Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 2c823b3c

History | View | Annotate | Download (80.7 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
#include <unistd.h>
36

    
37
#include "avcodec.h"
38
#include "dsputil.h"
39
#include "bitstream.h"
40

    
41
#include "vp3data.h"
42
#include "xiph.h"
43

    
44
#define FRAGMENT_PIXELS 8
45

    
46
typedef struct Coeff {
47
    struct Coeff *next;
48
    DCTELEM coeff;
49
    uint8_t index;
50
} Coeff;
51

    
52
//FIXME split things out into their own arrays
53
typedef struct Vp3Fragment {
54
    Coeff *next_coeff;
55
    /* address of first pixel taking into account which plane the fragment
56
     * lives on as well as the plane stride */
57
    int first_pixel;
58
    /* this is the macroblock that the fragment belongs to */
59
    uint16_t macroblock;
60
    uint8_t coding_method;
61
    int8_t motion_x;
62
    int8_t motion_y;
63
} Vp3Fragment;
64

    
65
#define SB_NOT_CODED        0
66
#define SB_PARTIALLY_CODED  1
67
#define SB_FULLY_CODED      2
68

    
69
#define MODE_INTER_NO_MV      0
70
#define MODE_INTRA            1
71
#define MODE_INTER_PLUS_MV    2
72
#define MODE_INTER_LAST_MV    3
73
#define MODE_INTER_PRIOR_LAST 4
74
#define MODE_USING_GOLDEN     5
75
#define MODE_GOLDEN_MV        6
76
#define MODE_INTER_FOURMV     7
77
#define CODING_MODE_COUNT     8
78

    
79
/* special internal mode */
80
#define MODE_COPY             8
81

    
82
/* There are 6 preset schemes, plus a free-form scheme */
83
static const int ModeAlphabet[6][CODING_MODE_COUNT] =
84
{
85
    /* scheme 1: Last motion vector dominates */
86
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
87
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
88
         MODE_INTRA,            MODE_USING_GOLDEN,
89
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
90

    
91
    /* scheme 2 */
92
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
93
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
94
         MODE_INTRA,            MODE_USING_GOLDEN,
95
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
96

    
97
    /* scheme 3 */
98
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
99
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
100
         MODE_INTRA,            MODE_USING_GOLDEN,
101
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
102

    
103
    /* scheme 4 */
104
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
105
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
106
         MODE_INTRA,            MODE_USING_GOLDEN,
107
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
108

    
109
    /* scheme 5: No motion vector dominates */
110
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
111
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
112
         MODE_INTRA,            MODE_USING_GOLDEN,
113
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
114

    
115
    /* scheme 6 */
116
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
117
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
118
         MODE_INTER_PLUS_MV,    MODE_INTRA,
119
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
120

    
121
};
122

    
123
#define MIN_DEQUANT_VAL 2
124

    
125
typedef struct Vp3DecodeContext {
126
    AVCodecContext *avctx;
127
    int theora, theora_tables;
128
    int version;
129
    int width, height;
130
    AVFrame golden_frame;
131
    AVFrame last_frame;
132
    AVFrame current_frame;
133
    int keyframe;
134
    DSPContext dsp;
135
    int flipped_image;
136

    
137
    int qis[3];
138
    int nqis;
139
    int quality_index;
140
    int last_quality_index;
141

    
142
    int superblock_count;
143
    int y_superblock_width;
144
    int y_superblock_height;
145
    int c_superblock_width;
146
    int c_superblock_height;
147
    int u_superblock_start;
148
    int v_superblock_start;
149
    unsigned char *superblock_coding;
150

    
151
    int macroblock_count;
152
    int macroblock_width;
153
    int macroblock_height;
154

    
155
    int fragment_count;
156
    int fragment_width;
157
    int fragment_height;
158

    
159
    Vp3Fragment *all_fragments;
160
    uint8_t *coeff_counts;
161
    Coeff *coeffs;
162
    Coeff *next_coeff;
163
    int fragment_start[3];
164

    
165
    ScanTable scantable;
166

    
167
    /* tables */
168
    uint16_t coded_dc_scale_factor[64];
169
    uint32_t coded_ac_scale_factor[64];
170
    uint8_t base_matrix[384][64];
171
    uint8_t qr_count[2][3];
172
    uint8_t qr_size [2][3][64];
173
    uint16_t qr_base[2][3][64];
174

    
175
    /* this is a list of indexes into the all_fragments array indicating
176
     * which of the fragments are coded */
177
    int *coded_fragment_list;
178
    int coded_fragment_list_index;
179
    int pixel_addresses_initialized;
180

    
181
    VLC dc_vlc[16];
182
    VLC ac_vlc_1[16];
183
    VLC ac_vlc_2[16];
184
    VLC ac_vlc_3[16];
185
    VLC ac_vlc_4[16];
186

    
187
    VLC superblock_run_length_vlc;
188
    VLC fragment_run_length_vlc;
189
    VLC mode_code_vlc;
190
    VLC motion_vector_vlc;
191

    
192
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
193
     * index into them */
194
    DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]);        //<qmat[is_inter][plane]
195

    
196
    /* This table contains superblock_count * 16 entries. Each set of 16
197
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
198
     * An entry will be -1 to indicate that no entry corresponds to that
199
     * index. */
200
    int *superblock_fragments;
201

    
202
    /* This table contains superblock_count * 4 entries. Each set of 4
203
     * numbers corresponds to the macroblock indexes 0..3 of the superblock.
204
     * An entry will be -1 to indicate that no entry corresponds to that
205
     * index. */
206
    int *superblock_macroblocks;
207

    
208
    /* This table contains macroblock_count * 6 entries. Each set of 6
209
     * numbers corresponds to the fragment indexes 0..5 which comprise
210
     * the macroblock (4 Y fragments and 2 C fragments). */
211
    int *macroblock_fragments;
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 *hilbert = NULL;
258
    int mapping_index = 0;
259

    
260
    int current_macroblock;
261
    int c_fragment;
262

    
263
    signed char travel_width[16] = {
264
         1,  1,  0, -1,
265
         0,  0,  1,  0,
266
         1,  0,  1,  0,
267
         0, -1,  0,  1
268
    };
269

    
270
    signed char travel_height[16] = {
271
         0,  0,  1,  0,
272
         1,  1,  0, -1,
273
         0,  1,  0, -1,
274
        -1,  0, -1,  0
275
    };
276

    
277
    signed char travel_width_mb[4] = {
278
         1,  0,  1,  0
279
    };
280

    
281
    signed char travel_height_mb[4] = {
282
         0,  1,  0, -1
283
    };
284

    
285
    hilbert_walk_mb[0] = 1;
286
    hilbert_walk_mb[1] = s->macroblock_width;
287
    hilbert_walk_mb[2] = 1;
288
    hilbert_walk_mb[3] = -s->macroblock_width;
289

    
290
    /* iterate through each superblock (all planes) and map the fragments */
291
    for (i = 0; i < s->superblock_count; i++) {
292
        /* time to re-assign the limits? */
293
        if (i == 0) {
294

    
295
            /* start of Y superblocks */
296
            right_edge = s->fragment_width;
297
            bottom_edge = s->fragment_height;
298
            current_width = -1;
299
            current_height = 0;
300
            superblock_row_inc = 3 * s->fragment_width -
301
                (s->y_superblock_width * 4 - s->fragment_width);
302

    
303
            /* the first operation for this variable is to advance by 1 */
304
            current_fragment = -1;
305

    
306
        } else if (i == s->u_superblock_start) {
307

    
308
            /* start of U superblocks */
309
            right_edge = s->fragment_width / 2;
310
            bottom_edge = s->fragment_height / 2;
311
            current_width = -1;
312
            current_height = 0;
313
            superblock_row_inc = 3 * (s->fragment_width / 2) -
314
                (s->c_superblock_width * 4 - s->fragment_width / 2);
315

    
316
            /* the first operation for this variable is to advance by 1 */
317
            current_fragment = s->fragment_start[1] - 1;
318

    
319
        } else if (i == s->v_superblock_start) {
320

    
321
            /* start of V superblocks */
322
            right_edge = s->fragment_width / 2;
323
            bottom_edge = s->fragment_height / 2;
324
            current_width = -1;
325
            current_height = 0;
326
            superblock_row_inc = 3 * (s->fragment_width / 2) -
327
                (s->c_superblock_width * 4 - s->fragment_width / 2);
328

    
329
            /* the first operation for this variable is to advance by 1 */
330
            current_fragment = s->fragment_start[2] - 1;
331

    
332
        }
333

    
334
        if (current_width >= right_edge - 1) {
335
            /* reset width and move to next superblock row */
336
            current_width = -1;
337
            current_height += 4;
338

    
339
            /* fragment is now at the start of a new superblock row */
340
            current_fragment += superblock_row_inc;
341
        }
342

    
343
        /* iterate through all 16 fragments in a superblock */
344
        for (j = 0; j < 16; j++) {
345
            current_fragment += travel_width[j] + right_edge * travel_height[j];
346
            current_width += travel_width[j];
347
            current_height += travel_height[j];
348

    
349
            /* check if the fragment is in bounds */
350
            if ((current_width < right_edge) &&
351
                (current_height < bottom_edge)) {
352
                s->superblock_fragments[mapping_index] = current_fragment;
353
            } else {
354
                s->superblock_fragments[mapping_index] = -1;
355
            }
356

    
357
            mapping_index++;
358
        }
359
    }
360

    
361
    /* initialize the superblock <-> macroblock mapping; iterate through
362
     * all of the Y plane superblocks to build this mapping */
363
    right_edge = s->macroblock_width;
364
    bottom_edge = s->macroblock_height;
365
    current_width = -1;
366
    current_height = 0;
367
    superblock_row_inc = s->macroblock_width -
368
        (s->y_superblock_width * 2 - s->macroblock_width);
369
    hilbert = hilbert_walk_mb;
370
    mapping_index = 0;
371
    current_macroblock = -1;
372
    for (i = 0; i < s->u_superblock_start; i++) {
373

    
374
        if (current_width >= right_edge - 1) {
375
            /* reset width and move to next superblock row */
376
            current_width = -1;
377
            current_height += 2;
378

    
379
            /* macroblock is now at the start of a new superblock row */
380
            current_macroblock += superblock_row_inc;
381
        }
382

    
383
        /* iterate through each potential macroblock in the superblock */
384
        for (j = 0; j < 4; j++) {
385
            current_macroblock += hilbert_walk_mb[j];
386
            current_width += travel_width_mb[j];
387
            current_height += travel_height_mb[j];
388

    
389
            /* check if the macroblock is in bounds */
390
            if ((current_width < right_edge) &&
391
                (current_height < bottom_edge)) {
392
                s->superblock_macroblocks[mapping_index] = current_macroblock;
393
            } else {
394
                s->superblock_macroblocks[mapping_index] = -1;
395
            }
396

    
397
            mapping_index++;
398
        }
399
    }
400

    
401
    /* initialize the macroblock <-> fragment mapping */
402
    current_fragment = 0;
403
    current_macroblock = 0;
404
    mapping_index = 0;
405
    for (i = 0; i < s->fragment_height; i += 2) {
406

    
407
        for (j = 0; j < s->fragment_width; j += 2) {
408

    
409
            s->all_fragments[current_fragment].macroblock = current_macroblock;
410
            s->macroblock_fragments[mapping_index++] = current_fragment;
411

    
412
            if (j + 1 < s->fragment_width) {
413
                s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
414
                s->macroblock_fragments[mapping_index++] = current_fragment + 1;
415
            } else
416
                s->macroblock_fragments[mapping_index++] = -1;
417

    
418
            if (i + 1 < s->fragment_height) {
419
                s->all_fragments[current_fragment + s->fragment_width].macroblock =
420
                    current_macroblock;
421
                s->macroblock_fragments[mapping_index++] =
422
                    current_fragment + s->fragment_width;
423
            } else
424
                s->macroblock_fragments[mapping_index++] = -1;
425

    
426
            if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
427
                s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
428
                    current_macroblock;
429
                s->macroblock_fragments[mapping_index++] =
430
                    current_fragment + s->fragment_width + 1;
431
            } else
432
                s->macroblock_fragments[mapping_index++] = -1;
433

    
434
            /* C planes */
435
            c_fragment = s->fragment_start[1] +
436
                (i * s->fragment_width / 4) + (j / 2);
437
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
438
            s->macroblock_fragments[mapping_index++] = c_fragment;
439

    
440
            c_fragment = s->fragment_start[2] +
441
                (i * s->fragment_width / 4) + (j / 2);
442
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
443
            s->macroblock_fragments[mapping_index++] = c_fragment;
444

    
445
            if (j + 2 <= s->fragment_width)
446
                current_fragment += 2;
447
            else
448
                current_fragment++;
449
            current_macroblock++;
450
        }
451

    
452
        current_fragment += s->fragment_width;
453
    }
454

    
455
    return 0;  /* successful path out */
456
}
457

    
458
/*
459
 * This function wipes out all of the fragment data.
460
 */
461
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
462
{
463
    int i;
464

    
465
    /* zero out all of the fragment information */
466
    s->coded_fragment_list_index = 0;
467
    for (i = 0; i < s->fragment_count; i++) {
468
        s->coeff_counts[i] = 0;
469
        s->all_fragments[i].motion_x = 127;
470
        s->all_fragments[i].motion_y = 127;
471
        s->all_fragments[i].next_coeff= NULL;
472
        s->coeffs[i].index=
473
        s->coeffs[i].coeff=0;
474
        s->coeffs[i].next= NULL;
475
    }
476
}
477

    
478
/*
479
 * This function sets up the dequantization tables used for a particular
480
 * frame.
481
 */
482
static void init_dequantizer(Vp3DecodeContext *s)
483
{
484
    int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
485
    int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
486
    int i, plane, inter, qri, bmi, bmj, qistart;
487

    
488
    for(inter=0; inter<2; inter++){
489
        for(plane=0; plane<3; plane++){
490
            int sum=0;
491
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
492
                sum+= s->qr_size[inter][plane][qri];
493
                if(s->quality_index <= sum)
494
                    break;
495
            }
496
            qistart= sum - s->qr_size[inter][plane][qri];
497
            bmi= s->qr_base[inter][plane][qri  ];
498
            bmj= s->qr_base[inter][plane][qri+1];
499
            for(i=0; i<64; i++){
500
                int coeff= (  2*(sum    -s->quality_index)*s->base_matrix[bmi][i]
501
                            - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
502
                            + s->qr_size[inter][plane][qri])
503
                           / (2*s->qr_size[inter][plane][qri]);
504

    
505
                int qmin= 8<<(inter + !i);
506
                int qscale= i ? ac_scale_factor : dc_scale_factor;
507

    
508
                s->qmat[inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
509
            }
510
        }
511
    }
512

    
513
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
514
}
515

    
516
/*
517
 * This function initializes the loop filter boundary limits if the frame's
518
 * quality index is different from the previous frame's.
519
 */
520
static void init_loop_filter(Vp3DecodeContext *s)
521
{
522
    int *bounding_values= s->bounding_values_array+127;
523
    int filter_limit;
524
    int x;
525

    
526
    filter_limit = s->filter_limit_values[s->quality_index];
527

    
528
    /* set up the bounding values */
529
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
530
    for (x = 0; x < filter_limit; x++) {
531
        bounding_values[-x - filter_limit] = -filter_limit + x;
532
        bounding_values[-x] = -x;
533
        bounding_values[x] = x;
534
        bounding_values[x + filter_limit] = filter_limit - x;
535
    }
536
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
537
}
538

    
539
/*
540
 * This function unpacks all of the superblock/macroblock/fragment coding
541
 * information from the bitstream.
542
 */
543
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
544
{
545
    int bit = 0;
546
    int current_superblock = 0;
547
    int current_run = 0;
548
    int decode_fully_flags = 0;
549
    int decode_partial_blocks = 0;
550
    int first_c_fragment_seen;
551

    
552
    int i, j;
553
    int current_fragment;
554

    
555
    if (s->keyframe) {
556
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
557

    
558
    } else {
559

    
560
        /* unpack the list of partially-coded superblocks */
561
        bit = get_bits1(gb);
562
        /* toggle the bit because as soon as the first run length is
563
         * fetched the bit will be toggled again */
564
        bit ^= 1;
565
        while (current_superblock < s->superblock_count) {
566
            if (current_run-- == 0) {
567
                bit ^= 1;
568
                current_run = get_vlc2(gb,
569
                    s->superblock_run_length_vlc.table, 6, 2);
570
                if (current_run == 33)
571
                    current_run += get_bits(gb, 12);
572

    
573
                /* if any of the superblocks are not partially coded, flag
574
                 * a boolean to decode the list of fully-coded superblocks */
575
                if (bit == 0) {
576
                    decode_fully_flags = 1;
577
                } else {
578

    
579
                    /* make a note of the fact that there are partially coded
580
                     * superblocks */
581
                    decode_partial_blocks = 1;
582
                }
583
            }
584
            s->superblock_coding[current_superblock++] = bit;
585
        }
586

    
587
        /* unpack the list of fully coded superblocks if any of the blocks were
588
         * not marked as partially coded in the previous step */
589
        if (decode_fully_flags) {
590

    
591
            current_superblock = 0;
592
            current_run = 0;
593
            bit = get_bits1(gb);
594
            /* toggle the bit because as soon as the first run length is
595
             * fetched the bit will be toggled again */
596
            bit ^= 1;
597
            while (current_superblock < s->superblock_count) {
598

    
599
                /* skip any superblocks already marked as partially coded */
600
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
601

    
602
                    if (current_run-- == 0) {
603
                        bit ^= 1;
604
                        current_run = get_vlc2(gb,
605
                            s->superblock_run_length_vlc.table, 6, 2);
606
                        if (current_run == 33)
607
                            current_run += get_bits(gb, 12);
608
                    }
609
                    s->superblock_coding[current_superblock] = 2*bit;
610
                }
611
                current_superblock++;
612
            }
613
        }
614

    
615
        /* if there were partial blocks, initialize bitstream for
616
         * unpacking fragment codings */
617
        if (decode_partial_blocks) {
618

    
619
            current_run = 0;
620
            bit = get_bits1(gb);
621
            /* toggle the bit because as soon as the first run length is
622
             * fetched the bit will be toggled again */
623
            bit ^= 1;
624
        }
625
    }
626

    
627
    /* figure out which fragments are coded; iterate through each
628
     * superblock (all planes) */
629
    s->coded_fragment_list_index = 0;
630
    s->next_coeff= s->coeffs + s->fragment_count;
631
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
632
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
633
    first_c_fragment_seen = 0;
634
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
635
    for (i = 0; i < s->superblock_count; i++) {
636

    
637
        /* iterate through all 16 fragments in a superblock */
638
        for (j = 0; j < 16; j++) {
639

    
640
            /* if the fragment is in bounds, check its coding status */
641
            current_fragment = s->superblock_fragments[i * 16 + j];
642
            if (current_fragment >= s->fragment_count) {
643
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
644
                    current_fragment, s->fragment_count);
645
                return 1;
646
            }
647
            if (current_fragment != -1) {
648
                if (s->superblock_coding[i] == SB_NOT_CODED) {
649

    
650
                    /* copy all the fragments from the prior frame */
651
                    s->all_fragments[current_fragment].coding_method =
652
                        MODE_COPY;
653

    
654
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
655

    
656
                    /* fragment may or may not be coded; this is the case
657
                     * that cares about the fragment coding runs */
658
                    if (current_run-- == 0) {
659
                        bit ^= 1;
660
                        current_run = get_vlc2(gb,
661
                            s->fragment_run_length_vlc.table, 5, 2);
662
                    }
663

    
664
                    if (bit) {
665
                        /* default mode; actual mode will be decoded in
666
                         * the next phase */
667
                        s->all_fragments[current_fragment].coding_method =
668
                            MODE_INTER_NO_MV;
669
                        s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
670
                        s->coded_fragment_list[s->coded_fragment_list_index] =
671
                            current_fragment;
672
                        if ((current_fragment >= s->fragment_start[1]) &&
673
                            (s->last_coded_y_fragment == -1) &&
674
                            (!first_c_fragment_seen)) {
675
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
676
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
677
                            first_c_fragment_seen = 1;
678
                        }
679
                        s->coded_fragment_list_index++;
680
                        s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
681
                    } else {
682
                        /* not coded; copy this fragment from the prior frame */
683
                        s->all_fragments[current_fragment].coding_method =
684
                            MODE_COPY;
685
                    }
686

    
687
                } else {
688

    
689
                    /* fragments are fully coded in this superblock; actual
690
                     * coding will be determined in next step */
691
                    s->all_fragments[current_fragment].coding_method =
692
                        MODE_INTER_NO_MV;
693
                    s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
694
                    s->coded_fragment_list[s->coded_fragment_list_index] =
695
                        current_fragment;
696
                    if ((current_fragment >= s->fragment_start[1]) &&
697
                        (s->last_coded_y_fragment == -1) &&
698
                        (!first_c_fragment_seen)) {
699
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
700
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
701
                        first_c_fragment_seen = 1;
702
                    }
703
                    s->coded_fragment_list_index++;
704
                    s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
705
                }
706
            }
707
        }
708
    }
709

    
710
    if (!first_c_fragment_seen)
711
        /* only Y fragments coded in this frame */
712
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
713
    else
714
        /* end the list of coded C fragments */
715
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
716

    
717
    return 0;
718
}
719

    
720
/*
721
 * This function unpacks all the coding mode data for individual macroblocks
722
 * from the bitstream.
723
 */
724
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
725
{
726
    int i, j, k;
727
    int scheme;
728
    int current_macroblock;
729
    int current_fragment;
730
    int coding_mode;
731
    int custom_mode_alphabet[CODING_MODE_COUNT];
732

    
733
    if (s->keyframe) {
734
        for (i = 0; i < s->fragment_count; i++)
735
            s->all_fragments[i].coding_method = MODE_INTRA;
736

    
737
    } else {
738

    
739
        /* fetch the mode coding scheme for this frame */
740
        scheme = get_bits(gb, 3);
741

    
742
        /* is it a custom coding scheme? */
743
        if (scheme == 0) {
744
            for (i = 0; i < 8; i++)
745
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
746
            for (i = 0; i < 8; i++)
747
                custom_mode_alphabet[get_bits(gb, 3)] = i;
748
        }
749

    
750
        /* iterate through all of the macroblocks that contain 1 or more
751
         * coded fragments */
752
        for (i = 0; i < s->u_superblock_start; i++) {
753

    
754
            for (j = 0; j < 4; j++) {
755
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
756
                if ((current_macroblock == -1) ||
757
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
758
                    continue;
759
                if (current_macroblock >= s->macroblock_count) {
760
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
761
                        current_macroblock, s->macroblock_count);
762
                    return 1;
763
                }
764

    
765
                /* mode 7 means get 3 bits for each coding mode */
766
                if (scheme == 7)
767
                    coding_mode = get_bits(gb, 3);
768
                else if(scheme == 0)
769
                    coding_mode = custom_mode_alphabet
770
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
771
                else
772
                    coding_mode = ModeAlphabet[scheme-1]
773
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
774

    
775
                s->macroblock_coding[current_macroblock] = coding_mode;
776
                for (k = 0; k < 6; k++) {
777
                    current_fragment =
778
                        s->macroblock_fragments[current_macroblock * 6 + k];
779
                    if (current_fragment == -1)
780
                        continue;
781
                    if (current_fragment >= s->fragment_count) {
782
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
783
                            current_fragment, s->fragment_count);
784
                        return 1;
785
                    }
786
                    if (s->all_fragments[current_fragment].coding_method !=
787
                        MODE_COPY)
788
                        s->all_fragments[current_fragment].coding_method =
789
                            coding_mode;
790
                }
791
            }
792
        }
793
    }
794

    
795
    return 0;
796
}
797

    
798
/*
799
 * This function unpacks all the motion vectors for the individual
800
 * macroblocks from the bitstream.
801
 */
802
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
803
{
804
    int i, j, k, l;
805
    int coding_mode;
806
    int motion_x[6];
807
    int motion_y[6];
808
    int last_motion_x = 0;
809
    int last_motion_y = 0;
810
    int prior_last_motion_x = 0;
811
    int prior_last_motion_y = 0;
812
    int current_macroblock;
813
    int current_fragment;
814

    
815
    if (s->keyframe)
816
        return 0;
817

    
818
    memset(motion_x, 0, 6 * sizeof(int));
819
    memset(motion_y, 0, 6 * sizeof(int));
820

    
821
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
822
    coding_mode = get_bits1(gb);
823

    
824
    /* iterate through all of the macroblocks that contain 1 or more
825
     * coded fragments */
826
    for (i = 0; i < s->u_superblock_start; i++) {
827

    
828
        for (j = 0; j < 4; j++) {
829
            current_macroblock = s->superblock_macroblocks[i * 4 + j];
830
            if ((current_macroblock == -1) ||
831
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
832
                continue;
833
            if (current_macroblock >= s->macroblock_count) {
834
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
835
                    current_macroblock, s->macroblock_count);
836
                return 1;
837
            }
838

    
839
            current_fragment = s->macroblock_fragments[current_macroblock * 6];
840
            if (current_fragment >= s->fragment_count) {
841
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
842
                    current_fragment, s->fragment_count);
843
                return 1;
844
            }
845
            switch (s->macroblock_coding[current_macroblock]) {
846

    
847
            case MODE_INTER_PLUS_MV:
848
            case MODE_GOLDEN_MV:
849
                /* all 6 fragments use the same motion vector */
850
                if (coding_mode == 0) {
851
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
852
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
853
                } else {
854
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
855
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
856
                }
857

    
858
                for (k = 1; k < 6; k++) {
859
                    motion_x[k] = motion_x[0];
860
                    motion_y[k] = motion_y[0];
861
                }
862

    
863
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
864
                if (s->macroblock_coding[current_macroblock] ==
865
                    MODE_INTER_PLUS_MV) {
866
                    prior_last_motion_x = last_motion_x;
867
                    prior_last_motion_y = last_motion_y;
868
                    last_motion_x = motion_x[0];
869
                    last_motion_y = motion_y[0];
870
                }
871
                break;
872

    
873
            case MODE_INTER_FOURMV:
874
                /* vector maintenance */
875
                prior_last_motion_x = last_motion_x;
876
                prior_last_motion_y = last_motion_y;
877

    
878
                /* fetch 4 vectors from the bitstream, one for each
879
                 * Y fragment, then average for the C fragment vectors */
880
                motion_x[4] = motion_y[4] = 0;
881
                for (k = 0; k < 4; k++) {
882
                    for (l = 0; l < s->coded_fragment_list_index; l++)
883
                        if (s->coded_fragment_list[l] == s->macroblock_fragments[6*current_macroblock + k])
884
                            break;
885
                    if (l < s->coded_fragment_list_index) {
886
                        if (coding_mode == 0) {
887
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
888
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
889
                        } else {
890
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
891
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
892
                        }
893
                        last_motion_x = motion_x[k];
894
                        last_motion_y = motion_y[k];
895
                    } else {
896
                        motion_x[k] = 0;
897
                        motion_y[k] = 0;
898
                    }
899
                    motion_x[4] += motion_x[k];
900
                    motion_y[4] += motion_y[k];
901
                }
902

    
903
                motion_x[5]=
904
                motion_x[4]= RSHIFT(motion_x[4], 2);
905
                motion_y[5]=
906
                motion_y[4]= RSHIFT(motion_y[4], 2);
907
                break;
908

    
909
            case MODE_INTER_LAST_MV:
910
                /* all 6 fragments use the last motion vector */
911
                motion_x[0] = last_motion_x;
912
                motion_y[0] = last_motion_y;
913
                for (k = 1; k < 6; k++) {
914
                    motion_x[k] = motion_x[0];
915
                    motion_y[k] = motion_y[0];
916
                }
917

    
918
                /* no vector maintenance (last vector remains the
919
                 * last vector) */
920
                break;
921

    
922
            case MODE_INTER_PRIOR_LAST:
923
                /* all 6 fragments use the motion vector prior to the
924
                 * last motion vector */
925
                motion_x[0] = prior_last_motion_x;
926
                motion_y[0] = prior_last_motion_y;
927
                for (k = 1; k < 6; k++) {
928
                    motion_x[k] = motion_x[0];
929
                    motion_y[k] = motion_y[0];
930
                }
931

    
932
                /* vector maintenance */
933
                prior_last_motion_x = last_motion_x;
934
                prior_last_motion_y = last_motion_y;
935
                last_motion_x = motion_x[0];
936
                last_motion_y = motion_y[0];
937
                break;
938

    
939
            default:
940
                /* covers intra, inter without MV, golden without MV */
941
                memset(motion_x, 0, 6 * sizeof(int));
942
                memset(motion_y, 0, 6 * sizeof(int));
943

    
944
                /* no vector maintenance */
945
                break;
946
            }
947

    
948
            /* assign the motion vectors to the correct fragments */
949
            for (k = 0; k < 6; k++) {
950
                current_fragment =
951
                    s->macroblock_fragments[current_macroblock * 6 + k];
952
                if (current_fragment == -1)
953
                    continue;
954
                if (current_fragment >= s->fragment_count) {
955
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
956
                        current_fragment, s->fragment_count);
957
                    return 1;
958
                }
959
                s->all_fragments[current_fragment].motion_x = motion_x[k];
960
                s->all_fragments[current_fragment].motion_y = motion_y[k];
961
            }
962
        }
963
    }
964

    
965
    return 0;
966
}
967

    
968
/*
969
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
970
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
971
 * data. This function unpacks all the VLCs for either the Y plane or both
972
 * C planes, and is called for DC coefficients or different AC coefficient
973
 * levels (since different coefficient types require different VLC tables.
974
 *
975
 * This function returns a residual eob run. E.g, if a particular token gave
976
 * instructions to EOB the next 5 fragments and there were only 2 fragments
977
 * left in the current fragment range, 3 would be returned so that it could
978
 * be passed into the next call to this same function.
979
 */
980
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
981
                        VLC *table, int coeff_index,
982
                        int first_fragment, int last_fragment,
983
                        int eob_run)
984
{
985
    int i;
986
    int token;
987
    int zero_run = 0;
988
    DCTELEM coeff = 0;
989
    Vp3Fragment *fragment;
990
    uint8_t *perm= s->scantable.permutated;
991
    int bits_to_get;
992

    
993
    if ((first_fragment >= s->fragment_count) ||
994
        (last_fragment >= s->fragment_count)) {
995

    
996
        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
997
            first_fragment, last_fragment);
998
        return 0;
999
    }
1000

    
1001
    for (i = first_fragment; i <= last_fragment; i++) {
1002
        int fragment_num = s->coded_fragment_list[i];
1003

    
1004
        if (s->coeff_counts[fragment_num] > coeff_index)
1005
            continue;
1006
        fragment = &s->all_fragments[fragment_num];
1007

    
1008
        if (!eob_run) {
1009
            /* decode a VLC into a token */
1010
            token = get_vlc2(gb, table->table, 5, 3);
1011
            /* use the token to get a zero run, a coefficient, and an eob run */
1012
            if (token <= 6) {
1013
                eob_run = eob_run_base[token];
1014
                if (eob_run_get_bits[token])
1015
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
1016
                coeff = zero_run = 0;
1017
            } else {
1018
                bits_to_get = coeff_get_bits[token];
1019
                if (!bits_to_get)
1020
                    coeff = coeff_tables[token][0];
1021
                else
1022
                    coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1023

    
1024
                zero_run = zero_run_base[token];
1025
                if (zero_run_get_bits[token])
1026
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
1027
            }
1028
        }
1029

    
1030
        if (!eob_run) {
1031
            s->coeff_counts[fragment_num] += zero_run;
1032
            if (s->coeff_counts[fragment_num] < 64){
1033
                fragment->next_coeff->coeff= coeff;
1034
                fragment->next_coeff->index= perm[s->coeff_counts[fragment_num]++]; //FIXME perm here already?
1035
                fragment->next_coeff->next= s->next_coeff;
1036
                s->next_coeff->next=NULL;
1037
                fragment->next_coeff= s->next_coeff++;
1038
            }
1039
        } else {
1040
            s->coeff_counts[fragment_num] |= 128;
1041
            eob_run--;
1042
        }
1043
    }
1044

    
1045
    return eob_run;
1046
}
1047

    
1048
/*
1049
 * This function unpacks all of the DCT coefficient data from the
1050
 * bitstream.
1051
 */
1052
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1053
{
1054
    int i;
1055
    int dc_y_table;
1056
    int dc_c_table;
1057
    int ac_y_table;
1058
    int ac_c_table;
1059
    int residual_eob_run = 0;
1060

    
1061
    /* fetch the DC table indexes */
1062
    dc_y_table = get_bits(gb, 4);
1063
    dc_c_table = get_bits(gb, 4);
1064

    
1065
    /* unpack the Y plane DC coefficients */
1066
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1067
        s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1068

    
1069
    /* unpack the C plane DC coefficients */
1070
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1071
        s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1072

    
1073
    /* fetch the AC table indexes */
1074
    ac_y_table = get_bits(gb, 4);
1075
    ac_c_table = get_bits(gb, 4);
1076

    
1077
    /* unpack the group 1 AC coefficients (coeffs 1-5) */
1078
    for (i = 1; i <= 5; i++) {
1079
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1080
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1081

    
1082
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1083
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1084
    }
1085

    
1086
    /* unpack the group 2 AC coefficients (coeffs 6-14) */
1087
    for (i = 6; i <= 14; i++) {
1088
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1089
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1090

    
1091
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1092
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1093
    }
1094

    
1095
    /* unpack the group 3 AC coefficients (coeffs 15-27) */
1096
    for (i = 15; i <= 27; i++) {
1097
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1098
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1099

    
1100
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1101
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1102
    }
1103

    
1104
    /* unpack the group 4 AC coefficients (coeffs 28-63) */
1105
    for (i = 28; i <= 63; i++) {
1106
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1107
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1108

    
1109
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1110
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1111
    }
1112

    
1113
    return 0;
1114
}
1115

    
1116
/*
1117
 * This function reverses the DC prediction for each coded fragment in
1118
 * the frame. Much of this function is adapted directly from the original
1119
 * VP3 source code.
1120
 */
1121
#define COMPATIBLE_FRAME(x) \
1122
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1123
#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1124
#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1125

    
1126
static void reverse_dc_prediction(Vp3DecodeContext *s,
1127
                                  int first_fragment,
1128
                                  int fragment_width,
1129
                                  int fragment_height)
1130
{
1131

    
1132
#define PUL 8
1133
#define PU 4
1134
#define PUR 2
1135
#define PL 1
1136

    
1137
    int x, y;
1138
    int i = first_fragment;
1139

    
1140
    int predicted_dc;
1141

    
1142
    /* DC values for the left, up-left, up, and up-right fragments */
1143
    int vl, vul, vu, vur;
1144

    
1145
    /* indexes for the left, up-left, up, and up-right fragments */
1146
    int l, ul, u, ur;
1147

    
1148
    /*
1149
     * The 6 fields mean:
1150
     *   0: up-left multiplier
1151
     *   1: up multiplier
1152
     *   2: up-right multiplier
1153
     *   3: left multiplier
1154
     */
1155
    int predictor_transform[16][4] = {
1156
        {  0,  0,  0,  0},
1157
        {  0,  0,  0,128},        // PL
1158
        {  0,  0,128,  0},        // PUR
1159
        {  0,  0, 53, 75},        // PUR|PL
1160
        {  0,128,  0,  0},        // PU
1161
        {  0, 64,  0, 64},        // PU|PL
1162
        {  0,128,  0,  0},        // PU|PUR
1163
        {  0,  0, 53, 75},        // PU|PUR|PL
1164
        {128,  0,  0,  0},        // PUL
1165
        {  0,  0,  0,128},        // PUL|PL
1166
        { 64,  0, 64,  0},        // PUL|PUR
1167
        {  0,  0, 53, 75},        // PUL|PUR|PL
1168
        {  0,128,  0,  0},        // PUL|PU
1169
       {-104,116,  0,116},        // PUL|PU|PL
1170
        { 24, 80, 24,  0},        // PUL|PU|PUR
1171
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1172
    };
1173

    
1174
    /* This table shows which types of blocks can use other blocks for
1175
     * prediction. For example, INTRA is the only mode in this table to
1176
     * have a frame number of 0. That means INTRA blocks can only predict
1177
     * from other INTRA blocks. There are 2 golden frame coding types;
1178
     * blocks encoding in these modes can only predict from other blocks
1179
     * that were encoded with these 1 of these 2 modes. */
1180
    unsigned char compatible_frame[8] = {
1181
        1,    /* MODE_INTER_NO_MV */
1182
        0,    /* MODE_INTRA */
1183
        1,    /* MODE_INTER_PLUS_MV */
1184
        1,    /* MODE_INTER_LAST_MV */
1185
        1,    /* MODE_INTER_PRIOR_MV */
1186
        2,    /* MODE_USING_GOLDEN */
1187
        2,    /* MODE_GOLDEN_MV */
1188
        1     /* MODE_INTER_FOUR_MV */
1189
    };
1190
    int current_frame_type;
1191

    
1192
    /* there is a last DC predictor for each of the 3 frame types */
1193
    short last_dc[3];
1194

    
1195
    int transform = 0;
1196

    
1197
    vul = vu = vur = vl = 0;
1198
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1199

    
1200
    /* for each fragment row... */
1201
    for (y = 0; y < fragment_height; y++) {
1202

    
1203
        /* for each fragment in a row... */
1204
        for (x = 0; x < fragment_width; x++, i++) {
1205

    
1206
            /* reverse prediction if this block was coded */
1207
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1208

    
1209
                current_frame_type =
1210
                    compatible_frame[s->all_fragments[i].coding_method];
1211

    
1212
                transform= 0;
1213
                if(x){
1214
                    l= i-1;
1215
                    vl = DC_COEFF(l);
1216
                    if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1217
                        transform |= PL;
1218
                }
1219
                if(y){
1220
                    u= i-fragment_width;
1221
                    vu = DC_COEFF(u);
1222
                    if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1223
                        transform |= PU;
1224
                    if(x){
1225
                        ul= i-fragment_width-1;
1226
                        vul = DC_COEFF(ul);
1227
                        if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1228
                            transform |= PUL;
1229
                    }
1230
                    if(x + 1 < fragment_width){
1231
                        ur= i-fragment_width+1;
1232
                        vur = DC_COEFF(ur);
1233
                        if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1234
                            transform |= PUR;
1235
                    }
1236
                }
1237

    
1238
                if (transform == 0) {
1239

    
1240
                    /* if there were no fragments to predict from, use last
1241
                     * DC saved */
1242
                    predicted_dc = last_dc[current_frame_type];
1243
                } else {
1244

    
1245
                    /* apply the appropriate predictor transform */
1246
                    predicted_dc =
1247
                        (predictor_transform[transform][0] * vul) +
1248
                        (predictor_transform[transform][1] * vu) +
1249
                        (predictor_transform[transform][2] * vur) +
1250
                        (predictor_transform[transform][3] * vl);
1251

    
1252
                    predicted_dc /= 128;
1253

    
1254
                    /* check for outranging on the [ul u l] and
1255
                     * [ul u ur l] predictors */
1256
                    if ((transform == 13) || (transform == 15)) {
1257
                        if (FFABS(predicted_dc - vu) > 128)
1258
                            predicted_dc = vu;
1259
                        else if (FFABS(predicted_dc - vl) > 128)
1260
                            predicted_dc = vl;
1261
                        else if (FFABS(predicted_dc - vul) > 128)
1262
                            predicted_dc = vul;
1263
                    }
1264
                }
1265

    
1266
                /* at long last, apply the predictor */
1267
                if(s->coeffs[i].index){
1268
                    *s->next_coeff= s->coeffs[i];
1269
                    s->coeffs[i].index=0;
1270
                    s->coeffs[i].coeff=0;
1271
                    s->coeffs[i].next= s->next_coeff++;
1272
                }
1273
                s->coeffs[i].coeff += predicted_dc;
1274
                /* save the DC */
1275
                last_dc[current_frame_type] = DC_COEFF(i);
1276
                if(DC_COEFF(i) && !(s->coeff_counts[i]&127)){
1277
                    s->coeff_counts[i]= 129;
1278
//                    s->all_fragments[i].next_coeff= s->next_coeff;
1279
                    s->coeffs[i].next= s->next_coeff;
1280
                    (s->next_coeff++)->next=NULL;
1281
                }
1282
            }
1283
        }
1284
    }
1285
}
1286

    
1287
/*
1288
 * Perform the final rendering for a particular slice of data.
1289
 * The slice number ranges from 0..(macroblock_height - 1).
1290
 */
1291
static void render_slice(Vp3DecodeContext *s, int slice)
1292
{
1293
    int x;
1294
    int16_t *dequantizer;
1295
    DECLARE_ALIGNED_16(DCTELEM, block[64]);
1296
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1297
    int motion_halfpel_index;
1298
    uint8_t *motion_source;
1299
    int plane;
1300
    int current_macroblock_entry = slice * s->macroblock_width * 6;
1301

    
1302
    if (slice >= s->macroblock_height)
1303
        return;
1304

    
1305
    for (plane = 0; plane < 3; plane++) {
1306
        uint8_t *output_plane = s->current_frame.data    [plane];
1307
        uint8_t *  last_plane = s->   last_frame.data    [plane];
1308
        uint8_t *golden_plane = s-> golden_frame.data    [plane];
1309
        int stride            = s->current_frame.linesize[plane];
1310
        int plane_width       = s->width  >> !!plane;
1311
        int plane_height      = s->height >> !!plane;
1312
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
1313
        int slice_height = y + (FRAGMENT_PIXELS << !plane);
1314
        int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1315

    
1316
        if (!s->flipped_image) stride = -stride;
1317

    
1318

    
1319
        if(FFABS(stride) > 2048)
1320
            return; //various tables are fixed size
1321

    
1322
        /* for each fragment row in the slice (both of them)... */
1323
        for (; y < slice_height; y += 8) {
1324

    
1325
            /* for each fragment in a row... */
1326
            for (x = 0; x < plane_width; x += 8, i++) {
1327

    
1328
                if ((i < 0) || (i >= s->fragment_count)) {
1329
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
1330
                    return;
1331
                }
1332

    
1333
                /* transform if this block was coded */
1334
                if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1335
                    !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1336

    
1337
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1338
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1339
                        motion_source= golden_plane;
1340
                    else
1341
                        motion_source= last_plane;
1342

    
1343
                    motion_source += s->all_fragments[i].first_pixel;
1344
                    motion_halfpel_index = 0;
1345

    
1346
                    /* sort out the motion vector if this fragment is coded
1347
                     * using a motion vector method */
1348
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1349
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1350
                        int src_x, src_y;
1351
                        motion_x = s->all_fragments[i].motion_x;
1352
                        motion_y = s->all_fragments[i].motion_y;
1353
                        if(plane){
1354
                            motion_x= (motion_x>>1) | (motion_x&1);
1355
                            motion_y= (motion_y>>1) | (motion_y&1);
1356
                        }
1357

    
1358
                        src_x= (motion_x>>1) + x;
1359
                        src_y= (motion_y>>1) + y;
1360
                        if ((motion_x == 127) || (motion_y == 127))
1361
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1362

    
1363
                        motion_halfpel_index = motion_x & 0x01;
1364
                        motion_source += (motion_x >> 1);
1365

    
1366
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1367
                        motion_source += ((motion_y >> 1) * stride);
1368

    
1369
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1370
                            uint8_t *temp= s->edge_emu_buffer;
1371
                            if(stride<0) temp -= 9*stride;
1372
                            else temp += 9*stride;
1373

    
1374
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1375
                            motion_source= temp;
1376
                        }
1377
                    }
1378

    
1379

    
1380
                    /* first, take care of copying a block from either the
1381
                     * previous or the golden frame */
1382
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1383
                        /* Note, it is possible to implement all MC cases with
1384
                           put_no_rnd_pixels_l2 which would look more like the
1385
                           VP3 source but this would be slower as
1386
                           put_no_rnd_pixels_tab is better optimzed */
1387
                        if(motion_halfpel_index != 3){
1388
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1389
                                output_plane + s->all_fragments[i].first_pixel,
1390
                                motion_source, stride, 8);
1391
                        }else{
1392
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1393
                            s->dsp.put_no_rnd_pixels_l2[1](
1394
                                output_plane + s->all_fragments[i].first_pixel,
1395
                                motion_source - d,
1396
                                motion_source + stride + 1 + d,
1397
                                stride, 8);
1398
                        }
1399
                        dequantizer = s->qmat[1][plane];
1400
                    }else{
1401
                        dequantizer = s->qmat[0][plane];
1402
                    }
1403

    
1404
                    /* dequantize the DCT coefficients */
1405
                    if(s->avctx->idct_algo==FF_IDCT_VP3){
1406
                        Coeff *coeff= s->coeffs + i;
1407
                        s->dsp.clear_block(block);
1408
                        while(coeff->next){
1409
                            block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1410
                            coeff= coeff->next;
1411
                        }
1412
                    }else{
1413
                        Coeff *coeff= s->coeffs + i;
1414
                        s->dsp.clear_block(block);
1415
                        while(coeff->next){
1416
                            block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1417
                            coeff= coeff->next;
1418
                        }
1419
                    }
1420

    
1421
                    /* invert DCT and place (or add) in final output */
1422

    
1423
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1424
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1425
                            block[0] += 128<<3;
1426
                        s->dsp.idct_put(
1427
                            output_plane + s->all_fragments[i].first_pixel,
1428
                            stride,
1429
                            block);
1430
                    } else {
1431
                        s->dsp.idct_add(
1432
                            output_plane + s->all_fragments[i].first_pixel,
1433
                            stride,
1434
                            block);
1435
                    }
1436
                } else {
1437

    
1438
                    /* copy directly from the previous frame */
1439
                    s->dsp.put_pixels_tab[1][0](
1440
                        output_plane + s->all_fragments[i].first_pixel,
1441
                        last_plane + s->all_fragments[i].first_pixel,
1442
                        stride, 8);
1443

    
1444
                }
1445
#if 0
1446
                /* perform the left edge filter if:
1447
                 *   - the fragment is not on the left column
1448
                 *   - the fragment is coded in this frame
1449
                 *   - the fragment is not coded in this frame but the left
1450
                 *     fragment is coded in this frame (this is done instead
1451
                 *     of a right edge filter when rendering the left fragment
1452
                 *     since this fragment is not available yet) */
1453
                if ((x > 0) &&
1454
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1455
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1456
                      (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1457
                    horizontal_filter(
1458
                        output_plane + s->all_fragments[i].first_pixel + 7*stride,
1459
                        -stride, s->bounding_values_array + 127);
1460
                }
1461

1462
                /* perform the top edge filter if:
1463
                 *   - the fragment is not on the top row
1464
                 *   - the fragment is coded in this frame
1465
                 *   - the fragment is not coded in this frame but the above
1466
                 *     fragment is coded in this frame (this is done instead
1467
                 *     of a bottom edge filter when rendering the above
1468
                 *     fragment since this fragment is not available yet) */
1469
                if ((y > 0) &&
1470
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1471
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1472
                      (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1473
                    vertical_filter(
1474
                        output_plane + s->all_fragments[i].first_pixel - stride,
1475
                        -stride, s->bounding_values_array + 127);
1476
                }
1477
#endif
1478
            }
1479
        }
1480
    }
1481

    
1482
     /* this looks like a good place for slice dispatch... */
1483
     /* algorithm:
1484
      *   if (slice == s->macroblock_height - 1)
1485
      *     dispatch (both last slice & 2nd-to-last slice);
1486
      *   else if (slice > 0)
1487
      *     dispatch (slice - 1);
1488
      */
1489

    
1490
    emms_c();
1491
}
1492

    
1493
static void apply_loop_filter(Vp3DecodeContext *s)
1494
{
1495
    int plane;
1496
    int x, y;
1497
    int *bounding_values= s->bounding_values_array+127;
1498

    
1499
#if 0
1500
    int bounding_values_array[256];
1501
    int filter_limit;
1502

1503
    /* find the right loop limit value */
1504
    for (x = 63; x >= 0; x--) {
1505
        if (vp31_ac_scale_factor[x] >= s->quality_index)
1506
            break;
1507
    }
1508
    filter_limit = vp31_filter_limit_values[s->quality_index];
1509

1510
    /* set up the bounding values */
1511
    memset(bounding_values_array, 0, 256 * sizeof(int));
1512
    for (x = 0; x < filter_limit; x++) {
1513
        bounding_values[-x - filter_limit] = -filter_limit + x;
1514
        bounding_values[-x] = -x;
1515
        bounding_values[x] = x;
1516
        bounding_values[x + filter_limit] = filter_limit - x;
1517
    }
1518
#endif
1519

    
1520
    for (plane = 0; plane < 3; plane++) {
1521
        int width           = s->fragment_width  >> !!plane;
1522
        int height          = s->fragment_height >> !!plane;
1523
        int fragment        = s->fragment_start        [plane];
1524
        int stride          = s->current_frame.linesize[plane];
1525
        uint8_t *plane_data = s->current_frame.data    [plane];
1526
        if (!s->flipped_image) stride = -stride;
1527

    
1528
        for (y = 0; y < height; y++) {
1529

    
1530
            for (x = 0; x < width; x++) {
1531
                /* do not perform left edge filter for left columns frags */
1532
                if ((x > 0) &&
1533
                    (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1534
                    s->dsp.vp3_h_loop_filter(
1535
                        plane_data + s->all_fragments[fragment].first_pixel,
1536
                        stride, bounding_values);
1537
                }
1538

    
1539
                /* do not perform top edge filter for top row fragments */
1540
                if ((y > 0) &&
1541
                    (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1542
                    s->dsp.vp3_v_loop_filter(
1543
                        plane_data + s->all_fragments[fragment].first_pixel,
1544
                        stride, bounding_values);
1545
                }
1546

    
1547
                /* do not perform right edge filter for right column
1548
                 * fragments or if right fragment neighbor is also coded
1549
                 * in this frame (it will be filtered in next iteration) */
1550
                if ((x < width - 1) &&
1551
                    (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1552
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1553
                    s->dsp.vp3_h_loop_filter(
1554
                        plane_data + s->all_fragments[fragment + 1].first_pixel,
1555
                        stride, bounding_values);
1556
                }
1557

    
1558
                /* do not perform bottom edge filter for bottom row
1559
                 * fragments or if bottom fragment neighbor is also coded
1560
                 * in this frame (it will be filtered in the next row) */
1561
                if ((y < height - 1) &&
1562
                    (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1563
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1564
                    s->dsp.vp3_v_loop_filter(
1565
                        plane_data + s->all_fragments[fragment + width].first_pixel,
1566
                        stride, bounding_values);
1567
                }
1568

    
1569
                fragment++;
1570
            }
1571
        }
1572
    }
1573
}
1574

    
1575
/*
1576
 * This function computes the first pixel addresses for each fragment.
1577
 * This function needs to be invoked after the first frame is allocated
1578
 * so that it has access to the plane strides.
1579
 */
1580
static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1581
{
1582
#define Y_INITIAL(chroma_shift)  s->flipped_image ? 1  : s->fragment_height >> chroma_shift
1583
#define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0
1584

    
1585
    int i, x, y;
1586
    const int y_inc = s->flipped_image ? 1 : -1;
1587

    
1588
    /* figure out the first pixel addresses for each of the fragments */
1589
    /* Y plane */
1590
    i = 0;
1591
    for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) {
1592
        for (x = 0; x < s->fragment_width; x++) {
1593
            s->all_fragments[i++].first_pixel =
1594
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1595
                    s->golden_frame.linesize[0] +
1596
                    x * FRAGMENT_PIXELS;
1597
        }
1598
    }
1599

    
1600
    /* U plane */
1601
    i = s->fragment_start[1];
1602
    for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
1603
        for (x = 0; x < s->fragment_width / 2; x++) {
1604
            s->all_fragments[i++].first_pixel =
1605
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1606
                    s->golden_frame.linesize[1] +
1607
                    x * FRAGMENT_PIXELS;
1608
        }
1609
    }
1610

    
1611
    /* V plane */
1612
    i = s->fragment_start[2];
1613
    for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
1614
        for (x = 0; x < s->fragment_width / 2; x++) {
1615
            s->all_fragments[i++].first_pixel =
1616
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1617
                    s->golden_frame.linesize[2] +
1618
                    x * FRAGMENT_PIXELS;
1619
        }
1620
    }
1621
}
1622

    
1623
/*
1624
 * This is the ffmpeg/libavcodec API init function.
1625
 */
1626
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1627
{
1628
    Vp3DecodeContext *s = avctx->priv_data;
1629
    int i, inter, plane;
1630
    int c_width;
1631
    int c_height;
1632
    int y_superblock_count;
1633
    int c_superblock_count;
1634

    
1635
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1636
        s->version = 0;
1637
    else
1638
        s->version = 1;
1639

    
1640
    s->avctx = avctx;
1641
    s->width = (avctx->width + 15) & 0xFFFFFFF0;
1642
    s->height = (avctx->height + 15) & 0xFFFFFFF0;
1643
    avctx->pix_fmt = PIX_FMT_YUV420P;
1644
    if(avctx->idct_algo==FF_IDCT_AUTO)
1645
        avctx->idct_algo=FF_IDCT_VP3;
1646
    dsputil_init(&s->dsp, avctx);
1647

    
1648
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1649

    
1650
    /* initialize to an impossible value which will force a recalculation
1651
     * in the first frame decode */
1652
    s->quality_index = -1;
1653

    
1654
    s->y_superblock_width = (s->width + 31) / 32;
1655
    s->y_superblock_height = (s->height + 31) / 32;
1656
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1657

    
1658
    /* work out the dimensions for the C planes */
1659
    c_width = s->width / 2;
1660
    c_height = s->height / 2;
1661
    s->c_superblock_width = (c_width + 31) / 32;
1662
    s->c_superblock_height = (c_height + 31) / 32;
1663
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1664

    
1665
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1666
    s->u_superblock_start = y_superblock_count;
1667
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1668
    s->superblock_coding = av_malloc(s->superblock_count);
1669

    
1670
    s->macroblock_width = (s->width + 15) / 16;
1671
    s->macroblock_height = (s->height + 15) / 16;
1672
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1673

    
1674
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1675
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1676

    
1677
    /* fragment count covers all 8x8 blocks for all 3 planes */
1678
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1679
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1680
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1681

    
1682
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1683
    s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
1684
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
1685
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
1686
    s->pixel_addresses_initialized = 0;
1687

    
1688
    if (!s->theora_tables)
1689
    {
1690
        for (i = 0; i < 64; i++) {
1691
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1692
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1693
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1694
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1695
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1696
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1697
        }
1698

    
1699
        for(inter=0; inter<2; inter++){
1700
            for(plane=0; plane<3; plane++){
1701
                s->qr_count[inter][plane]= 1;
1702
                s->qr_size [inter][plane][0]= 63;
1703
                s->qr_base [inter][plane][0]=
1704
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1705
            }
1706
        }
1707

    
1708
        /* init VLC tables */
1709
        for (i = 0; i < 16; i++) {
1710

    
1711
            /* DC histograms */
1712
            init_vlc(&s->dc_vlc[i], 5, 32,
1713
                &dc_bias[i][0][1], 4, 2,
1714
                &dc_bias[i][0][0], 4, 2, 0);
1715

    
1716
            /* group 1 AC histograms */
1717
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1718
                &ac_bias_0[i][0][1], 4, 2,
1719
                &ac_bias_0[i][0][0], 4, 2, 0);
1720

    
1721
            /* group 2 AC histograms */
1722
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1723
                &ac_bias_1[i][0][1], 4, 2,
1724
                &ac_bias_1[i][0][0], 4, 2, 0);
1725

    
1726
            /* group 3 AC histograms */
1727
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1728
                &ac_bias_2[i][0][1], 4, 2,
1729
                &ac_bias_2[i][0][0], 4, 2, 0);
1730

    
1731
            /* group 4 AC histograms */
1732
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1733
                &ac_bias_3[i][0][1], 4, 2,
1734
                &ac_bias_3[i][0][0], 4, 2, 0);
1735
        }
1736
    } else {
1737
        for (i = 0; i < 16; i++) {
1738

    
1739
            /* DC histograms */
1740
            init_vlc(&s->dc_vlc[i], 5, 32,
1741
                &s->huffman_table[i][0][1], 4, 2,
1742
                &s->huffman_table[i][0][0], 4, 2, 0);
1743

    
1744
            /* group 1 AC histograms */
1745
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1746
                &s->huffman_table[i+16][0][1], 4, 2,
1747
                &s->huffman_table[i+16][0][0], 4, 2, 0);
1748

    
1749
            /* group 2 AC histograms */
1750
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1751
                &s->huffman_table[i+16*2][0][1], 4, 2,
1752
                &s->huffman_table[i+16*2][0][0], 4, 2, 0);
1753

    
1754
            /* group 3 AC histograms */
1755
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1756
                &s->huffman_table[i+16*3][0][1], 4, 2,
1757
                &s->huffman_table[i+16*3][0][0], 4, 2, 0);
1758

    
1759
            /* group 4 AC histograms */
1760
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1761
                &s->huffman_table[i+16*4][0][1], 4, 2,
1762
                &s->huffman_table[i+16*4][0][0], 4, 2, 0);
1763
        }
1764
    }
1765

    
1766
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1767
        &superblock_run_length_vlc_table[0][1], 4, 2,
1768
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1769

    
1770
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1771
        &fragment_run_length_vlc_table[0][1], 4, 2,
1772
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1773

    
1774
    init_vlc(&s->mode_code_vlc, 3, 8,
1775
        &mode_code_vlc_table[0][1], 2, 1,
1776
        &mode_code_vlc_table[0][0], 2, 1, 0);
1777

    
1778
    init_vlc(&s->motion_vector_vlc, 6, 63,
1779
        &motion_vector_vlc_table[0][1], 2, 1,
1780
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1781

    
1782
    /* work out the block mapping tables */
1783
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1784
    s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
1785
    s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
1786
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1787
    init_block_mapping(s);
1788

    
1789
    for (i = 0; i < 3; i++) {
1790
        s->current_frame.data[i] = NULL;
1791
        s->last_frame.data[i] = NULL;
1792
        s->golden_frame.data[i] = NULL;
1793
    }
1794

    
1795
    return 0;
1796
}
1797

    
1798
/*
1799
 * This is the ffmpeg/libavcodec API frame decode function.
1800
 */
1801
static int vp3_decode_frame(AVCodecContext *avctx,
1802
                            void *data, int *data_size,
1803
                            const uint8_t *buf, int buf_size)
1804
{
1805
    Vp3DecodeContext *s = avctx->priv_data;
1806
    GetBitContext gb;
1807
    static int counter = 0;
1808
    int i;
1809

    
1810
    init_get_bits(&gb, buf, buf_size * 8);
1811

    
1812
    if (s->theora && get_bits1(&gb))
1813
    {
1814
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1815
        return -1;
1816
    }
1817

    
1818
    s->keyframe = !get_bits1(&gb);
1819
    if (!s->theora)
1820
        skip_bits(&gb, 1);
1821
    s->last_quality_index = s->quality_index;
1822

    
1823
    s->nqis=0;
1824
    do{
1825
        s->qis[s->nqis++]= get_bits(&gb, 6);
1826
    } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
1827

    
1828
    s->quality_index= s->qis[0];
1829

    
1830
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1831
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1832
            s->keyframe?"key":"", counter, s->quality_index);
1833
    counter++;
1834

    
1835
    if (s->quality_index != s->last_quality_index) {
1836
        init_dequantizer(s);
1837
        init_loop_filter(s);
1838
    }
1839

    
1840
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1841
        return buf_size;
1842

    
1843
    if (s->keyframe) {
1844
        if (!s->theora)
1845
        {
1846
            skip_bits(&gb, 4); /* width code */
1847
            skip_bits(&gb, 4); /* height code */
1848
            if (s->version)
1849
            {
1850
                s->version = get_bits(&gb, 5);
1851
                if (counter == 1)
1852
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1853
            }
1854
        }
1855
        if (s->version || s->theora)
1856
        {
1857
                if (get_bits1(&gb))
1858
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1859
            skip_bits(&gb, 2); /* reserved? */
1860
        }
1861

    
1862
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
1863
            if (s->golden_frame.data[0])
1864
                avctx->release_buffer(avctx, &s->golden_frame);
1865
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
1866
        } else {
1867
            if (s->golden_frame.data[0])
1868
                avctx->release_buffer(avctx, &s->golden_frame);
1869
            if (s->last_frame.data[0])
1870
                avctx->release_buffer(avctx, &s->last_frame);
1871
        }
1872

    
1873
        s->golden_frame.reference = 3;
1874
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1875
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1876
            return -1;
1877
        }
1878

    
1879
        /* golden frame is also the current frame */
1880
        s->current_frame= s->golden_frame;
1881

    
1882
        /* time to figure out pixel addresses? */
1883
        if (!s->pixel_addresses_initialized)
1884
        {
1885
            vp3_calculate_pixel_addresses(s);
1886
            s->pixel_addresses_initialized = 1;
1887
        }
1888
    } else {
1889
        /* allocate a new current frame */
1890
        s->current_frame.reference = 3;
1891
        if (!s->pixel_addresses_initialized) {
1892
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1893
            return -1;
1894
        }
1895
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
1896
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
1897
            return -1;
1898
        }
1899
    }
1900

    
1901
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1902
    s->current_frame.qstride= 0;
1903

    
1904
    init_frame(s, &gb);
1905

    
1906
    if (unpack_superblocks(s, &gb)){
1907
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1908
        return -1;
1909
    }
1910
    if (unpack_modes(s, &gb)){
1911
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1912
        return -1;
1913
    }
1914
    if (unpack_vectors(s, &gb)){
1915
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1916
        return -1;
1917
    }
1918
    if (unpack_dct_coeffs(s, &gb)){
1919
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1920
        return -1;
1921
    }
1922

    
1923
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1924
    if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
1925
        reverse_dc_prediction(s, s->fragment_start[1],
1926
            s->fragment_width / 2, s->fragment_height / 2);
1927
        reverse_dc_prediction(s, s->fragment_start[2],
1928
            s->fragment_width / 2, s->fragment_height / 2);
1929
    }
1930

    
1931
    for (i = 0; i < s->macroblock_height; i++)
1932
        render_slice(s, i);
1933

    
1934
    apply_loop_filter(s);
1935

    
1936
    *data_size=sizeof(AVFrame);
1937
    *(AVFrame*)data= s->current_frame;
1938

    
1939
    /* release the last frame, if it is allocated and if it is not the
1940
     * golden frame */
1941
    if ((s->last_frame.data[0]) &&
1942
        (s->last_frame.data[0] != s->golden_frame.data[0]))
1943
        avctx->release_buffer(avctx, &s->last_frame);
1944

    
1945
    /* shuffle frames (last = current) */
1946
    s->last_frame= s->current_frame;
1947
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1948

    
1949
    return buf_size;
1950
}
1951

    
1952
/*
1953
 * This is the ffmpeg/libavcodec API module cleanup function.
1954
 */
1955
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1956
{
1957
    Vp3DecodeContext *s = avctx->priv_data;
1958
    int i;
1959

    
1960
    av_free(s->superblock_coding);
1961
    av_free(s->all_fragments);
1962
    av_free(s->coeff_counts);
1963
    av_free(s->coeffs);
1964
    av_free(s->coded_fragment_list);
1965
    av_free(s->superblock_fragments);
1966
    av_free(s->superblock_macroblocks);
1967
    av_free(s->macroblock_fragments);
1968
    av_free(s->macroblock_coding);
1969

    
1970
    for (i = 0; i < 16; i++) {
1971
        free_vlc(&s->dc_vlc[i]);
1972
        free_vlc(&s->ac_vlc_1[i]);
1973
        free_vlc(&s->ac_vlc_2[i]);
1974
        free_vlc(&s->ac_vlc_3[i]);
1975
        free_vlc(&s->ac_vlc_4[i]);
1976
    }
1977

    
1978
    free_vlc(&s->superblock_run_length_vlc);
1979
    free_vlc(&s->fragment_run_length_vlc);
1980
    free_vlc(&s->mode_code_vlc);
1981
    free_vlc(&s->motion_vector_vlc);
1982

    
1983
    /* release all frames */
1984
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
1985
        avctx->release_buffer(avctx, &s->golden_frame);
1986
    if (s->last_frame.data[0])
1987
        avctx->release_buffer(avctx, &s->last_frame);
1988
    /* no need to release the current_frame since it will always be pointing
1989
     * to the same frame as either the golden or last frame */
1990

    
1991
    return 0;
1992
}
1993

    
1994
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1995
{
1996
    Vp3DecodeContext *s = avctx->priv_data;
1997

    
1998
    if (get_bits1(gb)) {
1999
        int token;
2000
        if (s->entries >= 32) { /* overflow */
2001
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2002
            return -1;
2003
        }
2004
        token = get_bits(gb, 5);
2005
        //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);
2006
        s->huffman_table[s->hti][token][0] = s->hbits;
2007
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
2008
        s->entries++;
2009
    }
2010
    else {
2011
        if (s->huff_code_size >= 32) {/* overflow */
2012
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2013
            return -1;
2014
        }
2015
        s->huff_code_size++;
2016
        s->hbits <<= 1;
2017
        read_huffman_tree(avctx, gb);
2018
        s->hbits |= 1;
2019
        read_huffman_tree(avctx, gb);
2020
        s->hbits >>= 1;
2021
        s->huff_code_size--;
2022
    }
2023
    return 0;
2024
}
2025

    
2026
#if CONFIG_THEORA_DECODER
2027
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2028
{
2029
    Vp3DecodeContext *s = avctx->priv_data;
2030
    int visible_width, visible_height;
2031

    
2032
    s->theora = get_bits_long(gb, 24);
2033
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2034

    
2035
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2036
    /* but previous versions have the image flipped relative to vp3 */
2037
    if (s->theora < 0x030200)
2038
    {
2039
        s->flipped_image = 1;
2040
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2041
    }
2042

    
2043
    visible_width  = s->width  = get_bits(gb, 16) << 4;
2044
    visible_height = s->height = get_bits(gb, 16) << 4;
2045

    
2046
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2047
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2048
        s->width= s->height= 0;
2049
        return -1;
2050
    }
2051

    
2052
    if (s->theora >= 0x030400)
2053
    {
2054
        skip_bits(gb, 32); /* total number of superblocks in a frame */
2055
        // fixme, the next field is 36bits long
2056
        skip_bits(gb, 32); /* total number of blocks in a frame */
2057
        skip_bits(gb, 4); /* total number of blocks in a frame */
2058
        skip_bits(gb, 32); /* total number of macroblocks in a frame */
2059
    }
2060

    
2061
    if (s->theora >= 0x030200) {
2062
        visible_width  = get_bits_long(gb, 24);
2063
        visible_height = get_bits_long(gb, 24);
2064

    
2065
        skip_bits(gb, 8); /* offset x */
2066
        skip_bits(gb, 8); /* offset y */
2067
    }
2068

    
2069
    skip_bits(gb, 32); /* fps numerator */
2070
    skip_bits(gb, 32); /* fps denumerator */
2071
    skip_bits(gb, 24); /* aspect numerator */
2072
    skip_bits(gb, 24); /* aspect denumerator */
2073

    
2074
    if (s->theora < 0x030200)
2075
        skip_bits(gb, 5); /* keyframe frequency force */
2076
    skip_bits(gb, 8); /* colorspace */
2077
    if (s->theora >= 0x030400)
2078
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2079
    skip_bits(gb, 24); /* bitrate */
2080

    
2081
    skip_bits(gb, 6); /* quality hint */
2082

    
2083
    if (s->theora >= 0x030200)
2084
    {
2085
        skip_bits(gb, 5); /* keyframe frequency force */
2086

    
2087
        if (s->theora < 0x030400)
2088
            skip_bits(gb, 5); /* spare bits */
2089
    }
2090

    
2091
//    align_get_bits(gb);
2092

    
2093
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2094
        && visible_height <= s->height && visible_height > s->height-16)
2095
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2096
    else
2097
        avcodec_set_dimensions(avctx, s->width, s->height);
2098

    
2099
    return 0;
2100
}
2101

    
2102
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2103
{
2104
    Vp3DecodeContext *s = avctx->priv_data;
2105
    int i, n, matrices, inter, plane;
2106

    
2107
    if (s->theora >= 0x030200) {
2108
        n = get_bits(gb, 3);
2109
        /* loop filter limit values table */
2110
        for (i = 0; i < 64; i++)
2111
            s->filter_limit_values[i] = get_bits(gb, n);
2112
    }
2113

    
2114
    if (s->theora >= 0x030200)
2115
        n = get_bits(gb, 4) + 1;
2116
    else
2117
        n = 16;
2118
    /* quality threshold table */
2119
    for (i = 0; i < 64; i++)
2120
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2121

    
2122
    if (s->theora >= 0x030200)
2123
        n = get_bits(gb, 4) + 1;
2124
    else
2125
        n = 16;
2126
    /* dc scale factor table */
2127
    for (i = 0; i < 64; i++)
2128
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2129

    
2130
    if (s->theora >= 0x030200)
2131
        matrices = get_bits(gb, 9) + 1;
2132
    else
2133
        matrices = 3;
2134

    
2135
    if(matrices > 384){
2136
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2137
        return -1;
2138
    }
2139

    
2140
    for(n=0; n<matrices; n++){
2141
        for (i = 0; i < 64; i++)
2142
            s->base_matrix[n][i]= get_bits(gb, 8);
2143
    }
2144

    
2145
    for (inter = 0; inter <= 1; inter++) {
2146
        for (plane = 0; plane <= 2; plane++) {
2147
            int newqr= 1;
2148
            if (inter || plane > 0)
2149
                newqr = get_bits1(gb);
2150
            if (!newqr) {
2151
                int qtj, plj;
2152
                if(inter && get_bits1(gb)){
2153
                    qtj = 0;
2154
                    plj = plane;
2155
                }else{
2156
                    qtj= (3*inter + plane - 1) / 3;
2157
                    plj= (plane + 2) % 3;
2158
                }
2159
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2160
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2161
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2162
            } else {
2163
                int qri= 0;
2164
                int qi = 0;
2165

    
2166
                for(;;){
2167
                    i= get_bits(gb, av_log2(matrices-1)+1);
2168
                    if(i>= matrices){
2169
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2170
                        return -1;
2171
                    }
2172
                    s->qr_base[inter][plane][qri]= i;
2173
                    if(qi >= 63)
2174
                        break;
2175
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2176
                    s->qr_size[inter][plane][qri++]= i;
2177
                    qi += i;
2178
                }
2179

    
2180
                if (qi > 63) {
2181
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2182
                    return -1;
2183
                }
2184
                s->qr_count[inter][plane]= qri;
2185
            }
2186
        }
2187
    }
2188

    
2189
    /* Huffman tables */
2190
    for (s->hti = 0; s->hti < 80; s->hti++) {
2191
        s->entries = 0;
2192
        s->huff_code_size = 1;
2193
        if (!get_bits1(gb)) {
2194
            s->hbits = 0;
2195
            read_huffman_tree(avctx, gb);
2196
            s->hbits = 1;
2197
            read_huffman_tree(avctx, gb);
2198
        }
2199
    }
2200

    
2201
    s->theora_tables = 1;
2202

    
2203
    return 0;
2204
}
2205

    
2206
static int theora_decode_init(AVCodecContext *avctx)
2207
{
2208
    Vp3DecodeContext *s = avctx->priv_data;
2209
    GetBitContext gb;
2210
    int ptype;
2211
    uint8_t *header_start[3];
2212
    int header_len[3];
2213
    int i;
2214

    
2215
    s->theora = 1;
2216

    
2217
    if (!avctx->extradata_size)
2218
    {
2219
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2220
        return -1;
2221
    }
2222

    
2223
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2224
                              42, header_start, header_len) < 0) {
2225
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2226
        return -1;
2227
    }
2228

    
2229
  for(i=0;i<3;i++) {
2230
    init_get_bits(&gb, header_start[i], header_len[i]);
2231

    
2232
    ptype = get_bits(&gb, 8);
2233

    
2234
     if (!(ptype & 0x80))
2235
     {
2236
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2237
//        return -1;
2238
     }
2239

    
2240
    // FIXME: Check for this as well.
2241
    skip_bits(&gb, 6*8); /* "theora" */
2242

    
2243
    switch(ptype)
2244
    {
2245
        case 0x80:
2246
            theora_decode_header(avctx, &gb);
2247
                break;
2248
        case 0x81:
2249
// FIXME: is this needed? it breaks sometimes
2250
//            theora_decode_comments(avctx, gb);
2251
            break;
2252
        case 0x82:
2253
            theora_decode_tables(avctx, &gb);
2254
            break;
2255
        default:
2256
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2257
            break;
2258
    }
2259
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2260
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2261
    if (s->theora < 0x030200)
2262
        break;
2263
  }
2264

    
2265
    vp3_decode_init(avctx);
2266
    return 0;
2267
}
2268

    
2269
AVCodec theora_decoder = {
2270
    "theora",
2271
    CODEC_TYPE_VIDEO,
2272
    CODEC_ID_THEORA,
2273
    sizeof(Vp3DecodeContext),
2274
    theora_decode_init,
2275
    NULL,
2276
    vp3_decode_end,
2277
    vp3_decode_frame,
2278
    0,
2279
    NULL,
2280
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2281
};
2282
#endif
2283

    
2284
AVCodec vp3_decoder = {
2285
    "vp3",
2286
    CODEC_TYPE_VIDEO,
2287
    CODEC_ID_VP3,
2288
    sizeof(Vp3DecodeContext),
2289
    vp3_decode_init,
2290
    NULL,
2291
    vp3_decode_end,
2292
    vp3_decode_frame,
2293
    0,
2294
    NULL,
2295
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2296
};