Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 356306ac

History | View | Annotate | Download (93.7 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 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://multimedia.cx/
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 "mpegvideo.h"
40

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

    
44
#define FRAGMENT_PIXELS 8
45

    
46
/*
47
 * Debugging Variables
48
 *
49
 * Define one or more of the following compile-time variables to 1 to obtain
50
 * elaborate information about certain aspects of the decoding process.
51
 *
52
 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
53
 * DEBUG_VP3: high-level decoding flow
54
 * DEBUG_INIT: initialization parameters
55
 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
56
 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
57
 * DEBUG_MODES: unpacking the coding modes for individual fragments
58
 * DEBUG_VECTORS: display the motion vectors
59
 * DEBUG_TOKEN: display exhaustive information about each DCT token
60
 * DEBUG_VLC: display the VLCs as they are extracted from the stream
61
 * DEBUG_DC_PRED: display the process of reversing DC prediction
62
 * DEBUG_IDCT: show every detail of the IDCT process
63
 */
64

    
65
#define KEYFRAMES_ONLY 0
66

    
67
#define DEBUG_VP3 0
68
#define DEBUG_INIT 0
69
#define DEBUG_DEQUANTIZERS 0
70
#define DEBUG_BLOCK_CODING 0
71
#define DEBUG_MODES 0
72
#define DEBUG_VECTORS 0
73
#define DEBUG_TOKEN 0
74
#define DEBUG_VLC 0
75
#define DEBUG_DC_PRED 0
76
#define DEBUG_IDCT 0
77

    
78
#if DEBUG_VP3
79
#define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
80
#else
81
static inline void debug_vp3(const char *format, ...) { }
82
#endif
83

    
84
#if DEBUG_INIT
85
#define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
86
#else
87
static inline void debug_init(const char *format, ...) { }
88
#endif
89

    
90
#if DEBUG_DEQUANTIZERS
91
#define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
92
#else
93
static inline void debug_dequantizers(const char *format, ...) { }
94
#endif
95

    
96
#if DEBUG_BLOCK_CODING
97
#define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
98
#else
99
static inline void debug_block_coding(const char *format, ...) { }
100
#endif
101

    
102
#if DEBUG_MODES
103
#define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
104
#else
105
static inline void debug_modes(const char *format, ...) { }
106
#endif
107

    
108
#if DEBUG_VECTORS
109
#define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
110
#else
111
static inline void debug_vectors(const char *format, ...) { }
112
#endif
113

    
114
#if DEBUG_TOKEN
115
#define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
116
#else
117
static inline void debug_token(const char *format, ...) { }
118
#endif
119

    
120
#if DEBUG_VLC
121
#define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
122
#else
123
static inline void debug_vlc(const char *format, ...) { }
124
#endif
125

    
126
#if DEBUG_DC_PRED
127
#define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
128
#else
129
static inline void debug_dc_pred(const char *format, ...) { }
130
#endif
131

    
132
#if DEBUG_IDCT
133
#define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
134
#else
135
static inline void debug_idct(const char *format, ...) { }
136
#endif
137

    
138
typedef struct Coeff {
139
    struct Coeff *next;
140
    DCTELEM coeff;
141
    uint8_t index;
142
} Coeff;
143

    
144
//FIXME split things out into their own arrays
145
typedef struct Vp3Fragment {
146
    Coeff *next_coeff;
147
    /* address of first pixel taking into account which plane the fragment
148
     * lives on as well as the plane stride */
149
    int first_pixel;
150
    /* this is the macroblock that the fragment belongs to */
151
    uint16_t macroblock;
152
    uint8_t coding_method;
153
    uint8_t coeff_count;
154
    int8_t motion_x;
155
    int8_t motion_y;
156
} Vp3Fragment;
157

    
158
#define SB_NOT_CODED        0
159
#define SB_PARTIALLY_CODED  1
160
#define SB_FULLY_CODED      2
161

    
162
#define MODE_INTER_NO_MV      0
163
#define MODE_INTRA            1
164
#define MODE_INTER_PLUS_MV    2
165
#define MODE_INTER_LAST_MV    3
166
#define MODE_INTER_PRIOR_LAST 4
167
#define MODE_USING_GOLDEN     5
168
#define MODE_GOLDEN_MV        6
169
#define MODE_INTER_FOURMV     7
170
#define CODING_MODE_COUNT     8
171

    
172
/* special internal mode */
173
#define MODE_COPY             8
174

    
175
/* There are 6 preset schemes, plus a free-form scheme */
176
static int ModeAlphabet[7][CODING_MODE_COUNT] =
177
{
178
    /* this is the custom scheme */
179
    { 0, 0, 0, 0, 0, 0, 0, 0 },
180

    
181
    /* scheme 1: Last motion vector dominates */
182
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
183
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
184
         MODE_INTRA,            MODE_USING_GOLDEN,
185
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
186

    
187
    /* scheme 2 */
188
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
189
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
190
         MODE_INTRA,            MODE_USING_GOLDEN,
191
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
192

    
193
    /* scheme 3 */
194
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
195
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
196
         MODE_INTRA,            MODE_USING_GOLDEN,
197
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
198

    
199
    /* scheme 4 */
200
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
201
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
202
         MODE_INTRA,            MODE_USING_GOLDEN,
203
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
204

    
205
    /* scheme 5: No motion vector dominates */
206
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
207
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
208
         MODE_INTRA,            MODE_USING_GOLDEN,
209
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
210

    
211
    /* scheme 6 */
212
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
213
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
214
         MODE_INTER_PLUS_MV,    MODE_INTRA,
215
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
216

    
217
};
218

    
219
#define MIN_DEQUANT_VAL 2
220

    
221
typedef struct Vp3DecodeContext {
222
    AVCodecContext *avctx;
223
    int theora, theora_tables;
224
    int version;
225
    int width, height;
226
    AVFrame golden_frame;
227
    AVFrame last_frame;
228
    AVFrame current_frame;
229
    int keyframe;
230
    DSPContext dsp;
231
    int flipped_image;
232

    
233
    int qis[3];
234
    int nqis;
235
    int quality_index;
236
    int last_quality_index;
237

    
238
    int superblock_count;
239
    int superblock_width;
240
    int superblock_height;
241
    int y_superblock_width;
242
    int y_superblock_height;
243
    int c_superblock_width;
244
    int c_superblock_height;
245
    int u_superblock_start;
246
    int v_superblock_start;
247
    unsigned char *superblock_coding;
248

    
249
    int macroblock_count;
250
    int macroblock_width;
251
    int macroblock_height;
252

    
253
    int fragment_count;
254
    int fragment_width;
255
    int fragment_height;
256

    
257
    Vp3Fragment *all_fragments;
258
    Coeff *coeffs;
259
    Coeff *next_coeff;
260
    int fragment_start[3];
261

    
262
    ScanTable scantable;
263

    
264
    /* tables */
265
    uint16_t coded_dc_scale_factor[64];
266
    uint32_t coded_ac_scale_factor[64];
267
    uint8_t base_matrix[384][64];
268
    uint8_t qr_count[2][3];
269
    uint8_t qr_size [2][3][64];
270
    uint16_t qr_base[2][3][64];
271

    
272
    /* this is a list of indices into the all_fragments array indicating
273
     * which of the fragments are coded */
274
    int *coded_fragment_list;
275
    int coded_fragment_list_index;
276
    int pixel_addresses_inited;
277

    
278
    VLC dc_vlc[16];
279
    VLC ac_vlc_1[16];
280
    VLC ac_vlc_2[16];
281
    VLC ac_vlc_3[16];
282
    VLC ac_vlc_4[16];
283

    
284
    VLC superblock_run_length_vlc;
285
    VLC fragment_run_length_vlc;
286
    VLC mode_code_vlc;
287
    VLC motion_vector_vlc;
288

    
289
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
290
     * index into them */
291
    DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]);        //<qmat[is_inter][plane]
292

    
293
    /* This table contains superblock_count * 16 entries. Each set of 16
294
     * numbers corresponds to the fragment indices 0..15 of the superblock.
295
     * An entry will be -1 to indicate that no entry corresponds to that
296
     * index. */
297
    int *superblock_fragments;
298

    
299
    /* This table contains superblock_count * 4 entries. Each set of 4
300
     * numbers corresponds to the macroblock indices 0..3 of the superblock.
301
     * An entry will be -1 to indicate that no entry corresponds to that
302
     * index. */
303
    int *superblock_macroblocks;
304

    
305
    /* This table contains macroblock_count * 6 entries. Each set of 6
306
     * numbers corresponds to the fragment indices 0..5 which comprise
307
     * the macroblock (4 Y fragments and 2 C fragments). */
308
    int *macroblock_fragments;
309
    /* This is an array that indicates how a particular macroblock
310
     * is coded. */
311
    unsigned char *macroblock_coding;
312

    
313
    int first_coded_y_fragment;
314
    int first_coded_c_fragment;
315
    int last_coded_y_fragment;
316
    int last_coded_c_fragment;
317

    
318
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
319
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
320

    
321
    /* Huffman decode */
322
    int hti;
323
    unsigned int hbits;
324
    int entries;
325
    int huff_code_size;
326
    uint16_t huffman_table[80][32][2];
327

    
328
    uint32_t filter_limit_values[64];
329
    int bounding_values_array[256];
330
} Vp3DecodeContext;
331

    
332
/************************************************************************
333
 * VP3 specific functions
334
 ************************************************************************/
335

    
336
/*
337
 * This function sets up all of the various blocks mappings:
338
 * superblocks <-> fragments, macroblocks <-> fragments,
339
 * superblocks <-> macroblocks
340
 *
341
 * Returns 0 is successful; returns 1 if *anything* went wrong.
342
 */
343
static int init_block_mapping(Vp3DecodeContext *s)
344
{
345
    int i, j;
346
    signed int hilbert_walk_mb[4];
347

    
348
    int current_fragment = 0;
349
    int current_width = 0;
350
    int current_height = 0;
351
    int right_edge = 0;
352
    int bottom_edge = 0;
353
    int superblock_row_inc = 0;
354
    int *hilbert = NULL;
355
    int mapping_index = 0;
356

    
357
    int current_macroblock;
358
    int c_fragment;
359

    
360
    signed char travel_width[16] = {
361
         1,  1,  0, -1,
362
         0,  0,  1,  0,
363
         1,  0,  1,  0,
364
         0, -1,  0,  1
365
    };
366

    
367
    signed char travel_height[16] = {
368
         0,  0,  1,  0,
369
         1,  1,  0, -1,
370
         0,  1,  0, -1,
371
        -1,  0, -1,  0
372
    };
373

    
374
    signed char travel_width_mb[4] = {
375
         1,  0,  1,  0
376
    };
377

    
378
    signed char travel_height_mb[4] = {
379
         0,  1,  0, -1
380
    };
381

    
382
    debug_vp3("  vp3: initialize block mapping tables\n");
383

    
384
    hilbert_walk_mb[0] = 1;
385
    hilbert_walk_mb[1] = s->macroblock_width;
386
    hilbert_walk_mb[2] = 1;
387
    hilbert_walk_mb[3] = -s->macroblock_width;
388

    
389
    /* iterate through each superblock (all planes) and map the fragments */
390
    for (i = 0; i < s->superblock_count; i++) {
391
        debug_init("    superblock %d (u starts @ %d, v starts @ %d)\n",
392
            i, s->u_superblock_start, s->v_superblock_start);
393

    
394
        /* time to re-assign the limits? */
395
        if (i == 0) {
396

    
397
            /* start of Y superblocks */
398
            right_edge = s->fragment_width;
399
            bottom_edge = s->fragment_height;
400
            current_width = -1;
401
            current_height = 0;
402
            superblock_row_inc = 3 * s->fragment_width -
403
                (s->y_superblock_width * 4 - s->fragment_width);
404

    
405
            /* the first operation for this variable is to advance by 1 */
406
            current_fragment = -1;
407

    
408
        } else if (i == s->u_superblock_start) {
409

    
410
            /* start of U superblocks */
411
            right_edge = s->fragment_width / 2;
412
            bottom_edge = s->fragment_height / 2;
413
            current_width = -1;
414
            current_height = 0;
415
            superblock_row_inc = 3 * (s->fragment_width / 2) -
416
                (s->c_superblock_width * 4 - s->fragment_width / 2);
417

    
418
            /* the first operation for this variable is to advance by 1 */
419
            current_fragment = s->fragment_start[1] - 1;
420

    
421
        } else if (i == s->v_superblock_start) {
422

    
423
            /* start of V superblocks */
424
            right_edge = s->fragment_width / 2;
425
            bottom_edge = s->fragment_height / 2;
426
            current_width = -1;
427
            current_height = 0;
428
            superblock_row_inc = 3 * (s->fragment_width / 2) -
429
                (s->c_superblock_width * 4 - s->fragment_width / 2);
430

    
431
            /* the first operation for this variable is to advance by 1 */
432
            current_fragment = s->fragment_start[2] - 1;
433

    
434
        }
435

    
436
        if (current_width >= right_edge - 1) {
437
            /* reset width and move to next superblock row */
438
            current_width = -1;
439
            current_height += 4;
440

    
441
            /* fragment is now at the start of a new superblock row */
442
            current_fragment += superblock_row_inc;
443
        }
444

    
445
        /* iterate through all 16 fragments in a superblock */
446
        for (j = 0; j < 16; j++) {
447
            current_fragment += travel_width[j] + right_edge * travel_height[j];
448
            current_width += travel_width[j];
449
            current_height += travel_height[j];
450

    
451
            /* check if the fragment is in bounds */
452
            if ((current_width < right_edge) &&
453
                (current_height < bottom_edge)) {
454
                s->superblock_fragments[mapping_index] = current_fragment;
455
                debug_init("    mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
456
                    s->superblock_fragments[mapping_index], i, j,
457
                    current_width, right_edge, current_height, bottom_edge);
458
            } else {
459
                s->superblock_fragments[mapping_index] = -1;
460
                debug_init("    superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
461
                    i, j,
462
                    current_width, right_edge, current_height, bottom_edge);
463
            }
464

    
465
            mapping_index++;
466
        }
467
    }
468

    
469
    /* initialize the superblock <-> macroblock mapping; iterate through
470
     * all of the Y plane superblocks to build this mapping */
471
    right_edge = s->macroblock_width;
472
    bottom_edge = s->macroblock_height;
473
    current_width = -1;
474
    current_height = 0;
475
    superblock_row_inc = s->macroblock_width -
476
        (s->y_superblock_width * 2 - s->macroblock_width);;
477
    hilbert = hilbert_walk_mb;
478
    mapping_index = 0;
479
    current_macroblock = -1;
480
    for (i = 0; i < s->u_superblock_start; i++) {
481

    
482
        if (current_width >= right_edge - 1) {
483
            /* reset width and move to next superblock row */
484
            current_width = -1;
485
            current_height += 2;
486

    
487
            /* macroblock is now at the start of a new superblock row */
488
            current_macroblock += superblock_row_inc;
489
        }
490

    
491
        /* iterate through each potential macroblock in the superblock */
492
        for (j = 0; j < 4; j++) {
493
            current_macroblock += hilbert_walk_mb[j];
494
            current_width += travel_width_mb[j];
495
            current_height += travel_height_mb[j];
496

    
497
            /* check if the macroblock is in bounds */
498
            if ((current_width < right_edge) &&
499
                (current_height < bottom_edge)) {
500
                s->superblock_macroblocks[mapping_index] = current_macroblock;
501
                debug_init("    mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
502
                    s->superblock_macroblocks[mapping_index], i, j,
503
                    current_width, right_edge, current_height, bottom_edge);
504
            } else {
505
                s->superblock_macroblocks[mapping_index] = -1;
506
                debug_init("    superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
507
                    i, j,
508
                    current_width, right_edge, current_height, bottom_edge);
509
            }
510

    
511
            mapping_index++;
512
        }
513
    }
514

    
515
    /* initialize the macroblock <-> fragment mapping */
516
    current_fragment = 0;
517
    current_macroblock = 0;
518
    mapping_index = 0;
519
    for (i = 0; i < s->fragment_height; i += 2) {
520

    
521
        for (j = 0; j < s->fragment_width; j += 2) {
522

    
523
            debug_init("    macroblock %d contains fragments: ", current_macroblock);
524
            s->all_fragments[current_fragment].macroblock = current_macroblock;
525
            s->macroblock_fragments[mapping_index++] = current_fragment;
526
            debug_init("%d ", current_fragment);
527

    
528
            if (j + 1 < s->fragment_width) {
529
                s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
530
                s->macroblock_fragments[mapping_index++] = current_fragment + 1;
531
                debug_init("%d ", current_fragment + 1);
532
            } else
533
                s->macroblock_fragments[mapping_index++] = -1;
534

    
535
            if (i + 1 < s->fragment_height) {
536
                s->all_fragments[current_fragment + s->fragment_width].macroblock =
537
                    current_macroblock;
538
                s->macroblock_fragments[mapping_index++] =
539
                    current_fragment + s->fragment_width;
540
                debug_init("%d ", current_fragment + s->fragment_width);
541
            } else
542
                s->macroblock_fragments[mapping_index++] = -1;
543

    
544
            if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
545
                s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
546
                    current_macroblock;
547
                s->macroblock_fragments[mapping_index++] =
548
                    current_fragment + s->fragment_width + 1;
549
                debug_init("%d ", current_fragment + s->fragment_width + 1);
550
            } else
551
                s->macroblock_fragments[mapping_index++] = -1;
552

    
553
            /* C planes */
554
            c_fragment = s->fragment_start[1] +
555
                (i * s->fragment_width / 4) + (j / 2);
556
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
557
            s->macroblock_fragments[mapping_index++] = c_fragment;
558
            debug_init("%d ", c_fragment);
559

    
560
            c_fragment = s->fragment_start[2] +
561
                (i * s->fragment_width / 4) + (j / 2);
562
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
563
            s->macroblock_fragments[mapping_index++] = c_fragment;
564
            debug_init("%d ", c_fragment);
565

    
566
            debug_init("\n");
567

    
568
            if (j + 2 <= s->fragment_width)
569
                current_fragment += 2;
570
            else
571
                current_fragment++;
572
            current_macroblock++;
573
        }
574

    
575
        current_fragment += s->fragment_width;
576
    }
577

    
578
    return 0;  /* successful path out */
579
}
580

    
581
/*
582
 * This function wipes out all of the fragment data.
583
 */
584
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
585
{
586
    int i;
587

    
588
    /* zero out all of the fragment information */
589
    s->coded_fragment_list_index = 0;
590
    for (i = 0; i < s->fragment_count; i++) {
591
        s->all_fragments[i].coeff_count = 0;
592
        s->all_fragments[i].motion_x = 127;
593
        s->all_fragments[i].motion_y = 127;
594
        s->all_fragments[i].next_coeff= NULL;
595
        s->coeffs[i].index=
596
        s->coeffs[i].coeff=0;
597
        s->coeffs[i].next= NULL;
598
    }
599
}
600

    
601
/*
602
 * This function sets up the dequantization tables used for a particular
603
 * frame.
604
 */
605
static void init_dequantizer(Vp3DecodeContext *s)
606
{
607
    int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
608
    int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
609
    int i, plane, inter, qri, bmi, bmj, qistart;
610

    
611
    debug_vp3("  vp3: initializing dequantization tables\n");
612

    
613
    for(inter=0; inter<2; inter++){
614
        for(plane=0; plane<3; plane++){
615
            int sum=0;
616
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
617
                sum+= s->qr_size[inter][plane][qri];
618
                if(s->quality_index <= sum)
619
                    break;
620
            }
621
            qistart= sum - s->qr_size[inter][plane][qri];
622
            bmi= s->qr_base[inter][plane][qri  ];
623
            bmj= s->qr_base[inter][plane][qri+1];
624
            for(i=0; i<64; i++){
625
                int coeff= (  2*(sum    -s->quality_index)*s->base_matrix[bmi][i]
626
                            - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
627
                            + s->qr_size[inter][plane][qri])
628
                           / (2*s->qr_size[inter][plane][qri]);
629

    
630
                int qmin= 8<<(inter + !i);
631
                int qscale= i ? ac_scale_factor : dc_scale_factor;
632

    
633
                s->qmat[inter][plane][i]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
634
            }
635
        }
636
    }
637

    
638
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
639
}
640

    
641
/*
642
 * This function initializes the loop filter boundary limits if the frame's
643
 * quality index is different from the previous frame's.
644
 */
645
static void init_loop_filter(Vp3DecodeContext *s)
646
{
647
    int *bounding_values= s->bounding_values_array+127;
648
    int filter_limit;
649
    int x;
650

    
651
    filter_limit = s->filter_limit_values[s->quality_index];
652

    
653
    /* set up the bounding values */
654
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
655
    for (x = 0; x < filter_limit; x++) {
656
        bounding_values[-x - filter_limit] = -filter_limit + x;
657
        bounding_values[-x] = -x;
658
        bounding_values[x] = x;
659
        bounding_values[x + filter_limit] = filter_limit - x;
660
    }
661
}
662

    
663
/*
664
 * This function unpacks all of the superblock/macroblock/fragment coding
665
 * information from the bitstream.
666
 */
667
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
668
{
669
    int bit = 0;
670
    int current_superblock = 0;
671
    int current_run = 0;
672
    int decode_fully_flags = 0;
673
    int decode_partial_blocks = 0;
674
    int first_c_fragment_seen;
675

    
676
    int i, j;
677
    int current_fragment;
678

    
679
    debug_vp3("  vp3: unpacking superblock coding\n");
680

    
681
    if (s->keyframe) {
682

    
683
        debug_vp3("    keyframe-- all superblocks are fully coded\n");
684
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
685

    
686
    } else {
687

    
688
        /* unpack the list of partially-coded superblocks */
689
        bit = get_bits1(gb);
690
        /* toggle the bit because as soon as the first run length is
691
         * fetched the bit will be toggled again */
692
        bit ^= 1;
693
        while (current_superblock < s->superblock_count) {
694
            if (current_run-- == 0) {
695
                bit ^= 1;
696
                current_run = get_vlc2(gb,
697
                    s->superblock_run_length_vlc.table, 6, 2);
698
                if (current_run == 33)
699
                    current_run += get_bits(gb, 12);
700
                debug_block_coding("      setting superblocks %d..%d to %s\n",
701
                    current_superblock,
702
                    current_superblock + current_run - 1,
703
                    (bit) ? "partially coded" : "not coded");
704

    
705
                /* if any of the superblocks are not partially coded, flag
706
                 * a boolean to decode the list of fully-coded superblocks */
707
                if (bit == 0) {
708
                    decode_fully_flags = 1;
709
                } else {
710

    
711
                    /* make a note of the fact that there are partially coded
712
                     * superblocks */
713
                    decode_partial_blocks = 1;
714
                }
715
            }
716
            s->superblock_coding[current_superblock++] = bit;
717
        }
718

    
719
        /* unpack the list of fully coded superblocks if any of the blocks were
720
         * not marked as partially coded in the previous step */
721
        if (decode_fully_flags) {
722

    
723
            current_superblock = 0;
724
            current_run = 0;
725
            bit = get_bits1(gb);
726
            /* toggle the bit because as soon as the first run length is
727
             * fetched the bit will be toggled again */
728
            bit ^= 1;
729
            while (current_superblock < s->superblock_count) {
730

    
731
                /* skip any superblocks already marked as partially coded */
732
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
733

    
734
                    if (current_run-- == 0) {
735
                        bit ^= 1;
736
                        current_run = get_vlc2(gb,
737
                            s->superblock_run_length_vlc.table, 6, 2);
738
                        if (current_run == 33)
739
                            current_run += get_bits(gb, 12);
740
                    }
741

    
742
                    debug_block_coding("      setting superblock %d to %s\n",
743
                        current_superblock,
744
                        (bit) ? "fully coded" : "not coded");
745
                    s->superblock_coding[current_superblock] = 2*bit;
746
                }
747
                current_superblock++;
748
            }
749
        }
750

    
751
        /* if there were partial blocks, initialize bitstream for
752
         * unpacking fragment codings */
753
        if (decode_partial_blocks) {
754

    
755
            current_run = 0;
756
            bit = get_bits1(gb);
757
            /* toggle the bit because as soon as the first run length is
758
             * fetched the bit will be toggled again */
759
            bit ^= 1;
760
        }
761
    }
762

    
763
    /* figure out which fragments are coded; iterate through each
764
     * superblock (all planes) */
765
    s->coded_fragment_list_index = 0;
766
    s->next_coeff= s->coeffs + s->fragment_count;
767
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
768
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
769
    first_c_fragment_seen = 0;
770
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
771
    for (i = 0; i < s->superblock_count; i++) {
772

    
773
        /* iterate through all 16 fragments in a superblock */
774
        for (j = 0; j < 16; j++) {
775

    
776
            /* if the fragment is in bounds, check its coding status */
777
            current_fragment = s->superblock_fragments[i * 16 + j];
778
            if (current_fragment >= s->fragment_count) {
779
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
780
                    current_fragment, s->fragment_count);
781
                return 1;
782
            }
783
            if (current_fragment != -1) {
784
                if (s->superblock_coding[i] == SB_NOT_CODED) {
785

    
786
                    /* copy all the fragments from the prior frame */
787
                    s->all_fragments[current_fragment].coding_method =
788
                        MODE_COPY;
789

    
790
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
791

    
792
                    /* fragment may or may not be coded; this is the case
793
                     * that cares about the fragment coding runs */
794
                    if (current_run-- == 0) {
795
                        bit ^= 1;
796
                        current_run = get_vlc2(gb,
797
                            s->fragment_run_length_vlc.table, 5, 2);
798
                    }
799

    
800
                    if (bit) {
801
                        /* default mode; actual mode will be decoded in
802
                         * the next phase */
803
                        s->all_fragments[current_fragment].coding_method =
804
                            MODE_INTER_NO_MV;
805
                        s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
806
                        s->coded_fragment_list[s->coded_fragment_list_index] =
807
                            current_fragment;
808
                        if ((current_fragment >= s->fragment_start[1]) &&
809
                            (s->last_coded_y_fragment == -1) &&
810
                            (!first_c_fragment_seen)) {
811
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
812
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
813
                            first_c_fragment_seen = 1;
814
                        }
815
                        s->coded_fragment_list_index++;
816
                        s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
817
                        debug_block_coding("      superblock %d is partially coded, fragment %d is coded\n",
818
                            i, current_fragment);
819
                    } else {
820
                        /* not coded; copy this fragment from the prior frame */
821
                        s->all_fragments[current_fragment].coding_method =
822
                            MODE_COPY;
823
                        debug_block_coding("      superblock %d is partially coded, fragment %d is not coded\n",
824
                            i, current_fragment);
825
                    }
826

    
827
                } else {
828

    
829
                    /* fragments are fully coded in this superblock; actual
830
                     * coding will be determined in next step */
831
                    s->all_fragments[current_fragment].coding_method =
832
                        MODE_INTER_NO_MV;
833
                    s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
834
                    s->coded_fragment_list[s->coded_fragment_list_index] =
835
                        current_fragment;
836
                    if ((current_fragment >= s->fragment_start[1]) &&
837
                        (s->last_coded_y_fragment == -1) &&
838
                        (!first_c_fragment_seen)) {
839
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
840
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
841
                        first_c_fragment_seen = 1;
842
                    }
843
                    s->coded_fragment_list_index++;
844
                    s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
845
                    debug_block_coding("      superblock %d is fully coded, fragment %d is coded\n",
846
                        i, current_fragment);
847
                }
848
            }
849
        }
850
    }
851

    
852
    if (!first_c_fragment_seen)
853
        /* only Y fragments coded in this frame */
854
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
855
    else
856
        /* end the list of coded C fragments */
857
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
858

    
859
    debug_block_coding("    %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
860
        s->coded_fragment_list_index,
861
        s->first_coded_y_fragment,
862
        s->last_coded_y_fragment,
863
        s->first_coded_c_fragment,
864
        s->last_coded_c_fragment);
865

    
866
    return 0;
867
}
868

    
869
/*
870
 * This function unpacks all the coding mode data for individual macroblocks
871
 * from the bitstream.
872
 */
873
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
874
{
875
    int i, j, k;
876
    int scheme;
877
    int current_macroblock;
878
    int current_fragment;
879
    int coding_mode;
880

    
881
    debug_vp3("  vp3: unpacking encoding modes\n");
882

    
883
    if (s->keyframe) {
884
        debug_vp3("    keyframe-- all blocks are coded as INTRA\n");
885

    
886
        for (i = 0; i < s->fragment_count; i++)
887
            s->all_fragments[i].coding_method = MODE_INTRA;
888

    
889
    } else {
890

    
891
        /* fetch the mode coding scheme for this frame */
892
        scheme = get_bits(gb, 3);
893
        debug_modes("    using mode alphabet %d\n", scheme);
894

    
895
        /* is it a custom coding scheme? */
896
        if (scheme == 0) {
897
            debug_modes("    custom mode alphabet ahead:\n");
898
            for (i = 0; i < 8; i++)
899
                ModeAlphabet[scheme][get_bits(gb, 3)] = i;
900
        }
901

    
902
        for (i = 0; i < 8; i++)
903
            debug_modes("      mode[%d][%d] = %d\n", scheme, i,
904
                ModeAlphabet[scheme][i]);
905

    
906
        /* iterate through all of the macroblocks that contain 1 or more
907
         * coded fragments */
908
        for (i = 0; i < s->u_superblock_start; i++) {
909

    
910
            for (j = 0; j < 4; j++) {
911
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
912
                if ((current_macroblock == -1) ||
913
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
914
                    continue;
915
                if (current_macroblock >= s->macroblock_count) {
916
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
917
                        current_macroblock, s->macroblock_count);
918
                    return 1;
919
                }
920

    
921
                /* mode 7 means get 3 bits for each coding mode */
922
                if (scheme == 7)
923
                    coding_mode = get_bits(gb, 3);
924
                else
925
                    coding_mode = ModeAlphabet[scheme]
926
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
927

    
928
                s->macroblock_coding[current_macroblock] = coding_mode;
929
                for (k = 0; k < 6; k++) {
930
                    current_fragment =
931
                        s->macroblock_fragments[current_macroblock * 6 + k];
932
                    if (current_fragment == -1)
933
                        continue;
934
                    if (current_fragment >= s->fragment_count) {
935
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
936
                            current_fragment, s->fragment_count);
937
                        return 1;
938
                    }
939
                    if (s->all_fragments[current_fragment].coding_method !=
940
                        MODE_COPY)
941
                        s->all_fragments[current_fragment].coding_method =
942
                            coding_mode;
943
                }
944

    
945
                debug_modes("    coding method for macroblock starting @ fragment %d = %d\n",
946
                    s->macroblock_fragments[current_macroblock * 6], coding_mode);
947
            }
948
        }
949
    }
950

    
951
    return 0;
952
}
953

    
954
/*
955
 * This function unpacks all the motion vectors for the individual
956
 * macroblocks from the bitstream.
957
 */
958
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
959
{
960
    int i, j, k;
961
    int coding_mode;
962
    int motion_x[6];
963
    int motion_y[6];
964
    int last_motion_x = 0;
965
    int last_motion_y = 0;
966
    int prior_last_motion_x = 0;
967
    int prior_last_motion_y = 0;
968
    int current_macroblock;
969
    int current_fragment;
970

    
971
    debug_vp3("  vp3: unpacking motion vectors\n");
972
    if (s->keyframe) {
973

    
974
        debug_vp3("    keyframe-- there are no motion vectors\n");
975

    
976
    } else {
977

    
978
        memset(motion_x, 0, 6 * sizeof(int));
979
        memset(motion_y, 0, 6 * sizeof(int));
980

    
981
        /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
982
        coding_mode = get_bits1(gb);
983
        debug_vectors("    using %s scheme for unpacking motion vectors\n",
984
            (coding_mode == 0) ? "VLC" : "fixed-length");
985

    
986
        /* iterate through all of the macroblocks that contain 1 or more
987
         * coded fragments */
988
        for (i = 0; i < s->u_superblock_start; i++) {
989

    
990
            for (j = 0; j < 4; j++) {
991
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
992
                if ((current_macroblock == -1) ||
993
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
994
                    continue;
995
                if (current_macroblock >= s->macroblock_count) {
996
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
997
                        current_macroblock, s->macroblock_count);
998
                    return 1;
999
                }
1000

    
1001
                current_fragment = s->macroblock_fragments[current_macroblock * 6];
1002
                if (current_fragment >= s->fragment_count) {
1003
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1004
                        current_fragment, s->fragment_count);
1005
                    return 1;
1006
                }
1007
                switch (s->macroblock_coding[current_macroblock]) {
1008

    
1009
                case MODE_INTER_PLUS_MV:
1010
                case MODE_GOLDEN_MV:
1011
                    /* all 6 fragments use the same motion vector */
1012
                    if (coding_mode == 0) {
1013
                        motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1014
                        motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1015
                    } else {
1016
                        motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1017
                        motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1018
                    }
1019

    
1020
                    for (k = 1; k < 6; k++) {
1021
                        motion_x[k] = motion_x[0];
1022
                        motion_y[k] = motion_y[0];
1023
                    }
1024

    
1025
                    /* vector maintenance, only on MODE_INTER_PLUS_MV */
1026
                    if (s->macroblock_coding[current_macroblock] ==
1027
                        MODE_INTER_PLUS_MV) {
1028
                        prior_last_motion_x = last_motion_x;
1029
                        prior_last_motion_y = last_motion_y;
1030
                        last_motion_x = motion_x[0];
1031
                        last_motion_y = motion_y[0];
1032
                    }
1033
                    break;
1034

    
1035
                case MODE_INTER_FOURMV:
1036
                    /* fetch 4 vectors from the bitstream, one for each
1037
                     * Y fragment, then average for the C fragment vectors */
1038
                    motion_x[4] = motion_y[4] = 0;
1039
                    for (k = 0; k < 4; k++) {
1040
                        if (coding_mode == 0) {
1041
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1042
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1043
                        } else {
1044
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1045
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1046
                        }
1047
                        motion_x[4] += motion_x[k];
1048
                        motion_y[4] += motion_y[k];
1049
                    }
1050

    
1051
                    motion_x[5]=
1052
                    motion_x[4]= RSHIFT(motion_x[4], 2);
1053
                    motion_y[5]=
1054
                    motion_y[4]= RSHIFT(motion_y[4], 2);
1055

    
1056
                    /* vector maintenance; vector[3] is treated as the
1057
                     * last vector in this case */
1058
                    prior_last_motion_x = last_motion_x;
1059
                    prior_last_motion_y = last_motion_y;
1060
                    last_motion_x = motion_x[3];
1061
                    last_motion_y = motion_y[3];
1062
                    break;
1063

    
1064
                case MODE_INTER_LAST_MV:
1065
                    /* all 6 fragments use the last motion vector */
1066
                    motion_x[0] = last_motion_x;
1067
                    motion_y[0] = last_motion_y;
1068
                    for (k = 1; k < 6; k++) {
1069
                        motion_x[k] = motion_x[0];
1070
                        motion_y[k] = motion_y[0];
1071
                    }
1072

    
1073
                    /* no vector maintenance (last vector remains the
1074
                     * last vector) */
1075
                    break;
1076

    
1077
                case MODE_INTER_PRIOR_LAST:
1078
                    /* all 6 fragments use the motion vector prior to the
1079
                     * last motion vector */
1080
                    motion_x[0] = prior_last_motion_x;
1081
                    motion_y[0] = prior_last_motion_y;
1082
                    for (k = 1; k < 6; k++) {
1083
                        motion_x[k] = motion_x[0];
1084
                        motion_y[k] = motion_y[0];
1085
                    }
1086

    
1087
                    /* vector maintenance */
1088
                    prior_last_motion_x = last_motion_x;
1089
                    prior_last_motion_y = last_motion_y;
1090
                    last_motion_x = motion_x[0];
1091
                    last_motion_y = motion_y[0];
1092
                    break;
1093

    
1094
                default:
1095
                    /* covers intra, inter without MV, golden without MV */
1096
                    memset(motion_x, 0, 6 * sizeof(int));
1097
                    memset(motion_y, 0, 6 * sizeof(int));
1098

    
1099
                    /* no vector maintenance */
1100
                    break;
1101
                }
1102

    
1103
                /* assign the motion vectors to the correct fragments */
1104
                debug_vectors("    vectors for macroblock starting @ fragment %d (coding method %d):\n",
1105
                    current_fragment,
1106
                    s->macroblock_coding[current_macroblock]);
1107
                for (k = 0; k < 6; k++) {
1108
                    current_fragment =
1109
                        s->macroblock_fragments[current_macroblock * 6 + k];
1110
                    if (current_fragment == -1)
1111
                        continue;
1112
                    if (current_fragment >= s->fragment_count) {
1113
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1114
                            current_fragment, s->fragment_count);
1115
                        return 1;
1116
                    }
1117
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
1118
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
1119
                    debug_vectors("    vector %d: fragment %d = (%d, %d)\n",
1120
                        k, current_fragment, motion_x[k], motion_y[k]);
1121
                }
1122
            }
1123
        }
1124
    }
1125

    
1126
    return 0;
1127
}
1128

    
1129
/*
1130
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1131
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1132
 * data. This function unpacks all the VLCs for either the Y plane or both
1133
 * C planes, and is called for DC coefficients or different AC coefficient
1134
 * levels (since different coefficient types require different VLC tables.
1135
 *
1136
 * This function returns a residual eob run. E.g, if a particular token gave
1137
 * instructions to EOB the next 5 fragments and there were only 2 fragments
1138
 * left in the current fragment range, 3 would be returned so that it could
1139
 * be passed into the next call to this same function.
1140
 */
1141
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1142
                        VLC *table, int coeff_index,
1143
                        int first_fragment, int last_fragment,
1144
                        int eob_run)
1145
{
1146
    int i;
1147
    int token;
1148
    int zero_run = 0;
1149
    DCTELEM coeff = 0;
1150
    Vp3Fragment *fragment;
1151
    uint8_t *perm= s->scantable.permutated;
1152
    int bits_to_get;
1153

    
1154
    if ((first_fragment >= s->fragment_count) ||
1155
        (last_fragment >= s->fragment_count)) {
1156

    
1157
        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1158
            first_fragment, last_fragment);
1159
        return 0;
1160
    }
1161

    
1162
    for (i = first_fragment; i <= last_fragment; i++) {
1163

    
1164
        fragment = &s->all_fragments[s->coded_fragment_list[i]];
1165
        if (fragment->coeff_count > coeff_index)
1166
            continue;
1167

    
1168
        if (!eob_run) {
1169
            /* decode a VLC into a token */
1170
            token = get_vlc2(gb, table->table, 5, 3);
1171
            debug_vlc(" token = %2d, ", token);
1172
            /* use the token to get a zero run, a coefficient, and an eob run */
1173
            if (token <= 6) {
1174
                eob_run = eob_run_base[token];
1175
                if (eob_run_get_bits[token])
1176
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
1177
                coeff = zero_run = 0;
1178
            } else {
1179
                bits_to_get = coeff_get_bits[token];
1180
                if (!bits_to_get)
1181
                    coeff = coeff_tables[token][0];
1182
                else
1183
                    coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1184

    
1185
                zero_run = zero_run_base[token];
1186
                if (zero_run_get_bits[token])
1187
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
1188
            }
1189
        }
1190

    
1191
        if (!eob_run) {
1192
            fragment->coeff_count += zero_run;
1193
            if (fragment->coeff_count < 64){
1194
                fragment->next_coeff->coeff= coeff;
1195
                fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1196
                fragment->next_coeff->next= s->next_coeff;
1197
                s->next_coeff->next=NULL;
1198
                fragment->next_coeff= s->next_coeff++;
1199
            }
1200
            debug_vlc(" fragment %d coeff = %d\n",
1201
                s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1202
        } else {
1203
            fragment->coeff_count |= 128;
1204
            debug_vlc(" fragment %d eob with %d coefficients\n",
1205
                s->coded_fragment_list[i], fragment->coeff_count&127);
1206
            eob_run--;
1207
        }
1208
    }
1209

    
1210
    return eob_run;
1211
}
1212

    
1213
/*
1214
 * This function unpacks all of the DCT coefficient data from the
1215
 * bitstream.
1216
 */
1217
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1218
{
1219
    int i;
1220
    int dc_y_table;
1221
    int dc_c_table;
1222
    int ac_y_table;
1223
    int ac_c_table;
1224
    int residual_eob_run = 0;
1225

    
1226
    /* fetch the DC table indices */
1227
    dc_y_table = get_bits(gb, 4);
1228
    dc_c_table = get_bits(gb, 4);
1229

    
1230
    /* unpack the Y plane DC coefficients */
1231
    debug_vp3("  vp3: unpacking Y plane DC coefficients using table %d\n",
1232
        dc_y_table);
1233
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1234
        s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1235

    
1236
    /* unpack the C plane DC coefficients */
1237
    debug_vp3("  vp3: unpacking C plane DC coefficients using table %d\n",
1238
        dc_c_table);
1239
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1240
        s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1241

    
1242
    /* fetch the AC table indices */
1243
    ac_y_table = get_bits(gb, 4);
1244
    ac_c_table = get_bits(gb, 4);
1245

    
1246
    /* unpack the group 1 AC coefficients (coeffs 1-5) */
1247
    for (i = 1; i <= 5; i++) {
1248

    
1249
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1250
            i, ac_y_table);
1251
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1252
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1253

    
1254
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1255
            i, ac_c_table);
1256
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1257
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1258
    }
1259

    
1260
    /* unpack the group 2 AC coefficients (coeffs 6-14) */
1261
    for (i = 6; i <= 14; i++) {
1262

    
1263
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1264
            i, ac_y_table);
1265
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1266
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1267

    
1268
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1269
            i, ac_c_table);
1270
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1271
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1272
    }
1273

    
1274
    /* unpack the group 3 AC coefficients (coeffs 15-27) */
1275
    for (i = 15; i <= 27; i++) {
1276

    
1277
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1278
            i, ac_y_table);
1279
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1280
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1281

    
1282
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1283
            i, ac_c_table);
1284
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1285
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1286
    }
1287

    
1288
    /* unpack the group 4 AC coefficients (coeffs 28-63) */
1289
    for (i = 28; i <= 63; i++) {
1290

    
1291
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1292
            i, ac_y_table);
1293
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1294
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1295

    
1296
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1297
            i, ac_c_table);
1298
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1299
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1300
    }
1301

    
1302
    return 0;
1303
}
1304

    
1305
/*
1306
 * This function reverses the DC prediction for each coded fragment in
1307
 * the frame. Much of this function is adapted directly from the original
1308
 * VP3 source code.
1309
 */
1310
#define COMPATIBLE_FRAME(x) \
1311
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1312
#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1313
#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1314

    
1315
static void reverse_dc_prediction(Vp3DecodeContext *s,
1316
                                  int first_fragment,
1317
                                  int fragment_width,
1318
                                  int fragment_height)
1319
{
1320

    
1321
#define PUL 8
1322
#define PU 4
1323
#define PUR 2
1324
#define PL 1
1325

    
1326
    int x, y;
1327
    int i = first_fragment;
1328

    
1329
    int predicted_dc;
1330

    
1331
    /* DC values for the left, up-left, up, and up-right fragments */
1332
    int vl, vul, vu, vur;
1333

    
1334
    /* indices for the left, up-left, up, and up-right fragments */
1335
    int l, ul, u, ur;
1336

    
1337
    /*
1338
     * The 6 fields mean:
1339
     *   0: up-left multiplier
1340
     *   1: up multiplier
1341
     *   2: up-right multiplier
1342
     *   3: left multiplier
1343
     */
1344
    int predictor_transform[16][4] = {
1345
        {  0,  0,  0,  0},
1346
        {  0,  0,  0,128},        // PL
1347
        {  0,  0,128,  0},        // PUR
1348
        {  0,  0, 53, 75},        // PUR|PL
1349
        {  0,128,  0,  0},        // PU
1350
        {  0, 64,  0, 64},        // PU|PL
1351
        {  0,128,  0,  0},        // PU|PUR
1352
        {  0,  0, 53, 75},        // PU|PUR|PL
1353
        {128,  0,  0,  0},        // PUL
1354
        {  0,  0,  0,128},        // PUL|PL
1355
        { 64,  0, 64,  0},        // PUL|PUR
1356
        {  0,  0, 53, 75},        // PUL|PUR|PL
1357
        {  0,128,  0,  0},        // PUL|PU
1358
       {-104,116,  0,116},        // PUL|PU|PL
1359
        { 24, 80, 24,  0},        // PUL|PU|PUR
1360
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1361
    };
1362

    
1363
    /* This table shows which types of blocks can use other blocks for
1364
     * prediction. For example, INTRA is the only mode in this table to
1365
     * have a frame number of 0. That means INTRA blocks can only predict
1366
     * from other INTRA blocks. There are 2 golden frame coding types;
1367
     * blocks encoding in these modes can only predict from other blocks
1368
     * that were encoded with these 1 of these 2 modes. */
1369
    unsigned char compatible_frame[8] = {
1370
        1,    /* MODE_INTER_NO_MV */
1371
        0,    /* MODE_INTRA */
1372
        1,    /* MODE_INTER_PLUS_MV */
1373
        1,    /* MODE_INTER_LAST_MV */
1374
        1,    /* MODE_INTER_PRIOR_MV */
1375
        2,    /* MODE_USING_GOLDEN */
1376
        2,    /* MODE_GOLDEN_MV */
1377
        1     /* MODE_INTER_FOUR_MV */
1378
    };
1379
    int current_frame_type;
1380

    
1381
    /* there is a last DC predictor for each of the 3 frame types */
1382
    short last_dc[3];
1383

    
1384
    int transform = 0;
1385

    
1386
    debug_vp3("  vp3: reversing DC prediction\n");
1387

    
1388
    vul = vu = vur = vl = 0;
1389
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1390

    
1391
    /* for each fragment row... */
1392
    for (y = 0; y < fragment_height; y++) {
1393

    
1394
        /* for each fragment in a row... */
1395
        for (x = 0; x < fragment_width; x++, i++) {
1396

    
1397
            /* reverse prediction if this block was coded */
1398
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1399

    
1400
                current_frame_type =
1401
                    compatible_frame[s->all_fragments[i].coding_method];
1402
                debug_dc_pred(" frag %d: orig DC = %d, ",
1403
                    i, DC_COEFF(i));
1404

    
1405
                transform= 0;
1406
                if(x){
1407
                    l= i-1;
1408
                    vl = DC_COEFF(l);
1409
                    if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1410
                        transform |= PL;
1411
                }
1412
                if(y){
1413
                    u= i-fragment_width;
1414
                    vu = DC_COEFF(u);
1415
                    if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1416
                        transform |= PU;
1417
                    if(x){
1418
                        ul= i-fragment_width-1;
1419
                        vul = DC_COEFF(ul);
1420
                        if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1421
                            transform |= PUL;
1422
                    }
1423
                    if(x + 1 < fragment_width){
1424
                        ur= i-fragment_width+1;
1425
                        vur = DC_COEFF(ur);
1426
                        if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1427
                            transform |= PUR;
1428
                    }
1429
                }
1430

    
1431
                debug_dc_pred("transform = %d, ", transform);
1432

    
1433
                if (transform == 0) {
1434

    
1435
                    /* if there were no fragments to predict from, use last
1436
                     * DC saved */
1437
                    predicted_dc = last_dc[current_frame_type];
1438
                    debug_dc_pred("from last DC (%d) = %d\n",
1439
                        current_frame_type, DC_COEFF(i));
1440

    
1441
                } else {
1442

    
1443
                    /* apply the appropriate predictor transform */
1444
                    predicted_dc =
1445
                        (predictor_transform[transform][0] * vul) +
1446
                        (predictor_transform[transform][1] * vu) +
1447
                        (predictor_transform[transform][2] * vur) +
1448
                        (predictor_transform[transform][3] * vl);
1449

    
1450
                    predicted_dc /= 128;
1451

    
1452
                    /* check for outranging on the [ul u l] and
1453
                     * [ul u ur l] predictors */
1454
                    if ((transform == 13) || (transform == 15)) {
1455
                        if (FFABS(predicted_dc - vu) > 128)
1456
                            predicted_dc = vu;
1457
                        else if (FFABS(predicted_dc - vl) > 128)
1458
                            predicted_dc = vl;
1459
                        else if (FFABS(predicted_dc - vul) > 128)
1460
                            predicted_dc = vul;
1461
                    }
1462

    
1463
                    debug_dc_pred("from pred DC = %d\n",
1464
                    DC_COEFF(i));
1465
                }
1466

    
1467
                /* at long last, apply the predictor */
1468
                if(s->coeffs[i].index){
1469
                    *s->next_coeff= s->coeffs[i];
1470
                    s->coeffs[i].index=0;
1471
                    s->coeffs[i].coeff=0;
1472
                    s->coeffs[i].next= s->next_coeff++;
1473
                }
1474
                s->coeffs[i].coeff += predicted_dc;
1475
                /* save the DC */
1476
                last_dc[current_frame_type] = DC_COEFF(i);
1477
                if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1478
                    s->all_fragments[i].coeff_count= 129;
1479
//                    s->all_fragments[i].next_coeff= s->next_coeff;
1480
                    s->coeffs[i].next= s->next_coeff;
1481
                    (s->next_coeff++)->next=NULL;
1482
                }
1483
            }
1484
        }
1485
    }
1486
}
1487

    
1488

    
1489
static void horizontal_filter(unsigned char *first_pixel, int stride,
1490
    int *bounding_values);
1491
static void vertical_filter(unsigned char *first_pixel, int stride,
1492
    int *bounding_values);
1493

    
1494
/*
1495
 * Perform the final rendering for a particular slice of data.
1496
 * The slice number ranges from 0..(macroblock_height - 1).
1497
 */
1498
static void render_slice(Vp3DecodeContext *s, int slice)
1499
{
1500
    int x;
1501
    int m, n;
1502
    int16_t *dequantizer;
1503
    DECLARE_ALIGNED_16(DCTELEM, block[64]);
1504
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1505
    int motion_halfpel_index;
1506
    uint8_t *motion_source;
1507
    int plane;
1508
    int current_macroblock_entry = slice * s->macroblock_width * 6;
1509

    
1510
    if (slice >= s->macroblock_height)
1511
        return;
1512

    
1513
    for (plane = 0; plane < 3; plane++) {
1514
        uint8_t *output_plane = s->current_frame.data    [plane];
1515
        uint8_t *  last_plane = s->   last_frame.data    [plane];
1516
        uint8_t *golden_plane = s-> golden_frame.data    [plane];
1517
        int stride            = s->current_frame.linesize[plane];
1518
        int plane_width       = s->width  >> !!plane;
1519
        int plane_height      = s->height >> !!plane;
1520
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
1521
        int slice_height = y + (FRAGMENT_PIXELS << !plane);
1522
        int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1523

    
1524
        if (!s->flipped_image) stride = -stride;
1525

    
1526

    
1527
        if(FFABS(stride) > 2048)
1528
            return; //various tables are fixed size
1529

    
1530
        /* for each fragment row in the slice (both of them)... */
1531
        for (; y < slice_height; y += 8) {
1532

    
1533
            /* for each fragment in a row... */
1534
            for (x = 0; x < plane_width; x += 8, i++) {
1535

    
1536
                if ((i < 0) || (i >= s->fragment_count)) {
1537
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
1538
                    return;
1539
                }
1540

    
1541
                /* transform if this block was coded */
1542
                if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1543
                    !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1544

    
1545
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1546
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1547
                        motion_source= golden_plane;
1548
                    else
1549
                        motion_source= last_plane;
1550

    
1551
                    motion_source += s->all_fragments[i].first_pixel;
1552
                    motion_halfpel_index = 0;
1553

    
1554
                    /* sort out the motion vector if this fragment is coded
1555
                     * using a motion vector method */
1556
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1557
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1558
                        int src_x, src_y;
1559
                        motion_x = s->all_fragments[i].motion_x;
1560
                        motion_y = s->all_fragments[i].motion_y;
1561
                        if(plane){
1562
                            motion_x= (motion_x>>1) | (motion_x&1);
1563
                            motion_y= (motion_y>>1) | (motion_y&1);
1564
                        }
1565

    
1566
                        src_x= (motion_x>>1) + x;
1567
                        src_y= (motion_y>>1) + y;
1568
                        if ((motion_x == 127) || (motion_y == 127))
1569
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1570

    
1571
                        motion_halfpel_index = motion_x & 0x01;
1572
                        motion_source += (motion_x >> 1);
1573

    
1574
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1575
                        motion_source += ((motion_y >> 1) * stride);
1576

    
1577
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1578
                            uint8_t *temp= s->edge_emu_buffer;
1579
                            if(stride<0) temp -= 9*stride;
1580
                            else temp += 9*stride;
1581

    
1582
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1583
                            motion_source= temp;
1584
                        }
1585
                    }
1586

    
1587

    
1588
                    /* first, take care of copying a block from either the
1589
                     * previous or the golden frame */
1590
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1591
                        /* Note, it is possible to implement all MC cases with
1592
                           put_no_rnd_pixels_l2 which would look more like the
1593
                           VP3 source but this would be slower as
1594
                           put_no_rnd_pixels_tab is better optimzed */
1595
                        if(motion_halfpel_index != 3){
1596
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1597
                                output_plane + s->all_fragments[i].first_pixel,
1598
                                motion_source, stride, 8);
1599
                        }else{
1600
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1601
                            s->dsp.put_no_rnd_pixels_l2[1](
1602
                                output_plane + s->all_fragments[i].first_pixel,
1603
                                motion_source - d,
1604
                                motion_source + stride + 1 + d,
1605
                                stride, 8);
1606
                        }
1607
                        dequantizer = s->qmat[1][plane];
1608
                    }else{
1609
                        dequantizer = s->qmat[0][plane];
1610
                    }
1611

    
1612
                    /* dequantize the DCT coefficients */
1613
                    debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1614
                        i, s->all_fragments[i].coding_method,
1615
                        DC_COEFF(i), dequantizer[0]);
1616

    
1617
                    if(s->avctx->idct_algo==FF_IDCT_VP3){
1618
                        Coeff *coeff= s->coeffs + i;
1619
                        memset(block, 0, sizeof(block));
1620
                        while(coeff->next){
1621
                            block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1622
                            coeff= coeff->next;
1623
                        }
1624
                    }else{
1625
                        Coeff *coeff= s->coeffs + i;
1626
                        memset(block, 0, sizeof(block));
1627
                        while(coeff->next){
1628
                            block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1629
                            coeff= coeff->next;
1630
                        }
1631
                    }
1632

    
1633
                    /* invert DCT and place (or add) in final output */
1634

    
1635
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1636
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1637
                            block[0] += 128<<3;
1638
                        s->dsp.idct_put(
1639
                            output_plane + s->all_fragments[i].first_pixel,
1640
                            stride,
1641
                            block);
1642
                    } else {
1643
                        s->dsp.idct_add(
1644
                            output_plane + s->all_fragments[i].first_pixel,
1645
                            stride,
1646
                            block);
1647
                    }
1648

    
1649
                    debug_idct("block after idct_%s():\n",
1650
                        (s->all_fragments[i].coding_method == MODE_INTRA)?
1651
                        "put" : "add");
1652
                    for (m = 0; m < 8; m++) {
1653
                        for (n = 0; n < 8; n++) {
1654
                            debug_idct(" %3d", *(output_plane +
1655
                                s->all_fragments[i].first_pixel + (m * stride + n)));
1656
                        }
1657
                        debug_idct("\n");
1658
                    }
1659
                    debug_idct("\n");
1660

    
1661
                } else {
1662

    
1663
                    /* copy directly from the previous frame */
1664
                    s->dsp.put_pixels_tab[1][0](
1665
                        output_plane + s->all_fragments[i].first_pixel,
1666
                        last_plane + s->all_fragments[i].first_pixel,
1667
                        stride, 8);
1668

    
1669
                }
1670
#if 0
1671
                /* perform the left edge filter if:
1672
                 *   - the fragment is not on the left column
1673
                 *   - the fragment is coded in this frame
1674
                 *   - the fragment is not coded in this frame but the left
1675
                 *     fragment is coded in this frame (this is done instead
1676
                 *     of a right edge filter when rendering the left fragment
1677
                 *     since this fragment is not available yet) */
1678
                if ((x > 0) &&
1679
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1680
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1681
                      (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1682
                    horizontal_filter(
1683
                        output_plane + s->all_fragments[i].first_pixel + 7*stride,
1684
                        -stride, s->bounding_values_array + 127);
1685
                }
1686

1687
                /* perform the top edge filter if:
1688
                 *   - the fragment is not on the top row
1689
                 *   - the fragment is coded in this frame
1690
                 *   - the fragment is not coded in this frame but the above
1691
                 *     fragment is coded in this frame (this is done instead
1692
                 *     of a bottom edge filter when rendering the above
1693
                 *     fragment since this fragment is not available yet) */
1694
                if ((y > 0) &&
1695
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1696
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1697
                      (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1698
                    vertical_filter(
1699
                        output_plane + s->all_fragments[i].first_pixel - stride,
1700
                        -stride, s->bounding_values_array + 127);
1701
                }
1702
#endif
1703
            }
1704
        }
1705
    }
1706

    
1707
     /* this looks like a good place for slice dispatch... */
1708
     /* algorithm:
1709
      *   if (slice == s->macroblock_height - 1)
1710
      *     dispatch (both last slice & 2nd-to-last slice);
1711
      *   else if (slice > 0)
1712
      *     dispatch (slice - 1);
1713
      */
1714

    
1715
    emms_c();
1716
}
1717

    
1718
static void horizontal_filter(unsigned char *first_pixel, int stride,
1719
    int *bounding_values)
1720
{
1721
    unsigned char *end;
1722
    int filter_value;
1723

    
1724
    for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1725
        filter_value =
1726
            (first_pixel[-2] - first_pixel[ 1])
1727
         +3*(first_pixel[ 0] - first_pixel[-1]);
1728
        filter_value = bounding_values[(filter_value + 4) >> 3];
1729
        first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
1730
        first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
1731
    }
1732
}
1733

    
1734
static void vertical_filter(unsigned char *first_pixel, int stride,
1735
    int *bounding_values)
1736
{
1737
    unsigned char *end;
1738
    int filter_value;
1739
    const int nstride= -stride;
1740

    
1741
    for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1742
        filter_value =
1743
            (first_pixel[2 * nstride] - first_pixel[ stride])
1744
         +3*(first_pixel[0          ] - first_pixel[nstride]);
1745
        filter_value = bounding_values[(filter_value + 4) >> 3];
1746
        first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value);
1747
        first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value);
1748
    }
1749
}
1750

    
1751
static void apply_loop_filter(Vp3DecodeContext *s)
1752
{
1753
    int plane;
1754
    int x, y;
1755
    int *bounding_values= s->bounding_values_array+127;
1756

    
1757
#if 0
1758
    int bounding_values_array[256];
1759
    int filter_limit;
1760

1761
    /* find the right loop limit value */
1762
    for (x = 63; x >= 0; x--) {
1763
        if (vp31_ac_scale_factor[x] >= s->quality_index)
1764
            break;
1765
    }
1766
    filter_limit = vp31_filter_limit_values[s->quality_index];
1767

1768
    /* set up the bounding values */
1769
    memset(bounding_values_array, 0, 256 * sizeof(int));
1770
    for (x = 0; x < filter_limit; x++) {
1771
        bounding_values[-x - filter_limit] = -filter_limit + x;
1772
        bounding_values[-x] = -x;
1773
        bounding_values[x] = x;
1774
        bounding_values[x + filter_limit] = filter_limit - x;
1775
    }
1776
#endif
1777

    
1778
    for (plane = 0; plane < 3; plane++) {
1779
        int width           = s->fragment_width  >> !!plane;
1780
        int height          = s->fragment_height >> !!plane;
1781
        int fragment        = s->fragment_start        [plane];
1782
        int stride          = s->current_frame.linesize[plane];
1783
        uint8_t *plane_data = s->current_frame.data    [plane];
1784
        if (!s->flipped_image) stride = -stride;
1785

    
1786
        for (y = 0; y < height; y++) {
1787

    
1788
            for (x = 0; x < width; x++) {
1789
START_TIMER
1790
                /* do not perform left edge filter for left columns frags */
1791
                if ((x > 0) &&
1792
                    (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1793
                    horizontal_filter(
1794
                        plane_data + s->all_fragments[fragment].first_pixel,
1795
                        stride, bounding_values);
1796
                }
1797

    
1798
                /* do not perform top edge filter for top row fragments */
1799
                if ((y > 0) &&
1800
                    (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1801
                    vertical_filter(
1802
                        plane_data + s->all_fragments[fragment].first_pixel,
1803
                        stride, bounding_values);
1804
                }
1805

    
1806
                /* do not perform right edge filter for right column
1807
                 * fragments or if right fragment neighbor is also coded
1808
                 * in this frame (it will be filtered in next iteration) */
1809
                if ((x < width - 1) &&
1810
                    (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1811
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1812
                    horizontal_filter(
1813
                        plane_data + s->all_fragments[fragment + 1].first_pixel,
1814
                        stride, bounding_values);
1815
                }
1816

    
1817
                /* do not perform bottom edge filter for bottom row
1818
                 * fragments or if bottom fragment neighbor is also coded
1819
                 * in this frame (it will be filtered in the next row) */
1820
                if ((y < height - 1) &&
1821
                    (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1822
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1823
                    vertical_filter(
1824
                        plane_data + s->all_fragments[fragment + width].first_pixel,
1825
                        stride, bounding_values);
1826
                }
1827

    
1828
                fragment++;
1829
STOP_TIMER("loop filter")
1830
            }
1831
        }
1832
    }
1833
}
1834

    
1835
/*
1836
 * This function computes the first pixel addresses for each fragment.
1837
 * This function needs to be invoked after the first frame is allocated
1838
 * so that it has access to the plane strides.
1839
 */
1840
static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1841
{
1842

    
1843
    int i, x, y;
1844

    
1845
    /* figure out the first pixel addresses for each of the fragments */
1846
    /* Y plane */
1847
    i = 0;
1848
    for (y = s->fragment_height; y > 0; y--) {
1849
        for (x = 0; x < s->fragment_width; x++) {
1850
            s->all_fragments[i++].first_pixel =
1851
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1852
                    s->golden_frame.linesize[0] +
1853
                    x * FRAGMENT_PIXELS;
1854
            debug_init("  fragment %d, first pixel @ %d\n",
1855
                i-1, s->all_fragments[i-1].first_pixel);
1856
        }
1857
    }
1858

    
1859
    /* U plane */
1860
    i = s->fragment_start[1];
1861
    for (y = s->fragment_height / 2; y > 0; y--) {
1862
        for (x = 0; x < s->fragment_width / 2; x++) {
1863
            s->all_fragments[i++].first_pixel =
1864
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1865
                    s->golden_frame.linesize[1] +
1866
                    x * FRAGMENT_PIXELS;
1867
            debug_init("  fragment %d, first pixel @ %d\n",
1868
                i-1, s->all_fragments[i-1].first_pixel);
1869
        }
1870
    }
1871

    
1872
    /* V plane */
1873
    i = s->fragment_start[2];
1874
    for (y = s->fragment_height / 2; y > 0; y--) {
1875
        for (x = 0; x < s->fragment_width / 2; x++) {
1876
            s->all_fragments[i++].first_pixel =
1877
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1878
                    s->golden_frame.linesize[2] +
1879
                    x * FRAGMENT_PIXELS;
1880
            debug_init("  fragment %d, first pixel @ %d\n",
1881
                i-1, s->all_fragments[i-1].first_pixel);
1882
        }
1883
    }
1884
}
1885

    
1886
/* FIXME: this should be merged with the above! */
1887
static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
1888
{
1889

    
1890
    int i, x, y;
1891

    
1892
    /* figure out the first pixel addresses for each of the fragments */
1893
    /* Y plane */
1894
    i = 0;
1895
    for (y = 1; y <= s->fragment_height; y++) {
1896
        for (x = 0; x < s->fragment_width; x++) {
1897
            s->all_fragments[i++].first_pixel =
1898
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1899
                    s->golden_frame.linesize[0] +
1900
                    x * FRAGMENT_PIXELS;
1901
            debug_init("  fragment %d, first pixel @ %d\n",
1902
                i-1, s->all_fragments[i-1].first_pixel);
1903
        }
1904
    }
1905

    
1906
    /* U plane */
1907
    i = s->fragment_start[1];
1908
    for (y = 1; y <= s->fragment_height / 2; y++) {
1909
        for (x = 0; x < s->fragment_width / 2; x++) {
1910
            s->all_fragments[i++].first_pixel =
1911
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1912
                    s->golden_frame.linesize[1] +
1913
                    x * FRAGMENT_PIXELS;
1914
            debug_init("  fragment %d, first pixel @ %d\n",
1915
                i-1, s->all_fragments[i-1].first_pixel);
1916
        }
1917
    }
1918

    
1919
    /* V plane */
1920
    i = s->fragment_start[2];
1921
    for (y = 1; y <= s->fragment_height / 2; y++) {
1922
        for (x = 0; x < s->fragment_width / 2; x++) {
1923
            s->all_fragments[i++].first_pixel =
1924
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1925
                    s->golden_frame.linesize[2] +
1926
                    x * FRAGMENT_PIXELS;
1927
            debug_init("  fragment %d, first pixel @ %d\n",
1928
                i-1, s->all_fragments[i-1].first_pixel);
1929
        }
1930
    }
1931
}
1932

    
1933
/*
1934
 * This is the ffmpeg/libavcodec API init function.
1935
 */
1936
static int vp3_decode_init(AVCodecContext *avctx)
1937
{
1938
    Vp3DecodeContext *s = avctx->priv_data;
1939
    int i, inter, plane;
1940
    int c_width;
1941
    int c_height;
1942
    int y_superblock_count;
1943
    int c_superblock_count;
1944

    
1945
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1946
        s->version = 0;
1947
    else
1948
        s->version = 1;
1949

    
1950
    s->avctx = avctx;
1951
    s->width = (avctx->width + 15) & 0xFFFFFFF0;
1952
    s->height = (avctx->height + 15) & 0xFFFFFFF0;
1953
    avctx->pix_fmt = PIX_FMT_YUV420P;
1954
    if(avctx->idct_algo==FF_IDCT_AUTO)
1955
        avctx->idct_algo=FF_IDCT_VP3;
1956
    dsputil_init(&s->dsp, avctx);
1957

    
1958
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1959

    
1960
    /* initialize to an impossible value which will force a recalculation
1961
     * in the first frame decode */
1962
    s->quality_index = -1;
1963

    
1964
    s->y_superblock_width = (s->width + 31) / 32;
1965
    s->y_superblock_height = (s->height + 31) / 32;
1966
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1967

    
1968
    /* work out the dimensions for the C planes */
1969
    c_width = s->width / 2;
1970
    c_height = s->height / 2;
1971
    s->c_superblock_width = (c_width + 31) / 32;
1972
    s->c_superblock_height = (c_height + 31) / 32;
1973
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1974

    
1975
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1976
    s->u_superblock_start = y_superblock_count;
1977
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1978
    s->superblock_coding = av_malloc(s->superblock_count);
1979

    
1980
    s->macroblock_width = (s->width + 15) / 16;
1981
    s->macroblock_height = (s->height + 15) / 16;
1982
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1983

    
1984
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1985
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1986

    
1987
    /* fragment count covers all 8x8 blocks for all 3 planes */
1988
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1989
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1990
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1991

    
1992
    debug_init("  Y plane: %d x %d\n", s->width, s->height);
1993
    debug_init("  C plane: %d x %d\n", c_width, c_height);
1994
    debug_init("  Y superblocks: %d x %d, %d total\n",
1995
        s->y_superblock_width, s->y_superblock_height, y_superblock_count);
1996
    debug_init("  C superblocks: %d x %d, %d total\n",
1997
        s->c_superblock_width, s->c_superblock_height, c_superblock_count);
1998
    debug_init("  total superblocks = %d, U starts @ %d, V starts @ %d\n",
1999
        s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2000
    debug_init("  macroblocks: %d x %d, %d total\n",
2001
        s->macroblock_width, s->macroblock_height, s->macroblock_count);
2002
    debug_init("  %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2003
        s->fragment_count,
2004
        s->fragment_width,
2005
        s->fragment_height,
2006
        s->fragment_start[1],
2007
        s->fragment_start[2]);
2008

    
2009
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2010
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2011
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2012
    s->pixel_addresses_inited = 0;
2013

    
2014
    if (!s->theora_tables)
2015
    {
2016
        for (i = 0; i < 64; i++) {
2017
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2018
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2019
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2020
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2021
            s->base_matrix[2][i] = vp31_inter_dequant[i];
2022
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
2023
        }
2024

    
2025
        for(inter=0; inter<2; inter++){
2026
            for(plane=0; plane<3; plane++){
2027
                s->qr_count[inter][plane]= 1;
2028
                s->qr_size [inter][plane][0]= 63;
2029
                s->qr_base [inter][plane][0]=
2030
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2031
            }
2032
        }
2033

    
2034
        /* init VLC tables */
2035
        for (i = 0; i < 16; i++) {
2036

    
2037
            /* DC histograms */
2038
            init_vlc(&s->dc_vlc[i], 5, 32,
2039
                &dc_bias[i][0][1], 4, 2,
2040
                &dc_bias[i][0][0], 4, 2, 0);
2041

    
2042
            /* group 1 AC histograms */
2043
            init_vlc(&s->ac_vlc_1[i], 5, 32,
2044
                &ac_bias_0[i][0][1], 4, 2,
2045
                &ac_bias_0[i][0][0], 4, 2, 0);
2046

    
2047
            /* group 2 AC histograms */
2048
            init_vlc(&s->ac_vlc_2[i], 5, 32,
2049
                &ac_bias_1[i][0][1], 4, 2,
2050
                &ac_bias_1[i][0][0], 4, 2, 0);
2051

    
2052
            /* group 3 AC histograms */
2053
            init_vlc(&s->ac_vlc_3[i], 5, 32,
2054
                &ac_bias_2[i][0][1], 4, 2,
2055
                &ac_bias_2[i][0][0], 4, 2, 0);
2056

    
2057
            /* group 4 AC histograms */
2058
            init_vlc(&s->ac_vlc_4[i], 5, 32,
2059
                &ac_bias_3[i][0][1], 4, 2,
2060
                &ac_bias_3[i][0][0], 4, 2, 0);
2061
        }
2062
    } else {
2063
        for (i = 0; i < 16; i++) {
2064

    
2065
            /* DC histograms */
2066
            init_vlc(&s->dc_vlc[i], 5, 32,
2067
                &s->huffman_table[i][0][1], 4, 2,
2068
                &s->huffman_table[i][0][0], 4, 2, 0);
2069

    
2070
            /* group 1 AC histograms */
2071
            init_vlc(&s->ac_vlc_1[i], 5, 32,
2072
                &s->huffman_table[i+16][0][1], 4, 2,
2073
                &s->huffman_table[i+16][0][0], 4, 2, 0);
2074

    
2075
            /* group 2 AC histograms */
2076
            init_vlc(&s->ac_vlc_2[i], 5, 32,
2077
                &s->huffman_table[i+16*2][0][1], 4, 2,
2078
                &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2079

    
2080
            /* group 3 AC histograms */
2081
            init_vlc(&s->ac_vlc_3[i], 5, 32,
2082
                &s->huffman_table[i+16*3][0][1], 4, 2,
2083
                &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2084

    
2085
            /* group 4 AC histograms */
2086
            init_vlc(&s->ac_vlc_4[i], 5, 32,
2087
                &s->huffman_table[i+16*4][0][1], 4, 2,
2088
                &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2089
        }
2090
    }
2091

    
2092
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
2093
        &superblock_run_length_vlc_table[0][1], 4, 2,
2094
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2095

    
2096
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
2097
        &fragment_run_length_vlc_table[0][1], 4, 2,
2098
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2099

    
2100
    init_vlc(&s->mode_code_vlc, 3, 8,
2101
        &mode_code_vlc_table[0][1], 2, 1,
2102
        &mode_code_vlc_table[0][0], 2, 1, 0);
2103

    
2104
    init_vlc(&s->motion_vector_vlc, 6, 63,
2105
        &motion_vector_vlc_table[0][1], 2, 1,
2106
        &motion_vector_vlc_table[0][0], 2, 1, 0);
2107

    
2108
    /* work out the block mapping tables */
2109
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2110
    s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2111
    s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2112
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2113
    init_block_mapping(s);
2114

    
2115
    for (i = 0; i < 3; i++) {
2116
        s->current_frame.data[i] = NULL;
2117
        s->last_frame.data[i] = NULL;
2118
        s->golden_frame.data[i] = NULL;
2119
    }
2120

    
2121
    return 0;
2122
}
2123

    
2124
/*
2125
 * This is the ffmpeg/libavcodec API frame decode function.
2126
 */
2127
static int vp3_decode_frame(AVCodecContext *avctx,
2128
                            void *data, int *data_size,
2129
                            const uint8_t *buf, int buf_size)
2130
{
2131
    Vp3DecodeContext *s = avctx->priv_data;
2132
    GetBitContext gb;
2133
    static int counter = 0;
2134
    int i;
2135

    
2136
    init_get_bits(&gb, buf, buf_size * 8);
2137

    
2138
    if (s->theora && get_bits1(&gb))
2139
    {
2140
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2141
        return -1;
2142
    }
2143

    
2144
    s->keyframe = !get_bits1(&gb);
2145
    if (!s->theora)
2146
        skip_bits(&gb, 1);
2147
    s->last_quality_index = s->quality_index;
2148

    
2149
    s->nqis=0;
2150
    do{
2151
        s->qis[s->nqis++]= get_bits(&gb, 6);
2152
    } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
2153

    
2154
    s->quality_index= s->qis[0];
2155

    
2156
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2157
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2158
            s->keyframe?"key":"", counter, s->quality_index);
2159
    counter++;
2160

    
2161
    if (s->quality_index != s->last_quality_index) {
2162
        init_dequantizer(s);
2163
        init_loop_filter(s);
2164
    }
2165

    
2166
    if (s->keyframe) {
2167
        if (!s->theora)
2168
        {
2169
            skip_bits(&gb, 4); /* width code */
2170
            skip_bits(&gb, 4); /* height code */
2171
            if (s->version)
2172
            {
2173
                s->version = get_bits(&gb, 5);
2174
                if (counter == 1)
2175
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2176
            }
2177
        }
2178
        if (s->version || s->theora)
2179
        {
2180
                if (get_bits1(&gb))
2181
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2182
            skip_bits(&gb, 2); /* reserved? */
2183
        }
2184

    
2185
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2186
            if (s->golden_frame.data[0])
2187
                avctx->release_buffer(avctx, &s->golden_frame);
2188
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2189
        } else {
2190
            if (s->golden_frame.data[0])
2191
                avctx->release_buffer(avctx, &s->golden_frame);
2192
            if (s->last_frame.data[0])
2193
                avctx->release_buffer(avctx, &s->last_frame);
2194
        }
2195

    
2196
        s->golden_frame.reference = 3;
2197
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2198
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2199
            return -1;
2200
        }
2201

    
2202
        /* golden frame is also the current frame */
2203
        s->current_frame= s->golden_frame;
2204

    
2205
        /* time to figure out pixel addresses? */
2206
        if (!s->pixel_addresses_inited)
2207
        {
2208
            if (!s->flipped_image)
2209
                vp3_calculate_pixel_addresses(s);
2210
            else
2211
                theora_calculate_pixel_addresses(s);
2212
            s->pixel_addresses_inited = 1;
2213
        }
2214
    } else {
2215
        /* allocate a new current frame */
2216
        s->current_frame.reference = 3;
2217
        if (!s->pixel_addresses_inited) {
2218
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2219
            return -1;
2220
        }
2221
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2222
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2223
            return -1;
2224
        }
2225
    }
2226

    
2227
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2228
    s->current_frame.qstride= 0;
2229

    
2230
    {START_TIMER
2231
    init_frame(s, &gb);
2232
    STOP_TIMER("init_frame")}
2233

    
2234
#if KEYFRAMES_ONLY
2235
if (!s->keyframe) {
2236

    
2237
    memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2238
        s->current_frame.linesize[0] * s->height);
2239
    memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2240
        s->current_frame.linesize[1] * s->height / 2);
2241
    memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2242
        s->current_frame.linesize[2] * s->height / 2);
2243

    
2244
} else {
2245
#endif
2246

    
2247
    {START_TIMER
2248
    if (unpack_superblocks(s, &gb)){
2249
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2250
        return -1;
2251
    }
2252
    STOP_TIMER("unpack_superblocks")}
2253
    {START_TIMER
2254
    if (unpack_modes(s, &gb)){
2255
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2256
        return -1;
2257
    }
2258
    STOP_TIMER("unpack_modes")}
2259
    {START_TIMER
2260
    if (unpack_vectors(s, &gb)){
2261
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2262
        return -1;
2263
    }
2264
    STOP_TIMER("unpack_vectors")}
2265
    {START_TIMER
2266
    if (unpack_dct_coeffs(s, &gb)){
2267
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2268
        return -1;
2269
    }
2270
    STOP_TIMER("unpack_dct_coeffs")}
2271
    {START_TIMER
2272

    
2273
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2274
    if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2275
        reverse_dc_prediction(s, s->fragment_start[1],
2276
            s->fragment_width / 2, s->fragment_height / 2);
2277
        reverse_dc_prediction(s, s->fragment_start[2],
2278
            s->fragment_width / 2, s->fragment_height / 2);
2279
    }
2280
    STOP_TIMER("reverse_dc_prediction")}
2281
    {START_TIMER
2282

    
2283
    for (i = 0; i < s->macroblock_height; i++)
2284
        render_slice(s, i);
2285
    STOP_TIMER("render_fragments")}
2286

    
2287
    {START_TIMER
2288
    apply_loop_filter(s);
2289
    STOP_TIMER("apply_loop_filter")}
2290
#if KEYFRAMES_ONLY
2291
}
2292
#endif
2293

    
2294
    *data_size=sizeof(AVFrame);
2295
    *(AVFrame*)data= s->current_frame;
2296

    
2297
    /* release the last frame, if it is allocated and if it is not the
2298
     * golden frame */
2299
    if ((s->last_frame.data[0]) &&
2300
        (s->last_frame.data[0] != s->golden_frame.data[0]))
2301
        avctx->release_buffer(avctx, &s->last_frame);
2302

    
2303
    /* shuffle frames (last = current) */
2304
    s->last_frame= s->current_frame;
2305
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2306

    
2307
    return buf_size;
2308
}
2309

    
2310
/*
2311
 * This is the ffmpeg/libavcodec API module cleanup function.
2312
 */
2313
static int vp3_decode_end(AVCodecContext *avctx)
2314
{
2315
    Vp3DecodeContext *s = avctx->priv_data;
2316

    
2317
    av_free(s->all_fragments);
2318
    av_free(s->coeffs);
2319
    av_free(s->coded_fragment_list);
2320
    av_free(s->superblock_fragments);
2321
    av_free(s->superblock_macroblocks);
2322
    av_free(s->macroblock_fragments);
2323
    av_free(s->macroblock_coding);
2324

    
2325
    /* release all frames */
2326
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2327
        avctx->release_buffer(avctx, &s->golden_frame);
2328
    if (s->last_frame.data[0])
2329
        avctx->release_buffer(avctx, &s->last_frame);
2330
    /* no need to release the current_frame since it will always be pointing
2331
     * to the same frame as either the golden or last frame */
2332

    
2333
    return 0;
2334
}
2335

    
2336
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2337
{
2338
    Vp3DecodeContext *s = avctx->priv_data;
2339

    
2340
    if (get_bits1(gb)) {
2341
        int token;
2342
        if (s->entries >= 32) { /* overflow */
2343
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2344
            return -1;
2345
        }
2346
        token = get_bits(gb, 5);
2347
        //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);
2348
        s->huffman_table[s->hti][token][0] = s->hbits;
2349
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
2350
        s->entries++;
2351
    }
2352
    else {
2353
        if (s->huff_code_size >= 32) {/* overflow */
2354
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2355
            return -1;
2356
        }
2357
        s->huff_code_size++;
2358
        s->hbits <<= 1;
2359
        read_huffman_tree(avctx, gb);
2360
        s->hbits |= 1;
2361
        read_huffman_tree(avctx, gb);
2362
        s->hbits >>= 1;
2363
        s->huff_code_size--;
2364
    }
2365
    return 0;
2366
}
2367

    
2368
#ifdef CONFIG_THEORA_DECODER
2369
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2370
{
2371
    Vp3DecodeContext *s = avctx->priv_data;
2372
    int visible_width, visible_height;
2373

    
2374
    s->theora = get_bits_long(gb, 24);
2375
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
2376

    
2377
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2378
    /* but previous versions have the image flipped relative to vp3 */
2379
    if (s->theora < 0x030200)
2380
    {
2381
        s->flipped_image = 1;
2382
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2383
    }
2384

    
2385
    s->width = get_bits(gb, 16) << 4;
2386
    s->height = get_bits(gb, 16) << 4;
2387

    
2388
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2389
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2390
        s->width= s->height= 0;
2391
        return -1;
2392
    }
2393

    
2394
    if (s->theora >= 0x030400)
2395
    {
2396
        skip_bits(gb, 32); /* total number of superblocks in a frame */
2397
        // fixme, the next field is 36bits long
2398
        skip_bits(gb, 32); /* total number of blocks in a frame */
2399
        skip_bits(gb, 4); /* total number of blocks in a frame */
2400
        skip_bits(gb, 32); /* total number of macroblocks in a frame */
2401
    }
2402

    
2403
    visible_width  = get_bits_long(gb, 24);
2404
    visible_height = get_bits_long(gb, 24);
2405

    
2406
    if (s->theora >= 0x030200) {
2407
        skip_bits(gb, 8); /* offset x */
2408
        skip_bits(gb, 8); /* offset y */
2409
    }
2410

    
2411
    skip_bits(gb, 32); /* fps numerator */
2412
    skip_bits(gb, 32); /* fps denumerator */
2413
    skip_bits(gb, 24); /* aspect numerator */
2414
    skip_bits(gb, 24); /* aspect denumerator */
2415

    
2416
    if (s->theora < 0x030200)
2417
        skip_bits(gb, 5); /* keyframe frequency force */
2418
    skip_bits(gb, 8); /* colorspace */
2419
    if (s->theora >= 0x030400)
2420
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2421
    skip_bits(gb, 24); /* bitrate */
2422

    
2423
    skip_bits(gb, 6); /* quality hint */
2424

    
2425
    if (s->theora >= 0x030200)
2426
    {
2427
        skip_bits(gb, 5); /* keyframe frequency force */
2428

    
2429
        if (s->theora < 0x030400)
2430
            skip_bits(gb, 5); /* spare bits */
2431
    }
2432

    
2433
//    align_get_bits(gb);
2434

    
2435
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2436
        && visible_height <= s->height && visible_height > s->height-16)
2437
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2438
    else
2439
        avcodec_set_dimensions(avctx, s->width, s->height);
2440

    
2441
    return 0;
2442
}
2443

    
2444
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2445
{
2446
    Vp3DecodeContext *s = avctx->priv_data;
2447
    int i, n, matrices, inter, plane;
2448

    
2449
    if (s->theora >= 0x030200) {
2450
        n = get_bits(gb, 3);
2451
        /* loop filter limit values table */
2452
        for (i = 0; i < 64; i++)
2453
            s->filter_limit_values[i] = get_bits(gb, n);
2454
    }
2455

    
2456
    if (s->theora >= 0x030200)
2457
        n = get_bits(gb, 4) + 1;
2458
    else
2459
        n = 16;
2460
    /* quality threshold table */
2461
    for (i = 0; i < 64; i++)
2462
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2463

    
2464
    if (s->theora >= 0x030200)
2465
        n = get_bits(gb, 4) + 1;
2466
    else
2467
        n = 16;
2468
    /* dc scale factor table */
2469
    for (i = 0; i < 64; i++)
2470
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2471

    
2472
    if (s->theora >= 0x030200)
2473
        matrices = get_bits(gb, 9) + 1;
2474
    else
2475
        matrices = 3;
2476

    
2477
    if(matrices > 384){
2478
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2479
        return -1;
2480
    }
2481

    
2482
    for(n=0; n<matrices; n++){
2483
        for (i = 0; i < 64; i++)
2484
            s->base_matrix[n][i]= get_bits(gb, 8);
2485
    }
2486

    
2487
    for (inter = 0; inter <= 1; inter++) {
2488
        for (plane = 0; plane <= 2; plane++) {
2489
            int newqr= 1;
2490
            if (inter || plane > 0)
2491
                newqr = get_bits1(gb);
2492
            if (!newqr) {
2493
                int qtj, plj;
2494
                if(inter && get_bits1(gb)){
2495
                    qtj = 0;
2496
                    plj = plane;
2497
                }else{
2498
                    qtj= (3*inter + plane - 1) / 3;
2499
                    plj= (plane + 2) % 3;
2500
                }
2501
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2502
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2503
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2504
            } else {
2505
                int qri= 0;
2506
                int qi = 0;
2507

    
2508
                for(;;){
2509
                    i= get_bits(gb, av_log2(matrices-1)+1);
2510
                    if(i>= matrices){
2511
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2512
                        return -1;
2513
                    }
2514
                    s->qr_base[inter][plane][qri]= i;
2515
                    if(qi >= 63)
2516
                        break;
2517
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2518
                    s->qr_size[inter][plane][qri++]= i;
2519
                    qi += i;
2520
                }
2521

    
2522
                if (qi > 63) {
2523
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2524
                    return -1;
2525
                }
2526
                s->qr_count[inter][plane]= qri;
2527
            }
2528
        }
2529
    }
2530

    
2531
    /* Huffman tables */
2532
    for (s->hti = 0; s->hti < 80; s->hti++) {
2533
        s->entries = 0;
2534
        s->huff_code_size = 1;
2535
        if (!get_bits1(gb)) {
2536
            s->hbits = 0;
2537
            read_huffman_tree(avctx, gb);
2538
            s->hbits = 1;
2539
            read_huffman_tree(avctx, gb);
2540
        }
2541
    }
2542

    
2543
    s->theora_tables = 1;
2544

    
2545
    return 0;
2546
}
2547

    
2548
static int theora_decode_init(AVCodecContext *avctx)
2549
{
2550
    Vp3DecodeContext *s = avctx->priv_data;
2551
    GetBitContext gb;
2552
    int ptype;
2553
    uint8_t *header_start[3];
2554
    int header_len[3];
2555
    int i;
2556

    
2557
    s->theora = 1;
2558

    
2559
    if (!avctx->extradata_size)
2560
    {
2561
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2562
        return -1;
2563
    }
2564

    
2565
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2566
                              42, header_start, header_len) < 0) {
2567
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2568
        return -1;
2569
    }
2570

    
2571
  for(i=0;i<3;i++) {
2572
    init_get_bits(&gb, header_start[i], header_len[i]);
2573

    
2574
    ptype = get_bits(&gb, 8);
2575
    debug_vp3("Theora headerpacket type: %x\n", ptype);
2576

    
2577
     if (!(ptype & 0x80))
2578
     {
2579
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2580
//        return -1;
2581
     }
2582

    
2583
    // FIXME: Check for this as well.
2584
    skip_bits(&gb, 6*8); /* "theora" */
2585

    
2586
    switch(ptype)
2587
    {
2588
        case 0x80:
2589
            theora_decode_header(avctx, &gb);
2590
                break;
2591
        case 0x81:
2592
// FIXME: is this needed? it breaks sometimes
2593
//            theora_decode_comments(avctx, gb);
2594
            break;
2595
        case 0x82:
2596
            theora_decode_tables(avctx, &gb);
2597
            break;
2598
        default:
2599
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2600
            break;
2601
    }
2602
    if(8*header_len[i] != get_bits_count(&gb))
2603
        av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2604
    if (s->theora < 0x030200)
2605
        break;
2606
  }
2607

    
2608
    vp3_decode_init(avctx);
2609
    return 0;
2610
}
2611

    
2612
AVCodec theora_decoder = {
2613
    "theora",
2614
    CODEC_TYPE_VIDEO,
2615
    CODEC_ID_THEORA,
2616
    sizeof(Vp3DecodeContext),
2617
    theora_decode_init,
2618
    NULL,
2619
    vp3_decode_end,
2620
    vp3_decode_frame,
2621
    0,
2622
    NULL
2623
};
2624
#endif
2625

    
2626
AVCodec vp3_decoder = {
2627
    "vp3",
2628
    CODEC_TYPE_VIDEO,
2629
    CODEC_ID_VP3,
2630
    sizeof(Vp3DecodeContext),
2631
    vp3_decode_init,
2632
    NULL,
2633
    vp3_decode_end,
2634
    vp3_decode_frame,
2635
    0,
2636
    NULL
2637
};