Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 161e8cf4

History | View | Annotate | Download (77.2 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
    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
#define MODE_INTER_NO_MV      0
67
#define MODE_INTRA            1
68
#define MODE_INTER_PLUS_MV    2
69
#define MODE_INTER_LAST_MV    3
70
#define MODE_INTER_PRIOR_LAST 4
71
#define MODE_USING_GOLDEN     5
72
#define MODE_GOLDEN_MV        6
73
#define MODE_INTER_FOURMV     7
74
#define CODING_MODE_COUNT     8
75

    
76
/* special internal mode */
77
#define MODE_COPY             8
78

    
79
/* There are 6 preset schemes, plus a free-form scheme */
80
static const int ModeAlphabet[6][CODING_MODE_COUNT] =
81
{
82
    /* scheme 1: Last motion vector dominates */
83
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
84
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
85
         MODE_INTRA,            MODE_USING_GOLDEN,
86
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
87

    
88
    /* scheme 2 */
89
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
90
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
91
         MODE_INTRA,            MODE_USING_GOLDEN,
92
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
93

    
94
    /* scheme 3 */
95
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
96
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
97
         MODE_INTRA,            MODE_USING_GOLDEN,
98
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
99

    
100
    /* scheme 4 */
101
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
102
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
103
         MODE_INTRA,            MODE_USING_GOLDEN,
104
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
105

    
106
    /* scheme 5: No motion vector dominates */
107
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
108
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
109
         MODE_INTRA,            MODE_USING_GOLDEN,
110
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
111

    
112
    /* scheme 6 */
113
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
114
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
115
         MODE_INTER_PLUS_MV,    MODE_INTRA,
116
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
117

    
118
};
119

    
120
#define MIN_DEQUANT_VAL 2
121

    
122
typedef struct Vp3DecodeContext {
123
    AVCodecContext *avctx;
124
    int theora, theora_tables;
125
    int version;
126
    int width, height;
127
    AVFrame golden_frame;
128
    AVFrame last_frame;
129
    AVFrame current_frame;
130
    int keyframe;
131
    DSPContext dsp;
132
    int flipped_image;
133
    int last_slice_end;
134

    
135
    int qps[3];
136
    int nqps;
137
    int last_qps[3];
138

    
139
    int superblock_count;
140
    int y_superblock_width;
141
    int y_superblock_height;
142
    int c_superblock_width;
143
    int c_superblock_height;
144
    int u_superblock_start;
145
    int v_superblock_start;
146
    unsigned char *superblock_coding;
147

    
148
    int macroblock_count;
149
    int macroblock_width;
150
    int macroblock_height;
151

    
152
    int fragment_count;
153
    int fragment_width;
154
    int fragment_height;
155

    
156
    Vp3Fragment *all_fragments;
157
    uint8_t *coeff_counts;
158
    Coeff *coeffs;
159
    Coeff *next_coeff;
160
    int fragment_start[3];
161
    int data_offset[3];
162

    
163
    ScanTable scantable;
164

    
165
    /* tables */
166
    uint16_t coded_dc_scale_factor[64];
167
    uint32_t coded_ac_scale_factor[64];
168
    uint8_t base_matrix[384][64];
169
    uint8_t qr_count[2][3];
170
    uint8_t qr_size [2][3][64];
171
    uint16_t qr_base[2][3][64];
172

    
173
    /* this is a list of indexes into the all_fragments array indicating
174
     * which of the fragments are coded */
175
    int *coded_fragment_list;
176
    int coded_fragment_list_index;
177

    
178
    /* track which fragments have already been decoded; called 'fast'
179
     * because this data structure avoids having to iterate through every
180
     * fragment in coded_fragment_list; once a fragment has been fully
181
     * decoded, it is removed from this list */
182
    int *fast_fragment_list;
183
    int fragment_list_y_head;
184
    int fragment_list_c_head;
185

    
186
    VLC dc_vlc[16];
187
    VLC ac_vlc_1[16];
188
    VLC ac_vlc_2[16];
189
    VLC ac_vlc_3[16];
190
    VLC ac_vlc_4[16];
191

    
192
    VLC superblock_run_length_vlc;
193
    VLC fragment_run_length_vlc;
194
    VLC mode_code_vlc;
195
    VLC motion_vector_vlc;
196

    
197
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
198
     * index into them */
199
    DECLARE_ALIGNED_16(int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
200

    
201
    /* This table contains superblock_count * 16 entries. Each set of 16
202
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
203
     * An entry will be -1 to indicate that no entry corresponds to that
204
     * index. */
205
    int *superblock_fragments;
206

    
207
    /* This is an array that indicates how a particular macroblock
208
     * is coded. */
209
    unsigned char *macroblock_coding;
210

    
211
    int first_coded_y_fragment;
212
    int first_coded_c_fragment;
213
    int last_coded_y_fragment;
214
    int last_coded_c_fragment;
215

    
216
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
217
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
218

    
219
    /* Huffman decode */
220
    int hti;
221
    unsigned int hbits;
222
    int entries;
223
    int huff_code_size;
224
    uint16_t huffman_table[80][32][2];
225

    
226
    uint8_t filter_limit_values[64];
227
    DECLARE_ALIGNED_8(int, bounding_values_array)[256+2];
228
} Vp3DecodeContext;
229

    
230
/************************************************************************
231
 * VP3 specific functions
232
 ************************************************************************/
233

    
234
/*
235
 * This function sets up all of the various blocks mappings:
236
 * superblocks <-> fragments, macroblocks <-> fragments,
237
 * superblocks <-> macroblocks
238
 *
239
 * Returns 0 is successful; returns 1 if *anything* went wrong.
240
 */
241
static int init_block_mapping(Vp3DecodeContext *s)
242
{
243
    int i, j;
244
    signed int hilbert_walk_mb[4];
245

    
246
    int current_fragment = 0;
247
    int current_width = 0;
248
    int current_height = 0;
249
    int right_edge = 0;
250
    int bottom_edge = 0;
251
    int superblock_row_inc = 0;
252
    int mapping_index = 0;
253

    
254
    int current_macroblock;
255
    int c_fragment;
256

    
257
    static const signed char travel_width[16] = {
258
         1,  1,  0, -1,
259
         0,  0,  1,  0,
260
         1,  0,  1,  0,
261
         0, -1,  0,  1
262
    };
263

    
264
    static const signed char travel_height[16] = {
265
         0,  0,  1,  0,
266
         1,  1,  0, -1,
267
         0,  1,  0, -1,
268
        -1,  0, -1,  0
269
    };
270

    
271
    hilbert_walk_mb[0] = 1;
272
    hilbert_walk_mb[1] = s->macroblock_width;
273
    hilbert_walk_mb[2] = 1;
274
    hilbert_walk_mb[3] = -s->macroblock_width;
275

    
276
    /* iterate through each superblock (all planes) and map the fragments */
277
    for (i = 0; i < s->superblock_count; i++) {
278
        /* time to re-assign the limits? */
279
        if (i == 0) {
280

    
281
            /* start of Y superblocks */
282
            right_edge = s->fragment_width;
283
            bottom_edge = s->fragment_height;
284
            current_width = -1;
285
            current_height = 0;
286
            superblock_row_inc = 3 * s->fragment_width -
287
                (s->y_superblock_width * 4 - s->fragment_width);
288

    
289
            /* the first operation for this variable is to advance by 1 */
290
            current_fragment = -1;
291

    
292
        } else if (i == s->u_superblock_start) {
293

    
294
            /* start of U superblocks */
295
            right_edge = s->fragment_width / 2;
296
            bottom_edge = s->fragment_height / 2;
297
            current_width = -1;
298
            current_height = 0;
299
            superblock_row_inc = 3 * (s->fragment_width / 2) -
300
                (s->c_superblock_width * 4 - s->fragment_width / 2);
301

    
302
            /* the first operation for this variable is to advance by 1 */
303
            current_fragment = s->fragment_start[1] - 1;
304

    
305
        } else if (i == s->v_superblock_start) {
306

    
307
            /* start of V superblocks */
308
            right_edge = s->fragment_width / 2;
309
            bottom_edge = s->fragment_height / 2;
310
            current_width = -1;
311
            current_height = 0;
312
            superblock_row_inc = 3 * (s->fragment_width / 2) -
313
                (s->c_superblock_width * 4 - s->fragment_width / 2);
314

    
315
            /* the first operation for this variable is to advance by 1 */
316
            current_fragment = s->fragment_start[2] - 1;
317

    
318
        }
319

    
320
        if (current_width >= right_edge - 1) {
321
            /* reset width and move to next superblock row */
322
            current_width = -1;
323
            current_height += 4;
324

    
325
            /* fragment is now at the start of a new superblock row */
326
            current_fragment += superblock_row_inc;
327
        }
328

    
329
        /* iterate through all 16 fragments in a superblock */
330
        for (j = 0; j < 16; j++) {
331
            current_fragment += travel_width[j] + right_edge * travel_height[j];
332
            current_width += travel_width[j];
333
            current_height += travel_height[j];
334

    
335
            /* check if the fragment is in bounds */
336
            if ((current_width < right_edge) &&
337
                (current_height < bottom_edge)) {
338
                s->superblock_fragments[mapping_index] = current_fragment;
339
            } else {
340
                s->superblock_fragments[mapping_index] = -1;
341
            }
342

    
343
            mapping_index++;
344
        }
345
    }
346

    
347
    return 0;  /* successful path out */
348
}
349

    
350
/*
351
 * This function wipes out all of the fragment data.
352
 */
353
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
354
{
355
    int i;
356

    
357
    /* zero out all of the fragment information */
358
    s->coded_fragment_list_index = 0;
359
    for (i = 0; i < s->fragment_count; i++) {
360
        s->coeff_counts[i] = 0;
361
        s->all_fragments[i].motion_x = 127;
362
        s->all_fragments[i].motion_y = 127;
363
        s->all_fragments[i].next_coeff= NULL;
364
        s->all_fragments[i].qpi = 0;
365
        s->coeffs[i].index=
366
        s->coeffs[i].coeff=0;
367
        s->coeffs[i].next= NULL;
368
    }
369
}
370

    
371
/*
372
 * This function sets up the dequantization tables used for a particular
373
 * frame.
374
 */
375
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
376
{
377
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
378
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
379
    int i, plane, inter, qri, bmi, bmj, qistart;
380

    
381
    for(inter=0; inter<2; inter++){
382
        for(plane=0; plane<3; plane++){
383
            int sum=0;
384
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
385
                sum+= s->qr_size[inter][plane][qri];
386
                if(s->qps[qpi] <= sum)
387
                    break;
388
            }
389
            qistart= sum - s->qr_size[inter][plane][qri];
390
            bmi= s->qr_base[inter][plane][qri  ];
391
            bmj= s->qr_base[inter][plane][qri+1];
392
            for(i=0; i<64; i++){
393
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
394
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
395
                            + s->qr_size[inter][plane][qri])
396
                           / (2*s->qr_size[inter][plane][qri]);
397

    
398
                int qmin= 8<<(inter + !i);
399
                int qscale= i ? ac_scale_factor : dc_scale_factor;
400

    
401
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
402
            }
403
            // all DC coefficients use the same quant so as not to interfere with DC prediction
404
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
405
        }
406
    }
407

    
408
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
409
}
410

    
411
/*
412
 * This function initializes the loop filter boundary limits if the frame's
413
 * quality index is different from the previous frame's.
414
 *
415
 * The filter_limit_values may not be larger than 127.
416
 */
417
static void init_loop_filter(Vp3DecodeContext *s)
418
{
419
    int *bounding_values= s->bounding_values_array+127;
420
    int filter_limit;
421
    int x;
422
    int value;
423

    
424
    filter_limit = s->filter_limit_values[s->qps[0]];
425

    
426
    /* set up the bounding values */
427
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
428
    for (x = 0; x < filter_limit; x++) {
429
        bounding_values[-x] = -x;
430
        bounding_values[x] = x;
431
    }
432
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
433
        bounding_values[ x] =  value;
434
        bounding_values[-x] = -value;
435
    }
436
    if (value)
437
        bounding_values[128] = value;
438
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
439
}
440

    
441
/*
442
 * This function unpacks all of the superblock/macroblock/fragment coding
443
 * information from the bitstream.
444
 */
445
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
446
{
447
    int bit = 0;
448
    int current_superblock = 0;
449
    int current_run = 0;
450
    int decode_fully_flags = 0;
451
    int decode_partial_blocks = 0;
452
    int first_c_fragment_seen;
453

    
454
    int i, j;
455
    int current_fragment;
456

    
457
    if (s->keyframe) {
458
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
459

    
460
    } else {
461

    
462
        /* unpack the list of partially-coded superblocks */
463
        bit = get_bits1(gb);
464
        /* toggle the bit because as soon as the first run length is
465
         * fetched the bit will be toggled again */
466
        bit ^= 1;
467
        while (current_superblock < s->superblock_count) {
468
            if (current_run-- == 0) {
469
                bit ^= 1;
470
                current_run = get_vlc2(gb,
471
                    s->superblock_run_length_vlc.table, 6, 2);
472
                if (current_run == 33)
473
                    current_run += get_bits(gb, 12);
474

    
475
                /* if any of the superblocks are not partially coded, flag
476
                 * a boolean to decode the list of fully-coded superblocks */
477
                if (bit == 0) {
478
                    decode_fully_flags = 1;
479
                } else {
480

    
481
                    /* make a note of the fact that there are partially coded
482
                     * superblocks */
483
                    decode_partial_blocks = 1;
484
                }
485
            }
486
            s->superblock_coding[current_superblock++] = bit;
487
        }
488

    
489
        /* unpack the list of fully coded superblocks if any of the blocks were
490
         * not marked as partially coded in the previous step */
491
        if (decode_fully_flags) {
492

    
493
            current_superblock = 0;
494
            current_run = 0;
495
            bit = get_bits1(gb);
496
            /* toggle the bit because as soon as the first run length is
497
             * fetched the bit will be toggled again */
498
            bit ^= 1;
499
            while (current_superblock < s->superblock_count) {
500

    
501
                /* skip any superblocks already marked as partially coded */
502
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
503

    
504
                    if (current_run-- == 0) {
505
                        bit ^= 1;
506
                        current_run = get_vlc2(gb,
507
                            s->superblock_run_length_vlc.table, 6, 2);
508
                        if (current_run == 33)
509
                            current_run += get_bits(gb, 12);
510
                    }
511
                    s->superblock_coding[current_superblock] = 2*bit;
512
                }
513
                current_superblock++;
514
            }
515
        }
516

    
517
        /* if there were partial blocks, initialize bitstream for
518
         * unpacking fragment codings */
519
        if (decode_partial_blocks) {
520

    
521
            current_run = 0;
522
            bit = get_bits1(gb);
523
            /* toggle the bit because as soon as the first run length is
524
             * fetched the bit will be toggled again */
525
            bit ^= 1;
526
        }
527
    }
528

    
529
    /* figure out which fragments are coded; iterate through each
530
     * superblock (all planes) */
531
    s->coded_fragment_list_index = 0;
532
    s->next_coeff= s->coeffs + s->fragment_count;
533
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
534
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
535
    first_c_fragment_seen = 0;
536
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
537
    for (i = 0; i < s->superblock_count; i++) {
538

    
539
        /* iterate through all 16 fragments in a superblock */
540
        for (j = 0; j < 16; j++) {
541

    
542
            /* if the fragment is in bounds, check its coding status */
543
            current_fragment = s->superblock_fragments[i * 16 + j];
544
            if (current_fragment >= s->fragment_count) {
545
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
546
                    current_fragment, s->fragment_count);
547
                return 1;
548
            }
549
            if (current_fragment != -1) {
550
                if (s->superblock_coding[i] == SB_NOT_CODED) {
551

    
552
                    /* copy all the fragments from the prior frame */
553
                    s->all_fragments[current_fragment].coding_method =
554
                        MODE_COPY;
555

    
556
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
557

    
558
                    /* fragment may or may not be coded; this is the case
559
                     * that cares about the fragment coding runs */
560
                    if (current_run-- == 0) {
561
                        bit ^= 1;
562
                        current_run = get_vlc2(gb,
563
                            s->fragment_run_length_vlc.table, 5, 2);
564
                    }
565

    
566
                    if (bit) {
567
                        /* default mode; actual mode will be decoded in
568
                         * the next phase */
569
                        s->all_fragments[current_fragment].coding_method =
570
                            MODE_INTER_NO_MV;
571
                        s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
572
                        s->coded_fragment_list[s->coded_fragment_list_index] =
573
                            current_fragment;
574
                        if ((current_fragment >= s->fragment_start[1]) &&
575
                            (s->last_coded_y_fragment == -1) &&
576
                            (!first_c_fragment_seen)) {
577
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
578
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
579
                            first_c_fragment_seen = 1;
580
                        }
581
                        s->coded_fragment_list_index++;
582
                    } else {
583
                        /* not coded; copy this fragment from the prior frame */
584
                        s->all_fragments[current_fragment].coding_method =
585
                            MODE_COPY;
586
                    }
587

    
588
                } else {
589

    
590
                    /* fragments are fully coded in this superblock; actual
591
                     * coding will be determined in next step */
592
                    s->all_fragments[current_fragment].coding_method =
593
                        MODE_INTER_NO_MV;
594
                    s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
595
                    s->coded_fragment_list[s->coded_fragment_list_index] =
596
                        current_fragment;
597
                    if ((current_fragment >= s->fragment_start[1]) &&
598
                        (s->last_coded_y_fragment == -1) &&
599
                        (!first_c_fragment_seen)) {
600
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
601
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
602
                        first_c_fragment_seen = 1;
603
                    }
604
                    s->coded_fragment_list_index++;
605
                }
606
            }
607
        }
608
    }
609

    
610
    if (!first_c_fragment_seen)
611
        /* only Y fragments coded in this frame */
612
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
613
    else
614
        /* end the list of coded C fragments */
615
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
616

    
617
    for (i = 0; i < s->fragment_count - 1; i++) {
618
        s->fast_fragment_list[i] = i + 1;
619
    }
620
    s->fast_fragment_list[s->fragment_count - 1] = -1;
621

    
622
    if (s->last_coded_y_fragment == -1)
623
        s->fragment_list_y_head = -1;
624
    else {
625
        s->fragment_list_y_head = s->first_coded_y_fragment;
626
        s->fast_fragment_list[s->last_coded_y_fragment] = -1;
627
    }
628

    
629
    if (s->last_coded_c_fragment == -1)
630
        s->fragment_list_c_head = -1;
631
    else {
632
        s->fragment_list_c_head = s->first_coded_c_fragment;
633
        s->fast_fragment_list[s->last_coded_c_fragment] = -1;
634
    }
635

    
636
    return 0;
637
}
638

    
639
/*
640
 * This function unpacks all the coding mode data for individual macroblocks
641
 * from the bitstream.
642
 */
643
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
644
{
645
    int i, j, k, sb_x, sb_y;
646
    int scheme;
647
    int current_macroblock;
648
    int current_fragment;
649
    int coding_mode;
650
    int custom_mode_alphabet[CODING_MODE_COUNT];
651
    const int *alphabet;
652

    
653
    if (s->keyframe) {
654
        for (i = 0; i < s->fragment_count; i++)
655
            s->all_fragments[i].coding_method = MODE_INTRA;
656

    
657
    } else {
658

    
659
        /* fetch the mode coding scheme for this frame */
660
        scheme = get_bits(gb, 3);
661

    
662
        /* is it a custom coding scheme? */
663
        if (scheme == 0) {
664
            for (i = 0; i < 8; i++)
665
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
666
            for (i = 0; i < 8; i++)
667
                custom_mode_alphabet[get_bits(gb, 3)] = i;
668
            alphabet = custom_mode_alphabet;
669
        } else
670
            alphabet = ModeAlphabet[scheme-1];
671

    
672
        /* iterate through all of the macroblocks that contain 1 or more
673
         * coded fragments */
674
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
675
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
676

    
677
            for (j = 0; j < 4; j++) {
678
                int mb_x = 2*sb_x +   (j>>1);
679
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
680
                int frags_coded = 0;
681
                current_macroblock = mb_y * s->macroblock_width + mb_x;
682

    
683
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
684
                    continue;
685

    
686
#define BLOCK_X (2*mb_x + (k&1))
687
#define BLOCK_Y (2*mb_y + (k>>1))
688
                /* coding modes are only stored if the macroblock has at least one
689
                 * luma block coded, otherwise it must be INTER_NO_MV */
690
                for (k = 0; k < 4; k++) {
691
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
692
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
693
                        break;
694
                }
695
                if (k == 4) {
696
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
697
                    continue;
698
                }
699

    
700
                /* mode 7 means get 3 bits for each coding mode */
701
                if (scheme == 7)
702
                    coding_mode = get_bits(gb, 3);
703
                else
704
                    coding_mode = alphabet
705
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
706

    
707
                s->macroblock_coding[current_macroblock] = coding_mode;
708
                for (k = 0; k < 4; k++) {
709
                    current_fragment =
710
                        BLOCK_Y*s->fragment_width + BLOCK_X;
711
                    if (s->all_fragments[current_fragment].coding_method !=
712
                        MODE_COPY)
713
                        s->all_fragments[current_fragment].coding_method =
714
                            coding_mode;
715
                }
716
                for (k = 0; k < 2; k++) {
717
                    current_fragment = s->fragment_start[k+1] +
718
                        mb_y*(s->fragment_width>>1) + mb_x;
719
                    if (s->all_fragments[current_fragment].coding_method !=
720
                        MODE_COPY)
721
                        s->all_fragments[current_fragment].coding_method =
722
                            coding_mode;
723
                }
724
            }
725
            }
726
        }
727
    }
728

    
729
    return 0;
730
}
731

    
732
/*
733
 * This function unpacks all the motion vectors for the individual
734
 * macroblocks from the bitstream.
735
 */
736
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
737
{
738
    int j, k, sb_x, sb_y;
739
    int coding_mode;
740
    int motion_x[6];
741
    int motion_y[6];
742
    int last_motion_x = 0;
743
    int last_motion_y = 0;
744
    int prior_last_motion_x = 0;
745
    int prior_last_motion_y = 0;
746
    int current_macroblock;
747
    int current_fragment;
748

    
749
    if (s->keyframe)
750
        return 0;
751

    
752
    memset(motion_x, 0, 6 * sizeof(int));
753
    memset(motion_y, 0, 6 * sizeof(int));
754

    
755
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
756
    coding_mode = get_bits1(gb);
757

    
758
    /* iterate through all of the macroblocks that contain 1 or more
759
     * coded fragments */
760
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
761
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
762

    
763
        for (j = 0; j < 4; j++) {
764
            int mb_x = 2*sb_x +   (j>>1);
765
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
766
            current_macroblock = mb_y * s->macroblock_width + mb_x;
767

    
768
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
769
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
770
                continue;
771

    
772
            switch (s->macroblock_coding[current_macroblock]) {
773

    
774
            case MODE_INTER_PLUS_MV:
775
            case MODE_GOLDEN_MV:
776
                /* all 6 fragments use the same motion vector */
777
                if (coding_mode == 0) {
778
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
779
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
780
                } else {
781
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
782
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
783
                }
784

    
785
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
786
                if (s->macroblock_coding[current_macroblock] ==
787
                    MODE_INTER_PLUS_MV) {
788
                    prior_last_motion_x = last_motion_x;
789
                    prior_last_motion_y = last_motion_y;
790
                    last_motion_x = motion_x[0];
791
                    last_motion_y = motion_y[0];
792
                }
793
                break;
794

    
795
            case MODE_INTER_FOURMV:
796
                /* vector maintenance */
797
                prior_last_motion_x = last_motion_x;
798
                prior_last_motion_y = last_motion_y;
799

    
800
                /* fetch 4 vectors from the bitstream, one for each
801
                 * Y fragment, then average for the C fragment vectors */
802
                motion_x[4] = motion_y[4] = 0;
803
                for (k = 0; k < 4; k++) {
804
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
805
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
806
                        if (coding_mode == 0) {
807
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
808
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
809
                        } else {
810
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
811
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
812
                        }
813
                        last_motion_x = motion_x[k];
814
                        last_motion_y = motion_y[k];
815
                    } else {
816
                        motion_x[k] = 0;
817
                        motion_y[k] = 0;
818
                    }
819
                    motion_x[4] += motion_x[k];
820
                    motion_y[4] += motion_y[k];
821
                }
822

    
823
                motion_x[5]=
824
                motion_x[4]= RSHIFT(motion_x[4], 2);
825
                motion_y[5]=
826
                motion_y[4]= RSHIFT(motion_y[4], 2);
827
                break;
828

    
829
            case MODE_INTER_LAST_MV:
830
                /* all 6 fragments use the last motion vector */
831
                motion_x[0] = last_motion_x;
832
                motion_y[0] = last_motion_y;
833

    
834
                /* no vector maintenance (last vector remains the
835
                 * last vector) */
836
                break;
837

    
838
            case MODE_INTER_PRIOR_LAST:
839
                /* all 6 fragments use the motion vector prior to the
840
                 * last motion vector */
841
                motion_x[0] = prior_last_motion_x;
842
                motion_y[0] = prior_last_motion_y;
843

    
844
                /* vector maintenance */
845
                prior_last_motion_x = last_motion_x;
846
                prior_last_motion_y = last_motion_y;
847
                last_motion_x = motion_x[0];
848
                last_motion_y = motion_y[0];
849
                break;
850

    
851
            default:
852
                /* covers intra, inter without MV, golden without MV */
853
                motion_x[0] = 0;
854
                motion_y[0] = 0;
855

    
856
                /* no vector maintenance */
857
                break;
858
            }
859

    
860
            /* assign the motion vectors to the correct fragments */
861
            for (k = 0; k < 4; k++) {
862
                current_fragment =
863
                    BLOCK_Y*s->fragment_width + BLOCK_X;
864
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
865
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
866
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
867
                } else {
868
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
869
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
870
                }
871
            }
872
            for (k = 0; k < 2; k++) {
873
                current_fragment = s->fragment_start[k+1] +
874
                    mb_y*(s->fragment_width>>1) + mb_x;
875
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
876
                    s->all_fragments[current_fragment].motion_x = motion_x[k+4];
877
                    s->all_fragments[current_fragment].motion_y = motion_y[k+4];
878
                } else {
879
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
880
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
881
                }
882
            }
883
        }
884
        }
885
    }
886

    
887
    return 0;
888
}
889

    
890
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
891
{
892
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
893
    int num_blocks = s->coded_fragment_list_index;
894

    
895
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
896
        i = blocks_decoded = num_blocks_at_qpi = 0;
897

    
898
        bit = get_bits1(gb);
899

    
900
        do {
901
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
902
            if (run_length == 34)
903
                run_length += get_bits(gb, 12);
904
            blocks_decoded += run_length;
905

    
906
            if (!bit)
907
                num_blocks_at_qpi += run_length;
908

    
909
            for (j = 0; j < run_length; i++) {
910
                if (i >= s->coded_fragment_list_index)
911
                    return -1;
912

    
913
                if (s->all_fragments[s->coded_fragment_list[i]].qpi == qpi) {
914
                    s->all_fragments[s->coded_fragment_list[i]].qpi += bit;
915
                    j++;
916
                }
917
            }
918

    
919
            if (run_length == 4129)
920
                bit = get_bits1(gb);
921
            else
922
                bit ^= 1;
923
        } while (blocks_decoded < num_blocks);
924

    
925
        num_blocks -= num_blocks_at_qpi;
926
    }
927

    
928
    return 0;
929
}
930

    
931
/*
932
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
933
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
934
 * data. This function unpacks all the VLCs for either the Y plane or both
935
 * C planes, and is called for DC coefficients or different AC coefficient
936
 * levels (since different coefficient types require different VLC tables.
937
 *
938
 * This function returns a residual eob run. E.g, if a particular token gave
939
 * instructions to EOB the next 5 fragments and there were only 2 fragments
940
 * left in the current fragment range, 3 would be returned so that it could
941
 * be passed into the next call to this same function.
942
 */
943
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
944
                        VLC *table, int coeff_index,
945
                        int y_plane,
946
                        int eob_run)
947
{
948
    int i;
949
    int token;
950
    int zero_run = 0;
951
    DCTELEM coeff = 0;
952
    Vp3Fragment *fragment;
953
    int bits_to_get;
954
    int next_fragment;
955
    int previous_fragment;
956
    int fragment_num;
957
    int *list_head;
958

    
959
    /* local references to structure members to avoid repeated deferences */
960
    uint8_t *perm= s->scantable.permutated;
961
    int *coded_fragment_list = s->coded_fragment_list;
962
    Vp3Fragment *all_fragments = s->all_fragments;
963
    uint8_t *coeff_counts = s->coeff_counts;
964
    VLC_TYPE (*vlc_table)[2] = table->table;
965
    int *fast_fragment_list = s->fast_fragment_list;
966

    
967
    if (y_plane) {
968
        next_fragment = s->fragment_list_y_head;
969
        list_head = &s->fragment_list_y_head;
970
    } else {
971
        next_fragment = s->fragment_list_c_head;
972
        list_head = &s->fragment_list_c_head;
973
    }
974

    
975
    i = next_fragment;
976
    previous_fragment = -1;  /* this indicates that the previous fragment is actually the list head */
977
    while (i != -1) {
978
        fragment_num = coded_fragment_list[i];
979

    
980
        if (coeff_counts[fragment_num] > coeff_index) {
981
            previous_fragment = i;
982
            i = fast_fragment_list[i];
983
            continue;
984
        }
985
        fragment = &all_fragments[fragment_num];
986

    
987
        if (!eob_run) {
988
            /* decode a VLC into a token */
989
            token = get_vlc2(gb, vlc_table, 5, 3);
990
            /* use the token to get a zero run, a coefficient, and an eob run */
991
            if (token <= 6) {
992
                eob_run = eob_run_base[token];
993
                if (eob_run_get_bits[token])
994
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
995
                coeff = zero_run = 0;
996
            } else {
997
                bits_to_get = coeff_get_bits[token];
998
                if (bits_to_get)
999
                    bits_to_get = get_bits(gb, bits_to_get);
1000
                coeff = coeff_tables[token][bits_to_get];
1001

    
1002
                zero_run = zero_run_base[token];
1003
                if (zero_run_get_bits[token])
1004
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
1005
            }
1006
        }
1007

    
1008
        if (!eob_run) {
1009
            coeff_counts[fragment_num] += zero_run;
1010
            if (coeff_counts[fragment_num] < 64){
1011
                fragment->next_coeff->coeff= coeff;
1012
                fragment->next_coeff->index= perm[coeff_counts[fragment_num]++]; //FIXME perm here already?
1013
                fragment->next_coeff->next= s->next_coeff;
1014
                s->next_coeff->next=NULL;
1015
                fragment->next_coeff= s->next_coeff++;
1016
            }
1017
            /* previous fragment is now this fragment */
1018
            previous_fragment = i;
1019
        } else {
1020
            coeff_counts[fragment_num] |= 128;
1021
            eob_run--;
1022
            /* remove this fragment from the list */
1023
            if (previous_fragment != -1)
1024
                fast_fragment_list[previous_fragment] = fast_fragment_list[i];
1025
            else
1026
                *list_head = fast_fragment_list[i];
1027
            /* previous fragment remains unchanged */
1028
        }
1029

    
1030
        i = fast_fragment_list[i];
1031
    }
1032

    
1033
    return eob_run;
1034
}
1035

    
1036
static void reverse_dc_prediction(Vp3DecodeContext *s,
1037
                                  int first_fragment,
1038
                                  int fragment_width,
1039
                                  int fragment_height);
1040
/*
1041
 * This function unpacks all of the DCT coefficient data from the
1042
 * bitstream.
1043
 */
1044
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1045
{
1046
    int i;
1047
    int dc_y_table;
1048
    int dc_c_table;
1049
    int ac_y_table;
1050
    int ac_c_table;
1051
    int residual_eob_run = 0;
1052
    VLC *y_tables[64];
1053
    VLC *c_tables[64];
1054

    
1055
    /* fetch the DC table indexes */
1056
    dc_y_table = get_bits(gb, 4);
1057
    dc_c_table = get_bits(gb, 4);
1058

    
1059
    /* unpack the Y plane DC coefficients */
1060
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1061
        1, residual_eob_run);
1062

    
1063
    /* reverse prediction of the Y-plane DC coefficients */
1064
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1065

    
1066
    /* unpack the C plane DC coefficients */
1067
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1068
        0, residual_eob_run);
1069

    
1070
    /* reverse prediction of the C-plane DC coefficients */
1071
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1072
    {
1073
        reverse_dc_prediction(s, s->fragment_start[1],
1074
            s->fragment_width / 2, s->fragment_height / 2);
1075
        reverse_dc_prediction(s, s->fragment_start[2],
1076
            s->fragment_width / 2, s->fragment_height / 2);
1077
    }
1078

    
1079
    /* fetch the AC table indexes */
1080
    ac_y_table = get_bits(gb, 4);
1081
    ac_c_table = get_bits(gb, 4);
1082

    
1083
    /* build tables of AC VLC tables */
1084
    for (i = 1; i <= 5; i++) {
1085
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1086
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1087
    }
1088
    for (i = 6; i <= 14; i++) {
1089
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1090
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1091
    }
1092
    for (i = 15; i <= 27; i++) {
1093
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1094
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1095
    }
1096
    for (i = 28; i <= 63; i++) {
1097
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1098
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1099
    }
1100

    
1101
    /* decode all AC coefficents */
1102
    for (i = 1; i <= 63; i++) {
1103
        if (s->fragment_list_y_head != -1)
1104
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1105
                1, residual_eob_run);
1106

    
1107
        if (s->fragment_list_c_head != -1)
1108
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1109
                0, residual_eob_run);
1110
    }
1111

    
1112
    return 0;
1113
}
1114

    
1115
/*
1116
 * This function reverses the DC prediction for each coded fragment in
1117
 * the frame. Much of this function is adapted directly from the original
1118
 * VP3 source code.
1119
 */
1120
#define COMPATIBLE_FRAME(x) \
1121
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1122
#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1123

    
1124
static void reverse_dc_prediction(Vp3DecodeContext *s,
1125
                                  int first_fragment,
1126
                                  int fragment_width,
1127
                                  int fragment_height)
1128
{
1129

    
1130
#define PUL 8
1131
#define PU 4
1132
#define PUR 2
1133
#define PL 1
1134

    
1135
    int x, y;
1136
    int i = first_fragment;
1137

    
1138
    int predicted_dc;
1139

    
1140
    /* DC values for the left, up-left, up, and up-right fragments */
1141
    int vl, vul, vu, vur;
1142

    
1143
    /* indexes for the left, up-left, up, and up-right fragments */
1144
    int l, ul, u, ur;
1145

    
1146
    /*
1147
     * The 6 fields mean:
1148
     *   0: up-left multiplier
1149
     *   1: up multiplier
1150
     *   2: up-right multiplier
1151
     *   3: left multiplier
1152
     */
1153
    static const int predictor_transform[16][4] = {
1154
        {  0,  0,  0,  0},
1155
        {  0,  0,  0,128},        // PL
1156
        {  0,  0,128,  0},        // PUR
1157
        {  0,  0, 53, 75},        // PUR|PL
1158
        {  0,128,  0,  0},        // PU
1159
        {  0, 64,  0, 64},        // PU|PL
1160
        {  0,128,  0,  0},        // PU|PUR
1161
        {  0,  0, 53, 75},        // PU|PUR|PL
1162
        {128,  0,  0,  0},        // PUL
1163
        {  0,  0,  0,128},        // PUL|PL
1164
        { 64,  0, 64,  0},        // PUL|PUR
1165
        {  0,  0, 53, 75},        // PUL|PUR|PL
1166
        {  0,128,  0,  0},        // PUL|PU
1167
       {-104,116,  0,116},        // PUL|PU|PL
1168
        { 24, 80, 24,  0},        // PUL|PU|PUR
1169
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1170
    };
1171

    
1172
    /* This table shows which types of blocks can use other blocks for
1173
     * prediction. For example, INTRA is the only mode in this table to
1174
     * have a frame number of 0. That means INTRA blocks can only predict
1175
     * from other INTRA blocks. There are 2 golden frame coding types;
1176
     * blocks encoding in these modes can only predict from other blocks
1177
     * that were encoded with these 1 of these 2 modes. */
1178
    static const unsigned char compatible_frame[9] = {
1179
        1,    /* MODE_INTER_NO_MV */
1180
        0,    /* MODE_INTRA */
1181
        1,    /* MODE_INTER_PLUS_MV */
1182
        1,    /* MODE_INTER_LAST_MV */
1183
        1,    /* MODE_INTER_PRIOR_MV */
1184
        2,    /* MODE_USING_GOLDEN */
1185
        2,    /* MODE_GOLDEN_MV */
1186
        1,    /* MODE_INTER_FOUR_MV */
1187
        3     /* MODE_COPY */
1188
    };
1189
    int current_frame_type;
1190

    
1191
    /* there is a last DC predictor for each of the 3 frame types */
1192
    short last_dc[3];
1193

    
1194
    int transform = 0;
1195

    
1196
    vul = vu = vur = vl = 0;
1197
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1198

    
1199
    /* for each fragment row... */
1200
    for (y = 0; y < fragment_height; y++) {
1201

    
1202
        /* for each fragment in a row... */
1203
        for (x = 0; x < fragment_width; x++, i++) {
1204

    
1205
            /* reverse prediction if this block was coded */
1206
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1207

    
1208
                current_frame_type =
1209
                    compatible_frame[s->all_fragments[i].coding_method];
1210

    
1211
                transform= 0;
1212
                if(x){
1213
                    l= i-1;
1214
                    vl = DC_COEFF(l);
1215
                    if(COMPATIBLE_FRAME(l))
1216
                        transform |= PL;
1217
                }
1218
                if(y){
1219
                    u= i-fragment_width;
1220
                    vu = DC_COEFF(u);
1221
                    if(COMPATIBLE_FRAME(u))
1222
                        transform |= PU;
1223
                    if(x){
1224
                        ul= i-fragment_width-1;
1225
                        vul = DC_COEFF(ul);
1226
                        if(COMPATIBLE_FRAME(ul))
1227
                            transform |= PUL;
1228
                    }
1229
                    if(x + 1 < fragment_width){
1230
                        ur= i-fragment_width+1;
1231
                        vur = DC_COEFF(ur);
1232
                        if(COMPATIBLE_FRAME(ur))
1233
                            transform |= PUR;
1234
                    }
1235
                }
1236

    
1237
                if (transform == 0) {
1238

    
1239
                    /* if there were no fragments to predict from, use last
1240
                     * DC saved */
1241
                    predicted_dc = last_dc[current_frame_type];
1242
                } else {
1243

    
1244
                    /* apply the appropriate predictor transform */
1245
                    predicted_dc =
1246
                        (predictor_transform[transform][0] * vul) +
1247
                        (predictor_transform[transform][1] * vu) +
1248
                        (predictor_transform[transform][2] * vur) +
1249
                        (predictor_transform[transform][3] * vl);
1250

    
1251
                    predicted_dc /= 128;
1252

    
1253
                    /* check for outranging on the [ul u l] and
1254
                     * [ul u ur l] predictors */
1255
                    if ((transform == 15) || (transform == 13)) {
1256
                        if (FFABS(predicted_dc - vu) > 128)
1257
                            predicted_dc = vu;
1258
                        else if (FFABS(predicted_dc - vl) > 128)
1259
                            predicted_dc = vl;
1260
                        else if (FFABS(predicted_dc - vul) > 128)
1261
                            predicted_dc = vul;
1262
                    }
1263
                }
1264

    
1265
                /* at long last, apply the predictor */
1266
                if(s->coeffs[i].index){
1267
                    *s->next_coeff= s->coeffs[i];
1268
                    s->coeffs[i].index=0;
1269
                    s->coeffs[i].coeff=0;
1270
                    s->coeffs[i].next= s->next_coeff++;
1271
                }
1272
                s->coeffs[i].coeff += predicted_dc;
1273
                /* save the DC */
1274
                last_dc[current_frame_type] = DC_COEFF(i);
1275
                if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1276
                    s->coeff_counts[i]= 129;
1277
//                    s->all_fragments[i].next_coeff= s->next_coeff;
1278
                    s->coeffs[i].next= s->next_coeff;
1279
                    (s->next_coeff++)->next=NULL;
1280
                }
1281
            }
1282
        }
1283
    }
1284
}
1285

    
1286
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1287
{
1288
    int x, y;
1289
    int *bounding_values= s->bounding_values_array+127;
1290

    
1291
    int width           = s->fragment_width  >> !!plane;
1292
    int height          = s->fragment_height >> !!plane;
1293
    int fragment        = s->fragment_start        [plane] + ystart * width;
1294
    int stride          = s->current_frame.linesize[plane];
1295
    uint8_t *plane_data = s->current_frame.data    [plane];
1296
    if (!s->flipped_image) stride = -stride;
1297
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1298

    
1299
    for (y = ystart; y < yend; y++) {
1300

    
1301
        for (x = 0; x < width; x++) {
1302
            /* This code basically just deblocks on the edges of coded blocks.
1303
             * However, it has to be much more complicated because of the
1304
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1305
             * because some pixels get filtered twice. */
1306
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1307
            {
1308
                /* do not perform left edge filter for left columns frags */
1309
                if (x > 0) {
1310
                    s->dsp.vp3_h_loop_filter(
1311
                        plane_data + 8*x,
1312
                        stride, bounding_values);
1313
                }
1314

    
1315
                /* do not perform top edge filter for top row fragments */
1316
                if (y > 0) {
1317
                    s->dsp.vp3_v_loop_filter(
1318
                        plane_data + 8*x,
1319
                        stride, bounding_values);
1320
                }
1321

    
1322
                /* do not perform right edge filter for right column
1323
                 * fragments or if right fragment neighbor is also coded
1324
                 * in this frame (it will be filtered in next iteration) */
1325
                if ((x < width - 1) &&
1326
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1327
                    s->dsp.vp3_h_loop_filter(
1328
                        plane_data + 8*x + 8,
1329
                        stride, bounding_values);
1330
                }
1331

    
1332
                /* do not perform bottom edge filter for bottom row
1333
                 * fragments or if bottom fragment neighbor is also coded
1334
                 * in this frame (it will be filtered in the next row) */
1335
                if ((y < height - 1) &&
1336
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1337
                    s->dsp.vp3_v_loop_filter(
1338
                        plane_data + 8*x + 8*stride,
1339
                        stride, bounding_values);
1340
                }
1341
            }
1342

    
1343
            fragment++;
1344
        }
1345
        plane_data += 8*stride;
1346
    }
1347
}
1348

    
1349
/**
1350
 * called when all pixels up to row y are complete
1351
 */
1352
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1353
{
1354
    int h, cy;
1355
    int offset[4];
1356

    
1357
    if(s->avctx->draw_horiz_band==NULL)
1358
        return;
1359

    
1360
    h= y - s->last_slice_end;
1361
    y -= h;
1362

    
1363
    if (!s->flipped_image) {
1364
        if (y == 0)
1365
            h -= s->height - s->avctx->height;  // account for non-mod16
1366
        y = s->height - y - h;
1367
    }
1368

    
1369
    cy = y >> 1;
1370
    offset[0] = s->current_frame.linesize[0]*y;
1371
    offset[1] = s->current_frame.linesize[1]*cy;
1372
    offset[2] = s->current_frame.linesize[2]*cy;
1373
    offset[3] = 0;
1374

    
1375
    emms_c();
1376
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1377
    s->last_slice_end= y + h;
1378
}
1379

    
1380
/*
1381
 * Perform the final rendering for a particular slice of data.
1382
 * The slice number ranges from 0..(macroblock_height - 1).
1383
 */
1384
static void render_slice(Vp3DecodeContext *s, int slice)
1385
{
1386
    int x;
1387
    int16_t *dequantizer;
1388
    DECLARE_ALIGNED_16(DCTELEM, block)[64];
1389
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1390
    int motion_halfpel_index;
1391
    uint8_t *motion_source;
1392
    int plane;
1393

    
1394
    if (slice >= s->macroblock_height)
1395
        return;
1396

    
1397
    for (plane = 0; plane < 3; plane++) {
1398
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1399
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1400
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1401
        int stride            = s->current_frame.linesize[plane];
1402
        int plane_width       = s->width  >> !!plane;
1403
        int plane_height      = s->height >> !!plane;
1404
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
1405
        int slice_height = y + (FRAGMENT_PIXELS << !plane);
1406
        int i = s->fragment_start[plane] + (y>>3)*(s->fragment_width>>!!plane);
1407

    
1408
        if (!s->flipped_image) stride = -stride;
1409
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1410
            continue;
1411

    
1412

    
1413
        if(FFABS(stride) > 2048)
1414
            return; //various tables are fixed size
1415

    
1416
        /* for each fragment row in the slice (both of them)... */
1417
        for (; y < slice_height; y += 8) {
1418

    
1419
            /* for each fragment in a row... */
1420
            for (x = 0; x < plane_width; x += 8, i++) {
1421
                int first_pixel = y*stride + x;
1422

    
1423
                if ((i < 0) || (i >= s->fragment_count)) {
1424
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
1425
                    return;
1426
                }
1427

    
1428
                /* transform if this block was coded */
1429
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1430

    
1431
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1432
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1433
                        motion_source= golden_plane;
1434
                    else
1435
                        motion_source= last_plane;
1436

    
1437
                    motion_source += first_pixel;
1438
                    motion_halfpel_index = 0;
1439

    
1440
                    /* sort out the motion vector if this fragment is coded
1441
                     * using a motion vector method */
1442
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1443
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1444
                        int src_x, src_y;
1445
                        motion_x = s->all_fragments[i].motion_x;
1446
                        motion_y = s->all_fragments[i].motion_y;
1447
                        if(plane){
1448
                            motion_x= (motion_x>>1) | (motion_x&1);
1449
                            motion_y= (motion_y>>1) | (motion_y&1);
1450
                        }
1451

    
1452
                        src_x= (motion_x>>1) + x;
1453
                        src_y= (motion_y>>1) + y;
1454
                        if ((motion_x == 127) || (motion_y == 127))
1455
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1456

    
1457
                        motion_halfpel_index = motion_x & 0x01;
1458
                        motion_source += (motion_x >> 1);
1459

    
1460
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1461
                        motion_source += ((motion_y >> 1) * stride);
1462

    
1463
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1464
                            uint8_t *temp= s->edge_emu_buffer;
1465
                            if(stride<0) temp -= 9*stride;
1466
                            else temp += 9*stride;
1467

    
1468
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1469
                            motion_source= temp;
1470
                        }
1471
                    }
1472

    
1473

    
1474
                    /* first, take care of copying a block from either the
1475
                     * previous or the golden frame */
1476
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1477
                        /* Note, it is possible to implement all MC cases with
1478
                           put_no_rnd_pixels_l2 which would look more like the
1479
                           VP3 source but this would be slower as
1480
                           put_no_rnd_pixels_tab is better optimzed */
1481
                        if(motion_halfpel_index != 3){
1482
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1483
                                output_plane + first_pixel,
1484
                                motion_source, stride, 8);
1485
                        }else{
1486
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1487
                            s->dsp.put_no_rnd_pixels_l2[1](
1488
                                output_plane + first_pixel,
1489
                                motion_source - d,
1490
                                motion_source + stride + 1 + d,
1491
                                stride, 8);
1492
                        }
1493
                        dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];
1494
                    }else{
1495
                        dequantizer = s->qmat[s->all_fragments[i].qpi][0][plane];
1496
                    }
1497

    
1498
                    /* dequantize the DCT coefficients */
1499
                    if(s->avctx->idct_algo==FF_IDCT_VP3){
1500
                        Coeff *coeff= s->coeffs + i;
1501
                        s->dsp.clear_block(block);
1502
                        while(coeff->next){
1503
                            block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1504
                            coeff= coeff->next;
1505
                        }
1506
                    }else{
1507
                        Coeff *coeff= s->coeffs + i;
1508
                        s->dsp.clear_block(block);
1509
                        while(coeff->next){
1510
                            block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1511
                            coeff= coeff->next;
1512
                        }
1513
                    }
1514

    
1515
                    /* invert DCT and place (or add) in final output */
1516

    
1517
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1518
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1519
                            block[0] += 128<<3;
1520
                        s->dsp.idct_put(
1521
                            output_plane + first_pixel,
1522
                            stride,
1523
                            block);
1524
                    } else {
1525
                        s->dsp.idct_add(
1526
                            output_plane + first_pixel,
1527
                            stride,
1528
                            block);
1529
                    }
1530
                } else {
1531

    
1532
                    /* copy directly from the previous frame */
1533
                    s->dsp.put_pixels_tab[1][0](
1534
                        output_plane + first_pixel,
1535
                        last_plane + first_pixel,
1536
                        stride, 8);
1537

    
1538
                }
1539
            }
1540
            // Filter the previous block row. We can't filter the current row yet
1541
            // since it needs pixels from the next row
1542
            if (y > 0)
1543
                apply_loop_filter(s, plane, (y>>3)-1, (y>>3));
1544
        }
1545
    }
1546

    
1547
     /* this looks like a good place for slice dispatch... */
1548
     /* algorithm:
1549
      *   if (slice == s->macroblock_height - 1)
1550
      *     dispatch (both last slice & 2nd-to-last slice);
1551
      *   else if (slice > 0)
1552
      *     dispatch (slice - 1);
1553
      */
1554

    
1555
    // now that we've filtered the last rows, they're safe to display
1556
    if (slice)
1557
        vp3_draw_horiz_band(s, 16*slice);
1558
}
1559

    
1560
/*
1561
 * This is the ffmpeg/libavcodec API init function.
1562
 */
1563
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1564
{
1565
    Vp3DecodeContext *s = avctx->priv_data;
1566
    int i, inter, plane;
1567
    int c_width;
1568
    int c_height;
1569
    int y_superblock_count;
1570
    int c_superblock_count;
1571

    
1572
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1573
        s->version = 0;
1574
    else
1575
        s->version = 1;
1576

    
1577
    s->avctx = avctx;
1578
    s->width = FFALIGN(avctx->width, 16);
1579
    s->height = FFALIGN(avctx->height, 16);
1580
    avctx->pix_fmt = PIX_FMT_YUV420P;
1581
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1582
    if(avctx->idct_algo==FF_IDCT_AUTO)
1583
        avctx->idct_algo=FF_IDCT_VP3;
1584
    dsputil_init(&s->dsp, avctx);
1585

    
1586
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1587

    
1588
    /* initialize to an impossible value which will force a recalculation
1589
     * in the first frame decode */
1590
    for (i = 0; i < 3; i++)
1591
        s->qps[i] = -1;
1592

    
1593
    s->y_superblock_width = (s->width + 31) / 32;
1594
    s->y_superblock_height = (s->height + 31) / 32;
1595
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1596

    
1597
    /* work out the dimensions for the C planes */
1598
    c_width = s->width / 2;
1599
    c_height = s->height / 2;
1600
    s->c_superblock_width = (c_width + 31) / 32;
1601
    s->c_superblock_height = (c_height + 31) / 32;
1602
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1603

    
1604
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1605
    s->u_superblock_start = y_superblock_count;
1606
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1607
    s->superblock_coding = av_malloc(s->superblock_count);
1608

    
1609
    s->macroblock_width = (s->width + 15) / 16;
1610
    s->macroblock_height = (s->height + 15) / 16;
1611
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1612

    
1613
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1614
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1615

    
1616
    /* fragment count covers all 8x8 blocks for all 3 planes */
1617
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1618
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1619
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1620

    
1621
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1622
    s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1623
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1624
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1625
    s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1626
    if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
1627
        !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
1628
        vp3_decode_end(avctx);
1629
        return -1;
1630
    }
1631

    
1632
    if (!s->theora_tables)
1633
    {
1634
        for (i = 0; i < 64; i++) {
1635
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1636
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1637
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1638
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1639
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1640
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1641
        }
1642

    
1643
        for(inter=0; inter<2; inter++){
1644
            for(plane=0; plane<3; plane++){
1645
                s->qr_count[inter][plane]= 1;
1646
                s->qr_size [inter][plane][0]= 63;
1647
                s->qr_base [inter][plane][0]=
1648
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1649
            }
1650
        }
1651

    
1652
        /* init VLC tables */
1653
        for (i = 0; i < 16; i++) {
1654

    
1655
            /* DC histograms */
1656
            init_vlc(&s->dc_vlc[i], 5, 32,
1657
                &dc_bias[i][0][1], 4, 2,
1658
                &dc_bias[i][0][0], 4, 2, 0);
1659

    
1660
            /* group 1 AC histograms */
1661
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1662
                &ac_bias_0[i][0][1], 4, 2,
1663
                &ac_bias_0[i][0][0], 4, 2, 0);
1664

    
1665
            /* group 2 AC histograms */
1666
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1667
                &ac_bias_1[i][0][1], 4, 2,
1668
                &ac_bias_1[i][0][0], 4, 2, 0);
1669

    
1670
            /* group 3 AC histograms */
1671
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1672
                &ac_bias_2[i][0][1], 4, 2,
1673
                &ac_bias_2[i][0][0], 4, 2, 0);
1674

    
1675
            /* group 4 AC histograms */
1676
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1677
                &ac_bias_3[i][0][1], 4, 2,
1678
                &ac_bias_3[i][0][0], 4, 2, 0);
1679
        }
1680
    } else {
1681
        for (i = 0; i < 16; i++) {
1682

    
1683
            /* DC histograms */
1684
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1685
                &s->huffman_table[i][0][1], 4, 2,
1686
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1687
                goto vlc_fail;
1688

    
1689
            /* group 1 AC histograms */
1690
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1691
                &s->huffman_table[i+16][0][1], 4, 2,
1692
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1693
                goto vlc_fail;
1694

    
1695
            /* group 2 AC histograms */
1696
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1697
                &s->huffman_table[i+16*2][0][1], 4, 2,
1698
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1699
                goto vlc_fail;
1700

    
1701
            /* group 3 AC histograms */
1702
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1703
                &s->huffman_table[i+16*3][0][1], 4, 2,
1704
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1705
                goto vlc_fail;
1706

    
1707
            /* group 4 AC histograms */
1708
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1709
                &s->huffman_table[i+16*4][0][1], 4, 2,
1710
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1711
                goto vlc_fail;
1712
        }
1713
    }
1714

    
1715
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1716
        &superblock_run_length_vlc_table[0][1], 4, 2,
1717
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1718

    
1719
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1720
        &fragment_run_length_vlc_table[0][1], 4, 2,
1721
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1722

    
1723
    init_vlc(&s->mode_code_vlc, 3, 8,
1724
        &mode_code_vlc_table[0][1], 2, 1,
1725
        &mode_code_vlc_table[0][0], 2, 1, 0);
1726

    
1727
    init_vlc(&s->motion_vector_vlc, 6, 63,
1728
        &motion_vector_vlc_table[0][1], 2, 1,
1729
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1730

    
1731
    /* work out the block mapping tables */
1732
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1733
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1734
    if (!s->superblock_fragments || !s->macroblock_coding) {
1735
        vp3_decode_end(avctx);
1736
        return -1;
1737
    }
1738
    init_block_mapping(s);
1739

    
1740
    for (i = 0; i < 3; i++) {
1741
        s->current_frame.data[i] = NULL;
1742
        s->last_frame.data[i] = NULL;
1743
        s->golden_frame.data[i] = NULL;
1744
    }
1745

    
1746
    return 0;
1747

    
1748
vlc_fail:
1749
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1750
    return -1;
1751
}
1752

    
1753
/*
1754
 * This is the ffmpeg/libavcodec API frame decode function.
1755
 */
1756
static int vp3_decode_frame(AVCodecContext *avctx,
1757
                            void *data, int *data_size,
1758
                            AVPacket *avpkt)
1759
{
1760
    const uint8_t *buf = avpkt->data;
1761
    int buf_size = avpkt->size;
1762
    Vp3DecodeContext *s = avctx->priv_data;
1763
    GetBitContext gb;
1764
    static int counter = 0;
1765
    int i;
1766

    
1767
    init_get_bits(&gb, buf, buf_size * 8);
1768

    
1769
    if (s->theora && get_bits1(&gb))
1770
    {
1771
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1772
        return -1;
1773
    }
1774

    
1775
    s->keyframe = !get_bits1(&gb);
1776
    if (!s->theora)
1777
        skip_bits(&gb, 1);
1778
    for (i = 0; i < 3; i++)
1779
        s->last_qps[i] = s->qps[i];
1780

    
1781
    s->nqps=0;
1782
    do{
1783
        s->qps[s->nqps++]= get_bits(&gb, 6);
1784
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1785
    for (i = s->nqps; i < 3; i++)
1786
        s->qps[i] = -1;
1787

    
1788
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1789
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1790
            s->keyframe?"key":"", counter, s->qps[0]);
1791
    counter++;
1792

    
1793
    if (s->qps[0] != s->last_qps[0])
1794
        init_loop_filter(s);
1795

    
1796
    for (i = 0; i < s->nqps; i++)
1797
        // reinit all dequantizers if the first one changed, because
1798
        // the DC of the first quantizer must be used for all matrices
1799
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1800
            init_dequantizer(s, i);
1801

    
1802
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1803
        return buf_size;
1804

    
1805
    if (s->keyframe) {
1806
        if (!s->theora)
1807
        {
1808
            skip_bits(&gb, 4); /* width code */
1809
            skip_bits(&gb, 4); /* height code */
1810
            if (s->version)
1811
            {
1812
                s->version = get_bits(&gb, 5);
1813
                if (counter == 1)
1814
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1815
            }
1816
        }
1817
        if (s->version || s->theora)
1818
        {
1819
                if (get_bits1(&gb))
1820
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1821
            skip_bits(&gb, 2); /* reserved? */
1822
        }
1823

    
1824
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1825
            if (s->golden_frame.data[0])
1826
                avctx->release_buffer(avctx, &s->golden_frame);
1827
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1828
        } else {
1829
            if (s->golden_frame.data[0])
1830
                avctx->release_buffer(avctx, &s->golden_frame);
1831
            if (s->last_frame.data[0])
1832
                avctx->release_buffer(avctx, &s->last_frame);
1833
        }
1834

    
1835
        s->golden_frame.reference = 3;
1836
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1837
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1838
            return -1;
1839
        }
1840

    
1841
        /* golden frame is also the current frame */
1842
        s->current_frame= s->golden_frame;
1843
    } else {
1844
        /* allocate a new current frame */
1845
        s->current_frame.reference = 3;
1846
        if (!s->golden_frame.data[0]) {
1847
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1848
            return -1;
1849
        }
1850
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1851
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1852
            return -1;
1853
        }
1854
    }
1855

    
1856
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1857
    s->current_frame.qstride= 0;
1858

    
1859
    init_frame(s, &gb);
1860

    
1861
    if (unpack_superblocks(s, &gb)){
1862
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1863
        return -1;
1864
    }
1865
    if (unpack_modes(s, &gb)){
1866
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1867
        return -1;
1868
    }
1869
    if (unpack_vectors(s, &gb)){
1870
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1871
        return -1;
1872
    }
1873
    if (unpack_block_qpis(s, &gb)){
1874
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1875
        return -1;
1876
    }
1877
    if (unpack_dct_coeffs(s, &gb)){
1878
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1879
        return -1;
1880
    }
1881

    
1882
    for (i = 0; i < 3; i++) {
1883
        if (s->flipped_image)
1884
            s->data_offset[i] = 0;
1885
        else
1886
            s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1887
    }
1888

    
1889
    s->last_slice_end = 0;
1890
    for (i = 0; i < s->macroblock_height; i++)
1891
        render_slice(s, i);
1892

    
1893
    // filter the last row
1894
    for (i = 0; i < 3; i++) {
1895
        int row = (s->height >> (3+!!i)) - 1;
1896
        apply_loop_filter(s, i, row, row+1);
1897
    }
1898
    vp3_draw_horiz_band(s, s->height);
1899

    
1900
    *data_size=sizeof(AVFrame);
1901
    *(AVFrame*)data= s->current_frame;
1902

    
1903
    /* release the last frame, if it is allocated and if it is not the
1904
     * golden frame */
1905
    if ((s->last_frame.data[0]) &&
1906
        (s->last_frame.data[0] != s->golden_frame.data[0]))
1907
        avctx->release_buffer(avctx, &s->last_frame);
1908

    
1909
    /* shuffle frames (last = current) */
1910
    s->last_frame= s->current_frame;
1911
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1912

    
1913
    return buf_size;
1914
}
1915

    
1916
/*
1917
 * This is the ffmpeg/libavcodec API module cleanup function.
1918
 */
1919
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1920
{
1921
    Vp3DecodeContext *s = avctx->priv_data;
1922
    int i;
1923

    
1924
    av_free(s->superblock_coding);
1925
    av_free(s->all_fragments);
1926
    av_free(s->coeff_counts);
1927
    av_free(s->coeffs);
1928
    av_free(s->coded_fragment_list);
1929
    av_free(s->fast_fragment_list);
1930
    av_free(s->superblock_fragments);
1931
    av_free(s->macroblock_coding);
1932

    
1933
    for (i = 0; i < 16; i++) {
1934
        free_vlc(&s->dc_vlc[i]);
1935
        free_vlc(&s->ac_vlc_1[i]);
1936
        free_vlc(&s->ac_vlc_2[i]);
1937
        free_vlc(&s->ac_vlc_3[i]);
1938
        free_vlc(&s->ac_vlc_4[i]);
1939
    }
1940

    
1941
    free_vlc(&s->superblock_run_length_vlc);
1942
    free_vlc(&s->fragment_run_length_vlc);
1943
    free_vlc(&s->mode_code_vlc);
1944
    free_vlc(&s->motion_vector_vlc);
1945

    
1946
    /* release all frames */
1947
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1948
        avctx->release_buffer(avctx, &s->golden_frame);
1949
    if (s->last_frame.data[0])
1950
        avctx->release_buffer(avctx, &s->last_frame);
1951
    /* no need to release the current_frame since it will always be pointing
1952
     * to the same frame as either the golden or last frame */
1953

    
1954
    return 0;
1955
}
1956

    
1957
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1958
{
1959
    Vp3DecodeContext *s = avctx->priv_data;
1960

    
1961
    if (get_bits1(gb)) {
1962
        int token;
1963
        if (s->entries >= 32) { /* overflow */
1964
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1965
            return -1;
1966
        }
1967
        token = get_bits(gb, 5);
1968
        //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);
1969
        s->huffman_table[s->hti][token][0] = s->hbits;
1970
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1971
        s->entries++;
1972
    }
1973
    else {
1974
        if (s->huff_code_size >= 32) {/* overflow */
1975
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1976
            return -1;
1977
        }
1978
        s->huff_code_size++;
1979
        s->hbits <<= 1;
1980
        if (read_huffman_tree(avctx, gb))
1981
            return -1;
1982
        s->hbits |= 1;
1983
        if (read_huffman_tree(avctx, gb))
1984
            return -1;
1985
        s->hbits >>= 1;
1986
        s->huff_code_size--;
1987
    }
1988
    return 0;
1989
}
1990

    
1991
#if CONFIG_THEORA_DECODER
1992
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1993
{
1994
    Vp3DecodeContext *s = avctx->priv_data;
1995
    int visible_width, visible_height, colorspace;
1996

    
1997
    s->theora = get_bits_long(gb, 24);
1998
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1999

    
2000
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2001
    /* but previous versions have the image flipped relative to vp3 */
2002
    if (s->theora < 0x030200)
2003
    {
2004
        s->flipped_image = 1;
2005
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2006
    }
2007

    
2008
    visible_width  = s->width  = get_bits(gb, 16) << 4;
2009
    visible_height = s->height = get_bits(gb, 16) << 4;
2010

    
2011
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2012
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2013
        s->width= s->height= 0;
2014
        return -1;
2015
    }
2016

    
2017
    if (s->theora >= 0x030200) {
2018
        visible_width  = get_bits_long(gb, 24);
2019
        visible_height = get_bits_long(gb, 24);
2020

    
2021
        skip_bits(gb, 8); /* offset x */
2022
        skip_bits(gb, 8); /* offset y */
2023
    }
2024

    
2025
    skip_bits(gb, 32); /* fps numerator */
2026
    skip_bits(gb, 32); /* fps denumerator */
2027
    skip_bits(gb, 24); /* aspect numerator */
2028
    skip_bits(gb, 24); /* aspect denumerator */
2029

    
2030
    if (s->theora < 0x030200)
2031
        skip_bits(gb, 5); /* keyframe frequency force */
2032
    colorspace = get_bits(gb, 8);
2033
    skip_bits(gb, 24); /* bitrate */
2034

    
2035
    skip_bits(gb, 6); /* quality hint */
2036

    
2037
    if (s->theora >= 0x030200)
2038
    {
2039
        skip_bits(gb, 5); /* keyframe frequency force */
2040
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2041
        skip_bits(gb, 3); /* reserved */
2042
    }
2043

    
2044
//    align_get_bits(gb);
2045

    
2046
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2047
        && visible_height <= s->height && visible_height > s->height-16)
2048
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2049
    else
2050
        avcodec_set_dimensions(avctx, s->width, s->height);
2051

    
2052
    if (colorspace == 1) {
2053
        avctx->color_primaries = AVCOL_PRI_BT470M;
2054
    } else if (colorspace == 2) {
2055
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2056
    }
2057
    if (colorspace == 1 || colorspace == 2) {
2058
        avctx->colorspace = AVCOL_SPC_BT470BG;
2059
        avctx->color_trc  = AVCOL_TRC_BT709;
2060
    }
2061

    
2062
    return 0;
2063
}
2064

    
2065
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2066
{
2067
    Vp3DecodeContext *s = avctx->priv_data;
2068
    int i, n, matrices, inter, plane;
2069

    
2070
    if (s->theora >= 0x030200) {
2071
        n = get_bits(gb, 3);
2072
        /* loop filter limit values table */
2073
        for (i = 0; i < 64; i++) {
2074
            s->filter_limit_values[i] = get_bits(gb, n);
2075
            if (s->filter_limit_values[i] > 127) {
2076
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2077
                s->filter_limit_values[i] = 127;
2078
            }
2079
        }
2080
    }
2081

    
2082
    if (s->theora >= 0x030200)
2083
        n = get_bits(gb, 4) + 1;
2084
    else
2085
        n = 16;
2086
    /* quality threshold table */
2087
    for (i = 0; i < 64; i++)
2088
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2089

    
2090
    if (s->theora >= 0x030200)
2091
        n = get_bits(gb, 4) + 1;
2092
    else
2093
        n = 16;
2094
    /* dc scale factor table */
2095
    for (i = 0; i < 64; i++)
2096
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2097

    
2098
    if (s->theora >= 0x030200)
2099
        matrices = get_bits(gb, 9) + 1;
2100
    else
2101
        matrices = 3;
2102

    
2103
    if(matrices > 384){
2104
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2105
        return -1;
2106
    }
2107

    
2108
    for(n=0; n<matrices; n++){
2109
        for (i = 0; i < 64; i++)
2110
            s->base_matrix[n][i]= get_bits(gb, 8);
2111
    }
2112

    
2113
    for (inter = 0; inter <= 1; inter++) {
2114
        for (plane = 0; plane <= 2; plane++) {
2115
            int newqr= 1;
2116
            if (inter || plane > 0)
2117
                newqr = get_bits1(gb);
2118
            if (!newqr) {
2119
                int qtj, plj;
2120
                if(inter && get_bits1(gb)){
2121
                    qtj = 0;
2122
                    plj = plane;
2123
                }else{
2124
                    qtj= (3*inter + plane - 1) / 3;
2125
                    plj= (plane + 2) % 3;
2126
                }
2127
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2128
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2129
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2130
            } else {
2131
                int qri= 0;
2132
                int qi = 0;
2133

    
2134
                for(;;){
2135
                    i= get_bits(gb, av_log2(matrices-1)+1);
2136
                    if(i>= matrices){
2137
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2138
                        return -1;
2139
                    }
2140
                    s->qr_base[inter][plane][qri]= i;
2141
                    if(qi >= 63)
2142
                        break;
2143
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2144
                    s->qr_size[inter][plane][qri++]= i;
2145
                    qi += i;
2146
                }
2147

    
2148
                if (qi > 63) {
2149
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2150
                    return -1;
2151
                }
2152
                s->qr_count[inter][plane]= qri;
2153
            }
2154
        }
2155
    }
2156

    
2157
    /* Huffman tables */
2158
    for (s->hti = 0; s->hti < 80; s->hti++) {
2159
        s->entries = 0;
2160
        s->huff_code_size = 1;
2161
        if (!get_bits1(gb)) {
2162
            s->hbits = 0;
2163
            if(read_huffman_tree(avctx, gb))
2164
                return -1;
2165
            s->hbits = 1;
2166
            if(read_huffman_tree(avctx, gb))
2167
                return -1;
2168
        }
2169
    }
2170

    
2171
    s->theora_tables = 1;
2172

    
2173
    return 0;
2174
}
2175

    
2176
static av_cold int theora_decode_init(AVCodecContext *avctx)
2177
{
2178
    Vp3DecodeContext *s = avctx->priv_data;
2179
    GetBitContext gb;
2180
    int ptype;
2181
    uint8_t *header_start[3];
2182
    int header_len[3];
2183
    int i;
2184

    
2185
    s->theora = 1;
2186

    
2187
    if (!avctx->extradata_size)
2188
    {
2189
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2190
        return -1;
2191
    }
2192

    
2193
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2194
                              42, header_start, header_len) < 0) {
2195
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2196
        return -1;
2197
    }
2198

    
2199
  for(i=0;i<3;i++) {
2200
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2201

    
2202
    ptype = get_bits(&gb, 8);
2203

    
2204
     if (!(ptype & 0x80))
2205
     {
2206
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2207
//        return -1;
2208
     }
2209

    
2210
    // FIXME: Check for this as well.
2211
    skip_bits_long(&gb, 6*8); /* "theora" */
2212

    
2213
    switch(ptype)
2214
    {
2215
        case 0x80:
2216
            theora_decode_header(avctx, &gb);
2217
                break;
2218
        case 0x81:
2219
// FIXME: is this needed? it breaks sometimes
2220
//            theora_decode_comments(avctx, gb);
2221
            break;
2222
        case 0x82:
2223
            if (theora_decode_tables(avctx, &gb))
2224
                return -1;
2225
            break;
2226
        default:
2227
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2228
            break;
2229
    }
2230
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2231
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2232
    if (s->theora < 0x030200)
2233
        break;
2234
  }
2235

    
2236
    return vp3_decode_init(avctx);
2237
}
2238

    
2239
AVCodec theora_decoder = {
2240
    "theora",
2241
    CODEC_TYPE_VIDEO,
2242
    CODEC_ID_THEORA,
2243
    sizeof(Vp3DecodeContext),
2244
    theora_decode_init,
2245
    NULL,
2246
    vp3_decode_end,
2247
    vp3_decode_frame,
2248
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2249
    NULL,
2250
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2251
};
2252
#endif
2253

    
2254
AVCodec vp3_decoder = {
2255
    "vp3",
2256
    CODEC_TYPE_VIDEO,
2257
    CODEC_ID_VP3,
2258
    sizeof(Vp3DecodeContext),
2259
    vp3_decode_init,
2260
    NULL,
2261
    vp3_decode_end,
2262
    vp3_decode_frame,
2263
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2264
    NULL,
2265
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2266
};