Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ a16389c1

History | View | Annotate | Download (74.6 KB)

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

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

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

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

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

    
43
#define FRAGMENT_PIXELS 8
44

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

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

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

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

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

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

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

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

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

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

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

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

    
117
};
118

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

    
126
#define MIN_DEQUANT_VAL 2
127

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

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

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

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

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

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

    
168
    ScanTable scantable;
169

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
266
    for (plane = 0; plane < 3; plane++) {
267
        int sb_width    = plane ? s->c_superblock_width  : s->y_superblock_width;
268
        int sb_height   = plane ? s->c_superblock_height : s->y_superblock_height;
269
        int frag_width  = s->fragment_width  >> !!plane;
270
        int frag_height = s->fragment_height >> !!plane;
271

    
272
        for (sb_y = 0; sb_y < sb_height; sb_y++)
273
            for (sb_x = 0; sb_x < sb_width; sb_x++)
274
                for (i = 0; i < 16; i++) {
275
                    x = 4*sb_x + hilbert_offset[i][0];
276
                    y = 4*sb_y + hilbert_offset[i][1];
277

    
278
                    if (x < frag_width && y < frag_height)
279
                        s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
280
                    else
281
                        s->superblock_fragments[j++] = -1;
282
                }
283
    }
284

    
285
    return 0;  /* successful path out */
286
}
287

    
288
/*
289
 * This function wipes out all of the fragment data.
290
 */
291
static void init_frame(Vp3DecodeContext *s, GetBitContext *gb)
292
{
293
    int i;
294

    
295
    /* zero out all of the fragment information */
296
    for (i = 0; i < s->fragment_count; i++) {
297
        s->all_fragments[i].motion_x = 127;
298
        s->all_fragments[i].motion_y = 127;
299
        s->all_fragments[i].dc = 0;
300
        s->all_fragments[i].qpi = 0;
301
    }
302
}
303

    
304
/*
305
 * This function sets up the dequantization tables used for a particular
306
 * frame.
307
 */
308
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
309
{
310
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
311
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
312
    int i, plane, inter, qri, bmi, bmj, qistart;
313

    
314
    for(inter=0; inter<2; inter++){
315
        for(plane=0; plane<3; plane++){
316
            int sum=0;
317
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
318
                sum+= s->qr_size[inter][plane][qri];
319
                if(s->qps[qpi] <= sum)
320
                    break;
321
            }
322
            qistart= sum - s->qr_size[inter][plane][qri];
323
            bmi= s->qr_base[inter][plane][qri  ];
324
            bmj= s->qr_base[inter][plane][qri+1];
325
            for(i=0; i<64; i++){
326
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
327
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
328
                            + s->qr_size[inter][plane][qri])
329
                           / (2*s->qr_size[inter][plane][qri]);
330

    
331
                int qmin= 8<<(inter + !i);
332
                int qscale= i ? ac_scale_factor : dc_scale_factor;
333

    
334
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
335
            }
336
            // all DC coefficients use the same quant so as not to interfere with DC prediction
337
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
338
        }
339
    }
340

    
341
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
342
}
343

    
344
/*
345
 * This function initializes the loop filter boundary limits if the frame's
346
 * quality index is different from the previous frame's.
347
 *
348
 * The filter_limit_values may not be larger than 127.
349
 */
350
static void init_loop_filter(Vp3DecodeContext *s)
351
{
352
    int *bounding_values= s->bounding_values_array+127;
353
    int filter_limit;
354
    int x;
355
    int value;
356

    
357
    filter_limit = s->filter_limit_values[s->qps[0]];
358

    
359
    /* set up the bounding values */
360
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
361
    for (x = 0; x < filter_limit; x++) {
362
        bounding_values[-x] = -x;
363
        bounding_values[x] = x;
364
    }
365
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
366
        bounding_values[ x] =  value;
367
        bounding_values[-x] = -value;
368
    }
369
    if (value)
370
        bounding_values[128] = value;
371
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
372
}
373

    
374
/*
375
 * This function unpacks all of the superblock/macroblock/fragment coding
376
 * information from the bitstream.
377
 */
378
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
379
{
380
    int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
381
    int bit = 0;
382
    int current_superblock = 0;
383
    int current_run = 0;
384
    int num_partial_superblocks = 0;
385

    
386
    int i, j;
387
    int current_fragment;
388
    int plane;
389

    
390
    if (s->keyframe) {
391
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
392

    
393
    } else {
394

    
395
        /* unpack the list of partially-coded superblocks */
396
        bit = get_bits1(gb);
397
        while (current_superblock < s->superblock_count) {
398
                current_run = get_vlc2(gb,
399
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
400
                if (current_run == 34)
401
                    current_run += get_bits(gb, 12);
402

    
403
            if (current_superblock + current_run > s->superblock_count) {
404
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
405
                return -1;
406
            }
407

    
408
            memset(s->superblock_coding + current_superblock, bit, current_run);
409

    
410
            current_superblock += current_run;
411
            if (bit)
412
                num_partial_superblocks += current_run;
413

    
414
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
415
                bit = get_bits1(gb);
416
            else
417
                bit ^= 1;
418
        }
419

    
420
        /* unpack the list of fully coded superblocks if any of the blocks were
421
         * not marked as partially coded in the previous step */
422
        if (num_partial_superblocks < s->superblock_count) {
423
            int superblocks_decoded = 0;
424

    
425
            current_superblock = 0;
426
            bit = get_bits1(gb);
427
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks) {
428
                        current_run = get_vlc2(gb,
429
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
430
                        if (current_run == 34)
431
                            current_run += get_bits(gb, 12);
432

    
433
                for (j = 0; j < current_run; current_superblock++) {
434
                    if (current_superblock >= s->superblock_count) {
435
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
436
                        return -1;
437
                    }
438

    
439
                /* skip any superblocks already marked as partially coded */
440
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
441
                    s->superblock_coding[current_superblock] = 2*bit;
442
                    j++;
443
                }
444
                }
445
                superblocks_decoded += current_run;
446

    
447
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
448
                    bit = get_bits1(gb);
449
                else
450
                    bit ^= 1;
451
            }
452
        }
453

    
454
        /* if there were partial blocks, initialize bitstream for
455
         * unpacking fragment codings */
456
        if (num_partial_superblocks) {
457

    
458
            current_run = 0;
459
            bit = get_bits1(gb);
460
            /* toggle the bit because as soon as the first run length is
461
             * fetched the bit will be toggled again */
462
            bit ^= 1;
463
        }
464
    }
465

    
466
    /* figure out which fragments are coded; iterate through each
467
     * superblock (all planes) */
468
    s->total_num_coded_frags = 0;
469
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
470

    
471
    for (plane = 0; plane < 3; plane++) {
472
        int sb_start = superblock_starts[plane];
473
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
474
        int num_coded_frags = 0;
475

    
476
    for (i = sb_start; i < sb_end; i++) {
477

    
478
        /* iterate through all 16 fragments in a superblock */
479
        for (j = 0; j < 16; j++) {
480

    
481
            /* if the fragment is in bounds, check its coding status */
482
            current_fragment = s->superblock_fragments[i * 16 + j];
483
            if (current_fragment >= s->fragment_count) {
484
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
485
                    current_fragment, s->fragment_count);
486
                return 1;
487
            }
488
            if (current_fragment != -1) {
489
                int coded = s->superblock_coding[i];
490

    
491
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
492

    
493
                    /* fragment may or may not be coded; this is the case
494
                     * that cares about the fragment coding runs */
495
                    if (current_run-- == 0) {
496
                        bit ^= 1;
497
                        current_run = get_vlc2(gb,
498
                            s->fragment_run_length_vlc.table, 5, 2);
499
                    }
500
                    coded = bit;
501
                }
502

    
503
                    if (coded) {
504
                        /* default mode; actual mode will be decoded in
505
                         * the next phase */
506
                        s->all_fragments[current_fragment].coding_method =
507
                            MODE_INTER_NO_MV;
508
                        s->coded_fragment_list[plane][num_coded_frags++] =
509
                            current_fragment;
510
                    } else {
511
                        /* not coded; copy this fragment from the prior frame */
512
                        s->all_fragments[current_fragment].coding_method =
513
                            MODE_COPY;
514
                    }
515
            }
516
        }
517
    }
518
        s->total_num_coded_frags += num_coded_frags;
519
        for (i = 0; i < 64; i++)
520
            s->num_coded_frags[plane][i] = num_coded_frags;
521
        if (plane < 2)
522
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
523
    }
524
    return 0;
525
}
526

    
527
/*
528
 * This function unpacks all the coding mode data for individual macroblocks
529
 * from the bitstream.
530
 */
531
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
532
{
533
    int i, j, k, sb_x, sb_y;
534
    int scheme;
535
    int current_macroblock;
536
    int current_fragment;
537
    int coding_mode;
538
    int custom_mode_alphabet[CODING_MODE_COUNT];
539
    const int *alphabet;
540

    
541
    if (s->keyframe) {
542
        for (i = 0; i < s->fragment_count; i++)
543
            s->all_fragments[i].coding_method = MODE_INTRA;
544

    
545
    } else {
546

    
547
        /* fetch the mode coding scheme for this frame */
548
        scheme = get_bits(gb, 3);
549

    
550
        /* is it a custom coding scheme? */
551
        if (scheme == 0) {
552
            for (i = 0; i < 8; i++)
553
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
554
            for (i = 0; i < 8; i++)
555
                custom_mode_alphabet[get_bits(gb, 3)] = i;
556
            alphabet = custom_mode_alphabet;
557
        } else
558
            alphabet = ModeAlphabet[scheme-1];
559

    
560
        /* iterate through all of the macroblocks that contain 1 or more
561
         * coded fragments */
562
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
563
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
564

    
565
            for (j = 0; j < 4; j++) {
566
                int mb_x = 2*sb_x +   (j>>1);
567
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
568
                current_macroblock = mb_y * s->macroblock_width + mb_x;
569

    
570
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
571
                    continue;
572

    
573
#define BLOCK_X (2*mb_x + (k&1))
574
#define BLOCK_Y (2*mb_y + (k>>1))
575
                /* coding modes are only stored if the macroblock has at least one
576
                 * luma block coded, otherwise it must be INTER_NO_MV */
577
                for (k = 0; k < 4; k++) {
578
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
579
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
580
                        break;
581
                }
582
                if (k == 4) {
583
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
584
                    continue;
585
                }
586

    
587
                /* mode 7 means get 3 bits for each coding mode */
588
                if (scheme == 7)
589
                    coding_mode = get_bits(gb, 3);
590
                else
591
                    coding_mode = alphabet
592
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
593

    
594
                s->macroblock_coding[current_macroblock] = coding_mode;
595
                for (k = 0; k < 4; k++) {
596
                    current_fragment =
597
                        BLOCK_Y*s->fragment_width + BLOCK_X;
598
                    if (s->all_fragments[current_fragment].coding_method !=
599
                        MODE_COPY)
600
                        s->all_fragments[current_fragment].coding_method =
601
                            coding_mode;
602
                }
603
                for (k = 0; k < 2; k++) {
604
                    current_fragment = s->fragment_start[k+1] +
605
                        mb_y*(s->fragment_width>>1) + mb_x;
606
                    if (s->all_fragments[current_fragment].coding_method !=
607
                        MODE_COPY)
608
                        s->all_fragments[current_fragment].coding_method =
609
                            coding_mode;
610
                }
611
            }
612
            }
613
        }
614
    }
615

    
616
    return 0;
617
}
618

    
619
/*
620
 * This function unpacks all the motion vectors for the individual
621
 * macroblocks from the bitstream.
622
 */
623
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
624
{
625
    int j, k, sb_x, sb_y;
626
    int coding_mode;
627
    int motion_x[6];
628
    int motion_y[6];
629
    int last_motion_x = 0;
630
    int last_motion_y = 0;
631
    int prior_last_motion_x = 0;
632
    int prior_last_motion_y = 0;
633
    int current_macroblock;
634
    int current_fragment;
635

    
636
    if (s->keyframe)
637
        return 0;
638

    
639
    memset(motion_x, 0, 6 * sizeof(int));
640
    memset(motion_y, 0, 6 * sizeof(int));
641

    
642
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
643
    coding_mode = get_bits1(gb);
644

    
645
    /* iterate through all of the macroblocks that contain 1 or more
646
     * coded fragments */
647
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
648
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
649

    
650
        for (j = 0; j < 4; j++) {
651
            int mb_x = 2*sb_x +   (j>>1);
652
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
653
            current_macroblock = mb_y * s->macroblock_width + mb_x;
654

    
655
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
656
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
657
                continue;
658

    
659
            switch (s->macroblock_coding[current_macroblock]) {
660

    
661
            case MODE_INTER_PLUS_MV:
662
            case MODE_GOLDEN_MV:
663
                /* all 6 fragments use the same motion vector */
664
                if (coding_mode == 0) {
665
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
666
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
667
                } else {
668
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
669
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
670
                }
671

    
672
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
673
                if (s->macroblock_coding[current_macroblock] ==
674
                    MODE_INTER_PLUS_MV) {
675
                    prior_last_motion_x = last_motion_x;
676
                    prior_last_motion_y = last_motion_y;
677
                    last_motion_x = motion_x[0];
678
                    last_motion_y = motion_y[0];
679
                }
680
                break;
681

    
682
            case MODE_INTER_FOURMV:
683
                /* vector maintenance */
684
                prior_last_motion_x = last_motion_x;
685
                prior_last_motion_y = last_motion_y;
686

    
687
                /* fetch 4 vectors from the bitstream, one for each
688
                 * Y fragment, then average for the C fragment vectors */
689
                motion_x[4] = motion_y[4] = 0;
690
                for (k = 0; k < 4; k++) {
691
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
692
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
693
                        if (coding_mode == 0) {
694
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
695
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
696
                        } else {
697
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
698
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
699
                        }
700
                        last_motion_x = motion_x[k];
701
                        last_motion_y = motion_y[k];
702
                    } else {
703
                        motion_x[k] = 0;
704
                        motion_y[k] = 0;
705
                    }
706
                    motion_x[4] += motion_x[k];
707
                    motion_y[4] += motion_y[k];
708
                }
709

    
710
                motion_x[5]=
711
                motion_x[4]= RSHIFT(motion_x[4], 2);
712
                motion_y[5]=
713
                motion_y[4]= RSHIFT(motion_y[4], 2);
714
                break;
715

    
716
            case MODE_INTER_LAST_MV:
717
                /* all 6 fragments use the last motion vector */
718
                motion_x[0] = last_motion_x;
719
                motion_y[0] = last_motion_y;
720

    
721
                /* no vector maintenance (last vector remains the
722
                 * last vector) */
723
                break;
724

    
725
            case MODE_INTER_PRIOR_LAST:
726
                /* all 6 fragments use the motion vector prior to the
727
                 * last motion vector */
728
                motion_x[0] = prior_last_motion_x;
729
                motion_y[0] = prior_last_motion_y;
730

    
731
                /* vector maintenance */
732
                prior_last_motion_x = last_motion_x;
733
                prior_last_motion_y = last_motion_y;
734
                last_motion_x = motion_x[0];
735
                last_motion_y = motion_y[0];
736
                break;
737

    
738
            default:
739
                /* covers intra, inter without MV, golden without MV */
740
                motion_x[0] = 0;
741
                motion_y[0] = 0;
742

    
743
                /* no vector maintenance */
744
                break;
745
            }
746

    
747
            /* assign the motion vectors to the correct fragments */
748
            for (k = 0; k < 4; k++) {
749
                current_fragment =
750
                    BLOCK_Y*s->fragment_width + BLOCK_X;
751
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
752
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
753
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
754
                } else {
755
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
756
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
757
                }
758
            }
759
            for (k = 0; k < 2; k++) {
760
                current_fragment = s->fragment_start[k+1] +
761
                    mb_y*(s->fragment_width>>1) + mb_x;
762
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
763
                    s->all_fragments[current_fragment].motion_x = motion_x[k+4];
764
                    s->all_fragments[current_fragment].motion_y = motion_y[k+4];
765
                } else {
766
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
767
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
768
                }
769
            }
770
        }
771
        }
772
    }
773

    
774
    return 0;
775
}
776

    
777
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
778
{
779
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
780
    int num_blocks = s->total_num_coded_frags;
781

    
782
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
783
        i = blocks_decoded = num_blocks_at_qpi = 0;
784

    
785
        bit = get_bits1(gb);
786

    
787
        do {
788
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
789
            if (run_length == 34)
790
                run_length += get_bits(gb, 12);
791
            blocks_decoded += run_length;
792

    
793
            if (!bit)
794
                num_blocks_at_qpi += run_length;
795

    
796
            for (j = 0; j < run_length; i++) {
797
                if (i >= s->total_num_coded_frags)
798
                    return -1;
799

    
800
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
801
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
802
                    j++;
803
                }
804
            }
805

    
806
            if (run_length == MAXIMUM_LONG_BIT_RUN)
807
                bit = get_bits1(gb);
808
            else
809
                bit ^= 1;
810
        } while (blocks_decoded < num_blocks);
811

    
812
        num_blocks -= num_blocks_at_qpi;
813
    }
814

    
815
    return 0;
816
}
817

    
818
/*
819
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
820
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
821
 * data. This function unpacks all the VLCs for either the Y plane or both
822
 * C planes, and is called for DC coefficients or different AC coefficient
823
 * levels (since different coefficient types require different VLC tables.
824
 *
825
 * This function returns a residual eob run. E.g, if a particular token gave
826
 * instructions to EOB the next 5 fragments and there were only 2 fragments
827
 * left in the current fragment range, 3 would be returned so that it could
828
 * be passed into the next call to this same function.
829
 */
830
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
831
                        VLC *table, int coeff_index,
832
                        int plane,
833
                        int eob_run)
834
{
835
    int i, j = 0;
836
    int token;
837
    int zero_run = 0;
838
    DCTELEM coeff = 0;
839
    int bits_to_get;
840
    int blocks_ended;
841
    int coeff_i = 0;
842
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
843
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
844

    
845
    /* local references to structure members to avoid repeated deferences */
846
    int *coded_fragment_list = s->coded_fragment_list[plane];
847
    Vp3Fragment *all_fragments = s->all_fragments;
848
    VLC_TYPE (*vlc_table)[2] = table->table;
849

    
850
    if (num_coeffs < 0)
851
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
852

    
853
    if (eob_run > num_coeffs) {
854
        coeff_i = blocks_ended = num_coeffs;
855
        eob_run -= num_coeffs;
856
    } else {
857
        coeff_i = blocks_ended = eob_run;
858
        eob_run = 0;
859
    }
860

    
861
    // insert fake EOB token to cover the split between planes or zzi
862
    if (blocks_ended)
863
        dct_tokens[j++] = blocks_ended << 2;
864

    
865
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
866
            /* decode a VLC into a token */
867
            token = get_vlc2(gb, vlc_table, 5, 3);
868
            /* use the token to get a zero run, a coefficient, and an eob run */
869
            if (token <= 6) {
870
                eob_run = eob_run_base[token];
871
                if (eob_run_get_bits[token])
872
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
873

    
874
                // record only the number of blocks ended in this plane,
875
                // any spill will be recorded in the next plane.
876
                if (eob_run > num_coeffs - coeff_i) {
877
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
878
                    blocks_ended   += num_coeffs - coeff_i;
879
                    eob_run        -= num_coeffs - coeff_i;
880
                    coeff_i         = num_coeffs;
881
                } else {
882
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
883
                    blocks_ended   += eob_run;
884
                    coeff_i        += eob_run;
885
                    eob_run = 0;
886
                }
887
            } else {
888
                bits_to_get = coeff_get_bits[token];
889
                if (bits_to_get)
890
                    bits_to_get = get_bits(gb, bits_to_get);
891
                coeff = coeff_tables[token][bits_to_get];
892

    
893
                zero_run = zero_run_base[token];
894
                if (zero_run_get_bits[token])
895
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
896

    
897
                if (zero_run) {
898
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
899
                } else {
900
                    // Save DC into the fragment structure. DC prediction is
901
                    // done in raster order, so the actual DC can't be in with
902
                    // other tokens. We still need the token in dct_tokens[]
903
                    // however, or else the structure collapses on itself.
904
                    if (!coeff_index)
905
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
906

    
907
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
908
                }
909

    
910
                if (coeff_index + zero_run > 64) {
911
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
912
                           " %d coeffs left\n", zero_run, 64-coeff_index);
913
                    zero_run = 64 - coeff_index;
914
                }
915

    
916
                // zero runs code multiple coefficients,
917
                // so don't try to decode coeffs for those higher levels
918
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
919
                    s->num_coded_frags[plane][i]--;
920
                coeff_i++;
921
            }
922
    }
923

    
924
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
925
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
926

    
927
    // decrement the number of blocks that have higher coeffecients for each
928
    // EOB run at this level
929
    if (blocks_ended)
930
        for (i = coeff_index+1; i < 64; i++)
931
            s->num_coded_frags[plane][i] -= blocks_ended;
932

    
933
    // setup the next buffer
934
    if (plane < 2)
935
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
936
    else if (coeff_index < 63)
937
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
938

    
939
    return eob_run;
940
}
941

    
942
static void reverse_dc_prediction(Vp3DecodeContext *s,
943
                                  int first_fragment,
944
                                  int fragment_width,
945
                                  int fragment_height);
946
/*
947
 * This function unpacks all of the DCT coefficient data from the
948
 * bitstream.
949
 */
950
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
951
{
952
    int i;
953
    int dc_y_table;
954
    int dc_c_table;
955
    int ac_y_table;
956
    int ac_c_table;
957
    int residual_eob_run = 0;
958
    VLC *y_tables[64];
959
    VLC *c_tables[64];
960

    
961
    s->dct_tokens[0][0] = s->dct_tokens_base;
962

    
963
    /* fetch the DC table indexes */
964
    dc_y_table = get_bits(gb, 4);
965
    dc_c_table = get_bits(gb, 4);
966

    
967
    /* unpack the Y plane DC coefficients */
968
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
969
        0, residual_eob_run);
970

    
971
    /* reverse prediction of the Y-plane DC coefficients */
972
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
973

    
974
    /* unpack the C plane DC coefficients */
975
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
976
        1, residual_eob_run);
977
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
978
        2, residual_eob_run);
979

    
980
    /* reverse prediction of the C-plane DC coefficients */
981
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
982
    {
983
        reverse_dc_prediction(s, s->fragment_start[1],
984
            s->fragment_width / 2, s->fragment_height / 2);
985
        reverse_dc_prediction(s, s->fragment_start[2],
986
            s->fragment_width / 2, s->fragment_height / 2);
987
    }
988

    
989
    /* fetch the AC table indexes */
990
    ac_y_table = get_bits(gb, 4);
991
    ac_c_table = get_bits(gb, 4);
992

    
993
    /* build tables of AC VLC tables */
994
    for (i = 1; i <= 5; i++) {
995
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
996
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
997
    }
998
    for (i = 6; i <= 14; i++) {
999
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1000
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1001
    }
1002
    for (i = 15; i <= 27; i++) {
1003
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1004
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1005
    }
1006
    for (i = 28; i <= 63; i++) {
1007
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1008
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1009
    }
1010

    
1011
    /* decode all AC coefficents */
1012
    for (i = 1; i <= 63; i++) {
1013
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1014
                0, residual_eob_run);
1015

    
1016
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1017
                1, residual_eob_run);
1018
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1019
                2, residual_eob_run);
1020
    }
1021

    
1022
    return 0;
1023
}
1024

    
1025
/*
1026
 * This function reverses the DC prediction for each coded fragment in
1027
 * the frame. Much of this function is adapted directly from the original
1028
 * VP3 source code.
1029
 */
1030
#define COMPATIBLE_FRAME(x) \
1031
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1032
#define DC_COEFF(u) s->all_fragments[u].dc
1033

    
1034
static void reverse_dc_prediction(Vp3DecodeContext *s,
1035
                                  int first_fragment,
1036
                                  int fragment_width,
1037
                                  int fragment_height)
1038
{
1039

    
1040
#define PUL 8
1041
#define PU 4
1042
#define PUR 2
1043
#define PL 1
1044

    
1045
    int x, y;
1046
    int i = first_fragment;
1047

    
1048
    int predicted_dc;
1049

    
1050
    /* DC values for the left, up-left, up, and up-right fragments */
1051
    int vl, vul, vu, vur;
1052

    
1053
    /* indexes for the left, up-left, up, and up-right fragments */
1054
    int l, ul, u, ur;
1055

    
1056
    /*
1057
     * The 6 fields mean:
1058
     *   0: up-left multiplier
1059
     *   1: up multiplier
1060
     *   2: up-right multiplier
1061
     *   3: left multiplier
1062
     */
1063
    static const int predictor_transform[16][4] = {
1064
        {  0,  0,  0,  0},
1065
        {  0,  0,  0,128},        // PL
1066
        {  0,  0,128,  0},        // PUR
1067
        {  0,  0, 53, 75},        // PUR|PL
1068
        {  0,128,  0,  0},        // PU
1069
        {  0, 64,  0, 64},        // PU|PL
1070
        {  0,128,  0,  0},        // PU|PUR
1071
        {  0,  0, 53, 75},        // PU|PUR|PL
1072
        {128,  0,  0,  0},        // PUL
1073
        {  0,  0,  0,128},        // PUL|PL
1074
        { 64,  0, 64,  0},        // PUL|PUR
1075
        {  0,  0, 53, 75},        // PUL|PUR|PL
1076
        {  0,128,  0,  0},        // PUL|PU
1077
       {-104,116,  0,116},        // PUL|PU|PL
1078
        { 24, 80, 24,  0},        // PUL|PU|PUR
1079
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1080
    };
1081

    
1082
    /* This table shows which types of blocks can use other blocks for
1083
     * prediction. For example, INTRA is the only mode in this table to
1084
     * have a frame number of 0. That means INTRA blocks can only predict
1085
     * from other INTRA blocks. There are 2 golden frame coding types;
1086
     * blocks encoding in these modes can only predict from other blocks
1087
     * that were encoded with these 1 of these 2 modes. */
1088
    static const unsigned char compatible_frame[9] = {
1089
        1,    /* MODE_INTER_NO_MV */
1090
        0,    /* MODE_INTRA */
1091
        1,    /* MODE_INTER_PLUS_MV */
1092
        1,    /* MODE_INTER_LAST_MV */
1093
        1,    /* MODE_INTER_PRIOR_MV */
1094
        2,    /* MODE_USING_GOLDEN */
1095
        2,    /* MODE_GOLDEN_MV */
1096
        1,    /* MODE_INTER_FOUR_MV */
1097
        3     /* MODE_COPY */
1098
    };
1099
    int current_frame_type;
1100

    
1101
    /* there is a last DC predictor for each of the 3 frame types */
1102
    short last_dc[3];
1103

    
1104
    int transform = 0;
1105

    
1106
    vul = vu = vur = vl = 0;
1107
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1108

    
1109
    /* for each fragment row... */
1110
    for (y = 0; y < fragment_height; y++) {
1111

    
1112
        /* for each fragment in a row... */
1113
        for (x = 0; x < fragment_width; x++, i++) {
1114

    
1115
            /* reverse prediction if this block was coded */
1116
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1117

    
1118
                current_frame_type =
1119
                    compatible_frame[s->all_fragments[i].coding_method];
1120

    
1121
                transform= 0;
1122
                if(x){
1123
                    l= i-1;
1124
                    vl = DC_COEFF(l);
1125
                    if(COMPATIBLE_FRAME(l))
1126
                        transform |= PL;
1127
                }
1128
                if(y){
1129
                    u= i-fragment_width;
1130
                    vu = DC_COEFF(u);
1131
                    if(COMPATIBLE_FRAME(u))
1132
                        transform |= PU;
1133
                    if(x){
1134
                        ul= i-fragment_width-1;
1135
                        vul = DC_COEFF(ul);
1136
                        if(COMPATIBLE_FRAME(ul))
1137
                            transform |= PUL;
1138
                    }
1139
                    if(x + 1 < fragment_width){
1140
                        ur= i-fragment_width+1;
1141
                        vur = DC_COEFF(ur);
1142
                        if(COMPATIBLE_FRAME(ur))
1143
                            transform |= PUR;
1144
                    }
1145
                }
1146

    
1147
                if (transform == 0) {
1148

    
1149
                    /* if there were no fragments to predict from, use last
1150
                     * DC saved */
1151
                    predicted_dc = last_dc[current_frame_type];
1152
                } else {
1153

    
1154
                    /* apply the appropriate predictor transform */
1155
                    predicted_dc =
1156
                        (predictor_transform[transform][0] * vul) +
1157
                        (predictor_transform[transform][1] * vu) +
1158
                        (predictor_transform[transform][2] * vur) +
1159
                        (predictor_transform[transform][3] * vl);
1160

    
1161
                    predicted_dc /= 128;
1162

    
1163
                    /* check for outranging on the [ul u l] and
1164
                     * [ul u ur l] predictors */
1165
                    if ((transform == 15) || (transform == 13)) {
1166
                        if (FFABS(predicted_dc - vu) > 128)
1167
                            predicted_dc = vu;
1168
                        else if (FFABS(predicted_dc - vl) > 128)
1169
                            predicted_dc = vl;
1170
                        else if (FFABS(predicted_dc - vul) > 128)
1171
                            predicted_dc = vul;
1172
                    }
1173
                }
1174

    
1175
                /* at long last, apply the predictor */
1176
                DC_COEFF(i) += predicted_dc;
1177
                /* save the DC */
1178
                last_dc[current_frame_type] = DC_COEFF(i);
1179
            }
1180
        }
1181
    }
1182
}
1183

    
1184
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1185
{
1186
    int x, y;
1187
    int *bounding_values= s->bounding_values_array+127;
1188

    
1189
    int width           = s->fragment_width  >> !!plane;
1190
    int height          = s->fragment_height >> !!plane;
1191
    int fragment        = s->fragment_start        [plane] + ystart * width;
1192
    int stride          = s->current_frame.linesize[plane];
1193
    uint8_t *plane_data = s->current_frame.data    [plane];
1194
    if (!s->flipped_image) stride = -stride;
1195
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1196

    
1197
    for (y = ystart; y < yend; y++) {
1198

    
1199
        for (x = 0; x < width; x++) {
1200
            /* This code basically just deblocks on the edges of coded blocks.
1201
             * However, it has to be much more complicated because of the
1202
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1203
             * because some pixels get filtered twice. */
1204
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1205
            {
1206
                /* do not perform left edge filter for left columns frags */
1207
                if (x > 0) {
1208
                    s->dsp.vp3_h_loop_filter(
1209
                        plane_data + 8*x,
1210
                        stride, bounding_values);
1211
                }
1212

    
1213
                /* do not perform top edge filter for top row fragments */
1214
                if (y > 0) {
1215
                    s->dsp.vp3_v_loop_filter(
1216
                        plane_data + 8*x,
1217
                        stride, bounding_values);
1218
                }
1219

    
1220
                /* do not perform right edge filter for right column
1221
                 * fragments or if right fragment neighbor is also coded
1222
                 * in this frame (it will be filtered in next iteration) */
1223
                if ((x < width - 1) &&
1224
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1225
                    s->dsp.vp3_h_loop_filter(
1226
                        plane_data + 8*x + 8,
1227
                        stride, bounding_values);
1228
                }
1229

    
1230
                /* do not perform bottom edge filter for bottom row
1231
                 * fragments or if bottom fragment neighbor is also coded
1232
                 * in this frame (it will be filtered in the next row) */
1233
                if ((y < height - 1) &&
1234
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1235
                    s->dsp.vp3_v_loop_filter(
1236
                        plane_data + 8*x + 8*stride,
1237
                        stride, bounding_values);
1238
                }
1239
            }
1240

    
1241
            fragment++;
1242
        }
1243
        plane_data += 8*stride;
1244
    }
1245
}
1246

    
1247
/**
1248
 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1249
 * for the next block in coding order
1250
 */
1251
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1252
                              int plane, int inter, DCTELEM block[64])
1253
{
1254
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1255
    uint8_t *perm = s->scantable.permutated;
1256
    int i = 0;
1257

    
1258
    do {
1259
        int token = *s->dct_tokens[plane][i];
1260
        switch (token & 3) {
1261
        case 0: // EOB
1262
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1263
                s->dct_tokens[plane][i]++;
1264
            else
1265
                *s->dct_tokens[plane][i] = token & ~3;
1266
            goto end;
1267
        case 1: // zero run
1268
            s->dct_tokens[plane][i]++;
1269
            i += (token >> 2) & 0x7f;
1270
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1271
            i++;
1272
            break;
1273
        case 2: // coeff
1274
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1275
            s->dct_tokens[plane][i++]++;
1276
            break;
1277
        default:
1278
            av_log(s->avctx, AV_LOG_ERROR, "internal: invalid token type\n");
1279
            return i;
1280
        }
1281
    } while (i < 64);
1282
end:
1283
    // the actual DC+prediction is in the fragment structure
1284
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1285
    return i;
1286
}
1287

    
1288
/**
1289
 * called when all pixels up to row y are complete
1290
 */
1291
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1292
{
1293
    int h, cy;
1294
    int offset[4];
1295

    
1296
    if(s->avctx->draw_horiz_band==NULL)
1297
        return;
1298

    
1299
    h= y - s->last_slice_end;
1300
    y -= h;
1301

    
1302
    if (!s->flipped_image) {
1303
        if (y == 0)
1304
            h -= s->height - s->avctx->height;  // account for non-mod16
1305
        y = s->height - y - h;
1306
    }
1307

    
1308
    cy = y >> 1;
1309
    offset[0] = s->current_frame.linesize[0]*y;
1310
    offset[1] = s->current_frame.linesize[1]*cy;
1311
    offset[2] = s->current_frame.linesize[2]*cy;
1312
    offset[3] = 0;
1313

    
1314
    emms_c();
1315
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1316
    s->last_slice_end= y + h;
1317
}
1318

    
1319
/*
1320
 * Perform the final rendering for a particular slice of data.
1321
 * The slice number ranges from 0..(c_superblock_height - 1).
1322
 */
1323
static void render_slice(Vp3DecodeContext *s, int slice)
1324
{
1325
    int x, y, i, j;
1326
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1327
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1328
    int motion_halfpel_index;
1329
    uint8_t *motion_source;
1330
    int plane, first_pixel;
1331

    
1332
    if (slice >= s->c_superblock_height)
1333
        return;
1334

    
1335
    for (plane = 0; plane < 3; plane++) {
1336
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1337
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1338
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1339
        int stride            = s->current_frame.linesize[plane];
1340
        int plane_width       = s->width  >> !!plane;
1341
        int plane_height      = s->height >> !!plane;
1342

    
1343
        int sb_x, sb_y        = slice << !plane;
1344
        int slice_height      = sb_y + (plane ? 1 : 2);
1345
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1346

    
1347
        int fragment_width    = s->fragment_width  >> !!plane;
1348
        int fragment_height   = s->fragment_height >> !!plane;
1349
        int fragment_start    = s->fragment_start[plane];
1350

    
1351
        if (!s->flipped_image) stride = -stride;
1352
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1353
            continue;
1354

    
1355

    
1356
        if(FFABS(stride) > 2048)
1357
            return; //various tables are fixed size
1358

    
1359
        /* for each superblock row in the slice (both of them)... */
1360
        for (; sb_y < slice_height; sb_y++) {
1361

    
1362
            /* for each superblock in a row... */
1363
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1364

    
1365
                /* for each block in a superblock... */
1366
                for (j = 0; j < 16; j++) {
1367
                    x = 4*sb_x + hilbert_offset[j][0];
1368
                    y = 4*sb_y + hilbert_offset[j][1];
1369

    
1370
                    i = fragment_start + y*fragment_width + x;
1371

    
1372
                    // bounds check
1373
                    if (x >= fragment_width || y >= fragment_height)
1374
                        continue;
1375

    
1376
                first_pixel = 8*y*stride + 8*x;
1377

    
1378
                /* transform if this block was coded */
1379
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1380
                    int intra = s->all_fragments[i].coding_method == MODE_INTRA;
1381

    
1382
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1383
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1384
                        motion_source= golden_plane;
1385
                    else
1386
                        motion_source= last_plane;
1387

    
1388
                    motion_source += first_pixel;
1389
                    motion_halfpel_index = 0;
1390

    
1391
                    /* sort out the motion vector if this fragment is coded
1392
                     * using a motion vector method */
1393
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1394
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1395
                        int src_x, src_y;
1396
                        motion_x = s->all_fragments[i].motion_x;
1397
                        motion_y = s->all_fragments[i].motion_y;
1398
                        if(plane){
1399
                            motion_x= (motion_x>>1) | (motion_x&1);
1400
                            motion_y= (motion_y>>1) | (motion_y&1);
1401
                        }
1402

    
1403
                        src_x= (motion_x>>1) + 8*x;
1404
                        src_y= (motion_y>>1) + 8*y;
1405
                        if ((motion_x == 127) || (motion_y == 127))
1406
                            av_log(s->avctx, AV_LOG_ERROR, " help! got invalid motion vector! (%X, %X)\n", motion_x, motion_y);
1407

    
1408
                        motion_halfpel_index = motion_x & 0x01;
1409
                        motion_source += (motion_x >> 1);
1410

    
1411
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1412
                        motion_source += ((motion_y >> 1) * stride);
1413

    
1414
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1415
                            uint8_t *temp= s->edge_emu_buffer;
1416
                            if(stride<0) temp -= 9*stride;
1417
                            else temp += 9*stride;
1418

    
1419
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1420
                            motion_source= temp;
1421
                        }
1422
                    }
1423

    
1424

    
1425
                    /* first, take care of copying a block from either the
1426
                     * previous or the golden frame */
1427
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1428
                        /* Note, it is possible to implement all MC cases with
1429
                           put_no_rnd_pixels_l2 which would look more like the
1430
                           VP3 source but this would be slower as
1431
                           put_no_rnd_pixels_tab is better optimzed */
1432
                        if(motion_halfpel_index != 3){
1433
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1434
                                output_plane + first_pixel,
1435
                                motion_source, stride, 8);
1436
                        }else{
1437
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1438
                            s->dsp.put_no_rnd_pixels_l2[1](
1439
                                output_plane + first_pixel,
1440
                                motion_source - d,
1441
                                motion_source + stride + 1 + d,
1442
                                stride, 8);
1443
                        }
1444
                    }
1445

    
1446
                        s->dsp.clear_block(block);
1447
                        vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
1448

    
1449
                    /* invert DCT and place (or add) in final output */
1450

    
1451
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1452
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1453
                            block[0] += 128<<3;
1454
                        s->dsp.idct_put(
1455
                            output_plane + first_pixel,
1456
                            stride,
1457
                            block);
1458
                    } else {
1459
                        s->dsp.idct_add(
1460
                            output_plane + first_pixel,
1461
                            stride,
1462
                            block);
1463
                    }
1464
                } else {
1465

    
1466
                    /* copy directly from the previous frame */
1467
                    s->dsp.put_pixels_tab[1][0](
1468
                        output_plane + first_pixel,
1469
                        last_plane + first_pixel,
1470
                        stride, 8);
1471

    
1472
                }
1473
                }
1474
            }
1475

    
1476
            // Filter up to the last row in the superblock row
1477
            apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1478
        }
1479
    }
1480

    
1481
     /* this looks like a good place for slice dispatch... */
1482
     /* algorithm:
1483
      *   if (slice == s->macroblock_height - 1)
1484
      *     dispatch (both last slice & 2nd-to-last slice);
1485
      *   else if (slice > 0)
1486
      *     dispatch (slice - 1);
1487
      */
1488

    
1489
    vp3_draw_horiz_band(s, FFMIN(64*slice + 64-16, s->height-16));
1490
}
1491

    
1492
/*
1493
 * This is the ffmpeg/libavcodec API init function.
1494
 */
1495
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1496
{
1497
    Vp3DecodeContext *s = avctx->priv_data;
1498
    int i, inter, plane;
1499
    int c_width;
1500
    int c_height;
1501

    
1502
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1503
        s->version = 0;
1504
    else
1505
        s->version = 1;
1506

    
1507
    s->avctx = avctx;
1508
    s->width = FFALIGN(avctx->width, 16);
1509
    s->height = FFALIGN(avctx->height, 16);
1510
    avctx->pix_fmt = PIX_FMT_YUV420P;
1511
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1512
    if(avctx->idct_algo==FF_IDCT_AUTO)
1513
        avctx->idct_algo=FF_IDCT_VP3;
1514
    dsputil_init(&s->dsp, avctx);
1515

    
1516
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1517

    
1518
    /* initialize to an impossible value which will force a recalculation
1519
     * in the first frame decode */
1520
    for (i = 0; i < 3; i++)
1521
        s->qps[i] = -1;
1522

    
1523
    s->y_superblock_width = (s->width + 31) / 32;
1524
    s->y_superblock_height = (s->height + 31) / 32;
1525
    s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1526

    
1527
    /* work out the dimensions for the C planes */
1528
    c_width = s->width / 2;
1529
    c_height = s->height / 2;
1530
    s->c_superblock_width = (c_width + 31) / 32;
1531
    s->c_superblock_height = (c_height + 31) / 32;
1532
    s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1533

    
1534
    s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1535
    s->u_superblock_start = s->y_superblock_count;
1536
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1537
    s->superblock_coding = av_malloc(s->superblock_count);
1538

    
1539
    s->macroblock_width = (s->width + 15) / 16;
1540
    s->macroblock_height = (s->height + 15) / 16;
1541
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1542

    
1543
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1544
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1545

    
1546
    /* fragment count covers all 8x8 blocks for all 3 planes */
1547
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1548
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1549
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1550

    
1551
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1552
    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1553
    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1554
    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1555
        !s->coded_fragment_list[0]) {
1556
        vp3_decode_end(avctx);
1557
        return -1;
1558
    }
1559

    
1560
    if (!s->theora_tables)
1561
    {
1562
        for (i = 0; i < 64; i++) {
1563
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1564
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1565
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1566
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1567
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1568
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1569
        }
1570

    
1571
        for(inter=0; inter<2; inter++){
1572
            for(plane=0; plane<3; plane++){
1573
                s->qr_count[inter][plane]= 1;
1574
                s->qr_size [inter][plane][0]= 63;
1575
                s->qr_base [inter][plane][0]=
1576
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1577
            }
1578
        }
1579

    
1580
        /* init VLC tables */
1581
        for (i = 0; i < 16; i++) {
1582

    
1583
            /* DC histograms */
1584
            init_vlc(&s->dc_vlc[i], 5, 32,
1585
                &dc_bias[i][0][1], 4, 2,
1586
                &dc_bias[i][0][0], 4, 2, 0);
1587

    
1588
            /* group 1 AC histograms */
1589
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1590
                &ac_bias_0[i][0][1], 4, 2,
1591
                &ac_bias_0[i][0][0], 4, 2, 0);
1592

    
1593
            /* group 2 AC histograms */
1594
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1595
                &ac_bias_1[i][0][1], 4, 2,
1596
                &ac_bias_1[i][0][0], 4, 2, 0);
1597

    
1598
            /* group 3 AC histograms */
1599
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1600
                &ac_bias_2[i][0][1], 4, 2,
1601
                &ac_bias_2[i][0][0], 4, 2, 0);
1602

    
1603
            /* group 4 AC histograms */
1604
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1605
                &ac_bias_3[i][0][1], 4, 2,
1606
                &ac_bias_3[i][0][0], 4, 2, 0);
1607
        }
1608
    } else {
1609
        for (i = 0; i < 16; i++) {
1610

    
1611
            /* DC histograms */
1612
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1613
                &s->huffman_table[i][0][1], 4, 2,
1614
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1615
                goto vlc_fail;
1616

    
1617
            /* group 1 AC histograms */
1618
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1619
                &s->huffman_table[i+16][0][1], 4, 2,
1620
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1621
                goto vlc_fail;
1622

    
1623
            /* group 2 AC histograms */
1624
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1625
                &s->huffman_table[i+16*2][0][1], 4, 2,
1626
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1627
                goto vlc_fail;
1628

    
1629
            /* group 3 AC histograms */
1630
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1631
                &s->huffman_table[i+16*3][0][1], 4, 2,
1632
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1633
                goto vlc_fail;
1634

    
1635
            /* group 4 AC histograms */
1636
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1637
                &s->huffman_table[i+16*4][0][1], 4, 2,
1638
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1639
                goto vlc_fail;
1640
        }
1641
    }
1642

    
1643
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1644
        &superblock_run_length_vlc_table[0][1], 4, 2,
1645
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1646

    
1647
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1648
        &fragment_run_length_vlc_table[0][1], 4, 2,
1649
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1650

    
1651
    init_vlc(&s->mode_code_vlc, 3, 8,
1652
        &mode_code_vlc_table[0][1], 2, 1,
1653
        &mode_code_vlc_table[0][0], 2, 1, 0);
1654

    
1655
    init_vlc(&s->motion_vector_vlc, 6, 63,
1656
        &motion_vector_vlc_table[0][1], 2, 1,
1657
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1658

    
1659
    /* work out the block mapping tables */
1660
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1661
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1662
    if (!s->superblock_fragments || !s->macroblock_coding) {
1663
        vp3_decode_end(avctx);
1664
        return -1;
1665
    }
1666
    init_block_mapping(s);
1667

    
1668
    for (i = 0; i < 3; i++) {
1669
        s->current_frame.data[i] = NULL;
1670
        s->last_frame.data[i] = NULL;
1671
        s->golden_frame.data[i] = NULL;
1672
    }
1673

    
1674
    return 0;
1675

    
1676
vlc_fail:
1677
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1678
    return -1;
1679
}
1680

    
1681
/*
1682
 * This is the ffmpeg/libavcodec API frame decode function.
1683
 */
1684
static int vp3_decode_frame(AVCodecContext *avctx,
1685
                            void *data, int *data_size,
1686
                            AVPacket *avpkt)
1687
{
1688
    const uint8_t *buf = avpkt->data;
1689
    int buf_size = avpkt->size;
1690
    Vp3DecodeContext *s = avctx->priv_data;
1691
    GetBitContext gb;
1692
    static int counter = 0;
1693
    int i;
1694

    
1695
    init_get_bits(&gb, buf, buf_size * 8);
1696

    
1697
    if (s->theora && get_bits1(&gb))
1698
    {
1699
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1700
        return -1;
1701
    }
1702

    
1703
    s->keyframe = !get_bits1(&gb);
1704
    if (!s->theora)
1705
        skip_bits(&gb, 1);
1706
    for (i = 0; i < 3; i++)
1707
        s->last_qps[i] = s->qps[i];
1708

    
1709
    s->nqps=0;
1710
    do{
1711
        s->qps[s->nqps++]= get_bits(&gb, 6);
1712
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1713
    for (i = s->nqps; i < 3; i++)
1714
        s->qps[i] = -1;
1715

    
1716
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1717
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1718
            s->keyframe?"key":"", counter, s->qps[0]);
1719
    counter++;
1720

    
1721
    if (s->qps[0] != s->last_qps[0])
1722
        init_loop_filter(s);
1723

    
1724
    for (i = 0; i < s->nqps; i++)
1725
        // reinit all dequantizers if the first one changed, because
1726
        // the DC of the first quantizer must be used for all matrices
1727
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1728
            init_dequantizer(s, i);
1729

    
1730
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1731
        return buf_size;
1732

    
1733
    s->current_frame.reference = 3;
1734
    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
1735
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1736
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1737
        goto error;
1738
    }
1739

    
1740
    if (s->keyframe) {
1741
        if (!s->theora)
1742
        {
1743
            skip_bits(&gb, 4); /* width code */
1744
            skip_bits(&gb, 4); /* height code */
1745
            if (s->version)
1746
            {
1747
                s->version = get_bits(&gb, 5);
1748
                if (counter == 1)
1749
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1750
            }
1751
        }
1752
        if (s->version || s->theora)
1753
        {
1754
                if (get_bits1(&gb))
1755
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1756
            skip_bits(&gb, 2); /* reserved? */
1757
        }
1758
    } else {
1759
        if (!s->golden_frame.data[0]) {
1760
            av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1761

    
1762
            s->golden_frame.reference = 3;
1763
            s->golden_frame.pict_type = FF_I_TYPE;
1764
            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1765
                av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1766
                goto error;
1767
            }
1768
            s->last_frame = s->golden_frame;
1769
            s->last_frame.type = FF_BUFFER_TYPE_COPY;
1770
        }
1771
    }
1772

    
1773
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1774
    s->current_frame.qstride= 0;
1775

    
1776
    init_frame(s, &gb);
1777

    
1778
    if (unpack_superblocks(s, &gb)){
1779
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1780
        goto error;
1781
    }
1782
    if (unpack_modes(s, &gb)){
1783
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1784
        goto error;
1785
    }
1786
    if (unpack_vectors(s, &gb)){
1787
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1788
        goto error;
1789
    }
1790
    if (unpack_block_qpis(s, &gb)){
1791
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1792
        goto error;
1793
    }
1794
    if (unpack_dct_coeffs(s, &gb)){
1795
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1796
        goto error;
1797
    }
1798

    
1799
    for (i = 0; i < 3; i++) {
1800
        if (s->flipped_image)
1801
            s->data_offset[i] = 0;
1802
        else
1803
            s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1804
    }
1805

    
1806
    s->last_slice_end = 0;
1807
    for (i = 0; i < s->c_superblock_height; i++)
1808
        render_slice(s, i);
1809

    
1810
    // filter the last row
1811
    for (i = 0; i < 3; i++) {
1812
        int row = (s->height >> (3+!!i)) - 1;
1813
        apply_loop_filter(s, i, row, row+1);
1814
    }
1815
    vp3_draw_horiz_band(s, s->height);
1816

    
1817
    *data_size=sizeof(AVFrame);
1818
    *(AVFrame*)data= s->current_frame;
1819

    
1820
    /* release the last frame, if it is allocated and if it is not the
1821
     * golden frame */
1822
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1823
        avctx->release_buffer(avctx, &s->last_frame);
1824

    
1825
    /* shuffle frames (last = current) */
1826
    s->last_frame= s->current_frame;
1827

    
1828
    if (s->keyframe) {
1829
        if (s->golden_frame.data[0])
1830
            avctx->release_buffer(avctx, &s->golden_frame);
1831
        s->golden_frame = s->current_frame;
1832
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1833
    }
1834

    
1835
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1836

    
1837
    return buf_size;
1838

    
1839
error:
1840
    if (s->current_frame.data[0])
1841
        avctx->release_buffer(avctx, &s->current_frame);
1842
    return -1;
1843
}
1844

    
1845
/*
1846
 * This is the ffmpeg/libavcodec API module cleanup function.
1847
 */
1848
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1849
{
1850
    Vp3DecodeContext *s = avctx->priv_data;
1851
    int i;
1852

    
1853
    av_free(s->superblock_coding);
1854
    av_free(s->all_fragments);
1855
    av_free(s->coded_fragment_list[0]);
1856
    av_free(s->dct_tokens_base);
1857
    av_free(s->superblock_fragments);
1858
    av_free(s->macroblock_coding);
1859

    
1860
    for (i = 0; i < 16; i++) {
1861
        free_vlc(&s->dc_vlc[i]);
1862
        free_vlc(&s->ac_vlc_1[i]);
1863
        free_vlc(&s->ac_vlc_2[i]);
1864
        free_vlc(&s->ac_vlc_3[i]);
1865
        free_vlc(&s->ac_vlc_4[i]);
1866
    }
1867

    
1868
    free_vlc(&s->superblock_run_length_vlc);
1869
    free_vlc(&s->fragment_run_length_vlc);
1870
    free_vlc(&s->mode_code_vlc);
1871
    free_vlc(&s->motion_vector_vlc);
1872

    
1873
    /* release all frames */
1874
    if (s->golden_frame.data[0])
1875
        avctx->release_buffer(avctx, &s->golden_frame);
1876
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1877
        avctx->release_buffer(avctx, &s->last_frame);
1878
    /* no need to release the current_frame since it will always be pointing
1879
     * to the same frame as either the golden or last frame */
1880

    
1881
    return 0;
1882
}
1883

    
1884
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1885
{
1886
    Vp3DecodeContext *s = avctx->priv_data;
1887

    
1888
    if (get_bits1(gb)) {
1889
        int token;
1890
        if (s->entries >= 32) { /* overflow */
1891
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1892
            return -1;
1893
        }
1894
        token = get_bits(gb, 5);
1895
        //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);
1896
        s->huffman_table[s->hti][token][0] = s->hbits;
1897
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1898
        s->entries++;
1899
    }
1900
    else {
1901
        if (s->huff_code_size >= 32) {/* overflow */
1902
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1903
            return -1;
1904
        }
1905
        s->huff_code_size++;
1906
        s->hbits <<= 1;
1907
        if (read_huffman_tree(avctx, gb))
1908
            return -1;
1909
        s->hbits |= 1;
1910
        if (read_huffman_tree(avctx, gb))
1911
            return -1;
1912
        s->hbits >>= 1;
1913
        s->huff_code_size--;
1914
    }
1915
    return 0;
1916
}
1917

    
1918
#if CONFIG_THEORA_DECODER
1919
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1920
{
1921
    Vp3DecodeContext *s = avctx->priv_data;
1922
    int visible_width, visible_height, colorspace;
1923

    
1924
    s->theora = get_bits_long(gb, 24);
1925
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1926

    
1927
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1928
    /* but previous versions have the image flipped relative to vp3 */
1929
    if (s->theora < 0x030200)
1930
    {
1931
        s->flipped_image = 1;
1932
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1933
    }
1934

    
1935
    visible_width  = s->width  = get_bits(gb, 16) << 4;
1936
    visible_height = s->height = get_bits(gb, 16) << 4;
1937

    
1938
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
1939
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
1940
        s->width= s->height= 0;
1941
        return -1;
1942
    }
1943

    
1944
    if (s->theora >= 0x030200) {
1945
        visible_width  = get_bits_long(gb, 24);
1946
        visible_height = get_bits_long(gb, 24);
1947

    
1948
        skip_bits(gb, 8); /* offset x */
1949
        skip_bits(gb, 8); /* offset y */
1950
    }
1951

    
1952
    skip_bits(gb, 32); /* fps numerator */
1953
    skip_bits(gb, 32); /* fps denumerator */
1954
    skip_bits(gb, 24); /* aspect numerator */
1955
    skip_bits(gb, 24); /* aspect denumerator */
1956

    
1957
    if (s->theora < 0x030200)
1958
        skip_bits(gb, 5); /* keyframe frequency force */
1959
    colorspace = get_bits(gb, 8);
1960
    skip_bits(gb, 24); /* bitrate */
1961

    
1962
    skip_bits(gb, 6); /* quality hint */
1963

    
1964
    if (s->theora >= 0x030200)
1965
    {
1966
        skip_bits(gb, 5); /* keyframe frequency force */
1967
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
1968
        skip_bits(gb, 3); /* reserved */
1969
    }
1970

    
1971
//    align_get_bits(gb);
1972

    
1973
    if (   visible_width  <= s->width  && visible_width  > s->width-16
1974
        && visible_height <= s->height && visible_height > s->height-16)
1975
        avcodec_set_dimensions(avctx, visible_width, visible_height);
1976
    else
1977
        avcodec_set_dimensions(avctx, s->width, s->height);
1978

    
1979
    if (colorspace == 1) {
1980
        avctx->color_primaries = AVCOL_PRI_BT470M;
1981
    } else if (colorspace == 2) {
1982
        avctx->color_primaries = AVCOL_PRI_BT470BG;
1983
    }
1984
    if (colorspace == 1 || colorspace == 2) {
1985
        avctx->colorspace = AVCOL_SPC_BT470BG;
1986
        avctx->color_trc  = AVCOL_TRC_BT709;
1987
    }
1988

    
1989
    return 0;
1990
}
1991

    
1992
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
1993
{
1994
    Vp3DecodeContext *s = avctx->priv_data;
1995
    int i, n, matrices, inter, plane;
1996

    
1997
    if (s->theora >= 0x030200) {
1998
        n = get_bits(gb, 3);
1999
        /* loop filter limit values table */
2000
        for (i = 0; i < 64; i++) {
2001
            s->filter_limit_values[i] = get_bits(gb, n);
2002
            if (s->filter_limit_values[i] > 127) {
2003
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2004
                s->filter_limit_values[i] = 127;
2005
            }
2006
        }
2007
    }
2008

    
2009
    if (s->theora >= 0x030200)
2010
        n = get_bits(gb, 4) + 1;
2011
    else
2012
        n = 16;
2013
    /* quality threshold table */
2014
    for (i = 0; i < 64; i++)
2015
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2016

    
2017
    if (s->theora >= 0x030200)
2018
        n = get_bits(gb, 4) + 1;
2019
    else
2020
        n = 16;
2021
    /* dc scale factor table */
2022
    for (i = 0; i < 64; i++)
2023
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2024

    
2025
    if (s->theora >= 0x030200)
2026
        matrices = get_bits(gb, 9) + 1;
2027
    else
2028
        matrices = 3;
2029

    
2030
    if(matrices > 384){
2031
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2032
        return -1;
2033
    }
2034

    
2035
    for(n=0; n<matrices; n++){
2036
        for (i = 0; i < 64; i++)
2037
            s->base_matrix[n][i]= get_bits(gb, 8);
2038
    }
2039

    
2040
    for (inter = 0; inter <= 1; inter++) {
2041
        for (plane = 0; plane <= 2; plane++) {
2042
            int newqr= 1;
2043
            if (inter || plane > 0)
2044
                newqr = get_bits1(gb);
2045
            if (!newqr) {
2046
                int qtj, plj;
2047
                if(inter && get_bits1(gb)){
2048
                    qtj = 0;
2049
                    plj = plane;
2050
                }else{
2051
                    qtj= (3*inter + plane - 1) / 3;
2052
                    plj= (plane + 2) % 3;
2053
                }
2054
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2055
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2056
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2057
            } else {
2058
                int qri= 0;
2059
                int qi = 0;
2060

    
2061
                for(;;){
2062
                    i= get_bits(gb, av_log2(matrices-1)+1);
2063
                    if(i>= matrices){
2064
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2065
                        return -1;
2066
                    }
2067
                    s->qr_base[inter][plane][qri]= i;
2068
                    if(qi >= 63)
2069
                        break;
2070
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2071
                    s->qr_size[inter][plane][qri++]= i;
2072
                    qi += i;
2073
                }
2074

    
2075
                if (qi > 63) {
2076
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2077
                    return -1;
2078
                }
2079
                s->qr_count[inter][plane]= qri;
2080
            }
2081
        }
2082
    }
2083

    
2084
    /* Huffman tables */
2085
    for (s->hti = 0; s->hti < 80; s->hti++) {
2086
        s->entries = 0;
2087
        s->huff_code_size = 1;
2088
        if (!get_bits1(gb)) {
2089
            s->hbits = 0;
2090
            if(read_huffman_tree(avctx, gb))
2091
                return -1;
2092
            s->hbits = 1;
2093
            if(read_huffman_tree(avctx, gb))
2094
                return -1;
2095
        }
2096
    }
2097

    
2098
    s->theora_tables = 1;
2099

    
2100
    return 0;
2101
}
2102

    
2103
static av_cold int theora_decode_init(AVCodecContext *avctx)
2104
{
2105
    Vp3DecodeContext *s = avctx->priv_data;
2106
    GetBitContext gb;
2107
    int ptype;
2108
    uint8_t *header_start[3];
2109
    int header_len[3];
2110
    int i;
2111

    
2112
    s->theora = 1;
2113

    
2114
    if (!avctx->extradata_size)
2115
    {
2116
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2117
        return -1;
2118
    }
2119

    
2120
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2121
                              42, header_start, header_len) < 0) {
2122
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2123
        return -1;
2124
    }
2125

    
2126
  for(i=0;i<3;i++) {
2127
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2128

    
2129
    ptype = get_bits(&gb, 8);
2130

    
2131
     if (!(ptype & 0x80))
2132
     {
2133
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2134
//        return -1;
2135
     }
2136

    
2137
    // FIXME: Check for this as well.
2138
    skip_bits_long(&gb, 6*8); /* "theora" */
2139

    
2140
    switch(ptype)
2141
    {
2142
        case 0x80:
2143
            theora_decode_header(avctx, &gb);
2144
                break;
2145
        case 0x81:
2146
// FIXME: is this needed? it breaks sometimes
2147
//            theora_decode_comments(avctx, gb);
2148
            break;
2149
        case 0x82:
2150
            if (theora_decode_tables(avctx, &gb))
2151
                return -1;
2152
            break;
2153
        default:
2154
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2155
            break;
2156
    }
2157
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2158
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2159
    if (s->theora < 0x030200)
2160
        break;
2161
  }
2162

    
2163
    return vp3_decode_init(avctx);
2164
}
2165

    
2166
AVCodec theora_decoder = {
2167
    "theora",
2168
    CODEC_TYPE_VIDEO,
2169
    CODEC_ID_THEORA,
2170
    sizeof(Vp3DecodeContext),
2171
    theora_decode_init,
2172
    NULL,
2173
    vp3_decode_end,
2174
    vp3_decode_frame,
2175
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2176
    NULL,
2177
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2178
};
2179
#endif
2180

    
2181
AVCodec vp3_decoder = {
2182
    "vp3",
2183
    CODEC_TYPE_VIDEO,
2184
    CODEC_ID_VP3,
2185
    sizeof(Vp3DecodeContext),
2186
    vp3_decode_init,
2187
    NULL,
2188
    vp3_decode_end,
2189
    vp3_decode_frame,
2190
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2191
    NULL,
2192
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2193
};