Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 8a515d2d

History | View | Annotate | Download (93.7 KB)

1
/*
2
 * Copyright (C) 2003-2004 the ffmpeg project
3
 *
4
 * This library is free software; you can redistribute it and/or
5
 * modify it under the terms of the GNU Lesser General Public
6
 * License as published by the Free Software Foundation; either
7
 * version 2 of the License, or (at your option) any later version.
8
 *
9
 * This library is distributed in the hope that it will be useful,
10
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12
 * Lesser General Public License for more details.
13
 *
14
 * You should have received a copy of the GNU Lesser General Public
15
 * License along with this library; if not, write to the Free Software
16
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
17
 *
18
 * VP3 Video Decoder by Mike Melanson (melanson@pcisys.net)
19
 * For more information about the VP3 coding process, visit:
20
 *   http://www.pcisys.net/~melanson/codecs/
21
 *
22
 * Theora decoder by Alex Beregszaszi
23
 *
24
 */
25

    
26
/**
27
 * @file vp3.c
28
 * On2 VP3 Video Decoder
29
 */
30

    
31
#include <stdio.h>
32
#include <stdlib.h>
33
#include <string.h>
34
#include <unistd.h>
35

    
36
#include "common.h"
37
#include "avcodec.h"
38
#include "dsputil.h"
39
#include "mpegvideo.h"
40

    
41
#include "vp3data.h"
42

    
43
#define FRAGMENT_PIXELS 8
44

    
45
/* 
46
 * Debugging Variables
47
 * 
48
 * Define one or more of the following compile-time variables to 1 to obtain
49
 * elaborate information about certain aspects of the decoding process.
50
 *
51
 * KEYFRAMES_ONLY: set this to 1 to only see keyframes (VP3 slideshow mode)
52
 * DEBUG_VP3: high-level decoding flow
53
 * DEBUG_INIT: initialization parameters
54
 * DEBUG_DEQUANTIZERS: display how the dequanization tables are built
55
 * DEBUG_BLOCK_CODING: unpacking the superblock/macroblock/fragment coding
56
 * DEBUG_MODES: unpacking the coding modes for individual fragments
57
 * DEBUG_VECTORS: display the motion vectors
58
 * DEBUG_TOKEN: display exhaustive information about each DCT token
59
 * DEBUG_VLC: display the VLCs as they are extracted from the stream
60
 * DEBUG_DC_PRED: display the process of reversing DC prediction
61
 * DEBUG_IDCT: show every detail of the IDCT process
62
 */
63

    
64
#define KEYFRAMES_ONLY 0
65

    
66
#define DEBUG_VP3 0
67
#define DEBUG_INIT 0
68
#define DEBUG_DEQUANTIZERS 0
69
#define DEBUG_BLOCK_CODING 0
70
#define DEBUG_MODES 0
71
#define DEBUG_VECTORS 0
72
#define DEBUG_TOKEN 0
73
#define DEBUG_VLC 0
74
#define DEBUG_DC_PRED 0
75
#define DEBUG_IDCT 0
76

    
77
#if DEBUG_VP3
78
#define debug_vp3 printf
79
#else
80
static inline void debug_vp3(const char *format, ...) { }
81
#endif
82

    
83
#if DEBUG_INIT
84
#define debug_init printf
85
#else
86
static inline void debug_init(const char *format, ...) { }
87
#endif
88

    
89
#if DEBUG_DEQUANTIZERS
90
#define debug_dequantizers printf 
91
#else
92
static inline void debug_dequantizers(const char *format, ...) { } 
93
#endif
94

    
95
#if DEBUG_BLOCK_CODING
96
#define debug_block_coding printf 
97
#else
98
static inline void debug_block_coding(const char *format, ...) { } 
99
#endif
100

    
101
#if DEBUG_MODES
102
#define debug_modes printf 
103
#else
104
static inline void debug_modes(const char *format, ...) { } 
105
#endif
106

    
107
#if DEBUG_VECTORS
108
#define debug_vectors printf 
109
#else
110
static inline void debug_vectors(const char *format, ...) { } 
111
#endif
112

    
113
#if DEBUG_TOKEN 
114
#define debug_token printf 
115
#else
116
static inline void debug_token(const char *format, ...) { } 
117
#endif
118

    
119
#if DEBUG_VLC
120
#define debug_vlc printf 
121
#else
122
static inline void debug_vlc(const char *format, ...) { } 
123
#endif
124

    
125
#if DEBUG_DC_PRED
126
#define debug_dc_pred printf 
127
#else
128
static inline void debug_dc_pred(const char *format, ...) { } 
129
#endif
130

    
131
#if DEBUG_IDCT
132
#define debug_idct printf 
133
#else
134
static inline void debug_idct(const char *format, ...) { } 
135
#endif
136

    
137
typedef struct Vp3Fragment {
138
    DCTELEM coeffs[64];
139
    int coding_method;
140
    int coeff_count;
141
    int last_coeff;
142
    int motion_x;
143
    int motion_y;
144
    /* address of first pixel taking into account which plane the fragment
145
     * lives on as well as the plane stride */
146
    int first_pixel;
147
    /* this is the macroblock that the fragment belongs to */
148
    int macroblock;
149
} Vp3Fragment;
150

    
151
#define SB_NOT_CODED        0
152
#define SB_PARTIALLY_CODED  1
153
#define SB_FULLY_CODED      2
154

    
155
#define MODE_INTER_NO_MV      0
156
#define MODE_INTRA            1
157
#define MODE_INTER_PLUS_MV    2
158
#define MODE_INTER_LAST_MV    3
159
#define MODE_INTER_PRIOR_LAST 4
160
#define MODE_USING_GOLDEN     5
161
#define MODE_GOLDEN_MV        6
162
#define MODE_INTER_FOURMV     7
163
#define CODING_MODE_COUNT     8
164

    
165
/* special internal mode */
166
#define MODE_COPY             8
167

    
168
/* There are 6 preset schemes, plus a free-form scheme */
169
static int ModeAlphabet[7][CODING_MODE_COUNT] =
170
{
171
    /* this is the custom scheme */
172
    { 0, 0, 0, 0, 0, 0, 0, 0 },
173

    
174
    /* scheme 1: Last motion vector dominates */
175
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,  
176
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
177
         MODE_INTRA,            MODE_USING_GOLDEN,      
178
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
179

    
180
    /* scheme 2 */
181
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,  
182
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
183
         MODE_INTRA,            MODE_USING_GOLDEN,      
184
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
185

    
186
    /* scheme 3 */
187
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,     
188
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
189
         MODE_INTRA,            MODE_USING_GOLDEN,      
190
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
191

    
192
    /* scheme 4 */
193
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,     
194
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
195
         MODE_INTRA,            MODE_USING_GOLDEN,      
196
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
197

    
198
    /* scheme 5: No motion vector dominates */
199
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,     
200
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
201
         MODE_INTRA,            MODE_USING_GOLDEN,      
202
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
203

    
204
    /* scheme 6 */
205
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,      
206
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
207
         MODE_INTER_PLUS_MV,    MODE_INTRA,             
208
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
209

    
210
};
211

    
212
#define MIN_DEQUANT_VAL 2
213

    
214
typedef struct Vp3DecodeContext {
215
    AVCodecContext *avctx;
216
    int theora, theora_tables;
217
    int version;
218
    int width, height;
219
    AVFrame golden_frame;
220
    AVFrame last_frame;
221
    AVFrame current_frame;
222
    int keyframe;
223
    DSPContext dsp;
224
    int flipped_image;
225

    
226
    int quality_index;
227
    int last_quality_index;
228

    
229
    int superblock_count;
230
    int superblock_width;
231
    int superblock_height;
232
    int y_superblock_width;
233
    int y_superblock_height;
234
    int c_superblock_width;
235
    int c_superblock_height;
236
    int u_superblock_start;
237
    int v_superblock_start;
238
    unsigned char *superblock_coding;
239

    
240
    int macroblock_count;
241
    int macroblock_width;
242
    int macroblock_height;
243

    
244
    int fragment_count;
245
    int fragment_width;
246
    int fragment_height;
247

    
248
    Vp3Fragment *all_fragments;
249
    int u_fragment_start;
250
    int v_fragment_start;
251
    
252
    /* tables */
253
    uint16_t coded_dc_scale_factor[64];
254
    uint32_t coded_ac_scale_factor[64];
255
    uint16_t coded_intra_y_dequant[64];
256
    uint16_t coded_intra_c_dequant[64];
257
    uint16_t coded_inter_dequant[64];
258

    
259
    /* this is a list of indices into the all_fragments array indicating
260
     * which of the fragments are coded */
261
    int *coded_fragment_list;
262
    int coded_fragment_list_index;
263
    int pixel_addresses_inited;
264

    
265
    VLC dc_vlc[16];
266
    VLC ac_vlc_1[16];
267
    VLC ac_vlc_2[16];
268
    VLC ac_vlc_3[16];
269
    VLC ac_vlc_4[16];
270

    
271
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
272
     * index into them */
273
    int16_t __align16 intra_y_dequant[64];
274
    int16_t __align16 intra_c_dequant[64];
275
    int16_t __align16 inter_dequant[64];
276

    
277
    /* This table contains superblock_count * 16 entries. Each set of 16
278
     * numbers corresponds to the fragment indices 0..15 of the superblock.
279
     * An entry will be -1 to indicate that no entry corresponds to that
280
     * index. */
281
    int *superblock_fragments;
282

    
283
    /* This table contains superblock_count * 4 entries. Each set of 4
284
     * numbers corresponds to the macroblock indices 0..3 of the superblock.
285
     * An entry will be -1 to indicate that no entry corresponds to that
286
     * index. */
287
    int *superblock_macroblocks;
288

    
289
    /* This table contains macroblock_count * 6 entries. Each set of 6
290
     * numbers corresponds to the fragment indices 0..5 which comprise
291
     * the macroblock (4 Y fragments and 2 C fragments). */
292
    int *macroblock_fragments;
293
    /* This is an array that indicates how a particular macroblock 
294
     * is coded. */
295
    unsigned char *macroblock_coding;
296

    
297
    int first_coded_y_fragment;
298
    int first_coded_c_fragment;
299
    int last_coded_y_fragment;
300
    int last_coded_c_fragment;
301

    
302
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
303
    uint8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
304
} Vp3DecodeContext;
305

    
306
static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb);
307
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb);
308

    
309
/************************************************************************
310
 * VP3 specific functions
311
 ************************************************************************/
312

    
313
/*
314
 * This function sets up all of the various blocks mappings:
315
 * superblocks <-> fragments, macroblocks <-> fragments,
316
 * superblocks <-> macroblocks
317
 *
318
 * Returns 0 is successful; returns 1 if *anything* went wrong.
319
 */
320
static int init_block_mapping(Vp3DecodeContext *s) 
321
{
322
    int i, j;
323
    signed int hilbert_walk_y[16];
324
    signed int hilbert_walk_c[16];
325
    signed int hilbert_walk_mb[4];
326

    
327
    int current_fragment = 0;
328
    int current_width = 0;
329
    int current_height = 0;
330
    int right_edge = 0;
331
    int bottom_edge = 0;
332
    int superblock_row_inc = 0;
333
    int *hilbert = NULL;
334
    int mapping_index = 0;
335

    
336
    int current_macroblock;
337
    int c_fragment;
338

    
339
    signed char travel_width[16] = {
340
         1,  1,  0, -1, 
341
         0,  0,  1,  0,
342
         1,  0,  1,  0,
343
         0, -1,  0,  1
344
    };
345

    
346
    signed char travel_height[16] = {
347
         0,  0,  1,  0,
348
         1,  1,  0, -1,
349
         0,  1,  0, -1,
350
        -1,  0, -1,  0
351
    };
352

    
353
    signed char travel_width_mb[4] = {
354
         1,  0,  1,  0
355
    };
356

    
357
    signed char travel_height_mb[4] = {
358
         0,  1,  0, -1
359
    };
360

    
361
    debug_vp3("  vp3: initialize block mapping tables\n");
362

    
363
    /* figure out hilbert pattern per these frame dimensions */
364
    hilbert_walk_y[0]  = 1;
365
    hilbert_walk_y[1]  = 1;
366
    hilbert_walk_y[2]  = s->fragment_width;
367
    hilbert_walk_y[3]  = -1;
368
    hilbert_walk_y[4]  = s->fragment_width;
369
    hilbert_walk_y[5]  = s->fragment_width;
370
    hilbert_walk_y[6]  = 1;
371
    hilbert_walk_y[7]  = -s->fragment_width;
372
    hilbert_walk_y[8]  = 1;
373
    hilbert_walk_y[9]  = s->fragment_width;
374
    hilbert_walk_y[10]  = 1;
375
    hilbert_walk_y[11] = -s->fragment_width;
376
    hilbert_walk_y[12] = -s->fragment_width;
377
    hilbert_walk_y[13] = -1;
378
    hilbert_walk_y[14] = -s->fragment_width;
379
    hilbert_walk_y[15] = 1;
380

    
381
    hilbert_walk_c[0]  = 1;
382
    hilbert_walk_c[1]  = 1;
383
    hilbert_walk_c[2]  = s->fragment_width / 2;
384
    hilbert_walk_c[3]  = -1;
385
    hilbert_walk_c[4]  = s->fragment_width / 2;
386
    hilbert_walk_c[5]  = s->fragment_width / 2;
387
    hilbert_walk_c[6]  = 1;
388
    hilbert_walk_c[7]  = -s->fragment_width / 2;
389
    hilbert_walk_c[8]  = 1;
390
    hilbert_walk_c[9]  = s->fragment_width / 2;
391
    hilbert_walk_c[10]  = 1;
392
    hilbert_walk_c[11] = -s->fragment_width / 2;
393
    hilbert_walk_c[12] = -s->fragment_width / 2;
394
    hilbert_walk_c[13] = -1;
395
    hilbert_walk_c[14] = -s->fragment_width / 2;
396
    hilbert_walk_c[15] = 1;
397

    
398
    hilbert_walk_mb[0] = 1;
399
    hilbert_walk_mb[1] = s->macroblock_width;
400
    hilbert_walk_mb[2] = 1;
401
    hilbert_walk_mb[3] = -s->macroblock_width;
402

    
403
    /* iterate through each superblock (all planes) and map the fragments */
404
    for (i = 0; i < s->superblock_count; i++) {
405
        debug_init("    superblock %d (u starts @ %d, v starts @ %d)\n",
406
            i, s->u_superblock_start, s->v_superblock_start);
407

    
408
        /* time to re-assign the limits? */
409
        if (i == 0) {
410

    
411
            /* start of Y superblocks */
412
            right_edge = s->fragment_width;
413
            bottom_edge = s->fragment_height;
414
            current_width = -1;
415
            current_height = 0;
416
            superblock_row_inc = 3 * s->fragment_width - 
417
                (s->y_superblock_width * 4 - s->fragment_width);
418
            hilbert = hilbert_walk_y;
419

    
420
            /* the first operation for this variable is to advance by 1 */
421
            current_fragment = -1;
422

    
423
        } else if (i == s->u_superblock_start) {
424

    
425
            /* start of U superblocks */
426
            right_edge = s->fragment_width / 2;
427
            bottom_edge = s->fragment_height / 2;
428
            current_width = -1;
429
            current_height = 0;
430
            superblock_row_inc = 3 * (s->fragment_width / 2) - 
431
                (s->c_superblock_width * 4 - s->fragment_width / 2);
432
            hilbert = hilbert_walk_c;
433

    
434
            /* the first operation for this variable is to advance by 1 */
435
            current_fragment = s->u_fragment_start - 1;
436

    
437
        } else if (i == s->v_superblock_start) {
438

    
439
            /* start of V superblocks */
440
            right_edge = s->fragment_width / 2;
441
            bottom_edge = s->fragment_height / 2;
442
            current_width = -1;
443
            current_height = 0;
444
            superblock_row_inc = 3 * (s->fragment_width / 2) - 
445
                (s->c_superblock_width * 4 - s->fragment_width / 2);
446
            hilbert = hilbert_walk_c;
447

    
448
            /* the first operation for this variable is to advance by 1 */
449
            current_fragment = s->v_fragment_start - 1;
450

    
451
        }
452

    
453
        if (current_width >= right_edge - 1) {
454
            /* reset width and move to next superblock row */
455
            current_width = -1;
456
            current_height += 4;
457

    
458
            /* fragment is now at the start of a new superblock row */
459
            current_fragment += superblock_row_inc;
460
        }
461

    
462
        /* iterate through all 16 fragments in a superblock */
463
        for (j = 0; j < 16; j++) {
464
            current_fragment += hilbert[j];
465
            current_width += travel_width[j];
466
            current_height += travel_height[j];
467

    
468
            /* check if the fragment is in bounds */
469
            if ((current_width < right_edge) &&
470
                (current_height < bottom_edge)) {
471
                s->superblock_fragments[mapping_index] = current_fragment;
472
                debug_init("    mapping fragment %d to superblock %d, position %d (%d/%d x %d/%d)\n", 
473
                    s->superblock_fragments[mapping_index], i, j,
474
                    current_width, right_edge, current_height, bottom_edge);
475
            } else {
476
                s->superblock_fragments[mapping_index] = -1;
477
                debug_init("    superblock %d, position %d has no fragment (%d/%d x %d/%d)\n", 
478
                    i, j,
479
                    current_width, right_edge, current_height, bottom_edge);
480
            }
481

    
482
            mapping_index++;
483
        }
484
    }
485

    
486
    /* initialize the superblock <-> macroblock mapping; iterate through
487
     * all of the Y plane superblocks to build this mapping */
488
    right_edge = s->macroblock_width;
489
    bottom_edge = s->macroblock_height;
490
    current_width = -1;
491
    current_height = 0;
492
    superblock_row_inc = s->macroblock_width -
493
        (s->y_superblock_width * 2 - s->macroblock_width);;
494
    hilbert = hilbert_walk_mb;
495
    mapping_index = 0;
496
    current_macroblock = -1;
497
    for (i = 0; i < s->u_superblock_start; i++) {
498

    
499
        if (current_width >= right_edge - 1) {
500
            /* reset width and move to next superblock row */
501
            current_width = -1;
502
            current_height += 2;
503

    
504
            /* macroblock is now at the start of a new superblock row */
505
            current_macroblock += superblock_row_inc;
506
        }
507

    
508
        /* iterate through each potential macroblock in the superblock */
509
        for (j = 0; j < 4; j++) {
510
            current_macroblock += hilbert_walk_mb[j];
511
            current_width += travel_width_mb[j];
512
            current_height += travel_height_mb[j];
513

    
514
            /* check if the macroblock is in bounds */
515
            if ((current_width < right_edge) &&
516
                (current_height < bottom_edge)) {
517
                s->superblock_macroblocks[mapping_index] = current_macroblock;
518
                debug_init("    mapping macroblock %d to superblock %d, position %d (%d/%d x %d/%d)\n",
519
                    s->superblock_macroblocks[mapping_index], i, j,
520
                    current_width, right_edge, current_height, bottom_edge);
521
            } else {
522
                s->superblock_macroblocks[mapping_index] = -1;
523
                debug_init("    superblock %d, position %d has no macroblock (%d/%d x %d/%d)\n",
524
                    i, j,
525
                    current_width, right_edge, current_height, bottom_edge);
526
            }
527

    
528
            mapping_index++;
529
        }
530
    }
531

    
532
    /* initialize the macroblock <-> fragment mapping */
533
    current_fragment = 0;
534
    current_macroblock = 0;
535
    mapping_index = 0;
536
    for (i = 0; i < s->fragment_height; i += 2) {
537

    
538
        for (j = 0; j < s->fragment_width; j += 2) {
539

    
540
            debug_init("    macroblock %d contains fragments: ", current_macroblock);
541
            s->all_fragments[current_fragment].macroblock = current_macroblock;
542
            s->macroblock_fragments[mapping_index++] = current_fragment;
543
            debug_init("%d ", current_fragment);
544

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

    
552
            if (i + 1 < s->fragment_height) {
553
                s->all_fragments[current_fragment + s->fragment_width].macroblock = 
554
                    current_macroblock;
555
                s->macroblock_fragments[mapping_index++] = 
556
                    current_fragment + s->fragment_width;
557
                debug_init("%d ", current_fragment + s->fragment_width);
558
            } else
559
                s->macroblock_fragments[mapping_index++] = -1;
560

    
561
            if ((j + 1 < s->fragment_width) && (i + 1 < s->fragment_height)) {
562
                s->all_fragments[current_fragment + s->fragment_width + 1].macroblock = 
563
                    current_macroblock;
564
                s->macroblock_fragments[mapping_index++] = 
565
                    current_fragment + s->fragment_width + 1;
566
                debug_init("%d ", current_fragment + s->fragment_width + 1);
567
            } else
568
                s->macroblock_fragments[mapping_index++] = -1;
569

    
570
            /* C planes */
571
            c_fragment = s->u_fragment_start + 
572
                (i * s->fragment_width / 4) + (j / 2);
573
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
574
            s->macroblock_fragments[mapping_index++] = c_fragment;
575
            debug_init("%d ", c_fragment);
576

    
577
            c_fragment = s->v_fragment_start + 
578
                (i * s->fragment_width / 4) + (j / 2);
579
            s->all_fragments[c_fragment].macroblock = s->macroblock_count;
580
            s->macroblock_fragments[mapping_index++] = c_fragment;
581
            debug_init("%d ", c_fragment);
582

    
583
            debug_init("\n");
584

    
585
            if (j + 2 <= s->fragment_width)
586
                current_fragment += 2;
587
            else 
588
                current_fragment++;
589
            current_macroblock++;
590
        }
591

    
592
        current_fragment += s->fragment_width;
593
    }
594

    
595
    return 0;  /* successful path out */
596
}
597

    
598
/*
599
 * This function unpacks a single token (which should be in the range 0..31)
600
 * and returns a zero run (number of zero coefficients in current DCT matrix
601
 * before next non-zero coefficient), the next DCT coefficient, and the
602
 * number of consecutive, non-EOB'd DCT blocks to EOB.
603
 */
604
static void unpack_token(GetBitContext *gb, int token, int *zero_run,
605
                         DCTELEM *coeff, int *eob_run) 
606
{
607
    int sign;
608

    
609
    *zero_run = 0;
610
    *eob_run = 0;
611
    *coeff = 0;
612

    
613
    debug_token("    vp3 token %d: ", token);
614
    switch (token) {
615

    
616
    case 0:
617
        debug_token("DCT_EOB_TOKEN, EOB next block\n");
618
        *eob_run = 1;
619
        break;
620

    
621
    case 1:
622
        debug_token("DCT_EOB_PAIR_TOKEN, EOB next 2 blocks\n");
623
        *eob_run = 2;
624
        break;
625

    
626
    case 2:
627
        debug_token("DCT_EOB_TRIPLE_TOKEN, EOB next 3 blocks\n");
628
        *eob_run = 3;
629
        break;
630

    
631
    case 3:
632
        debug_token("DCT_REPEAT_RUN_TOKEN, ");
633
        *eob_run = get_bits(gb, 2) + 4;
634
        debug_token("EOB the next %d blocks\n", *eob_run);
635
        break;
636

    
637
    case 4:
638
        debug_token("DCT_REPEAT_RUN2_TOKEN, ");
639
        *eob_run = get_bits(gb, 3) + 8;
640
        debug_token("EOB the next %d blocks\n", *eob_run);
641
        break;
642

    
643
    case 5:
644
        debug_token("DCT_REPEAT_RUN3_TOKEN, ");
645
        *eob_run = get_bits(gb, 4) + 16;
646
        debug_token("EOB the next %d blocks\n", *eob_run);
647
        break;
648

    
649
    case 6:
650
        debug_token("DCT_REPEAT_RUN4_TOKEN, ");
651
        *eob_run = get_bits(gb, 12);
652
        debug_token("EOB the next %d blocks\n", *eob_run);
653
        break;
654

    
655
    case 7:
656
        debug_token("DCT_SHORT_ZRL_TOKEN, ");
657
        /* note that this token actually indicates that (3 extra bits) + 1 0s
658
         * should be output; this case specifies a run of (3 EBs) 0s and a
659
         * coefficient of 0. */
660
        *zero_run = get_bits(gb, 3);
661
        *coeff = 0;
662
        debug_token("skip the next %d positions in output matrix\n", *zero_run + 1);
663
        break;
664

    
665
    case 8:
666
        debug_token("DCT_ZRL_TOKEN, ");
667
        /* note that this token actually indicates that (6 extra bits) + 1 0s
668
         * should be output; this case specifies a run of (6 EBs) 0s and a
669
         * coefficient of 0. */
670
        *zero_run = get_bits(gb, 6);
671
        *coeff = 0;
672
        debug_token("skip the next %d positions in output matrix\n", *zero_run + 1);
673
        break;
674

    
675
    case 9:
676
        debug_token("ONE_TOKEN, output 1\n");
677
        *coeff = 1;
678
        break;
679

    
680
    case 10:
681
        debug_token("MINUS_ONE_TOKEN, output -1\n");
682
        *coeff = -1;
683
        break;
684

    
685
    case 11:
686
        debug_token("TWO_TOKEN, output 2\n");
687
        *coeff = 2;
688
        break;
689

    
690
    case 12:
691
        debug_token("MINUS_TWO_TOKEN, output -2\n");
692
        *coeff = -2;
693
        break;
694

    
695
    case 13:
696
    case 14:
697
    case 15:
698
    case 16:
699
        debug_token("LOW_VAL_TOKENS, ");
700
        if (get_bits(gb, 1))
701
            *coeff = -(3 + (token - 13));
702
        else
703
            *coeff = 3 + (token - 13);
704
        debug_token("output %d\n", *coeff);
705
        break;
706

    
707
    case 17:
708
        debug_token("DCT_VAL_CATEGORY3, ");
709
        sign = get_bits(gb, 1);
710
        *coeff = 7 + get_bits(gb, 1);
711
        if (sign)
712
            *coeff = -(*coeff);
713
        debug_token("output %d\n", *coeff);
714
        break;
715

    
716
    case 18:
717
        debug_token("DCT_VAL_CATEGORY4, ");
718
        sign = get_bits(gb, 1);
719
        *coeff = 9 + get_bits(gb, 2);
720
        if (sign)
721
            *coeff = -(*coeff);
722
        debug_token("output %d\n", *coeff);
723
        break;
724

    
725
    case 19:
726
        debug_token("DCT_VAL_CATEGORY5, ");
727
        sign = get_bits(gb, 1);
728
        *coeff = 13 + get_bits(gb, 3);
729
        if (sign)
730
            *coeff = -(*coeff);
731
        debug_token("output %d\n", *coeff);
732
        break;
733

    
734
    case 20:
735
        debug_token("DCT_VAL_CATEGORY6, ");
736
        sign = get_bits(gb, 1);
737
        *coeff = 21 + get_bits(gb, 4);
738
        if (sign)
739
            *coeff = -(*coeff);
740
        debug_token("output %d\n", *coeff);
741
        break;
742

    
743
    case 21:
744
        debug_token("DCT_VAL_CATEGORY7, ");
745
        sign = get_bits(gb, 1);
746
        *coeff = 37 + get_bits(gb, 5);
747
        if (sign)
748
            *coeff = -(*coeff);
749
        debug_token("output %d\n", *coeff);
750
        break;
751

    
752
    case 22:
753
        debug_token("DCT_VAL_CATEGORY8, ");
754
        sign = get_bits(gb, 1);
755
        *coeff = 69 + get_bits(gb, 9);
756
        if (sign)
757
            *coeff = -(*coeff);
758
        debug_token("output %d\n", *coeff);
759
        break;
760

    
761
    case 23:
762
    case 24:
763
    case 25:
764
    case 26:
765
    case 27:
766
        debug_token("DCT_RUN_CATEGORY1, ");
767
        *zero_run = token - 22;
768
        if (get_bits(gb, 1))
769
            *coeff = -1;
770
        else
771
            *coeff = 1;
772
        debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
773
        break;
774

    
775
    case 28:
776
        debug_token("DCT_RUN_CATEGORY1B, ");
777
        if (get_bits(gb, 1))
778
            *coeff = -1;
779
        else
780
            *coeff = 1;
781
        *zero_run = 6 + get_bits(gb, 2);
782
        debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
783
        break;
784

    
785
    case 29:
786
        debug_token("DCT_RUN_CATEGORY1C, ");
787
        if (get_bits(gb, 1))
788
            *coeff = -1;
789
        else
790
            *coeff = 1;
791
        *zero_run = 10 + get_bits(gb, 3);
792
        debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
793
        break;
794

    
795
    case 30:
796
        debug_token("DCT_RUN_CATEGORY2, ");
797
        sign = get_bits(gb, 1);
798
        *coeff = 2 + get_bits(gb, 1);
799
        if (sign)
800
            *coeff = -(*coeff);
801
        *zero_run = 1;
802
        debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
803
        break;
804

    
805
    case 31:
806
        debug_token("DCT_RUN_CATEGORY2, ");
807
        sign = get_bits(gb, 1);
808
        *coeff = 2 + get_bits(gb, 1);
809
        if (sign)
810
            *coeff = -(*coeff);
811
        *zero_run = 2 + get_bits(gb, 1);
812
        debug_token("output %d 0s, then %d\n", *zero_run, *coeff);
813
        break;
814

    
815
    default:
816
        av_log(NULL, AV_LOG_ERROR, "  vp3: help! Got a bad token: %d > 31\n", token);
817
        break;
818

    
819
  }
820
}
821

    
822
/*
823
 * This function wipes out all of the fragment data.
824
 */
825
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
826
{
827
    int i;
828

    
829
    /* zero out all of the fragment information */
830
    s->coded_fragment_list_index = 0;
831
    for (i = 0; i < s->fragment_count; i++) {
832
        memset(s->all_fragments[i].coeffs, 0, 64 * sizeof(DCTELEM));
833
        s->all_fragments[i].coeff_count = 0;
834
        s->all_fragments[i].last_coeff = 0;
835
s->all_fragments[i].motion_x = 0xbeef;
836
s->all_fragments[i].motion_y = 0xbeef;
837
    }
838
}
839

    
840
/*
841
 * This function sets of the dequantization tables used for a particular
842
 * frame.
843
 */
844
static void init_dequantizer(Vp3DecodeContext *s)
845
{
846

    
847
    int ac_scale_factor = s->coded_ac_scale_factor[s->quality_index];
848
    int dc_scale_factor = s->coded_dc_scale_factor[s->quality_index];
849
    int i, j;
850

    
851
    debug_vp3("  vp3: initializing dequantization tables\n");
852

    
853
    /* 
854
     * Scale dequantizers:
855
     *
856
     *   quantizer * sf
857
     *   --------------
858
     *        100
859
     *
860
     * where sf = dc_scale_factor for DC quantizer
861
     *         or ac_scale_factor for AC quantizer
862
     *
863
     * Then, saturate the result to a lower limit of MIN_DEQUANT_VAL.
864
     */
865
#define SCALER 4
866

    
867
    /* scale DC quantizers */
868
    s->intra_y_dequant[0] = s->coded_intra_y_dequant[0] * dc_scale_factor / 100;
869
    if (s->intra_y_dequant[0] < MIN_DEQUANT_VAL * 2)
870
        s->intra_y_dequant[0] = MIN_DEQUANT_VAL * 2;
871
    s->intra_y_dequant[0] *= SCALER;
872

    
873
    s->intra_c_dequant[0] = s->coded_intra_c_dequant[0] * dc_scale_factor / 100;
874
    if (s->intra_c_dequant[0] < MIN_DEQUANT_VAL * 2)
875
        s->intra_c_dequant[0] = MIN_DEQUANT_VAL * 2;
876
    s->intra_c_dequant[0] *= SCALER;
877

    
878
    s->inter_dequant[0] = s->coded_inter_dequant[0] * dc_scale_factor / 100;
879
    if (s->inter_dequant[0] < MIN_DEQUANT_VAL * 4)
880
        s->inter_dequant[0] = MIN_DEQUANT_VAL * 4;
881
    s->inter_dequant[0] *= SCALER;
882

    
883
    /* scale AC quantizers, zigzag at the same time in preparation for
884
     * the dequantization phase */
885
    for (i = 1; i < 64; i++) {
886

    
887
        j = zigzag_index[i];
888

    
889
        s->intra_y_dequant[j] = s->coded_intra_y_dequant[i] * ac_scale_factor / 100;
890
        if (s->intra_y_dequant[j] < MIN_DEQUANT_VAL)
891
            s->intra_y_dequant[j] = MIN_DEQUANT_VAL;
892
        s->intra_y_dequant[j] *= SCALER;
893

    
894
        s->intra_c_dequant[j] = s->coded_intra_c_dequant[i] * ac_scale_factor / 100;
895
        if (s->intra_c_dequant[j] < MIN_DEQUANT_VAL)
896
            s->intra_c_dequant[j] = MIN_DEQUANT_VAL;
897
        s->intra_c_dequant[j] *= SCALER;
898

    
899
        s->inter_dequant[j] = s->coded_inter_dequant[i] * ac_scale_factor / 100;
900
        if (s->inter_dequant[j] < MIN_DEQUANT_VAL * 2)
901
            s->inter_dequant[j] = MIN_DEQUANT_VAL * 2;
902
        s->inter_dequant[j] *= SCALER;
903
    }
904
    
905
    memset(s->qscale_table, (FFMAX(s->intra_y_dequant[1], s->intra_c_dequant[1])+8)/16, 512); //FIXME finetune
906

    
907
    /* print debug information as requested */
908
    debug_dequantizers("intra Y dequantizers:\n");
909
    for (i = 0; i < 8; i++) {
910
      for (j = i * 8; j < i * 8 + 8; j++) {
911
        debug_dequantizers(" %4d,", s->intra_y_dequant[j]);
912
      }
913
      debug_dequantizers("\n");
914
    }
915
    debug_dequantizers("\n");
916

    
917
    debug_dequantizers("intra C dequantizers:\n");
918
    for (i = 0; i < 8; i++) {
919
      for (j = i * 8; j < i * 8 + 8; j++) {
920
        debug_dequantizers(" %4d,", s->intra_c_dequant[j]);
921
      }
922
      debug_dequantizers("\n");
923
    }
924
    debug_dequantizers("\n");
925

    
926
    debug_dequantizers("interframe dequantizers:\n");
927
    for (i = 0; i < 8; i++) {
928
      for (j = i * 8; j < i * 8 + 8; j++) {
929
        debug_dequantizers(" %4d,", s->inter_dequant[j]);
930
      }
931
      debug_dequantizers("\n");
932
    }
933
    debug_dequantizers("\n");
934
}
935

    
936
/*
937
 * This function is used to fetch runs of 1s or 0s from the bitstream for
938
 * use in determining which superblocks are fully and partially coded.
939
 *
940
 *  Codeword                RunLength
941
 *  0                       1
942
 *  10x                     2-3
943
 *  110x                    4-5
944
 *  1110xx                  6-9
945
 *  11110xxx                10-17
946
 *  111110xxxx              18-33
947
 *  111111xxxxxxxxxxxx      34-4129
948
 */
949
static int get_superblock_run_length(GetBitContext *gb)
950
{
951

    
952
    if (get_bits(gb, 1) == 0)
953
        return 1;
954

    
955
    else if (get_bits(gb, 1) == 0)
956
        return (2 + get_bits(gb, 1));
957

    
958
    else if (get_bits(gb, 1) == 0)
959
        return (4 + get_bits(gb, 1));
960

    
961
    else if (get_bits(gb, 1) == 0)
962
        return (6 + get_bits(gb, 2));
963

    
964
    else if (get_bits(gb, 1) == 0)
965
        return (10 + get_bits(gb, 3));
966

    
967
    else if (get_bits(gb, 1) == 0)
968
        return (18 + get_bits(gb, 4));
969

    
970
    else
971
        return (34 + get_bits(gb, 12));
972

    
973
}
974

    
975
/*
976
 * This function is used to fetch runs of 1s or 0s from the bitstream for
977
 * use in determining which particular fragments are coded.
978
 *
979
 * Codeword                RunLength
980
 * 0x                      1-2
981
 * 10x                     3-4
982
 * 110x                    5-6
983
 * 1110xx                  7-10
984
 * 11110xx                 11-14
985
 * 11111xxxx               15-30
986
 */
987
static int get_fragment_run_length(GetBitContext *gb)
988
{
989

    
990
    if (get_bits(gb, 1) == 0)
991
        return (1 + get_bits(gb, 1));
992

    
993
    else if (get_bits(gb, 1) == 0)
994
        return (3 + get_bits(gb, 1));
995

    
996
    else if (get_bits(gb, 1) == 0)
997
        return (5 + get_bits(gb, 1));
998

    
999
    else if (get_bits(gb, 1) == 0)
1000
        return (7 + get_bits(gb, 2));
1001

    
1002
    else if (get_bits(gb, 1) == 0)
1003
        return (11 + get_bits(gb, 2));
1004

    
1005
    else
1006
        return (15 + get_bits(gb, 4));
1007

    
1008
}
1009

    
1010
/*
1011
 * This function decodes a VLC from the bitstream and returns a number
1012
 * that ranges from 0..7. The number indicates which of the 8 coding
1013
 * modes to use.
1014
 *
1015
 *  VLC       Number
1016
 *  0            0
1017
 *  10           1
1018
 *  110          2
1019
 *  1110         3
1020
 *  11110        4
1021
 *  111110       5
1022
 *  1111110      6
1023
 *  1111111      7
1024
 *
1025
 */
1026
static int get_mode_code(GetBitContext *gb)
1027
{
1028

    
1029
    if (get_bits(gb, 1) == 0)
1030
        return 0;
1031

    
1032
    else if (get_bits(gb, 1) == 0)
1033
        return 1;
1034

    
1035
    else if (get_bits(gb, 1) == 0)
1036
        return 2;
1037

    
1038
    else if (get_bits(gb, 1) == 0)
1039
        return 3;
1040

    
1041
    else if (get_bits(gb, 1) == 0)
1042
        return 4;
1043

    
1044
    else if (get_bits(gb, 1) == 0)
1045
        return 5;
1046

    
1047
    else if (get_bits(gb, 1) == 0)
1048
        return 6;
1049

    
1050
    else
1051
        return 7;
1052

    
1053
}
1054

    
1055
/*
1056
 * This function extracts a motion vector from the bitstream using a VLC
1057
 * scheme. 3 bits are fetched from the bitstream and 1 of 8 actions is
1058
 * taken depending on the value on those 3 bits:
1059
 *
1060
 *  0: return 0
1061
 *  1: return 1
1062
 *  2: return -1
1063
 *  3: if (next bit is 1) return -2, else return 2
1064
 *  4: if (next bit is 1) return -3, else return 3
1065
 *  5: return 4 + (next 2 bits), next bit is sign
1066
 *  6: return 8 + (next 3 bits), next bit is sign
1067
 *  7: return 16 + (next 4 bits), next bit is sign
1068
 */
1069
static int get_motion_vector_vlc(GetBitContext *gb)
1070
{
1071
    int bits;
1072

    
1073
    bits = get_bits(gb, 3);
1074

    
1075
    switch(bits) {
1076

    
1077
    case 0:
1078
        bits = 0;
1079
        break;
1080

    
1081
    case 1:
1082
        bits = 1;
1083
        break;
1084

    
1085
    case 2:
1086
        bits = -1;
1087
        break;
1088

    
1089
    case 3:
1090
        if (get_bits(gb, 1) == 0)
1091
            bits = 2;
1092
        else
1093
            bits = -2;
1094
        break;
1095

    
1096
    case 4:
1097
        if (get_bits(gb, 1) == 0)
1098
            bits = 3;
1099
        else
1100
            bits = -3;
1101
        break;
1102

    
1103
    case 5:
1104
        bits = 4 + get_bits(gb, 2);
1105
        if (get_bits(gb, 1) == 1)
1106
            bits = -bits;
1107
        break;
1108

    
1109
    case 6:
1110
        bits = 8 + get_bits(gb, 3);
1111
        if (get_bits(gb, 1) == 1)
1112
            bits = -bits;
1113
        break;
1114

    
1115
    case 7:
1116
        bits = 16 + get_bits(gb, 4);
1117
        if (get_bits(gb, 1) == 1)
1118
            bits = -bits;
1119
        break;
1120

    
1121
    }
1122

    
1123
    return bits;
1124
}
1125

    
1126
/*
1127
 * This function fetches a 5-bit number from the stream followed by
1128
 * a sign and calls it a motion vector.
1129
 */
1130
static int get_motion_vector_fixed(GetBitContext *gb)
1131
{
1132

    
1133
    int bits;
1134

    
1135
    bits = get_bits(gb, 5);
1136

    
1137
    if (get_bits(gb, 1) == 1)
1138
        bits = -bits;
1139

    
1140
    return bits;
1141
}
1142

    
1143
/*
1144
 * This function unpacks all of the superblock/macroblock/fragment coding 
1145
 * information from the bitstream.
1146
 */
1147
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
1148
{
1149
    int bit = 0;
1150
    int current_superblock = 0;
1151
    int current_run = 0;
1152
    int decode_fully_flags = 0;
1153
    int decode_partial_blocks = 0;
1154
    int first_c_fragment_seen;
1155

    
1156
    int i, j;
1157
    int current_fragment;
1158

    
1159
    debug_vp3("  vp3: unpacking superblock coding\n");
1160

    
1161
    if (s->keyframe) {
1162

    
1163
        debug_vp3("    keyframe-- all superblocks are fully coded\n");
1164
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
1165

    
1166
    } else {
1167

    
1168
        /* unpack the list of partially-coded superblocks */
1169
        bit = get_bits(gb, 1);
1170
        /* toggle the bit because as soon as the first run length is 
1171
         * fetched the bit will be toggled again */
1172
        bit ^= 1;
1173
        while (current_superblock < s->superblock_count) {
1174
            if (current_run == 0) {
1175
                bit ^= 1;
1176
                current_run = get_superblock_run_length(gb);
1177
                debug_block_coding("      setting superblocks %d..%d to %s\n",
1178
                    current_superblock,
1179
                    current_superblock + current_run - 1,
1180
                    (bit) ? "partially coded" : "not coded");
1181

    
1182
                /* if any of the superblocks are not partially coded, flag
1183
                 * a boolean to decode the list of fully-coded superblocks */
1184
                if (bit == 0) {
1185
                    decode_fully_flags = 1;
1186
                } else {
1187

    
1188
                    /* make a note of the fact that there are partially coded
1189
                     * superblocks */
1190
                    decode_partial_blocks = 1;
1191
                }
1192
            }
1193
            s->superblock_coding[current_superblock++] = 
1194
                (bit) ? SB_PARTIALLY_CODED : SB_NOT_CODED;
1195
            current_run--;
1196
        }
1197

    
1198
        /* unpack the list of fully coded superblocks if any of the blocks were
1199
         * not marked as partially coded in the previous step */
1200
        if (decode_fully_flags) {
1201

    
1202
            current_superblock = 0;
1203
            current_run = 0;
1204
            bit = get_bits(gb, 1);
1205
            /* toggle the bit because as soon as the first run length is 
1206
             * fetched the bit will be toggled again */
1207
            bit ^= 1;
1208
            while (current_superblock < s->superblock_count) {
1209

    
1210
                /* skip any superblocks already marked as partially coded */
1211
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
1212

    
1213
                    if (current_run == 0) {
1214
                        bit ^= 1;
1215
                        current_run = get_superblock_run_length(gb);
1216
                    }
1217

    
1218
                    debug_block_coding("      setting superblock %d to %s\n",
1219
                        current_superblock,
1220
                        (bit) ? "fully coded" : "not coded");
1221
                    s->superblock_coding[current_superblock] = 
1222
                        (bit) ? SB_FULLY_CODED : SB_NOT_CODED;
1223
                    current_run--;
1224
                }
1225
                current_superblock++;
1226
            }
1227
        }
1228

    
1229
        /* if there were partial blocks, initialize bitstream for
1230
         * unpacking fragment codings */
1231
        if (decode_partial_blocks) {
1232

    
1233
            current_run = 0;
1234
            bit = get_bits(gb, 1);
1235
            /* toggle the bit because as soon as the first run length is 
1236
             * fetched the bit will be toggled again */
1237
            bit ^= 1;
1238
        }
1239
    }
1240

    
1241
    /* figure out which fragments are coded; iterate through each
1242
     * superblock (all planes) */
1243
    s->coded_fragment_list_index = 0;
1244
    s->first_coded_y_fragment = s->first_coded_c_fragment = 0;
1245
    s->last_coded_y_fragment = s->last_coded_c_fragment = -1;
1246
    first_c_fragment_seen = 0;
1247
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
1248
    for (i = 0; i < s->superblock_count; i++) {
1249

    
1250
        /* iterate through all 16 fragments in a superblock */
1251
        for (j = 0; j < 16; j++) {
1252

    
1253
            /* if the fragment is in bounds, check its coding status */
1254
            current_fragment = s->superblock_fragments[i * 16 + j];
1255
            if (current_fragment >= s->fragment_count) {
1256
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
1257
                    current_fragment, s->fragment_count);
1258
                return 1;
1259
            }
1260
            if (current_fragment != -1) {
1261
                if (s->superblock_coding[i] == SB_NOT_CODED) {
1262

    
1263
                    /* copy all the fragments from the prior frame */
1264
                    s->all_fragments[current_fragment].coding_method = 
1265
                        MODE_COPY;
1266

    
1267
                } else if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
1268

    
1269
                    /* fragment may or may not be coded; this is the case
1270
                     * that cares about the fragment coding runs */
1271
                    if (current_run == 0) {
1272
                        bit ^= 1;
1273
                        current_run = get_fragment_run_length(gb);
1274
                    }
1275

    
1276
                    if (bit) {
1277
                        /* default mode; actual mode will be decoded in 
1278
                         * the next phase */
1279
                        s->all_fragments[current_fragment].coding_method = 
1280
                            MODE_INTER_NO_MV;
1281
                        s->coded_fragment_list[s->coded_fragment_list_index] = 
1282
                            current_fragment;
1283
                        if ((current_fragment >= s->u_fragment_start) &&
1284
                            (s->last_coded_y_fragment == -1) &&
1285
                            (!first_c_fragment_seen)) {
1286
                            s->first_coded_c_fragment = s->coded_fragment_list_index;
1287
                            s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
1288
                            first_c_fragment_seen = 1;
1289
                        }
1290
                        s->coded_fragment_list_index++;
1291
                        s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
1292
                        debug_block_coding("      superblock %d is partially coded, fragment %d is coded\n",
1293
                            i, current_fragment);
1294
                    } else {
1295
                        /* not coded; copy this fragment from the prior frame */
1296
                        s->all_fragments[current_fragment].coding_method =
1297
                            MODE_COPY;
1298
                        debug_block_coding("      superblock %d is partially coded, fragment %d is not coded\n",
1299
                            i, current_fragment);
1300
                    }
1301

    
1302
                    current_run--;
1303

    
1304
                } else {
1305

    
1306
                    /* fragments are fully coded in this superblock; actual
1307
                     * coding will be determined in next step */
1308
                    s->all_fragments[current_fragment].coding_method = 
1309
                        MODE_INTER_NO_MV;
1310
                    s->coded_fragment_list[s->coded_fragment_list_index] = 
1311
                        current_fragment;
1312
                    if ((current_fragment >= s->u_fragment_start) &&
1313
                        (s->last_coded_y_fragment == -1) &&
1314
                        (!first_c_fragment_seen)) {
1315
                        s->first_coded_c_fragment = s->coded_fragment_list_index;
1316
                        s->last_coded_y_fragment = s->first_coded_c_fragment - 1;
1317
                        first_c_fragment_seen = 1;
1318
                    }
1319
                    s->coded_fragment_list_index++;
1320
                    s->macroblock_coding[s->all_fragments[current_fragment].macroblock] = MODE_INTER_NO_MV;
1321
                    debug_block_coding("      superblock %d is fully coded, fragment %d is coded\n",
1322
                        i, current_fragment);
1323
                }
1324
            }
1325
        }
1326
    }
1327

    
1328
    if (!first_c_fragment_seen)
1329
        /* only Y fragments coded in this frame */
1330
        s->last_coded_y_fragment = s->coded_fragment_list_index - 1;
1331
    else 
1332
        /* end the list of coded C fragments */
1333
        s->last_coded_c_fragment = s->coded_fragment_list_index - 1;
1334

    
1335
    debug_block_coding("    %d total coded fragments, y: %d -> %d, c: %d -> %d\n",
1336
        s->coded_fragment_list_index,
1337
        s->first_coded_y_fragment,
1338
        s->last_coded_y_fragment,
1339
        s->first_coded_c_fragment,
1340
        s->last_coded_c_fragment);
1341

    
1342
    return 0;
1343
}
1344

    
1345
/*
1346
 * This function unpacks all the coding mode data for individual macroblocks
1347
 * from the bitstream.
1348
 */
1349
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
1350
{
1351
    int i, j, k;
1352
    int scheme;
1353
    int current_macroblock;
1354
    int current_fragment;
1355
    int coding_mode;
1356

    
1357
    debug_vp3("  vp3: unpacking encoding modes\n");
1358

    
1359
    if (s->keyframe) {
1360
        debug_vp3("    keyframe-- all blocks are coded as INTRA\n");
1361

    
1362
        for (i = 0; i < s->fragment_count; i++)
1363
            s->all_fragments[i].coding_method = MODE_INTRA;
1364

    
1365
    } else {
1366

    
1367
        /* fetch the mode coding scheme for this frame */
1368
        scheme = get_bits(gb, 3);
1369
        debug_modes("    using mode alphabet %d\n", scheme);
1370

    
1371
        /* is it a custom coding scheme? */
1372
        if (scheme == 0) {
1373
            debug_modes("    custom mode alphabet ahead:\n");
1374
            for (i = 0; i < 8; i++)
1375
                ModeAlphabet[scheme][get_bits(gb, 3)] = i;
1376
        }
1377

    
1378
        for (i = 0; i < 8; i++)
1379
            debug_modes("      mode[%d][%d] = %d\n", scheme, i, 
1380
                ModeAlphabet[scheme][i]);
1381

    
1382
        /* iterate through all of the macroblocks that contain 1 or more
1383
         * coded fragments */
1384
        for (i = 0; i < s->u_superblock_start; i++) {
1385

    
1386
            for (j = 0; j < 4; j++) {
1387
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
1388
                if ((current_macroblock == -1) ||
1389
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
1390
                    continue;
1391
                if (current_macroblock >= s->macroblock_count) {
1392
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad macroblock number (%d >= %d)\n",
1393
                        current_macroblock, s->macroblock_count);
1394
                    return 1;
1395
                }
1396

    
1397
                /* mode 7 means get 3 bits for each coding mode */
1398
                if (scheme == 7)
1399
                    coding_mode = get_bits(gb, 3);
1400
                else
1401
                    coding_mode = ModeAlphabet[scheme][get_mode_code(gb)];
1402

    
1403
                s->macroblock_coding[current_macroblock] = coding_mode;
1404
                for (k = 0; k < 6; k++) {
1405
                    current_fragment = 
1406
                        s->macroblock_fragments[current_macroblock * 6 + k];
1407
                    if (current_fragment == -1)
1408
                        continue;
1409
                    if (current_fragment >= s->fragment_count) {
1410
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_modes(): bad fragment number (%d >= %d)\n",
1411
                            current_fragment, s->fragment_count);
1412
                        return 1;
1413
                    }
1414
                    if (s->all_fragments[current_fragment].coding_method != 
1415
                        MODE_COPY)
1416
                        s->all_fragments[current_fragment].coding_method =
1417
                            coding_mode;
1418
                }
1419

    
1420
                debug_modes("    coding method for macroblock starting @ fragment %d = %d\n",
1421
                    s->macroblock_fragments[current_macroblock * 6], coding_mode);
1422
            }
1423
        }
1424
    }
1425

    
1426
    return 0;
1427
}
1428

    
1429
/*
1430
 * This function unpacks all the motion vectors for the individual
1431
 * macroblocks from the bitstream.
1432
 */
1433
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
1434
{
1435
    int i, j, k;
1436
    int coding_mode;
1437
    int motion_x[6];
1438
    int motion_y[6];
1439
    int last_motion_x = 0;
1440
    int last_motion_y = 0;
1441
    int prior_last_motion_x = 0;
1442
    int prior_last_motion_y = 0;
1443
    int current_macroblock;
1444
    int current_fragment;
1445

    
1446
    debug_vp3("  vp3: unpacking motion vectors\n");
1447
    if (s->keyframe) {
1448

    
1449
        debug_vp3("    keyframe-- there are no motion vectors\n");
1450

    
1451
    } else {
1452

    
1453
        memset(motion_x, 0, 6 * sizeof(int));
1454
        memset(motion_y, 0, 6 * sizeof(int));
1455

    
1456
        /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
1457
        coding_mode = get_bits(gb, 1);
1458
        debug_vectors("    using %s scheme for unpacking motion vectors\n",
1459
            (coding_mode == 0) ? "VLC" : "fixed-length");
1460

    
1461
        /* iterate through all of the macroblocks that contain 1 or more
1462
         * coded fragments */
1463
        for (i = 0; i < s->u_superblock_start; i++) {
1464

    
1465
            for (j = 0; j < 4; j++) {
1466
                current_macroblock = s->superblock_macroblocks[i * 4 + j];
1467
                if ((current_macroblock == -1) ||
1468
                    (s->macroblock_coding[current_macroblock] == MODE_COPY))
1469
                    continue;
1470
                if (current_macroblock >= s->macroblock_count) {
1471
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad macroblock number (%d >= %d)\n",
1472
                        current_macroblock, s->macroblock_count);
1473
                    return 1;
1474
                }
1475

    
1476
                current_fragment = s->macroblock_fragments[current_macroblock * 6];
1477
                if (current_fragment >= s->fragment_count) {
1478
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d\n",
1479
                        current_fragment, s->fragment_count);
1480
                    return 1;
1481
                }
1482
                switch (s->macroblock_coding[current_macroblock]) {
1483

    
1484
                case MODE_INTER_PLUS_MV:
1485
                case MODE_GOLDEN_MV:
1486
                    /* all 6 fragments use the same motion vector */
1487
                    if (coding_mode == 0) {
1488
                        motion_x[0] = get_motion_vector_vlc(gb);
1489
                        motion_y[0] = get_motion_vector_vlc(gb);
1490
                    } else {
1491
                        motion_x[0] = get_motion_vector_fixed(gb);
1492
                        motion_y[0] = get_motion_vector_fixed(gb);
1493
                    }
1494
                    for (k = 1; k < 6; k++) {
1495
                        motion_x[k] = motion_x[0];
1496
                        motion_y[k] = motion_y[0];
1497
                    }
1498

    
1499
                    /* vector maintenance, only on MODE_INTER_PLUS_MV */
1500
                    if (s->macroblock_coding[current_macroblock] ==
1501
                        MODE_INTER_PLUS_MV) {
1502
                        prior_last_motion_x = last_motion_x;
1503
                        prior_last_motion_y = last_motion_y;
1504
                        last_motion_x = motion_x[0];
1505
                        last_motion_y = motion_y[0];
1506
                    }
1507
                    break;
1508

    
1509
                case MODE_INTER_FOURMV:
1510
                    /* fetch 4 vectors from the bitstream, one for each
1511
                     * Y fragment, then average for the C fragment vectors */
1512
                    motion_x[4] = motion_y[4] = 0;
1513
                    for (k = 0; k < 4; k++) {
1514
                        if (coding_mode == 0) {
1515
                            motion_x[k] = get_motion_vector_vlc(gb);
1516
                            motion_y[k] = get_motion_vector_vlc(gb);
1517
                        } else {
1518
                            motion_x[k] = get_motion_vector_fixed(gb);
1519
                            motion_y[k] = get_motion_vector_fixed(gb);
1520
                        }
1521
                        motion_x[4] += motion_x[k];
1522
                        motion_y[4] += motion_y[k];
1523
                    }
1524

    
1525
                    if (motion_x[4] >= 0) 
1526
                        motion_x[4] = (motion_x[4] + 2) / 4;
1527
                    else
1528
                        motion_x[4] = (motion_x[4] - 2) / 4;
1529
                    motion_x[5] = motion_x[4];
1530

    
1531
                    if (motion_y[4] >= 0) 
1532
                        motion_y[4] = (motion_y[4] + 2) / 4;
1533
                    else
1534
                        motion_y[4] = (motion_y[4] - 2) / 4;
1535
                    motion_y[5] = motion_y[4];
1536

    
1537
                    /* vector maintenance; vector[3] is treated as the
1538
                     * last vector in this case */
1539
                    prior_last_motion_x = last_motion_x;
1540
                    prior_last_motion_y = last_motion_y;
1541
                    last_motion_x = motion_x[3];
1542
                    last_motion_y = motion_y[3];
1543
                    break;
1544

    
1545
                case MODE_INTER_LAST_MV:
1546
                    /* all 6 fragments use the last motion vector */
1547
                    motion_x[0] = last_motion_x;
1548
                    motion_y[0] = last_motion_y;
1549
                    for (k = 1; k < 6; k++) {
1550
                        motion_x[k] = motion_x[0];
1551
                        motion_y[k] = motion_y[0];
1552
                    }
1553

    
1554
                    /* no vector maintenance (last vector remains the
1555
                     * last vector) */
1556
                    break;
1557

    
1558
                case MODE_INTER_PRIOR_LAST:
1559
                    /* all 6 fragments use the motion vector prior to the
1560
                     * last motion vector */
1561
                    motion_x[0] = prior_last_motion_x;
1562
                    motion_y[0] = prior_last_motion_y;
1563
                    for (k = 1; k < 6; k++) {
1564
                        motion_x[k] = motion_x[0];
1565
                        motion_y[k] = motion_y[0];
1566
                    }
1567

    
1568
                    /* vector maintenance */
1569
                    prior_last_motion_x = last_motion_x;
1570
                    prior_last_motion_y = last_motion_y;
1571
                    last_motion_x = motion_x[0];
1572
                    last_motion_y = motion_y[0];
1573
                    break;
1574

    
1575
                default:
1576
                    /* covers intra, inter without MV, golden without MV */
1577
                    memset(motion_x, 0, 6 * sizeof(int));
1578
                    memset(motion_y, 0, 6 * sizeof(int));
1579

    
1580
                    /* no vector maintenance */
1581
                    break;
1582
                }
1583

    
1584
                /* assign the motion vectors to the correct fragments */
1585
                debug_vectors("    vectors for macroblock starting @ fragment %d (coding method %d):\n",
1586
                    current_fragment,
1587
                    s->macroblock_coding[current_macroblock]);
1588
                for (k = 0; k < 6; k++) {
1589
                    current_fragment = 
1590
                        s->macroblock_fragments[current_macroblock * 6 + k];
1591
                    if (current_fragment == -1)
1592
                        continue;
1593
                    if (current_fragment >= s->fragment_count) {
1594
                        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vectors(): bad fragment number (%d >= %d)\n",
1595
                            current_fragment, s->fragment_count);
1596
                        return 1;
1597
                    }
1598
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
1599
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
1600
                    debug_vectors("    vector %d: fragment %d = (%d, %d)\n",
1601
                        k, current_fragment, motion_x[k], motion_y[k]);
1602
                }
1603
            }
1604
        }
1605
    }
1606

    
1607
    return 0;
1608
}
1609

    
1610
/* 
1611
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
1612
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
1613
 * data. This function unpacks all the VLCs for either the Y plane or both
1614
 * C planes, and is called for DC coefficients or different AC coefficient
1615
 * levels (since different coefficient types require different VLC tables.
1616
 *
1617
 * This function returns a residual eob run. E.g, if a particular token gave
1618
 * instructions to EOB the next 5 fragments and there were only 2 fragments
1619
 * left in the current fragment range, 3 would be returned so that it could
1620
 * be passed into the next call to this same function.
1621
 */
1622
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
1623
                        VLC *table, int coeff_index,
1624
                        int first_fragment, int last_fragment,
1625
                        int eob_run)
1626
{
1627
    int i;
1628
    int token;
1629
    int zero_run;
1630
    DCTELEM coeff;
1631
    Vp3Fragment *fragment;
1632

    
1633
    if ((first_fragment >= s->fragment_count) ||
1634
        (last_fragment >= s->fragment_count)) {
1635

    
1636
        av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_vlcs(): bad fragment number (%d -> %d ?)\n",
1637
            first_fragment, last_fragment);
1638
        return 0;
1639
    }
1640

    
1641
    for (i = first_fragment; i <= last_fragment; i++) {
1642

    
1643
        fragment = &s->all_fragments[s->coded_fragment_list[i]];
1644
        if (fragment->coeff_count > coeff_index)
1645
            continue;
1646

    
1647
        if (!eob_run) {
1648
            /* decode a VLC into a token */
1649
            token = get_vlc2(gb, table->table, 5, 3);
1650
            debug_vlc(" token = %2d, ", token);
1651
            /* use the token to get a zero run, a coefficient, and an eob run */
1652
            unpack_token(gb, token, &zero_run, &coeff, &eob_run);
1653
        }
1654

    
1655
        if (!eob_run) {
1656
            fragment->coeff_count += zero_run;
1657
            if (fragment->coeff_count < 64)
1658
                fragment->coeffs[fragment->coeff_count++] = coeff;
1659
            debug_vlc(" fragment %d coeff = %d\n",
1660
                s->coded_fragment_list[i], fragment->coeffs[coeff_index]);
1661
        } else {
1662
            fragment->last_coeff = fragment->coeff_count;
1663
            fragment->coeff_count = 64;
1664
            debug_vlc(" fragment %d eob with %d coefficients\n", 
1665
                s->coded_fragment_list[i], fragment->last_coeff);
1666
            eob_run--;
1667
        }
1668
    }
1669

    
1670
    return eob_run;
1671
}
1672

    
1673
/*
1674
 * This function unpacks all of the DCT coefficient data from the
1675
 * bitstream.
1676
 */
1677
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1678
{
1679
    int i;
1680
    int dc_y_table;
1681
    int dc_c_table;
1682
    int ac_y_table;
1683
    int ac_c_table;
1684
    int residual_eob_run = 0;
1685

    
1686
    /* fetch the DC table indices */
1687
    dc_y_table = get_bits(gb, 4);
1688
    dc_c_table = get_bits(gb, 4);
1689

    
1690
    /* unpack the Y plane DC coefficients */
1691
    debug_vp3("  vp3: unpacking Y plane DC coefficients using table %d\n",
1692
        dc_y_table);
1693
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0, 
1694
        s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1695

    
1696
    /* unpack the C plane DC coefficients */
1697
    debug_vp3("  vp3: unpacking C plane DC coefficients using table %d\n",
1698
        dc_c_table);
1699
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1700
        s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1701

    
1702
    /* fetch the AC table indices */
1703
    ac_y_table = get_bits(gb, 4);
1704
    ac_c_table = get_bits(gb, 4);
1705

    
1706
    /* unpack the group 1 AC coefficients (coeffs 1-5) */
1707
    for (i = 1; i <= 5; i++) {
1708

    
1709
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1710
            i, ac_y_table);
1711
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i, 
1712
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1713

    
1714
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1715
            i, ac_c_table);
1716
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i, 
1717
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1718
    }
1719

    
1720
    /* unpack the group 2 AC coefficients (coeffs 6-14) */
1721
    for (i = 6; i <= 14; i++) {
1722

    
1723
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1724
            i, ac_y_table);
1725
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i, 
1726
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1727

    
1728
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1729
            i, ac_c_table);
1730
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i, 
1731
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1732
    }
1733

    
1734
    /* unpack the group 3 AC coefficients (coeffs 15-27) */
1735
    for (i = 15; i <= 27; i++) {
1736

    
1737
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1738
            i, ac_y_table);
1739
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i, 
1740
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1741

    
1742
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1743
            i, ac_c_table);
1744
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i, 
1745
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1746
    }
1747

    
1748
    /* unpack the group 4 AC coefficients (coeffs 28-63) */
1749
    for (i = 28; i <= 63; i++) {
1750

    
1751
        debug_vp3("  vp3: unpacking level %d Y plane AC coefficients using table %d\n",
1752
            i, ac_y_table);
1753
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i, 
1754
            s->first_coded_y_fragment, s->last_coded_y_fragment, residual_eob_run);
1755

    
1756
        debug_vp3("  vp3: unpacking level %d C plane AC coefficients using table %d\n",
1757
            i, ac_c_table);
1758
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i, 
1759
            s->first_coded_c_fragment, s->last_coded_c_fragment, residual_eob_run);
1760
    }
1761

    
1762
    return 0;
1763
}
1764

    
1765
/*
1766
 * This function reverses the DC prediction for each coded fragment in
1767
 * the frame. Much of this function is adapted directly from the original 
1768
 * VP3 source code.
1769
 */
1770
#define COMPATIBLE_FRAME(x) \
1771
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1772
#define FRAME_CODED(x) (s->all_fragments[x].coding_method != MODE_COPY)
1773
static inline int iabs (int x) { return ((x < 0) ? -x : x); }
1774

    
1775
static void reverse_dc_prediction(Vp3DecodeContext *s,
1776
                                  int first_fragment,
1777
                                  int fragment_width,
1778
                                  int fragment_height) 
1779
{
1780

    
1781
#define PUL 8
1782
#define PU 4
1783
#define PUR 2
1784
#define PL 1
1785

    
1786
    int x, y;
1787
    int i = first_fragment;
1788

    
1789
    /*
1790
     * Fragment prediction groups:
1791
     *
1792
     * 32222222226
1793
     * 10000000004
1794
     * 10000000004
1795
     * 10000000004
1796
     * 10000000004
1797
     *
1798
     * Note: Groups 5 and 7 do not exist as it would mean that the 
1799
     * fragment's x coordinate is both 0 and (width - 1) at the same time.
1800
     */
1801
    int predictor_group;
1802
    short predicted_dc;
1803

    
1804
    /* validity flags for the left, up-left, up, and up-right fragments */
1805
    int fl, ful, fu, fur;
1806

    
1807
    /* DC values for the left, up-left, up, and up-right fragments */
1808
    int vl, vul, vu, vur;
1809

    
1810
    /* indices for the left, up-left, up, and up-right fragments */
1811
    int l, ul, u, ur;
1812

    
1813
    /* 
1814
     * The 6 fields mean:
1815
     *   0: up-left multiplier
1816
     *   1: up multiplier
1817
     *   2: up-right multiplier
1818
     *   3: left multiplier
1819
     *   4: mask
1820
     *   5: right bit shift divisor (e.g., 7 means >>=7, a.k.a. div by 128)
1821
     */
1822
    int predictor_transform[16][6] = {
1823
        {  0,  0,  0,  0,   0,  0 },
1824
        {  0,  0,  0,  1,   0,  0 },        // PL
1825
        {  0,  0,  1,  0,   0,  0 },        // PUR
1826
        {  0,  0, 53, 75, 127,  7 },        // PUR|PL
1827
        {  0,  1,  0,  0,   0,  0 },        // PU
1828
        {  0,  1,  0,  1,   1,  1 },        // PU|PL
1829
        {  0,  1,  0,  0,   0,  0 },        // PU|PUR
1830
        {  0,  0, 53, 75, 127,  7 },        // PU|PUR|PL
1831
        {  1,  0,  0,  0,   0,  0 },        // PUL
1832
        {  0,  0,  0,  1,   0,  0 },        // PUL|PL
1833
        {  1,  0,  1,  0,   1,  1 },        // PUL|PUR
1834
        {  0,  0, 53, 75, 127,  7 },        // PUL|PUR|PL
1835
        {  0,  1,  0,  0,   0,  0 },        // PUL|PU
1836
        {-26, 29,  0, 29,  31,  5 },        // PUL|PU|PL
1837
        {  3, 10,  3,  0,  15,  4 },        // PUL|PU|PUR
1838
        {-26, 29,  0, 29,  31,  5 }         // PUL|PU|PUR|PL
1839
    };
1840

    
1841
    /* This table shows which types of blocks can use other blocks for
1842
     * prediction. For example, INTRA is the only mode in this table to
1843
     * have a frame number of 0. That means INTRA blocks can only predict
1844
     * from other INTRA blocks. There are 2 golden frame coding types; 
1845
     * blocks encoding in these modes can only predict from other blocks
1846
     * that were encoded with these 1 of these 2 modes. */
1847
    unsigned char compatible_frame[8] = {
1848
        1,    /* MODE_INTER_NO_MV */
1849
        0,    /* MODE_INTRA */
1850
        1,    /* MODE_INTER_PLUS_MV */
1851
        1,    /* MODE_INTER_LAST_MV */
1852
        1,    /* MODE_INTER_PRIOR_MV */
1853
        2,    /* MODE_USING_GOLDEN */
1854
        2,    /* MODE_GOLDEN_MV */
1855
        1     /* MODE_INTER_FOUR_MV */
1856
    };
1857
    int current_frame_type;
1858

    
1859
    /* there is a last DC predictor for each of the 3 frame types */
1860
    short last_dc[3];
1861

    
1862
    int transform = 0;
1863

    
1864
    debug_vp3("  vp3: reversing DC prediction\n");
1865

    
1866
    vul = vu = vur = vl = 0;
1867
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1868

    
1869
    /* for each fragment row... */
1870
    for (y = 0; y < fragment_height; y++) {
1871

    
1872
        /* for each fragment in a row... */
1873
        for (x = 0; x < fragment_width; x++, i++) {
1874

    
1875
            /* reverse prediction if this block was coded */
1876
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1877

    
1878
                current_frame_type = 
1879
                    compatible_frame[s->all_fragments[i].coding_method];
1880
                predictor_group = (x == 0) + ((y == 0) << 1) +
1881
                    ((x + 1 == fragment_width) << 2);
1882
                debug_dc_pred(" frag %d: group %d, orig DC = %d, ",
1883
                    i, predictor_group, s->all_fragments[i].coeffs[0]);
1884

    
1885
                switch (predictor_group) {
1886

    
1887
                case 0:
1888
                    /* main body of fragments; consider all 4 possible
1889
                     * fragments for prediction */
1890

    
1891
                    /* calculate the indices of the predicting fragments */
1892
                    ul = i - fragment_width - 1;
1893
                    u = i - fragment_width;
1894
                    ur = i - fragment_width + 1;
1895
                    l = i - 1;
1896

    
1897
                    /* fetch the DC values for the predicting fragments */
1898
                    vul = s->all_fragments[ul].coeffs[0];
1899
                    vu = s->all_fragments[u].coeffs[0];
1900
                    vur = s->all_fragments[ur].coeffs[0];
1901
                    vl = s->all_fragments[l].coeffs[0];
1902

    
1903
                    /* figure out which fragments are valid */
1904
                    ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1905
                    fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1906
                    fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1907
                    fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1908

    
1909
                    /* decide which predictor transform to use */
1910
                    transform = (fl*PL) | (fu*PU) | (ful*PUL) | (fur*PUR);
1911

    
1912
                    break;
1913

    
1914
                case 1:
1915
                    /* left column of fragments, not including top corner;
1916
                     * only consider up and up-right fragments */
1917

    
1918
                    /* calculate the indices of the predicting fragments */
1919
                    u = i - fragment_width;
1920
                    ur = i - fragment_width + 1;
1921

    
1922
                    /* fetch the DC values for the predicting fragments */
1923
                    vu = s->all_fragments[u].coeffs[0];
1924
                    vur = s->all_fragments[ur].coeffs[0];
1925

    
1926
                    /* figure out which fragments are valid */
1927
                    fur = FRAME_CODED(ur) && COMPATIBLE_FRAME(ur);
1928
                    fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1929

    
1930
                    /* decide which predictor transform to use */
1931
                    transform = (fu*PU) | (fur*PUR);
1932

    
1933
                    break;
1934

    
1935
                case 2:
1936
                case 6:
1937
                    /* top row of fragments, not including top-left frag;
1938
                     * only consider the left fragment for prediction */
1939

    
1940
                    /* calculate the indices of the predicting fragments */
1941
                    l = i - 1;
1942

    
1943
                    /* fetch the DC values for the predicting fragments */
1944
                    vl = s->all_fragments[l].coeffs[0];
1945

    
1946
                    /* figure out which fragments are valid */
1947
                    fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1948

    
1949
                    /* decide which predictor transform to use */
1950
                    transform = (fl*PL);
1951

    
1952
                    break;
1953

    
1954
                case 3:
1955
                    /* top-left fragment */
1956

    
1957
                    /* nothing to predict from in this case */
1958
                    transform = 0;
1959

    
1960
                    break;
1961

    
1962
                case 4:
1963
                    /* right column of fragments, not including top corner;
1964
                     * consider up-left, up, and left fragments for
1965
                     * prediction */
1966

    
1967
                    /* calculate the indices of the predicting fragments */
1968
                    ul = i - fragment_width - 1;
1969
                    u = i - fragment_width;
1970
                    l = i - 1;
1971

    
1972
                    /* fetch the DC values for the predicting fragments */
1973
                    vul = s->all_fragments[ul].coeffs[0];
1974
                    vu = s->all_fragments[u].coeffs[0];
1975
                    vl = s->all_fragments[l].coeffs[0];
1976

    
1977
                    /* figure out which fragments are valid */
1978
                    ful = FRAME_CODED(ul) && COMPATIBLE_FRAME(ul);
1979
                    fu = FRAME_CODED(u) && COMPATIBLE_FRAME(u);
1980
                    fl = FRAME_CODED(l) && COMPATIBLE_FRAME(l);
1981

    
1982
                    /* decide which predictor transform to use */
1983
                    transform = (fl*PL) | (fu*PU) | (ful*PUL);
1984

    
1985
                    break;
1986

    
1987
                }
1988

    
1989
                debug_dc_pred("transform = %d, ", transform);
1990

    
1991
                if (transform == 0) {
1992

    
1993
                    /* if there were no fragments to predict from, use last
1994
                     * DC saved */
1995
                    s->all_fragments[i].coeffs[0] += last_dc[current_frame_type];
1996
                    debug_dc_pred("from last DC (%d) = %d\n", 
1997
                        current_frame_type, s->all_fragments[i].coeffs[0]);
1998

    
1999
                } else {
2000

    
2001
                    /* apply the appropriate predictor transform */
2002
                    predicted_dc =
2003
                        (predictor_transform[transform][0] * vul) +
2004
                        (predictor_transform[transform][1] * vu) +
2005
                        (predictor_transform[transform][2] * vur) +
2006
                        (predictor_transform[transform][3] * vl);
2007

    
2008
                    /* if there is a shift value in the transform, add
2009
                     * the sign bit before the shift */
2010
                    if (predictor_transform[transform][5] != 0) {
2011
                        predicted_dc += ((predicted_dc >> 15) & 
2012
                            predictor_transform[transform][4]);
2013
                        predicted_dc >>= predictor_transform[transform][5];
2014
                    }
2015

    
2016
                    /* check for outranging on the [ul u l] and
2017
                     * [ul u ur l] predictors */
2018
                    if ((transform == 13) || (transform == 15)) {
2019
                        if (iabs(predicted_dc - vu) > 128)
2020
                            predicted_dc = vu;
2021
                        else if (iabs(predicted_dc - vl) > 128)
2022
                            predicted_dc = vl;
2023
                        else if (iabs(predicted_dc - vul) > 128)
2024
                            predicted_dc = vul;
2025
                    }
2026

    
2027
                    /* at long last, apply the predictor */
2028
                    s->all_fragments[i].coeffs[0] += predicted_dc;
2029
                    debug_dc_pred("from pred DC = %d\n", 
2030
                    s->all_fragments[i].coeffs[0]);
2031
                }
2032

    
2033
                /* save the DC */
2034
                last_dc[current_frame_type] = s->all_fragments[i].coeffs[0];
2035
            }
2036
        }
2037
    }
2038
}
2039

    
2040
/*
2041
 * This function performs the final rendering of each fragment's data
2042
 * onto the output frame.
2043
 */
2044
static void render_fragments(Vp3DecodeContext *s,
2045
                             int first_fragment,
2046
                             int width,
2047
                             int height,
2048
                             int plane /* 0 = Y, 1 = U, 2 = V */) 
2049
{
2050
    int x, y;
2051
    int m, n;
2052
    int i = first_fragment;
2053
    int16_t *dequantizer;
2054
    DCTELEM __align16 output_samples[64];
2055
    unsigned char *output_plane;
2056
    unsigned char *last_plane;
2057
    unsigned char *golden_plane;
2058
    int stride;
2059
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
2060
    int upper_motion_limit, lower_motion_limit;
2061
    int motion_halfpel_index;
2062
    uint8_t *motion_source;
2063

    
2064
    debug_vp3("  vp3: rendering final fragments for %s\n",
2065
        (plane == 0) ? "Y plane" : (plane == 1) ? "U plane" : "V plane");
2066

    
2067
    /* set up plane-specific parameters */
2068
    if (plane == 0) {
2069
        dequantizer = s->intra_y_dequant;
2070
        output_plane = s->current_frame.data[0];
2071
        last_plane = s->last_frame.data[0];
2072
        golden_plane = s->golden_frame.data[0];
2073
        stride = s->current_frame.linesize[0];
2074
        if (!s->flipped_image) stride = -stride;
2075
        upper_motion_limit = 7 * s->current_frame.linesize[0];
2076
        lower_motion_limit = height * s->current_frame.linesize[0] + width - 8;
2077
    } else if (plane == 1) {
2078
        dequantizer = s->intra_c_dequant;
2079
        output_plane = s->current_frame.data[1];
2080
        last_plane = s->last_frame.data[1];
2081
        golden_plane = s->golden_frame.data[1];
2082
        stride = s->current_frame.linesize[1];
2083
        if (!s->flipped_image) stride = -stride;
2084
        upper_motion_limit = 7 * s->current_frame.linesize[1];
2085
        lower_motion_limit = height * s->current_frame.linesize[1] + width - 8;
2086
    } else {
2087
        dequantizer = s->intra_c_dequant;
2088
        output_plane = s->current_frame.data[2];
2089
        last_plane = s->last_frame.data[2];
2090
        golden_plane = s->golden_frame.data[2];
2091
        stride = s->current_frame.linesize[2];
2092
        if (!s->flipped_image) stride = -stride;
2093
        upper_motion_limit = 7 * s->current_frame.linesize[2];
2094
        lower_motion_limit = height * s->current_frame.linesize[2] + width - 8;
2095
    }
2096

    
2097
    /* for each fragment row... */
2098
    for (y = 0; y < height; y += 8) {
2099

    
2100
        /* for each fragment in a row... */
2101
        for (x = 0; x < width; x += 8, i++) {
2102

    
2103
            if ((i < 0) || (i >= s->fragment_count)) {
2104
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_fragments(): bad fragment number (%d)\n", i);
2105
                return;
2106
            }
2107

    
2108
            /* transform if this block was coded */
2109
            if ((s->all_fragments[i].coding_method != MODE_COPY) &&
2110
                !((s->avctx->flags & CODEC_FLAG_GRAY) && plane)) {
2111

    
2112
                if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
2113
                    (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
2114
                    motion_source= golden_plane;
2115
                else 
2116
                    motion_source= last_plane;
2117

    
2118
                motion_source += s->all_fragments[i].first_pixel;
2119
                motion_halfpel_index = 0;
2120

    
2121
                /* sort out the motion vector if this fragment is coded
2122
                 * using a motion vector method */
2123
                if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
2124
                    (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
2125
                    int src_x, src_y;
2126
                    motion_x = s->all_fragments[i].motion_x;
2127
                    motion_y = s->all_fragments[i].motion_y;
2128
                    if(plane){
2129
                        motion_x= (motion_x>>1) | (motion_x&1);
2130
                        motion_y= (motion_y>>1) | (motion_y&1);
2131
                    }
2132

    
2133
                    src_x= (motion_x>>1) + x;
2134
                    src_y= (motion_y>>1) + y;
2135
if ((motion_x == 0xbeef) || (motion_y == 0xbeef))
2136
av_log(s->avctx, AV_LOG_ERROR, " help! got beefy vector! (%X, %X)\n", motion_x, motion_y);
2137

    
2138
                    motion_halfpel_index = motion_x & 0x01;
2139
                    motion_source += (motion_x >> 1);
2140

    
2141
//                    motion_y = -motion_y;
2142
                    motion_halfpel_index |= (motion_y & 0x01) << 1;
2143
                    motion_source += ((motion_y >> 1) * stride);
2144

    
2145
                    if(src_x<0 || src_y<0 || src_x + 9 >= width || src_y + 9 >= height){
2146
                        uint8_t *temp= s->edge_emu_buffer;
2147
                        if(stride<0) temp -= 9*stride;
2148
                        else temp += 9*stride;
2149

    
2150
                        ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, width, height);
2151
                        motion_source= temp;
2152
                    }
2153
                }
2154
                
2155

    
2156
                /* first, take care of copying a block from either the
2157
                 * previous or the golden frame */
2158
                if (s->all_fragments[i].coding_method != MODE_INTRA) {
2159
                    //Note, it is possible to implement all MC cases with put_no_rnd_pixels_l2 which would look more like the VP3 source but this would be slower as put_no_rnd_pixels_tab is better optimzed
2160
                    if(motion_halfpel_index != 3){
2161
                        s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
2162
                            output_plane + s->all_fragments[i].first_pixel,
2163
                            motion_source, stride, 8);
2164
                    }else{
2165
                        int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
2166
                        s->dsp.put_no_rnd_pixels_l2[1](
2167
                            output_plane + s->all_fragments[i].first_pixel,
2168
                            motion_source - d, 
2169
                            motion_source + stride + 1 + d, 
2170
                            stride, 8);
2171
                    }
2172
                }
2173

    
2174
                /* dequantize the DCT coefficients */
2175
                debug_idct("fragment %d, coding mode %d, DC = %d, dequant = %d:\n", 
2176
                    i, s->all_fragments[i].coding_method, 
2177
                    s->all_fragments[i].coeffs[0], dequantizer[0]);
2178

    
2179
                /* invert DCT and place (or add) in final output */
2180
                s->dsp.vp3_idct(s->all_fragments[i].coeffs,
2181
                    dequantizer,
2182
                    s->all_fragments[i].coeff_count,
2183
                    output_samples);
2184
                if (s->all_fragments[i].coding_method == MODE_INTRA) {
2185
                    s->dsp.put_signed_pixels_clamped(output_samples,
2186
                        output_plane + s->all_fragments[i].first_pixel,
2187
                        stride);
2188
                } else {
2189
                    s->dsp.add_pixels_clamped(output_samples,
2190
                        output_plane + s->all_fragments[i].first_pixel,
2191
                        stride);
2192
                }
2193

    
2194
                debug_idct("block after idct_%s():\n",
2195
                    (s->all_fragments[i].coding_method == MODE_INTRA)?
2196
                    "put" : "add");
2197
                for (m = 0; m < 8; m++) {
2198
                    for (n = 0; n < 8; n++) {
2199
                        debug_idct(" %3d", *(output_plane + 
2200
                            s->all_fragments[i].first_pixel + (m * stride + n)));
2201
                    }
2202
                    debug_idct("\n");
2203
                }
2204
                debug_idct("\n");
2205

    
2206
            } else {
2207

    
2208
                /* copy directly from the previous frame */
2209
                s->dsp.put_pixels_tab[1][0](
2210
                    output_plane + s->all_fragments[i].first_pixel,
2211
                    last_plane + s->all_fragments[i].first_pixel,
2212
                    stride, 8);
2213

    
2214
            }
2215
        }
2216
    }
2217

    
2218
    emms_c();
2219

    
2220
}
2221

    
2222
/* 
2223
 * This function computes the first pixel addresses for each fragment.
2224
 * This function needs to be invoked after the first frame is allocated
2225
 * so that it has access to the plane strides.
2226
 */
2227
static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s) 
2228
{
2229

    
2230
    int i, x, y;
2231

    
2232
    /* figure out the first pixel addresses for each of the fragments */
2233
    /* Y plane */
2234
    i = 0;
2235
    for (y = s->fragment_height; y > 0; y--) {
2236
        for (x = 0; x < s->fragment_width; x++) {
2237
            s->all_fragments[i++].first_pixel = 
2238
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2239
                    s->golden_frame.linesize[0] +
2240
                    x * FRAGMENT_PIXELS;
2241
            debug_init("  fragment %d, first pixel @ %d\n", 
2242
                i-1, s->all_fragments[i-1].first_pixel);
2243
        }
2244
    }
2245

    
2246
    /* U plane */
2247
    i = s->u_fragment_start;
2248
    for (y = s->fragment_height / 2; y > 0; y--) {
2249
        for (x = 0; x < s->fragment_width / 2; x++) {
2250
            s->all_fragments[i++].first_pixel = 
2251
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2252
                    s->golden_frame.linesize[1] +
2253
                    x * FRAGMENT_PIXELS;
2254
            debug_init("  fragment %d, first pixel @ %d\n", 
2255
                i-1, s->all_fragments[i-1].first_pixel);
2256
        }
2257
    }
2258

    
2259
    /* V plane */
2260
    i = s->v_fragment_start;
2261
    for (y = s->fragment_height / 2; y > 0; y--) {
2262
        for (x = 0; x < s->fragment_width / 2; x++) {
2263
            s->all_fragments[i++].first_pixel = 
2264
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2265
                    s->golden_frame.linesize[2] +
2266
                    x * FRAGMENT_PIXELS;
2267
            debug_init("  fragment %d, first pixel @ %d\n", 
2268
                i-1, s->all_fragments[i-1].first_pixel);
2269
        }
2270
    }
2271
}
2272

    
2273
/* FIXME: this should be merged with the above! */
2274
static void theora_calculate_pixel_addresses(Vp3DecodeContext *s) 
2275
{
2276

    
2277
    int i, x, y;
2278

    
2279
    /* figure out the first pixel addresses for each of the fragments */
2280
    /* Y plane */
2281
    i = 0;
2282
    for (y = 1; y <= s->fragment_height; y++) {
2283
        for (x = 0; x < s->fragment_width; x++) {
2284
            s->all_fragments[i++].first_pixel = 
2285
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
2286
                    s->golden_frame.linesize[0] +
2287
                    x * FRAGMENT_PIXELS;
2288
            debug_init("  fragment %d, first pixel @ %d\n", 
2289
                i-1, s->all_fragments[i-1].first_pixel);
2290
        }
2291
    }
2292

    
2293
    /* U plane */
2294
    i = s->u_fragment_start;
2295
    for (y = 1; y <= s->fragment_height / 2; y++) {
2296
        for (x = 0; x < s->fragment_width / 2; x++) {
2297
            s->all_fragments[i++].first_pixel = 
2298
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
2299
                    s->golden_frame.linesize[1] +
2300
                    x * FRAGMENT_PIXELS;
2301
            debug_init("  fragment %d, first pixel @ %d\n", 
2302
                i-1, s->all_fragments[i-1].first_pixel);
2303
        }
2304
    }
2305

    
2306
    /* V plane */
2307
    i = s->v_fragment_start;
2308
    for (y = 1; y <= s->fragment_height / 2; y++) {
2309
        for (x = 0; x < s->fragment_width / 2; x++) {
2310
            s->all_fragments[i++].first_pixel = 
2311
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
2312
                    s->golden_frame.linesize[2] +
2313
                    x * FRAGMENT_PIXELS;
2314
            debug_init("  fragment %d, first pixel @ %d\n", 
2315
                i-1, s->all_fragments[i-1].first_pixel);
2316
        }
2317
    }
2318
}
2319

    
2320
/*
2321
 * This is the ffmpeg/libavcodec API init function.
2322
 */
2323
static int vp3_decode_init(AVCodecContext *avctx)
2324
{
2325
    Vp3DecodeContext *s = avctx->priv_data;
2326
    int i;
2327
    int c_width;
2328
    int c_height;
2329
    int y_superblock_count;
2330
    int c_superblock_count;
2331

    
2332
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
2333
        s->version = 0;
2334
    else
2335
        s->version = 1;
2336

    
2337
    s->avctx = avctx;
2338
#if 0
2339
    s->width = avctx->width;
2340
    s->height = avctx->height;
2341
#else
2342
    s->width = (avctx->width + 15) & 0xFFFFFFF0;
2343
    s->height = (avctx->height + 15) & 0xFFFFFFF0;
2344
#endif
2345
    avctx->pix_fmt = PIX_FMT_YUV420P;
2346
    avctx->has_b_frames = 0;
2347
    dsputil_init(&s->dsp, avctx);
2348
    s->dsp.vp3_dsp_init();
2349

    
2350
    /* initialize to an impossible value which will force a recalculation
2351
     * in the first frame decode */
2352
    s->quality_index = -1;
2353

    
2354
    s->y_superblock_width = (s->width + 31) / 32;
2355
    s->y_superblock_height = (s->height + 31) / 32;
2356
    y_superblock_count = s->y_superblock_width * s->y_superblock_height;
2357

    
2358
    /* work out the dimensions for the C planes */
2359
    c_width = s->width / 2;
2360
    c_height = s->height / 2;
2361
    s->c_superblock_width = (c_width + 31) / 32;
2362
    s->c_superblock_height = (c_height + 31) / 32;
2363
    c_superblock_count = s->c_superblock_width * s->c_superblock_height;
2364

    
2365
    s->superblock_count = y_superblock_count + (c_superblock_count * 2);
2366
    s->u_superblock_start = y_superblock_count;
2367
    s->v_superblock_start = s->u_superblock_start + c_superblock_count;
2368
    s->superblock_coding = av_malloc(s->superblock_count);
2369

    
2370
    s->macroblock_width = (s->width + 15) / 16;
2371
    s->macroblock_height = (s->height + 15) / 16;
2372
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
2373

    
2374
    s->fragment_width = s->width / FRAGMENT_PIXELS;
2375
    s->fragment_height = s->height / FRAGMENT_PIXELS;
2376

    
2377
    /* fragment count covers all 8x8 blocks for all 3 planes */
2378
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
2379
    s->u_fragment_start = s->fragment_width * s->fragment_height;
2380
    s->v_fragment_start = s->fragment_width * s->fragment_height * 5 / 4;
2381

    
2382
    debug_init("  Y plane: %d x %d\n", s->width, s->height);
2383
    debug_init("  C plane: %d x %d\n", c_width, c_height);
2384
    debug_init("  Y superblocks: %d x %d, %d total\n",
2385
        s->y_superblock_width, s->y_superblock_height, y_superblock_count);
2386
    debug_init("  C superblocks: %d x %d, %d total\n",
2387
        s->c_superblock_width, s->c_superblock_height, c_superblock_count);
2388
    debug_init("  total superblocks = %d, U starts @ %d, V starts @ %d\n", 
2389
        s->superblock_count, s->u_superblock_start, s->v_superblock_start);
2390
    debug_init("  macroblocks: %d x %d, %d total\n",
2391
        s->macroblock_width, s->macroblock_height, s->macroblock_count);
2392
    debug_init("  %d fragments, %d x %d, u starts @ %d, v starts @ %d\n",
2393
        s->fragment_count,
2394
        s->fragment_width,
2395
        s->fragment_height,
2396
        s->u_fragment_start,
2397
        s->v_fragment_start);
2398

    
2399
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
2400
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
2401
    s->pixel_addresses_inited = 0;
2402

    
2403
    if (!s->theora_tables)
2404
    {
2405
        for (i = 0; i < 64; i++)
2406
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
2407
        for (i = 0; i < 64; i++)
2408
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
2409
        for (i = 0; i < 64; i++)
2410
            s->coded_intra_y_dequant[i] = vp31_intra_y_dequant[i];
2411
        for (i = 0; i < 64; i++)
2412
            s->coded_intra_c_dequant[i] = vp31_intra_c_dequant[i];
2413
        for (i = 0; i < 64; i++)
2414
            s->coded_inter_dequant[i] = vp31_inter_dequant[i];
2415
    }
2416

    
2417
    /* init VLC tables */
2418
    for (i = 0; i < 16; i++) {
2419

    
2420
        /* DC histograms */
2421
        init_vlc(&s->dc_vlc[i], 5, 32,
2422
            &dc_bias[i][0][1], 4, 2,
2423
            &dc_bias[i][0][0], 4, 2);
2424

    
2425
        /* group 1 AC histograms */
2426
        init_vlc(&s->ac_vlc_1[i], 5, 32,
2427
            &ac_bias_0[i][0][1], 4, 2,
2428
            &ac_bias_0[i][0][0], 4, 2);
2429

    
2430
        /* group 2 AC histograms */
2431
        init_vlc(&s->ac_vlc_2[i], 5, 32,
2432
            &ac_bias_1[i][0][1], 4, 2,
2433
            &ac_bias_1[i][0][0], 4, 2);
2434

    
2435
        /* group 3 AC histograms */
2436
        init_vlc(&s->ac_vlc_3[i], 5, 32,
2437
            &ac_bias_2[i][0][1], 4, 2,
2438
            &ac_bias_2[i][0][0], 4, 2);
2439

    
2440
        /* group 4 AC histograms */
2441
        init_vlc(&s->ac_vlc_4[i], 5, 32,
2442
            &ac_bias_3[i][0][1], 4, 2,
2443
            &ac_bias_3[i][0][0], 4, 2);
2444
    }
2445

    
2446
    /* build quantization zigzag table */
2447
    for (i = 0; i < 64; i++)
2448
        zigzag_index[dezigzag_index[i]] = i;
2449

    
2450
    /* work out the block mapping tables */
2451
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
2452
    s->superblock_macroblocks = av_malloc(s->superblock_count * 4 * sizeof(int));
2453
    s->macroblock_fragments = av_malloc(s->macroblock_count * 6 * sizeof(int));
2454
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
2455
    init_block_mapping(s);
2456

    
2457
    for (i = 0; i < 3; i++) {
2458
        s->current_frame.data[i] = NULL;
2459
        s->last_frame.data[i] = NULL;
2460
        s->golden_frame.data[i] = NULL;
2461
    }
2462

    
2463
    return 0;
2464
}
2465

    
2466
/*
2467
 * This is the ffmpeg/libavcodec API frame decode function.
2468
 */
2469
static int vp3_decode_frame(AVCodecContext *avctx, 
2470
                            void *data, int *data_size,
2471
                            uint8_t *buf, int buf_size)
2472
{
2473
    Vp3DecodeContext *s = avctx->priv_data;
2474
    GetBitContext gb;
2475
    static int counter = 0;
2476

    
2477
    init_get_bits(&gb, buf, buf_size * 8);
2478
    
2479
    if (s->theora && get_bits1(&gb))
2480
    {
2481
        int ptype = get_bits(&gb, 7);
2482

    
2483
        skip_bits(&gb, 6*8); /* "theora" */
2484
        
2485
        switch(ptype)
2486
        {
2487
            case 1:
2488
                theora_decode_comments(avctx, gb);
2489
                break;
2490
            case 2:
2491
                theora_decode_tables(avctx, gb);
2492
                    init_dequantizer(s);
2493
                break;
2494
            default:
2495
                av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype);
2496
        }
2497
        return buf_size;
2498
    }
2499

    
2500
    s->keyframe = !get_bits1(&gb);
2501
    if (!s->theora)
2502
        skip_bits(&gb, 1);
2503
    s->last_quality_index = s->quality_index;
2504
    s->quality_index = get_bits(&gb, 6);
2505
    if (s->theora >= 0x030300)
2506
        skip_bits1(&gb);
2507

    
2508
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
2509
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
2510
            s->keyframe?"key":"", counter, s->quality_index);
2511
    counter++;
2512

    
2513
    if (s->quality_index != s->last_quality_index)
2514
        init_dequantizer(s);
2515

    
2516
    if (s->keyframe) {
2517
        if (!s->theora)
2518
        {
2519
            skip_bits(&gb, 4); /* width code */
2520
            skip_bits(&gb, 4); /* height code */
2521
            if (s->version)
2522
            {
2523
                s->version = get_bits(&gb, 5);
2524
                if (counter == 1)
2525
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
2526
            }
2527
        }
2528
        if (s->version || s->theora)
2529
        {
2530
                if (get_bits1(&gb))
2531
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
2532
            skip_bits(&gb, 2); /* reserved? */
2533
        }
2534

    
2535
        if (s->last_frame.data[0] == s->golden_frame.data[0]) {
2536
            if (s->golden_frame.data[0])
2537
                avctx->release_buffer(avctx, &s->golden_frame);
2538
            s->last_frame= s->golden_frame; /* ensure that we catch any access to this released frame */
2539
        } else {
2540
            if (s->golden_frame.data[0])
2541
                avctx->release_buffer(avctx, &s->golden_frame);
2542
            if (s->last_frame.data[0])
2543
                avctx->release_buffer(avctx, &s->last_frame);
2544
        }
2545

    
2546
        s->golden_frame.reference = 3;
2547
        if(avctx->get_buffer(avctx, &s->golden_frame) < 0) {
2548
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2549
            return -1;
2550
        }
2551

    
2552
        /* golden frame is also the current frame */
2553
        memcpy(&s->current_frame, &s->golden_frame, sizeof(AVFrame));
2554

    
2555
        /* time to figure out pixel addresses? */
2556
        if (!s->pixel_addresses_inited)
2557
        {
2558
            if (!s->flipped_image)
2559
                vp3_calculate_pixel_addresses(s);
2560
            else
2561
                theora_calculate_pixel_addresses(s);
2562
        }
2563
    } else {
2564
        /* allocate a new current frame */
2565
        s->current_frame.reference = 3;
2566
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
2567
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");
2568
            return -1;
2569
        }
2570
    }
2571

    
2572
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
2573
    s->current_frame.qstride= 0;
2574

    
2575
    init_frame(s, &gb);
2576

    
2577
#if KEYFRAMES_ONLY
2578
if (!s->keyframe) {
2579

    
2580
    memcpy(s->current_frame.data[0], s->golden_frame.data[0],
2581
        s->current_frame.linesize[0] * s->height);
2582
    memcpy(s->current_frame.data[1], s->golden_frame.data[1],
2583
        s->current_frame.linesize[1] * s->height / 2);
2584
    memcpy(s->current_frame.data[2], s->golden_frame.data[2],
2585
        s->current_frame.linesize[2] * s->height / 2);
2586

    
2587
} else {
2588
#endif
2589

    
2590
    if (unpack_superblocks(s, &gb) ||
2591
        unpack_modes(s, &gb) ||
2592
        unpack_vectors(s, &gb) ||
2593
        unpack_dct_coeffs(s, &gb)) {
2594

    
2595
        av_log(s->avctx, AV_LOG_ERROR, "  vp3: could not decode frame\n");
2596
        return -1;
2597
    }
2598

    
2599
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
2600
    render_fragments(s, 0, s->width, s->height, 0);
2601

    
2602
    if ((avctx->flags & CODEC_FLAG_GRAY) == 0) {
2603
        reverse_dc_prediction(s, s->u_fragment_start,
2604
            s->fragment_width / 2, s->fragment_height / 2);
2605
        reverse_dc_prediction(s, s->v_fragment_start,
2606
            s->fragment_width / 2, s->fragment_height / 2);
2607
        render_fragments(s, s->u_fragment_start, s->width / 2, s->height / 2, 1);
2608
        render_fragments(s, s->v_fragment_start, s->width / 2, s->height / 2, 2);
2609
    } else {
2610
        memset(s->current_frame.data[1], 0x80, s->width * s->height / 4);
2611
        memset(s->current_frame.data[2], 0x80, s->width * s->height / 4);
2612
    }
2613

    
2614
#if KEYFRAMES_ONLY
2615
}
2616
#endif
2617

    
2618
    *data_size=sizeof(AVFrame);
2619
    *(AVFrame*)data= s->current_frame;
2620

    
2621
    /* release the last frame, if it is allocated and if it is not the
2622
     * golden frame */
2623
    if ((s->last_frame.data[0]) &&
2624
        (s->last_frame.data[0] != s->golden_frame.data[0]))
2625
        avctx->release_buffer(avctx, &s->last_frame);
2626

    
2627
    /* shuffle frames (last = current) */
2628
    memcpy(&s->last_frame, &s->current_frame, sizeof(AVFrame));
2629
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
2630

    
2631
    return buf_size;
2632
}
2633

    
2634
/*
2635
 * This is the ffmpeg/libavcodec API module cleanup function.
2636
 */
2637
static int vp3_decode_end(AVCodecContext *avctx)
2638
{
2639
    Vp3DecodeContext *s = avctx->priv_data;
2640

    
2641
    av_free(s->all_fragments);
2642
    av_free(s->coded_fragment_list);
2643
    av_free(s->superblock_fragments);
2644
    av_free(s->superblock_macroblocks);
2645
    av_free(s->macroblock_fragments);
2646
    av_free(s->macroblock_coding);
2647
    
2648
    /* release all frames */
2649
    if (s->golden_frame.data[0] && s->golden_frame.data[0] != s->last_frame.data[0])
2650
        avctx->release_buffer(avctx, &s->golden_frame);
2651
    if (s->last_frame.data[0])
2652
        avctx->release_buffer(avctx, &s->last_frame);
2653
    /* no need to release the current_frame since it will always be pointing
2654
     * to the same frame as either the golden or last frame */
2655

    
2656
    return 0;
2657
}
2658

    
2659
static int theora_decode_header(AVCodecContext *avctx, GetBitContext gb)
2660
{
2661
    Vp3DecodeContext *s = avctx->priv_data;
2662
    int major, minor, micro;
2663

    
2664
    major = get_bits(&gb, 8); /* version major */
2665
    minor = get_bits(&gb, 8); /* version minor */
2666
    micro = get_bits(&gb, 8); /* version micro */
2667
    av_log(avctx, AV_LOG_INFO, "Theora bitstream version %d.%d.%d\n",
2668
        major, minor, micro);
2669

    
2670
    /* FIXME: endianess? */
2671
    s->theora = (major << 16) | (minor << 8) | micro;
2672

    
2673
    /* 3.3.0 aka alpha3 has the same frame orientation as original vp3 */
2674
    /* but previous versions have the image flipped relative to vp3 */
2675
    if (s->theora < 0x030300)
2676
    {
2677
        s->flipped_image = 1;
2678
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2679
    }
2680

    
2681
    s->width = get_bits(&gb, 16) << 4;
2682
    s->height = get_bits(&gb, 16) << 4;
2683
    
2684
    skip_bits(&gb, 24); /* frame width */
2685
    skip_bits(&gb, 24); /* frame height */
2686

    
2687
    skip_bits(&gb, 8); /* offset x */
2688
    skip_bits(&gb, 8); /* offset y */
2689

    
2690
    skip_bits(&gb, 32); /* fps numerator */
2691
    skip_bits(&gb, 32); /* fps denumerator */
2692
    skip_bits(&gb, 24); /* aspect numerator */
2693
    skip_bits(&gb, 24); /* aspect denumerator */
2694
    
2695
    if (s->theora < 0x030300)
2696
        skip_bits(&gb, 5); /* keyframe frequency force */
2697
    skip_bits(&gb, 8); /* colorspace */
2698
    skip_bits(&gb, 24); /* bitrate */
2699

    
2700
    skip_bits(&gb, 6); /* last(?) quality index */
2701
    
2702
    if (s->theora >= 0x030300)
2703
    {
2704
        skip_bits(&gb, 5); /* keyframe frequency force */
2705
        skip_bits(&gb, 5); /* spare bits */
2706
    }
2707
    
2708
//    align_get_bits(&gb);
2709
    
2710
    avctx->width = s->width;
2711
    avctx->height = s->height;
2712

    
2713
    vp3_decode_init(avctx);
2714

    
2715
    return 0;
2716
}
2717

    
2718
static int theora_decode_comments(AVCodecContext *avctx, GetBitContext gb)
2719
{
2720
    int nb_comments, i, tmp;
2721

    
2722
    tmp = get_bits_long(&gb, 32);
2723
    tmp = be2me_32(tmp);
2724
    while(tmp--)
2725
            skip_bits(&gb, 8);
2726

    
2727
    nb_comments = get_bits_long(&gb, 32);
2728
    nb_comments = be2me_32(nb_comments);
2729
    for (i = 0; i < nb_comments; i++)
2730
    {
2731
        tmp = get_bits_long(&gb, 32);
2732
        tmp = be2me_32(tmp);
2733
        while(tmp--)
2734
            skip_bits(&gb, 8);
2735
    }
2736
    
2737
    return 0;
2738
}
2739

    
2740
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext gb)
2741
{
2742
    Vp3DecodeContext *s = avctx->priv_data;
2743
    int i;
2744
    
2745
    /* quality threshold table */
2746
    for (i = 0; i < 64; i++)
2747
        s->coded_ac_scale_factor[i] = get_bits(&gb, 16);
2748

    
2749
    /* dc scale factor table */
2750
    for (i = 0; i < 64; i++)
2751
        s->coded_dc_scale_factor[i] = get_bits(&gb, 16);
2752

    
2753
    /* y coeffs */
2754
    for (i = 0; i < 64; i++)
2755
        s->coded_intra_y_dequant[i] = get_bits(&gb, 8);
2756

    
2757
    /* uv coeffs */
2758
    for (i = 0; i < 64; i++)
2759
        s->coded_intra_c_dequant[i] = get_bits(&gb, 8);
2760

    
2761
    /* inter coeffs */
2762
    for (i = 0; i < 64; i++)
2763
        s->coded_inter_dequant[i] = get_bits(&gb, 8);
2764

    
2765
    /* FIXME: read huffmann tree.. */
2766
    
2767
    s->theora_tables = 1;
2768
    
2769
    return 0;
2770
}
2771

    
2772
static int theora_decode_init(AVCodecContext *avctx)
2773
{
2774
    Vp3DecodeContext *s = avctx->priv_data;
2775
    GetBitContext gb;
2776
    int ptype;
2777
    
2778
    s->theora = 1;
2779

    
2780
    if (!avctx->extradata_size)
2781
        return -1;
2782

    
2783
    init_get_bits(&gb, avctx->extradata, avctx->extradata_size);
2784

    
2785
    ptype = get_bits(&gb, 8);
2786
    debug_vp3("Theora headerpacket type: %x\n", ptype);
2787
            
2788
    if (!(ptype & 0x80))
2789
        return -1;
2790
        
2791
    skip_bits(&gb, 6*8); /* "theora" */
2792
        
2793
    switch(ptype)
2794
    {
2795
        case 0x80:
2796
            theora_decode_header(avctx, gb);
2797
            vp3_decode_init(avctx);
2798
                break;
2799
        case 0x81:
2800
            theora_decode_comments(avctx, gb);
2801
            break;
2802
        case 0x82:
2803
            theora_decode_tables(avctx, gb);
2804
            break;
2805
    }
2806

    
2807
    return 0;
2808
}
2809

    
2810
AVCodec vp3_decoder = {
2811
    "vp3",
2812
    CODEC_TYPE_VIDEO,
2813
    CODEC_ID_VP3,
2814
    sizeof(Vp3DecodeContext),
2815
    vp3_decode_init,
2816
    NULL,
2817
    vp3_decode_end,
2818
    vp3_decode_frame,
2819
    0,
2820
    NULL
2821
};
2822

    
2823
AVCodec theora_decoder = {
2824
    "theora",
2825
    CODEC_TYPE_VIDEO,
2826
    CODEC_ID_THEORA,
2827
    sizeof(Vp3DecodeContext),
2828
    theora_decode_init,
2829
    NULL,
2830
    vp3_decode_end,
2831
    vp3_decode_frame,
2832
    0,
2833
    NULL
2834
};