Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 2029f312

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
/**
23
 * @file vp3.c
24
 * On2 VP3 Video Decoder
25
 *
26
 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
27
 * For more information about the VP3 coding process, visit:
28
 *   http://multimedia.cx/
29
 *
30
 * Theora decoder by Alex Beregszaszi
31
 */
32

    
33
#include <stdio.h>
34
#include <stdlib.h>
35
#include <string.h>
36
#include <unistd.h>
37

    
38
#include "avcodec.h"
39
#include "dsputil.h"
40
#include "mpegvideo.h"
41

    
42
#include "vp3data.h"
43
#include "xiph.h"
44

    
45
#define FRAGMENT_PIXELS 8
46

    
47
/*
48
 * Debugging Variables
49
 *
50
 * Define one or more of the following compile-time variables to 1 to obtain
51
 * elaborate information about certain aspects of the decoding process.
52
 *
53
 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
54
 * DEBUG_VP3: high-level decoding flow
55
 * DEBUG_INIT: initialization parameters
56
 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
57
 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
58
 * DEBUG_MODES: unpacking the coding modes for individual fragments
59
 * DEBUG_VECTORS: display the motion vectors
60
 * DEBUG_TOKEN: display exhaustive information about each DCT token
61
 * DEBUG_VLC: display the VLCs as they are extracted from the stream
62
 * DEBUG_DC_PRED: display the process of reversing DC prediction
63
 * DEBUG_IDCT: show every detail of the IDCT process
64
 */
65

    
66
#define KEYFRAMES_ONLY 0
67

    
68
#define DEBUG_VP3 0
69
#define DEBUG_INIT 0
70
#define DEBUG_DEQUANTIZERS 0
71
#define DEBUG_BLOCK_CODING 0
72
#define DEBUG_MODES 0
73
#define DEBUG_VECTORS 0
74
#define DEBUG_TOKEN 0
75
#define DEBUG_VLC 0
76
#define DEBUG_DC_PRED 0
77
#define DEBUG_IDCT 0
78

    
79
#if DEBUG_VP3
80
#define debug_vp3(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
81
#else
82
static inline void debug_vp3(const char *format, ...) { }
83
#endif
84

    
85
#if DEBUG_INIT
86
#define debug_init(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
87
#else
88
static inline void debug_init(const char *format, ...) { }
89
#endif
90

    
91
#if DEBUG_DEQUANTIZERS
92
#define debug_dequantizers(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
93
#else
94
static inline void debug_dequantizers(const char *format, ...) { }
95
#endif
96

    
97
#if DEBUG_BLOCK_CODING
98
#define debug_block_coding(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
99
#else
100
static inline void debug_block_coding(const char *format, ...) { }
101
#endif
102

    
103
#if DEBUG_MODES
104
#define debug_modes(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
105
#else
106
static inline void debug_modes(const char *format, ...) { }
107
#endif
108

    
109
#if DEBUG_VECTORS
110
#define debug_vectors(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
111
#else
112
static inline void debug_vectors(const char *format, ...) { }
113
#endif
114

    
115
#if DEBUG_TOKEN
116
#define debug_token(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
117
#else
118
static inline void debug_token(const char *format, ...) { }
119
#endif
120

    
121
#if DEBUG_VLC
122
#define debug_vlc(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
123
#else
124
static inline void debug_vlc(const char *format, ...) { }
125
#endif
126

    
127
#if DEBUG_DC_PRED
128
#define debug_dc_pred(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
129
#else
130
static inline void debug_dc_pred(const char *format, ...) { }
131
#endif
132

    
133
#if DEBUG_IDCT
134
#define debug_idct(args...) av_log(NULL, AV_LOG_DEBUG, ## args)
135
#else
136
static inline void debug_idct(const char *format, ...) { }
137
#endif
138

    
139
typedef struct Coeff {
140
    struct Coeff *next;
141
    DCTELEM coeff;
142
    uint8_t index;
143
} Coeff;
144

    
145
//FIXME split things out into their own arrays
146
typedef struct Vp3Fragment {
147
    Coeff *next_coeff;
148
    /* address of first pixel taking into account which plane the fragment
149
     * lives on as well as the plane stride */
150
    int first_pixel;
151
    /* this is the macroblock that the fragment belongs to */
152
    uint16_t macroblock;
153
    uint8_t coding_method;
154
    uint8_t coeff_count;
155
    int8_t motion_x;
156
    int8_t motion_y;
157
} Vp3Fragment;
158

    
159
#define SB_NOT_CODED        0
160
#define SB_PARTIALLY_CODED  1
161
#define SB_FULLY_CODED      2
162

    
163
#define MODE_INTER_NO_MV      0
164
#define MODE_INTRA            1
165
#define MODE_INTER_PLUS_MV    2
166
#define MODE_INTER_LAST_MV    3
167
#define MODE_INTER_PRIOR_LAST 4
168
#define MODE_USING_GOLDEN     5
169
#define MODE_GOLDEN_MV        6
170
#define MODE_INTER_FOURMV     7
171
#define CODING_MODE_COUNT     8
172

    
173
/* special internal mode */
174
#define MODE_COPY             8
175

    
176
/* There are 6 preset schemes, plus a free-form scheme */
177
static int ModeAlphabet[7][CODING_MODE_COUNT] =
178
{
179
    /* this is the custom scheme */
180
    { 0, 0, 0, 0, 0, 0, 0, 0 },
181

    
182
    /* scheme 1: Last motion vector dominates */
183
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
184
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
185
         MODE_INTRA,            MODE_USING_GOLDEN,
186
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
187

    
188
    /* scheme 2 */
189
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
190
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
191
         MODE_INTRA,            MODE_USING_GOLDEN,
192
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
193

    
194
    /* scheme 3 */
195
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
196
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
197
         MODE_INTRA,            MODE_USING_GOLDEN,
198
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
199

    
200
    /* scheme 4 */
201
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
202
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
203
         MODE_INTRA,            MODE_USING_GOLDEN,
204
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
205

    
206
    /* scheme 5: No motion vector dominates */
207
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
208
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
209
         MODE_INTRA,            MODE_USING_GOLDEN,
210
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
211

    
212
    /* scheme 6 */
213
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
214
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
215
         MODE_INTER_PLUS_MV,    MODE_INTRA,
216
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
217

    
218
};
219

    
220
#define MIN_DEQUANT_VAL 2
221

    
222
typedef struct Vp3DecodeContext {
223
    AVCodecContext *avctx;
224
    int theora, theora_tables;
225
    int version;
226
    int width, height;
227
    AVFrame golden_frame;
228
    AVFrame last_frame;
229
    AVFrame current_frame;
230
    int keyframe;
231
    DSPContext dsp;
232
    int flipped_image;
233

    
234
    int qis[3];
235
    int nqis;
236
    int quality_index;
237
    int last_quality_index;
238

    
239
    int superblock_count;
240
    int superblock_width;
241
    int superblock_height;
242
    int y_superblock_width;
243
    int y_superblock_height;
244
    int c_superblock_width;
245
    int c_superblock_height;
246
    int u_superblock_start;
247
    int v_superblock_start;
248
    unsigned char *superblock_coding;
249

    
250
    int macroblock_count;
251
    int macroblock_width;
252
    int macroblock_height;
253

    
254
    int fragment_count;
255
    int fragment_width;
256
    int fragment_height;
257

    
258
    Vp3Fragment *all_fragments;
259
    Coeff *coeffs;
260
    Coeff *next_coeff;
261
    int fragment_start[3];
262

    
263
    ScanTable scantable;
264

    
265
    /* tables */
266
    uint16_t coded_dc_scale_factor[64];
267
    uint32_t coded_ac_scale_factor[64];
268
    uint8_t base_matrix[384][64];
269
    uint8_t qr_count[2][3];
270
    uint8_t qr_size [2][3][64];
271
    uint16_t qr_base[2][3][64];
272

    
273
    /* this is a list of indices into the all_fragments array indicating
274
     * which of the fragments are coded */
275
    int *coded_fragment_list;
276
    int coded_fragment_list_index;
277
    int pixel_addresses_inited;
278

    
279
    VLC dc_vlc[16];
280
    VLC ac_vlc_1[16];
281
    VLC ac_vlc_2[16];
282
    VLC ac_vlc_3[16];
283
    VLC ac_vlc_4[16];
284

    
285
    VLC superblock_run_length_vlc;
286
    VLC fragment_run_length_vlc;
287
    VLC mode_code_vlc;
288
    VLC motion_vector_vlc;
289

    
290
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
291
     * index into them */
292
    DECLARE_ALIGNED_16(int16_t, qmat[2][4][64]);        //<qmat[is_inter][plane]
293

    
294
    /* This table contains superblock_count * 16 entries. Each set of 16
295
     * numbers corresponds to the fragment indices 0..15 of the superblock.
296
     * An entry will be -1 to indicate that no entry corresponds to that
297
     * index. */
298
    int *superblock_fragments;
299

    
300
    /* This table contains superblock_count * 4 entries. Each set of 4
301
     * numbers corresponds to the macroblock indices 0..3 of the superblock.
302
     * An entry will be -1 to indicate that no entry corresponds to that
303
     * index. */
304
    int *superblock_macroblocks;
305

    
306
    /* This table contains macroblock_count * 6 entries. Each set of 6
307
     * numbers corresponds to the fragment indices 0..5 which comprise
308
     * the macroblock (4 Y fragments and 2 C fragments). */
309
    int *macroblock_fragments;
310
    /* This is an array that indicates how a particular macroblock
311
     * is coded. */
312
    unsigned char *macroblock_coding;
313

    
314
    int first_coded_y_fragment;
315
    int first_coded_c_fragment;
316
    int last_coded_y_fragment;
317
    int last_coded_c_fragment;
318

    
319
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
320
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
321

    
322
    /* Huffman decode */
323
    int hti;
324
    unsigned int hbits;
325
    int entries;
326
    int huff_code_size;
327
    uint16_t huffman_table[80][32][2];
328

    
329
    uint32_t filter_limit_values[64];
330
    int bounding_values_array[256];
331
} Vp3DecodeContext;
332

    
333
/************************************************************************
334
 * VP3 specific functions
335
 ************************************************************************/
336

    
337
/*
338
 * This function sets up all of the various blocks mappings:
339
 * superblocks <-> fragments, macroblocks <-> fragments,
340
 * superblocks <-> macroblocks
341
 *
342
 * Returns 0 is successful; returns 1 if *anything* went wrong.
343
 */
344
static int init_block_mapping(Vp3DecodeContext *s)
345
{
346
    int i, j;
347
    signed int hilbert_walk_mb[4];
348

    
349
    int current_fragment = 0;
350
    int current_width = 0;
351
    int current_height = 0;
352
    int right_edge = 0;
353
    int bottom_edge = 0;
354
    int superblock_row_inc = 0;
355
    int *hilbert = NULL;
356
    int mapping_index = 0;
357

    
358
    int current_macroblock;
359
    int c_fragment;
360

    
361
    signed char travel_width[16] = {
362
         1,  1,  0, -1,
363
         0,  0,  1,  0,
364
         1,  0,  1,  0,
365
         0, -1,  0,  1
366
    };
367

    
368
    signed char travel_height[16] = {
369
         0,  0,  1,  0,
370
         1,  1,  0, -1,
371
         0,  1,  0, -1,
372
        -1,  0, -1,  0
373
    };
374

    
375
    signed char travel_width_mb[4] = {
376
         1,  0,  1,  0
377
    };
378

    
379
    signed char travel_height_mb[4] = {
380
         0,  1,  0, -1
381
    };
382

    
383
    debug_vp3("  vp3: initialize block mapping tables\n");
384

    
385
    hilbert_walk_mb[0] = 1;
386
    hilbert_walk_mb[1] = s->macroblock_width;
387
    hilbert_walk_mb[2] = 1;
388
    hilbert_walk_mb[3] = -s->macroblock_width;
389

    
390
    /* iterate through each superblock (all planes) and map the fragments */
391
    for (i = 0; i < s->superblock_count; i++) {
392
        debug_init("    superblock %d (u starts @ %d, v starts @ %d)\n",
393
            i, s->u_superblock_start, s->v_superblock_start);
394

    
395
        /* time to re-assign the limits? */
396
        if (i == 0) {
397

    
398
            /* start of Y superblocks */
399
            right_edge = s->fragment_width;
400
            bottom_edge = s->fragment_height;
401
            current_width = -1;
402
            current_height = 0;
403
            superblock_row_inc = 3 * s->fragment_width -
404
                (s->y_superblock_width * 4 - s->fragment_width);
405

    
406
            /* the first operation for this variable is to advance by 1 */
407
            current_fragment = -1;
408

    
409
        } else if (i == s->u_superblock_start) {
410

    
411
            /* start of U superblocks */
412
            right_edge = s->fragment_width / 2;
413
            bottom_edge = s->fragment_height / 2;
414
            current_width = -1;
415
            current_height = 0;
416
            superblock_row_inc = 3 * (s->fragment_width / 2) -
417
                (s->c_superblock_width * 4 - s->fragment_width / 2);
418

    
419
            /* the first operation for this variable is to advance by 1 */
420
            current_fragment = s->fragment_start[1] - 1;
421

    
422
        } else if (i == s->v_superblock_start) {
423

    
424
            /* start of V superblocks */
425
            right_edge = s->fragment_width / 2;
426
            bottom_edge = s->fragment_height / 2;
427
            current_width = -1;
428
            current_height = 0;
429
            superblock_row_inc = 3 * (s->fragment_width / 2) -
430
                (s->c_superblock_width * 4 - s->fragment_width / 2);
431

    
432
            /* the first operation for this variable is to advance by 1 */
433
            current_fragment = s->fragment_start[2] - 1;
434

    
435
        }
436

    
437
        if (current_width >= right_edge - 1) {
438
            /* reset width and move to next superblock row */
439
            current_width = -1;
440
            current_height += 4;
441

    
442
            /* fragment is now at the start of a new superblock row */
443
            current_fragment += superblock_row_inc;
444
        }
445

    
446
        /* iterate through all 16 fragments in a superblock */
447
        for (j = 0; j < 16; j++) {
448
            current_fragment += travel_width[j] + right_edge * travel_height[j];
449
            current_width += travel_width[j];
450
            current_height += travel_height[j];
451

    
452
            /* check if the fragment is in bounds */
453
            if ((current_width < right_edge) &&
454
                (current_height < bottom_edge)) {
455
                s->superblock_fragments[mapping_index] = current_fragment;
456
                debug_init("    mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n",
457
                    s->superblock_fragments[mapping_index], i, j,
458
                    current_width, right_edge, current_height, bottom_edge);
459
            } else {
460
                s->superblock_fragments[mapping_index] = -1;
461
                debug_init("    superblock %d, position %d has no fragment (%d/%d x %d/%d)\n",
462
                    i, j,
463
                    current_width, right_edge, current_height, bottom_edge);
464
            }
465

    
466
            mapping_index++;
467
        }
468
    }
469

    
470
    /* initialize the superblock <-> macroblock mapping; iterate through
471
     * all of the Y plane superblocks to build this mapping */
472
    right_edge = s->macroblock_width;
473
    bottom_edge = s->macroblock_height;
474
    current_width = -1;
475
    current_height = 0;
476
    superblock_row_inc = s->macroblock_width -
477
        (s->y_superblock_width * 2 - s->macroblock_width);;
478
    hilbert = hilbert_walk_mb;
479
    mapping_index = 0;
480
    current_macroblock = -1;
481
    for (i = 0; i < s->u_superblock_start; i++) {
482

    
483
        if (current_width >= right_edge - 1) {
484
            /* reset width and move to next superblock row */
485
            current_width = -1;
486
            current_height += 2;
487

    
488
            /* macroblock is now at the start of a new superblock row */
489
            current_macroblock += superblock_row_inc;
490
        }
491

    
492
        /* iterate through each potential macroblock in the superblock */
493
        for (j = 0; j < 4; j++) {
494
            current_macroblock += hilbert_walk_mb[j];
495
            current_width += travel_width_mb[j];
496
            current_height += travel_height_mb[j];
497

    
498
            /* check if the macroblock is in bounds */
499
            if ((current_width < right_edge) &&
500
                (current_height < bottom_edge)) {
501
                s->superblock_macroblocks[mapping_index] = current_macroblock;
502
                debug_init("    mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
503
                    s->superblock_macroblocks[mapping_index], i, j,
504
                    current_width, right_edge, current_height, bottom_edge);
505
            } else {
506
                s->superblock_macroblocks[mapping_index] = -1;
507
                debug_init("    superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
508
                    i, j,
509
                    current_width, right_edge, current_height, bottom_edge);
510
            }
511

    
512
            mapping_index++;
513
        }
514
    }
515

    
516
    /* initialize the macroblock <-> fragment mapping */
517
    current_fragment = 0;
518
    current_macroblock = 0;
519
    mapping_index = 0;
520
    for (i = 0; i < s->fragment_height; i += 2) {
521

    
522
        for (j = 0; j < s->fragment_width; j += 2) {
523

    
524
            debug_init("    macroblock %d contains fragments: ", current_macroblock);
525
            s->all_fragments[current_fragment].macroblock = current_macroblock;
526
            s->macroblock_fragments[mapping_index++] = current_fragment;
527
            debug_init("%d ", current_fragment);
528

    
529
            if (j + 1 < s->fragment_width) {
530
                s->all_fragments[current_fragment + 1].macroblock = current_macroblock;
531
                s->macroblock_fragments[mapping_index++] = current_fragment + 1;
532
                debug_init("%d ", current_fragment + 1);
533
            } else
534
                s->macroblock_fragments[mapping_index++] = -1;
535

    
536
            if (i + 1 < s->fragment_height) {
537
                s->all_fragments[current_fragment + s->fragment_width].macroblock =
538
                    current_macroblock;
539
                s->macroblock_fragments[mapping_index++] =
540
                    current_fragment + s->fragment_width;
541
                debug_init("%d ", current_fragment + s->fragment_width);
542
            } else
543
                s->macroblock_fragments[mapping_index++] = -1;
544

    
545
            if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
546
                s->all_fragments[current_fragment + s->fragment_width + 1].macroblock =
547
                    current_macroblock;
548
                s->macroblock_fragments[mapping_index++] =
549
                    current_fragment + s->fragment_width + 1;
550
                debug_init("%d ", current_fragment + s->fragment_width + 1);
551
            } else
552
                s->macroblock_fragments[mapping_index++] = -1;
553

    
554
            /* C planes */
555
            c_fragment = s->fragment_start[1] +
556
                (i * s->fragment_width / 4) + (j / 2);
557
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
558
            s->macroblock_fragments[mapping_index++] = c_fragment;
559
            debug_init("%d ", c_fragment);
560

    
561
            c_fragment = s->fragment_start[2] +
562
                (i * s->fragment_width / 4) + (j / 2);
563
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
564
            s->macroblock_fragments[mapping_index++] = c_fragment;
565
            debug_init("%d ", c_fragment);
566

    
567
            debug_init("\n");
568

    
569
            if (j + 2 <= s->fragment_width)
570
                current_fragment += 2;
571
            else
572
                current_fragment++;
573
            current_macroblock++;
574
        }
575

    
576
        current_fragment += s->fragment_width;
577
    }
578

    
579
    return 0;  /* successful path out */
580
}
581

    
582
/*
583
 * This function wipes out all of the fragment data.
584
 */
585
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
586
{
587
    int i;
588

    
589
    /* zero out all of the fragment information */
590
    s->coded_fragment_list_index = 0;
591
    for (i = 0; i < s->fragment_count; i++) {
592
        s->all_fragments[i].coeff_count = 0;
593
        s->all_fragments[i].motion_x = 127;
594
        s->all_fragments[i].motion_y = 127;
595
        s->all_fragments[i].next_coeff= NULL;
596
        s->coeffs[i].index=
597
        s->coeffs[i].coeff=0;
598
        s->coeffs[i].next= NULL;
599
    }
600
}
601

    
602
/*
603
 * This function sets up the dequantization tables used for a particular
604
 * frame.
605
 */
606
static void init_dequantizer(Vp3DecodeContext *s)
607
{
608
    int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
609
    int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
610
    int i, plane, inter, qri, bmi, bmj, qistart;
611

    
612
    debug_vp3("  vp3: initializing dequantization tables\n");
613

    
614
    for(inter=0; inter<2; inter++){
615
        for(plane=0; plane<3; plane++){
616
            int sum=0;
617
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
618
                sum+= s->qr_size[inter][plane][qri];
619
                if(s->quality_index <= sum)
620
                    break;
621
            }
622
            qistart= sum - s->qr_size[inter][plane][qri];
623
            bmi= s->qr_base[inter][plane][qri  ];
624
            bmj= s->qr_base[inter][plane][qri+1];
625
            for(i=0; i<64; i++){
626
                int coeff= (  2*(sum    -s->quality_index)*s->base_matrix[bmi][i]
627
                            - 2*(qistart-s->quality_index)*s->base_matrix[bmj][i]
628
                            + s->qr_size[inter][plane][qri])
629
                           / (2*s->qr_size[inter][plane][qri]);
630

    
631
                int qmin= 8<<(inter + !i);
632
                int qscale= i ? ac_scale_factor : dc_scale_factor;
633

    
634
                s->qmat[inter][plane][i]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
635
            }
636
        }
637
    }
638

    
639
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][1], s->qmat[0][1][1])+8)/16, 512); //FIXME finetune
640
}
641

    
642
/*
643
 * This function initializes the loop filter boundary limits if the frame's
644
 * quality index is different from the previous frame's.
645
 */
646
static void init_loop_filter(Vp3DecodeContext *s)
647
{
648
    int *bounding_values= s->bounding_values_array+127;
649
    int filter_limit;
650
    int x;
651

    
652
    filter_limit = s->filter_limit_values[s->quality_index];
653

    
654
    /* set up the bounding values */
655
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
656
    for (x = 0; x < filter_limit; x++) {
657
        bounding_values[-x - filter_limit] = -filter_limit + x;
658
        bounding_values[-x] = -x;
659
        bounding_values[x] = x;
660
        bounding_values[x + filter_limit] = filter_limit - x;
661
    }
662
}
663

    
664
/*
665
 * This function unpacks all of the superblock/macroblock/fragment coding
666
 * information from the bitstream.
667
 */
668
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
669
{
670
    int bit = 0;
671
    int current_superblock = 0;
672
    int current_run = 0;
673
    int decode_fully_flags = 0;
674
    int decode_partial_blocks = 0;
675
    int first_c_fragment_seen;
676

    
677
    int i, j;
678
    int current_fragment;
679

    
680
    debug_vp3("  vp3: unpacking superblock coding\n");
681

    
682
    if (s->keyframe) {
683

    
684
        debug_vp3("    keyframe-- all superblocks are fully coded\n");
685
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
686

    
687
    } else {
688

    
689
        /* unpack the list of partially-coded superblocks */
690
        bit = get_bits(gb, 1);
691
        /* toggle the bit because as soon as the first run length is
692
         * fetched the bit will be toggled again */
693
        bit ^= 1;
694
        while (current_superblock < s->superblock_count) {
695
            if (current_run-- == 0) {
696
                bit ^= 1;
697
                current_run = get_vlc2(gb,
698
                    s->superblock_run_length_vlc.table, 6, 2);
699
                if (current_run == 33)
700
                    current_run += get_bits(gb, 12);
701
                debug_block_coding("      setting superblocks %d..%d to %s\n",
702
                    current_superblock,
703
                    current_superblock + current_run - 1,
704
                    (bit) ? "partially coded" : "not coded");
705

    
706
                /* if any of the superblocks are not partially coded, flag
707
                 * a boolean to decode the list of fully-coded superblocks */
708
                if (bit == 0) {
709
                    decode_fully_flags = 1;
710
                } else {
711

    
712
                    /* make a note of the fact that there are partially coded
713
                     * superblocks */
714
                    decode_partial_blocks = 1;
715
                }
716
            }
717
            s->superblock_coding[current_superblock++] = bit;
718
        }
719

    
720
        /* unpack the list of fully coded superblocks if any of the blocks were
721
         * not marked as partially coded in the previous step */
722
        if (decode_fully_flags) {
723

    
724
            current_superblock = 0;
725
            current_run = 0;
726
            bit = get_bits(gb, 1);
727
            /* toggle the bit because as soon as the first run length is
728
             * fetched the bit will be toggled again */
729
            bit ^= 1;
730
            while (current_superblock < s->superblock_count) {
731

    
732
                /* skip any superblocks already marked as partially coded */
733
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
734

    
735
                    if (current_run-- == 0) {
736
                        bit ^= 1;
737
                        current_run = get_vlc2(gb,
738
                            s->superblock_run_length_vlc.table, 6, 2);
739
                        if (current_run == 33)
740
                            current_run += get_bits(gb, 12);
741
                    }
742

    
743
                    debug_block_coding("      setting superblock %d to %s\n",
744
                        current_superblock,
745
                        (bit) ? "fully coded" : "not coded");
746
                    s->superblock_coding[current_superblock] = 2*bit;
747
                }
748
                current_superblock++;
749
            }
750
        }
751

    
752
        /* if there were partial blocks, initialize bitstream for
753
         * unpacking fragment codings */
754
        if (decode_partial_blocks) {
755

    
756
            current_run = 0;
757
            bit = get_bits(gb, 1);
758
            /* toggle the bit because as soon as the first run length is
759
             * fetched the bit will be toggled again */
760
            bit ^= 1;
761
        }
762
    }
763

    
764
    /* figure out which fragments are coded; iterate through each
765
     * superblock (all planes) */
766
    s->coded_fragment_list_index = 0;
767
    s->next_coeff= s->coeffs + s->fragment_count;
768
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
769
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
770
    first_c_fragment_seen = 0;
771
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
772
    for (i = 0; i < s->superblock_count; i++) {
773

    
774
        /* iterate through all 16 fragments in a superblock */
775
        for (j = 0; j < 16; j++) {
776

    
777
            /* if the fragment is in bounds, check its coding status */
778
            current_fragment = s->superblock_fragments[i * 16 + j];
779
            if (current_fragment >= s->fragment_count) {
780
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
781
                    current_fragment, s->fragment_count);
782
                return 1;
783
            }
784
            if (current_fragment != -1) {
785
                if (s->superblock_coding[i] == SB_NOT_CODED) {
786

    
787
                    /* copy all the fragments from the prior frame */
788
                    s->all_fragments[current_fragment].coding_method =
789
                        MODE_COPY;
790

    
791
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
792

    
793
                    /* fragment may or may not be coded; this is the case
794
                     * that cares about the fragment coding runs */
795
                    if (current_run-- == 0) {
796
                        bit ^= 1;
797
                        current_run = get_vlc2(gb,
798
                            s->fragment_run_length_vlc.table, 5, 2);
799
                    }
800

    
801
                    if (bit) {
802
                        /* default mode; actual mode will be decoded in
803
                         * the next phase */
804
                        s->all_fragments[current_fragment].coding_method =
805
                            MODE_INTER_NO_MV;
806
                        s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
807
                        s->coded_fragment_list[s->coded_fragment_list_index] =
808
                            current_fragment;
809
                        if ((current_fragment >= s->fragment_start[1]) &&
810
                            (s->last_coded_y_fragment == -1) &&
811
                            (!first_c_fragment_seen)) {
812
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
813
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
814
                            first_c_fragment_seen = 1;
815
                        }
816
                        s->coded_fragment_list_index++;
817
                        s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
818
                        debug_block_coding("      superblock %d is partially coded, fragment %d is coded\n",
819
                            i, current_fragment);
820
                    } else {
821
                        /* not coded; copy this fragment from the prior frame */
822
                        s->all_fragments[current_fragment].coding_method =
823
                            MODE_COPY;
824
                        debug_block_coding("      superblock %d is partially coded, fragment %d is not coded\n",
825
                            i, current_fragment);
826
                    }
827

    
828
                } else {
829

    
830
                    /* fragments are fully coded in this superblock; actual
831
                     * coding will be determined in next step */
832
                    s->all_fragments[current_fragment].coding_method =
833
                        MODE_INTER_NO_MV;
834
                    s->all_fragments[current_fragment].next_coeff= s->coeffs + current_fragment;
835
                    s->coded_fragment_list[s->coded_fragment_list_index] =
836
                        current_fragment;
837
                    if ((current_fragment >= s->fragment_start[1]) &&
838
                        (s->last_coded_y_fragment == -1) &&
839
                        (!first_c_fragment_seen)) {
840
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
841
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
842
                        first_c_fragment_seen = 1;
843
                    }
844
                    s->coded_fragment_list_index++;
845
                    s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
846
                    debug_block_coding("      superblock %d is fully coded, fragment %d is coded\n",
847
                        i, current_fragment);
848
                }
849
            }
850
        }
851
    }
852

    
853
    if (!first_c_fragment_seen)
854
        /* only Y fragments coded in this frame */
855
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
856
    else
857
        /* end the list of coded C fragments */
858
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
859

    
860
    debug_block_coding("    %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
861
        s->coded_fragment_list_index,
862
        s->first_coded_y_fragment,
863
        s->last_coded_y_fragment,
864
        s->first_coded_c_fragment,
865
        s->last_coded_c_fragment);
866

    
867
    return 0;
868
}
869

    
870
/*
871
 * This function unpacks all the coding mode data for individual macroblocks
872
 * from the bitstream.
873
 */
874
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
875
{
876
    int i, j, k;
877
    int scheme;
878
    int current_macroblock;
879
    int current_fragment;
880
    int coding_mode;
881

    
882
    debug_vp3("  vp3: unpacking encoding modes\n");
883

    
884
    if (s->keyframe) {
885
        debug_vp3("    keyframe-- all blocks are coded as INTRA\n");
886

    
887
        for (i = 0; i < s->fragment_count; i++)
888
            s->all_fragments[i].coding_method = MODE_INTRA;
889

    
890
    } else {
891

    
892
        /* fetch the mode coding scheme for this frame */
893
        scheme = get_bits(gb, 3);
894
        debug_modes("    using mode alphabet %d\n", scheme);
895

    
896
        /* is it a custom coding scheme? */
897
        if (scheme == 0) {
898
            debug_modes("    custom mode alphabet ahead:\n");
899
            for (i = 0; i < 8; i++)
900
                ModeAlphabet[scheme][get_bits(gb, 3)] = i;
901
        }
902

    
903
        for (i = 0; i < 8; i++)
904
            debug_modes("      mode[%d][%d] = %d\n", scheme, i,
905
                ModeAlphabet[scheme][i]);
906

    
907
        /* iterate through all of the macroblocks that contain 1 or more
908
         * coded fragments */
909
        for (i = 0; i < s->u_superblock_start; i++) {
910

    
911
            for (j = 0; j < 4; j++) {
912
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
913
                if ((current_macroblock == -1) ||
914
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
915
                    continue;
916
                if (current_macroblock >= s->macroblock_count) {
917
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
918
                        current_macroblock, s->macroblock_count);
919
                    return 1;
920
                }
921

    
922
                /* mode 7 means get 3 bits for each coding mode */
923
                if (scheme == 7)
924
                    coding_mode = get_bits(gb, 3);
925
                else
926
                    coding_mode = ModeAlphabet[scheme]
927
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
928

    
929
                s->macroblock_coding[current_macroblock] = coding_mode;
930
                for (k = 0; k < 6; k++) {
931
                    current_fragment =
932
                        s->macroblock_fragments[current_macroblock * 6 + k];
933
                    if (current_fragment == -1)
934
                        continue;
935
                    if (current_fragment >= s->fragment_count) {
936
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
937
                            current_fragment, s->fragment_count);
938
                        return 1;
939
                    }
940
                    if (s->all_fragments[current_fragment].coding_method !=
941
                        MODE_COPY)
942
                        s->all_fragments[current_fragment].coding_method =
943
                            coding_mode;
944
                }
945

    
946
                debug_modes("    coding method for macroblock starting @ fragment %d = %d\n",
947
                    s->macroblock_fragments[current_macroblock * 6], coding_mode);
948
            }
949
        }
950
    }
951

    
952
    return 0;
953
}
954

    
955
/*
956
 * This function unpacks all the motion vectors for the individual
957
 * macroblocks from the bitstream.
958
 */
959
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
960
{
961
    int i, j, k;
962
    int coding_mode;
963
    int motion_x[6];
964
    int motion_y[6];
965
    int last_motion_x = 0;
966
    int last_motion_y = 0;
967
    int prior_last_motion_x = 0;
968
    int prior_last_motion_y = 0;
969
    int current_macroblock;
970
    int current_fragment;
971

    
972
    debug_vp3("  vp3: unpacking motion vectors\n");
973
    if (s->keyframe) {
974

    
975
        debug_vp3("    keyframe-- there are no motion vectors\n");
976

    
977
    } else {
978

    
979
        memset(motion_x, 0, 6 * sizeof(int));
980
        memset(motion_y, 0, 6 * sizeof(int));
981

    
982
        /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
983
        coding_mode = get_bits(gb, 1);
984
        debug_vectors("    using %s scheme for unpacking motion vectors\n",
985
            (coding_mode == 0) ? "VLC" : "fixed-length");
986

    
987
        /* iterate through all of the macroblocks that contain 1 or more
988
         * coded fragments */
989
        for (i = 0; i < s->u_superblock_start; i++) {
990

    
991
            for (j = 0; j < 4; j++) {
992
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
993
                if ((current_macroblock == -1) ||
994
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
995
                    continue;
996
                if (current_macroblock >= s->macroblock_count) {
997
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
998
                        current_macroblock, s->macroblock_count);
999
                    return 1;
1000
                }
1001

    
1002
                current_fragment = s->macroblock_fragments[current_macroblock * 6];
1003
                if (current_fragment >= s->fragment_count) {
1004
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1005
                        current_fragment, s->fragment_count);
1006
                    return 1;
1007
                }
1008
                switch (s->macroblock_coding[current_macroblock]) {
1009

    
1010
                case MODE_INTER_PLUS_MV:
1011
                case MODE_GOLDEN_MV:
1012
                    /* all 6 fragments use the same motion vector */
1013
                    if (coding_mode == 0) {
1014
                        motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1015
                        motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1016
                    } else {
1017
                        motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1018
                        motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
1019
                    }
1020

    
1021
                    for (k = 1; k < 6; k++) {
1022
                        motion_x[k] = motion_x[0];
1023
                        motion_y[k] = motion_y[0];
1024
                    }
1025

    
1026
                    /* vector maintenance, only on MODE_INTER_PLUS_MV */
1027
                    if (s->macroblock_coding[current_macroblock] ==
1028
                        MODE_INTER_PLUS_MV) {
1029
                        prior_last_motion_x = last_motion_x;
1030
                        prior_last_motion_y = last_motion_y;
1031
                        last_motion_x = motion_x[0];
1032
                        last_motion_y = motion_y[0];
1033
                    }
1034
                    break;
1035

    
1036
                case MODE_INTER_FOURMV:
1037
                    /* fetch 4 vectors from the bitstream, one for each
1038
                     * Y fragment, then average for the C fragment vectors */
1039
                    motion_x[4] = motion_y[4] = 0;
1040
                    for (k = 0; k < 4; k++) {
1041
                        if (coding_mode == 0) {
1042
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1043
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
1044
                        } else {
1045
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1046
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
1047
                        }
1048
                        motion_x[4] += motion_x[k];
1049
                        motion_y[4] += motion_y[k];
1050
                    }
1051

    
1052
                    motion_x[5]=
1053
                    motion_x[4]= RSHIFT(motion_x[4], 2);
1054
                    motion_y[5]=
1055
                    motion_y[4]= RSHIFT(motion_y[4], 2);
1056

    
1057
                    /* vector maintenance; vector[3] is treated as the
1058
                     * last vector in this case */
1059
                    prior_last_motion_x = last_motion_x;
1060
                    prior_last_motion_y = last_motion_y;
1061
                    last_motion_x = motion_x[3];
1062
                    last_motion_y = motion_y[3];
1063
                    break;
1064

    
1065
                case MODE_INTER_LAST_MV:
1066
                    /* all 6 fragments use the last motion vector */
1067
                    motion_x[0] = last_motion_x;
1068
                    motion_y[0] = last_motion_y;
1069
                    for (k = 1; k < 6; k++) {
1070
                        motion_x[k] = motion_x[0];
1071
                        motion_y[k] = motion_y[0];
1072
                    }
1073

    
1074
                    /* no vector maintenance (last vector remains the
1075
                     * last vector) */
1076
                    break;
1077

    
1078
                case MODE_INTER_PRIOR_LAST:
1079
                    /* all 6 fragments use the motion vector prior to the
1080
                     * last motion vector */
1081
                    motion_x[0] = prior_last_motion_x;
1082
                    motion_y[0] = prior_last_motion_y;
1083
                    for (k = 1; k < 6; k++) {
1084
                        motion_x[k] = motion_x[0];
1085
                        motion_y[k] = motion_y[0];
1086
                    }
1087

    
1088
                    /* vector maintenance */
1089
                    prior_last_motion_x = last_motion_x;
1090
                    prior_last_motion_y = last_motion_y;
1091
                    last_motion_x = motion_x[0];
1092
                    last_motion_y = motion_y[0];
1093
                    break;
1094

    
1095
                default:
1096
                    /* covers intra, inter without MV, golden without MV */
1097
                    memset(motion_x, 0, 6 * sizeof(int));
1098
                    memset(motion_y, 0, 6 * sizeof(int));
1099

    
1100
                    /* no vector maintenance */
1101
                    break;
1102
                }
1103

    
1104
                /* assign the motion vectors to the correct fragments */
1105
                debug_vectors("    vectors for macroblock starting @ fragment %d (coding method %d):\n",
1106
                    current_fragment,
1107
                    s->macroblock_coding[current_macroblock]);
1108
                for (k = 0; k < 6; k++) {
1109
                    current_fragment =
1110
                        s->macroblock_fragments[current_macroblock * 6 + k];
1111
                    if (current_fragment == -1)
1112
                        continue;
1113
                    if (current_fragment >= s->fragment_count) {
1114
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1115
                            current_fragment, s->fragment_count);
1116
                        return 1;
1117
                    }
1118
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
1119
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
1120
                    debug_vectors("    vector %d: fragment %d = (%d, %d)\n",
1121
                        k, current_fragment, motion_x[k], motion_y[k]);
1122
                }
1123
            }
1124
        }
1125
    }
1126

    
1127
    return 0;
1128
}
1129

    
1130
/*
1131
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1132
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1133
 * data. This function unpacks all the VLCs for either the Y plane or both
1134
 * C planes, and is called for DC coefficients or different AC coefficient
1135
 * levels (since different coefficient types require different VLC tables.
1136
 *
1137
 * This function returns a residual eob run. E.g, if a particular token gave
1138
 * instructions to EOB the next 5 fragments and there were only 2 fragments
1139
 * left in the current fragment range, 3 would be returned so that it could
1140
 * be passed into the next call to this same function.
1141
 */
1142
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1143
                        VLC *table, int coeff_index,
1144
                        int first_fragment, int last_fragment,
1145
                        int eob_run)
1146
{
1147
    int i;
1148
    int token;
1149
    int zero_run = 0;
1150
    DCTELEM coeff = 0;
1151
    Vp3Fragment *fragment;
1152
    uint8_t *perm= s->scantable.permutated;
1153
    int bits_to_get;
1154

    
1155
    if ((first_fragment >= s->fragment_count) ||
1156
        (last_fragment >= s->fragment_count)) {
1157

    
1158
        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1159
            first_fragment, last_fragment);
1160
        return 0;
1161
    }
1162

    
1163
    for (i = first_fragment; i <= last_fragment; i++) {
1164

    
1165
        fragment = &s->all_fragments[s->coded_fragment_list[i]];
1166
        if (fragment->coeff_count > coeff_index)
1167
            continue;
1168

    
1169
        if (!eob_run) {
1170
            /* decode a VLC into a token */
1171
            token = get_vlc2(gb, table->table, 5, 3);
1172
            debug_vlc(" token = %2d, ", token);
1173
            /* use the token to get a zero run, a coefficient, and an eob run */
1174
            if (token <= 6) {
1175
                eob_run = eob_run_base[token];
1176
                if (eob_run_get_bits[token])
1177
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
1178
                coeff = zero_run = 0;
1179
            } else {
1180
                bits_to_get = coeff_get_bits[token];
1181
                if (!bits_to_get)
1182
                    coeff = coeff_tables[token][0];
1183
                else
1184
                    coeff = coeff_tables[token][get_bits(gb, bits_to_get)];
1185

    
1186
                zero_run = zero_run_base[token];
1187
                if (zero_run_get_bits[token])
1188
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
1189
            }
1190
        }
1191

    
1192
        if (!eob_run) {
1193
            fragment->coeff_count += zero_run;
1194
            if (fragment->coeff_count < 64){
1195
                fragment->next_coeff->coeff= coeff;
1196
                fragment->next_coeff->index= perm[fragment->coeff_count++]; //FIXME perm here already?
1197
                fragment->next_coeff->next= s->next_coeff;
1198
                s->next_coeff->next=NULL;
1199
                fragment->next_coeff= s->next_coeff++;
1200
            }
1201
            debug_vlc(" fragment %d coeff = %d\n",
1202
                s->coded_fragment_list[i], fragment->next_coeff[coeff_index]);
1203
        } else {
1204
            fragment->coeff_count |= 128;
1205
            debug_vlc(" fragment %d eob with %d coefficients\n",
1206
                s->coded_fragment_list[i], fragment->coeff_count&127);
1207
            eob_run--;
1208
        }
1209
    }
1210

    
1211
    return eob_run;
1212
}
1213

    
1214
/*
1215
 * This function unpacks all of the DCT coefficient data from the
1216
 * bitstream.
1217
 */
1218
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1219
{
1220
    int i;
1221
    int dc_y_table;
1222
    int dc_c_table;
1223
    int ac_y_table;
1224
    int ac_c_table;
1225
    int residual_eob_run = 0;
1226

    
1227
    /* fetch the DC table indices */
1228
    dc_y_table = get_bits(gb, 4);
1229
    dc_c_table = get_bits(gb, 4);
1230

    
1231
    /* unpack the Y plane DC coefficients */
1232
    debug_vp3("  vp3: unpacking Y plane DC coefficients using table %d\n",
1233
        dc_y_table);
1234
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1235
        s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1236

    
1237
    /* unpack the C plane DC coefficients */
1238
    debug_vp3("  vp3: unpacking C plane DC coefficients using table %d\n",
1239
        dc_c_table);
1240
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1241
        s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1242

    
1243
    /* fetch the AC table indices */
1244
    ac_y_table = get_bits(gb, 4);
1245
    ac_c_table = get_bits(gb, 4);
1246

    
1247
    /* unpack the group 1 AC coefficients (coeffs 1-5) */
1248
    for (i = 1; i <= 5; i++) {
1249

    
1250
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1251
            i, ac_y_table);
1252
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
1253
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1254

    
1255
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1256
            i, ac_c_table);
1257
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
1258
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1259
    }
1260

    
1261
    /* unpack the group 2 AC coefficients (coeffs 6-14) */
1262
    for (i = 6; i <= 14; i++) {
1263

    
1264
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1265
            i, ac_y_table);
1266
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
1267
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1268

    
1269
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1270
            i, ac_c_table);
1271
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
1272
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1273
    }
1274

    
1275
    /* unpack the group 3 AC coefficients (coeffs 15-27) */
1276
    for (i = 15; i <= 27; i++) {
1277

    
1278
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1279
            i, ac_y_table);
1280
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
1281
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1282

    
1283
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1284
            i, ac_c_table);
1285
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
1286
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1287
    }
1288

    
1289
    /* unpack the group 4 AC coefficients (coeffs 28-63) */
1290
    for (i = 28; i <= 63; i++) {
1291

    
1292
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1293
            i, ac_y_table);
1294
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
1295
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1296

    
1297
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1298
            i, ac_c_table);
1299
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
1300
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1301
    }
1302

    
1303
    return 0;
1304
}
1305

    
1306
/*
1307
 * This function reverses the DC prediction for each coded fragment in
1308
 * the frame. Much of this function is adapted directly from the original
1309
 * VP3 source code.
1310
 */
1311
#define COMPATIBLE_FRAME(x) \
1312
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1313
#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1314
#define DC_COEFF(u) (s->coeffs[u].index ? 0 : s->coeffs[u].coeff) //FIXME do somethin to simplify this
1315

    
1316
static void reverse_dc_prediction(Vp3DecodeContext *s,
1317
                                  int first_fragment,
1318
                                  int fragment_width,
1319
                                  int fragment_height)
1320
{
1321

    
1322
#define PUL 8
1323
#define PU 4
1324
#define PUR 2
1325
#define PL 1
1326

    
1327
    int x, y;
1328
    int i = first_fragment;
1329

    
1330
    int predicted_dc;
1331

    
1332
    /* DC values for the left, up-left, up, and up-right fragments */
1333
    int vl, vul, vu, vur;
1334

    
1335
    /* indices for the left, up-left, up, and up-right fragments */
1336
    int l, ul, u, ur;
1337

    
1338
    /*
1339
     * The 6 fields mean:
1340
     *   0: up-left multiplier
1341
     *   1: up multiplier
1342
     *   2: up-right multiplier
1343
     *   3: left multiplier
1344
     */
1345
    int predictor_transform[16][4] = {
1346
        {  0,  0,  0,  0},
1347
        {  0,  0,  0,128},        // PL
1348
        {  0,  0,128,  0},        // PUR
1349
        {  0,  0, 53, 75},        // PUR|PL
1350
        {  0,128,  0,  0},        // PU
1351
        {  0, 64,  0, 64},        // PU|PL
1352
        {  0,128,  0,  0},        // PU|PUR
1353
        {  0,  0, 53, 75},        // PU|PUR|PL
1354
        {128,  0,  0,  0},        // PUL
1355
        {  0,  0,  0,128},        // PUL|PL
1356
        { 64,  0, 64,  0},        // PUL|PUR
1357
        {  0,  0, 53, 75},        // PUL|PUR|PL
1358
        {  0,128,  0,  0},        // PUL|PU
1359
       {-104,116,  0,116},        // PUL|PU|PL
1360
        { 24, 80, 24,  0},        // PUL|PU|PUR
1361
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1362
    };
1363

    
1364
    /* This table shows which types of blocks can use other blocks for
1365
     * prediction. For example, INTRA is the only mode in this table to
1366
     * have a frame number of 0. That means INTRA blocks can only predict
1367
     * from other INTRA blocks. There are 2 golden frame coding types;
1368
     * blocks encoding in these modes can only predict from other blocks
1369
     * that were encoded with these 1 of these 2 modes. */
1370
    unsigned char compatible_frame[8] = {
1371
        1,    /* MODE_INTER_NO_MV */
1372
        0,    /* MODE_INTRA */
1373
        1,    /* MODE_INTER_PLUS_MV */
1374
        1,    /* MODE_INTER_LAST_MV */
1375
        1,    /* MODE_INTER_PRIOR_MV */
1376
        2,    /* MODE_USING_GOLDEN */
1377
        2,    /* MODE_GOLDEN_MV */
1378
        1     /* MODE_INTER_FOUR_MV */
1379
    };
1380
    int current_frame_type;
1381

    
1382
    /* there is a last DC predictor for each of the 3 frame types */
1383
    short last_dc[3];
1384

    
1385
    int transform = 0;
1386

    
1387
    debug_vp3("  vp3: reversing DC prediction\n");
1388

    
1389
    vul = vu = vur = vl = 0;
1390
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1391

    
1392
    /* for each fragment row... */
1393
    for (y = 0; y < fragment_height; y++) {
1394

    
1395
        /* for each fragment in a row... */
1396
        for (x = 0; x < fragment_width; x++, i++) {
1397

    
1398
            /* reverse prediction if this block was coded */
1399
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1400

    
1401
                current_frame_type =
1402
                    compatible_frame[s->all_fragments[i].coding_method];
1403
                debug_dc_pred(" frag %d: orig DC = %d, ",
1404
                    i, DC_COEFF(i));
1405

    
1406
                transform= 0;
1407
                if(x){
1408
                    l= i-1;
1409
                    vl = DC_COEFF(l);
1410
                    if(FRAME_CODED(l) && COMPATIBLE_FRAME(l))
1411
                        transform |= PL;
1412
                }
1413
                if(y){
1414
                    u= i-fragment_width;
1415
                    vu = DC_COEFF(u);
1416
                    if(FRAME_CODED(u) && COMPATIBLE_FRAME(u))
1417
                        transform |= PU;
1418
                    if(x){
1419
                        ul= i-fragment_width-1;
1420
                        vul = DC_COEFF(ul);
1421
                        if(FRAME_CODED(ul) && COMPATIBLE_FRAME(ul))
1422
                            transform |= PUL;
1423
                    }
1424
                    if(x + 1 < fragment_width){
1425
                        ur= i-fragment_width+1;
1426
                        vur = DC_COEFF(ur);
1427
                        if(FRAME_CODED(ur) && COMPATIBLE_FRAME(ur))
1428
                            transform |= PUR;
1429
                    }
1430
                }
1431

    
1432
                debug_dc_pred("transform = %d, ", transform);
1433

    
1434
                if (transform == 0) {
1435

    
1436
                    /* if there were no fragments to predict from, use last
1437
                     * DC saved */
1438
                    predicted_dc = last_dc[current_frame_type];
1439
                    debug_dc_pred("from last DC (%d) = %d\n",
1440
                        current_frame_type, DC_COEFF(i));
1441

    
1442
                } else {
1443

    
1444
                    /* apply the appropriate predictor transform */
1445
                    predicted_dc =
1446
                        (predictor_transform[transform][0] * vul) +
1447
                        (predictor_transform[transform][1] * vu) +
1448
                        (predictor_transform[transform][2] * vur) +
1449
                        (predictor_transform[transform][3] * vl);
1450

    
1451
                    predicted_dc /= 128;
1452

    
1453
                    /* check for outranging on the [ul u l] and
1454
                     * [ul u ur l] predictors */
1455
                    if ((transform == 13) || (transform == 15)) {
1456
                        if (FFABS(predicted_dc - vu) > 128)
1457
                            predicted_dc = vu;
1458
                        else if (FFABS(predicted_dc - vl) > 128)
1459
                            predicted_dc = vl;
1460
                        else if (FFABS(predicted_dc - vul) > 128)
1461
                            predicted_dc = vul;
1462
                    }
1463

    
1464
                    debug_dc_pred("from pred DC = %d\n",
1465
                    DC_COEFF(i));
1466
                }
1467

    
1468
                /* at long last, apply the predictor */
1469
                if(s->coeffs[i].index){
1470
                    *s->next_coeff= s->coeffs[i];
1471
                    s->coeffs[i].index=0;
1472
                    s->coeffs[i].coeff=0;
1473
                    s->coeffs[i].next= s->next_coeff++;
1474
                }
1475
                s->coeffs[i].coeff += predicted_dc;
1476
                /* save the DC */
1477
                last_dc[current_frame_type] = DC_COEFF(i);
1478
                if(DC_COEFF(i) && !(s->all_fragments[i].coeff_count&127)){
1479
                    s->all_fragments[i].coeff_count= 129;
1480
//                    s->all_fragments[i].next_coeff= s->next_coeff;
1481
                    s->coeffs[i].next= s->next_coeff;
1482
                    (s->next_coeff++)->next=NULL;
1483
                }
1484
            }
1485
        }
1486
    }
1487
}
1488

    
1489

    
1490
static void horizontal_filter(unsigned char *first_pixel, int stride,
1491
    int *bounding_values);
1492
static void vertical_filter(unsigned char *first_pixel, int stride,
1493
    int *bounding_values);
1494

    
1495
/*
1496
 * Perform the final rendering for a particular slice of data.
1497
 * The slice number ranges from 0..(macroblock_height - 1).
1498
 */
1499
static void render_slice(Vp3DecodeContext *s, int slice)
1500
{
1501
    int x;
1502
    int m, n;
1503
    int16_t *dequantizer;
1504
    DECLARE_ALIGNED_16(DCTELEM, block[64]);
1505
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1506
    int motion_halfpel_index;
1507
    uint8_t *motion_source;
1508
    int plane;
1509
    int current_macroblock_entry = slice * s->macroblock_width * 6;
1510

    
1511
    if (slice >= s->macroblock_height)
1512
        return;
1513

    
1514
    for (plane = 0; plane < 3; plane++) {
1515
        uint8_t *output_plane = s->current_frame.data    [plane];
1516
        uint8_t *  last_plane = s->   last_frame.data    [plane];
1517
        uint8_t *golden_plane = s-> golden_frame.data    [plane];
1518
        int stride            = s->current_frame.linesize[plane];
1519
        int plane_width       = s->width  >> !!plane;
1520
        int plane_height      = s->height >> !!plane;
1521
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
1522
        int slice_height = y + (FRAGMENT_PIXELS << !plane);
1523
        int i = s->macroblock_fragments[current_macroblock_entry + plane + 3*!!plane];
1524

    
1525
        if (!s->flipped_image) stride = -stride;
1526

    
1527

    
1528
        if(FFABS(stride) > 2048)
1529
            return; //various tables are fixed size
1530

    
1531
        /* for each fragment row in the slice (both of them)... */
1532
        for (; y < slice_height; y += 8) {
1533

    
1534
            /* for each fragment in a row... */
1535
            for (x = 0; x < plane_width; x += 8, i++) {
1536

    
1537
                if ((i < 0) || (i >= s->fragment_count)) {
1538
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
1539
                    return;
1540
                }
1541

    
1542
                /* transform if this block was coded */
1543
                if ((s->all_fragments[i].coding_method != MODE_COPY) &&
1544
                    !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
1545

    
1546
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1547
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1548
                        motion_source= golden_plane;
1549
                    else
1550
                        motion_source= last_plane;
1551

    
1552
                    motion_source += s->all_fragments[i].first_pixel;
1553
                    motion_halfpel_index = 0;
1554

    
1555
                    /* sort out the motion vector if this fragment is coded
1556
                     * using a motion vector method */
1557
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1558
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1559
                        int src_x, src_y;
1560
                        motion_x = s->all_fragments[i].motion_x;
1561
                        motion_y = s->all_fragments[i].motion_y;
1562
                        if(plane){
1563
                            motion_x= (motion_x>>1) | (motion_x&1);
1564
                            motion_y= (motion_y>>1) | (motion_y&1);
1565
                        }
1566

    
1567
                        src_x= (motion_x>>1) + x;
1568
                        src_y= (motion_y>>1) + y;
1569
                        if ((motion_x == 127) || (motion_y == 127))
1570
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1571

    
1572
                        motion_halfpel_index = motion_x & 0x01;
1573
                        motion_source += (motion_x >> 1);
1574

    
1575
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1576
                        motion_source += ((motion_y >> 1) * stride);
1577

    
1578
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1579
                            uint8_t *temp= s->edge_emu_buffer;
1580
                            if(stride<0) temp -= 9*stride;
1581
                            else temp += 9*stride;
1582

    
1583
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1584
                            motion_source= temp;
1585
                        }
1586
                    }
1587

    
1588

    
1589
                    /* first, take care of copying a block from either the
1590
                     * previous or the golden frame */
1591
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1592
                        /* Note, it is possible to implement all MC cases with
1593
                           put_no_rnd_pixels_l2 which would look more like the
1594
                           VP3 source but this would be slower as
1595
                           put_no_rnd_pixels_tab is better optimzed */
1596
                        if(motion_halfpel_index != 3){
1597
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1598
                                output_plane + s->all_fragments[i].first_pixel,
1599
                                motion_source, stride, 8);
1600
                        }else{
1601
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1602
                            s->dsp.put_no_rnd_pixels_l2[1](
1603
                                output_plane + s->all_fragments[i].first_pixel,
1604
                                motion_source - d,
1605
                                motion_source + stride + 1 + d,
1606
                                stride, 8);
1607
                        }
1608
                        dequantizer = s->qmat[1][plane];
1609
                    }else{
1610
                        dequantizer = s->qmat[0][plane];
1611
                    }
1612

    
1613
                    /* dequantize the DCT coefficients */
1614
                    debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n",
1615
                        i, s->all_fragments[i].coding_method,
1616
                        DC_COEFF(i), dequantizer[0]);
1617

    
1618
                    if(s->avctx->idct_algo==FF_IDCT_VP3){
1619
                        Coeff *coeff= s->coeffs + i;
1620
                        memset(block, 0, sizeof(block));
1621
                        while(coeff->next){
1622
                            block[coeff->index]= coeff->coeff * dequantizer[coeff->index];
1623
                            coeff= coeff->next;
1624
                        }
1625
                    }else{
1626
                        Coeff *coeff= s->coeffs + i;
1627
                        memset(block, 0, sizeof(block));
1628
                        while(coeff->next){
1629
                            block[coeff->index]= (coeff->coeff * dequantizer[coeff->index] + 2)>>2;
1630
                            coeff= coeff->next;
1631
                        }
1632
                    }
1633

    
1634
                    /* invert DCT and place (or add) in final output */
1635

    
1636
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1637
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1638
                            block[0] += 128<<3;
1639
                        s->dsp.idct_put(
1640
                            output_plane + s->all_fragments[i].first_pixel,
1641
                            stride,
1642
                            block);
1643
                    } else {
1644
                        s->dsp.idct_add(
1645
                            output_plane + s->all_fragments[i].first_pixel,
1646
                            stride,
1647
                            block);
1648
                    }
1649

    
1650
                    debug_idct("block after idct_%s():\n",
1651
                        (s->all_fragments[i].coding_method == MODE_INTRA)?
1652
                        "put" : "add");
1653
                    for (m = 0; m < 8; m++) {
1654
                        for (n = 0; n < 8; n++) {
1655
                            debug_idct(" %3d", *(output_plane +
1656
                                s->all_fragments[i].first_pixel + (m * stride + n)));
1657
                        }
1658
                        debug_idct("\n");
1659
                    }
1660
                    debug_idct("\n");
1661

    
1662
                } else {
1663

    
1664
                    /* copy directly from the previous frame */
1665
                    s->dsp.put_pixels_tab[1][0](
1666
                        output_plane + s->all_fragments[i].first_pixel,
1667
                        last_plane + s->all_fragments[i].first_pixel,
1668
                        stride, 8);
1669

    
1670
                }
1671
#if 0
1672
                /* perform the left edge filter if:
1673
                 *   - the fragment is not on the left column
1674
                 *   - the fragment is coded in this frame
1675
                 *   - the fragment is not coded in this frame but the left
1676
                 *     fragment is coded in this frame (this is done instead
1677
                 *     of a right edge filter when rendering the left fragment
1678
                 *     since this fragment is not available yet) */
1679
                if ((x > 0) &&
1680
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1681
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1682
                      (s->all_fragments[i - 1].coding_method != MODE_COPY)) )) {
1683
                    horizontal_filter(
1684
                        output_plane + s->all_fragments[i].first_pixel + 7*stride,
1685
                        -stride, s->bounding_values_array + 127);
1686
                }
1687

1688
                /* perform the top edge filter if:
1689
                 *   - the fragment is not on the top row
1690
                 *   - the fragment is coded in this frame
1691
                 *   - the fragment is not coded in this frame but the above
1692
                 *     fragment is coded in this frame (this is done instead
1693
                 *     of a bottom edge filter when rendering the above
1694
                 *     fragment since this fragment is not available yet) */
1695
                if ((y > 0) &&
1696
                    ((s->all_fragments[i].coding_method != MODE_COPY) ||
1697
                     ((s->all_fragments[i].coding_method == MODE_COPY) &&
1698
                      (s->all_fragments[i - fragment_width].coding_method != MODE_COPY)) )) {
1699
                    vertical_filter(
1700
                        output_plane + s->all_fragments[i].first_pixel - stride,
1701
                        -stride, s->bounding_values_array + 127);
1702
                }
1703
#endif
1704
            }
1705
        }
1706
    }
1707

    
1708
     /* this looks like a good place for slice dispatch... */
1709
     /* algorithm:
1710
      *   if (slice == s->macroblock_height - 1)
1711
      *     dispatch (both last slice & 2nd-to-last slice);
1712
      *   else if (slice > 0)
1713
      *     dispatch (slice - 1);
1714
      */
1715

    
1716
    emms_c();
1717
}
1718

    
1719
static void horizontal_filter(unsigned char *first_pixel, int stride,
1720
    int *bounding_values)
1721
{
1722
    unsigned char *end;
1723
    int filter_value;
1724

    
1725
    for (end= first_pixel + 8*stride; first_pixel != end; first_pixel += stride) {
1726
        filter_value =
1727
            (first_pixel[-2] - first_pixel[ 1])
1728
         +3*(first_pixel[ 0] - first_pixel[-1]);
1729
        filter_value = bounding_values[(filter_value + 4) >> 3];
1730
        first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
1731
        first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
1732
    }
1733
}
1734

    
1735
static void vertical_filter(unsigned char *first_pixel, int stride,
1736
    int *bounding_values)
1737
{
1738
    unsigned char *end;
1739
    int filter_value;
1740
    const int nstride= -stride;
1741

    
1742
    for (end= first_pixel + 8; first_pixel < end; first_pixel++) {
1743
        filter_value =
1744
            (first_pixel[2 * nstride] - first_pixel[ stride])
1745
         +3*(first_pixel[0          ] - first_pixel[nstride]);
1746
        filter_value = bounding_values[(filter_value + 4) >> 3];
1747
        first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value);
1748
        first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value);
1749
    }
1750
}
1751

    
1752
static void apply_loop_filter(Vp3DecodeContext *s)
1753
{
1754
    int plane;
1755
    int x, y;
1756
    int *bounding_values= s->bounding_values_array+127;
1757

    
1758
#if 0
1759
    int bounding_values_array[256];
1760
    int filter_limit;
1761

1762
    /* find the right loop limit value */
1763
    for (x = 63; x >= 0; x--) {
1764
        if (vp31_ac_scale_factor[x] >= s->quality_index)
1765
            break;
1766
    }
1767
    filter_limit = vp31_filter_limit_values[s->quality_index];
1768

1769
    /* set up the bounding values */
1770
    memset(bounding_values_array, 0, 256 * sizeof(int));
1771
    for (x = 0; x < filter_limit; x++) {
1772
        bounding_values[-x - filter_limit] = -filter_limit + x;
1773
        bounding_values[-x] = -x;
1774
        bounding_values[x] = x;
1775
        bounding_values[x + filter_limit] = filter_limit - x;
1776
    }
1777
#endif
1778

    
1779
    for (plane = 0; plane < 3; plane++) {
1780
        int width           = s->fragment_width  >> !!plane;
1781
        int height          = s->fragment_height >> !!plane;
1782
        int fragment        = s->fragment_start        [plane];
1783
        int stride          = s->current_frame.linesize[plane];
1784
        uint8_t *plane_data = s->current_frame.data    [plane];
1785
        if (!s->flipped_image) stride = -stride;
1786

    
1787
        for (y = 0; y < height; y++) {
1788

    
1789
            for (x = 0; x < width; x++) {
1790
START_TIMER
1791
                /* do not perform left edge filter for left columns frags */
1792
                if ((x > 0) &&
1793
                    (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1794
                    horizontal_filter(
1795
                        plane_data + s->all_fragments[fragment].first_pixel,
1796
                        stride, bounding_values);
1797
                }
1798

    
1799
                /* do not perform top edge filter for top row fragments */
1800
                if ((y > 0) &&
1801
                    (s->all_fragments[fragment].coding_method != MODE_COPY)) {
1802
                    vertical_filter(
1803
                        plane_data + s->all_fragments[fragment].first_pixel,
1804
                        stride, bounding_values);
1805
                }
1806

    
1807
                /* do not perform right edge filter for right column
1808
                 * fragments or if right fragment neighbor is also coded
1809
                 * in this frame (it will be filtered in next iteration) */
1810
                if ((x < width - 1) &&
1811
                    (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1812
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1813
                    horizontal_filter(
1814
                        plane_data + s->all_fragments[fragment + 1].first_pixel,
1815
                        stride, bounding_values);
1816
                }
1817

    
1818
                /* do not perform bottom edge filter for bottom row
1819
                 * fragments or if bottom fragment neighbor is also coded
1820
                 * in this frame (it will be filtered in the next row) */
1821
                if ((y < height - 1) &&
1822
                    (s->all_fragments[fragment].coding_method != MODE_COPY) &&
1823
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1824
                    vertical_filter(
1825
                        plane_data + s->all_fragments[fragment + width].first_pixel,
1826
                        stride, bounding_values);
1827
                }
1828

    
1829
                fragment++;
1830
STOP_TIMER("loop filter")
1831
            }
1832
        }
1833
    }
1834
}
1835

    
1836
/*
1837
 * This function computes the first pixel addresses for each fragment.
1838
 * This function needs to be invoked after the first frame is allocated
1839
 * so that it has access to the plane strides.
1840
 */
1841
static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
1842
{
1843

    
1844
    int i, x, y;
1845

    
1846
    /* figure out the first pixel addresses for each of the fragments */
1847
    /* Y plane */
1848
    i = 0;
1849
    for (y = s->fragment_height; y > 0; y--) {
1850
        for (x = 0; x < s->fragment_width; x++) {
1851
            s->all_fragments[i++].first_pixel =
1852
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1853
                    s->golden_frame.linesize[0] +
1854
                    x * FRAGMENT_PIXELS;
1855
            debug_init("  fragment %d, first pixel @ %d\n",
1856
                i-1, s->all_fragments[i-1].first_pixel);
1857
        }
1858
    }
1859

    
1860
    /* U plane */
1861
    i = s->fragment_start[1];
1862
    for (y = s->fragment_height / 2; y > 0; y--) {
1863
        for (x = 0; x < s->fragment_width / 2; x++) {
1864
            s->all_fragments[i++].first_pixel =
1865
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1866
                    s->golden_frame.linesize[1] +
1867
                    x * FRAGMENT_PIXELS;
1868
            debug_init("  fragment %d, first pixel @ %d\n",
1869
                i-1, s->all_fragments[i-1].first_pixel);
1870
        }
1871
    }
1872

    
1873
    /* V plane */
1874
    i = s->fragment_start[2];
1875
    for (y = s->fragment_height / 2; y > 0; y--) {
1876
        for (x = 0; x < s->fragment_width / 2; x++) {
1877
            s->all_fragments[i++].first_pixel =
1878
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1879
                    s->golden_frame.linesize[2] +
1880
                    x * FRAGMENT_PIXELS;
1881
            debug_init("  fragment %d, first pixel @ %d\n",
1882
                i-1, s->all_fragments[i-1].first_pixel);
1883
        }
1884
    }
1885
}
1886

    
1887
/* FIXME: this should be merged with the above! */
1888
static void theora_calculate_pixel_addresses(Vp3DecodeContext *s)
1889
{
1890

    
1891
    int i, x, y;
1892

    
1893
    /* figure out the first pixel addresses for each of the fragments */
1894
    /* Y plane */
1895
    i = 0;
1896
    for (y = 1; y <= s->fragment_height; y++) {
1897
        for (x = 0; x < s->fragment_width; x++) {
1898
            s->all_fragments[i++].first_pixel =
1899
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
1900
                    s->golden_frame.linesize[0] +
1901
                    x * FRAGMENT_PIXELS;
1902
            debug_init("  fragment %d, first pixel @ %d\n",
1903
                i-1, s->all_fragments[i-1].first_pixel);
1904
        }
1905
    }
1906

    
1907
    /* U plane */
1908
    i = s->fragment_start[1];
1909
    for (y = 1; y <= s->fragment_height / 2; y++) {
1910
        for (x = 0; x < s->fragment_width / 2; x++) {
1911
            s->all_fragments[i++].first_pixel =
1912
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
1913
                    s->golden_frame.linesize[1] +
1914
                    x * FRAGMENT_PIXELS;
1915
            debug_init("  fragment %d, first pixel @ %d\n",
1916
                i-1, s->all_fragments[i-1].first_pixel);
1917
        }
1918
    }
1919

    
1920
    /* V plane */
1921
    i = s->fragment_start[2];
1922
    for (y = 1; y <= s->fragment_height / 2; y++) {
1923
        for (x = 0; x < s->fragment_width / 2; x++) {
1924
            s->all_fragments[i++].first_pixel =
1925
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
1926
                    s->golden_frame.linesize[2] +
1927
                    x * FRAGMENT_PIXELS;
1928
            debug_init("  fragment %d, first pixel @ %d\n",
1929
                i-1, s->all_fragments[i-1].first_pixel);
1930
        }
1931
    }
1932
}
1933

    
1934
/*
1935
 * This is the ffmpeg/libavcodec API init function.
1936
 */
1937
static int vp3_decode_init(AVCodecContext *avctx)
1938
{
1939
    Vp3DecodeContext *s = avctx->priv_data;
1940
    int i, inter, plane;
1941
    int c_width;
1942
    int c_height;
1943
    int y_superblock_count;
1944
    int c_superblock_count;
1945

    
1946
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1947
        s->version = 0;
1948
    else
1949
        s->version = 1;
1950

    
1951
    s->avctx = avctx;
1952
    s->width = (avctx->width + 15) & 0xFFFFFFF0;
1953
    s->height = (avctx->height + 15) & 0xFFFFFFF0;
1954
    avctx->pix_fmt = PIX_FMT_YUV420P;
1955
    if(avctx->idct_algo==FF_IDCT_AUTO)
1956
        avctx->idct_algo=FF_IDCT_VP3;
1957
    dsputil_init(&s->dsp, avctx);
1958

    
1959
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1960

    
1961
    /* initialize to an impossible value which will force a recalculation
1962
     * in the first frame decode */
1963
    s->quality_index = -1;
1964

    
1965
    s->y_superblock_width = (s->width + 31) / 32;
1966
    s->y_superblock_height = (s->height + 31) / 32;
1967
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1968

    
1969
    /* work out the dimensions for the C planes */
1970
    c_width = s->width / 2;
1971
    c_height = s->height / 2;
1972
    s->c_superblock_width = (c_width + 31) / 32;
1973
    s->c_superblock_height = (c_height + 31) / 32;
1974
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1975

    
1976
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
1977
    s->u_superblock_start = y_superblock_count;
1978
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
1979
    s->superblock_coding = av_malloc(s->superblock_count);
1980

    
1981
    s->macroblock_width = (s->width + 15) / 16;
1982
    s->macroblock_height = (s->height + 15) / 16;
1983
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1984

    
1985
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1986
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1987

    
1988
    /* fragment count covers all 8x8 blocks for all 3 planes */
1989
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1990
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1991
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1992

    
1993
    debug_init("  Y plane: %d x %d\n", s->width, s->height);
1994
    debug_init("  C plane: %d x %d\n", c_width, c_height);
1995
    debug_init("  Y superblocks: %d x %d, %d total\n",
1996
        s->y_superblock_width, s->y_superblock_height, y_superblock_count);
1997
    debug_init("  C superblocks: %d x %d, %d total\n",
1998
        s->c_superblock_width, s->c_superblock_height, c_superblock_count);
1999
    debug_init("  total superblocks = %d, U starts @ %d, V starts @ %d\n",
2000
        s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2001
    debug_init("  macroblocks: %d x %d, %d total\n",
2002
        s->macroblock_width, s->macroblock_height, s->macroblock_count);
2003
    debug_init("  %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2004
        s->fragment_count,
2005
        s->fragment_width,
2006
        s->fragment_height,
2007
        s->fragment_start[1],
2008
        s->fragment_start[2]);
2009

    
2010
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2011
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
2012
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2013
    s->pixel_addresses_inited = 0;
2014

    
2015
    if (!s->theora_tables)
2016
    {
2017
        for (i = 0; i < 64; i++) {
2018
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2019
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2020
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
2021
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
2022
            s->base_matrix[2][i] = vp31_inter_dequant[i];
2023
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
2024
        }
2025

    
2026
        for(inter=0; inter<2; inter++){
2027
            for(plane=0; plane<3; plane++){
2028
                s->qr_count[inter][plane]= 1;
2029
                s->qr_size [inter][plane][0]= 63;
2030
                s->qr_base [inter][plane][0]=
2031
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
2032
            }
2033
        }
2034

    
2035
        /* init VLC tables */
2036
        for (i = 0; i < 16; i++) {
2037

    
2038
            /* DC histograms */
2039
            init_vlc(&s->dc_vlc[i], 5, 32,
2040
                &dc_bias[i][0][1], 4, 2,
2041
                &dc_bias[i][0][0], 4, 2, 0);
2042

    
2043
            /* group 1 AC histograms */
2044
            init_vlc(&s->ac_vlc_1[i], 5, 32,
2045
                &ac_bias_0[i][0][1], 4, 2,
2046
                &ac_bias_0[i][0][0], 4, 2, 0);
2047

    
2048
            /* group 2 AC histograms */
2049
            init_vlc(&s->ac_vlc_2[i], 5, 32,
2050
                &ac_bias_1[i][0][1], 4, 2,
2051
                &ac_bias_1[i][0][0], 4, 2, 0);
2052

    
2053
            /* group 3 AC histograms */
2054
            init_vlc(&s->ac_vlc_3[i], 5, 32,
2055
                &ac_bias_2[i][0][1], 4, 2,
2056
                &ac_bias_2[i][0][0], 4, 2, 0);
2057

    
2058
            /* group 4 AC histograms */
2059
            init_vlc(&s->ac_vlc_4[i], 5, 32,
2060
                &ac_bias_3[i][0][1], 4, 2,
2061
                &ac_bias_3[i][0][0], 4, 2, 0);
2062
        }
2063
    } else {
2064
        for (i = 0; i < 16; i++) {
2065

    
2066
            /* DC histograms */
2067
            init_vlc(&s->dc_vlc[i], 5, 32,
2068
                &s->huffman_table[i][0][1], 4, 2,
2069
                &s->huffman_table[i][0][0], 4, 2, 0);
2070

    
2071
            /* group 1 AC histograms */
2072
            init_vlc(&s->ac_vlc_1[i], 5, 32,
2073
                &s->huffman_table[i+16][0][1], 4, 2,
2074
                &s->huffman_table[i+16][0][0], 4, 2, 0);
2075

    
2076
            /* group 2 AC histograms */
2077
            init_vlc(&s->ac_vlc_2[i], 5, 32,
2078
                &s->huffman_table[i+16*2][0][1], 4, 2,
2079
                &s->huffman_table[i+16*2][0][0], 4, 2, 0);
2080

    
2081
            /* group 3 AC histograms */
2082
            init_vlc(&s->ac_vlc_3[i], 5, 32,
2083
                &s->huffman_table[i+16*3][0][1], 4, 2,
2084
                &s->huffman_table[i+16*3][0][0], 4, 2, 0);
2085

    
2086
            /* group 4 AC histograms */
2087
            init_vlc(&s->ac_vlc_4[i], 5, 32,
2088
                &s->huffman_table[i+16*4][0][1], 4, 2,
2089
                &s->huffman_table[i+16*4][0][0], 4, 2, 0);
2090
        }
2091
    }
2092

    
2093
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
2094
        &superblock_run_length_vlc_table[0][1], 4, 2,
2095
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
2096

    
2097
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
2098
        &fragment_run_length_vlc_table[0][1], 4, 2,
2099
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
2100

    
2101
    init_vlc(&s->mode_code_vlc, 3, 8,
2102
        &mode_code_vlc_table[0][1], 2, 1,
2103
        &mode_code_vlc_table[0][0], 2, 1, 0);
2104

    
2105
    init_vlc(&s->motion_vector_vlc, 6, 63,
2106
        &motion_vector_vlc_table[0][1], 2, 1,
2107
        &motion_vector_vlc_table[0][0], 2, 1, 0);
2108

    
2109
    /* work out the block mapping tables */
2110
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2111
    s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2112
    s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2113
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2114
    init_block_mapping(s);
2115

    
2116
    for (i = 0; i < 3; i++) {
2117
        s->current_frame.data[i] = NULL;
2118
        s->last_frame.data[i] = NULL;
2119
        s->golden_frame.data[i] = NULL;
2120
    }
2121

    
2122
    return 0;
2123
}
2124

    
2125
/*
2126
 * This is the ffmpeg/libavcodec API frame decode function.
2127
 */
2128
static int vp3_decode_frame(AVCodecContext *avctx,
2129
                            void *data, int *data_size,
2130
                            uint8_t *buf, int buf_size)
2131
{
2132
    Vp3DecodeContext *s = avctx->priv_data;
2133
    GetBitContext gb;
2134
    static int counter = 0;
2135
    int i;
2136

    
2137
    init_get_bits(&gb, buf, buf_size * 8);
2138

    
2139
    if (s->theora && get_bits1(&gb))
2140
    {
2141
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
2142
        return -1;
2143
    }
2144

    
2145
    s->keyframe = !get_bits1(&gb);
2146
    if (!s->theora)
2147
        skip_bits(&gb, 1);
2148
    s->last_quality_index = s->quality_index;
2149

    
2150
    s->nqis=0;
2151
    do{
2152
        s->qis[s->nqis++]= get_bits(&gb, 6);
2153
    } while(s->theora >= 0x030200 && s->nqis<3 && get_bits1(&gb));
2154

    
2155
    s->quality_index= s->qis[0];
2156

    
2157
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2158
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2159
            s->keyframe?"key":"", counter, s->quality_index);
2160
    counter++;
2161

    
2162
    if (s->quality_index != s->last_quality_index) {
2163
        init_dequantizer(s);
2164
        init_loop_filter(s);
2165
    }
2166

    
2167
    if (s->keyframe) {
2168
        if (!s->theora)
2169
        {
2170
            skip_bits(&gb, 4); /* width code */
2171
            skip_bits(&gb, 4); /* height code */
2172
            if (s->version)
2173
            {
2174
                s->version = get_bits(&gb, 5);
2175
                if (counter == 1)
2176
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2177
            }
2178
        }
2179
        if (s->version || s->theora)
2180
        {
2181
                if (get_bits1(&gb))
2182
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2183
            skip_bits(&gb, 2); /* reserved? */
2184
        }
2185

    
2186
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2187
            if (s->golden_frame.data[0])
2188
                avctx->release_buffer(avctx, &s->golden_frame);
2189
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2190
        } else {
2191
            if (s->golden_frame.data[0])
2192
                avctx->release_buffer(avctx, &s->golden_frame);
2193
            if (s->last_frame.data[0])
2194
                avctx->release_buffer(avctx, &s->last_frame);
2195
        }
2196

    
2197
        s->golden_frame.reference = 3;
2198
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2199
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2200
            return -1;
2201
        }
2202

    
2203
        /* golden frame is also the current frame */
2204
        s->current_frame= s->golden_frame;
2205

    
2206
        /* time to figure out pixel addresses? */
2207
        if (!s->pixel_addresses_inited)
2208
        {
2209
            if (!s->flipped_image)
2210
                vp3_calculate_pixel_addresses(s);
2211
            else
2212
                theora_calculate_pixel_addresses(s);
2213
            s->pixel_addresses_inited = 1;
2214
        }
2215
    } else {
2216
        /* allocate a new current frame */
2217
        s->current_frame.reference = 3;
2218
        if (!s->pixel_addresses_inited) {
2219
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
2220
            return -1;
2221
        }
2222
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2223
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2224
            return -1;
2225
        }
2226
    }
2227

    
2228
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2229
    s->current_frame.qstride= 0;
2230

    
2231
    {START_TIMER
2232
    init_frame(s, &gb);
2233
    STOP_TIMER("init_frame")}
2234

    
2235
#if KEYFRAMES_ONLY
2236
if (!s->keyframe) {
2237

    
2238
    memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2239
        s->current_frame.linesize[0] * s->height);
2240
    memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2241
        s->current_frame.linesize[1] * s->height / 2);
2242
    memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2243
        s->current_frame.linesize[2] * s->height / 2);
2244

    
2245
} else {
2246
#endif
2247

    
2248
    {START_TIMER
2249
    if (unpack_superblocks(s, &gb)){
2250
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
2251
        return -1;
2252
    }
2253
    STOP_TIMER("unpack_superblocks")}
2254
    {START_TIMER
2255
    if (unpack_modes(s, &gb)){
2256
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
2257
        return -1;
2258
    }
2259
    STOP_TIMER("unpack_modes")}
2260
    {START_TIMER
2261
    if (unpack_vectors(s, &gb)){
2262
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
2263
        return -1;
2264
    }
2265
    STOP_TIMER("unpack_vectors")}
2266
    {START_TIMER
2267
    if (unpack_dct_coeffs(s, &gb)){
2268
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
2269
        return -1;
2270
    }
2271
    STOP_TIMER("unpack_dct_coeffs")}
2272
    {START_TIMER
2273

    
2274
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2275
    if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2276
        reverse_dc_prediction(s, s->fragment_start[1],
2277
            s->fragment_width / 2, s->fragment_height / 2);
2278
        reverse_dc_prediction(s, s->fragment_start[2],
2279
            s->fragment_width / 2, s->fragment_height / 2);
2280
    }
2281
    STOP_TIMER("reverse_dc_prediction")}
2282
    {START_TIMER
2283

    
2284
    for (i = 0; i < s->macroblock_height; i++)
2285
        render_slice(s, i);
2286
    STOP_TIMER("render_fragments")}
2287

    
2288
    {START_TIMER
2289
    apply_loop_filter(s);
2290
    STOP_TIMER("apply_loop_filter")}
2291
#if KEYFRAMES_ONLY
2292
}
2293
#endif
2294

    
2295
    *data_size=sizeof(AVFrame);
2296
    *(AVFrame*)data= s->current_frame;
2297

    
2298
    /* release the last frame, if it is allocated and if it is not the
2299
     * golden frame */
2300
    if ((s->last_frame.data[0]) &&
2301
        (s->last_frame.data[0] != s->golden_frame.data[0]))
2302
        avctx->release_buffer(avctx, &s->last_frame);
2303

    
2304
    /* shuffle frames (last = current) */
2305
    s->last_frame= s->current_frame;
2306
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2307

    
2308
    return buf_size;
2309
}
2310

    
2311
/*
2312
 * This is the ffmpeg/libavcodec API module cleanup function.
2313
 */
2314
static int vp3_decode_end(AVCodecContext *avctx)
2315
{
2316
    Vp3DecodeContext *s = avctx->priv_data;
2317

    
2318
    av_free(s->all_fragments);
2319
    av_free(s->coeffs);
2320
    av_free(s->coded_fragment_list);
2321
    av_free(s->superblock_fragments);
2322
    av_free(s->superblock_macroblocks);
2323
    av_free(s->macroblock_fragments);
2324
    av_free(s->macroblock_coding);
2325

    
2326
    /* release all frames */
2327
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2328
        avctx->release_buffer(avctx, &s->golden_frame);
2329
    if (s->last_frame.data[0])
2330
        avctx->release_buffer(avctx, &s->last_frame);
2331
    /* no need to release the current_frame since it will always be pointing
2332
     * to the same frame as either the golden or last frame */
2333

    
2334
    return 0;
2335
}
2336

    
2337
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
2338
{
2339
    Vp3DecodeContext *s = avctx->priv_data;
2340

    
2341
    if (get_bits(gb, 1)) {
2342
        int token;
2343
        if (s->entries >= 32) { /* overflow */
2344
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2345
            return -1;
2346
        }
2347
        token = get_bits(gb, 5);
2348
        //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);
2349
        s->huffman_table[s->hti][token][0] = s->hbits;
2350
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
2351
        s->entries++;
2352
    }
2353
    else {
2354
        if (s->huff_code_size >= 32) {/* overflow */
2355
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
2356
            return -1;
2357
        }
2358
        s->huff_code_size++;
2359
        s->hbits <<= 1;
2360
        read_huffman_tree(avctx, gb);
2361
        s->hbits |= 1;
2362
        read_huffman_tree(avctx, gb);
2363
        s->hbits >>= 1;
2364
        s->huff_code_size--;
2365
    }
2366
    return 0;
2367
}
2368

    
2369
#ifdef CONFIG_THEORA_DECODER
2370
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
2371
{
2372
    Vp3DecodeContext *s = avctx->priv_data;
2373
    int visible_width, visible_height;
2374

    
2375
    s->theora = get_bits_long(gb, 24);
2376
    av_log(avctx, AV_LOG_INFO, "Theora bitstream version %X\n", s->theora);
2377

    
2378
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
2379
    /* but previous versions have the image flipped relative to vp3 */
2380
    if (s->theora < 0x030200)
2381
    {
2382
        s->flipped_image = 1;
2383
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2384
    }
2385

    
2386
    s->width = get_bits(gb, 16) << 4;
2387
    s->height = get_bits(gb, 16) << 4;
2388

    
2389
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2390
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2391
        s->width= s->height= 0;
2392
        return -1;
2393
    }
2394

    
2395
    if (s->theora >= 0x030400)
2396
    {
2397
        skip_bits(gb, 32); /* total number of superblocks in a frame */
2398
        // fixme, the next field is 36bits long
2399
        skip_bits(gb, 32); /* total number of blocks in a frame */
2400
        skip_bits(gb, 4); /* total number of blocks in a frame */
2401
        skip_bits(gb, 32); /* total number of macroblocks in a frame */
2402
    }
2403

    
2404
    visible_width  = get_bits_long(gb, 24);
2405
    visible_height = get_bits_long(gb, 24);
2406

    
2407
    if (s->theora >= 0x030200) {
2408
        skip_bits(gb, 8); /* offset x */
2409
        skip_bits(gb, 8); /* offset y */
2410
    }
2411

    
2412
    skip_bits(gb, 32); /* fps numerator */
2413
    skip_bits(gb, 32); /* fps denumerator */
2414
    skip_bits(gb, 24); /* aspect numerator */
2415
    skip_bits(gb, 24); /* aspect denumerator */
2416

    
2417
    if (s->theora < 0x030200)
2418
        skip_bits(gb, 5); /* keyframe frequency force */
2419
    skip_bits(gb, 8); /* colorspace */
2420
    if (s->theora >= 0x030400)
2421
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2422
    skip_bits(gb, 24); /* bitrate */
2423

    
2424
    skip_bits(gb, 6); /* quality hint */
2425

    
2426
    if (s->theora >= 0x030200)
2427
    {
2428
        skip_bits(gb, 5); /* keyframe frequency force */
2429

    
2430
        if (s->theora < 0x030400)
2431
            skip_bits(gb, 5); /* spare bits */
2432
    }
2433

    
2434
//    align_get_bits(gb);
2435

    
2436
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2437
        && visible_height <= s->height && visible_height > s->height-16)
2438
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2439
    else
2440
        avcodec_set_dimensions(avctx, s->width, s->height);
2441

    
2442
    return 0;
2443
}
2444

    
2445
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2446
{
2447
    Vp3DecodeContext *s = avctx->priv_data;
2448
    int i, n, matrices, inter, plane;
2449

    
2450
    if (s->theora >= 0x030200) {
2451
        n = get_bits(gb, 3);
2452
        /* loop filter limit values table */
2453
        for (i = 0; i < 64; i++)
2454
            s->filter_limit_values[i] = get_bits(gb, n);
2455
    }
2456

    
2457
    if (s->theora >= 0x030200)
2458
        n = get_bits(gb, 4) + 1;
2459
    else
2460
        n = 16;
2461
    /* quality threshold table */
2462
    for (i = 0; i < 64; i++)
2463
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2464

    
2465
    if (s->theora >= 0x030200)
2466
        n = get_bits(gb, 4) + 1;
2467
    else
2468
        n = 16;
2469
    /* dc scale factor table */
2470
    for (i = 0; i < 64; i++)
2471
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2472

    
2473
    if (s->theora >= 0x030200)
2474
        matrices = get_bits(gb, 9) + 1;
2475
    else
2476
        matrices = 3;
2477

    
2478
    if(matrices > 384){
2479
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2480
        return -1;
2481
    }
2482

    
2483
    for(n=0; n<matrices; n++){
2484
        for (i = 0; i < 64; i++)
2485
            s->base_matrix[n][i]= get_bits(gb, 8);
2486
    }
2487

    
2488
    for (inter = 0; inter <= 1; inter++) {
2489
        for (plane = 0; plane <= 2; plane++) {
2490
            int newqr= 1;
2491
            if (inter || plane > 0)
2492
                newqr = get_bits(gb, 1);
2493
            if (!newqr) {
2494
                int qtj, plj;
2495
                if(inter && get_bits(gb, 1)){
2496
                    qtj = 0;
2497
                    plj = plane;
2498
                }else{
2499
                    qtj= (3*inter + plane - 1) / 3;
2500
                    plj= (plane + 2) % 3;
2501
                }
2502
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2503
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2504
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2505
            } else {
2506
                int qri= 0;
2507
                int qi = 0;
2508

    
2509
                for(;;){
2510
                    i= get_bits(gb, av_log2(matrices-1)+1);
2511
                    if(i>= matrices){
2512
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2513
                        return -1;
2514
                    }
2515
                    s->qr_base[inter][plane][qri]= i;
2516
                    if(qi >= 63)
2517
                        break;
2518
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2519
                    s->qr_size[inter][plane][qri++]= i;
2520
                    qi += i;
2521
                }
2522

    
2523
                if (qi > 63) {
2524
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2525
                    return -1;
2526
                }
2527
                s->qr_count[inter][plane]= qri;
2528
            }
2529
        }
2530
    }
2531

    
2532
    /* Huffman tables */
2533
    for (s->hti = 0; s->hti < 80; s->hti++) {
2534
        s->entries = 0;
2535
        s->huff_code_size = 1;
2536
        if (!get_bits(gb, 1)) {
2537
            s->hbits = 0;
2538
            read_huffman_tree(avctx, gb);
2539
            s->hbits = 1;
2540
            read_huffman_tree(avctx, gb);
2541
        }
2542
    }
2543

    
2544
    s->theora_tables = 1;
2545

    
2546
    return 0;
2547
}
2548

    
2549
static int theora_decode_init(AVCodecContext *avctx)
2550
{
2551
    Vp3DecodeContext *s = avctx->priv_data;
2552
    GetBitContext gb;
2553
    int ptype;
2554
    uint8_t *header_start[3];
2555
    int header_len[3];
2556
    int i;
2557

    
2558
    s->theora = 1;
2559

    
2560
    if (!avctx->extradata_size)
2561
    {
2562
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2563
        return -1;
2564
    }
2565

    
2566
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2567
                              42, header_start, header_len) < 0) {
2568
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2569
        return -1;
2570
    }
2571

    
2572
  for(i=0;i<3;i++) {
2573
    init_get_bits(&gb, header_start[i], header_len[i]);
2574

    
2575
    ptype = get_bits(&gb, 8);
2576
    debug_vp3("Theora headerpacket type: %x\n", ptype);
2577

    
2578
     if (!(ptype & 0x80))
2579
     {
2580
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2581
//        return -1;
2582
     }
2583

    
2584
    // FIXME: Check for this as well.
2585
    skip_bits(&gb, 6*8); /* "theora" */
2586

    
2587
    switch(ptype)
2588
    {
2589
        case 0x80:
2590
            theora_decode_header(avctx, &gb);
2591
                break;
2592
        case 0x81:
2593
// FIXME: is this needed? it breaks sometimes
2594
//            theora_decode_comments(avctx, gb);
2595
            break;
2596
        case 0x82:
2597
            theora_decode_tables(avctx, &gb);
2598
            break;
2599
        default:
2600
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2601
            break;
2602
    }
2603
    if(8*header_len[i] != get_bits_count(&gb))
2604
        av_log(avctx, AV_LOG_ERROR, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2605
    if (s->theora < 0x030200)
2606
        break;
2607
  }
2608

    
2609
    vp3_decode_init(avctx);
2610
    return 0;
2611
}
2612

    
2613
AVCodec theora_decoder = {
2614
    "theora",
2615
    CODEC_TYPE_VIDEO,
2616
    CODEC_ID_THEORA,
2617
    sizeof(Vp3DecodeContext),
2618
    theora_decode_init,
2619
    NULL,
2620
    vp3_decode_end,
2621
    vp3_decode_frame,
2622
    0,
2623
    NULL
2624
};
2625
#endif
2626

    
2627
AVCodec vp3_decoder = {
2628
    "vp3",
2629
    CODEC_TYPE_VIDEO,
2630
    CODEC_ID_VP3,
2631
    sizeof(Vp3DecodeContext),
2632
    vp3_decode_init,
2633
    NULL,
2634
    vp3_decode_end,
2635
    vp3_decode_frame,
2636
    0,
2637
    NULL
2638
};