Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 60867312

History | View | Annotate | Download (76.5 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
    uint8_t qpi;
52
} Vp3Fragment;
53

    
54
#define SB_NOT_CODED        0
55
#define SB_PARTIALLY_CODED  1
56
#define SB_FULLY_CODED      2
57

    
58
// This is the maximum length of a single long bit run that can be encoded
59
// for superblock coding or block qps. Theora special-cases this to read a
60
// bit instead of flipping the current bit to allow for runs longer than 4129.
61
#define MAXIMUM_LONG_BIT_RUN 4129
62

    
63
#define MODE_INTER_NO_MV      0
64
#define MODE_INTRA            1
65
#define MODE_INTER_PLUS_MV    2
66
#define MODE_INTER_LAST_MV    3
67
#define MODE_INTER_PRIOR_LAST 4
68
#define MODE_USING_GOLDEN     5
69
#define MODE_GOLDEN_MV        6
70
#define MODE_INTER_FOURMV     7
71
#define CODING_MODE_COUNT     8
72

    
73
/* special internal mode */
74
#define MODE_COPY             8
75

    
76
/* There are 6 preset schemes, plus a free-form scheme */
77
static const int ModeAlphabet[6][CODING_MODE_COUNT] =
78
{
79
    /* scheme 1: Last motion vector dominates */
80
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
81
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
82
         MODE_INTRA,            MODE_USING_GOLDEN,
83
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
84

    
85
    /* scheme 2 */
86
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
87
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
88
         MODE_INTRA,            MODE_USING_GOLDEN,
89
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
90

    
91
    /* scheme 3 */
92
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
93
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
94
         MODE_INTRA,            MODE_USING_GOLDEN,
95
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
96

    
97
    /* scheme 4 */
98
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
99
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
100
         MODE_INTRA,            MODE_USING_GOLDEN,
101
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
102

    
103
    /* scheme 5: No motion vector dominates */
104
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
105
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
106
         MODE_INTRA,            MODE_USING_GOLDEN,
107
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
108

    
109
    /* scheme 6 */
110
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
111
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
112
         MODE_INTER_PLUS_MV,    MODE_INTRA,
113
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
114

    
115
};
116

    
117
static const uint8_t hilbert_offset[16][2] = {
118
    {0,0}, {1,0}, {1,1}, {0,1},
119
    {0,2}, {0,3}, {1,3}, {1,2},
120
    {2,2}, {2,3}, {3,3}, {3,2},
121
    {3,1}, {2,1}, {2,0}, {3,0}
122
};
123

    
124
#define MIN_DEQUANT_VAL 2
125

    
126
typedef struct Vp3DecodeContext {
127
    AVCodecContext *avctx;
128
    int theora, theora_tables;
129
    int version;
130
    int width, height;
131
    int chroma_x_shift, chroma_y_shift;
132
    AVFrame golden_frame;
133
    AVFrame last_frame;
134
    AVFrame current_frame;
135
    int keyframe;
136
    DSPContext dsp;
137
    int flipped_image;
138
    int last_slice_end;
139

    
140
    int qps[3];
141
    int nqps;
142
    int last_qps[3];
143

    
144
    int superblock_count;
145
    int y_superblock_width;
146
    int y_superblock_height;
147
    int y_superblock_count;
148
    int c_superblock_width;
149
    int c_superblock_height;
150
    int c_superblock_count;
151
    int u_superblock_start;
152
    int v_superblock_start;
153
    unsigned char *superblock_coding;
154

    
155
    int macroblock_count;
156
    int macroblock_width;
157
    int macroblock_height;
158

    
159
    int fragment_count;
160
    int fragment_width[2];
161
    int fragment_height[2];
162

    
163
    Vp3Fragment *all_fragments;
164
    int fragment_start[3];
165
    int data_offset[3];
166

    
167
    int8_t (*motion_val[2])[2];
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 && get_bits_left(gb) > 0) {
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
                   && get_bits_left(gb) > 0) {
414
                        current_run = get_vlc2(gb,
415
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
416
                        if (current_run == 34)
417
                            current_run += get_bits(gb, 12);
418

    
419
                for (j = 0; j < current_run; current_superblock++) {
420
                    if (current_superblock >= s->superblock_count) {
421
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
422
                        return -1;
423
                    }
424

    
425
                /* skip any superblocks already marked as partially coded */
426
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
427
                    s->superblock_coding[current_superblock] = 2*bit;
428
                    j++;
429
                }
430
                }
431
                superblocks_decoded += current_run;
432

    
433
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
434
                    bit = get_bits1(gb);
435
                else
436
                    bit ^= 1;
437
            }
438
        }
439

    
440
        /* if there were partial blocks, initialize bitstream for
441
         * unpacking fragment codings */
442
        if (num_partial_superblocks) {
443

    
444
            current_run = 0;
445
            bit = get_bits1(gb);
446
            /* toggle the bit because as soon as the first run length is
447
             * fetched the bit will be toggled again */
448
            bit ^= 1;
449
        }
450
    }
451

    
452
    /* figure out which fragments are coded; iterate through each
453
     * superblock (all planes) */
454
    s->total_num_coded_frags = 0;
455
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
456

    
457
    for (plane = 0; plane < 3; plane++) {
458
        int sb_start = superblock_starts[plane];
459
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
460
        int num_coded_frags = 0;
461

    
462
    for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
463

    
464
        /* iterate through all 16 fragments in a superblock */
465
        for (j = 0; j < 16; j++) {
466

    
467
            /* if the fragment is in bounds, check its coding status */
468
            current_fragment = s->superblock_fragments[i * 16 + j];
469
            if (current_fragment != -1) {
470
                int coded = s->superblock_coding[i];
471

    
472
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
473

    
474
                    /* fragment may or may not be coded; this is the case
475
                     * that cares about the fragment coding runs */
476
                    if (current_run-- == 0) {
477
                        bit ^= 1;
478
                        current_run = get_vlc2(gb,
479
                            s->fragment_run_length_vlc.table, 5, 2);
480
                    }
481
                    coded = bit;
482
                }
483

    
484
                    if (coded) {
485
                        /* default mode; actual mode will be decoded in
486
                         * the next phase */
487
                        s->all_fragments[current_fragment].coding_method =
488
                            MODE_INTER_NO_MV;
489
                        s->coded_fragment_list[plane][num_coded_frags++] =
490
                            current_fragment;
491
                    } else {
492
                        /* not coded; copy this fragment from the prior frame */
493
                        s->all_fragments[current_fragment].coding_method =
494
                            MODE_COPY;
495
                    }
496
            }
497
        }
498
    }
499
        s->total_num_coded_frags += num_coded_frags;
500
        for (i = 0; i < 64; i++)
501
            s->num_coded_frags[plane][i] = num_coded_frags;
502
        if (plane < 2)
503
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
504
    }
505
    return 0;
506
}
507

    
508
/*
509
 * This function unpacks all the coding mode data for individual macroblocks
510
 * from the bitstream.
511
 */
512
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
513
{
514
    int i, j, k, sb_x, sb_y;
515
    int scheme;
516
    int current_macroblock;
517
    int current_fragment;
518
    int coding_mode;
519
    int custom_mode_alphabet[CODING_MODE_COUNT];
520
    const int *alphabet;
521
    Vp3Fragment *frag;
522

    
523
    if (s->keyframe) {
524
        for (i = 0; i < s->fragment_count; i++)
525
            s->all_fragments[i].coding_method = MODE_INTRA;
526

    
527
    } else {
528

    
529
        /* fetch the mode coding scheme for this frame */
530
        scheme = get_bits(gb, 3);
531

    
532
        /* is it a custom coding scheme? */
533
        if (scheme == 0) {
534
            for (i = 0; i < 8; i++)
535
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
536
            for (i = 0; i < 8; i++)
537
                custom_mode_alphabet[get_bits(gb, 3)] = i;
538
            alphabet = custom_mode_alphabet;
539
        } else
540
            alphabet = ModeAlphabet[scheme-1];
541

    
542
        /* iterate through all of the macroblocks that contain 1 or more
543
         * coded fragments */
544
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
545
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
546
                if (get_bits_left(gb) <= 0)
547
                    return -1;
548

    
549
            for (j = 0; j < 4; j++) {
550
                int mb_x = 2*sb_x +   (j>>1);
551
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
552
                current_macroblock = mb_y * s->macroblock_width + mb_x;
553

    
554
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
555
                    continue;
556

    
557
#define BLOCK_X (2*mb_x + (k&1))
558
#define BLOCK_Y (2*mb_y + (k>>1))
559
                /* coding modes are only stored if the macroblock has at least one
560
                 * luma block coded, otherwise it must be INTER_NO_MV */
561
                for (k = 0; k < 4; k++) {
562
                    current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
563
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
564
                        break;
565
                }
566
                if (k == 4) {
567
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
568
                    continue;
569
                }
570

    
571
                /* mode 7 means get 3 bits for each coding mode */
572
                if (scheme == 7)
573
                    coding_mode = get_bits(gb, 3);
574
                else
575
                    coding_mode = alphabet
576
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
577

    
578
                s->macroblock_coding[current_macroblock] = coding_mode;
579
                for (k = 0; k < 4; k++) {
580
                    frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
581
                    if (frag->coding_method != MODE_COPY)
582
                        frag->coding_method = coding_mode;
583
                }
584

    
585
#define SET_CHROMA_MODES \
586
    if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
587
        frag[s->fragment_start[1]].coding_method = coding_mode;\
588
    if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
589
        frag[s->fragment_start[2]].coding_method = coding_mode;
590

    
591
                if (s->chroma_y_shift) {
592
                    frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
593
                    SET_CHROMA_MODES
594
                } else if (s->chroma_x_shift) {
595
                    frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
596
                    for (k = 0; k < 2; k++) {
597
                        SET_CHROMA_MODES
598
                        frag += s->fragment_width[1];
599
                    }
600
                } else {
601
                    for (k = 0; k < 4; k++) {
602
                        frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
603
                        SET_CHROMA_MODES
604
                    }
605
                }
606
            }
607
            }
608
        }
609
    }
610

    
611
    return 0;
612
}
613

    
614
/*
615
 * This function unpacks all the motion vectors for the individual
616
 * macroblocks from the bitstream.
617
 */
618
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
619
{
620
    int j, k, sb_x, sb_y;
621
    int coding_mode;
622
    int motion_x[4];
623
    int motion_y[4];
624
    int last_motion_x = 0;
625
    int last_motion_y = 0;
626
    int prior_last_motion_x = 0;
627
    int prior_last_motion_y = 0;
628
    int current_macroblock;
629
    int current_fragment;
630
    int frag;
631

    
632
    if (s->keyframe)
633
        return 0;
634

    
635
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
636
    coding_mode = get_bits1(gb);
637

    
638
    /* iterate through all of the macroblocks that contain 1 or more
639
     * coded fragments */
640
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
641
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
642
            if (get_bits_left(gb) <= 0)
643
                return -1;
644

    
645
        for (j = 0; j < 4; j++) {
646
            int mb_x = 2*sb_x +   (j>>1);
647
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
648
            current_macroblock = mb_y * s->macroblock_width + mb_x;
649

    
650
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
651
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
652
                continue;
653

    
654
            switch (s->macroblock_coding[current_macroblock]) {
655

    
656
            case MODE_INTER_PLUS_MV:
657
            case MODE_GOLDEN_MV:
658
                /* all 6 fragments use the same motion vector */
659
                if (coding_mode == 0) {
660
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
661
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
662
                } else {
663
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
664
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
665
                }
666

    
667
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
668
                if (s->macroblock_coding[current_macroblock] ==
669
                    MODE_INTER_PLUS_MV) {
670
                    prior_last_motion_x = last_motion_x;
671
                    prior_last_motion_y = last_motion_y;
672
                    last_motion_x = motion_x[0];
673
                    last_motion_y = motion_y[0];
674
                }
675
                break;
676

    
677
            case MODE_INTER_FOURMV:
678
                /* vector maintenance */
679
                prior_last_motion_x = last_motion_x;
680
                prior_last_motion_y = last_motion_y;
681

    
682
                /* fetch 4 vectors from the bitstream, one for each
683
                 * Y fragment, then average for the C fragment vectors */
684
                for (k = 0; k < 4; k++) {
685
                    current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
686
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
687
                        if (coding_mode == 0) {
688
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
689
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
690
                        } else {
691
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
692
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
693
                        }
694
                        last_motion_x = motion_x[k];
695
                        last_motion_y = motion_y[k];
696
                    } else {
697
                        motion_x[k] = 0;
698
                        motion_y[k] = 0;
699
                    }
700
                }
701
                break;
702

    
703
            case MODE_INTER_LAST_MV:
704
                /* all 6 fragments use the last motion vector */
705
                motion_x[0] = last_motion_x;
706
                motion_y[0] = last_motion_y;
707

    
708
                /* no vector maintenance (last vector remains the
709
                 * last vector) */
710
                break;
711

    
712
            case MODE_INTER_PRIOR_LAST:
713
                /* all 6 fragments use the motion vector prior to the
714
                 * last motion vector */
715
                motion_x[0] = prior_last_motion_x;
716
                motion_y[0] = prior_last_motion_y;
717

    
718
                /* vector maintenance */
719
                prior_last_motion_x = last_motion_x;
720
                prior_last_motion_y = last_motion_y;
721
                last_motion_x = motion_x[0];
722
                last_motion_y = motion_y[0];
723
                break;
724

    
725
            default:
726
                /* covers intra, inter without MV, golden without MV */
727
                motion_x[0] = 0;
728
                motion_y[0] = 0;
729

    
730
                /* no vector maintenance */
731
                break;
732
            }
733

    
734
            /* assign the motion vectors to the correct fragments */
735
            for (k = 0; k < 4; k++) {
736
                current_fragment =
737
                    BLOCK_Y*s->fragment_width[0] + BLOCK_X;
738
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
739
                    s->motion_val[0][current_fragment][0] = motion_x[k];
740
                    s->motion_val[0][current_fragment][1] = motion_y[k];
741
                } else {
742
                    s->motion_val[0][current_fragment][0] = motion_x[0];
743
                    s->motion_val[0][current_fragment][1] = motion_y[0];
744
                }
745
            }
746

    
747
            if (s->chroma_y_shift) {
748
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
749
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
750
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
751
                }
752
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
753
                motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
754
                frag = mb_y*s->fragment_width[1] + mb_x;
755
                s->motion_val[1][frag][0] = motion_x[0];
756
                s->motion_val[1][frag][1] = 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 = 2*mb_y*s->fragment_width[1] + mb_x;
771
                for (k = 0; k < 2; k++) {
772
                    s->motion_val[1][frag][0] = motion_x[k];
773
                    s->motion_val[1][frag][1] = motion_y[k];
774
                    frag += s->fragment_width[1];
775
                }
776
            } else {
777
                for (k = 0; k < 4; k++) {
778
                    frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
779
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
780
                        s->motion_val[1][frag][0] = motion_x[k];
781
                        s->motion_val[1][frag][1] = motion_y[k];
782
                    } else {
783
                        s->motion_val[1][frag][0] = motion_x[0];
784
                        s->motion_val[1][frag][1] = motion_y[0];
785
                    }
786
                }
787
            }
788
        }
789
        }
790
    }
791

    
792
    return 0;
793
}
794

    
795
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
796
{
797
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
798
    int num_blocks = s->total_num_coded_frags;
799

    
800
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
801
        i = blocks_decoded = num_blocks_at_qpi = 0;
802

    
803
        bit = get_bits1(gb);
804

    
805
        do {
806
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
807
            if (run_length == 34)
808
                run_length += get_bits(gb, 12);
809
            blocks_decoded += run_length;
810

    
811
            if (!bit)
812
                num_blocks_at_qpi += run_length;
813

    
814
            for (j = 0; j < run_length; i++) {
815
                if (i >= s->total_num_coded_frags)
816
                    return -1;
817

    
818
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
819
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
820
                    j++;
821
                }
822
            }
823

    
824
            if (run_length == MAXIMUM_LONG_BIT_RUN)
825
                bit = get_bits1(gb);
826
            else
827
                bit ^= 1;
828
        } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
829

    
830
        num_blocks -= num_blocks_at_qpi;
831
    }
832

    
833
    return 0;
834
}
835

    
836
/*
837
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
838
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
839
 * data. This function unpacks all the VLCs for either the Y plane or both
840
 * C planes, and is called for DC coefficients or different AC coefficient
841
 * levels (since different coefficient types require different VLC tables.
842
 *
843
 * This function returns a residual eob run. E.g, if a particular token gave
844
 * instructions to EOB the next 5 fragments and there were only 2 fragments
845
 * left in the current fragment range, 3 would be returned so that it could
846
 * be passed into the next call to this same function.
847
 */
848
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
849
                        VLC *table, int coeff_index,
850
                        int plane,
851
                        int eob_run)
852
{
853
    int i, j = 0;
854
    int token;
855
    int zero_run = 0;
856
    DCTELEM coeff = 0;
857
    int bits_to_get;
858
    int blocks_ended;
859
    int coeff_i = 0;
860
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
861
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
862

    
863
    /* local references to structure members to avoid repeated deferences */
864
    int *coded_fragment_list = s->coded_fragment_list[plane];
865
    Vp3Fragment *all_fragments = s->all_fragments;
866
    VLC_TYPE (*vlc_table)[2] = table->table;
867

    
868
    if (num_coeffs < 0)
869
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
870

    
871
    if (eob_run > num_coeffs) {
872
        coeff_i = blocks_ended = num_coeffs;
873
        eob_run -= num_coeffs;
874
    } else {
875
        coeff_i = blocks_ended = eob_run;
876
        eob_run = 0;
877
    }
878

    
879
    // insert fake EOB token to cover the split between planes or zzi
880
    if (blocks_ended)
881
        dct_tokens[j++] = blocks_ended << 2;
882

    
883
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
884
            /* decode a VLC into a token */
885
            token = get_vlc2(gb, vlc_table, 5, 3);
886
            /* use the token to get a zero run, a coefficient, and an eob run */
887
            if (token <= 6) {
888
                eob_run = eob_run_base[token];
889
                if (eob_run_get_bits[token])
890
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
891

    
892
                // record only the number of blocks ended in this plane,
893
                // any spill will be recorded in the next plane.
894
                if (eob_run > num_coeffs - coeff_i) {
895
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
896
                    blocks_ended   += num_coeffs - coeff_i;
897
                    eob_run        -= num_coeffs - coeff_i;
898
                    coeff_i         = num_coeffs;
899
                } else {
900
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
901
                    blocks_ended   += eob_run;
902
                    coeff_i        += eob_run;
903
                    eob_run = 0;
904
                }
905
            } else {
906
                bits_to_get = coeff_get_bits[token];
907
                if (bits_to_get)
908
                    bits_to_get = get_bits(gb, bits_to_get);
909
                coeff = coeff_tables[token][bits_to_get];
910

    
911
                zero_run = zero_run_base[token];
912
                if (zero_run_get_bits[token])
913
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
914

    
915
                if (zero_run) {
916
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
917
                } else {
918
                    // Save DC into the fragment structure. DC prediction is
919
                    // done in raster order, so the actual DC can't be in with
920
                    // other tokens. We still need the token in dct_tokens[]
921
                    // however, or else the structure collapses on itself.
922
                    if (!coeff_index)
923
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
924

    
925
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
926
                }
927

    
928
                if (coeff_index + zero_run > 64) {
929
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
930
                           " %d coeffs left\n", zero_run, 64-coeff_index);
931
                    zero_run = 64 - coeff_index;
932
                }
933

    
934
                // zero runs code multiple coefficients,
935
                // so don't try to decode coeffs for those higher levels
936
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
937
                    s->num_coded_frags[plane][i]--;
938
                coeff_i++;
939
            }
940
    }
941

    
942
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
943
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
944

    
945
    // decrement the number of blocks that have higher coeffecients for each
946
    // EOB run at this level
947
    if (blocks_ended)
948
        for (i = coeff_index+1; i < 64; i++)
949
            s->num_coded_frags[plane][i] -= blocks_ended;
950

    
951
    // setup the next buffer
952
    if (plane < 2)
953
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
954
    else if (coeff_index < 63)
955
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
956

    
957
    return eob_run;
958
}
959

    
960
static void reverse_dc_prediction(Vp3DecodeContext *s,
961
                                  int first_fragment,
962
                                  int fragment_width,
963
                                  int fragment_height);
964
/*
965
 * This function unpacks all of the DCT coefficient data from the
966
 * bitstream.
967
 */
968
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
969
{
970
    int i;
971
    int dc_y_table;
972
    int dc_c_table;
973
    int ac_y_table;
974
    int ac_c_table;
975
    int residual_eob_run = 0;
976
    VLC *y_tables[64];
977
    VLC *c_tables[64];
978

    
979
    s->dct_tokens[0][0] = s->dct_tokens_base;
980

    
981
    /* fetch the DC table indexes */
982
    dc_y_table = get_bits(gb, 4);
983
    dc_c_table = get_bits(gb, 4);
984

    
985
    /* unpack the Y plane DC coefficients */
986
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
987
        0, residual_eob_run);
988

    
989
    /* reverse prediction of the Y-plane DC coefficients */
990
    reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
991

    
992
    /* unpack the C plane DC coefficients */
993
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
994
        1, residual_eob_run);
995
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
996
        2, residual_eob_run);
997

    
998
    /* reverse prediction of the C-plane DC coefficients */
999
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1000
    {
1001
        reverse_dc_prediction(s, s->fragment_start[1],
1002
            s->fragment_width[1], s->fragment_height[1]);
1003
        reverse_dc_prediction(s, s->fragment_start[2],
1004
            s->fragment_width[1], s->fragment_height[1]);
1005
    }
1006

    
1007
    /* fetch the AC table indexes */
1008
    ac_y_table = get_bits(gb, 4);
1009
    ac_c_table = get_bits(gb, 4);
1010

    
1011
    /* build tables of AC VLC tables */
1012
    for (i = 1; i <= 5; i++) {
1013
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1014
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1015
    }
1016
    for (i = 6; i <= 14; i++) {
1017
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1018
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1019
    }
1020
    for (i = 15; i <= 27; i++) {
1021
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1022
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1023
    }
1024
    for (i = 28; i <= 63; i++) {
1025
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1026
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1027
    }
1028

    
1029
    /* decode all AC coefficents */
1030
    for (i = 1; i <= 63; i++) {
1031
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1032
                0, residual_eob_run);
1033

    
1034
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1035
                1, residual_eob_run);
1036
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1037
                2, residual_eob_run);
1038
    }
1039

    
1040
    return 0;
1041
}
1042

    
1043
/*
1044
 * This function reverses the DC prediction for each coded fragment in
1045
 * the frame. Much of this function is adapted directly from the original
1046
 * VP3 source code.
1047
 */
1048
#define COMPATIBLE_FRAME(x) \
1049
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1050
#define DC_COEFF(u) s->all_fragments[u].dc
1051

    
1052
static void reverse_dc_prediction(Vp3DecodeContext *s,
1053
                                  int first_fragment,
1054
                                  int fragment_width,
1055
                                  int fragment_height)
1056
{
1057

    
1058
#define PUL 8
1059
#define PU 4
1060
#define PUR 2
1061
#define PL 1
1062

    
1063
    int x, y;
1064
    int i = first_fragment;
1065

    
1066
    int predicted_dc;
1067

    
1068
    /* DC values for the left, up-left, up, and up-right fragments */
1069
    int vl, vul, vu, vur;
1070

    
1071
    /* indexes for the left, up-left, up, and up-right fragments */
1072
    int l, ul, u, ur;
1073

    
1074
    /*
1075
     * The 6 fields mean:
1076
     *   0: up-left multiplier
1077
     *   1: up multiplier
1078
     *   2: up-right multiplier
1079
     *   3: left multiplier
1080
     */
1081
    static const int predictor_transform[16][4] = {
1082
        {  0,  0,  0,  0},
1083
        {  0,  0,  0,128},        // PL
1084
        {  0,  0,128,  0},        // PUR
1085
        {  0,  0, 53, 75},        // PUR|PL
1086
        {  0,128,  0,  0},        // PU
1087
        {  0, 64,  0, 64},        // PU|PL
1088
        {  0,128,  0,  0},        // PU|PUR
1089
        {  0,  0, 53, 75},        // PU|PUR|PL
1090
        {128,  0,  0,  0},        // PUL
1091
        {  0,  0,  0,128},        // PUL|PL
1092
        { 64,  0, 64,  0},        // PUL|PUR
1093
        {  0,  0, 53, 75},        // PUL|PUR|PL
1094
        {  0,128,  0,  0},        // PUL|PU
1095
       {-104,116,  0,116},        // PUL|PU|PL
1096
        { 24, 80, 24,  0},        // PUL|PU|PUR
1097
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1098
    };
1099

    
1100
    /* This table shows which types of blocks can use other blocks for
1101
     * prediction. For example, INTRA is the only mode in this table to
1102
     * have a frame number of 0. That means INTRA blocks can only predict
1103
     * from other INTRA blocks. There are 2 golden frame coding types;
1104
     * blocks encoding in these modes can only predict from other blocks
1105
     * that were encoded with these 1 of these 2 modes. */
1106
    static const unsigned char compatible_frame[9] = {
1107
        1,    /* MODE_INTER_NO_MV */
1108
        0,    /* MODE_INTRA */
1109
        1,    /* MODE_INTER_PLUS_MV */
1110
        1,    /* MODE_INTER_LAST_MV */
1111
        1,    /* MODE_INTER_PRIOR_MV */
1112
        2,    /* MODE_USING_GOLDEN */
1113
        2,    /* MODE_GOLDEN_MV */
1114
        1,    /* MODE_INTER_FOUR_MV */
1115
        3     /* MODE_COPY */
1116
    };
1117
    int current_frame_type;
1118

    
1119
    /* there is a last DC predictor for each of the 3 frame types */
1120
    short last_dc[3];
1121

    
1122
    int transform = 0;
1123

    
1124
    vul = vu = vur = vl = 0;
1125
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1126

    
1127
    /* for each fragment row... */
1128
    for (y = 0; y < fragment_height; y++) {
1129

    
1130
        /* for each fragment in a row... */
1131
        for (x = 0; x < fragment_width; x++, i++) {
1132

    
1133
            /* reverse prediction if this block was coded */
1134
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1135

    
1136
                current_frame_type =
1137
                    compatible_frame[s->all_fragments[i].coding_method];
1138

    
1139
                transform= 0;
1140
                if(x){
1141
                    l= i-1;
1142
                    vl = DC_COEFF(l);
1143
                    if(COMPATIBLE_FRAME(l))
1144
                        transform |= PL;
1145
                }
1146
                if(y){
1147
                    u= i-fragment_width;
1148
                    vu = DC_COEFF(u);
1149
                    if(COMPATIBLE_FRAME(u))
1150
                        transform |= PU;
1151
                    if(x){
1152
                        ul= i-fragment_width-1;
1153
                        vul = DC_COEFF(ul);
1154
                        if(COMPATIBLE_FRAME(ul))
1155
                            transform |= PUL;
1156
                    }
1157
                    if(x + 1 < fragment_width){
1158
                        ur= i-fragment_width+1;
1159
                        vur = DC_COEFF(ur);
1160
                        if(COMPATIBLE_FRAME(ur))
1161
                            transform |= PUR;
1162
                    }
1163
                }
1164

    
1165
                if (transform == 0) {
1166

    
1167
                    /* if there were no fragments to predict from, use last
1168
                     * DC saved */
1169
                    predicted_dc = last_dc[current_frame_type];
1170
                } else {
1171

    
1172
                    /* apply the appropriate predictor transform */
1173
                    predicted_dc =
1174
                        (predictor_transform[transform][0] * vul) +
1175
                        (predictor_transform[transform][1] * vu) +
1176
                        (predictor_transform[transform][2] * vur) +
1177
                        (predictor_transform[transform][3] * vl);
1178

    
1179
                    predicted_dc /= 128;
1180

    
1181
                    /* check for outranging on the [ul u l] and
1182
                     * [ul u ur l] predictors */
1183
                    if ((transform == 15) || (transform == 13)) {
1184
                        if (FFABS(predicted_dc - vu) > 128)
1185
                            predicted_dc = vu;
1186
                        else if (FFABS(predicted_dc - vl) > 128)
1187
                            predicted_dc = vl;
1188
                        else if (FFABS(predicted_dc - vul) > 128)
1189
                            predicted_dc = vul;
1190
                    }
1191
                }
1192

    
1193
                /* at long last, apply the predictor */
1194
                DC_COEFF(i) += predicted_dc;
1195
                /* save the DC */
1196
                last_dc[current_frame_type] = DC_COEFF(i);
1197
            }
1198
        }
1199
    }
1200
}
1201

    
1202
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1203
{
1204
    int x, y;
1205
    int *bounding_values= s->bounding_values_array+127;
1206

    
1207
    int width           = s->fragment_width[!!plane];
1208
    int height          = s->fragment_height[!!plane];
1209
    int fragment        = s->fragment_start        [plane] + ystart * width;
1210
    int stride          = s->current_frame.linesize[plane];
1211
    uint8_t *plane_data = s->current_frame.data    [plane];
1212
    if (!s->flipped_image) stride = -stride;
1213
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1214

    
1215
    for (y = ystart; y < yend; y++) {
1216

    
1217
        for (x = 0; x < width; x++) {
1218
            /* This code basically just deblocks on the edges of coded blocks.
1219
             * However, it has to be much more complicated because of the
1220
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1221
             * because some pixels get filtered twice. */
1222
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1223
            {
1224
                /* do not perform left edge filter for left columns frags */
1225
                if (x > 0) {
1226
                    s->dsp.vp3_h_loop_filter(
1227
                        plane_data + 8*x,
1228
                        stride, bounding_values);
1229
                }
1230

    
1231
                /* do not perform top edge filter for top row fragments */
1232
                if (y > 0) {
1233
                    s->dsp.vp3_v_loop_filter(
1234
                        plane_data + 8*x,
1235
                        stride, bounding_values);
1236
                }
1237

    
1238
                /* do not perform right edge filter for right column
1239
                 * fragments or if right fragment neighbor is also coded
1240
                 * in this frame (it will be filtered in next iteration) */
1241
                if ((x < width - 1) &&
1242
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1243
                    s->dsp.vp3_h_loop_filter(
1244
                        plane_data + 8*x + 8,
1245
                        stride, bounding_values);
1246
                }
1247

    
1248
                /* do not perform bottom edge filter for bottom row
1249
                 * fragments or if bottom fragment neighbor is also coded
1250
                 * in this frame (it will be filtered in the next row) */
1251
                if ((y < height - 1) &&
1252
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1253
                    s->dsp.vp3_v_loop_filter(
1254
                        plane_data + 8*x + 8*stride,
1255
                        stride, bounding_values);
1256
                }
1257
            }
1258

    
1259
            fragment++;
1260
        }
1261
        plane_data += 8*stride;
1262
    }
1263
}
1264

    
1265
/**
1266
 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1267
 * for the next block in coding order
1268
 */
1269
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1270
                              int plane, int inter, DCTELEM block[64])
1271
{
1272
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1273
    uint8_t *perm = s->scantable.permutated;
1274
    int i = 0;
1275

    
1276
    do {
1277
        int token = *s->dct_tokens[plane][i];
1278
        switch (token & 3) {
1279
        case 0: // EOB
1280
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1281
                s->dct_tokens[plane][i]++;
1282
            else
1283
                *s->dct_tokens[plane][i] = token & ~3;
1284
            goto end;
1285
        case 1: // zero run
1286
            s->dct_tokens[plane][i]++;
1287
            i += (token >> 2) & 0x7f;
1288
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1289
            i++;
1290
            break;
1291
        case 2: // coeff
1292
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1293
            s->dct_tokens[plane][i++]++;
1294
            break;
1295
        default: // shouldn't happen
1296
            return i;
1297
        }
1298
    } while (i < 64);
1299
end:
1300
    // the actual DC+prediction is in the fragment structure
1301
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1302
    return i;
1303
}
1304

    
1305
/**
1306
 * called when all pixels up to row y are complete
1307
 */
1308
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1309
{
1310
    int h, cy;
1311
    int offset[4];
1312

    
1313
    if(s->avctx->draw_horiz_band==NULL)
1314
        return;
1315

    
1316
    h= y - s->last_slice_end;
1317
    y -= h;
1318

    
1319
    if (!s->flipped_image) {
1320
        if (y == 0)
1321
            h -= s->height - s->avctx->height;  // account for non-mod16
1322
        y = s->height - y - h;
1323
    }
1324

    
1325
    cy = y >> 1;
1326
    offset[0] = s->current_frame.linesize[0]*y;
1327
    offset[1] = s->current_frame.linesize[1]*cy;
1328
    offset[2] = s->current_frame.linesize[2]*cy;
1329
    offset[3] = 0;
1330

    
1331
    emms_c();
1332
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1333
    s->last_slice_end= y + h;
1334
}
1335

    
1336
/*
1337
 * Perform the final rendering for a particular slice of data.
1338
 * The slice number ranges from 0..(c_superblock_height - 1).
1339
 */
1340
static void render_slice(Vp3DecodeContext *s, int slice)
1341
{
1342
    int x, y, i, j;
1343
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1344
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1345
    int motion_halfpel_index;
1346
    uint8_t *motion_source;
1347
    int plane, first_pixel;
1348

    
1349
    if (slice >= s->c_superblock_height)
1350
        return;
1351

    
1352
    for (plane = 0; plane < 3; plane++) {
1353
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1354
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1355
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1356
        int stride            = s->current_frame.linesize[plane];
1357
        int plane_width       = s->width  >> (plane && s->chroma_x_shift);
1358
        int plane_height      = s->height >> (plane && s->chroma_y_shift);
1359
        int8_t (*motion_val)[2] = s->motion_val[!!plane];
1360

    
1361
        int sb_x, sb_y        = slice << (!plane && s->chroma_y_shift);
1362
        int slice_height      = sb_y + 1 + (!plane && s->chroma_y_shift);
1363
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1364

    
1365
        int fragment_width    = s->fragment_width[!!plane];
1366
        int fragment_height   = s->fragment_height[!!plane];
1367
        int fragment_start    = s->fragment_start[plane];
1368

    
1369
        if (!s->flipped_image) stride = -stride;
1370
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1371
            continue;
1372

    
1373

    
1374
        if(FFABS(stride) > 2048)
1375
            return; //various tables are fixed size
1376

    
1377
        /* for each superblock row in the slice (both of them)... */
1378
        for (; sb_y < slice_height; sb_y++) {
1379

    
1380
            /* for each superblock in a row... */
1381
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1382

    
1383
                /* for each block in a superblock... */
1384
                for (j = 0; j < 16; j++) {
1385
                    x = 4*sb_x + hilbert_offset[j][0];
1386
                    y = 4*sb_y + hilbert_offset[j][1];
1387

    
1388
                    i = fragment_start + y*fragment_width + x;
1389

    
1390
                    // bounds check
1391
                    if (x >= fragment_width || y >= fragment_height)
1392
                        continue;
1393

    
1394
                first_pixel = 8*y*stride + 8*x;
1395

    
1396
                /* transform if this block was coded */
1397
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1398
                    int intra = s->all_fragments[i].coding_method == MODE_INTRA;
1399

    
1400
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1401
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1402
                        motion_source= golden_plane;
1403
                    else
1404
                        motion_source= last_plane;
1405

    
1406
                    motion_source += first_pixel;
1407
                    motion_halfpel_index = 0;
1408

    
1409
                    /* sort out the motion vector if this fragment is coded
1410
                     * using a motion vector method */
1411
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1412
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1413
                        int src_x, src_y;
1414
                        motion_x = motion_val[y*fragment_width + x][0];
1415
                        motion_y = motion_val[y*fragment_width + x][1];
1416

    
1417
                        src_x= (motion_x>>1) + 8*x;
1418
                        src_y= (motion_y>>1) + 8*y;
1419

    
1420
                        motion_halfpel_index = motion_x & 0x01;
1421
                        motion_source += (motion_x >> 1);
1422

    
1423
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1424
                        motion_source += ((motion_y >> 1) * stride);
1425

    
1426
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1427
                            uint8_t *temp= s->edge_emu_buffer;
1428
                            if(stride<0) temp -= 9*stride;
1429
                            else temp += 9*stride;
1430

    
1431
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1432
                            motion_source= temp;
1433
                        }
1434
                    }
1435

    
1436

    
1437
                    /* first, take care of copying a block from either the
1438
                     * previous or the golden frame */
1439
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1440
                        /* Note, it is possible to implement all MC cases with
1441
                           put_no_rnd_pixels_l2 which would look more like the
1442
                           VP3 source but this would be slower as
1443
                           put_no_rnd_pixels_tab is better optimzed */
1444
                        if(motion_halfpel_index != 3){
1445
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1446
                                output_plane + first_pixel,
1447
                                motion_source, stride, 8);
1448
                        }else{
1449
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1450
                            s->dsp.put_no_rnd_pixels_l2[1](
1451
                                output_plane + first_pixel,
1452
                                motion_source - d,
1453
                                motion_source + stride + 1 + d,
1454
                                stride, 8);
1455
                        }
1456
                    }
1457

    
1458
                        s->dsp.clear_block(block);
1459
                        vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
1460

    
1461
                    /* invert DCT and place (or add) in final output */
1462

    
1463
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1464
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1465
                            block[0] += 128<<3;
1466
                        s->dsp.idct_put(
1467
                            output_plane + first_pixel,
1468
                            stride,
1469
                            block);
1470
                    } else {
1471
                        s->dsp.idct_add(
1472
                            output_plane + first_pixel,
1473
                            stride,
1474
                            block);
1475
                    }
1476
                } else {
1477

    
1478
                    /* copy directly from the previous frame */
1479
                    s->dsp.put_pixels_tab[1][0](
1480
                        output_plane + first_pixel,
1481
                        last_plane + first_pixel,
1482
                        stride, 8);
1483

    
1484
                }
1485
                }
1486
            }
1487

    
1488
            // Filter up to the last row in the superblock row
1489
            apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1490
        }
1491
    }
1492

    
1493
     /* this looks like a good place for slice dispatch... */
1494
     /* algorithm:
1495
      *   if (slice == s->macroblock_height - 1)
1496
      *     dispatch (both last slice & 2nd-to-last slice);
1497
      *   else if (slice > 0)
1498
      *     dispatch (slice - 1);
1499
      */
1500

    
1501
    vp3_draw_horiz_band(s, FFMIN(64*slice + 64-16, s->height-16));
1502
}
1503

    
1504
/*
1505
 * This is the ffmpeg/libavcodec API init function.
1506
 */
1507
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1508
{
1509
    Vp3DecodeContext *s = avctx->priv_data;
1510
    int i, inter, plane;
1511
    int c_width;
1512
    int c_height;
1513
    int y_fragment_count, c_fragment_count;
1514

    
1515
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1516
        s->version = 0;
1517
    else
1518
        s->version = 1;
1519

    
1520
    s->avctx = avctx;
1521
    s->width = FFALIGN(avctx->width, 16);
1522
    s->height = FFALIGN(avctx->height, 16);
1523
    if (avctx->pix_fmt == PIX_FMT_NONE)
1524
        avctx->pix_fmt = PIX_FMT_YUV420P;
1525
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1526
    if(avctx->idct_algo==FF_IDCT_AUTO)
1527
        avctx->idct_algo=FF_IDCT_VP3;
1528
    dsputil_init(&s->dsp, avctx);
1529

    
1530
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1531

    
1532
    /* initialize to an impossible value which will force a recalculation
1533
     * in the first frame decode */
1534
    for (i = 0; i < 3; i++)
1535
        s->qps[i] = -1;
1536

    
1537
    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1538

    
1539
    s->y_superblock_width = (s->width + 31) / 32;
1540
    s->y_superblock_height = (s->height + 31) / 32;
1541
    s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1542

    
1543
    /* work out the dimensions for the C planes */
1544
    c_width = s->width >> s->chroma_x_shift;
1545
    c_height = s->height >> s->chroma_y_shift;
1546
    s->c_superblock_width = (c_width + 31) / 32;
1547
    s->c_superblock_height = (c_height + 31) / 32;
1548
    s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1549

    
1550
    s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1551
    s->u_superblock_start = s->y_superblock_count;
1552
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1553
    s->superblock_coding = av_malloc(s->superblock_count);
1554

    
1555
    s->macroblock_width = (s->width + 15) / 16;
1556
    s->macroblock_height = (s->height + 15) / 16;
1557
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1558

    
1559
    s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1560
    s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1561
    s->fragment_width[1]  = s->fragment_width[0]  >> s->chroma_x_shift;
1562
    s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1563

    
1564
    /* fragment count covers all 8x8 blocks for all 3 planes */
1565
    y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
1566
    c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
1567
    s->fragment_count    = y_fragment_count + 2*c_fragment_count;
1568
    s->fragment_start[1] = y_fragment_count;
1569
    s->fragment_start[2] = y_fragment_count + c_fragment_count;
1570

    
1571
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1572
    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1573
    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1574
    s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1575
    s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1576

    
1577
    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1578
        !s->coded_fragment_list[0] || !s->motion_val[0] || !s->motion_val[1]) {
1579
        vp3_decode_end(avctx);
1580
        return -1;
1581
    }
1582

    
1583
    if (!s->theora_tables)
1584
    {
1585
        for (i = 0; i < 64; i++) {
1586
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1587
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1588
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1589
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1590
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1591
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1592
        }
1593

    
1594
        for(inter=0; inter<2; inter++){
1595
            for(plane=0; plane<3; plane++){
1596
                s->qr_count[inter][plane]= 1;
1597
                s->qr_size [inter][plane][0]= 63;
1598
                s->qr_base [inter][plane][0]=
1599
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1600
            }
1601
        }
1602

    
1603
        /* init VLC tables */
1604
        for (i = 0; i < 16; i++) {
1605

    
1606
            /* DC histograms */
1607
            init_vlc(&s->dc_vlc[i], 5, 32,
1608
                &dc_bias[i][0][1], 4, 2,
1609
                &dc_bias[i][0][0], 4, 2, 0);
1610

    
1611
            /* group 1 AC histograms */
1612
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1613
                &ac_bias_0[i][0][1], 4, 2,
1614
                &ac_bias_0[i][0][0], 4, 2, 0);
1615

    
1616
            /* group 2 AC histograms */
1617
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1618
                &ac_bias_1[i][0][1], 4, 2,
1619
                &ac_bias_1[i][0][0], 4, 2, 0);
1620

    
1621
            /* group 3 AC histograms */
1622
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1623
                &ac_bias_2[i][0][1], 4, 2,
1624
                &ac_bias_2[i][0][0], 4, 2, 0);
1625

    
1626
            /* group 4 AC histograms */
1627
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1628
                &ac_bias_3[i][0][1], 4, 2,
1629
                &ac_bias_3[i][0][0], 4, 2, 0);
1630
        }
1631
    } else {
1632
        for (i = 0; i < 16; i++) {
1633

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

    
1640
            /* group 1 AC histograms */
1641
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1642
                &s->huffman_table[i+16][0][1], 4, 2,
1643
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1644
                goto vlc_fail;
1645

    
1646
            /* group 2 AC histograms */
1647
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1648
                &s->huffman_table[i+16*2][0][1], 4, 2,
1649
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1650
                goto vlc_fail;
1651

    
1652
            /* group 3 AC histograms */
1653
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1654
                &s->huffman_table[i+16*3][0][1], 4, 2,
1655
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1656
                goto vlc_fail;
1657

    
1658
            /* group 4 AC histograms */
1659
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1660
                &s->huffman_table[i+16*4][0][1], 4, 2,
1661
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1662
                goto vlc_fail;
1663
        }
1664
    }
1665

    
1666
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1667
        &superblock_run_length_vlc_table[0][1], 4, 2,
1668
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1669

    
1670
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1671
        &fragment_run_length_vlc_table[0][1], 4, 2,
1672
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1673

    
1674
    init_vlc(&s->mode_code_vlc, 3, 8,
1675
        &mode_code_vlc_table[0][1], 2, 1,
1676
        &mode_code_vlc_table[0][0], 2, 1, 0);
1677

    
1678
    init_vlc(&s->motion_vector_vlc, 6, 63,
1679
        &motion_vector_vlc_table[0][1], 2, 1,
1680
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1681

    
1682
    /* work out the block mapping tables */
1683
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1684
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1685
    if (!s->superblock_fragments || !s->macroblock_coding) {
1686
        vp3_decode_end(avctx);
1687
        return -1;
1688
    }
1689
    init_block_mapping(s);
1690

    
1691
    for (i = 0; i < 3; i++) {
1692
        s->current_frame.data[i] = NULL;
1693
        s->last_frame.data[i] = NULL;
1694
        s->golden_frame.data[i] = NULL;
1695
    }
1696

    
1697
    return 0;
1698

    
1699
vlc_fail:
1700
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1701
    return -1;
1702
}
1703

    
1704
/*
1705
 * This is the ffmpeg/libavcodec API frame decode function.
1706
 */
1707
static int vp3_decode_frame(AVCodecContext *avctx,
1708
                            void *data, int *data_size,
1709
                            AVPacket *avpkt)
1710
{
1711
    const uint8_t *buf = avpkt->data;
1712
    int buf_size = avpkt->size;
1713
    Vp3DecodeContext *s = avctx->priv_data;
1714
    GetBitContext gb;
1715
    static int counter = 0;
1716
    int i;
1717

    
1718
    init_get_bits(&gb, buf, buf_size * 8);
1719

    
1720
    if (s->theora && get_bits1(&gb))
1721
    {
1722
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1723
        return -1;
1724
    }
1725

    
1726
    s->keyframe = !get_bits1(&gb);
1727
    if (!s->theora)
1728
        skip_bits(&gb, 1);
1729
    for (i = 0; i < 3; i++)
1730
        s->last_qps[i] = s->qps[i];
1731

    
1732
    s->nqps=0;
1733
    do{
1734
        s->qps[s->nqps++]= get_bits(&gb, 6);
1735
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1736
    for (i = s->nqps; i < 3; i++)
1737
        s->qps[i] = -1;
1738

    
1739
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1740
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1741
            s->keyframe?"key":"", counter, s->qps[0]);
1742
    counter++;
1743

    
1744
    if (s->qps[0] != s->last_qps[0])
1745
        init_loop_filter(s);
1746

    
1747
    for (i = 0; i < s->nqps; i++)
1748
        // reinit all dequantizers if the first one changed, because
1749
        // the DC of the first quantizer must be used for all matrices
1750
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1751
            init_dequantizer(s, i);
1752

    
1753
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1754
        return buf_size;
1755

    
1756
    s->current_frame.reference = 3;
1757
    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
1758
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1759
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1760
        goto error;
1761
    }
1762

    
1763
    if (s->keyframe) {
1764
        if (!s->theora)
1765
        {
1766
            skip_bits(&gb, 4); /* width code */
1767
            skip_bits(&gb, 4); /* height code */
1768
            if (s->version)
1769
            {
1770
                s->version = get_bits(&gb, 5);
1771
                if (counter == 1)
1772
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1773
            }
1774
        }
1775
        if (s->version || s->theora)
1776
        {
1777
                if (get_bits1(&gb))
1778
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1779
            skip_bits(&gb, 2); /* reserved? */
1780
        }
1781
    } else {
1782
        if (!s->golden_frame.data[0]) {
1783
            av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1784

    
1785
            s->golden_frame.reference = 3;
1786
            s->golden_frame.pict_type = FF_I_TYPE;
1787
            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1788
                av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1789
                goto error;
1790
            }
1791
            s->last_frame = s->golden_frame;
1792
            s->last_frame.type = FF_BUFFER_TYPE_COPY;
1793
        }
1794
    }
1795

    
1796
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1797
    s->current_frame.qstride= 0;
1798

    
1799
    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1800

    
1801
    if (unpack_superblocks(s, &gb)){
1802
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1803
        goto error;
1804
    }
1805
    if (unpack_modes(s, &gb)){
1806
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1807
        goto error;
1808
    }
1809
    if (unpack_vectors(s, &gb)){
1810
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1811
        goto error;
1812
    }
1813
    if (unpack_block_qpis(s, &gb)){
1814
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1815
        goto error;
1816
    }
1817
    if (unpack_dct_coeffs(s, &gb)){
1818
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1819
        goto error;
1820
    }
1821

    
1822
    for (i = 0; i < 3; i++) {
1823
        int height = s->height >> (i && s->chroma_y_shift);
1824
        if (s->flipped_image)
1825
            s->data_offset[i] = 0;
1826
        else
1827
            s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1828
    }
1829

    
1830
    s->last_slice_end = 0;
1831
    for (i = 0; i < s->c_superblock_height; i++)
1832
        render_slice(s, i);
1833

    
1834
    // filter the last row
1835
    for (i = 0; i < 3; i++) {
1836
        int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1837
        apply_loop_filter(s, i, row, row+1);
1838
    }
1839
    vp3_draw_horiz_band(s, s->height);
1840

    
1841
    *data_size=sizeof(AVFrame);
1842
    *(AVFrame*)data= s->current_frame;
1843

    
1844
    /* release the last frame, if it is allocated and if it is not the
1845
     * golden frame */
1846
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1847
        avctx->release_buffer(avctx, &s->last_frame);
1848

    
1849
    /* shuffle frames (last = current) */
1850
    s->last_frame= s->current_frame;
1851

    
1852
    if (s->keyframe) {
1853
        if (s->golden_frame.data[0])
1854
            avctx->release_buffer(avctx, &s->golden_frame);
1855
        s->golden_frame = s->current_frame;
1856
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1857
    }
1858

    
1859
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1860

    
1861
    return buf_size;
1862

    
1863
error:
1864
    if (s->current_frame.data[0])
1865
        avctx->release_buffer(avctx, &s->current_frame);
1866
    return -1;
1867
}
1868

    
1869
/*
1870
 * This is the ffmpeg/libavcodec API module cleanup function.
1871
 */
1872
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1873
{
1874
    Vp3DecodeContext *s = avctx->priv_data;
1875
    int i;
1876

    
1877
    av_free(s->superblock_coding);
1878
    av_free(s->all_fragments);
1879
    av_free(s->coded_fragment_list[0]);
1880
    av_free(s->dct_tokens_base);
1881
    av_free(s->superblock_fragments);
1882
    av_free(s->macroblock_coding);
1883
    av_free(s->motion_val[0]);
1884
    av_free(s->motion_val[1]);
1885

    
1886
    for (i = 0; i < 16; i++) {
1887
        free_vlc(&s->dc_vlc[i]);
1888
        free_vlc(&s->ac_vlc_1[i]);
1889
        free_vlc(&s->ac_vlc_2[i]);
1890
        free_vlc(&s->ac_vlc_3[i]);
1891
        free_vlc(&s->ac_vlc_4[i]);
1892
    }
1893

    
1894
    free_vlc(&s->superblock_run_length_vlc);
1895
    free_vlc(&s->fragment_run_length_vlc);
1896
    free_vlc(&s->mode_code_vlc);
1897
    free_vlc(&s->motion_vector_vlc);
1898

    
1899
    /* release all frames */
1900
    if (s->golden_frame.data[0])
1901
        avctx->release_buffer(avctx, &s->golden_frame);
1902
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1903
        avctx->release_buffer(avctx, &s->last_frame);
1904
    /* no need to release the current_frame since it will always be pointing
1905
     * to the same frame as either the golden or last frame */
1906

    
1907
    return 0;
1908
}
1909

    
1910
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1911
{
1912
    Vp3DecodeContext *s = avctx->priv_data;
1913

    
1914
    if (get_bits1(gb)) {
1915
        int token;
1916
        if (s->entries >= 32) { /* overflow */
1917
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1918
            return -1;
1919
        }
1920
        token = get_bits(gb, 5);
1921
        //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);
1922
        s->huffman_table[s->hti][token][0] = s->hbits;
1923
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1924
        s->entries++;
1925
    }
1926
    else {
1927
        if (s->huff_code_size >= 32) {/* overflow */
1928
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1929
            return -1;
1930
        }
1931
        s->huff_code_size++;
1932
        s->hbits <<= 1;
1933
        if (read_huffman_tree(avctx, gb))
1934
            return -1;
1935
        s->hbits |= 1;
1936
        if (read_huffman_tree(avctx, gb))
1937
            return -1;
1938
        s->hbits >>= 1;
1939
        s->huff_code_size--;
1940
    }
1941
    return 0;
1942
}
1943

    
1944
#if CONFIG_THEORA_DECODER
1945
static const enum PixelFormat theora_pix_fmts[4] = {
1946
    PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
1947
};
1948

    
1949
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1950
{
1951
    Vp3DecodeContext *s = avctx->priv_data;
1952
    int visible_width, visible_height, colorspace;
1953

    
1954
    s->theora = get_bits_long(gb, 24);
1955
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1956

    
1957
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1958
    /* but previous versions have the image flipped relative to vp3 */
1959
    if (s->theora < 0x030200)
1960
    {
1961
        s->flipped_image = 1;
1962
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1963
    }
1964

    
1965
    visible_width  = s->width  = get_bits(gb, 16) << 4;
1966
    visible_height = s->height = get_bits(gb, 16) << 4;
1967

    
1968
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
1969
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
1970
        s->width= s->height= 0;
1971
        return -1;
1972
    }
1973

    
1974
    if (s->theora >= 0x030200) {
1975
        visible_width  = get_bits_long(gb, 24);
1976
        visible_height = get_bits_long(gb, 24);
1977

    
1978
        skip_bits(gb, 8); /* offset x */
1979
        skip_bits(gb, 8); /* offset y */
1980
    }
1981

    
1982
    skip_bits(gb, 32); /* fps numerator */
1983
    skip_bits(gb, 32); /* fps denumerator */
1984
    skip_bits(gb, 24); /* aspect numerator */
1985
    skip_bits(gb, 24); /* aspect denumerator */
1986

    
1987
    if (s->theora < 0x030200)
1988
        skip_bits(gb, 5); /* keyframe frequency force */
1989
    colorspace = get_bits(gb, 8);
1990
    skip_bits(gb, 24); /* bitrate */
1991

    
1992
    skip_bits(gb, 6); /* quality hint */
1993

    
1994
    if (s->theora >= 0x030200)
1995
    {
1996
        skip_bits(gb, 5); /* keyframe frequency force */
1997
        avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
1998
        skip_bits(gb, 3); /* reserved */
1999
    }
2000

    
2001
//    align_get_bits(gb);
2002

    
2003
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2004
        && visible_height <= s->height && visible_height > s->height-16)
2005
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2006
    else
2007
        avcodec_set_dimensions(avctx, s->width, s->height);
2008

    
2009
    if (colorspace == 1) {
2010
        avctx->color_primaries = AVCOL_PRI_BT470M;
2011
    } else if (colorspace == 2) {
2012
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2013
    }
2014
    if (colorspace == 1 || colorspace == 2) {
2015
        avctx->colorspace = AVCOL_SPC_BT470BG;
2016
        avctx->color_trc  = AVCOL_TRC_BT709;
2017
    }
2018

    
2019
    return 0;
2020
}
2021

    
2022
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2023
{
2024
    Vp3DecodeContext *s = avctx->priv_data;
2025
    int i, n, matrices, inter, plane;
2026

    
2027
    if (s->theora >= 0x030200) {
2028
        n = get_bits(gb, 3);
2029
        /* loop filter limit values table */
2030
        for (i = 0; i < 64; i++) {
2031
            s->filter_limit_values[i] = get_bits(gb, n);
2032
            if (s->filter_limit_values[i] > 127) {
2033
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2034
                s->filter_limit_values[i] = 127;
2035
            }
2036
        }
2037
    }
2038

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

    
2047
    if (s->theora >= 0x030200)
2048
        n = get_bits(gb, 4) + 1;
2049
    else
2050
        n = 16;
2051
    /* dc scale factor table */
2052
    for (i = 0; i < 64; i++)
2053
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2054

    
2055
    if (s->theora >= 0x030200)
2056
        matrices = get_bits(gb, 9) + 1;
2057
    else
2058
        matrices = 3;
2059

    
2060
    if(matrices > 384){
2061
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2062
        return -1;
2063
    }
2064

    
2065
    for(n=0; n<matrices; n++){
2066
        for (i = 0; i < 64; i++)
2067
            s->base_matrix[n][i]= get_bits(gb, 8);
2068
    }
2069

    
2070
    for (inter = 0; inter <= 1; inter++) {
2071
        for (plane = 0; plane <= 2; plane++) {
2072
            int newqr= 1;
2073
            if (inter || plane > 0)
2074
                newqr = get_bits1(gb);
2075
            if (!newqr) {
2076
                int qtj, plj;
2077
                if(inter && get_bits1(gb)){
2078
                    qtj = 0;
2079
                    plj = plane;
2080
                }else{
2081
                    qtj= (3*inter + plane - 1) / 3;
2082
                    plj= (plane + 2) % 3;
2083
                }
2084
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2085
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2086
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2087
            } else {
2088
                int qri= 0;
2089
                int qi = 0;
2090

    
2091
                for(;;){
2092
                    i= get_bits(gb, av_log2(matrices-1)+1);
2093
                    if(i>= matrices){
2094
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2095
                        return -1;
2096
                    }
2097
                    s->qr_base[inter][plane][qri]= i;
2098
                    if(qi >= 63)
2099
                        break;
2100
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2101
                    s->qr_size[inter][plane][qri++]= i;
2102
                    qi += i;
2103
                }
2104

    
2105
                if (qi > 63) {
2106
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2107
                    return -1;
2108
                }
2109
                s->qr_count[inter][plane]= qri;
2110
            }
2111
        }
2112
    }
2113

    
2114
    /* Huffman tables */
2115
    for (s->hti = 0; s->hti < 80; s->hti++) {
2116
        s->entries = 0;
2117
        s->huff_code_size = 1;
2118
        if (!get_bits1(gb)) {
2119
            s->hbits = 0;
2120
            if(read_huffman_tree(avctx, gb))
2121
                return -1;
2122
            s->hbits = 1;
2123
            if(read_huffman_tree(avctx, gb))
2124
                return -1;
2125
        }
2126
    }
2127

    
2128
    s->theora_tables = 1;
2129

    
2130
    return 0;
2131
}
2132

    
2133
static av_cold int theora_decode_init(AVCodecContext *avctx)
2134
{
2135
    Vp3DecodeContext *s = avctx->priv_data;
2136
    GetBitContext gb;
2137
    int ptype;
2138
    uint8_t *header_start[3];
2139
    int header_len[3];
2140
    int i;
2141

    
2142
    s->theora = 1;
2143

    
2144
    if (!avctx->extradata_size)
2145
    {
2146
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2147
        return -1;
2148
    }
2149

    
2150
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2151
                              42, header_start, header_len) < 0) {
2152
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2153
        return -1;
2154
    }
2155

    
2156
  for(i=0;i<3;i++) {
2157
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2158

    
2159
    ptype = get_bits(&gb, 8);
2160

    
2161
     if (!(ptype & 0x80))
2162
     {
2163
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2164
//        return -1;
2165
     }
2166

    
2167
    // FIXME: Check for this as well.
2168
    skip_bits_long(&gb, 6*8); /* "theora" */
2169

    
2170
    switch(ptype)
2171
    {
2172
        case 0x80:
2173
            theora_decode_header(avctx, &gb);
2174
                break;
2175
        case 0x81:
2176
// FIXME: is this needed? it breaks sometimes
2177
//            theora_decode_comments(avctx, gb);
2178
            break;
2179
        case 0x82:
2180
            if (theora_decode_tables(avctx, &gb))
2181
                return -1;
2182
            break;
2183
        default:
2184
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2185
            break;
2186
    }
2187
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2188
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2189
    if (s->theora < 0x030200)
2190
        break;
2191
  }
2192

    
2193
    return vp3_decode_init(avctx);
2194
}
2195

    
2196
AVCodec theora_decoder = {
2197
    "theora",
2198
    AVMEDIA_TYPE_VIDEO,
2199
    CODEC_ID_THEORA,
2200
    sizeof(Vp3DecodeContext),
2201
    theora_decode_init,
2202
    NULL,
2203
    vp3_decode_end,
2204
    vp3_decode_frame,
2205
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2206
    NULL,
2207
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2208
};
2209
#endif
2210

    
2211
AVCodec vp3_decoder = {
2212
    "vp3",
2213
    AVMEDIA_TYPE_VIDEO,
2214
    CODEC_ID_VP3,
2215
    sizeof(Vp3DecodeContext),
2216
    vp3_decode_init,
2217
    NULL,
2218
    vp3_decode_end,
2219
    vp3_decode_frame,
2220
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2221
    NULL,
2222
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2223
};