Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 1e76a1da

History | View | Annotate | Download (76.1 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
    int chroma_x_shift, chroma_y_shift;
134
    AVFrame golden_frame;
135
    AVFrame last_frame;
136
    AVFrame current_frame;
137
    int keyframe;
138
    DSPContext dsp;
139
    int flipped_image;
140
    int last_slice_end;
141

    
142
    int qps[3];
143
    int nqps;
144
    int last_qps[3];
145

    
146
    int superblock_count;
147
    int y_superblock_width;
148
    int y_superblock_height;
149
    int y_superblock_count;
150
    int c_superblock_width;
151
    int c_superblock_height;
152
    int c_superblock_count;
153
    int u_superblock_start;
154
    int v_superblock_start;
155
    unsigned char *superblock_coding;
156

    
157
    int macroblock_count;
158
    int macroblock_width;
159
    int macroblock_height;
160

    
161
    int fragment_count;
162
    int fragment_width[2];
163
    int fragment_height[2];
164

    
165
    Vp3Fragment *all_fragments;
166
    int fragment_start[3];
167
    int data_offset[3];
168

    
169
    ScanTable scantable;
170

    
171
    /* tables */
172
    uint16_t coded_dc_scale_factor[64];
173
    uint32_t coded_ac_scale_factor[64];
174
    uint8_t base_matrix[384][64];
175
    uint8_t qr_count[2][3];
176
    uint8_t qr_size [2][3][64];
177
    uint16_t qr_base[2][3][64];
178

    
179
    /**
180
     * This is a list of all tokens in bitstream order. Reordering takes place
181
     * by pulling from each level during IDCT. As a consequence, IDCT must be
182
     * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
183
     * otherwise. The 32 different tokens with up to 12 bits of extradata are
184
     * collapsed into 3 types, packed as follows:
185
     *   (from the low to high bits)
186
     *
187
     * 2 bits: type (0,1,2)
188
     *   0: EOB run, 14 bits for run length (12 needed)
189
     *   1: zero run, 7 bits for run length
190
     *                7 bits for the next coefficient (3 needed)
191
     *   2: coefficient, 14 bits (11 needed)
192
     *
193
     * Coefficients are signed, so are packed in the highest bits for automatic
194
     * sign extension.
195
     */
196
    int16_t *dct_tokens[3][64];
197
    int16_t *dct_tokens_base;
198
#define TOKEN_EOB(eob_run)              ((eob_run) << 2)
199
#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
200
#define TOKEN_COEFF(coeff)              (((coeff) << 2) + 2)
201

    
202
    /**
203
     * number of blocks that contain DCT coefficients at the given level or higher
204
     */
205
    int num_coded_frags[3][64];
206
    int total_num_coded_frags;
207

    
208
    /* this is a list of indexes into the all_fragments array indicating
209
     * which of the fragments are coded */
210
    int *coded_fragment_list[3];
211

    
212
    VLC dc_vlc[16];
213
    VLC ac_vlc_1[16];
214
    VLC ac_vlc_2[16];
215
    VLC ac_vlc_3[16];
216
    VLC ac_vlc_4[16];
217

    
218
    VLC superblock_run_length_vlc;
219
    VLC fragment_run_length_vlc;
220
    VLC mode_code_vlc;
221
    VLC motion_vector_vlc;
222

    
223
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
224
     * index into them */
225
    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
226

    
227
    /* This table contains superblock_count * 16 entries. Each set of 16
228
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
229
     * An entry will be -1 to indicate that no entry corresponds to that
230
     * index. */
231
    int *superblock_fragments;
232

    
233
    /* This is an array that indicates how a particular macroblock
234
     * is coded. */
235
    unsigned char *macroblock_coding;
236

    
237
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
238
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
239

    
240
    /* Huffman decode */
241
    int hti;
242
    unsigned int hbits;
243
    int entries;
244
    int huff_code_size;
245
    uint16_t huffman_table[80][32][2];
246

    
247
    uint8_t filter_limit_values[64];
248
    DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
249
} Vp3DecodeContext;
250

    
251
/************************************************************************
252
 * VP3 specific functions
253
 ************************************************************************/
254

    
255
/*
256
 * This function sets up all of the various blocks mappings:
257
 * superblocks <-> fragments, macroblocks <-> fragments,
258
 * superblocks <-> macroblocks
259
 *
260
 * Returns 0 is successful; returns 1 if *anything* went wrong.
261
 */
262
static int init_block_mapping(Vp3DecodeContext *s)
263
{
264
    int sb_x, sb_y, plane;
265
    int x, y, i, j = 0;
266

    
267
    for (plane = 0; plane < 3; plane++) {
268
        int sb_width    = plane ? s->c_superblock_width  : s->y_superblock_width;
269
        int sb_height   = plane ? s->c_superblock_height : s->y_superblock_height;
270
        int frag_width  = s->fragment_width[!!plane];
271
        int frag_height = s->fragment_height[!!plane];
272

    
273
        for (sb_y = 0; sb_y < sb_height; sb_y++)
274
            for (sb_x = 0; sb_x < sb_width; sb_x++)
275
                for (i = 0; i < 16; i++) {
276
                    x = 4*sb_x + hilbert_offset[i][0];
277
                    y = 4*sb_y + hilbert_offset[i][1];
278

    
279
                    if (x < frag_width && y < frag_height)
280
                        s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
281
                    else
282
                        s->superblock_fragments[j++] = -1;
283
                }
284
    }
285

    
286
    return 0;  /* successful path out */
287
}
288

    
289
/*
290
 * This function sets up the dequantization tables used for a particular
291
 * frame.
292
 */
293
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
294
{
295
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
296
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
297
    int i, plane, inter, qri, bmi, bmj, qistart;
298

    
299
    for(inter=0; inter<2; inter++){
300
        for(plane=0; plane<3; plane++){
301
            int sum=0;
302
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
303
                sum+= s->qr_size[inter][plane][qri];
304
                if(s->qps[qpi] <= sum)
305
                    break;
306
            }
307
            qistart= sum - s->qr_size[inter][plane][qri];
308
            bmi= s->qr_base[inter][plane][qri  ];
309
            bmj= s->qr_base[inter][plane][qri+1];
310
            for(i=0; i<64; i++){
311
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
312
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
313
                            + s->qr_size[inter][plane][qri])
314
                           / (2*s->qr_size[inter][plane][qri]);
315

    
316
                int qmin= 8<<(inter + !i);
317
                int qscale= i ? ac_scale_factor : dc_scale_factor;
318

    
319
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
320
            }
321
            // all DC coefficients use the same quant so as not to interfere with DC prediction
322
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
323
        }
324
    }
325

    
326
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
327
}
328

    
329
/*
330
 * This function initializes the loop filter boundary limits if the frame's
331
 * quality index is different from the previous frame's.
332
 *
333
 * The filter_limit_values may not be larger than 127.
334
 */
335
static void init_loop_filter(Vp3DecodeContext *s)
336
{
337
    int *bounding_values= s->bounding_values_array+127;
338
    int filter_limit;
339
    int x;
340
    int value;
341

    
342
    filter_limit = s->filter_limit_values[s->qps[0]];
343

    
344
    /* set up the bounding values */
345
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
346
    for (x = 0; x < filter_limit; x++) {
347
        bounding_values[-x] = -x;
348
        bounding_values[x] = x;
349
    }
350
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
351
        bounding_values[ x] =  value;
352
        bounding_values[-x] = -value;
353
    }
354
    if (value)
355
        bounding_values[128] = value;
356
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
357
}
358

    
359
/*
360
 * This function unpacks all of the superblock/macroblock/fragment coding
361
 * information from the bitstream.
362
 */
363
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
364
{
365
    int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
366
    int bit = 0;
367
    int current_superblock = 0;
368
    int current_run = 0;
369
    int num_partial_superblocks = 0;
370

    
371
    int i, j;
372
    int current_fragment;
373
    int plane;
374

    
375
    if (s->keyframe) {
376
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
377

    
378
    } else {
379

    
380
        /* unpack the list of partially-coded superblocks */
381
        bit = get_bits1(gb);
382
        while (current_superblock < s->superblock_count) {
383
                current_run = get_vlc2(gb,
384
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
385
                if (current_run == 34)
386
                    current_run += get_bits(gb, 12);
387

    
388
            if (current_superblock + current_run > s->superblock_count) {
389
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
390
                return -1;
391
            }
392

    
393
            memset(s->superblock_coding + current_superblock, bit, current_run);
394

    
395
            current_superblock += current_run;
396
            if (bit)
397
                num_partial_superblocks += current_run;
398

    
399
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
400
                bit = get_bits1(gb);
401
            else
402
                bit ^= 1;
403
        }
404

    
405
        /* unpack the list of fully coded superblocks if any of the blocks were
406
         * not marked as partially coded in the previous step */
407
        if (num_partial_superblocks < s->superblock_count) {
408
            int superblocks_decoded = 0;
409

    
410
            current_superblock = 0;
411
            bit = get_bits1(gb);
412
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks) {
413
                        current_run = get_vlc2(gb,
414
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
415
                        if (current_run == 34)
416
                            current_run += get_bits(gb, 12);
417

    
418
                for (j = 0; j < current_run; current_superblock++) {
419
                    if (current_superblock >= s->superblock_count) {
420
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
421
                        return -1;
422
                    }
423

    
424
                /* skip any superblocks already marked as partially coded */
425
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
426
                    s->superblock_coding[current_superblock] = 2*bit;
427
                    j++;
428
                }
429
                }
430
                superblocks_decoded += current_run;
431

    
432
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
433
                    bit = get_bits1(gb);
434
                else
435
                    bit ^= 1;
436
            }
437
        }
438

    
439
        /* if there were partial blocks, initialize bitstream for
440
         * unpacking fragment codings */
441
        if (num_partial_superblocks) {
442

    
443
            current_run = 0;
444
            bit = get_bits1(gb);
445
            /* toggle the bit because as soon as the first run length is
446
             * fetched the bit will be toggled again */
447
            bit ^= 1;
448
        }
449
    }
450

    
451
    /* figure out which fragments are coded; iterate through each
452
     * superblock (all planes) */
453
    s->total_num_coded_frags = 0;
454
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
455

    
456
    for (plane = 0; plane < 3; plane++) {
457
        int sb_start = superblock_starts[plane];
458
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
459
        int num_coded_frags = 0;
460

    
461
    for (i = sb_start; i < sb_end; i++) {
462

    
463
        /* iterate through all 16 fragments in a superblock */
464
        for (j = 0; j < 16; j++) {
465

    
466
            /* if the fragment is in bounds, check its coding status */
467
            current_fragment = s->superblock_fragments[i * 16 + j];
468
            if (current_fragment != -1) {
469
                int coded = s->superblock_coding[i];
470

    
471
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
472

    
473
                    /* fragment may or may not be coded; this is the case
474
                     * that cares about the fragment coding runs */
475
                    if (current_run-- == 0) {
476
                        bit ^= 1;
477
                        current_run = get_vlc2(gb,
478
                            s->fragment_run_length_vlc.table, 5, 2);
479
                    }
480
                    coded = bit;
481
                }
482

    
483
                    if (coded) {
484
                        /* default mode; actual mode will be decoded in
485
                         * the next phase */
486
                        s->all_fragments[current_fragment].coding_method =
487
                            MODE_INTER_NO_MV;
488
                        s->coded_fragment_list[plane][num_coded_frags++] =
489
                            current_fragment;
490
                    } else {
491
                        /* not coded; copy this fragment from the prior frame */
492
                        s->all_fragments[current_fragment].coding_method =
493
                            MODE_COPY;
494
                    }
495
            }
496
        }
497
    }
498
        s->total_num_coded_frags += num_coded_frags;
499
        for (i = 0; i < 64; i++)
500
            s->num_coded_frags[plane][i] = num_coded_frags;
501
        if (plane < 2)
502
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
503
    }
504
    return 0;
505
}
506

    
507
/*
508
 * This function unpacks all the coding mode data for individual macroblocks
509
 * from the bitstream.
510
 */
511
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
512
{
513
    int i, j, k, sb_x, sb_y;
514
    int scheme;
515
    int current_macroblock;
516
    int current_fragment;
517
    int coding_mode;
518
    int custom_mode_alphabet[CODING_MODE_COUNT];
519
    const int *alphabet;
520
    Vp3Fragment *frag;
521

    
522
    if (s->keyframe) {
523
        for (i = 0; i < s->fragment_count; i++)
524
            s->all_fragments[i].coding_method = MODE_INTRA;
525

    
526
    } else {
527

    
528
        /* fetch the mode coding scheme for this frame */
529
        scheme = get_bits(gb, 3);
530

    
531
        /* is it a custom coding scheme? */
532
        if (scheme == 0) {
533
            for (i = 0; i < 8; i++)
534
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
535
            for (i = 0; i < 8; i++)
536
                custom_mode_alphabet[get_bits(gb, 3)] = i;
537
            alphabet = custom_mode_alphabet;
538
        } else
539
            alphabet = ModeAlphabet[scheme-1];
540

    
541
        /* iterate through all of the macroblocks that contain 1 or more
542
         * coded fragments */
543
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
544
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
545

    
546
            for (j = 0; j < 4; j++) {
547
                int mb_x = 2*sb_x +   (j>>1);
548
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
549
                current_macroblock = mb_y * s->macroblock_width + mb_x;
550

    
551
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
552
                    continue;
553

    
554
#define BLOCK_X (2*mb_x + (k&1))
555
#define BLOCK_Y (2*mb_y + (k>>1))
556
                /* coding modes are only stored if the macroblock has at least one
557
                 * luma block coded, otherwise it must be INTER_NO_MV */
558
                for (k = 0; k < 4; k++) {
559
                    current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
560
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
561
                        break;
562
                }
563
                if (k == 4) {
564
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
565
                    continue;
566
                }
567

    
568
                /* mode 7 means get 3 bits for each coding mode */
569
                if (scheme == 7)
570
                    coding_mode = get_bits(gb, 3);
571
                else
572
                    coding_mode = alphabet
573
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
574

    
575
                s->macroblock_coding[current_macroblock] = coding_mode;
576
                for (k = 0; k < 4; k++) {
577
                    frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
578
                    if (frag->coding_method != MODE_COPY)
579
                        frag->coding_method = coding_mode;
580
                }
581

    
582
#define SET_CHROMA_MODES \
583
    if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
584
        frag[s->fragment_start[1]].coding_method = coding_mode;\
585
    if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
586
        frag[s->fragment_start[2]].coding_method = coding_mode;
587

    
588
                if (s->chroma_y_shift) {
589
                    frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
590
                    SET_CHROMA_MODES
591
                } else if (s->chroma_x_shift) {
592
                    frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
593
                    for (k = 0; k < 2; k++) {
594
                        SET_CHROMA_MODES
595
                        frag += s->fragment_width[1];
596
                    }
597
                } else {
598
                    for (k = 0; k < 4; k++) {
599
                        frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
600
                        SET_CHROMA_MODES
601
                    }
602
                }
603
            }
604
            }
605
        }
606
    }
607

    
608
    return 0;
609
}
610

    
611
/*
612
 * This function unpacks all the motion vectors for the individual
613
 * macroblocks from the bitstream.
614
 */
615
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
616
{
617
    int j, k, sb_x, sb_y;
618
    int coding_mode;
619
    int motion_x[4];
620
    int motion_y[4];
621
    int last_motion_x = 0;
622
    int last_motion_y = 0;
623
    int prior_last_motion_x = 0;
624
    int prior_last_motion_y = 0;
625
    int current_macroblock;
626
    int current_fragment;
627
    Vp3Fragment *frag;
628

    
629
    if (s->keyframe)
630
        return 0;
631

    
632
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
633
    coding_mode = get_bits1(gb);
634

    
635
    /* iterate through all of the macroblocks that contain 1 or more
636
     * coded fragments */
637
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
638
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
639

    
640
        for (j = 0; j < 4; j++) {
641
            int mb_x = 2*sb_x +   (j>>1);
642
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
643
            current_macroblock = mb_y * s->macroblock_width + mb_x;
644

    
645
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
646
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
647
                continue;
648

    
649
            switch (s->macroblock_coding[current_macroblock]) {
650

    
651
            case MODE_INTER_PLUS_MV:
652
            case MODE_GOLDEN_MV:
653
                /* all 6 fragments use the same motion vector */
654
                if (coding_mode == 0) {
655
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
656
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
657
                } else {
658
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
659
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
660
                }
661

    
662
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
663
                if (s->macroblock_coding[current_macroblock] ==
664
                    MODE_INTER_PLUS_MV) {
665
                    prior_last_motion_x = last_motion_x;
666
                    prior_last_motion_y = last_motion_y;
667
                    last_motion_x = motion_x[0];
668
                    last_motion_y = motion_y[0];
669
                }
670
                break;
671

    
672
            case MODE_INTER_FOURMV:
673
                /* vector maintenance */
674
                prior_last_motion_x = last_motion_x;
675
                prior_last_motion_y = last_motion_y;
676

    
677
                /* fetch 4 vectors from the bitstream, one for each
678
                 * Y fragment, then average for the C fragment vectors */
679
                for (k = 0; k < 4; k++) {
680
                    current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
681
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
682
                        if (coding_mode == 0) {
683
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
684
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
685
                        } else {
686
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
687
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
688
                        }
689
                        last_motion_x = motion_x[k];
690
                        last_motion_y = motion_y[k];
691
                    } else {
692
                        motion_x[k] = 0;
693
                        motion_y[k] = 0;
694
                    }
695
                }
696
                break;
697

    
698
            case MODE_INTER_LAST_MV:
699
                /* all 6 fragments use the last motion vector */
700
                motion_x[0] = last_motion_x;
701
                motion_y[0] = last_motion_y;
702

    
703
                /* no vector maintenance (last vector remains the
704
                 * last vector) */
705
                break;
706

    
707
            case MODE_INTER_PRIOR_LAST:
708
                /* all 6 fragments use the motion vector prior to the
709
                 * last motion vector */
710
                motion_x[0] = prior_last_motion_x;
711
                motion_y[0] = prior_last_motion_y;
712

    
713
                /* vector maintenance */
714
                prior_last_motion_x = last_motion_x;
715
                prior_last_motion_y = last_motion_y;
716
                last_motion_x = motion_x[0];
717
                last_motion_y = motion_y[0];
718
                break;
719

    
720
            default:
721
                /* covers intra, inter without MV, golden without MV */
722
                motion_x[0] = 0;
723
                motion_y[0] = 0;
724

    
725
                /* no vector maintenance */
726
                break;
727
            }
728

    
729
            /* assign the motion vectors to the correct fragments */
730
            for (k = 0; k < 4; k++) {
731
                current_fragment =
732
                    BLOCK_Y*s->fragment_width[0] + BLOCK_X;
733
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
734
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
735
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
736
                } else {
737
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
738
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
739
                }
740
            }
741

    
742
#define SET_CHROMA_MV(mx, my) \
743
    frag[s->fragment_start[1]].motion_x = mx; \
744
    frag[s->fragment_start[1]].motion_y = my; \
745
    frag[s->fragment_start[2]].motion_x = mx; \
746
    frag[s->fragment_start[2]].motion_y = my
747

    
748
            if (s->chroma_y_shift) {
749
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
750
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
751
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
752
                }
753
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
754
                motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
755
                frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
756
                SET_CHROMA_MV(motion_x[0], motion_y[0]);
757
            } else if (s->chroma_x_shift) {
758
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
759
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
760
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
761
                    motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
762
                    motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
763
                } else {
764
                    motion_x[1] = motion_x[0];
765
                    motion_y[1] = motion_y[0];
766
                }
767
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
768
                motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
769

    
770
                frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
771
                for (k = 0; k < 2; k++) {
772
                    SET_CHROMA_MV(motion_x[k], motion_y[k]);
773
                    frag += s->fragment_width[1];
774
                }
775
            } else {
776
                for (k = 0; k < 4; k++) {
777
                    frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
778
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
779
                        SET_CHROMA_MV(motion_x[k], motion_y[k]);
780
                    } else {
781
                        SET_CHROMA_MV(motion_x[0], motion_y[0]);
782
                    }
783
                }
784
            }
785
        }
786
        }
787
    }
788

    
789
    return 0;
790
}
791

    
792
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
793
{
794
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
795
    int num_blocks = s->total_num_coded_frags;
796

    
797
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
798
        i = blocks_decoded = num_blocks_at_qpi = 0;
799

    
800
        bit = get_bits1(gb);
801

    
802
        do {
803
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
804
            if (run_length == 34)
805
                run_length += get_bits(gb, 12);
806
            blocks_decoded += run_length;
807

    
808
            if (!bit)
809
                num_blocks_at_qpi += run_length;
810

    
811
            for (j = 0; j < run_length; i++) {
812
                if (i >= s->total_num_coded_frags)
813
                    return -1;
814

    
815
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
816
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
817
                    j++;
818
                }
819
            }
820

    
821
            if (run_length == MAXIMUM_LONG_BIT_RUN)
822
                bit = get_bits1(gb);
823
            else
824
                bit ^= 1;
825
        } while (blocks_decoded < num_blocks);
826

    
827
        num_blocks -= num_blocks_at_qpi;
828
    }
829

    
830
    return 0;
831
}
832

    
833
/*
834
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
835
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
836
 * data. This function unpacks all the VLCs for either the Y plane or both
837
 * C planes, and is called for DC coefficients or different AC coefficient
838
 * levels (since different coefficient types require different VLC tables.
839
 *
840
 * This function returns a residual eob run. E.g, if a particular token gave
841
 * instructions to EOB the next 5 fragments and there were only 2 fragments
842
 * left in the current fragment range, 3 would be returned so that it could
843
 * be passed into the next call to this same function.
844
 */
845
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
846
                        VLC *table, int coeff_index,
847
                        int plane,
848
                        int eob_run)
849
{
850
    int i, j = 0;
851
    int token;
852
    int zero_run = 0;
853
    DCTELEM coeff = 0;
854
    int bits_to_get;
855
    int blocks_ended;
856
    int coeff_i = 0;
857
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
858
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
859

    
860
    /* local references to structure members to avoid repeated deferences */
861
    int *coded_fragment_list = s->coded_fragment_list[plane];
862
    Vp3Fragment *all_fragments = s->all_fragments;
863
    VLC_TYPE (*vlc_table)[2] = table->table;
864

    
865
    if (num_coeffs < 0)
866
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
867

    
868
    if (eob_run > num_coeffs) {
869
        coeff_i = blocks_ended = num_coeffs;
870
        eob_run -= num_coeffs;
871
    } else {
872
        coeff_i = blocks_ended = eob_run;
873
        eob_run = 0;
874
    }
875

    
876
    // insert fake EOB token to cover the split between planes or zzi
877
    if (blocks_ended)
878
        dct_tokens[j++] = blocks_ended << 2;
879

    
880
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
881
            /* decode a VLC into a token */
882
            token = get_vlc2(gb, vlc_table, 5, 3);
883
            /* use the token to get a zero run, a coefficient, and an eob run */
884
            if (token <= 6) {
885
                eob_run = eob_run_base[token];
886
                if (eob_run_get_bits[token])
887
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
888

    
889
                // record only the number of blocks ended in this plane,
890
                // any spill will be recorded in the next plane.
891
                if (eob_run > num_coeffs - coeff_i) {
892
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
893
                    blocks_ended   += num_coeffs - coeff_i;
894
                    eob_run        -= num_coeffs - coeff_i;
895
                    coeff_i         = num_coeffs;
896
                } else {
897
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
898
                    blocks_ended   += eob_run;
899
                    coeff_i        += eob_run;
900
                    eob_run = 0;
901
                }
902
            } else {
903
                bits_to_get = coeff_get_bits[token];
904
                if (bits_to_get)
905
                    bits_to_get = get_bits(gb, bits_to_get);
906
                coeff = coeff_tables[token][bits_to_get];
907

    
908
                zero_run = zero_run_base[token];
909
                if (zero_run_get_bits[token])
910
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
911

    
912
                if (zero_run) {
913
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
914
                } else {
915
                    // Save DC into the fragment structure. DC prediction is
916
                    // done in raster order, so the actual DC can't be in with
917
                    // other tokens. We still need the token in dct_tokens[]
918
                    // however, or else the structure collapses on itself.
919
                    if (!coeff_index)
920
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
921

    
922
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
923
                }
924

    
925
                if (coeff_index + zero_run > 64) {
926
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
927
                           " %d coeffs left\n", zero_run, 64-coeff_index);
928
                    zero_run = 64 - coeff_index;
929
                }
930

    
931
                // zero runs code multiple coefficients,
932
                // so don't try to decode coeffs for those higher levels
933
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
934
                    s->num_coded_frags[plane][i]--;
935
                coeff_i++;
936
            }
937
    }
938

    
939
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
940
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
941

    
942
    // decrement the number of blocks that have higher coeffecients for each
943
    // EOB run at this level
944
    if (blocks_ended)
945
        for (i = coeff_index+1; i < 64; i++)
946
            s->num_coded_frags[plane][i] -= blocks_ended;
947

    
948
    // setup the next buffer
949
    if (plane < 2)
950
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
951
    else if (coeff_index < 63)
952
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
953

    
954
    return eob_run;
955
}
956

    
957
static void reverse_dc_prediction(Vp3DecodeContext *s,
958
                                  int first_fragment,
959
                                  int fragment_width,
960
                                  int fragment_height);
961
/*
962
 * This function unpacks all of the DCT coefficient data from the
963
 * bitstream.
964
 */
965
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
966
{
967
    int i;
968
    int dc_y_table;
969
    int dc_c_table;
970
    int ac_y_table;
971
    int ac_c_table;
972
    int residual_eob_run = 0;
973
    VLC *y_tables[64];
974
    VLC *c_tables[64];
975

    
976
    s->dct_tokens[0][0] = s->dct_tokens_base;
977

    
978
    /* fetch the DC table indexes */
979
    dc_y_table = get_bits(gb, 4);
980
    dc_c_table = get_bits(gb, 4);
981

    
982
    /* unpack the Y plane DC coefficients */
983
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
984
        0, residual_eob_run);
985

    
986
    /* reverse prediction of the Y-plane DC coefficients */
987
    reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
988

    
989
    /* unpack the C plane DC coefficients */
990
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
991
        1, residual_eob_run);
992
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
993
        2, residual_eob_run);
994

    
995
    /* reverse prediction of the C-plane DC coefficients */
996
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
997
    {
998
        reverse_dc_prediction(s, s->fragment_start[1],
999
            s->fragment_width[1], s->fragment_height[1]);
1000
        reverse_dc_prediction(s, s->fragment_start[2],
1001
            s->fragment_width[1], s->fragment_height[1]);
1002
    }
1003

    
1004
    /* fetch the AC table indexes */
1005
    ac_y_table = get_bits(gb, 4);
1006
    ac_c_table = get_bits(gb, 4);
1007

    
1008
    /* build tables of AC VLC tables */
1009
    for (i = 1; i <= 5; i++) {
1010
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1011
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1012
    }
1013
    for (i = 6; i <= 14; i++) {
1014
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1015
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1016
    }
1017
    for (i = 15; i <= 27; i++) {
1018
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1019
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1020
    }
1021
    for (i = 28; i <= 63; i++) {
1022
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1023
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1024
    }
1025

    
1026
    /* decode all AC coefficents */
1027
    for (i = 1; i <= 63; i++) {
1028
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1029
                0, residual_eob_run);
1030

    
1031
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1032
                1, residual_eob_run);
1033
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1034
                2, residual_eob_run);
1035
    }
1036

    
1037
    return 0;
1038
}
1039

    
1040
/*
1041
 * This function reverses the DC prediction for each coded fragment in
1042
 * the frame. Much of this function is adapted directly from the original
1043
 * VP3 source code.
1044
 */
1045
#define COMPATIBLE_FRAME(x) \
1046
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1047
#define DC_COEFF(u) s->all_fragments[u].dc
1048

    
1049
static void reverse_dc_prediction(Vp3DecodeContext *s,
1050
                                  int first_fragment,
1051
                                  int fragment_width,
1052
                                  int fragment_height)
1053
{
1054

    
1055
#define PUL 8
1056
#define PU 4
1057
#define PUR 2
1058
#define PL 1
1059

    
1060
    int x, y;
1061
    int i = first_fragment;
1062

    
1063
    int predicted_dc;
1064

    
1065
    /* DC values for the left, up-left, up, and up-right fragments */
1066
    int vl, vul, vu, vur;
1067

    
1068
    /* indexes for the left, up-left, up, and up-right fragments */
1069
    int l, ul, u, ur;
1070

    
1071
    /*
1072
     * The 6 fields mean:
1073
     *   0: up-left multiplier
1074
     *   1: up multiplier
1075
     *   2: up-right multiplier
1076
     *   3: left multiplier
1077
     */
1078
    static const int predictor_transform[16][4] = {
1079
        {  0,  0,  0,  0},
1080
        {  0,  0,  0,128},        // PL
1081
        {  0,  0,128,  0},        // PUR
1082
        {  0,  0, 53, 75},        // PUR|PL
1083
        {  0,128,  0,  0},        // PU
1084
        {  0, 64,  0, 64},        // PU|PL
1085
        {  0,128,  0,  0},        // PU|PUR
1086
        {  0,  0, 53, 75},        // PU|PUR|PL
1087
        {128,  0,  0,  0},        // PUL
1088
        {  0,  0,  0,128},        // PUL|PL
1089
        { 64,  0, 64,  0},        // PUL|PUR
1090
        {  0,  0, 53, 75},        // PUL|PUR|PL
1091
        {  0,128,  0,  0},        // PUL|PU
1092
       {-104,116,  0,116},        // PUL|PU|PL
1093
        { 24, 80, 24,  0},        // PUL|PU|PUR
1094
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1095
    };
1096

    
1097
    /* This table shows which types of blocks can use other blocks for
1098
     * prediction. For example, INTRA is the only mode in this table to
1099
     * have a frame number of 0. That means INTRA blocks can only predict
1100
     * from other INTRA blocks. There are 2 golden frame coding types;
1101
     * blocks encoding in these modes can only predict from other blocks
1102
     * that were encoded with these 1 of these 2 modes. */
1103
    static const unsigned char compatible_frame[9] = {
1104
        1,    /* MODE_INTER_NO_MV */
1105
        0,    /* MODE_INTRA */
1106
        1,    /* MODE_INTER_PLUS_MV */
1107
        1,    /* MODE_INTER_LAST_MV */
1108
        1,    /* MODE_INTER_PRIOR_MV */
1109
        2,    /* MODE_USING_GOLDEN */
1110
        2,    /* MODE_GOLDEN_MV */
1111
        1,    /* MODE_INTER_FOUR_MV */
1112
        3     /* MODE_COPY */
1113
    };
1114
    int current_frame_type;
1115

    
1116
    /* there is a last DC predictor for each of the 3 frame types */
1117
    short last_dc[3];
1118

    
1119
    int transform = 0;
1120

    
1121
    vul = vu = vur = vl = 0;
1122
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1123

    
1124
    /* for each fragment row... */
1125
    for (y = 0; y < fragment_height; y++) {
1126

    
1127
        /* for each fragment in a row... */
1128
        for (x = 0; x < fragment_width; x++, i++) {
1129

    
1130
            /* reverse prediction if this block was coded */
1131
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1132

    
1133
                current_frame_type =
1134
                    compatible_frame[s->all_fragments[i].coding_method];
1135

    
1136
                transform= 0;
1137
                if(x){
1138
                    l= i-1;
1139
                    vl = DC_COEFF(l);
1140
                    if(COMPATIBLE_FRAME(l))
1141
                        transform |= PL;
1142
                }
1143
                if(y){
1144
                    u= i-fragment_width;
1145
                    vu = DC_COEFF(u);
1146
                    if(COMPATIBLE_FRAME(u))
1147
                        transform |= PU;
1148
                    if(x){
1149
                        ul= i-fragment_width-1;
1150
                        vul = DC_COEFF(ul);
1151
                        if(COMPATIBLE_FRAME(ul))
1152
                            transform |= PUL;
1153
                    }
1154
                    if(x + 1 < fragment_width){
1155
                        ur= i-fragment_width+1;
1156
                        vur = DC_COEFF(ur);
1157
                        if(COMPATIBLE_FRAME(ur))
1158
                            transform |= PUR;
1159
                    }
1160
                }
1161

    
1162
                if (transform == 0) {
1163

    
1164
                    /* if there were no fragments to predict from, use last
1165
                     * DC saved */
1166
                    predicted_dc = last_dc[current_frame_type];
1167
                } else {
1168

    
1169
                    /* apply the appropriate predictor transform */
1170
                    predicted_dc =
1171
                        (predictor_transform[transform][0] * vul) +
1172
                        (predictor_transform[transform][1] * vu) +
1173
                        (predictor_transform[transform][2] * vur) +
1174
                        (predictor_transform[transform][3] * vl);
1175

    
1176
                    predicted_dc /= 128;
1177

    
1178
                    /* check for outranging on the [ul u l] and
1179
                     * [ul u ur l] predictors */
1180
                    if ((transform == 15) || (transform == 13)) {
1181
                        if (FFABS(predicted_dc - vu) > 128)
1182
                            predicted_dc = vu;
1183
                        else if (FFABS(predicted_dc - vl) > 128)
1184
                            predicted_dc = vl;
1185
                        else if (FFABS(predicted_dc - vul) > 128)
1186
                            predicted_dc = vul;
1187
                    }
1188
                }
1189

    
1190
                /* at long last, apply the predictor */
1191
                DC_COEFF(i) += predicted_dc;
1192
                /* save the DC */
1193
                last_dc[current_frame_type] = DC_COEFF(i);
1194
            }
1195
        }
1196
    }
1197
}
1198

    
1199
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1200
{
1201
    int x, y;
1202
    int *bounding_values= s->bounding_values_array+127;
1203

    
1204
    int width           = s->fragment_width[!!plane];
1205
    int height          = s->fragment_height[!!plane];
1206
    int fragment        = s->fragment_start        [plane] + ystart * width;
1207
    int stride          = s->current_frame.linesize[plane];
1208
    uint8_t *plane_data = s->current_frame.data    [plane];
1209
    if (!s->flipped_image) stride = -stride;
1210
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1211

    
1212
    for (y = ystart; y < yend; y++) {
1213

    
1214
        for (x = 0; x < width; x++) {
1215
            /* This code basically just deblocks on the edges of coded blocks.
1216
             * However, it has to be much more complicated because of the
1217
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1218
             * because some pixels get filtered twice. */
1219
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1220
            {
1221
                /* do not perform left edge filter for left columns frags */
1222
                if (x > 0) {
1223
                    s->dsp.vp3_h_loop_filter(
1224
                        plane_data + 8*x,
1225
                        stride, bounding_values);
1226
                }
1227

    
1228
                /* do not perform top edge filter for top row fragments */
1229
                if (y > 0) {
1230
                    s->dsp.vp3_v_loop_filter(
1231
                        plane_data + 8*x,
1232
                        stride, bounding_values);
1233
                }
1234

    
1235
                /* do not perform right edge filter for right column
1236
                 * fragments or if right fragment neighbor is also coded
1237
                 * in this frame (it will be filtered in next iteration) */
1238
                if ((x < width - 1) &&
1239
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1240
                    s->dsp.vp3_h_loop_filter(
1241
                        plane_data + 8*x + 8,
1242
                        stride, bounding_values);
1243
                }
1244

    
1245
                /* do not perform bottom edge filter for bottom row
1246
                 * fragments or if bottom fragment neighbor is also coded
1247
                 * in this frame (it will be filtered in the next row) */
1248
                if ((y < height - 1) &&
1249
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1250
                    s->dsp.vp3_v_loop_filter(
1251
                        plane_data + 8*x + 8*stride,
1252
                        stride, bounding_values);
1253
                }
1254
            }
1255

    
1256
            fragment++;
1257
        }
1258
        plane_data += 8*stride;
1259
    }
1260
}
1261

    
1262
/**
1263
 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1264
 * for the next block in coding order
1265
 */
1266
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1267
                              int plane, int inter, DCTELEM block[64])
1268
{
1269
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1270
    uint8_t *perm = s->scantable.permutated;
1271
    int i = 0;
1272

    
1273
    do {
1274
        int token = *s->dct_tokens[plane][i];
1275
        switch (token & 3) {
1276
        case 0: // EOB
1277
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1278
                s->dct_tokens[plane][i]++;
1279
            else
1280
                *s->dct_tokens[plane][i] = token & ~3;
1281
            goto end;
1282
        case 1: // zero run
1283
            s->dct_tokens[plane][i]++;
1284
            i += (token >> 2) & 0x7f;
1285
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1286
            i++;
1287
            break;
1288
        case 2: // coeff
1289
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1290
            s->dct_tokens[plane][i++]++;
1291
            break;
1292
        default:
1293
            av_log(s->avctx, AV_LOG_ERROR, "internal: invalid token type\n");
1294
            return i;
1295
        }
1296
    } while (i < 64);
1297
end:
1298
    // the actual DC+prediction is in the fragment structure
1299
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1300
    return i;
1301
}
1302

    
1303
/**
1304
 * called when all pixels up to row y are complete
1305
 */
1306
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1307
{
1308
    int h, cy;
1309
    int offset[4];
1310

    
1311
    if(s->avctx->draw_horiz_band==NULL)
1312
        return;
1313

    
1314
    h= y - s->last_slice_end;
1315
    y -= h;
1316

    
1317
    if (!s->flipped_image) {
1318
        if (y == 0)
1319
            h -= s->height - s->avctx->height;  // account for non-mod16
1320
        y = s->height - y - h;
1321
    }
1322

    
1323
    cy = y >> 1;
1324
    offset[0] = s->current_frame.linesize[0]*y;
1325
    offset[1] = s->current_frame.linesize[1]*cy;
1326
    offset[2] = s->current_frame.linesize[2]*cy;
1327
    offset[3] = 0;
1328

    
1329
    emms_c();
1330
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1331
    s->last_slice_end= y + h;
1332
}
1333

    
1334
/*
1335
 * Perform the final rendering for a particular slice of data.
1336
 * The slice number ranges from 0..(c_superblock_height - 1).
1337
 */
1338
static void render_slice(Vp3DecodeContext *s, int slice)
1339
{
1340
    int x, y, i, j;
1341
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1342
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1343
    int motion_halfpel_index;
1344
    uint8_t *motion_source;
1345
    int plane, first_pixel;
1346

    
1347
    if (slice >= s->c_superblock_height)
1348
        return;
1349

    
1350
    for (plane = 0; plane < 3; plane++) {
1351
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1352
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1353
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1354
        int stride            = s->current_frame.linesize[plane];
1355
        int plane_width       = s->width  >> (plane && s->chroma_x_shift);
1356
        int plane_height      = s->height >> (plane && s->chroma_y_shift);
1357

    
1358
        int sb_x, sb_y        = slice << (!plane && s->chroma_y_shift);
1359
        int slice_height      = sb_y + 1 + (!plane && s->chroma_y_shift);
1360
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1361

    
1362
        int fragment_width    = s->fragment_width[!!plane];
1363
        int fragment_height   = s->fragment_height[!!plane];
1364
        int fragment_start    = s->fragment_start[plane];
1365

    
1366
        if (!s->flipped_image) stride = -stride;
1367
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1368
            continue;
1369

    
1370

    
1371
        if(FFABS(stride) > 2048)
1372
            return; //various tables are fixed size
1373

    
1374
        /* for each superblock row in the slice (both of them)... */
1375
        for (; sb_y < slice_height; sb_y++) {
1376

    
1377
            /* for each superblock in a row... */
1378
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1379

    
1380
                /* for each block in a superblock... */
1381
                for (j = 0; j < 16; j++) {
1382
                    x = 4*sb_x + hilbert_offset[j][0];
1383
                    y = 4*sb_y + hilbert_offset[j][1];
1384

    
1385
                    i = fragment_start + y*fragment_width + x;
1386

    
1387
                    // bounds check
1388
                    if (x >= fragment_width || y >= fragment_height)
1389
                        continue;
1390

    
1391
                first_pixel = 8*y*stride + 8*x;
1392

    
1393
                /* transform if this block was coded */
1394
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1395
                    int intra = s->all_fragments[i].coding_method == MODE_INTRA;
1396

    
1397
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1398
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1399
                        motion_source= golden_plane;
1400
                    else
1401
                        motion_source= last_plane;
1402

    
1403
                    motion_source += first_pixel;
1404
                    motion_halfpel_index = 0;
1405

    
1406
                    /* sort out the motion vector if this fragment is coded
1407
                     * using a motion vector method */
1408
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1409
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1410
                        int src_x, src_y;
1411
                        motion_x = s->all_fragments[i].motion_x;
1412
                        motion_y = s->all_fragments[i].motion_y;
1413

    
1414
                        src_x= (motion_x>>1) + 8*x;
1415
                        src_y= (motion_y>>1) + 8*y;
1416

    
1417
                        motion_halfpel_index = motion_x & 0x01;
1418
                        motion_source += (motion_x >> 1);
1419

    
1420
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1421
                        motion_source += ((motion_y >> 1) * stride);
1422

    
1423
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1424
                            uint8_t *temp= s->edge_emu_buffer;
1425
                            if(stride<0) temp -= 9*stride;
1426
                            else temp += 9*stride;
1427

    
1428
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1429
                            motion_source= temp;
1430
                        }
1431
                    }
1432

    
1433

    
1434
                    /* first, take care of copying a block from either the
1435
                     * previous or the golden frame */
1436
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1437
                        /* Note, it is possible to implement all MC cases with
1438
                           put_no_rnd_pixels_l2 which would look more like the
1439
                           VP3 source but this would be slower as
1440
                           put_no_rnd_pixels_tab is better optimzed */
1441
                        if(motion_halfpel_index != 3){
1442
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1443
                                output_plane + first_pixel,
1444
                                motion_source, stride, 8);
1445
                        }else{
1446
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1447
                            s->dsp.put_no_rnd_pixels_l2[1](
1448
                                output_plane + first_pixel,
1449
                                motion_source - d,
1450
                                motion_source + stride + 1 + d,
1451
                                stride, 8);
1452
                        }
1453
                    }
1454

    
1455
                        s->dsp.clear_block(block);
1456
                        vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
1457

    
1458
                    /* invert DCT and place (or add) in final output */
1459

    
1460
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1461
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1462
                            block[0] += 128<<3;
1463
                        s->dsp.idct_put(
1464
                            output_plane + first_pixel,
1465
                            stride,
1466
                            block);
1467
                    } else {
1468
                        s->dsp.idct_add(
1469
                            output_plane + first_pixel,
1470
                            stride,
1471
                            block);
1472
                    }
1473
                } else {
1474

    
1475
                    /* copy directly from the previous frame */
1476
                    s->dsp.put_pixels_tab[1][0](
1477
                        output_plane + first_pixel,
1478
                        last_plane + first_pixel,
1479
                        stride, 8);
1480

    
1481
                }
1482
                }
1483
            }
1484

    
1485
            // Filter up to the last row in the superblock row
1486
            apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1487
        }
1488
    }
1489

    
1490
     /* this looks like a good place for slice dispatch... */
1491
     /* algorithm:
1492
      *   if (slice == s->macroblock_height - 1)
1493
      *     dispatch (both last slice & 2nd-to-last slice);
1494
      *   else if (slice > 0)
1495
      *     dispatch (slice - 1);
1496
      */
1497

    
1498
    vp3_draw_horiz_band(s, FFMIN(64*slice + 64-16, s->height-16));
1499
}
1500

    
1501
/*
1502
 * This is the ffmpeg/libavcodec API init function.
1503
 */
1504
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1505
{
1506
    Vp3DecodeContext *s = avctx->priv_data;
1507
    int i, inter, plane;
1508
    int c_width;
1509
    int c_height;
1510
    int y_fragment_count, c_fragment_count;
1511

    
1512
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1513
        s->version = 0;
1514
    else
1515
        s->version = 1;
1516

    
1517
    s->avctx = avctx;
1518
    s->width = FFALIGN(avctx->width, 16);
1519
    s->height = FFALIGN(avctx->height, 16);
1520
    if (avctx->pix_fmt == PIX_FMT_NONE)
1521
        avctx->pix_fmt = PIX_FMT_YUV420P;
1522
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1523
    if(avctx->idct_algo==FF_IDCT_AUTO)
1524
        avctx->idct_algo=FF_IDCT_VP3;
1525
    dsputil_init(&s->dsp, avctx);
1526

    
1527
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1528

    
1529
    /* initialize to an impossible value which will force a recalculation
1530
     * in the first frame decode */
1531
    for (i = 0; i < 3; i++)
1532
        s->qps[i] = -1;
1533

    
1534
    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1535

    
1536
    s->y_superblock_width = (s->width + 31) / 32;
1537
    s->y_superblock_height = (s->height + 31) / 32;
1538
    s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1539

    
1540
    /* work out the dimensions for the C planes */
1541
    c_width = s->width >> s->chroma_x_shift;
1542
    c_height = s->height >> s->chroma_y_shift;
1543
    s->c_superblock_width = (c_width + 31) / 32;
1544
    s->c_superblock_height = (c_height + 31) / 32;
1545
    s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1546

    
1547
    s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1548
    s->u_superblock_start = s->y_superblock_count;
1549
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1550
    s->superblock_coding = av_malloc(s->superblock_count);
1551

    
1552
    s->macroblock_width = (s->width + 15) / 16;
1553
    s->macroblock_height = (s->height + 15) / 16;
1554
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1555

    
1556
    s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1557
    s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1558
    s->fragment_width[1]  = s->fragment_width[0]  >> s->chroma_x_shift;
1559
    s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1560

    
1561
    /* fragment count covers all 8x8 blocks for all 3 planes */
1562
    y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
1563
    c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
1564
    s->fragment_count    = y_fragment_count + 2*c_fragment_count;
1565
    s->fragment_start[1] = y_fragment_count;
1566
    s->fragment_start[2] = y_fragment_count + c_fragment_count;
1567

    
1568
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1569
    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1570
    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1571
    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1572
        !s->coded_fragment_list[0]) {
1573
        vp3_decode_end(avctx);
1574
        return -1;
1575
    }
1576

    
1577
    if (!s->theora_tables)
1578
    {
1579
        for (i = 0; i < 64; i++) {
1580
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1581
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1582
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1583
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1584
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1585
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1586
        }
1587

    
1588
        for(inter=0; inter<2; inter++){
1589
            for(plane=0; plane<3; plane++){
1590
                s->qr_count[inter][plane]= 1;
1591
                s->qr_size [inter][plane][0]= 63;
1592
                s->qr_base [inter][plane][0]=
1593
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1594
            }
1595
        }
1596

    
1597
        /* init VLC tables */
1598
        for (i = 0; i < 16; i++) {
1599

    
1600
            /* DC histograms */
1601
            init_vlc(&s->dc_vlc[i], 5, 32,
1602
                &dc_bias[i][0][1], 4, 2,
1603
                &dc_bias[i][0][0], 4, 2, 0);
1604

    
1605
            /* group 1 AC histograms */
1606
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1607
                &ac_bias_0[i][0][1], 4, 2,
1608
                &ac_bias_0[i][0][0], 4, 2, 0);
1609

    
1610
            /* group 2 AC histograms */
1611
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1612
                &ac_bias_1[i][0][1], 4, 2,
1613
                &ac_bias_1[i][0][0], 4, 2, 0);
1614

    
1615
            /* group 3 AC histograms */
1616
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1617
                &ac_bias_2[i][0][1], 4, 2,
1618
                &ac_bias_2[i][0][0], 4, 2, 0);
1619

    
1620
            /* group 4 AC histograms */
1621
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1622
                &ac_bias_3[i][0][1], 4, 2,
1623
                &ac_bias_3[i][0][0], 4, 2, 0);
1624
        }
1625
    } else {
1626
        for (i = 0; i < 16; i++) {
1627

    
1628
            /* DC histograms */
1629
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1630
                &s->huffman_table[i][0][1], 4, 2,
1631
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1632
                goto vlc_fail;
1633

    
1634
            /* group 1 AC histograms */
1635
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1636
                &s->huffman_table[i+16][0][1], 4, 2,
1637
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1638
                goto vlc_fail;
1639

    
1640
            /* group 2 AC histograms */
1641
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1642
                &s->huffman_table[i+16*2][0][1], 4, 2,
1643
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1644
                goto vlc_fail;
1645

    
1646
            /* group 3 AC histograms */
1647
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1648
                &s->huffman_table[i+16*3][0][1], 4, 2,
1649
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1650
                goto vlc_fail;
1651

    
1652
            /* group 4 AC histograms */
1653
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1654
                &s->huffman_table[i+16*4][0][1], 4, 2,
1655
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1656
                goto vlc_fail;
1657
        }
1658
    }
1659

    
1660
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1661
        &superblock_run_length_vlc_table[0][1], 4, 2,
1662
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1663

    
1664
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1665
        &fragment_run_length_vlc_table[0][1], 4, 2,
1666
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1667

    
1668
    init_vlc(&s->mode_code_vlc, 3, 8,
1669
        &mode_code_vlc_table[0][1], 2, 1,
1670
        &mode_code_vlc_table[0][0], 2, 1, 0);
1671

    
1672
    init_vlc(&s->motion_vector_vlc, 6, 63,
1673
        &motion_vector_vlc_table[0][1], 2, 1,
1674
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1675

    
1676
    /* work out the block mapping tables */
1677
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1678
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1679
    if (!s->superblock_fragments || !s->macroblock_coding) {
1680
        vp3_decode_end(avctx);
1681
        return -1;
1682
    }
1683
    init_block_mapping(s);
1684

    
1685
    for (i = 0; i < 3; i++) {
1686
        s->current_frame.data[i] = NULL;
1687
        s->last_frame.data[i] = NULL;
1688
        s->golden_frame.data[i] = NULL;
1689
    }
1690

    
1691
    return 0;
1692

    
1693
vlc_fail:
1694
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1695
    return -1;
1696
}
1697

    
1698
/*
1699
 * This is the ffmpeg/libavcodec API frame decode function.
1700
 */
1701
static int vp3_decode_frame(AVCodecContext *avctx,
1702
                            void *data, int *data_size,
1703
                            AVPacket *avpkt)
1704
{
1705
    const uint8_t *buf = avpkt->data;
1706
    int buf_size = avpkt->size;
1707
    Vp3DecodeContext *s = avctx->priv_data;
1708
    GetBitContext gb;
1709
    static int counter = 0;
1710
    int i;
1711

    
1712
    init_get_bits(&gb, buf, buf_size * 8);
1713

    
1714
    if (s->theora && get_bits1(&gb))
1715
    {
1716
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1717
        return -1;
1718
    }
1719

    
1720
    s->keyframe = !get_bits1(&gb);
1721
    if (!s->theora)
1722
        skip_bits(&gb, 1);
1723
    for (i = 0; i < 3; i++)
1724
        s->last_qps[i] = s->qps[i];
1725

    
1726
    s->nqps=0;
1727
    do{
1728
        s->qps[s->nqps++]= get_bits(&gb, 6);
1729
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1730
    for (i = s->nqps; i < 3; i++)
1731
        s->qps[i] = -1;
1732

    
1733
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1734
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1735
            s->keyframe?"key":"", counter, s->qps[0]);
1736
    counter++;
1737

    
1738
    if (s->qps[0] != s->last_qps[0])
1739
        init_loop_filter(s);
1740

    
1741
    for (i = 0; i < s->nqps; i++)
1742
        // reinit all dequantizers if the first one changed, because
1743
        // the DC of the first quantizer must be used for all matrices
1744
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1745
            init_dequantizer(s, i);
1746

    
1747
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1748
        return buf_size;
1749

    
1750
    s->current_frame.reference = 3;
1751
    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
1752
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1753
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1754
        goto error;
1755
    }
1756

    
1757
    if (s->keyframe) {
1758
        if (!s->theora)
1759
        {
1760
            skip_bits(&gb, 4); /* width code */
1761
            skip_bits(&gb, 4); /* height code */
1762
            if (s->version)
1763
            {
1764
                s->version = get_bits(&gb, 5);
1765
                if (counter == 1)
1766
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1767
            }
1768
        }
1769
        if (s->version || s->theora)
1770
        {
1771
                if (get_bits1(&gb))
1772
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1773
            skip_bits(&gb, 2); /* reserved? */
1774
        }
1775
    } else {
1776
        if (!s->golden_frame.data[0]) {
1777
            av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1778

    
1779
            s->golden_frame.reference = 3;
1780
            s->golden_frame.pict_type = FF_I_TYPE;
1781
            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1782
                av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1783
                goto error;
1784
            }
1785
            s->last_frame = s->golden_frame;
1786
            s->last_frame.type = FF_BUFFER_TYPE_COPY;
1787
        }
1788
    }
1789

    
1790
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1791
    s->current_frame.qstride= 0;
1792

    
1793
    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1794

    
1795
    if (unpack_superblocks(s, &gb)){
1796
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1797
        goto error;
1798
    }
1799
    if (unpack_modes(s, &gb)){
1800
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1801
        goto error;
1802
    }
1803
    if (unpack_vectors(s, &gb)){
1804
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1805
        goto error;
1806
    }
1807
    if (unpack_block_qpis(s, &gb)){
1808
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1809
        goto error;
1810
    }
1811
    if (unpack_dct_coeffs(s, &gb)){
1812
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1813
        goto error;
1814
    }
1815

    
1816
    for (i = 0; i < 3; i++) {
1817
        int height = s->height >> (i && s->chroma_y_shift);
1818
        if (s->flipped_image)
1819
            s->data_offset[i] = 0;
1820
        else
1821
            s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1822
    }
1823

    
1824
    s->last_slice_end = 0;
1825
    for (i = 0; i < s->c_superblock_height; i++)
1826
        render_slice(s, i);
1827

    
1828
    // filter the last row
1829
    for (i = 0; i < 3; i++) {
1830
        int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1831
        apply_loop_filter(s, i, row, row+1);
1832
    }
1833
    vp3_draw_horiz_band(s, s->height);
1834

    
1835
    *data_size=sizeof(AVFrame);
1836
    *(AVFrame*)data= s->current_frame;
1837

    
1838
    /* release the last frame, if it is allocated and if it is not the
1839
     * golden frame */
1840
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1841
        avctx->release_buffer(avctx, &s->last_frame);
1842

    
1843
    /* shuffle frames (last = current) */
1844
    s->last_frame= s->current_frame;
1845

    
1846
    if (s->keyframe) {
1847
        if (s->golden_frame.data[0])
1848
            avctx->release_buffer(avctx, &s->golden_frame);
1849
        s->golden_frame = s->current_frame;
1850
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1851
    }
1852

    
1853
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1854

    
1855
    return buf_size;
1856

    
1857
error:
1858
    if (s->current_frame.data[0])
1859
        avctx->release_buffer(avctx, &s->current_frame);
1860
    return -1;
1861
}
1862

    
1863
/*
1864
 * This is the ffmpeg/libavcodec API module cleanup function.
1865
 */
1866
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1867
{
1868
    Vp3DecodeContext *s = avctx->priv_data;
1869
    int i;
1870

    
1871
    av_free(s->superblock_coding);
1872
    av_free(s->all_fragments);
1873
    av_free(s->coded_fragment_list[0]);
1874
    av_free(s->dct_tokens_base);
1875
    av_free(s->superblock_fragments);
1876
    av_free(s->macroblock_coding);
1877

    
1878
    for (i = 0; i < 16; i++) {
1879
        free_vlc(&s->dc_vlc[i]);
1880
        free_vlc(&s->ac_vlc_1[i]);
1881
        free_vlc(&s->ac_vlc_2[i]);
1882
        free_vlc(&s->ac_vlc_3[i]);
1883
        free_vlc(&s->ac_vlc_4[i]);
1884
    }
1885

    
1886
    free_vlc(&s->superblock_run_length_vlc);
1887
    free_vlc(&s->fragment_run_length_vlc);
1888
    free_vlc(&s->mode_code_vlc);
1889
    free_vlc(&s->motion_vector_vlc);
1890

    
1891
    /* release all frames */
1892
    if (s->golden_frame.data[0])
1893
        avctx->release_buffer(avctx, &s->golden_frame);
1894
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1895
        avctx->release_buffer(avctx, &s->last_frame);
1896
    /* no need to release the current_frame since it will always be pointing
1897
     * to the same frame as either the golden or last frame */
1898

    
1899
    return 0;
1900
}
1901

    
1902
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1903
{
1904
    Vp3DecodeContext *s = avctx->priv_data;
1905

    
1906
    if (get_bits1(gb)) {
1907
        int token;
1908
        if (s->entries >= 32) { /* overflow */
1909
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1910
            return -1;
1911
        }
1912
        token = get_bits(gb, 5);
1913
        //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);
1914
        s->huffman_table[s->hti][token][0] = s->hbits;
1915
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1916
        s->entries++;
1917
    }
1918
    else {
1919
        if (s->huff_code_size >= 32) {/* overflow */
1920
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1921
            return -1;
1922
        }
1923
        s->huff_code_size++;
1924
        s->hbits <<= 1;
1925
        if (read_huffman_tree(avctx, gb))
1926
            return -1;
1927
        s->hbits |= 1;
1928
        if (read_huffman_tree(avctx, gb))
1929
            return -1;
1930
        s->hbits >>= 1;
1931
        s->huff_code_size--;
1932
    }
1933
    return 0;
1934
}
1935

    
1936
#if CONFIG_THEORA_DECODER
1937
static const enum PixelFormat theora_pix_fmts[4] = {
1938
    PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
1939
};
1940

    
1941
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1942
{
1943
    Vp3DecodeContext *s = avctx->priv_data;
1944
    int visible_width, visible_height, colorspace;
1945

    
1946
    s->theora = get_bits_long(gb, 24);
1947
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1948

    
1949
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1950
    /* but previous versions have the image flipped relative to vp3 */
1951
    if (s->theora < 0x030200)
1952
    {
1953
        s->flipped_image = 1;
1954
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1955
    }
1956

    
1957
    visible_width  = s->width  = get_bits(gb, 16) << 4;
1958
    visible_height = s->height = get_bits(gb, 16) << 4;
1959

    
1960
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
1961
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
1962
        s->width= s->height= 0;
1963
        return -1;
1964
    }
1965

    
1966
    if (s->theora >= 0x030200) {
1967
        visible_width  = get_bits_long(gb, 24);
1968
        visible_height = get_bits_long(gb, 24);
1969

    
1970
        skip_bits(gb, 8); /* offset x */
1971
        skip_bits(gb, 8); /* offset y */
1972
    }
1973

    
1974
    skip_bits(gb, 32); /* fps numerator */
1975
    skip_bits(gb, 32); /* fps denumerator */
1976
    skip_bits(gb, 24); /* aspect numerator */
1977
    skip_bits(gb, 24); /* aspect denumerator */
1978

    
1979
    if (s->theora < 0x030200)
1980
        skip_bits(gb, 5); /* keyframe frequency force */
1981
    colorspace = get_bits(gb, 8);
1982
    skip_bits(gb, 24); /* bitrate */
1983

    
1984
    skip_bits(gb, 6); /* quality hint */
1985

    
1986
    if (s->theora >= 0x030200)
1987
    {
1988
        skip_bits(gb, 5); /* keyframe frequency force */
1989
        avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
1990
        skip_bits(gb, 3); /* reserved */
1991
    }
1992

    
1993
//    align_get_bits(gb);
1994

    
1995
    if (   visible_width  <= s->width  && visible_width  > s->width-16
1996
        && visible_height <= s->height && visible_height > s->height-16)
1997
        avcodec_set_dimensions(avctx, visible_width, visible_height);
1998
    else
1999
        avcodec_set_dimensions(avctx, s->width, s->height);
2000

    
2001
    if (colorspace == 1) {
2002
        avctx->color_primaries = AVCOL_PRI_BT470M;
2003
    } else if (colorspace == 2) {
2004
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2005
    }
2006
    if (colorspace == 1 || colorspace == 2) {
2007
        avctx->colorspace = AVCOL_SPC_BT470BG;
2008
        avctx->color_trc  = AVCOL_TRC_BT709;
2009
    }
2010

    
2011
    return 0;
2012
}
2013

    
2014
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2015
{
2016
    Vp3DecodeContext *s = avctx->priv_data;
2017
    int i, n, matrices, inter, plane;
2018

    
2019
    if (s->theora >= 0x030200) {
2020
        n = get_bits(gb, 3);
2021
        /* loop filter limit values table */
2022
        for (i = 0; i < 64; i++) {
2023
            s->filter_limit_values[i] = get_bits(gb, n);
2024
            if (s->filter_limit_values[i] > 127) {
2025
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2026
                s->filter_limit_values[i] = 127;
2027
            }
2028
        }
2029
    }
2030

    
2031
    if (s->theora >= 0x030200)
2032
        n = get_bits(gb, 4) + 1;
2033
    else
2034
        n = 16;
2035
    /* quality threshold table */
2036
    for (i = 0; i < 64; i++)
2037
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2038

    
2039
    if (s->theora >= 0x030200)
2040
        n = get_bits(gb, 4) + 1;
2041
    else
2042
        n = 16;
2043
    /* dc scale factor table */
2044
    for (i = 0; i < 64; i++)
2045
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2046

    
2047
    if (s->theora >= 0x030200)
2048
        matrices = get_bits(gb, 9) + 1;
2049
    else
2050
        matrices = 3;
2051

    
2052
    if(matrices > 384){
2053
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2054
        return -1;
2055
    }
2056

    
2057
    for(n=0; n<matrices; n++){
2058
        for (i = 0; i < 64; i++)
2059
            s->base_matrix[n][i]= get_bits(gb, 8);
2060
    }
2061

    
2062
    for (inter = 0; inter <= 1; inter++) {
2063
        for (plane = 0; plane <= 2; plane++) {
2064
            int newqr= 1;
2065
            if (inter || plane > 0)
2066
                newqr = get_bits1(gb);
2067
            if (!newqr) {
2068
                int qtj, plj;
2069
                if(inter && get_bits1(gb)){
2070
                    qtj = 0;
2071
                    plj = plane;
2072
                }else{
2073
                    qtj= (3*inter + plane - 1) / 3;
2074
                    plj= (plane + 2) % 3;
2075
                }
2076
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2077
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2078
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2079
            } else {
2080
                int qri= 0;
2081
                int qi = 0;
2082

    
2083
                for(;;){
2084
                    i= get_bits(gb, av_log2(matrices-1)+1);
2085
                    if(i>= matrices){
2086
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2087
                        return -1;
2088
                    }
2089
                    s->qr_base[inter][plane][qri]= i;
2090
                    if(qi >= 63)
2091
                        break;
2092
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2093
                    s->qr_size[inter][plane][qri++]= i;
2094
                    qi += i;
2095
                }
2096

    
2097
                if (qi > 63) {
2098
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2099
                    return -1;
2100
                }
2101
                s->qr_count[inter][plane]= qri;
2102
            }
2103
        }
2104
    }
2105

    
2106
    /* Huffman tables */
2107
    for (s->hti = 0; s->hti < 80; s->hti++) {
2108
        s->entries = 0;
2109
        s->huff_code_size = 1;
2110
        if (!get_bits1(gb)) {
2111
            s->hbits = 0;
2112
            if(read_huffman_tree(avctx, gb))
2113
                return -1;
2114
            s->hbits = 1;
2115
            if(read_huffman_tree(avctx, gb))
2116
                return -1;
2117
        }
2118
    }
2119

    
2120
    s->theora_tables = 1;
2121

    
2122
    return 0;
2123
}
2124

    
2125
static av_cold int theora_decode_init(AVCodecContext *avctx)
2126
{
2127
    Vp3DecodeContext *s = avctx->priv_data;
2128
    GetBitContext gb;
2129
    int ptype;
2130
    uint8_t *header_start[3];
2131
    int header_len[3];
2132
    int i;
2133

    
2134
    s->theora = 1;
2135

    
2136
    if (!avctx->extradata_size)
2137
    {
2138
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2139
        return -1;
2140
    }
2141

    
2142
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2143
                              42, header_start, header_len) < 0) {
2144
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2145
        return -1;
2146
    }
2147

    
2148
  for(i=0;i<3;i++) {
2149
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2150

    
2151
    ptype = get_bits(&gb, 8);
2152

    
2153
     if (!(ptype & 0x80))
2154
     {
2155
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2156
//        return -1;
2157
     }
2158

    
2159
    // FIXME: Check for this as well.
2160
    skip_bits_long(&gb, 6*8); /* "theora" */
2161

    
2162
    switch(ptype)
2163
    {
2164
        case 0x80:
2165
            theora_decode_header(avctx, &gb);
2166
                break;
2167
        case 0x81:
2168
// FIXME: is this needed? it breaks sometimes
2169
//            theora_decode_comments(avctx, gb);
2170
            break;
2171
        case 0x82:
2172
            if (theora_decode_tables(avctx, &gb))
2173
                return -1;
2174
            break;
2175
        default:
2176
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2177
            break;
2178
    }
2179
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2180
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2181
    if (s->theora < 0x030200)
2182
        break;
2183
  }
2184

    
2185
    return vp3_decode_init(avctx);
2186
}
2187

    
2188
AVCodec theora_decoder = {
2189
    "theora",
2190
    CODEC_TYPE_VIDEO,
2191
    CODEC_ID_THEORA,
2192
    sizeof(Vp3DecodeContext),
2193
    theora_decode_init,
2194
    NULL,
2195
    vp3_decode_end,
2196
    vp3_decode_frame,
2197
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2198
    NULL,
2199
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2200
};
2201
#endif
2202

    
2203
AVCodec vp3_decoder = {
2204
    "vp3",
2205
    CODEC_TYPE_VIDEO,
2206
    CODEC_ID_VP3,
2207
    sizeof(Vp3DecodeContext),
2208
    vp3_decode_init,
2209
    NULL,
2210
    vp3_decode_end,
2211
    vp3_decode_frame,
2212
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2213
    NULL,
2214
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2215
};