment;
218
    int last_coded_y_fragment;
219
    int last_coded_c_fragment;
220

    
221
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
222
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
223

    
224
    /* Huffman decode */
225
    int hti;
226
    unsigned int hbits;
227
    int entries;
228
    int huff_code_size;
229
    uint16_t huffman_table[80][32][2];
230

    
231
    uint8_t filter_limit_values[64];
232
    DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
233
} Vp3DecodeContext;
234

    
235
/************************************************************************
236
 * VP3 specific functions
237
 ************************************************************************/
238

    
239
/*
240
 * This function sets up all of the various blocks mappings:
241
 * superblocks <-> fragments, macroblocks <-> fragments,
242
 * superblocks <-> macroblocks
243
 *
244
 * Returns 0 is successful; returns 1 if *anything* went wrong.
245
 */
246
static int init_block_mapping(Vp3DecodeContext *s)
247
{
248
    int i, j;
249
    signed int hilbert_walk_mb[4];
250

    
251
    int current_fragment = 0;
252
    int current_width = 0;
253
    int current_height = 0;
254
    int right_edge = 0;
255
    int bottom_edge = 0;
256
    int superblock_row_inc = 0;
257
    int mapping_index = 0;
258

    
259
    int current_macroblock;
260
    int c_fragment;
261

    
262
    static const signed char travel_width[16] = {
263
         1,  1,  0, -1,
264
         0,  0,  1,  0,
265
         1,  0,  1,  0,
266
         0, -1,  0,  1
267
    };
268

    
269
    static const signed char travel_height[16] = {
270
         0,  0,  1,  0,
271
         1,  1,  0, -1,
272
         0,  1,  0, -1,
273
        -1,  0, -1,  0
274
    };
275

    
276
    hilbert_walk_mb[0] = 1;
277
    hilbert_walk_mb[1] = s->macroblock_width;
278
    hilbert_walk_mb[2] = 1;
279
    hilbert_walk_mb[3] = -s->macroblock_width;
280

    
281
    /* iterate through each superblock (all planes) and map the fragments */
282
    for (i = 0; i < s->superblock_count; i++) {
283
        /* time to re-assign the limits? */
284
        if (i == 0) {
285

    
286
            /* start of Y superblocks */
287
            right_edge = s->fragment_width;
288
            bottom_edge = s->fragment_height;
289
            current_width = -1;
290
            current_height = 0;
291
            superblock_row_inc = 3 * s->fragment_width -
292
                (s->y_superblock_width * 4 - s->fragment_width);
293

    
294
            /* the first operation for this variable is to advance by 1 */
295
            current_fragment = -1;
296

    
297
        } else if (i == s->u_superblock_start) {
298

    
299
            /* start of U superblocks */
300
            right_edge = s->fragment_width / 2;
301
            bottom_edge = s->fragment_height / 2;
302
            current_width = -1;
303
            current_height = 0;
304
            superblock_row_inc = 3 * (s->fragment_width / 2) -
305
                (s->c_superblock_width * 4 - s->fragment_width / 2);
306

    
307
            /* the first operation for this variable is to advance by 1 */
308
            current_fragment = s->fragment_start[1] - 1;
309

    
310
        } else if (i == s->v_superblock_start) {
311

    
312
            /* start of V superblocks */
313
            right_edge = s->fragment_width / 2;
314
            bottom_edge = s->fragment_height / 2;
315
            current_width = -1;
316
            current_height = 0;
317
            superblock_row_inc = 3 * (s->fragment_width / 2) -
318
                (s->c_superblock_width * 4 - s->fragment_width / 2);
319

    
320
            /* the first operation for this variable is to advance by 1 */
321
            current_fragment = s->fragment_start[2] - 1;
322

    
323
        }
324

    
325
        if (current_width >= right_edge - 1) {
326
            /* reset width and move to next superblock row */
327
            current_width = -1;
328
            current_height += 4;
329

    
330
            /* fragment is now at the start of a new superblock row */
331
            current_fragment += superblock_row_inc;
332
        }
333

    
334
        /* iterate through all 16 fragments in a superblock */
335
        for (j = 0; j < 16; j++) {
336
            current_fragment += travel_width[j] + right_edge * travel_height[j];
337
            current_width += travel_width[j];
338
            current_height += travel_height[j];
339

    
340
            /* check if the fragment is in bounds */
341
            if ((current_width < right_edge) &&
342
                (current_height < bottom_edge)) {
343
                s->superblock_fragments[mapping_index] = current_fragment;
344
            } else {
345
                s->superblock_fragments[mapping_index] = -1;
346
            }
347

    
348
            mapping_index++;
349
        }
350
    }
351

    
352
    return 0;  /* successful path out */
353
}
354

    
355
/*
356
 * This function wipes out all of the fragment data.
357
 */
358
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
359
{
360
    int i;
361

    
362
    /* zero out all of the fragment information */
363
    s->coded_fragment_list_index = 0;
364
    for (i = 0; i < s->fragment_count; i++) {
365
        s->coeff_counts[i] = 0;
366
        s->all_fragments[i].motion_x = 127;
367
        s->all_fragments[i].motion_y = 127;
368
        s->all_fragments[i].next_coeff= NULL;
369
        s->all_fragments[i].qpi = 0;
370
        s->coeffs[i].index=
371
        s->coeffs[i].coeff=0;
372
        s->coeffs[i].next= NULL;
373
    }
374
}
375

    
376
/*
377
 * This function sets up the dequantization tables used for a particular
378
 * frame.
379
 */
380
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
381
{
382
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
383
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
384
    int i, plane, inter, qri, bmi, bmj, qistart;
385

    
386
    for(inter=0; inter<2; inter++){
387
        for(plane=0; plane<3; plane++){
388
            int sum=0;
389
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
390
                sum+= s->qr_size[inter][plane][qri];
391
                if(s->qps[qpi] <= sum)
392
                    break;
393
            }
394
            qistart= sum - s->qr_size[inter][plane][qri];
395
            bmi= s->qr_base[inter][plane][qri  ];
396
            bmj= s->qr_base[inter][plane][qri+1];
397
            for(i=0; i<64; i++){
398
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
399
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
400
                            + s->qr_size[inter][plane][qri])
401
                           / (2*s->qr_size[inter][plane][qri]);
402

    
403
                int qmin= 8<<(inter + !i);
404
                int qscale= i ? ac_scale_factor : dc_scale_factor;
405

    
406
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
407
            }
408
            // all DC coefficients use the same quant so as not to interfere with DC prediction
409
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
410
        }
411
    }
412

    
413
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
414
}
415

    
416
/*
417
 * This function initializes the loop filter boundary limits if the frame's
418
 * quality index is different from the previous frame's.
419
 *
420
 * The filter_limit_values may not be larger than 127.
421
 */
422
static void init_loop_filter(Vp3DecodeContext *s)
423
{
424
    int *bounding_values= s->bounding_values_array+127;
425
    int filter_limit;
426
    int x;
427
    int value;
428

    
429
    filter_limit = s->filter_limit_values[s->qps[0]];
430

    
431
    /* set up the bounding values */
432
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
433
    for (x = 0; x < filter_limit; x++) {
434
        bounding_values[-x] = -x;
435
        bounding_values[x] = x;
436
    }
437
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
438
        bounding_values[ x] =  value;
439
        bounding_values[-x] = -value;
440
    }
441
    if (value)
442
        bounding_values[128] = value;
443
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
444
}
445

    
446
/*
447
 * This function unpacks all of the superblock/macroblock/fragment coding
448
 * information from the bitstream.
449
 */
450
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
451
{
452
    int bit = 0;
453
    int current_superblock = 0;
454
    int current_run = 0;
455
    int decode_fully_flags = 0;
456
    int decode_partial_blocks = 0;
457
    int first_c_fragment_seen;
458

    
459
    int i, j;
460
    int current_fragment;
461

    
462
    if (s->keyframe) {
463
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
464

    
465
    } else {
466

    
467
        /* unpack the list of partially-coded superblocks */
468
        bit = get_bits1(gb);
469
        while (current_superblock < s->superblock_count) {
470
                current_run = get_vlc2(gb,
471
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
472
                if (current_run == 34)
473
                    current_run += get_bits(gb, 12);
474

    
475
            if (current_superblock + current_run > s->superblock_count) {
476
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
477
                return -1;
478
            }
479

    
480
            memset(s->superblock_coding + current_superblock, bit, current_run);
481

    
482
            current_superblock += current_run;
483

    
484
                /* if any of the superblocks are not partially coded, flag
485
                 * a boolean to decode the list of fully-coded superblocks */
486
                if (bit == 0) {
487
                    decode_fully_flags = 1;
488
                } else {
489

    
490
                    /* make a note of the fact that there are partially coded
491
                     * superblocks */
492
                    decode_partial_blocks = 1;
493
                }
494

    
495
                bit ^= 1;
496
        }
497

    
498
        /* unpack the list of fully coded superblocks if any of the blocks were
499
         * not marked as partially coded in the previous step */
500
        if (decode_fully_flags) {
501

    
502
            current_superblock = 0;
503
            current_run = 0;
504
            bit = get_bits1(gb);
505
            /* toggle the bit because as soon as the first run length is
506
             * fetched the bit will be toggled again */
507
            bit ^= 1;
508
            while (current_superblock < s->superblock_count) {
509

    
510
                /* skip any superblocks already marked as partially coded */
511
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
512

    
513
                    if (current_run-- == 0) {
514
                        bit ^= 1;
515
                        current_run = get_vlc2(gb,
516
                            s->superblock_run_length_vlc.table, 6, 2);
517
                        if (current_run == 33)
518
                            current_run += get_bits(gb, 12);
519
                    }
520
                    s->superblock_coding[current_superblock] = 2*bit;
521
                }
522
                current_superblock++;
523
            }
524
        }
525

    
526
        /* if there were partial blocks, initialize bitstream for
527
         * unpacking fragment codings */
528
        if (decode_partial_blocks) {
529

    
530
            current_run = 0;
531
            bit = get_bits1(gb);
532
            /* toggle the bit because as soon as the first run length is
533
             * fetched the bit will be toggled again */
534
            bit ^= 1;
535
        }
536
    }
537

    
538
    /* figure out which fragments are coded; iterate through each
539
     * superblock (all planes) */
540
    s->coded_fragment_list_index = 0;
541
    s->next_coeff= s->coeffs + s->fragment_count;
542
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
543
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
544
    first_c_fragment_seen = 0;
545
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
546
    for (i = 0; i < s->superblock_count; i++) {
547

    
548
        /* iterate through all 16 fragments in a superblock */
549
        for (j = 0; j < 16; j++) {
550

    
551
            /* if the fragment is in bounds, check its coding status */
552
            current_fragment = s->superblock_fragments[i * 16 + j];
553
            if (current_fragment >= s->fragment_count) {
554
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
555
                    current_fragment, s->fragment_count);
556
                return 1;
557
            }
558
            if (current_fragment != -1) {
559
                if (s->superblock_coding[i] == SB_NOT_CODED) {
560

    
561
                    /* copy all the fragments from the prior frame */
562
                    s->all_fragments[current_fragment].coding_method =
563
                        MODE_COPY;
564

    
565
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
566

    
567
                    /* fragment may or may not be coded; this is the case
568
                     * that cares about the fragment coding runs */
569
                    if (current_run-- == 0) {
570
                        bit ^= 1;
571
                        current_run = get_vlc2(gb,
572
                            s->fragment_run_length_vlc.table, 5, 2);
573
                    }
574

    
575
                    if (bit) {
576
                        /* default mode; actual mode will be decoded in
577
                         * the next phase */
578
                        s->all_fragments[current_fragment].coding_method =
579
                            MODE_INTER_NO_MV;
580
                        s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
581
                        s->coded_fragment_list[s->coded_fragment_list_index] =
582
                            current_fragment;
583
                        if ((current_fragment >= s->fragment_start[1]) &&
584
                            (s->last_coded_y_fragment == -1) &&
585
                            (!first_c_fragment_seen)) {
586
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
587
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
588
                            first_c_fragment_seen = 1;
589
                        }
590
                        s->coded_fragment_list_index++;
591
                    } else {
592
                        /* not coded; copy this fragment from the prior frame */
593
                        s->all_fragments[current_fragment].coding_method =
594
                            MODE_COPY;
595
                    }
596

    
597
                } else {
598

    
599
                    /* fragments are fully coded in this superblock; actual
600
                     * coding will be determined in next step */
601
                    s->all_fragments[current_fragment].coding_method =
602
                        MODE_INTER_NO_MV;
603
                    s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
604
                    s->coded_fragment_list[s->coded_fragment_list_index] =
605
                        current_fragment;
606
                    if ((current_fragment >= s->fragment_start[1]) &&
607
                        (s->last_coded_y_fragment == -1) &&
608
                        (!first_c_fragment_seen)) {
609
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
610
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
611
                        first_c_fragment_seen = 1;
612
                    }
613
                    s->coded_fragment_list_index++;
614
                }
615
            }
616
        }
617
    }
618

    
619
    if (!first_c_fragment_seen)
620
        /* only Y fragments coded in this frame */
621
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
622
    else
623
        /* end the list of coded C fragments */
624
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
625

    
626
    for (i = 0; i < s->fragment_count - 1; i++) {
627
        s->fast_fragment_list[i] = i + 1;
628
    }
629
    s->fast_fragment_list[s->fragment_count - 1] = -1;
630

    
631
    if (s->last_coded_y_fragment == -1)
632
        s->fragment_list_y_head = -1;
633
    else {
634
        s->fragment_list_y_head = s->first_coded_y_fragment;
635
        s->fast_fragment_list[s->last_coded_y_fragment] = -1;
636
    }
637

    
638
    if (s->last_coded_c_fragment == -1)
639
        s->fragment_list_c_head = -1;
640
    else {
641
        s->fragment_list_c_head = s->first_coded_c_fragment;
642
        s->fast_fragment_list[s->last_coded_c_fragment] = -1;
643
    }
644

    
645
    return 0;
646
}
647

    
648
/*
649
 * This function unpacks all the coding mode data for individual macroblocks
650
 * from the bitstream.
651
 */
652
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
653
{
654
    int i, j, k, sb_x, sb_y;
655
    int scheme;
656
    int current_macroblock;
657
    int current_fragment;
658
    int coding_mode;
659
    int custom_mode_alphabet[CODING_MODE_COUNT];
660
    const int *alphabet;
661

    
662
    if (s->keyframe) {
663
        for (i = 0; i < s->fragment_count; i++)
664
            s->all_fragments[i].coding_method = MODE_INTRA;
665

    
666
    } else {
667

    
668
        /* fetch the mode coding scheme for this frame */
669
        scheme = get_bits(gb, 3);
670

    
671
        /* is it a custom coding scheme? */
672
        if (scheme == 0) {
673
            for (i = 0; i < 8; i++)
674
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
675
            for (i = 0; i < 8; i++)
676
                custom_mode_alphabet[get_bits(gb, 3)] = i;
677
            alphabet = custom_mode_alphabet;
678
        } else
679
            alphabet = ModeAlphabet[scheme-1];
680

    
681
        /* iterate through all of the macroblocks that contain 1 or more
682
         * coded fragments */
683
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
684
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
685

    
686
            for (j = 0; j < 4; j++) {
687
                int mb_x = 2*sb_x +   (j>>1);
688
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
689
                int frags_coded = 0;
690
                current_macroblock = mb_y * s->macroblock_width + mb_x;
691

    
692
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
693
                    continue;
694

    
695
#define BLOCK_X (2*mb_x + (k&1))
696
#define BLOCK_Y (2*mb_y + (k>>1))
697
                /* coding modes are only stored if the macroblock has at least one
698
                 * luma block coded, otherwise it must be INTER_NO_MV */
699
                for (k = 0; k < 4; k++) {
700
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
701
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
702
                        break;
703
                }
704
                if (k == 4) {
705
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
706
                    continue;
707
                }
708

    
709
                /* mode 7 means get 3 bits for each coding mode */
710
                if (scheme == 7)
711
                    coding_mode = get_bits(gb, 3);
712
                else
713
                    coding_mode = alphabet
714
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
715

    
716
                s->macroblock_coding[current_macroblock] = coding_mode;
717
                for (k = 0; k < 4; k++) {
718
                    current_fragment =
719
                        BLOCK_Y*s->fragment_width + BLOCK_X;
720
                    if (s->all_fragments[current_fragment].coding_method !=
721
                        MODE_COPY)
722
                        s->all_fragments[current_fragment].coding_method =
723
                            coding_mode;
724
                }
725
                for (k = 0; k < 2; k++) {
726
                    current_fragment = s->fragment_start[k+1] +
727
                        mb_y*(s->fragment_width>>1) + mb_x;
728
                    if (s->all_fragments[current_fragment].coding_method !=
729
                        MODE_COPY)
730
                        s->all_fragments[current_fragment].coding_method =
731
                            coding_mode;
732
                }
733
            }
734
            }
735
        }
736
    }
737

    
738
    return 0;
739
}
740

    
741
/*
742
 * This function unpacks all the motion vectors for the individual
743
 * macroblocks from the bitstream.
744
 */
745
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
746
{
747
    int j, k, sb_x, sb_y;
748
    int coding_mode;
749
    int motion_x[6];
750
    int motion_y[6];
751
    int last_motion_x = 0;
752
    int last_motion_y = 0;
753
    int prior_last_motion_x = 0;
754
    int prior_last_motion_y = 0;
755
    int current_macroblock;
756
    int current_fragment;
757

    
758
    if (s->keyframe)
759
        return 0;
760

    
761
    memset(motion_x, 0, 6 * sizeof(int));
762
    memset(motion_y, 0, 6 * sizeof(int));
763

    
764
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
765
    coding_mode = get_bits1(gb);
766

    
767
    /* iterate through all of the macroblocks that contain 1 or more
768
     * coded fragments */
769
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
770
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
771

    
772
        for (j = 0; j < 4; j++) {
773
            int mb_x = 2*sb_x +   (j>>1);
774
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
775
            current_macroblock = mb_y * s->macroblock_width + mb_x;
776

    
777
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
778
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
779
                continue;
780

    
781
            switch (s->macroblock_coding[current_macroblock]) {
782

    
783
            case MODE_INTER_PLUS_MV:
784
            case MODE_GOLDEN_MV:
785
                /* all 6 fragments use the same motion vector */
786
                if (coding_mode == 0) {
787
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
788
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
789
                } else {
790
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
791
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
792
                }
793

    
794
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
795
                if (s->macroblock_coding[current_macroblock] ==
796
                    MODE_INTER_PLUS_MV) {
797
                    prior_last_motion_x = last_motion_x;
798
                    prior_last_motion_y = last_motion_y;
799
                    last_motion_x = motion_x[0];
800
                    last_motion_y = motion_y[0];
801
                }
802
                break;
803

    
804
            case MODE_INTER_FOURMV:
805
                /* vector maintenance */
806
                prior_last_motion_x = last_motion_x;
807
                prior_last_motion_y = last_motion_y;
808

    
809
                /* fetch 4 vectors from the bitstream, one for each
810
                 * Y fragment, then average for the C fragment vectors */
811
                motion_x[4] = motion_y[4] = 0;
812
                for (k = 0; k < 4; k++) {
813
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
814
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
815
                        if (coding_mode == 0) {
816
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
817
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
818
                        } else {
819
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
820
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
821
                        }
822
                        last_motion_x = motion_x[k];
823
                        last_motion_y = motion_y[k];
824
                    } else {
825
                        motion_x[k] = 0;
826
                        motion_y[k] = 0;
827
                    }
828
                    motion_x[4] += motion_x[k];
829
                    motion_y[4] += motion_y[k];
830
                }
831

    
832
                motion_x[5]=
833
                motion_x[4]= RSHIFT(motion_x[4], 2);
834
                motion_y[5]=
835
                motion_y[4]= RSHIFT(motion_y[4], 2);
836
                break;
837

    
838
            case MODE_INTER_LAST_MV:
839
                /* all 6 fragments use the last motion vector */
840
                motion_x[0] = last_motion_x;
841
                motion_y[0] = last_motion_y;
842

    
843
                /* no vector maintenance (last vector remains the
844
                 * last vector) */
845
                break;
846

    
847
            case MODE_INTER_PRIOR_LAST:
848
                /* all 6 fragments use the motion vector prior to the
849
                 * last motion vector */
850
                motion_x[0] = prior_last_motion_x;
851
                motion_y[0] = prior_last_motion_y;
852

    
853
                /* vector maintenance */
854
                prior_last_motion_x = last_motion_x;
855
                prior_last_motion_y = last_motion_y;
856
                last_motion_x = motion_x[0];
857
                last_motion_y = motion_y[0];
858
                break;
859

    
860
            default:
861
                /* covers intra, inter without MV, golden without MV */
862
                motion_x[0] = 0;
863
                motion_y[0] = 0;
864

    
865
                /* no vector maintenance */
866
                break;
867
            }
868

    
869
            /* assign the motion vectors to the correct fragments */
870
            for (k = 0; k < 4; k++) {
871
                current_fragment =
872
                    BLOCK_Y*s->fragment_width + BLOCK_X;
873
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
874
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
875
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
876
                } else {
877
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
878
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
879
                }
880
            }
881
            for (k = 0; k < 2; k++) {
882
                current_fragment = s->fragment_start[k+1] +
883
                    mb_y*(s->fragment_width>>1) + mb_x;
884
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
885
                    s->all_fragments[current_fragment].motion_x = motion_x[k+4];
886
                    s->all_fragments[current_fragment].motion_y = motion_y[k+4];
887
                } else {
888
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
889
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
890
                }
891
            }
892
        }
893
        }
894
    }
895

    
896
    return 0;
897
}
898

    
899
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
900
{
901
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
902
    int num_blocks = s->coded_fragment_list_index;
903

    
904
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
905
        i = blocks_decoded = num_blocks_at_qpi = 0;
906

    
907
        bit = get_bits1(gb);
908

    
909
        do {
910
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
911
            if (run_length == 34)
912
                run_length += get_bits(gb, 12);
913
            blocks_decoded += run_length;
914

    
915
            if (!bit)
916
                num_blocks_at_qpi += run_length;
917

    
918
            for (j = 0; j < run_length; i++) {
919
                if (i >= s->coded_fragment_list_index)
920
                    return -1;
921

    
922
                if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
923
                    s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
924
                    j++;
925
                }
926
            }
927

    
928
            if (run_length == MAXIMUM_LONG_BIT_RUN)
929
                bit = get_bits1(gb);
930
            else
931
                bit ^= 1;
932
        } while (blocks_decoded < num_blocks);
933

    
934
        num_blocks -= num_blocks_at_qpi;
935
    }
936

    
937
    return 0;
938
}
939

    
940
/*
941
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
942
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
943
 * data. This function unpacks all the VLCs for either the Y plane or both
944
 * C planes, and is called for DC coefficients or different AC coefficient
945
 * levels (since different coefficient types require different VLC tables.
946
 *
947
 * This function returns a residual eob run. E.g, if a particular token gave
948
 * instructions to EOB the next 5 fragments and there were only 2 fragments
949
 * left in the current fragment range, 3 would be returned so that it could
950
 * be passed into the next call to this same function.
951
 */
952
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
953
                        VLC *table, int coeff_index,
954
                        int y_plane,
955
                        int eob_run)
956
{
957
    int i;
958
    int token;
959
    int zero_run = 0;
960
    DCTELEM coeff = 0;
961
    Vp3Fragment *fragment;
962
    int bits_to_get;
963
    int next_fragment;
964
    int previous_fragment;
965
    int fragment_num;
966
    int *list_head;
967

    
968
    /* local references to structure members to avoid repeated deferences */
969
    uint8_t *perm= s->scantable.permutated;
970
    int *coded_fragment_list = s->coded_fragment_list;
971
    Vp3Fragment *all_fragments = s->all_fragments;
972
    uint8_t *coeff_counts = s->coeff_counts;
973
    VLC_TYPE (*vlc_table)[2] = table->table;
974
    int *fast_fragment_list = s->fast_fragment_list;
975

    
976
    if (y_plane) {
977
        next_fragment = s->fragment_list_y_head;
978
        list_head = &s->fragment_list_y_head;
979
    } else {
980
        next_fragment = s->fragment_list_c_head;
981
        list_head = &s->fragment_list_c_head;
982
    }
983

    
984
    i = next_fragment;
985
    previous_fragment = -1;  /* this indicates that the previous fragment is actually the list head */
986
    while (i != -1) {
987
        fragment_num = coded_fragment_list[i];
988

    
989
        if (coeff_counts[fragment_num] > coeff_index) {
990
            previous_fragment = i;
991
            i = fast_fragment_list[i];
992
            continue;
993
        }
994
        fragment = &all_fragments[fragment_num];
995

    
996
        if (!eob_run) {
997
            /* decode a VLC into a token */
998
            token = get_vlc2(gb, vlc_table, 5, 3);
999
            /* use the token to get a zero run, a coefficient, and an eob run */
1000
            if (token <= 6) {
1001
                eob_run = eob_run_base[token];
1002
                if (eob_run_get_bits[token])
1003
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
1004
                coeff = zero_run = 0;
1005
            } else {
1006
                bits_to_get = coeff_get_bits[token];
1007
                if (bits_to_get)
1008
                    bits_to_get = get_bits(gb, bits_to_get);
1009
                coeff = coeff_tables[token][bits_to_get];
1010

    
1011
                zero_run = zero_run_base[token];
1012
                if (zero_run_get_bits[token])
1013
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
1014
            }
1015
        }
1016

    
1017
        if (!eob_run) {
1018
            coeff_counts[fragment_num] += zero_run;
1019
            if (coeff_counts[fragment_num] < 64){
1020
                fragment->next_coeff->coeff= coeff;
1021
                fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1022
                fragment->next_coeff->next= s->next_coeff;
1023
                s->next_coeff->next=NULL;
1024
                fragment->next_coeff= s->next_coeff++;
1025
            }
1026
            /* previous fragment is now this fragment */
1027
            previous_fragment = i;
1028
        } else {
1029
            coeff_counts[fragment_num] |= 128;
1030
            eob_run--;
1031
            /* remove this fragment from the list */
1032
            if (previous_fragment != -1)
1033
                fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1034
            else
1035
                *list_head = fast_fragment_list[i];
1036
            /* previous fragment remains unchanged */
1037
        }
1038

    
1039
        i = fast_fragment_list[i];
1040
    }
1041

    
1042
    return eob_run;
1043
}
1044

    
1045
static void reverse_dc_prediction(Vp3DecodeContext *s,
1046
                                  int first_fragment,
1047
                                  int fragment_width,
1048
                                  int fragment_height);
1049
/*
1050
 * This function unpacks all of the DCT coefficient data from the
1051
 * bitstream.
1052
 */
1053
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1054
{
1055
    int i;
1056
    int dc_y_table;
1057
    int dc_c_table;
1058
    int ac_y_table;
1059
    int ac_c_table;
1060
    int residual_eob_run = 0;
1061
    VLC *y_tables[64];
1062
    VLC *c_tables[64];
1063

    
1064
    /* fetch the DC table indexes */
1065
    dc_y_table = get_bits(gb, 4);
1066
    dc_c_table = get_bits(gb, 4);
1067

    
1068
    /* unpack the Y plane DC coefficients */
1069
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1070
        1, residual_eob_run);
1071

    
1072
    /* reverse prediction of the Y-plane DC coefficients */
1073
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1074

    
1075
    /* unpack the C plane DC coefficients */
1076
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1077
        0, residual_eob_run);
1078

    
1079
    /* reverse prediction of the C-plane DC coefficients */
1080
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1081
    {
1082
        reverse_dc_prediction(s, s->fragment_start[1],
1083
            s->fragment_width / 2, s->fragment_height / 2);
1084
        reverse_dc_prediction(s, s->fragment_start[2],
1085
            s->fragment_width / 2, s->fragment_height / 2);
1086
    }
1087

    
1088
    /* fetch the AC table indexes */
1089
    ac_y_table = get_bits(gb, 4);
1090
    ac_c_table = get_bits(gb, 4);
1091

    
1092
    /* build tables of AC VLC tables */
1093
    for (i = 1; i <= 5; i++) {
1094
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1095
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1096
    }
1097
    for (i = 6; i <= 14; i++) {
1098
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1099
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1100
    }
1101
    for (i = 15; i <= 27; i++) {
1102
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1103
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1104
    }
1105
    for (i = 28; i <= 63; i++) {
1106
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1107
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1108
    }
1109

    
1110
    /* decode all AC coefficents */
1111
    for (i = 1; i <= 63; i++) {
1112
        if (s->fragment_list_y_head != -1)
1113
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1114
                1, residual_eob_run);
1115

    
1116
        if (s->fragment_list_c_head != -1)
1117
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1118
                0, residual_eob_run);
1119
    }
1120

    
1121
    return 0;
1122
}
1123

    
1124
/*
1125
 * This function reverses the DC prediction for each coded fragment in
1126
 * the frame. Much of this function is adapted directly from the original
1127
 * VP3 source code.
1128
 */
1129
#define COMPATIBLE_FRAME(x) \
1130
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1131
#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1132

    
1133
static void reverse_dc_prediction(Vp3DecodeContext *s,
1134
                                  int first_fragment,
1135
                                  int fragment_width,
1136
                                  int fragment_height)
1137
{
1138

    
1139
#define PUL 8
1140
#define PU 4
1141
#define PUR 2
1142
#define PL 1
1143

    
1144
    int x, y;
1145
    int i = first_fragment;
1146

    
1147
    int predicted_dc;
1148

    
1149
    /* DC values for the left, up-left, up, and up-right fragments */
1150
    int vl, vul, vu, vur;
1151

    
1152
    /* indexes for the left, up-left, up, and up-right fragments */
1153
    int l, ul, u, ur;
1154

    
1155
    /*
1156
     * The 6 fields mean:
1157
     *   0: up-left multiplier
1158
     *   1: up multiplier
1159
     *   2: up-right multiplier
1160
     *   3: left multiplier
1161
     */
1162
    static const int predictor_transform[16][4] = {
1163
        {  0,  0,  0,  0},
1164
        {  0,  0,  0,128},        // PL
1165
        {  0,  0,128,  0},        // PUR
1166
        {  0,  0, 53, 75},        // PUR|PL
1167
        {  0,128,  0,  0},        // PU
1168
        {  0, 64,  0, 64},        // PU|PL
1169
        {  0,128,  0,  0},        // PU|PUR
1170
        {  0,  0, 53, 75},        // PU|PUR|PL
1171
        {128,  0,  0,  0},        // PUL
1172
        {  0,  0,  0,128},        // PUL|PL
1173
        { 64,  0, 64,  0},        // PUL|PUR
1174
        {  0,  0, 53, 75},        // PUL|PUR|PL
1175
        {  0,128,  0,  0},        // PUL|PU
1176
       {-104,116,  0,116},        // PUL|PU|PL
1177
        { 24, 80, 24,  0},        // PUL|PU|PUR
1178
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1179
    };
1180

    
1181
    /* This table shows which types of blocks can use other blocks for
1182
     * prediction. For example, INTRA is the only mode in this table to
1183
     * have a frame number of 0. That means INTRA blocks can only predict
1184
     * from other INTRA blocks. There are 2 golden frame coding types;
1185
     * blocks encoding in these modes can only predict from other blocks
1186
     * that were encoded with these 1 of these 2 modes. */
1187
    static const unsigned char compatible_frame[9] = {
1188
        1,    /* MODE_INTER_NO_MV */
1189
        0,    /* MODE_INTRA */
1190
        1,    /* MODE_INTER_PLUS_MV */
1191
        1,    /* MODE_INTER_LAST_MV */
1192
        1,    /* MODE_INTER_PRIOR_MV */
1193
        2,    /* MODE_USING_GOLDEN */
1194
        2,    /* MODE_GOLDEN_MV */
1195
        1,    /* MODE_INTER_FOUR_MV */
1196
        3     /* MODE_COPY */
1197
    };
1198
    int current_frame_type;
1199

    
1200
    /* there is a last DC predictor for each of the 3 frame types */
1201
    short last_dc[3];
1202

    
1203
    int transform = 0;
1204

    
1205
    vul = vu = vur = vl = 0;
1206
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1207

    
1208
    /* for each fragment row... */
1209
    for (y = 0; y < fragment_height; y++) {
1210

    
1211
        /* for each fragment in a row... */
1212
        for (x = 0; x < fragment_width; x++, i++) {
1213

    
1214
            /* reverse prediction if this block was coded */
1215
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1216

    
1217
                current_frame_type =
1218
                    compatible_frame[s->all_fragments[i].coding_method];
1219

    
1220
                transform= 0;
1221
                if(x){
1222
                    l= i-1;
1223
                    vl = DC_COEFF(l);
1224
                    if(COMPATIBLE_FRAME(l))
1225
                        transform |= PL;
1226
                }
1227
                if(y){
1228
                    u= i-fragment_width;
1229
                    vu = DC_COEFF(u);
1230
                    if(COMPATIBLE_FRAME(u))
1231
                        transform |= PU;
1232
                    if(x){
1233
                        ul= i-fragment_width-1;
1234
                        vul = DC_COEFF(ul);
1235
                        if(COMPATIBLE_FRAME(ul))
1236
                            transform |= PUL;
1237
                    }
1238
                    if(x + 1 < fragment_width){
1239
                        ur= i-fragment_width+1;
1240
                        vur = DC_COEFF(ur);
1241
                        if(COMPATIBLE_FRAME(ur))
1242
                            transform |= PUR;
1243
                    }
1244
                }
1245

    
1246
                if (transform == 0) {
1247

    
1248
                    /* if there were no fragments to predict from, use last
1249
                     * DC saved */
1250
                    predicted_dc = last_dc[current_frame_type];
1251
                } else {
1252

    
1253
                    /* apply the appropriate predictor transform */
1254
                    predicted_dc =
1255
                        (predictor_transform[transform][0] * vul) +
1256
                        (predictor_transform[transform][1] * vu) +
1257
                        (predictor_transform[transform][2] * vur) +
1258
                        (predictor_transform[transform][3] * vl);
1259

    
1260
                    predicted_dc /= 128;
1261

    
1262
                    /* check for outranging on the [ul u l] and
1263
                     * [ul u ur l] predictors */
1264
                    if ((transform == 15) || (transform == 13)) {
1265
                        if (FFABS(predicted_dc - vu) > 128)
1266
                            predicted_dc = vu;
1267
                        else if (FFABS(predicted_dc - vl) > 128)
1268
                            predicted_dc = vl;
1269
                        else if (FFABS(predicted_dc - vul) > 128)
1270
                            predicted_dc = vul;
1271
                    }
1272
                }
1273

    
1274
                /* at long last, apply the predictor */
1275
                if(s->coeffs[i].index){
1276
                    *s->next_coeff= s->coeffs[i];
1277
                    s->coeffs[i].index=0;
1278
                    s->coeffs[i].coeff=0;
1279
                    s->coeffs[i].next= s->next_coeff++;
1280
                }
1281
                s->coeffs[i].coeff += predicted_dc;
1282
                /* save the DC */
1283
                last_dc[current_frame_type] = DC_COEFF(i);
1284
                if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1285
                    s->coeff_counts[i]= 129;
1286
//                    s->all_fragments[i].next_coeff= s->next_coeff;
1287
                    s->coeffs[i].next= s->next_coeff;
1288
                    (s->next_coeff++)->next=NULL;
1289
                }
1290
            }
1291
        }
1292
    }
1293
}
1294

    
1295
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1296
{
1297
    int x, y;
1298
    int *bounding_values= s->bounding_values_array+127;
1299

    
1300
    int width           = s->fragment_width  >> !!plane;
1301
    int height          = s->fragment_height >> !!plane;
1302
    int fragment        = s->fragment_start        [plane] + ystart * width;
1303
    int stride          = s->current_frame.linesize[plane];
1304
    uint8_t *plane_data = s->current_frame.data    [plane];
1305
    if (!s->flipped_image) stride = -stride;
1306
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1307

    
1308
    for (y = ystart; y < yend; y++) {
1309

    
1310
        for (x = 0; x < width; x++) {
1311
            /* This code basically just deblocks on the edges of coded blocks.
1312
             * However, it has to be much more complicated because of the
1313
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1314
             * because some pixels get filtered twice. */
1315
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1316
            {
1317
                /* do not perform left edge filter for left columns frags */
1318
                if (x > 0) {
1319
                    s->dsp.vp3_h_loop_filter(
1320
                        plane_data + 8*x,
1321
                        stride, bounding_values);
1322
                }
1323

    
1324
                /* do not perform top edge filter for top row fragments */
1325
                if (y > 0) {
1326
                    s->dsp.vp3_v_loop_filter(
1327
                        plane_data + 8*x,
1328
                        stride, bounding_values);
1329
                }
1330

    
1331
                /* do not perform right edge filter for right column
1332
                 * fragments or if right fragment neighbor is also coded
1333
                 * in this frame (it will be filtered in next iteration) */
1334
                if ((x < width - 1) &&
1335
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1336
                    s->dsp.vp3_h_loop_filter(
1337
                        plane_data + 8*x + 8,
1338
                        stride, bounding_values);
1339
                }
1340

    
1341
                /* do not perform bottom edge filter for bottom row
1342
                 * fragments or if bottom fragment neighbor is also coded
1343
                 * in this frame (it will be filtered in the next row) */
1344
                if ((y < height - 1) &&
1345
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1346
                    s->dsp.vp3_v_loop_filter(
1347
                        plane_data + 8*x + 8*stride,
1348
                        stride, bounding_values);
1349
                }
1350
            }
1351

    
1352
            fragment++;
1353
        }
1354
        plane_data += 8*stride;
1355
    }
1356
}
1357

    
1358
/**
1359
 * called when all pixels up to row y are complete
1360
 */
1361
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1362
{
1363
    int h, cy;
1364
    int offset[4];
1365

    
1366
    if(s->avctx->draw_horiz_band==NULL)
1367
        return;
1368

    
1369
    h= y - s->last_slice_end;
1370
    y -= h;
1371

    
1372
    if (!s->flipped_image) {
1373
        if (y == 0)
1374
            h -= s->height - s->avctx->height;  // account for non-mod16
1375
        y = s->height - y - h;
1376
    }
1377

    
1378
    cy = y >> 1;
1379
    offset[0] = s->current_frame.linesize[0]*y;
1380
    offset[1] = s->current_frame.linesize[1]*cy;
1381
    offset[2] = s->current_frame.linesize[2]*cy;
1382
    offset[3] = 0;
1383

    
1384
    emms_c();
1385
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1386
    s->last_slice_end= y + h;
1387
}
1388

    
1389
/*
1390
 * Perform the final rendering for a particular slice of data.
1391
 * The slice number ranges from 0..(macroblock_height - 1).
1392
 */
1393
static void render_slice(Vp3DecodeContext *s, int slice)
1394
{
1395
    int x;
1396
    int16_t *dequantizer;
1397
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1398
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1399
    int motion_halfpel_index;
1400
    uint8_t *motion_source;
1401
    int plane;
1402

    
1403
    if (slice >= s->macroblock_height)
1404
        return;
1405

    
1406
    for (plane = 0; plane < 3; plane++) {
1407
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1408
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1409
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1410
        int stride            = s->current_frame.linesize[plane];
1411
        int plane_width       = s->width  >> !!plane;
1412
        int plane_height      = s->height >> !!plane;
1413
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
1414
        int slice_height = y + (FRAGMENT_PIXELS << !plane);
1415
        int i = s->fragment_start[plane] + (y>>3)*(s->fragment_width>>!!plane);
1416

    
1417
        if (!s->flipped_image) stride = -stride;
1418
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1419
            continue;
1420

    
1421

    
1422
        if(FFABS(stride) > 2048)
1423
            return; //various tables are fixed size
1424

    
1425
        /* for each fragment row in the slice (both of them)... */
1426
        for (; y < slice_height; y += 8) {
1427

    
1428
            /* for each fragment in a row... */
1429
            for (x = 0; x < plane_width; x += 8, i++) {
1430
                int first_pixel = y*stride + x;
1431

    
1432
                if ((i < 0) || (i >= s->fragment_count)) {
1433
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
1434
                    return;
1435
                }
1436

    
1437
                /* transform if this block was coded */
1438
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1439

    
1440
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1441
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1442
                        motion_source= golden_plane;
1443
                    else
1444
                        motion_source= last_plane;
1445

    
1446
                    motion_source += first_pixel;
1447
                    motion_halfpel_index = 0;
1448

    
1449
                    /* sort out the motion vector if this fragment is coded
1450
                     * using a motion vector method */
1451
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1452
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1453
                        int src_x, src_y;
1454
                        motion_x = s->all_fragments[i].motion_x;
1455
                        motion_y = s->all_fragments[i].motion_y;
1456
                        if(plane){
1457
                            motion_x= (motion_x>>1) | (motion_x&1);
1458
                            motion_y= (motion_y>>1) | (motion_y&1);
1459
                        }
1460

    
1461
                        src_x= (motion_x>>1) + x;
1462
                        src_y= (motion_y>>1) + y;
1463
                        if ((motion_x == 127) || (motion_y == 127))
1464
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1465

    
1466
                        motion_halfpel_index = motion_x & 0x01;
1467
                        motion_source += (motion_x >> 1);
1468

    
1469
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1470
                        motion_source += ((motion_y >> 1) * stride);
1471

    
1472
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1473
                            uint8_t *temp= s->edge_emu_buffer;
1474
                            if(stride<0) temp -= 9*stride;
1475
                            else temp += 9*stride;
1476

    
1477
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1478
                            motion_source= temp;
1479
                        }
1480
                    }
1481

    
1482

    
1483
                    /* first, take care of copying a block from either the
1484
                     * previous or the golden frame */
1485
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1486
                        /* Note, it is possible to implement all MC cases with
1487
                           put_no_rnd_pixels_l2 which would look more like the
1488
                           VP3 source but this would be slower as
1489
                           put_no_rnd_pixels_tab is better optimzed */
1490
                        if(motion_halfpel_index != 3){
1491
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1492
                                output_plane + first_pixel,
1493
                                motion_source, stride, 8);
1494
                        }else{
1495
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1496
                            s->dsp.put_no_rnd_pixels_l2[1](
1497
                                output_plane + first_pixel,
1498
                                motion_source - d,
1499
                                motion_source + stride + 1 + d,
1500
                                stride, 8);
1501
                        }
1502
                        dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1503
                    }else{
1504
                        dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1505
                    }
1506

    
1507
                    /* dequantize the DCT coefficients */
1508
                    if(s->avctx->idct_algo==FF_IDCT_VP3){
1509
                        Coeff *coeff= s->coeffs + i;
1510
                        s->dsp.clear_block(block);
1511
                        while(coeff->next){
1512
                            block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1513
                            coeff= coeff->next;
1514
                        }
1515
                    }else{
1516
                        Coeff *coeff= s->coeffs + i;
1517
                        s->dsp.clear_block(block);
1518
                        while(coeff->next){
1519
                            block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1520
                            coeff= coeff->next;
1521
                        }
1522
                    }
1523

    
1524
                    /* invert DCT and place (or add) in final output */
1525

    
1526
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1527
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1528
                            block[0] += 128<<3;
1529
                        s->dsp.idct_put(
1530
                            output_plane + first_pixel,
1531
                            stride,
1532
                            block);
1533
                    } else {
1534
                        s->dsp.idct_add(
1535
                            output_plane + first_pixel,
1536
                            stride,
1537
                            block);
1538
                    }
1539
                } else {
1540

    
1541
                    /* copy directly from the previous frame */
1542
                    s->dsp.put_pixels_tab[1][0](
1543
                        output_plane + first_pixel,
1544
                        last_plane + first_pixel,
1545
                        stride, 8);
1546

    
1547
                }
1548
            }
1549
            // Filter the previous block row. We can't filter the current row yet
1550
            // since it needs pixels from the next row
1551
            if (y > 0)
1552
                apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
1553
        }
1554
    }
1555

    
1556
     /* this looks like a good place for slice dispatch... */
1557
     /* algorithm:
1558
      *   if (slice == s->macroblock_height - 1)
1559
      *     dispatch (both last slice & 2nd-to-last slice);
1560
      *   else if (slice > 0)
1561
      *     dispatch (slice - 1);
1562
      */
1563

    
1564
    // now that we've filtered the last rows, they're safe to display
1565
    if (slice)
1566
        vp3_draw_horiz_band(s, 16*slice);
1567
}
1568

    
1569
/*
1570
 * This is the ffmpeg/libavcodec API init function.
1571
 */
1572
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1573
{
1574
    Vp3DecodeContext *s = avctx->priv_data;
1575
    int i, inter, plane;
1576
    int c_width;
1577
    int c_height;
1578
    int y_superblock_count;
1579
    int c_superblock_count;
1580

    
1581
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1582
        s->version = 0;
1583
    else
1584
        s->version = 1;
1585

    
1586
    s->avctx = avctx;
1587
    s->width = FFALIGN(avctx->width, 16);
1588
    s->height = FFALIGN(avctx->height, 16);
1589
    avctx->pix_fmt = PIX_FMT_YUV420P;
1590
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1591
    if(avctx->idct_algo==FF_IDCT_AUTO)
1592
        avctx->idct_algo=FF_IDCT_VP3;
1593
    dsputil_init(&s->dsp, avctx);
1594

    
1595
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1596

    
1597
    /* initialize to an impossible value which will force a recalculation
1598
     * in the first frame decode */
1599
    for (i = 0; i < 3; i++)
1600
        s->qps[i] = -1;
1601

    
1602
    s->y_superblock_width = (s->width + 31) / 32;
1603
    s->y_superblock_height = (s->height + 31) / 32;
1604
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1605

    
1606
    /* work out the dimensions for the C planes */
1607
    c_width = s->width / 2;
1608
    c_height = s->height / 2;
1609
    s->c_superblock_width = (c_width + 31) / 32;
1610
    s->c_superblock_height = (c_height + 31) / 32;
1611
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1612

    
1613
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1614
    s->u_superblock_start = y_superblock_count;
1615
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1616
    s->superblock_coding = av_malloc(s->superblock_count);
1617

    
1618
    s->macroblock_width = (s->width + 15) / 16;
1619
    s->macroblock_height = (s->height + 15) / 16;
1620
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1621

    
1622
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1623
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1624

    
1625
    /* fragment count covers all 8x8 blocks for all 3 planes */
1626
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1627
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1628
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1629

    
1630
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1631
    s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1632
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1633
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1634
    s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1635
    if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1636
        !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1637
        vp3_decode_end(avctx);
1638
        return -1;
1639
    }
1640

    
1641
    if (!s->theora_tables)
1642
    {
1643
        for (i = 0; i < 64; i++) {
1644
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1645
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1646
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1647
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1648
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1649
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1650
        }
1651

    
1652
        for(inter=0; inter<2; inter++){
1653
            for(plane=0; plane<3; plane++){
1654
                s->qr_count[inter][plane]= 1;
1655
                s->qr_size [inter][plane][0]= 63;
1656
                s->qr_base [inter][plane][0]=
1657
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1658
            }
1659
        }
1660

    
1661
        /* init VLC tables */
1662
        for (i = 0; i < 16; i++) {
1663

    
1664
            /* DC histograms */
1665
            init_vlc(&s->dc_vlc[i], 5, 32,
1666
                &dc_bias[i][0][1], 4, 2,
1667
                &dc_bias[i][0][0], 4, 2, 0);
1668

    
1669
            /* group 1 AC histograms */
1670
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1671
                &ac_bias_0[i][0][1], 4, 2,
1672
                &ac_bias_0[i][0][0], 4, 2, 0);
1673

    
1674
            /* group 2 AC histograms */
1675
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1676
                &ac_bias_1[i][0][1], 4, 2,
1677
                &ac_bias_1[i][0][0], 4, 2, 0);
1678

    
1679
            /* group 3 AC histograms */
1680
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1681
                &ac_bias_2[i][0][1], 4, 2,
1682
                &ac_bias_2[i][0][0], 4, 2, 0);
1683

    
1684
            /* group 4 AC histograms */
1685
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1686
                &ac_bias_3[i][0][1], 4, 2,
1687
                &ac_bias_3[i][0][0], 4, 2, 0);
1688
        }
1689
    } else {
1690
        for (i = 0; i < 16; i++) {
1691

    
1692
            /* DC histograms */
1693
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1694
                &s->huffman_table[i][0][1], 4, 2,
1695
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1696
                goto vlc_fail;
1697

    
1698
            /* group 1 AC histograms */
1699
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1700
                &s->huffman_table[i+16][0][1], 4, 2,
1701
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1702
                goto vlc_fail;
1703

    
1704
            /* group 2 AC histograms */
1705
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1706
                &s->huffman_table[i+16*2][0][1], 4, 2,
1707
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1708
                goto vlc_fail;
1709

    
1710
            /* group 3 AC histograms */
1711
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1712
                &s->huffman_table[i+16*3][0][1], 4, 2,
1713
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1714
                goto vlc_fail;
1715

    
1716
            /* group 4 AC histograms */
1717
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1718
                &s->huffman_table[i+16*4][0][1], 4, 2,
1719
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1720
                goto vlc_fail;
1721
        }
1722
    }
1723

    
1724
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1725
        &superblock_run_length_vlc_table[0][1], 4, 2,
1726
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1727

    
1728
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1729
        &fragment_run_length_vlc_table[0][1], 4, 2,
1730
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1731

    
1732
    init_vlc(&s->mode_code_vlc, 3, 8,
1733
        &mode_code_vlc_table[0][1], 2, 1,
1734
        &mode_code_vlc_table[0][0], 2, 1, 0);
1735

    
1736
    init_vlc(&s->motion_vector_vlc, 6, 63,
1737
        &motion_vector_vlc_table[0][1], 2, 1,
1738
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1739

    
1740
    /* work out the block mapping tables */
1741
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1742
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1743
    if (!s->superblock_fragments || !s->macroblock_coding) {
1744
        vp3_decode_end(avctx);
1745
        return -1;
1746
    }
1747
    init_block_mapping(s);
1748

    
1749
    for (i = 0; i < 3; i++) {
1750
        s->current_frame.data[i] = NULL;
1751
        s->last_frame.data[i] = NULL;
1752
        s->golden_frame.data[i] = NULL;
1753
    }
1754

    
1755
    return 0;
1756

    
1757
vlc_fail:
1758
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1759
    return -1;
1760
}
1761

    
1762
/*
1763
 * This is the ffmpeg/libavcodec API frame decode function.
1764
 */
1765
static int vp3_decode_frame(AVCodecContext *avctx,
1766
                            void *data, int *data_size,
1767
                            AVPacket *avpkt)
1768
{
1769
    const uint8_t *buf = avpkt->data;
1770
    int buf_size = avpkt->size;
1771
    Vp3DecodeContext *s = avctx->priv_data;
1772
    GetBitContext gb;
1773
    static int counter = 0;
1774
    int i;
1775

    
1776
    init_get_bits(&gb, buf, buf_size * 8);
1777

    
1778
    if (s->theora && get_bits1(&gb))
1779
    {
1780
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1781
        return -1;
1782
    }
1783

    
1784
    s->keyframe = !get_bits1(&gb);
1785
    if (!s->theora)
1786
        skip_bits(&gb, 1);
1787
    for (i = 0; i < 3; i++)
1788
        s->last_qps[i] = s->qps[i];
1789

    
1790
    s->nqps=0;
1791
    do{
1792
        s->qps[s->nqps++]= get_bits(&gb, 6);
1793
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1794
    for (i = s->nqps; i < 3; i++)
1795
        s->qps[i] = -1;
1796

    
1797
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1798
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1799
            s->keyframe?"key":"", counter, s->qps[0]);
1800
    counter++;
1801

    
1802
    if (s->qps[0] != s->last_qps[0])
1803
        init_loop_filter(s);
1804

    
1805
    for (i = 0; i < s->nqps; i++)
1806
        // reinit all dequantizers if the first one changed, because
1807
        // the DC of the first quantizer must be used for all matrices
1808
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1809
            init_dequantizer(s, i);
1810

    
1811
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1812
        return buf_size;
1813

    
1814
    if (s->keyframe) {
1815
        if (!s->theora)
1816
        {
1817
            skip_bits(&gb, 4); /* width code */
1818
            skip_bits(&gb, 4); /* height code */
1819
            if (s->version)
1820
            {
1821
                s->version = get_bits(&gb, 5);
1822
                if (counter == 1)
1823
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1824
            }
1825
        }
1826
        if (s->version || s->theora)
1827
        {
1828
                if (get_bits1(&gb))
1829
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1830
            skip_bits(&gb, 2); /* reserved? */
1831
        }
1832

    
1833
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1834
            if (s->golden_frame.data[0])
1835
                avctx->release_buffer(avctx, &s->golden_frame);
1836
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1837
        } else {
1838
            if (s->golden_frame.data[0])
1839
                avctx->release_buffer(avctx, &s->golden_frame);
1840
            if (s->last_frame.data[0])
1841
                avctx->release_buffer(avctx, &s->last_frame);
1842
        }
1843

    
1844
        s->golden_frame.reference = 3;
1845
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1846
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1847
            return -1;
1848
        }
1849

    
1850
        /* golden frame is also the current frame */
1851
        s->current_frame= s->golden_frame;
1852
    } else {
1853
        /* allocate a new current frame */
1854
        s->current_frame.reference = 3;
1855
        if (!s->golden_frame.data[0]) {
1856
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1857
            return -1;
1858
        }
1859
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1860
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1861
            return -1;
1862
        }
1863
    }
1864

    
1865
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1866
    s->current_frame.qstride= 0;
1867

    
1868
    init_frame(s, &gb);
1869

    
1870
    if (unpack_superblocks(s, &gb)){
1871
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1872
        return -1;
1873
    }
1874
    if (unpack_modes(s, &gb)){
1875
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1876
        return -1;
1877
    }
1878
    if (unpack_vectors(s, &gb)){
1879
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1880
        return -1;
1881
    }
1882
    if (unpack_block_qpis(s, &gb)){
1883
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1884
        return -1;
1885
    }
1886
    if (unpack_dct_coeffs(s, &gb)){
1887
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1888
        return -1;
1889
    }
1890

    
1891
    for (i = 0; i < 3; i++) {
1892
        if (s->flipped_image)
1893
            s->data_offset[i] = 0;
1894
        else
1895
            s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1896
    }
1897

    
1898
    s->last_slice_end = 0;
1899
    for (i = 0; i < s->macroblock_height; i++)
1900
        render_slice(s, i);
1901

    
1902
    // filter the last row
1903
    for (i = 0; i < 3; i++) {
1904
        int row = (s->height >> (3+!!i)) - 1;
1905
        apply_loop_filter(s, i, row, row+1);
1906
    }
1907
    vp3_draw_horiz_band(s, s->height);
1908

    
1909
    *data_size=sizeof(AVFrame);
1910
    *(AVFrame*)data= s->current_frame;
1911

    
1912
    /* release the last frame, if it is allocated and if it is not the
1913
     * golden frame */
1914
    if ((s->last_frame.data[0]) &&
1915
        (s->last_frame.data[0] != s->golden_frame.data[0]))
1916
        avctx->release_buffer(avctx, &s->last_frame);
1917

    
1918
    /* shuffle frames (last = current) */
1919
    s->last_frame= s->current_frame;
1920
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1921

    
1922
    return buf_size;
1923
}
1924

    
1925
/*
1926
 * This is the ffmpeg/libavcodec API module cleanup function.
1927
 */
1928
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1929
{
1930
    Vp3DecodeContext *s = avctx->priv_data;
1931
    int i;
1932

    
1933
    av_free(s->superblock_coding);
1934
    av_free(s->all_fragments);
1935
    av_free(s->coeff_counts);
1936
    av_free(s->coeffs);
1937
    av_free(s->coded_fragment_list);
1938
    av_free(s->fast_fragment_list);
1939
    av_free(s->superblock_fragments);
1940
    av_free(s->macroblock_coding);
1941

    
1942
    for (i = 0; i < 16; i++) {
1943
        free_vlc(&s->dc_vlc[i]);
1944
        free_vlc(&s->ac_vlc_1[i]);
1945
        free_vlc(&s->ac_vlc_2[i]);
1946
        free_vlc(&s->ac_vlc_3[i]);
1947
        free_vlc(&s->ac_vlc_4[i]);
1948
    }
1949

    
1950
    free_vlc(&s->superblock_run_length_vlc);
1951
    free_vlc(&s->fragment_run_length_vlc);
1952
    free_vlc(&s->mode_code_vlc);
1953
    free_vlc(&s->motion_vector_vlc);
1954

    
1955
    /* release all frames */
1956
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1957
        avctx->release_buffer(avctx, &s->golden_frame);
1958
    if (s->last_frame.data[0])
1959
        avctx->release_buffer(avctx, &s->last_frame);
1960
    /* no need to release the current_frame since it will always be pointing
1961
     * to the same frame as either the golden or last frame */
1962

    
1963
    return 0;
1964
}
1965

    
1966
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1967
{
1968
    Vp3DecodeContext *s = avctx->priv_data;
1969

    
1970
    if (get_bits1(gb)) {
1971
        int token;
1972
        if (s->entries >= 32) { /* overflow */
1973
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1974
            return -1;
1975
        }
1976
        token = get_bits(gb, 5);
1977
        //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);
1978
        s->huffman_table[s->hti][token][0] = s->hbits;
1979
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1980
        s->entries++;
1981
    }
1982
    else {
1983
        if (s->huff_code_size >= 32) {/* overflow */
1984
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1985
            return -1;
1986
        }
1987
        s->huff_code_size++;
1988
        s->hbits <<= 1;
1989
        if (read_huffman_tree(avctx, gb))
1990
            return -1;
1991
        s->hbits |= 1;
1992
        if (read_huffman_tree(avctx, gb))
1993
            return -1;
1994
        s->hbits >>= 1;
1995
        s->huff_code_size--;
1996
    }
1997
    return 0;
1998
}
1999

    
2000
#if CONFIG_THEORA_DECODER
2001
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2002
{
2003
    Vp3DecodeContext *s = avctx->priv_data;
2004
    int visible_width, visible_height, colorspace;
2005

    
2006
    s->theora = get_bits_long(gb, 24);
2007
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2008

    
2009
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2010
    /* but previous versions have the image flipped relative to vp3 */
2011
    if (s->theora < 0x030200)
2012
    {
2013
        s->flipped_image = 1;
2014
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2015
    }
2016

    
2017
    visible_width  = s->width  = get_bits(gb, 16) << 4;
2018
    visible_height = s->height = get_bits(gb, 16) << 4;
2019

    
2020
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2021
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2022
        s->width= s->height= 0;
2023
        return -1;
2024
    }
2025

    
2026
    if (s->theora >= 0x030200) {
2027
        visible_width  = get_bits_long(gb, 24);
2028
        visible_height = get_bits_long(gb, 24);
2029

    
2030
        skip_bits(gb, 8); /* offset x */
2031
        skip_bits(gb, 8); /* offset y */
2032
    }
2033

    
2034
    skip_bits(gb, 32); /* fps numerator */
2035
    skip_bits(gb, 32); /* fps denumerator */
2036
    skip_bits(gb, 24); /* aspect numerator */
2037
    skip_bits(gb, 24); /* aspect denumerator */
2038

    
2039
    if (s->theora < 0x030200)
2040
        skip_bits(gb, 5); /* keyframe frequency force */
2041
    colorspace = get_bits(gb, 8);
2042
    skip_bits(gb, 24); /* bitrate */
2043

    
2044
    skip_bits(gb, 6); /* quality hint */
2045

    
2046
    if (s->theora >= 0x030200)
2047
    {
2048
        skip_bits(gb, 5); /* keyframe frequency force */
2049
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2050
        skip_bits(gb, 3); /* reserved */
2051
    }
2052

    
2053
//    align_get_bits(gb);
2054

    
2055
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2056
        && visible_height <= s->height && visible_height > s->height-16)
2057
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2058
    else
2059
        avcodec_set_dimensions(avctx, s->width, s->height);
2060

    
2061
    if (colorspace == 1) {
2062
        avctx->color_primaries = AVCOL_PRI_BT470M;
2063
    } else if (colorspace == 2) {
2064
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2065
    }
2066
    if (colorspace == 1 || colorspace == 2) {
2067
        avctx->colorspace = AVCOL_SPC_BT470BG;
2068
        avctx->color_trc  = AVCOL_TRC_BT709;
2069
    }
2070

    
2071
    return 0;
2072
}
2073

    
2074
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2075
{
2076
    Vp3DecodeContext *s = avctx->priv_data;
2077
    int i, n, matrices, inter, plane;
2078

    
2079
    if (s->theora >= 0x030200) {
2080
        n = get_bits(gb, 3);
2081
        /* loop filter limit values table */
2082
        for (i = 0; i < 64; i++) {
2083
            s->filter_limit_values[i] = get_bits(gb, n);
2084
            if (s->filter_limit_values[i] > 127) {
2085
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2086
                s->filter_limit_values[i] = 127;
2087
            }
2088
        }
2089
    }
2090

    
2091
    if (s->theora >= 0x030200)
2092
        n = get_bits(gb, 4) + 1;
2093
    else
2094
        n = 16;
2095
    /* quality threshold table */
2096
    for (i = 0; i < 64; i++)
2097
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2098

    
2099
    if (s->theora >= 0x030200)
2100
        n = get_bits(gb, 4) + 1;
2101
    else
2102
        n = 16;
2103
    /* dc scale factor table */
2104
    for (i = 0; i < 64; i++)
2105
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2106

    
2107
    if (s->theora >= 0x030200)
2108
        matrices = get_bits(gb, 9) + 1;
2109
    else
2110
        matrices = 3;
2111

    
2112
    if(matrices > 384){
2113
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2114
        return -1;
2115
    }
2116

    
2117
    for(n=0; n<matrices; n++){
2118
        for (i = 0; i < 64; i++)
2119
            s->base_matrix[n][i]= get_bits(gb, 8);
2120
    }
2121

    
2122
    for (inter = 0; inter <= 1; inter++) {
2123
        for (plane = 0; plane <= 2; plane++) {
2124
            int newqr= 1;
2125
            if (inter || plane > 0)
2126
                newqr = get_bits1(gb);
2127
            if (!newqr) {
2128
                int qtj, plj;
2129
                if(inter && get_bits1(gb)){
2130
                    qtj = 0;
2131
                    plj = plane;
2132
                }else{
2133
                    qtj= (3*inter + plane - 1) / 3;
2134
                    plj= (plane + 2) % 3;
2135
                }
2136
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2137
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2138
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2139
            } else {
2140
                int qri= 0;
2141
                int qi = 0;
2142

    
2143
                for(;;){
2144
                    i= get_bits(gb, av_log2(matrices-1)+1);
2145
                    if(i>= matrices){
2146
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2147
                        return -1;
2148
                    }
2149
                    s->qr_base[inter][plane][qri]= i;
2150
                    if(qi >= 63)
2151
                        break;
2152
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2153
                    s->qr_size[inter][plane][qri++]= i;
2154
                    qi += i;
2155
                }
2156

    
2157
                if (qi > 63) {
2158
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2159
                    return -1;
2160
                }
2161
                s->qr_count[inter][plane]= qri;
2162
            }
2163
        }
2164
    }
2165

    
2166
    /* Huffman tables */
2167
    for (s->hti = 0; s->hti < 80; s->hti++) {
2168
        s->entries = 0;
2169
        s->huff_code_size = 1;
2170
        if (!get_bits1(gb)) {
2171
            s->hbits = 0;
2172
            if(read_huffman_tree(avctx, gb))
2173
                return -1;
2174
            s->hbits = 1;
2175
            if(read_huffman_tree(avctx, gb))
2176
                return -1;
2177
        }
2178
    }
2179

    
2180
    s->theora_tables = 1;
2181

    
2182
    return 0;
2183
}
2184

    
2185
static av_cold int theora_decode_init(AVCodecContext *avctx)
2186
{
2187
    Vp3DecodeContext *s = avctx->priv_data;
2188
    GetBitContext gb;
2189
    int ptype;
2190
    uint8_t *header_start[3];
2191
    int header_len[3];
2192
    int i;
2193

    
2194
    s->theora = 1;
2195

    
2196
    if (!avctx->extradata_size)
2197
    {
2198
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2199
        return -1;
2200
    }
2201

    
2202
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2203
                              42, header_start, header_len) < 0) {
2204
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2205
        return -1;
2206
    }
2207

    
2208
  for(i=0;i<3;i++) {
2209
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2210

    
2211
    ptype = get_bits(&gb, 8);
2212

    
2213
     if (!(ptype & 0x80))
2214
     {
2215
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2216
//        return -1;
2217
     }
2218

    
2219
    // FIXME: Check for this as well.
2220
    skip_bits_long(&gb, 6*8); /* "theora" */
2221

    
2222
    switch(ptype)
2223
    {
2224
        case 0x80:
2225
            theora_decode_header(avctx, &gb);
2226
                break;
2227
        case 0x81:
2228
// FIXME: is this needed? it breaks sometimes
2229
//            theora_decode_comments(avctx, gb);
2230
            break;
2231
        case 0x82:
2232
            if (theora_decode_tables(avctx, &gb))
2233
                return -1;
2234
            break;
2235
        default:
2236
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2237
            break;
2238
    }
2239
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2240
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2241
    if (s->theora < 0x030200)
2242
        break;
2243
  }
2244

    
2245
    return vp3_decode_init(avctx);
2246
}
2247

    
2248
AVCodec theora_decoder = {
2249
    "theora",
2250
    CODEC_TYPE_VIDEO,
2251
    CODEC_ID_THEORA,
2252
    sizeof(Vp3DecodeContext),
2253
    theora_decode_init,
2254
    NULL,
2255
    vp3_decode_end,
2256
    vp3_decode_frame,
2257
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2258
    NULL,
2259
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2260
};
2261
#endif
2262

    
2263
AVCodec vp3_decoder = {
2264
    "vp3",
2265
    CODEC_TYPE_VIDEO,
2266
    CODEC_ID_VP3,
2267
    sizeof(Vp3DecodeContext),
2268
    vp3_decode_init,
2269
    NULL,
2270
    vp3_decode_end,
2271
    vp3_decode_frame,
2272
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2273
    NULL,
2274
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2275
};