Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 0c9afacb

History | View | Annotate | Download (76.6 KB)

1
/*
2
 * Copyright (C) 2003-2004 the ffmpeg project
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

    
21
/**
22
 * @file libavcodec/vp3.c
23
 * On2 VP3 Video Decoder
24
 *
25
 * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
26
 * For more information about the VP3 coding process, visit:
27
 *   http://wiki.multimedia.cx/index.php?title=On2_VP3
28
 *
29
 * Theora decoder by Alex Beregszaszi
30
 */
31

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

    
36
#include "avcodec.h"
37
#include "dsputil.h"
38
#include "get_bits.h"
39

    
40
#include "vp3data.h"
41
#include "xiph.h"
42

    
43
#define FRAGMENT_PIXELS 8
44

    
45
static av_cold int vp3_decode_end(AVCodecContext *avctx);
46

    
47
//FIXME split things out into their own arrays
48
typedef struct Vp3Fragment {
49
    int16_t dc;
50
    uint8_t coding_method;
51
    int8_t motion_x;
52
    int8_t motion_y;
53
    uint8_t qpi;
54
} Vp3Fragment;
55

    
56
#define SB_NOT_CODED        0
57
#define SB_PARTIALLY_CODED  1
58
#define SB_FULLY_CODED      2
59

    
60
// This is the maximum length of a single long bit run that can be encoded
61
// for superblock coding or block qps. Theora special-cases this to read a
62
// bit instead of flipping the current bit to allow for runs longer than 4129.
63
#define MAXIMUM_LONG_BIT_RUN 4129
64

    
65
#define MODE_INTER_NO_MV      0
66
#define MODE_INTRA            1
67
#define MODE_INTER_PLUS_MV    2
68
#define MODE_INTER_LAST_MV    3
69
#define MODE_INTER_PRIOR_LAST 4
70
#define MODE_USING_GOLDEN     5
71
#define MODE_GOLDEN_MV        6
72
#define MODE_INTER_FOURMV     7
73
#define CODING_MODE_COUNT     8
74

    
75
/* special internal mode */
76
#define MODE_COPY             8
77

    
78
/* There are 6 preset schemes, plus a free-form scheme */
79
static const int ModeAlphabet[6][CODING_MODE_COUNT] =
80
{
81
    /* scheme 1: Last motion vector dominates */
82
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
83
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
84
         MODE_INTRA,            MODE_USING_GOLDEN,
85
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
86

    
87
    /* scheme 2 */
88
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
89
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
90
         MODE_INTRA,            MODE_USING_GOLDEN,
91
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
92

    
93
    /* scheme 3 */
94
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
95
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
96
         MODE_INTRA,            MODE_USING_GOLDEN,
97
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
98

    
99
    /* scheme 4 */
100
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
101
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
102
         MODE_INTRA,            MODE_USING_GOLDEN,
103
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
104

    
105
    /* scheme 5: No motion vector dominates */
106
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
107
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
108
         MODE_INTRA,            MODE_USING_GOLDEN,
109
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
110

    
111
    /* scheme 6 */
112
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
113
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
114
         MODE_INTER_PLUS_MV,    MODE_INTRA,
115
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
116

    
117
};
118

    
119
static const uint8_t hilbert_offset[16][2] = {
120
    {0,0}, {1,0}, {1,1}, {0,1},
121
    {0,2}, {0,3}, {1,3}, {1,2},
122
    {2,2}, {2,3}, {3,3}, {3,2},
123
    {3,1}, {2,1}, {2,0}, {3,0}
124
};
125

    
126
#define MIN_DEQUANT_VAL 2
127

    
128
typedef struct Vp3DecodeContext {
129
    AVCodecContext *avctx;
130
    int theora, theora_tables;
131
    int version;
132
    int width, height;
133
    AVFrame golden_frame;
134
    AVFrame last_frame;
135
    AVFrame current_frame;
136
    int keyframe;
137
    DSPContext dsp;
138
    int flipped_image;
139
    int last_slice_end;
140

    
141
    int qps[3];
142
    int nqps;
143
    int last_qps[3];
144

    
145
    int superblock_count;
146
    int y_superblock_width;
147
    int y_superblock_height;
148
    int y_superblock_count;
149
    int c_superblock_width;
150
    int c_superblock_height;
151
    int c_superblock_count;
152
    int u_superblock_start;
153
    int v_superblock_start;
154
    unsigned char *superblock_coding;
155

    
156
    int macroblock_count;
157
    int macroblock_width;
158
    int macroblock_height;
159

    
160
    int fragment_count;
161
    int fragment_width;
162
    int fragment_height;
163

    
164
    Vp3Fragment *all_fragments;
165
    int fragment_start[3];
166
    int data_offset[3];
167

    
168
    ScanTable scantable;
169

    
170
    /* tables */
171
    uint16_t coded_dc_scale_factor[64];
172
    uint32_t coded_ac_scale_factor[64];
173
    uint8_t base_matrix[384][64];
174
    uint8_t qr_count[2][3];
175
    uint8_t qr_size [2][3][64];
176
    uint16_t qr_base[2][3][64];
177

    
178
    /**
179
     * This is a list of all tokens in bitstream order. Reordering takes place
180
     * by pulling from each level during IDCT. As a consequence, IDCT must be
181
     * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
182
     * otherwise. The 32 different tokens with up to 12 bits of extradata are
183
     * collapsed into 3 types, packed as follows:
184
     *   (from the low to high bits)
185
     *
186
     * 2 bits: type (0,1,2)
187
     *   0: EOB run, 14 bits for run length (12 needed)
188
     *   1: zero run, 7 bits for run length
189
     *                7 bits for the next coefficient (3 needed)
190
     *   2: coefficient, 14 bits (11 needed)
191
     *
192
     * Coefficients are signed, so are packed in the highest bits for automatic
193
     * sign extension.
194
     */
195
    int16_t *dct_tokens[3][64];
196
    int16_t *dct_tokens_base;
197
#define TOKEN_EOB(eob_run)              ((eob_run) << 2)
198
#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
199
#define TOKEN_COEFF(coeff)              (((coeff) << 2) + 2)
200

    
201
    /**
202
     * number of blocks that contain DCT coefficients at the given level or higher
203
     */
204
    int num_coded_frags[3][64];
205
    int total_num_coded_frags;
206

    
207
    /* this is a list of indexes into the all_fragments array indicating
208
     * which of the fragments are coded */
209
    int *coded_fragment_list[3];
210

    
211
    VLC dc_vlc[16];
212
    VLC ac_vlc_1[16];
213
    VLC ac_vlc_2[16];
214
    VLC ac_vlc_3[16];
215
    VLC ac_vlc_4[16];
216

    
217
    VLC superblock_run_length_vlc;
218
    VLC fragment_run_length_vlc;
219
    VLC mode_code_vlc;
220
    VLC motion_vector_vlc;
221

    
222
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
223
     * index into them */
224
    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
225

    
226
    /* This table contains superblock_count * 16 entries. Each set of 16
227
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
228
     * An entry will be -1 to indicate that no entry corresponds to that
229
     * index. */
230
    int *superblock_fragments;
231

    
232
    /* This is an array that indicates how a particular macroblock
233
     * is coded. */
234
    unsigned char *macroblock_coding;
235

    
236
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
237
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
238

    
239
    /* Huffman decode */
240
    int hti;
241
    unsigned int hbits;
242
    int entries;
243
    int huff_code_size;
244
    uint16_t huffman_table[80][32][2];
245

    
246
    uint8_t filter_limit_values[64];
247
    DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
248
} Vp3DecodeContext;
249

    
250
/************************************************************************
251
 * VP3 specific functions
252
 ************************************************************************/
253

    
254
/*
255
 * This function sets up all of the various blocks mappings:
256
 * superblocks <-> fragments, macroblocks <-> fragments,
257
 * superblocks <-> macroblocks
258
 *
259
 * Returns 0 is successful; returns 1 if *anything* went wrong.
260
 */
261
static int init_block_mapping(Vp3DecodeContext *s)
262
{
263
    int i, j;
264
    signed int hilbert_walk_mb[4];
265

    
266
    int current_fragment = 0;
267
    int current_width = 0;
268
    int current_height = 0;
269
    int right_edge = 0;
270
    int bottom_edge = 0;
271
    int superblock_row_inc = 0;
272
    int mapping_index = 0;
273

    
274
    static const signed char travel_width[16] = {
275
         1,  1,  0, -1,
276
         0,  0,  1,  0,
277
         1,  0,  1,  0,
278
         0, -1,  0,  1
279
    };
280

    
281
    static const signed char travel_height[16] = {
282
         0,  0,  1,  0,
283
         1,  1,  0, -1,
284
         0,  1,  0, -1,
285
        -1,  0, -1,  0
286
    };
287

    
288
    hilbert_walk_mb[0] = 1;
289
    hilbert_walk_mb[1] = s->macroblock_width;
290
    hilbert_walk_mb[2] = 1;
291
    hilbert_walk_mb[3] = -s->macroblock_width;
292

    
293
    /* iterate through each superblock (all planes) and map the fragments */
294
    for (i = 0; i < s->superblock_count; i++) {
295
        /* time to re-assign the limits? */
296
        if (i == 0) {
297

    
298
            /* start of Y superblocks */
299
            right_edge = s->fragment_width;
300
            bottom_edge = s->fragment_height;
301
            current_width = -1;
302
            current_height = 0;
303
            superblock_row_inc = 3 * s->fragment_width -
304
                (s->y_superblock_width * 4 - s->fragment_width);
305

    
306
            /* the first operation for this variable is to advance by 1 */
307
            current_fragment = -1;
308

    
309
        } else if (i == s->u_superblock_start) {
310

    
311
            /* start of U superblocks */
312
            right_edge = s->fragment_width / 2;
313
            bottom_edge = s->fragment_height / 2;
314
            current_width = -1;
315
            current_height = 0;
316
            superblock_row_inc = 3 * (s->fragment_width / 2) -
317
                (s->c_superblock_width * 4 - s->fragment_width / 2);
318

    
319
            /* the first operation for this variable is to advance by 1 */
320
            current_fragment = s->fragment_start[1] - 1;
321

    
322
        } else if (i == s->v_superblock_start) {
323

    
324
            /* start of V superblocks */
325
            right_edge = s->fragment_width / 2;
326
            bottom_edge = s->fragment_height / 2;
327
            current_width = -1;
328
            current_height = 0;
329
            superblock_row_inc = 3 * (s->fragment_width / 2) -
330
                (s->c_superblock_width * 4 - s->fragment_width / 2);
331

    
332
            /* the first operation for this variable is to advance by 1 */
333
            current_fragment = s->fragment_start[2] - 1;
334

    
335
        }
336

    
337
        if (current_width >= right_edge - 1) {
338
            /* reset width and move to next superblock row */
339
            current_width = -1;
340
            current_height += 4;
341

    
342
            /* fragment is now at the start of a new superblock row */
343
            current_fragment += superblock_row_inc;
344
        }
345

    
346
        /* iterate through all 16 fragments in a superblock */
347
        for (j = 0; j < 16; j++) {
348
            current_fragment += travel_width[j] + right_edge * travel_height[j];
349
            current_width += travel_width[j];
350
            current_height += travel_height[j];
351

    
352
            /* check if the fragment is in bounds */
353
            if ((current_width < right_edge) &&
354
                (current_height < bottom_edge)) {
355
                s->superblock_fragments[mapping_index] = current_fragment;
356
            } else {
357
                s->superblock_fragments[mapping_index] = -1;
358
            }
359

    
360
            mapping_index++;
361
        }
362
    }
363

    
364
    return 0;  /* successful path out */
365
}
366

    
367
/*
368
 * This function wipes out all of the fragment data.
369
 */
370
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
371
{
372
    int i;
373

    
374
    /* zero out all of the fragment information */
375
    for (i = 0; i < s->fragment_count; i++) {
376
        s->all_fragments[i].motion_x = 127;
377
        s->all_fragments[i].motion_y = 127;
378
        s->all_fragments[i].dc = 0;
379
        s->all_fragments[i].qpi = 0;
380
    }
381
}
382

    
383
/*
384
 * This function sets up the dequantization tables used for a particular
385
 * frame.
386
 */
387
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
388
{
389
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
390
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
391
    int i, plane, inter, qri, bmi, bmj, qistart;
392

    
393
    for(inter=0; inter<2; inter++){
394
        for(plane=0; plane<3; plane++){
395
            int sum=0;
396
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
397
                sum+= s->qr_size[inter][plane][qri];
398
                if(s->qps[qpi] <= sum)
399
                    break;
400
            }
401
            qistart= sum - s->qr_size[inter][plane][qri];
402
            bmi= s->qr_base[inter][plane][qri  ];
403
            bmj= s->qr_base[inter][plane][qri+1];
404
            for(i=0; i<64; i++){
405
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
406
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
407
                            + s->qr_size[inter][plane][qri])
408
                           / (2*s->qr_size[inter][plane][qri]);
409

    
410
                int qmin= 8<<(inter + !i);
411
                int qscale= i ? ac_scale_factor : dc_scale_factor;
412

    
413
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
414
            }
415
            // all DC coefficients use the same quant so as not to interfere with DC prediction
416
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
417
        }
418
    }
419

    
420
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
421
}
422

    
423
/*
424
 * This function initializes the loop filter boundary limits if the frame's
425
 * quality index is different from the previous frame's.
426
 *
427
 * The filter_limit_values may not be larger than 127.
428
 */
429
static void init_loop_filter(Vp3DecodeContext *s)
430
{
431
    int *bounding_values= s->bounding_values_array+127;
432
    int filter_limit;
433
    int x;
434
    int value;
435

    
436
    filter_limit = s->filter_limit_values[s->qps[0]];
437

    
438
    /* set up the bounding values */
439
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
440
    for (x = 0; x < filter_limit; x++) {
441
        bounding_values[-x] = -x;
442
        bounding_values[x] = x;
443
    }
444
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
445
        bounding_values[ x] =  value;
446
        bounding_values[-x] = -value;
447
    }
448
    if (value)
449
        bounding_values[128] = value;
450
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
451
}
452

    
453
/*
454
 * This function unpacks all of the superblock/macroblock/fragment coding
455
 * information from the bitstream.
456
 */
457
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
458
{
459
    int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
460
    int bit = 0;
461
    int current_superblock = 0;
462
    int current_run = 0;
463
    int num_partial_superblocks = 0;
464

    
465
    int i, j;
466
    int current_fragment;
467
    int plane;
468

    
469
    if (s->keyframe) {
470
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
471

    
472
    } else {
473

    
474
        /* unpack the list of partially-coded superblocks */
475
        bit = get_bits1(gb);
476
        while (current_superblock < s->superblock_count) {
477
                current_run = get_vlc2(gb,
478
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
479
                if (current_run == 34)
480
                    current_run += get_bits(gb, 12);
481

    
482
            if (current_superblock + current_run > s->superblock_count) {
483
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
484
                return -1;
485
            }
486

    
487
            memset(s->superblock_coding + current_superblock, bit, current_run);
488

    
489
            current_superblock += current_run;
490
            if (bit)
491
                num_partial_superblocks += current_run;
492

    
493
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
494
                bit = get_bits1(gb);
495
            else
496
                bit ^= 1;
497
        }
498

    
499
        /* unpack the list of fully coded superblocks if any of the blocks were
500
         * not marked as partially coded in the previous step */
501
        if (num_partial_superblocks < s->superblock_count) {
502
            int superblocks_decoded = 0;
503

    
504
            current_superblock = 0;
505
            bit = get_bits1(gb);
506
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks) {
507
                        current_run = get_vlc2(gb,
508
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
509
                        if (current_run == 34)
510
                            current_run += get_bits(gb, 12);
511

    
512
                for (j = 0; j < current_run; current_superblock++) {
513
                    if (current_superblock >= s->superblock_count) {
514
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
515
                        return -1;
516
                    }
517

    
518
                /* skip any superblocks already marked as partially coded */
519
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
520
                    s->superblock_coding[current_superblock] = 2*bit;
521
                    j++;
522
                }
523
                }
524
                superblocks_decoded += current_run;
525

    
526
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
527
                    bit = get_bits1(gb);
528
                else
529
                    bit ^= 1;
530
            }
531
        }
532

    
533
        /* if there were partial blocks, initialize bitstream for
534
         * unpacking fragment codings */
535
        if (num_partial_superblocks) {
536

    
537
            current_run = 0;
538
            bit = get_bits1(gb);
539
            /* toggle the bit because as soon as the first run length is
540
             * fetched the bit will be toggled again */
541
            bit ^= 1;
542
        }
543
    }
544

    
545
    /* figure out which fragments are coded; iterate through each
546
     * superblock (all planes) */
547
    s->total_num_coded_frags = 0;
548
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
549

    
550
    for (plane = 0; plane < 3; plane++) {
551
        int sb_start = superblock_starts[plane];
552
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
553
        int num_coded_frags = 0;
554

    
555
    for (i = sb_start; i < sb_end; i++) {
556

    
557
        /* iterate through all 16 fragments in a superblock */
558
        for (j = 0; j < 16; j++) {
559

    
560
            /* if the fragment is in bounds, check its coding status */
561
            current_fragment = s->superblock_fragments[i * 16 + j];
562
            if (current_fragment >= s->fragment_count) {
563
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
564
                    current_fragment, s->fragment_count);
565
                return 1;
566
            }
567
            if (current_fragment != -1) {
568
                int coded = s->superblock_coding[i];
569

    
570
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
571

    
572
                    /* fragment may or may not be coded; this is the case
573
                     * that cares about the fragment coding runs */
574
                    if (current_run-- == 0) {
575
                        bit ^= 1;
576
                        current_run = get_vlc2(gb,
577
                            s->fragment_run_length_vlc.table, 5, 2);
578
                    }
579
                    coded = bit;
580
                }
581

    
582
                    if (coded) {
583
                        /* default mode; actual mode will be decoded in
584
                         * the next phase */
585
                        s->all_fragments[current_fragment].coding_method =
586
                            MODE_INTER_NO_MV;
587
                        s->coded_fragment_list[plane][num_coded_frags++] =
588
                            current_fragment;
589
                    } else {
590
                        /* not coded; copy this fragment from the prior frame */
591
                        s->all_fragments[current_fragment].coding_method =
592
                            MODE_COPY;
593
                    }
594
            }
595
        }
596
    }
597
        s->total_num_coded_frags += num_coded_frags;
598
        for (i = 0; i < 64; i++)
599
            s->num_coded_frags[plane][i] = num_coded_frags;
600
        if (plane < 2)
601
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
602
    }
603
    return 0;
604
}
605

    
606
/*
607
 * This function unpacks all the coding mode data for individual macroblocks
608
 * from the bitstream.
609
 */
610
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
611
{
612
    int i, j, k, sb_x, sb_y;
613
    int scheme;
614
    int current_macroblock;
615
    int current_fragment;
616
    int coding_mode;
617
    int custom_mode_alphabet[CODING_MODE_COUNT];
618
    const int *alphabet;
619

    
620
    if (s->keyframe) {
621
        for (i = 0; i < s->fragment_count; i++)
622
            s->all_fragments[i].coding_method = MODE_INTRA;
623

    
624
    } else {
625

    
626
        /* fetch the mode coding scheme for this frame */
627
        scheme = get_bits(gb, 3);
628

    
629
        /* is it a custom coding scheme? */
630
        if (scheme == 0) {
631
            for (i = 0; i < 8; i++)
632
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
633
            for (i = 0; i < 8; i++)
634
                custom_mode_alphabet[get_bits(gb, 3)] = i;
635
            alphabet = custom_mode_alphabet;
636
        } else
637
            alphabet = ModeAlphabet[scheme-1];
638

    
639
        /* iterate through all of the macroblocks that contain 1 or more
640
         * coded fragments */
641
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
642
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
643

    
644
            for (j = 0; j < 4; j++) {
645
                int mb_x = 2*sb_x +   (j>>1);
646
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
647
                current_macroblock = mb_y * s->macroblock_width + mb_x;
648

    
649
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
650
                    continue;
651

    
652
#define BLOCK_X (2*mb_x + (k&1))
653
#define BLOCK_Y (2*mb_y + (k>>1))
654
                /* coding modes are only stored if the macroblock has at least one
655
                 * luma block coded, otherwise it must be INTER_NO_MV */
656
                for (k = 0; k < 4; k++) {
657
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
658
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
659
                        break;
660
                }
661
                if (k == 4) {
662
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
663
                    continue;
664
                }
665

    
666
                /* mode 7 means get 3 bits for each coding mode */
667
                if (scheme == 7)
668
                    coding_mode = get_bits(gb, 3);
669
                else
670
                    coding_mode = alphabet
671
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
672

    
673
                s->macroblock_coding[current_macroblock] = coding_mode;
674
                for (k = 0; k < 4; k++) {
675
                    current_fragment =
676
                        BLOCK_Y*s->fragment_width + BLOCK_X;
677
                    if (s->all_fragments[current_fragment].coding_method !=
678
                        MODE_COPY)
679
                        s->all_fragments[current_fragment].coding_method =
680
                            coding_mode;
681
                }
682
                for (k = 0; k < 2; k++) {
683
                    current_fragment = s->fragment_start[k+1] +
684
                        mb_y*(s->fragment_width>>1) + mb_x;
685
                    if (s->all_fragments[current_fragment].coding_method !=
686
                        MODE_COPY)
687
                        s->all_fragments[current_fragment].coding_method =
688
                            coding_mode;
689
                }
690
            }
691
            }
692
        }
693
    }
694

    
695
    return 0;
696
}
697

    
698
/*
699
 * This function unpacks all the motion vectors for the individual
700
 * macroblocks from the bitstream.
701
 */
702
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
703
{
704
    int j, k, sb_x, sb_y;
705
    int coding_mode;
706
    int motion_x[6];
707
    int motion_y[6];
708
    int last_motion_x = 0;
709
    int last_motion_y = 0;
710
    int prior_last_motion_x = 0;
711
    int prior_last_motion_y = 0;
712
    int current_macroblock;
713
    int current_fragment;
714

    
715
    if (s->keyframe)
716
        return 0;
717

    
718
    memset(motion_x, 0, 6 * sizeof(int));
719
    memset(motion_y, 0, 6 * sizeof(int));
720

    
721
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
722
    coding_mode = get_bits1(gb);
723

    
724
    /* iterate through all of the macroblocks that contain 1 or more
725
     * coded fragments */
726
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
727
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
728

    
729
        for (j = 0; j < 4; j++) {
730
            int mb_x = 2*sb_x +   (j>>1);
731
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
732
            current_macroblock = mb_y * s->macroblock_width + mb_x;
733

    
734
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
735
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
736
                continue;
737

    
738
            switch (s->macroblock_coding[current_macroblock]) {
739

    
740
            case MODE_INTER_PLUS_MV:
741
            case MODE_GOLDEN_MV:
742
                /* all 6 fragments use the same motion vector */
743
                if (coding_mode == 0) {
744
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
745
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
746
                } else {
747
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
748
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
749
                }
750

    
751
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
752
                if (s->macroblock_coding[current_macroblock] ==
753
                    MODE_INTER_PLUS_MV) {
754
                    prior_last_motion_x = last_motion_x;
755
                    prior_last_motion_y = last_motion_y;
756
                    last_motion_x = motion_x[0];
757
                    last_motion_y = motion_y[0];
758
                }
759
                break;
760

    
761
            case MODE_INTER_FOURMV:
762
                /* vector maintenance */
763
                prior_last_motion_x = last_motion_x;
764
                prior_last_motion_y = last_motion_y;
765

    
766
                /* fetch 4 vectors from the bitstream, one for each
767
                 * Y fragment, then average for the C fragment vectors */
768
                motion_x[4] = motion_y[4] = 0;
769
                for (k = 0; k < 4; k++) {
770
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
771
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
772
                        if (coding_mode == 0) {
773
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
774
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
775
                        } else {
776
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
777
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
778
                        }
779
                        last_motion_x = motion_x[k];
780
                        last_motion_y = motion_y[k];
781
                    } else {
782
                        motion_x[k] = 0;
783
                        motion_y[k] = 0;
784
                    }
785
                    motion_x[4] += motion_x[k];
786
                    motion_y[4] += motion_y[k];
787
                }
788

    
789
                motion_x[5]=
790
                motion_x[4]= RSHIFT(motion_x[4], 2);
791
                motion_y[5]=
792
                motion_y[4]= RSHIFT(motion_y[4], 2);
793
                break;
794

    
795
            case MODE_INTER_LAST_MV:
796
                /* all 6 fragments use the last motion vector */
797
                motion_x[0] = last_motion_x;
798
                motion_y[0] = last_motion_y;
799

    
800
                /* no vector maintenance (last vector remains the
801
                 * last vector) */
802
                break;
803

    
804
            case MODE_INTER_PRIOR_LAST:
805
                /* all 6 fragments use the motion vector prior to the
806
                 * last motion vector */
807
                motion_x[0] = prior_last_motion_x;
808
                motion_y[0] = prior_last_motion_y;
809

    
810
                /* vector maintenance */
811
                prior_last_motion_x = last_motion_x;
812
                prior_last_motion_y = last_motion_y;
813
                last_motion_x = motion_x[0];
814
                last_motion_y = motion_y[0];
815
                break;
816

    
817
            default:
818
                /* covers intra, inter without MV, golden without MV */
819
                motion_x[0] = 0;
820
                motion_y[0] = 0;
821

    
822
                /* no vector maintenance */
823
                break;
824
            }
825

    
826
            /* assign the motion vectors to the correct fragments */
827
            for (k = 0; k < 4; k++) {
828
                current_fragment =
829
                    BLOCK_Y*s->fragment_width + BLOCK_X;
830
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
831
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
832
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
833
                } else {
834
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
835
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
836
                }
837
            }
838
            for (k = 0; k < 2; k++) {
839
                current_fragment = s->fragment_start[k+1] +
840
                    mb_y*(s->fragment_width>>1) + mb_x;
841
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
842
                    s->all_fragments[current_fragment].motion_x = motion_x[k+4];
843
                    s->all_fragments[current_fragment].motion_y = motion_y[k+4];
844
                } else {
845
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
846
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
847
                }
848
            }
849
        }
850
        }
851
    }
852

    
853
    return 0;
854
}
855

    
856
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
857
{
858
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
859
    int num_blocks = s->total_num_coded_frags;
860

    
861
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
862
        i = blocks_decoded = num_blocks_at_qpi = 0;
863

    
864
        bit = get_bits1(gb);
865

    
866
        do {
867
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
868
            if (run_length == 34)
869
                run_length += get_bits(gb, 12);
870
            blocks_decoded += run_length;
871

    
872
            if (!bit)
873
                num_blocks_at_qpi += run_length;
874

    
875
            for (j = 0; j < run_length; i++) {
876
                if (i >= s->total_num_coded_frags)
877
                    return -1;
878

    
879
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
880
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
881
                    j++;
882
                }
883
            }
884

    
885
            if (run_length == MAXIMUM_LONG_BIT_RUN)
886
                bit = get_bits1(gb);
887
            else
888
                bit ^= 1;
889
        } while (blocks_decoded < num_blocks);
890

    
891
        num_blocks -= num_blocks_at_qpi;
892
    }
893

    
894
    return 0;
895
}
896

    
897
/*
898
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
899
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
900
 * data. This function unpacks all the VLCs for either the Y plane or both
901
 * C planes, and is called for DC coefficients or different AC coefficient
902
 * levels (since different coefficient types require different VLC tables.
903
 *
904
 * This function returns a residual eob run. E.g, if a particular token gave
905
 * instructions to EOB the next 5 fragments and there were only 2 fragments
906
 * left in the current fragment range, 3 would be returned so that it could
907
 * be passed into the next call to this same function.
908
 */
909
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
910
                        VLC *table, int coeff_index,
911
                        int plane,
912
                        int eob_run)
913
{
914
    int i, j = 0;
915
    int token;
916
    int zero_run = 0;
917
    DCTELEM coeff = 0;
918
    int bits_to_get;
919
    int blocks_ended;
920
    int coeff_i = 0;
921
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
922
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
923

    
924
    /* local references to structure members to avoid repeated deferences */
925
    int *coded_fragment_list = s->coded_fragment_list[plane];
926
    Vp3Fragment *all_fragments = s->all_fragments;
927
    VLC_TYPE (*vlc_table)[2] = table->table;
928

    
929
    if (num_coeffs < 0)
930
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
931

    
932
    if (eob_run > num_coeffs) {
933
        coeff_i = blocks_ended = num_coeffs;
934
        eob_run -= num_coeffs;
935
    } else {
936
        coeff_i = blocks_ended = eob_run;
937
        eob_run = 0;
938
    }
939

    
940
    // insert fake EOB token to cover the split between planes or zzi
941
    if (blocks_ended)
942
        dct_tokens[j++] = blocks_ended << 2;
943

    
944
    while (coeff_i < num_coeffs) {
945
            /* decode a VLC into a token */
946
            token = get_vlc2(gb, vlc_table, 5, 3);
947
            /* use the token to get a zero run, a coefficient, and an eob run */
948
            if (token <= 6) {
949
                eob_run = eob_run_base[token];
950
                if (eob_run_get_bits[token])
951
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
952

    
953
                // record only the number of blocks ended in this plane,
954
                // any spill will be recorded in the next plane.
955
                if (eob_run > num_coeffs - coeff_i) {
956
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
957
                    blocks_ended   += num_coeffs - coeff_i;
958
                    eob_run        -= num_coeffs - coeff_i;
959
                    coeff_i         = num_coeffs;
960
                } else {
961
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
962
                    blocks_ended   += eob_run;
963
                    coeff_i        += eob_run;
964
                    eob_run = 0;
965
                }
966
            } else {
967
                bits_to_get = coeff_get_bits[token];
968
                if (bits_to_get)
969
                    bits_to_get = get_bits(gb, bits_to_get);
970
                coeff = coeff_tables[token][bits_to_get];
971

    
972
                zero_run = zero_run_base[token];
973
                if (zero_run_get_bits[token])
974
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
975

    
976
                if (zero_run) {
977
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
978
                } else {
979
                    // Save DC into the fragment structure. DC prediction is
980
                    // done in raster order, so the actual DC can't be in with
981
                    // other tokens. We still need the token in dct_tokens[]
982
                    // however, or else the structure collapses on itself.
983
                    if (!coeff_index)
984
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
985

    
986
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
987
                }
988

    
989
                if (coeff_index + zero_run > 64) {
990
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
991
                           " %d coeffs left\n", zero_run, 64-coeff_index);
992
                    zero_run = 64 - coeff_index;
993
                }
994

    
995
                // zero runs code multiple coefficients,
996
                // so don't try to decode coeffs for those higher levels
997
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
998
                    s->num_coded_frags[plane][i]--;
999
                coeff_i++;
1000
            }
1001
    }
1002

    
1003
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
1004
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
1005

    
1006
    // decrement the number of blocks that have higher coeffecients for each
1007
    // EOB run at this level
1008
    if (blocks_ended)
1009
        for (i = coeff_index+1; i < 64; i++)
1010
            s->num_coded_frags[plane][i] -= blocks_ended;
1011

    
1012
    // setup the next buffer
1013
    if (plane < 2)
1014
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
1015
    else if (coeff_index < 63)
1016
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
1017

    
1018
    return eob_run;
1019
}
1020

    
1021
static void reverse_dc_prediction(Vp3DecodeContext *s,
1022
                                  int first_fragment,
1023
                                  int fragment_width,
1024
                                  int fragment_height);
1025
/*
1026
 * This function unpacks all of the DCT coefficient data from the
1027
 * bitstream.
1028
 */
1029
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
1030
{
1031
    int i;
1032
    int dc_y_table;
1033
    int dc_c_table;
1034
    int ac_y_table;
1035
    int ac_c_table;
1036
    int residual_eob_run = 0;
1037
    VLC *y_tables[64];
1038
    VLC *c_tables[64];
1039

    
1040
    s->dct_tokens[0][0] = s->dct_tokens_base;
1041

    
1042
    /* fetch the DC table indexes */
1043
    dc_y_table = get_bits(gb, 4);
1044
    dc_c_table = get_bits(gb, 4);
1045

    
1046
    /* unpack the Y plane DC coefficients */
1047
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
1048
        0, residual_eob_run);
1049

    
1050
    /* reverse prediction of the Y-plane DC coefficients */
1051
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
1052

    
1053
    /* unpack the C plane DC coefficients */
1054
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1055
        1, residual_eob_run);
1056
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1057
        2, residual_eob_run);
1058

    
1059
    /* reverse prediction of the C-plane DC coefficients */
1060
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1061
    {
1062
        reverse_dc_prediction(s, s->fragment_start[1],
1063
            s->fragment_width / 2, s->fragment_height / 2);
1064
        reverse_dc_prediction(s, s->fragment_start[2],
1065
            s->fragment_width / 2, s->fragment_height / 2);
1066
    }
1067

    
1068
    /* fetch the AC table indexes */
1069
    ac_y_table = get_bits(gb, 4);
1070
    ac_c_table = get_bits(gb, 4);
1071

    
1072
    /* build tables of AC VLC tables */
1073
    for (i = 1; i <= 5; i++) {
1074
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1075
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1076
    }
1077
    for (i = 6; i <= 14; i++) {
1078
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1079
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1080
    }
1081
    for (i = 15; i <= 27; i++) {
1082
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1083
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1084
    }
1085
    for (i = 28; i <= 63; i++) {
1086
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1087
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1088
    }
1089

    
1090
    /* decode all AC coefficents */
1091
    for (i = 1; i <= 63; i++) {
1092
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1093
                0, residual_eob_run);
1094

    
1095
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1096
                1, residual_eob_run);
1097
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1098
                2, residual_eob_run);
1099
    }
1100

    
1101
    return 0;
1102
}
1103

    
1104
/*
1105
 * This function reverses the DC prediction for each coded fragment in
1106
 * the frame. Much of this function is adapted directly from the original
1107
 * VP3 source code.
1108
 */
1109
#define COMPATIBLE_FRAME(x) \
1110
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1111
#define DC_COEFF(u) s->all_fragments[u].dc
1112

    
1113
static void reverse_dc_prediction(Vp3DecodeContext *s,
1114
                                  int first_fragment,
1115
                                  int fragment_width,
1116
                                  int fragment_height)
1117
{
1118

    
1119
#define PUL 8
1120
#define PU 4
1121
#define PUR 2
1122
#define PL 1
1123

    
1124
    int x, y;
1125
    int i = first_fragment;
1126

    
1127
    int predicted_dc;
1128

    
1129
    /* DC values for the left, up-left, up, and up-right fragments */
1130
    int vl, vul, vu, vur;
1131

    
1132
    /* indexes for the left, up-left, up, and up-right fragments */
1133
    int l, ul, u, ur;
1134

    
1135
    /*
1136
     * The 6 fields mean:
1137
     *   0: up-left multiplier
1138
     *   1: up multiplier
1139
     *   2: up-right multiplier
1140
     *   3: left multiplier
1141
     */
1142
    static const int predictor_transform[16][4] = {
1143
        {  0,  0,  0,  0},
1144
        {  0,  0,  0,128},        // PL
1145
        {  0,  0,128,  0},        // PUR
1146
        {  0,  0, 53, 75},        // PUR|PL
1147
        {  0,128,  0,  0},        // PU
1148
        {  0, 64,  0, 64},        // PU|PL
1149
        {  0,128,  0,  0},        // PU|PUR
1150
        {  0,  0, 53, 75},        // PU|PUR|PL
1151
        {128,  0,  0,  0},        // PUL
1152
        {  0,  0,  0,128},        // PUL|PL
1153
        { 64,  0, 64,  0},        // PUL|PUR
1154
        {  0,  0, 53, 75},        // PUL|PUR|PL
1155
        {  0,128,  0,  0},        // PUL|PU
1156
       {-104,116,  0,116},        // PUL|PU|PL
1157
        { 24, 80, 24,  0},        // PUL|PU|PUR
1158
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1159
    };
1160

    
1161
    /* This table shows which types of blocks can use other blocks for
1162
     * prediction. For example, INTRA is the only mode in this table to
1163
     * have a frame number of 0. That means INTRA blocks can only predict
1164
     * from other INTRA blocks. There are 2 golden frame coding types;
1165
     * blocks encoding in these modes can only predict from other blocks
1166
     * that were encoded with these 1 of these 2 modes. */
1167
    static const unsigned char compatible_frame[9] = {
1168
        1,    /* MODE_INTER_NO_MV */
1169
        0,    /* MODE_INTRA */
1170
        1,    /* MODE_INTER_PLUS_MV */
1171
        1,    /* MODE_INTER_LAST_MV */
1172
        1,    /* MODE_INTER_PRIOR_MV */
1173
        2,    /* MODE_USING_GOLDEN */
1174
        2,    /* MODE_GOLDEN_MV */
1175
        1,    /* MODE_INTER_FOUR_MV */
1176
        3     /* MODE_COPY */
1177
    };
1178
    int current_frame_type;
1179

    
1180
    /* there is a last DC predictor for each of the 3 frame types */
1181
    short last_dc[3];
1182

    
1183
    int transform = 0;
1184

    
1185
    vul = vu = vur = vl = 0;
1186
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1187

    
1188
    /* for each fragment row... */
1189
    for (y = 0; y < fragment_height; y++) {
1190

    
1191
        /* for each fragment in a row... */
1192
        for (x = 0; x < fragment_width; x++, i++) {
1193

    
1194
            /* reverse prediction if this block was coded */
1195
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1196

    
1197
                current_frame_type =
1198
                    compatible_frame[s->all_fragments[i].coding_method];
1199

    
1200
                transform= 0;
1201
                if(x){
1202
                    l= i-1;
1203
                    vl = DC_COEFF(l);
1204
                    if(COMPATIBLE_FRAME(l))
1205
                        transform |= PL;
1206
                }
1207
                if(y){
1208
                    u= i-fragment_width;
1209
                    vu = DC_COEFF(u);
1210
                    if(COMPATIBLE_FRAME(u))
1211
                        transform |= PU;
1212
                    if(x){
1213
                        ul= i-fragment_width-1;
1214
                        vul = DC_COEFF(ul);
1215
                        if(COMPATIBLE_FRAME(ul))
1216
                            transform |= PUL;
1217
                    }
1218
                    if(x + 1 < fragment_width){
1219
                        ur= i-fragment_width+1;
1220
                        vur = DC_COEFF(ur);
1221
                        if(COMPATIBLE_FRAME(ur))
1222
                            transform |= PUR;
1223
                    }
1224
                }
1225

    
1226
                if (transform == 0) {
1227

    
1228
                    /* if there were no fragments to predict from, use last
1229
                     * DC saved */
1230
                    predicted_dc = last_dc[current_frame_type];
1231
                } else {
1232

    
1233
                    /* apply the appropriate predictor transform */
1234
                    predicted_dc =
1235
                        (predictor_transform[transform][0] * vul) +
1236
                        (predictor_transform[transform][1] * vu) +
1237
                        (predictor_transform[transform][2] * vur) +
1238
                        (predictor_transform[transform][3] * vl);
1239

    
1240
                    predicted_dc /= 128;
1241

    
1242
                    /* check for outranging on the [ul u l] and
1243
                     * [ul u ur l] predictors */
1244
                    if ((transform == 15) || (transform == 13)) {
1245
                        if (FFABS(predicted_dc - vu) > 128)
1246
                            predicted_dc = vu;
1247
                        else if (FFABS(predicted_dc - vl) > 128)
1248
                            predicted_dc = vl;
1249
                        else if (FFABS(predicted_dc - vul) > 128)
1250
                            predicted_dc = vul;
1251
                    }
1252
                }
1253

    
1254
                /* at long last, apply the predictor */
1255
                DC_COEFF(i) += predicted_dc;
1256
                /* save the DC */
1257
                last_dc[current_frame_type] = DC_COEFF(i);
1258
            }
1259
        }
1260
    }
1261
}
1262

    
1263
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1264
{
1265
    int x, y;
1266
    int *bounding_values= s->bounding_values_array+127;
1267

    
1268
    int width           = s->fragment_width  >> !!plane;
1269
    int height          = s->fragment_height >> !!plane;
1270
    int fragment        = s->fragment_start        [plane] + ystart * width;
1271
    int stride          = s->current_frame.linesize[plane];
1272
    uint8_t *plane_data = s->current_frame.data    [plane];
1273
    if (!s->flipped_image) stride = -stride;
1274
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1275

    
1276
    for (y = ystart; y < yend; y++) {
1277

    
1278
        for (x = 0; x < width; x++) {
1279
            /* This code basically just deblocks on the edges of coded blocks.
1280
             * However, it has to be much more complicated because of the
1281
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1282
             * because some pixels get filtered twice. */
1283
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1284
            {
1285
                /* do not perform left edge filter for left columns frags */
1286
                if (x > 0) {
1287
                    s->dsp.vp3_h_loop_filter(
1288
                        plane_data + 8*x,
1289
                        stride, bounding_values);
1290
                }
1291

    
1292
                /* do not perform top edge filter for top row fragments */
1293
                if (y > 0) {
1294
                    s->dsp.vp3_v_loop_filter(
1295
                        plane_data + 8*x,
1296
                        stride, bounding_values);
1297
                }
1298

    
1299
                /* do not perform right edge filter for right column
1300
                 * fragments or if right fragment neighbor is also coded
1301
                 * in this frame (it will be filtered in next iteration) */
1302
                if ((x < width - 1) &&
1303
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1304
                    s->dsp.vp3_h_loop_filter(
1305
                        plane_data + 8*x + 8,
1306
                        stride, bounding_values);
1307
                }
1308

    
1309
                /* do not perform bottom edge filter for bottom row
1310
                 * fragments or if bottom fragment neighbor is also coded
1311
                 * in this frame (it will be filtered in the next row) */
1312
                if ((y < height - 1) &&
1313
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1314
                    s->dsp.vp3_v_loop_filter(
1315
                        plane_data + 8*x + 8*stride,
1316
                        stride, bounding_values);
1317
                }
1318
            }
1319

    
1320
            fragment++;
1321
        }
1322
        plane_data += 8*stride;
1323
    }
1324
}
1325

    
1326
/**
1327
 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1328
 * for the next block in coding order
1329
 */
1330
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1331
                              int plane, int inter, DCTELEM block[64])
1332
{
1333
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1334
    uint8_t *perm = s->scantable.permutated;
1335
    int i = 0;
1336

    
1337
    do {
1338
        int token = *s->dct_tokens[plane][i];
1339
        switch (token & 3) {
1340
        case 0: // EOB
1341
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1342
                s->dct_tokens[plane][i]++;
1343
            else
1344
                *s->dct_tokens[plane][i] = token & ~3;
1345
            goto end;
1346
        case 1: // zero run
1347
            s->dct_tokens[plane][i]++;
1348
            i += (token >> 2) & 0x7f;
1349
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1350
            i++;
1351
            break;
1352
        case 2: // coeff
1353
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1354
            s->dct_tokens[plane][i++]++;
1355
            break;
1356
        default:
1357
            av_log(s->avctx, AV_LOG_ERROR, "internal: invalid token type\n");
1358
            return i;
1359
        }
1360
    } while (i < 64);
1361
end:
1362
    // the actual DC+prediction is in the fragment structure
1363
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1364
    return i;
1365
}
1366

    
1367
/**
1368
 * called when all pixels up to row y are complete
1369
 */
1370
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1371
{
1372
    int h, cy;
1373
    int offset[4];
1374

    
1375
    if(s->avctx->draw_horiz_band==NULL)
1376
        return;
1377

    
1378
    h= y - s->last_slice_end;
1379
    y -= h;
1380

    
1381
    if (!s->flipped_image) {
1382
        if (y == 0)
1383
            h -= s->height - s->avctx->height;  // account for non-mod16
1384
        y = s->height - y - h;
1385
    }
1386

    
1387
    cy = y >> 1;
1388
    offset[0] = s->current_frame.linesize[0]*y;
1389
    offset[1] = s->current_frame.linesize[1]*cy;
1390
    offset[2] = s->current_frame.linesize[2]*cy;
1391
    offset[3] = 0;
1392

    
1393
    emms_c();
1394
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1395
    s->last_slice_end= y + h;
1396
}
1397

    
1398
/*
1399
 * Perform the final rendering for a particular slice of data.
1400
 * The slice number ranges from 0..(c_superblock_height - 1).
1401
 */
1402
static void render_slice(Vp3DecodeContext *s, int slice)
1403
{
1404
    int x, y, i, j;
1405
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1406
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1407
    int motion_halfpel_index;
1408
    uint8_t *motion_source;
1409
    int plane, first_pixel;
1410

    
1411
    if (slice >= s->c_superblock_height)
1412
        return;
1413

    
1414
    for (plane = 0; plane < 3; plane++) {
1415
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1416
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1417
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1418
        int stride            = s->current_frame.linesize[plane];
1419
        int plane_width       = s->width  >> !!plane;
1420
        int plane_height      = s->height >> !!plane;
1421

    
1422
        int sb_x, sb_y        = slice << !plane;
1423
        int slice_height      = sb_y + (plane ? 1 : 2);
1424
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1425

    
1426
        int fragment_width    = s->fragment_width  >> !!plane;
1427
        int fragment_height   = s->fragment_height >> !!plane;
1428
        int fragment_start    = s->fragment_start[plane];
1429

    
1430
        if (!s->flipped_image) stride = -stride;
1431
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1432
            continue;
1433

    
1434

    
1435
        if(FFABS(stride) > 2048)
1436
            return; //various tables are fixed size
1437

    
1438
        /* for each superblock row in the slice (both of them)... */
1439
        for (; sb_y < slice_height; sb_y++) {
1440

    
1441
            /* for each superblock in a row... */
1442
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1443

    
1444
                /* for each block in a superblock... */
1445
                for (j = 0; j < 16; j++) {
1446
                    x = 4*sb_x + hilbert_offset[j][0];
1447
                    y = 4*sb_y + hilbert_offset[j][1];
1448

    
1449
                    i = fragment_start + y*fragment_width + x;
1450

    
1451
                    // bounds check
1452
                    if (x >= fragment_width || y >= fragment_height)
1453
                        continue;
1454

    
1455
                first_pixel = 8*y*stride + 8*x;
1456

    
1457
                /* transform if this block was coded */
1458
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1459
                    int intra = s->all_fragments[i].coding_method == MODE_INTRA;
1460

    
1461
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1462
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1463
                        motion_source= golden_plane;
1464
                    else
1465
                        motion_source= last_plane;
1466

    
1467
                    motion_source += first_pixel;
1468
                    motion_halfpel_index = 0;
1469

    
1470
                    /* sort out the motion vector if this fragment is coded
1471
                     * using a motion vector method */
1472
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1473
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1474
                        int src_x, src_y;
1475
                        motion_x = s->all_fragments[i].motion_x;
1476
                        motion_y = s->all_fragments[i].motion_y;
1477
                        if(plane){
1478
                            motion_x= (motion_x>>1) | (motion_x&1);
1479
                            motion_y= (motion_y>>1) | (motion_y&1);
1480
                        }
1481

    
1482
                        src_x= (motion_x>>1) + 8*x;
1483
                        src_y= (motion_y>>1) + 8*y;
1484
                        if ((motion_x == 127) || (motion_y == 127))
1485
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1486

    
1487
                        motion_halfpel_index = motion_x & 0x01;
1488
                        motion_source += (motion_x >> 1);
1489

    
1490
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1491
                        motion_source += ((motion_y >> 1) * stride);
1492

    
1493
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1494
                            uint8_t *temp= s->edge_emu_buffer;
1495
                            if(stride<0) temp -= 9*stride;
1496
                            else temp += 9*stride;
1497

    
1498
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1499
                            motion_source= temp;
1500
                        }
1501
                    }
1502

    
1503

    
1504
                    /* first, take care of copying a block from either the
1505
                     * previous or the golden frame */
1506
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1507
                        /* Note, it is possible to implement all MC cases with
1508
                           put_no_rnd_pixels_l2 which would look more like the
1509
                           VP3 source but this would be slower as
1510
                           put_no_rnd_pixels_tab is better optimzed */
1511
                        if(motion_halfpel_index != 3){
1512
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1513
                                output_plane + first_pixel,
1514
                                motion_source, stride, 8);
1515
                        }else{
1516
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1517
                            s->dsp.put_no_rnd_pixels_l2[1](
1518
                                output_plane + first_pixel,
1519
                                motion_source - d,
1520
                                motion_source + stride + 1 + d,
1521
                                stride, 8);
1522
                        }
1523
                    }
1524

    
1525
                        s->dsp.clear_block(block);
1526
                        vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
1527

    
1528
                    /* invert DCT and place (or add) in final output */
1529

    
1530
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1531
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1532
                            block[0] += 128<<3;
1533
                        s->dsp.idct_put(
1534
                            output_plane + first_pixel,
1535
                            stride,
1536
                            block);
1537
                    } else {
1538
                        s->dsp.idct_add(
1539
                            output_plane + first_pixel,
1540
                            stride,
1541
                            block);
1542
                    }
1543
                } else {
1544

    
1545
                    /* copy directly from the previous frame */
1546
                    s->dsp.put_pixels_tab[1][0](
1547
                        output_plane + first_pixel,
1548
                        last_plane + first_pixel,
1549
                        stride, 8);
1550

    
1551
                }
1552
                }
1553
            }
1554

    
1555
            // Filter up to the last row in the superblock row
1556
            apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1557
        }
1558
    }
1559

    
1560
     /* this looks like a good place for slice dispatch... */
1561
     /* algorithm:
1562
      *   if (slice == s->macroblock_height - 1)
1563
      *     dispatch (both last slice & 2nd-to-last slice);
1564
      *   else if (slice > 0)
1565
      *     dispatch (slice - 1);
1566
      */
1567

    
1568
    vp3_draw_horiz_band(s, 64*slice + 64-16);
1569
}
1570

    
1571
/*
1572
 * This is the ffmpeg/libavcodec API init function.
1573
 */
1574
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1575
{
1576
    Vp3DecodeContext *s = avctx->priv_data;
1577
    int i, inter, plane;
1578
    int c_width;
1579
    int c_height;
1580

    
1581
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1582
        s->version = 0;
1583
    else
1584
        s->version = 1;
1585

    
1586
    s->avctx = avctx;
1587
    s->width = FFALIGN(avctx->width, 16);
1588
    s->height = FFALIGN(avctx->height, 16);
1589
    avctx->pix_fmt = PIX_FMT_YUV420P;
1590
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1591
    if(avctx->idct_algo==FF_IDCT_AUTO)
1592
        avctx->idct_algo=FF_IDCT_VP3;
1593
    dsputil_init(&s->dsp, avctx);
1594

    
1595
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1596

    
1597
    /* initialize to an impossible value which will force a recalculation
1598
     * in the first frame decode */
1599
    for (i = 0; i < 3; i++)
1600
        s->qps[i] = -1;
1601

    
1602
    s->y_superblock_width = (s->width + 31) / 32;
1603
    s->y_superblock_height = (s->height + 31) / 32;
1604
    s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1605

    
1606
    /* work out the dimensions for the C planes */
1607
    c_width = s->width / 2;
1608
    c_height = s->height / 2;
1609
    s->c_superblock_width = (c_width + 31) / 32;
1610
    s->c_superblock_height = (c_height + 31) / 32;
1611
    s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1612

    
1613
    s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1614
    s->u_superblock_start = s->y_superblock_count;
1615
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1616
    s->superblock_coding = av_malloc(s->superblock_count);
1617

    
1618
    s->macroblock_width = (s->width + 15) / 16;
1619
    s->macroblock_height = (s->height + 15) / 16;
1620
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1621

    
1622
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1623
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1624

    
1625
    /* fragment count covers all 8x8 blocks for all 3 planes */
1626
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1627
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1628
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1629

    
1630
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1631
    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1632
    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1633
    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1634
        !s->coded_fragment_list[0]) {
1635
        vp3_decode_end(avctx);
1636
        return -1;
1637
    }
1638

    
1639
    if (!s->theora_tables)
1640
    {
1641
        for (i = 0; i < 64; i++) {
1642
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1643
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1644
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1645
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1646
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1647
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1648
        }
1649

    
1650
        for(inter=0; inter<2; inter++){
1651
            for(plane=0; plane<3; plane++){
1652
                s->qr_count[inter][plane]= 1;
1653
                s->qr_size [inter][plane][0]= 63;
1654
                s->qr_base [inter][plane][0]=
1655
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1656
            }
1657
        }
1658

    
1659
        /* init VLC tables */
1660
        for (i = 0; i < 16; i++) {
1661

    
1662
            /* DC histograms */
1663
            init_vlc(&s->dc_vlc[i], 5, 32,
1664
                &dc_bias[i][0][1], 4, 2,
1665
                &dc_bias[i][0][0], 4, 2, 0);
1666

    
1667
            /* group 1 AC histograms */
1668
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1669
                &ac_bias_0[i][0][1], 4, 2,
1670
                &ac_bias_0[i][0][0], 4, 2, 0);
1671

    
1672
            /* group 2 AC histograms */
1673
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1674
                &ac_bias_1[i][0][1], 4, 2,
1675
                &ac_bias_1[i][0][0], 4, 2, 0);
1676

    
1677
            /* group 3 AC histograms */
1678
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1679
                &ac_bias_2[i][0][1], 4, 2,
1680
                &ac_bias_2[i][0][0], 4, 2, 0);
1681

    
1682
            /* group 4 AC histograms */
1683
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1684
                &ac_bias_3[i][0][1], 4, 2,
1685
                &ac_bias_3[i][0][0], 4, 2, 0);
1686
        }
1687
    } else {
1688
        for (i = 0; i < 16; i++) {
1689

    
1690
            /* DC histograms */
1691
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1692
                &s->huffman_table[i][0][1], 4, 2,
1693
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1694
                goto vlc_fail;
1695

    
1696
            /* group 1 AC histograms */
1697
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1698
                &s->huffman_table[i+16][0][1], 4, 2,
1699
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1700
                goto vlc_fail;
1701

    
1702
            /* group 2 AC histograms */
1703
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1704
                &s->huffman_table[i+16*2][0][1], 4, 2,
1705
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1706
                goto vlc_fail;
1707

    
1708
            /* group 3 AC histograms */
1709
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1710
                &s->huffman_table[i+16*3][0][1], 4, 2,
1711
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1712
                goto vlc_fail;
1713

    
1714
            /* group 4 AC histograms */
1715
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1716
                &s->huffman_table[i+16*4][0][1], 4, 2,
1717
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1718
                goto vlc_fail;
1719
        }
1720
    }
1721

    
1722
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1723
        &superblock_run_length_vlc_table[0][1], 4, 2,
1724
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1725

    
1726
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1727
        &fragment_run_length_vlc_table[0][1], 4, 2,
1728
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1729

    
1730
    init_vlc(&s->mode_code_vlc, 3, 8,
1731
        &mode_code_vlc_table[0][1], 2, 1,
1732
        &mode_code_vlc_table[0][0], 2, 1, 0);
1733

    
1734
    init_vlc(&s->motion_vector_vlc, 6, 63,
1735
        &motion_vector_vlc_table[0][1], 2, 1,
1736
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1737

    
1738
    /* work out the block mapping tables */
1739
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1740
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1741
    if (!s->superblock_fragments || !s->macroblock_coding) {
1742
        vp3_decode_end(avctx);
1743
        return -1;
1744
    }
1745
    init_block_mapping(s);
1746

    
1747
    for (i = 0; i < 3; i++) {
1748
        s->current_frame.data[i] = NULL;
1749
        s->last_frame.data[i] = NULL;
1750
        s->golden_frame.data[i] = NULL;
1751
    }
1752

    
1753
    return 0;
1754

    
1755
vlc_fail:
1756
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1757
    return -1;
1758
}
1759

    
1760
/*
1761
 * This is the ffmpeg/libavcodec API frame decode function.
1762
 */
1763
static int vp3_decode_frame(AVCodecContext *avctx,
1764
                            void *data, int *data_size,
1765
                            AVPacket *avpkt)
1766
{
1767
    const uint8_t *buf = avpkt->data;
1768
    int buf_size = avpkt->size;
1769
    Vp3DecodeContext *s = avctx->priv_data;
1770
    GetBitContext gb;
1771
    static int counter = 0;
1772
    int i;
1773

    
1774
    init_get_bits(&gb, buf, buf_size * 8);
1775

    
1776
    if (s->theora && get_bits1(&gb))
1777
    {
1778
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1779
        return -1;
1780
    }
1781

    
1782
    s->keyframe = !get_bits1(&gb);
1783
    if (!s->theora)
1784
        skip_bits(&gb, 1);
1785
    for (i = 0; i < 3; i++)
1786
        s->last_qps[i] = s->qps[i];
1787

    
1788
    s->nqps=0;
1789
    do{
1790
        s->qps[s->nqps++]= get_bits(&gb, 6);
1791
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1792
    for (i = s->nqps; i < 3; i++)
1793
        s->qps[i] = -1;
1794

    
1795
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1796
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1797
            s->keyframe?"key":"", counter, s->qps[0]);
1798
    counter++;
1799

    
1800
    if (s->qps[0] != s->last_qps[0])
1801
        init_loop_filter(s);
1802

    
1803
    for (i = 0; i < s->nqps; i++)
1804
        // reinit all dequantizers if the first one changed, because
1805
        // the DC of the first quantizer must be used for all matrices
1806
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1807
            init_dequantizer(s, i);
1808

    
1809
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1810
        return buf_size;
1811

    
1812
    s->current_frame.reference = 3;
1813
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1814
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1815
        goto error;
1816
    }
1817

    
1818
    if (s->keyframe) {
1819
        if (!s->theora)
1820
        {
1821
            skip_bits(&gb, 4); /* width code */
1822
            skip_bits(&gb, 4); /* height code */
1823
            if (s->version)
1824
            {
1825
                s->version = get_bits(&gb, 5);
1826
                if (counter == 1)
1827
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1828
            }
1829
        }
1830
        if (s->version || s->theora)
1831
        {
1832
                if (get_bits1(&gb))
1833
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1834
            skip_bits(&gb, 2); /* reserved? */
1835
        }
1836
    } else {
1837
        if (!s->golden_frame.data[0]) {
1838
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
1839
            goto error;
1840
        }
1841
    }
1842

    
1843
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1844
    s->current_frame.qstride= 0;
1845

    
1846
    init_frame(s, &gb);
1847

    
1848
    if (unpack_superblocks(s, &gb)){
1849
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1850
        goto error;
1851
    }
1852
    if (unpack_modes(s, &gb)){
1853
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1854
        goto error;
1855
    }
1856
    if (unpack_vectors(s, &gb)){
1857
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1858
        goto error;
1859
    }
1860
    if (unpack_block_qpis(s, &gb)){
1861
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1862
        goto error;
1863
    }
1864
    if (unpack_dct_coeffs(s, &gb)){
1865
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1866
        goto error;
1867
    }
1868

    
1869
    for (i = 0; i < 3; i++) {
1870
        if (s->flipped_image)
1871
            s->data_offset[i] = 0;
1872
        else
1873
            s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1874
    }
1875

    
1876
    s->last_slice_end = 0;
1877
    for (i = 0; i < s->c_superblock_height; i++)
1878
        render_slice(s, i);
1879

    
1880
    // filter the last row
1881
    for (i = 0; i < 3; i++) {
1882
        int row = (s->height >> (3+!!i)) - 1;
1883
        apply_loop_filter(s, i, row, row+1);
1884
    }
1885
    vp3_draw_horiz_band(s, s->height);
1886

    
1887
    *data_size=sizeof(AVFrame);
1888
    *(AVFrame*)data= s->current_frame;
1889

    
1890
    /* release the last frame, if it is allocated and if it is not the
1891
     * golden frame */
1892
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1893
        avctx->release_buffer(avctx, &s->last_frame);
1894

    
1895
    /* shuffle frames (last = current) */
1896
    s->last_frame= s->current_frame;
1897

    
1898
    if (s->keyframe) {
1899
        if (s->golden_frame.data[0])
1900
            avctx->release_buffer(avctx, &s->golden_frame);
1901
        s->golden_frame = s->current_frame;
1902
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1903
    }
1904

    
1905
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1906

    
1907
    return buf_size;
1908

    
1909
error:
1910
    if (s->current_frame.data[0])
1911
        avctx->release_buffer(avctx, &s->current_frame);
1912
    return -1;
1913
}
1914

    
1915
/*
1916
 * This is the ffmpeg/libavcodec API module cleanup function.
1917
 */
1918
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1919
{
1920
    Vp3DecodeContext *s = avctx->priv_data;
1921
    int i;
1922

    
1923
    av_free(s->superblock_coding);
1924
    av_free(s->all_fragments);
1925
    av_free(s->coded_fragment_list[0]);
1926
    av_free(s->dct_tokens_base);
1927
    av_free(s->superblock_fragments);
1928
    av_free(s->macroblock_coding);
1929

    
1930
    for (i = 0; i < 16; i++) {
1931
        free_vlc(&s->dc_vlc[i]);
1932
        free_vlc(&s->ac_vlc_1[i]);
1933
        free_vlc(&s->ac_vlc_2[i]);
1934
        free_vlc(&s->ac_vlc_3[i]);
1935
        free_vlc(&s->ac_vlc_4[i]);
1936
    }
1937

    
1938
    free_vlc(&s->superblock_run_length_vlc);
1939
    free_vlc(&s->fragment_run_length_vlc);
1940
    free_vlc(&s->mode_code_vlc);
1941
    free_vlc(&s->motion_vector_vlc);
1942

    
1943
    /* release all frames */
1944
    if (s->golden_frame.data[0])
1945
        avctx->release_buffer(avctx, &s->golden_frame);
1946
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1947
        avctx->release_buffer(avctx, &s->last_frame);
1948
    /* no need to release the current_frame since it will always be pointing
1949
     * to the same frame as either the golden or last frame */
1950

    
1951
    return 0;
1952
}
1953

    
1954
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1955
{
1956
    Vp3DecodeContext *s = avctx->priv_data;
1957

    
1958
    if (get_bits1(gb)) {
1959
        int token;
1960
        if (s->entries >= 32) { /* overflow */
1961
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1962
            return -1;
1963
        }
1964
        token = get_bits(gb, 5);
1965
        //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
1966
        s->huffman_table[s->hti][token][0] = s->hbits;
1967
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1968
        s->entries++;
1969
    }
1970
    else {
1971
        if (s->huff_code_size >= 32) {/* overflow */
1972
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1973
            return -1;
1974
        }
1975
        s->huff_code_size++;
1976
        s->hbits <<= 1;
1977
        if (read_huffman_tree(avctx, gb))
1978
            return -1;
1979
        s->hbits |= 1;
1980
        if (read_huffman_tree(avctx, gb))
1981
            return -1;
1982
        s->hbits >>= 1;
1983
        s->huff_code_size--;
1984
    }
1985
    return 0;
1986
}
1987

    
1988
#if CONFIG_THEORA_DECODER
1989
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1990
{
1991
    Vp3DecodeContext *s = avctx->priv_data;
1992
    int visible_width, visible_height, colorspace;
1993

    
1994
    s->theora = get_bits_long(gb, 24);
1995
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1996

    
1997
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1998
    /* but previous versions have the image flipped relative to vp3 */
1999
    if (s->theora < 0x030200)
2000
    {
2001
        s->flipped_image = 1;
2002
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
2003
    }
2004

    
2005
    visible_width  = s->width  = get_bits(gb, 16) << 4;
2006
    visible_height = s->height = get_bits(gb, 16) << 4;
2007

    
2008
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
2009
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
2010
        s->width= s->height= 0;
2011
        return -1;
2012
    }
2013

    
2014
    if (s->theora >= 0x030200) {
2015
        visible_width  = get_bits_long(gb, 24);
2016
        visible_height = get_bits_long(gb, 24);
2017

    
2018
        skip_bits(gb, 8); /* offset x */
2019
        skip_bits(gb, 8); /* offset y */
2020
    }
2021

    
2022
    skip_bits(gb, 32); /* fps numerator */
2023
    skip_bits(gb, 32); /* fps denumerator */
2024
    skip_bits(gb, 24); /* aspect numerator */
2025
    skip_bits(gb, 24); /* aspect denumerator */
2026

    
2027
    if (s->theora < 0x030200)
2028
        skip_bits(gb, 5); /* keyframe frequency force */
2029
    colorspace = get_bits(gb, 8);
2030
    skip_bits(gb, 24); /* bitrate */
2031

    
2032
    skip_bits(gb, 6); /* quality hint */
2033

    
2034
    if (s->theora >= 0x030200)
2035
    {
2036
        skip_bits(gb, 5); /* keyframe frequency force */
2037
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
2038
        skip_bits(gb, 3); /* reserved */
2039
    }
2040

    
2041
//    align_get_bits(gb);
2042

    
2043
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2044
        && visible_height <= s->height && visible_height > s->height-16)
2045
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2046
    else
2047
        avcodec_set_dimensions(avctx, s->width, s->height);
2048

    
2049
    if (colorspace == 1) {
2050
        avctx->color_primaries = AVCOL_PRI_BT470M;
2051
    } else if (colorspace == 2) {
2052
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2053
    }
2054
    if (colorspace == 1 || colorspace == 2) {
2055
        avctx->colorspace = AVCOL_SPC_BT470BG;
2056
        avctx->color_trc  = AVCOL_TRC_BT709;
2057
    }
2058

    
2059
    return 0;
2060
}
2061

    
2062
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2063
{
2064
    Vp3DecodeContext *s = avctx->priv_data;
2065
    int i, n, matrices, inter, plane;
2066

    
2067
    if (s->theora >= 0x030200) {
2068
        n = get_bits(gb, 3);
2069
        /* loop filter limit values table */
2070
        for (i = 0; i < 64; i++) {
2071
            s->filter_limit_values[i] = get_bits(gb, n);
2072
            if (s->filter_limit_values[i] > 127) {
2073
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2074
                s->filter_limit_values[i] = 127;
2075
            }
2076
        }
2077
    }
2078

    
2079
    if (s->theora >= 0x030200)
2080
        n = get_bits(gb, 4) + 1;
2081
    else
2082
        n = 16;
2083
    /* quality threshold table */
2084
    for (i = 0; i < 64; i++)
2085
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2086

    
2087
    if (s->theora >= 0x030200)
2088
        n = get_bits(gb, 4) + 1;
2089
    else
2090
        n = 16;
2091
    /* dc scale factor table */
2092
    for (i = 0; i < 64; i++)
2093
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2094

    
2095
    if (s->theora >= 0x030200)
2096
        matrices = get_bits(gb, 9) + 1;
2097
    else
2098
        matrices = 3;
2099

    
2100
    if(matrices > 384){
2101
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2102
        return -1;
2103
    }
2104

    
2105
    for(n=0; n<matrices; n++){
2106
        for (i = 0; i < 64; i++)
2107
            s->base_matrix[n][i]= get_bits(gb, 8);
2108
    }
2109

    
2110
    for (inter = 0; inter <= 1; inter++) {
2111
        for (plane = 0; plane <= 2; plane++) {
2112
            int newqr= 1;
2113
            if (inter || plane > 0)
2114
                newqr = get_bits1(gb);
2115
            if (!newqr) {
2116
                int qtj, plj;
2117
                if(inter && get_bits1(gb)){
2118
                    qtj = 0;
2119
                    plj = plane;
2120
                }else{
2121
                    qtj= (3*inter + plane - 1) / 3;
2122
                    plj= (plane + 2) % 3;
2123
                }
2124
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2125
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2126
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2127
            } else {
2128
                int qri= 0;
2129
                int qi = 0;
2130

    
2131
                for(;;){
2132
                    i= get_bits(gb, av_log2(matrices-1)+1);
2133
                    if(i>= matrices){
2134
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2135
                        return -1;
2136
                    }
2137
                    s->qr_base[inter][plane][qri]= i;
2138
                    if(qi >= 63)
2139
                        break;
2140
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2141
                    s->qr_size[inter][plane][qri++]= i;
2142
                    qi += i;
2143
                }
2144

    
2145
                if (qi > 63) {
2146
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2147
                    return -1;
2148
                }
2149
                s->qr_count[inter][plane]= qri;
2150
            }
2151
        }
2152
    }
2153

    
2154
    /* Huffman tables */
2155
    for (s->hti = 0; s->hti < 80; s->hti++) {
2156
        s->entries = 0;
2157
        s->huff_code_size = 1;
2158
        if (!get_bits1(gb)) {
2159
            s->hbits = 0;
2160
            if(read_huffman_tree(avctx, gb))
2161
                return -1;
2162
            s->hbits = 1;
2163
            if(read_huffman_tree(avctx, gb))
2164
                return -1;
2165
        }
2166
    }
2167

    
2168
    s->theora_tables = 1;
2169

    
2170
    return 0;
2171
}
2172

    
2173
static av_cold int theora_decode_init(AVCodecContext *avctx)
2174
{
2175
    Vp3DecodeContext *s = avctx->priv_data;
2176
    GetBitContext gb;
2177
    int ptype;
2178
    uint8_t *header_start[3];
2179
    int header_len[3];
2180
    int i;
2181

    
2182
    s->theora = 1;
2183

    
2184
    if (!avctx->extradata_size)
2185
    {
2186
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2187
        return -1;
2188
    }
2189

    
2190
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2191
                              42, header_start, header_len) < 0) {
2192
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2193
        return -1;
2194
    }
2195

    
2196
  for(i=0;i<3;i++) {
2197
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2198

    
2199
    ptype = get_bits(&gb, 8);
2200

    
2201
     if (!(ptype & 0x80))
2202
     {
2203
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2204
//        return -1;
2205
     }
2206

    
2207
    // FIXME: Check for this as well.
2208
    skip_bits_long(&gb, 6*8); /* "theora" */
2209

    
2210
    switch(ptype)
2211
    {
2212
        case 0x80:
2213
            theora_decode_header(avctx, &gb);
2214
                break;
2215
        case 0x81:
2216
// FIXME: is this needed? it breaks sometimes
2217
//            theora_decode_comments(avctx, gb);
2218
            break;
2219
        case 0x82:
2220
            if (theora_decode_tables(avctx, &gb))
2221
                return -1;
2222
            break;
2223
        default:
2224
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2225
            break;
2226
    }
2227
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2228
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2229
    if (s->theora < 0x030200)
2230
        break;
2231
  }
2232

    
2233
    return vp3_decode_init(avctx);
2234
}
2235

    
2236
AVCodec theora_decoder = {
2237
    "theora",
2238
    CODEC_TYPE_VIDEO,
2239
    CODEC_ID_THEORA,
2240
    sizeof(Vp3DecodeContext),
2241
    theora_decode_init,
2242
    NULL,
2243
    vp3_decode_end,
2244
    vp3_decode_frame,
2245
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2246
    NULL,
2247
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2248
};
2249
#endif
2250

    
2251
AVCodec vp3_decoder = {
2252
    "vp3",
2253
    CODEC_TYPE_VIDEO,
2254
    CODEC_ID_VP3,
2255
    sizeof(Vp3DecodeContext),
2256
    vp3_decode_init,
2257
    NULL,
2258
    vp3_decode_end,
2259
    vp3_decode_frame,
2260
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2261
    NULL,
2262
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2263
};