Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 703acd54

History | View | Annotate | Download (74.1 KB)

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

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

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

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

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

    
43
#define FRAGMENT_PIXELS 8
44

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

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

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

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

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

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

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

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

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

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

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

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

    
117
};
118

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

    
126
#define MIN_DEQUANT_VAL 2
127

    
128
typedef struct Vp3DecodeContext {
129
    AVCodecContext *avctx;
130
    int theora, theora_tables;
131
    int version;
132
    int width, height;
133
    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 sets up the dequantization tables used for a particular
290
 * frame.
291
 */
292
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
293
{
294
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
295
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
296
    int i, plane, inter, qri, bmi, bmj, qistart;
297

    
298
    for(inter=0; inter<2; inter++){
299
        for(plane=0; plane<3; plane++){
300
            int sum=0;
301
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
302
                sum+= s->qr_size[inter][plane][qri];
303
                if(s->qps[qpi] <= sum)
304
                    break;
305
            }
306
            qistart= sum - s->qr_size[inter][plane][qri];
307
            bmi= s->qr_base[inter][plane][qri  ];
308
            bmj= s->qr_base[inter][plane][qri+1];
309
            for(i=0; i<64; i++){
310
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
311
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
312
                            + s->qr_size[inter][plane][qri])
313
                           / (2*s->qr_size[inter][plane][qri]);
314

    
315
                int qmin= 8<<(inter + !i);
316
                int qscale= i ? ac_scale_factor : dc_scale_factor;
317

    
318
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
319
            }
320
            // all DC coefficients use the same quant so as not to interfere with DC prediction
321
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
322
        }
323
    }
324

    
325
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
326
}
327

    
328
/*
329
 * This function initializes the loop filter boundary limits if the frame's
330
 * quality index is different from the previous frame's.
331
 *
332
 * The filter_limit_values may not be larger than 127.
333
 */
334
static void init_loop_filter(Vp3DecodeContext *s)
335
{
336
    int *bounding_values= s->bounding_values_array+127;
337
    int filter_limit;
338
    int x;
339
    int value;
340

    
341
    filter_limit = s->filter_limit_values[s->qps[0]];
342

    
343
    /* set up the bounding values */
344
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
345
    for (x = 0; x < filter_limit; x++) {
346
        bounding_values[-x] = -x;
347
        bounding_values[x] = x;
348
    }
349
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
350
        bounding_values[ x] =  value;
351
        bounding_values[-x] = -value;
352
    }
353
    if (value)
354
        bounding_values[128] = value;
355
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
356
}
357

    
358
/*
359
 * This function unpacks all of the superblock/macroblock/fragment coding
360
 * information from the bitstream.
361
 */
362
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
363
{
364
    int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
365
    int bit = 0;
366
    int current_superblock = 0;
367
    int current_run = 0;
368
    int num_partial_superblocks = 0;
369

    
370
    int i, j;
371
    int current_fragment;
372
    int plane;
373

    
374
    if (s->keyframe) {
375
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
376

    
377
    } else {
378

    
379
        /* unpack the list of partially-coded superblocks */
380
        bit = get_bits1(gb);
381
        while (current_superblock < s->superblock_count) {
382
                current_run = get_vlc2(gb,
383
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
384
                if (current_run == 34)
385
                    current_run += get_bits(gb, 12);
386

    
387
            if (current_superblock + current_run > s->superblock_count) {
388
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
389
                return -1;
390
            }
391

    
392
            memset(s->superblock_coding + current_superblock, bit, current_run);
393

    
394
            current_superblock += current_run;
395
            if (bit)
396
                num_partial_superblocks += current_run;
397

    
398
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
399
                bit = get_bits1(gb);
400
            else
401
                bit ^= 1;
402
        }
403

    
404
        /* unpack the list of fully coded superblocks if any of the blocks were
405
         * not marked as partially coded in the previous step */
406
        if (num_partial_superblocks < s->superblock_count) {
407
            int superblocks_decoded = 0;
408

    
409
            current_superblock = 0;
410
            bit = get_bits1(gb);
411
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks) {
412
                        current_run = get_vlc2(gb,
413
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
414
                        if (current_run == 34)
415
                            current_run += get_bits(gb, 12);
416

    
417
                for (j = 0; j < current_run; current_superblock++) {
418
                    if (current_superblock >= s->superblock_count) {
419
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
420
                        return -1;
421
                    }
422

    
423
                /* skip any superblocks already marked as partially coded */
424
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
425
                    s->superblock_coding[current_superblock] = 2*bit;
426
                    j++;
427
                }
428
                }
429
                superblocks_decoded += current_run;
430

    
431
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
432
                    bit = get_bits1(gb);
433
                else
434
                    bit ^= 1;
435
            }
436
        }
437

    
438
        /* if there were partial blocks, initialize bitstream for
439
         * unpacking fragment codings */
440
        if (num_partial_superblocks) {
441

    
442
            current_run = 0;
443
            bit = get_bits1(gb);
444
            /* toggle the bit because as soon as the first run length is
445
             * fetched the bit will be toggled again */
446
            bit ^= 1;
447
        }
448
    }
449

    
450
    /* figure out which fragments are coded; iterate through each
451
     * superblock (all planes) */
452
    s->total_num_coded_frags = 0;
453
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
454

    
455
    for (plane = 0; plane < 3; plane++) {
456
        int sb_start = superblock_starts[plane];
457
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
458
        int num_coded_frags = 0;
459

    
460
    for (i = sb_start; i < sb_end; i++) {
461

    
462
        /* iterate through all 16 fragments in a superblock */
463
        for (j = 0; j < 16; j++) {
464

    
465
            /* if the fragment is in bounds, check its coding status */
466
            current_fragment = s->superblock_fragments[i * 16 + j];
467
            if (current_fragment >= s->fragment_count) {
468
                av_log(s->avctx, AV_LOG_ERROR, "  vp3:unpack_superblocks(): bad fragment number (%d >= %d)\n",
469
                    current_fragment, s->fragment_count);
470
                return 1;
471
            }
472
            if (current_fragment != -1) {
473
                int coded = s->superblock_coding[i];
474

    
475
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
476

    
477
                    /* fragment may or may not be coded; this is the case
478
                     * that cares about the fragment coding runs */
479
                    if (current_run-- == 0) {
480
                        bit ^= 1;
481
                        current_run = get_vlc2(gb,
482
                            s->fragment_run_length_vlc.table, 5, 2);
483
                    }
484
                    coded = bit;
485
                }
486

    
487
                    if (coded) {
488
                        /* default mode; actual mode will be decoded in
489
                         * the next phase */
490
                        s->all_fragments[current_fragment].coding_method =
491
                            MODE_INTER_NO_MV;
492
                        s->coded_fragment_list[plane][num_coded_frags++] =
493
                            current_fragment;
494
                    } else {
495
                        /* not coded; copy this fragment from the prior frame */
496
                        s->all_fragments[current_fragment].coding_method =
497
                            MODE_COPY;
498
                    }
499
            }
500
        }
501
    }
502
        s->total_num_coded_frags += num_coded_frags;
503
        for (i = 0; i < 64; i++)
504
            s->num_coded_frags[plane][i] = num_coded_frags;
505
        if (plane < 2)
506
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
507
    }
508
    return 0;
509
}
510

    
511
/*
512
 * This function unpacks all the coding mode data for individual macroblocks
513
 * from the bitstream.
514
 */
515
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
516
{
517
    int i, j, k, sb_x, sb_y;
518
    int scheme;
519
    int current_macroblock;
520
    int current_fragment;
521
    int coding_mode;
522
    int custom_mode_alphabet[CODING_MODE_COUNT];
523
    const int *alphabet;
524

    
525
    if (s->keyframe) {
526
        for (i = 0; i < s->fragment_count; i++)
527
            s->all_fragments[i].coding_method = MODE_INTRA;
528

    
529
    } else {
530

    
531
        /* fetch the mode coding scheme for this frame */
532
        scheme = get_bits(gb, 3);
533

    
534
        /* is it a custom coding scheme? */
535
        if (scheme == 0) {
536
            for (i = 0; i < 8; i++)
537
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
538
            for (i = 0; i < 8; i++)
539
                custom_mode_alphabet[get_bits(gb, 3)] = i;
540
            alphabet = custom_mode_alphabet;
541
        } else
542
            alphabet = ModeAlphabet[scheme-1];
543

    
544
        /* iterate through all of the macroblocks that contain 1 or more
545
         * coded fragments */
546
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
547
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
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 + 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
                    current_fragment =
581
                        BLOCK_Y*s->fragment_width + BLOCK_X;
582
                    if (s->all_fragments[current_fragment].coding_method !=
583
                        MODE_COPY)
584
                        s->all_fragments[current_fragment].coding_method =
585
                            coding_mode;
586
                }
587
                for (k = 0; k < 2; k++) {
588
                    current_fragment = s->fragment_start[k+1] +
589
                        mb_y*(s->fragment_width>>1) + mb_x;
590
                    if (s->all_fragments[current_fragment].coding_method !=
591
                        MODE_COPY)
592
                        s->all_fragments[current_fragment].coding_method =
593
                            coding_mode;
594
                }
595
            }
596
            }
597
        }
598
    }
599

    
600
    return 0;
601
}
602

    
603
/*
604
 * This function unpacks all the motion vectors for the individual
605
 * macroblocks from the bitstream.
606
 */
607
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
608
{
609
    int j, k, sb_x, sb_y;
610
    int coding_mode;
611
    int motion_x[6];
612
    int motion_y[6];
613
    int last_motion_x = 0;
614
    int last_motion_y = 0;
615
    int prior_last_motion_x = 0;
616
    int prior_last_motion_y = 0;
617
    int current_macroblock;
618
    int current_fragment;
619

    
620
    if (s->keyframe)
621
        return 0;
622

    
623
    memset(motion_x, 0, 6 * sizeof(int));
624
    memset(motion_y, 0, 6 * sizeof(int));
625

    
626
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
627
    coding_mode = get_bits1(gb);
628

    
629
    /* iterate through all of the macroblocks that contain 1 or more
630
     * coded fragments */
631
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
632
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
633

    
634
        for (j = 0; j < 4; j++) {
635
            int mb_x = 2*sb_x +   (j>>1);
636
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
637
            current_macroblock = mb_y * s->macroblock_width + mb_x;
638

    
639
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
640
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
641
                continue;
642

    
643
            switch (s->macroblock_coding[current_macroblock]) {
644

    
645
            case MODE_INTER_PLUS_MV:
646
            case MODE_GOLDEN_MV:
647
                /* all 6 fragments use the same motion vector */
648
                if (coding_mode == 0) {
649
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
650
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
651
                } else {
652
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
653
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
654
                }
655

    
656
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
657
                if (s->macroblock_coding[current_macroblock] ==
658
                    MODE_INTER_PLUS_MV) {
659
                    prior_last_motion_x = last_motion_x;
660
                    prior_last_motion_y = last_motion_y;
661
                    last_motion_x = motion_x[0];
662
                    last_motion_y = motion_y[0];
663
                }
664
                break;
665

    
666
            case MODE_INTER_FOURMV:
667
                /* vector maintenance */
668
                prior_last_motion_x = last_motion_x;
669
                prior_last_motion_y = last_motion_y;
670

    
671
                /* fetch 4 vectors from the bitstream, one for each
672
                 * Y fragment, then average for the C fragment vectors */
673
                motion_x[4] = motion_y[4] = 0;
674
                for (k = 0; k < 4; k++) {
675
                    current_fragment = BLOCK_Y*s->fragment_width + BLOCK_X;
676
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
677
                        if (coding_mode == 0) {
678
                            motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
679
                            motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
680
                        } else {
681
                            motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
682
                            motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
683
                        }
684
                        last_motion_x = motion_x[k];
685
                        last_motion_y = motion_y[k];
686
                    } else {
687
                        motion_x[k] = 0;
688
                        motion_y[k] = 0;
689
                    }
690
                    motion_x[4] += motion_x[k];
691
                    motion_y[4] += motion_y[k];
692
                }
693

    
694
                motion_x[5]=
695
                motion_x[4]= RSHIFT(motion_x[4], 2);
696
                motion_y[5]=
697
                motion_y[4]= RSHIFT(motion_y[4], 2);
698
                break;
699

    
700
            case MODE_INTER_LAST_MV:
701
                /* all 6 fragments use the last motion vector */
702
                motion_x[0] = last_motion_x;
703
                motion_y[0] = last_motion_y;
704

    
705
                /* no vector maintenance (last vector remains the
706
                 * last vector) */
707
                break;
708

    
709
            case MODE_INTER_PRIOR_LAST:
710
                /* all 6 fragments use the motion vector prior to the
711
                 * last motion vector */
712
                motion_x[0] = prior_last_motion_x;
713
                motion_y[0] = prior_last_motion_y;
714

    
715
                /* vector maintenance */
716
                prior_last_motion_x = last_motion_x;
717
                prior_last_motion_y = last_motion_y;
718
                last_motion_x = motion_x[0];
719
                last_motion_y = motion_y[0];
720
                break;
721

    
722
            default:
723
                /* covers intra, inter without MV, golden without MV */
724
                motion_x[0] = 0;
725
                motion_y[0] = 0;
726

    
727
                /* no vector maintenance */
728
                break;
729
            }
730

    
731
            /* assign the motion vectors to the correct fragments */
732
            for (k = 0; k < 4; k++) {
733
                current_fragment =
734
                    BLOCK_Y*s->fragment_width + BLOCK_X;
735
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
736
                    s->all_fragments[current_fragment].motion_x = motion_x[k];
737
                    s->all_fragments[current_fragment].motion_y = motion_y[k];
738
                } else {
739
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
740
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
741
                }
742
            }
743
            for (k = 0; k < 2; k++) {
744
                current_fragment = s->fragment_start[k+1] +
745
                    mb_y*(s->fragment_width>>1) + mb_x;
746
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
747
                    s->all_fragments[current_fragment].motion_x = motion_x[k+4];
748
                    s->all_fragments[current_fragment].motion_y = motion_y[k+4];
749
                } else {
750
                    s->all_fragments[current_fragment].motion_x = motion_x[0];
751
                    s->all_fragments[current_fragment].motion_y = motion_y[0];
752
                }
753
            }
754
        }
755
        }
756
    }
757

    
758
    return 0;
759
}
760

    
761
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
762
{
763
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
764
    int num_blocks = s->total_num_coded_frags;
765

    
766
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
767
        i = blocks_decoded = num_blocks_at_qpi = 0;
768

    
769
        bit = get_bits1(gb);
770

    
771
        do {
772
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
773
            if (run_length == 34)
774
                run_length += get_bits(gb, 12);
775
            blocks_decoded += run_length;
776

    
777
            if (!bit)
778
                num_blocks_at_qpi += run_length;
779

    
780
            for (j = 0; j < run_length; i++) {
781
                if (i >= s->total_num_coded_frags)
782
                    return -1;
783

    
784
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
785
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
786
                    j++;
787
                }
788
            }
789

    
790
            if (run_length == MAXIMUM_LONG_BIT_RUN)
791
                bit = get_bits1(gb);
792
            else
793
                bit ^= 1;
794
        } while (blocks_decoded < num_blocks);
795

    
796
        num_blocks -= num_blocks_at_qpi;
797
    }
798

    
799
    return 0;
800
}
801

    
802
/*
803
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
804
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
805
 * data. This function unpacks all the VLCs for either the Y plane or both
806
 * C planes, and is called for DC coefficients or different AC coefficient
807
 * levels (since different coefficient types require different VLC tables.
808
 *
809
 * This function returns a residual eob run. E.g, if a particular token gave
810
 * instructions to EOB the next 5 fragments and there were only 2 fragments
811
 * left in the current fragment range, 3 would be returned so that it could
812
 * be passed into the next call to this same function.
813
 */
814
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
815
                        VLC *table, int coeff_index,
816
                        int plane,
817
                        int eob_run)
818
{
819
    int i, j = 0;
820
    int token;
821
    int zero_run = 0;
822
    DCTELEM coeff = 0;
823
    int bits_to_get;
824
    int blocks_ended;
825
    int coeff_i = 0;
826
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
827
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
828

    
829
    /* local references to structure members to avoid repeated deferences */
830
    int *coded_fragment_list = s->coded_fragment_list[plane];
831
    Vp3Fragment *all_fragments = s->all_fragments;
832
    VLC_TYPE (*vlc_table)[2] = table->table;
833

    
834
    if (num_coeffs < 0)
835
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
836

    
837
    if (eob_run > num_coeffs) {
838
        coeff_i = blocks_ended = num_coeffs;
839
        eob_run -= num_coeffs;
840
    } else {
841
        coeff_i = blocks_ended = eob_run;
842
        eob_run = 0;
843
    }
844

    
845
    // insert fake EOB token to cover the split between planes or zzi
846
    if (blocks_ended)
847
        dct_tokens[j++] = blocks_ended << 2;
848

    
849
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
850
            /* decode a VLC into a token */
851
            token = get_vlc2(gb, vlc_table, 5, 3);
852
            /* use the token to get a zero run, a coefficient, and an eob run */
853
            if (token <= 6) {
854
                eob_run = eob_run_base[token];
855
                if (eob_run_get_bits[token])
856
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
857

    
858
                // record only the number of blocks ended in this plane,
859
                // any spill will be recorded in the next plane.
860
                if (eob_run > num_coeffs - coeff_i) {
861
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
862
                    blocks_ended   += num_coeffs - coeff_i;
863
                    eob_run        -= num_coeffs - coeff_i;
864
                    coeff_i         = num_coeffs;
865
                } else {
866
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
867
                    blocks_ended   += eob_run;
868
                    coeff_i        += eob_run;
869
                    eob_run = 0;
870
                }
871
            } else {
872
                bits_to_get = coeff_get_bits[token];
873
                if (bits_to_get)
874
                    bits_to_get = get_bits(gb, bits_to_get);
875
                coeff = coeff_tables[token][bits_to_get];
876

    
877
                zero_run = zero_run_base[token];
878
                if (zero_run_get_bits[token])
879
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
880

    
881
                if (zero_run) {
882
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
883
                } else {
884
                    // Save DC into the fragment structure. DC prediction is
885
                    // done in raster order, so the actual DC can't be in with
886
                    // other tokens. We still need the token in dct_tokens[]
887
                    // however, or else the structure collapses on itself.
888
                    if (!coeff_index)
889
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
890

    
891
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
892
                }
893

    
894
                if (coeff_index + zero_run > 64) {
895
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
896
                           " %d coeffs left\n", zero_run, 64-coeff_index);
897
                    zero_run = 64 - coeff_index;
898
                }
899

    
900
                // zero runs code multiple coefficients,
901
                // so don't try to decode coeffs for those higher levels
902
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
903
                    s->num_coded_frags[plane][i]--;
904
                coeff_i++;
905
            }
906
    }
907

    
908
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
909
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
910

    
911
    // decrement the number of blocks that have higher coeffecients for each
912
    // EOB run at this level
913
    if (blocks_ended)
914
        for (i = coeff_index+1; i < 64; i++)
915
            s->num_coded_frags[plane][i] -= blocks_ended;
916

    
917
    // setup the next buffer
918
    if (plane < 2)
919
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
920
    else if (coeff_index < 63)
921
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
922

    
923
    return eob_run;
924
}
925

    
926
static void reverse_dc_prediction(Vp3DecodeContext *s,
927
                                  int first_fragment,
928
                                  int fragment_width,
929
                                  int fragment_height);
930
/*
931
 * This function unpacks all of the DCT coefficient data from the
932
 * bitstream.
933
 */
934
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
935
{
936
    int i;
937
    int dc_y_table;
938
    int dc_c_table;
939
    int ac_y_table;
940
    int ac_c_table;
941
    int residual_eob_run = 0;
942
    VLC *y_tables[64];
943
    VLC *c_tables[64];
944

    
945
    s->dct_tokens[0][0] = s->dct_tokens_base;
946

    
947
    /* fetch the DC table indexes */
948
    dc_y_table = get_bits(gb, 4);
949
    dc_c_table = get_bits(gb, 4);
950

    
951
    /* unpack the Y plane DC coefficients */
952
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
953
        0, residual_eob_run);
954

    
955
    /* reverse prediction of the Y-plane DC coefficients */
956
    reverse_dc_prediction(s, 0, s->fragment_width, s->fragment_height);
957

    
958
    /* unpack the C plane DC coefficients */
959
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
960
        1, residual_eob_run);
961
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
962
        2, residual_eob_run);
963

    
964
    /* reverse prediction of the C-plane DC coefficients */
965
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
966
    {
967
        reverse_dc_prediction(s, s->fragment_start[1],
968
            s->fragment_width / 2, s->fragment_height / 2);
969
        reverse_dc_prediction(s, s->fragment_start[2],
970
            s->fragment_width / 2, s->fragment_height / 2);
971
    }
972

    
973
    /* fetch the AC table indexes */
974
    ac_y_table = get_bits(gb, 4);
975
    ac_c_table = get_bits(gb, 4);
976

    
977
    /* build tables of AC VLC tables */
978
    for (i = 1; i <= 5; i++) {
979
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
980
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
981
    }
982
    for (i = 6; i <= 14; i++) {
983
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
984
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
985
    }
986
    for (i = 15; i <= 27; i++) {
987
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
988
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
989
    }
990
    for (i = 28; i <= 63; i++) {
991
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
992
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
993
    }
994

    
995
    /* decode all AC coefficents */
996
    for (i = 1; i <= 63; i++) {
997
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
998
                0, residual_eob_run);
999

    
1000
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1001
                1, residual_eob_run);
1002
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1003
                2, residual_eob_run);
1004
    }
1005

    
1006
    return 0;
1007
}
1008

    
1009
/*
1010
 * This function reverses the DC prediction for each coded fragment in
1011
 * the frame. Much of this function is adapted directly from the original
1012
 * VP3 source code.
1013
 */
1014
#define COMPATIBLE_FRAME(x) \
1015
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1016
#define DC_COEFF(u) s->all_fragments[u].dc
1017

    
1018
static void reverse_dc_prediction(Vp3DecodeContext *s,
1019
                                  int first_fragment,
1020
                                  int fragment_width,
1021
                                  int fragment_height)
1022
{
1023

    
1024
#define PUL 8
1025
#define PU 4
1026
#define PUR 2
1027
#define PL 1
1028

    
1029
    int x, y;
1030
    int i = first_fragment;
1031

    
1032
    int predicted_dc;
1033

    
1034
    /* DC values for the left, up-left, up, and up-right fragments */
1035
    int vl, vul, vu, vur;
1036

    
1037
    /* indexes for the left, up-left, up, and up-right fragments */
1038
    int l, ul, u, ur;
1039

    
1040
    /*
1041
     * The 6 fields mean:
1042
     *   0: up-left multiplier
1043
     *   1: up multiplier
1044
     *   2: up-right multiplier
1045
     *   3: left multiplier
1046
     */
1047
    static const int predictor_transform[16][4] = {
1048
        {  0,  0,  0,  0},
1049
        {  0,  0,  0,128},        // PL
1050
        {  0,  0,128,  0},        // PUR
1051
        {  0,  0, 53, 75},        // PUR|PL
1052
        {  0,128,  0,  0},        // PU
1053
        {  0, 64,  0, 64},        // PU|PL
1054
        {  0,128,  0,  0},        // PU|PUR
1055
        {  0,  0, 53, 75},        // PU|PUR|PL
1056
        {128,  0,  0,  0},        // PUL
1057
        {  0,  0,  0,128},        // PUL|PL
1058
        { 64,  0, 64,  0},        // PUL|PUR
1059
        {  0,  0, 53, 75},        // PUL|PUR|PL
1060
        {  0,128,  0,  0},        // PUL|PU
1061
       {-104,116,  0,116},        // PUL|PU|PL
1062
        { 24, 80, 24,  0},        // PUL|PU|PUR
1063
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1064
    };
1065

    
1066
    /* This table shows which types of blocks can use other blocks for
1067
     * prediction. For example, INTRA is the only mode in this table to
1068
     * have a frame number of 0. That means INTRA blocks can only predict
1069
     * from other INTRA blocks. There are 2 golden frame coding types;
1070
     * blocks encoding in these modes can only predict from other blocks
1071
     * that were encoded with these 1 of these 2 modes. */
1072
    static const unsigned char compatible_frame[9] = {
1073
        1,    /* MODE_INTER_NO_MV */
1074
        0,    /* MODE_INTRA */
1075
        1,    /* MODE_INTER_PLUS_MV */
1076
        1,    /* MODE_INTER_LAST_MV */
1077
        1,    /* MODE_INTER_PRIOR_MV */
1078
        2,    /* MODE_USING_GOLDEN */
1079
        2,    /* MODE_GOLDEN_MV */
1080
        1,    /* MODE_INTER_FOUR_MV */
1081
        3     /* MODE_COPY */
1082
    };
1083
    int current_frame_type;
1084

    
1085
    /* there is a last DC predictor for each of the 3 frame types */
1086
    short last_dc[3];
1087

    
1088
    int transform = 0;
1089

    
1090
    vul = vu = vur = vl = 0;
1091
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1092

    
1093
    /* for each fragment row... */
1094
    for (y = 0; y < fragment_height; y++) {
1095

    
1096
        /* for each fragment in a row... */
1097
        for (x = 0; x < fragment_width; x++, i++) {
1098

    
1099
            /* reverse prediction if this block was coded */
1100
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1101

    
1102
                current_frame_type =
1103
                    compatible_frame[s->all_fragments[i].coding_method];
1104

    
1105
                transform= 0;
1106
                if(x){
1107
                    l= i-1;
1108
                    vl = DC_COEFF(l);
1109
                    if(COMPATIBLE_FRAME(l))
1110
                        transform |= PL;
1111
                }
1112
                if(y){
1113
                    u= i-fragment_width;
1114
                    vu = DC_COEFF(u);
1115
                    if(COMPATIBLE_FRAME(u))
1116
                        transform |= PU;
1117
                    if(x){
1118
                        ul= i-fragment_width-1;
1119
                        vul = DC_COEFF(ul);
1120
                        if(COMPATIBLE_FRAME(ul))
1121
                            transform |= PUL;
1122
                    }
1123
                    if(x + 1 < fragment_width){
1124
                        ur= i-fragment_width+1;
1125
                        vur = DC_COEFF(ur);
1126
                        if(COMPATIBLE_FRAME(ur))
1127
                            transform |= PUR;
1128
                    }
1129
                }
1130

    
1131
                if (transform == 0) {
1132

    
1133
                    /* if there were no fragments to predict from, use last
1134
                     * DC saved */
1135
                    predicted_dc = last_dc[current_frame_type];
1136
                } else {
1137

    
1138
                    /* apply the appropriate predictor transform */
1139
                    predicted_dc =
1140
                        (predictor_transform[transform][0] * vul) +
1141
                        (predictor_transform[transform][1] * vu) +
1142
                        (predictor_transform[transform][2] * vur) +
1143
                        (predictor_transform[transform][3] * vl);
1144

    
1145
                    predicted_dc /= 128;
1146

    
1147
                    /* check for outranging on the [ul u l] and
1148
                     * [ul u ur l] predictors */
1149
                    if ((transform == 15) || (transform == 13)) {
1150
                        if (FFABS(predicted_dc - vu) > 128)
1151
                            predicted_dc = vu;
1152
                        else if (FFABS(predicted_dc - vl) > 128)
1153
                            predicted_dc = vl;
1154
                        else if (FFABS(predicted_dc - vul) > 128)
1155
                            predicted_dc = vul;
1156
                    }
1157
                }
1158

    
1159
                /* at long last, apply the predictor */
1160
                DC_COEFF(i) += predicted_dc;
1161
                /* save the DC */
1162
                last_dc[current_frame_type] = DC_COEFF(i);
1163
            }
1164
        }
1165
    }
1166
}
1167

    
1168
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1169
{
1170
    int x, y;
1171
    int *bounding_values= s->bounding_values_array+127;
1172

    
1173
    int width           = s->fragment_width  >> !!plane;
1174
    int height          = s->fragment_height >> !!plane;
1175
    int fragment        = s->fragment_start        [plane] + ystart * width;
1176
    int stride          = s->current_frame.linesize[plane];
1177
    uint8_t *plane_data = s->current_frame.data    [plane];
1178
    if (!s->flipped_image) stride = -stride;
1179
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1180

    
1181
    for (y = ystart; y < yend; y++) {
1182

    
1183
        for (x = 0; x < width; x++) {
1184
            /* This code basically just deblocks on the edges of coded blocks.
1185
             * However, it has to be much more complicated because of the
1186
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1187
             * because some pixels get filtered twice. */
1188
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1189
            {
1190
                /* do not perform left edge filter for left columns frags */
1191
                if (x > 0) {
1192
                    s->dsp.vp3_h_loop_filter(
1193
                        plane_data + 8*x,
1194
                        stride, bounding_values);
1195
                }
1196

    
1197
                /* do not perform top edge filter for top row fragments */
1198
                if (y > 0) {
1199
                    s->dsp.vp3_v_loop_filter(
1200
                        plane_data + 8*x,
1201
                        stride, bounding_values);
1202
                }
1203

    
1204
                /* do not perform right edge filter for right column
1205
                 * fragments or if right fragment neighbor is also coded
1206
                 * in this frame (it will be filtered in next iteration) */
1207
                if ((x < width - 1) &&
1208
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1209
                    s->dsp.vp3_h_loop_filter(
1210
                        plane_data + 8*x + 8,
1211
                        stride, bounding_values);
1212
                }
1213

    
1214
                /* do not perform bottom edge filter for bottom row
1215
                 * fragments or if bottom fragment neighbor is also coded
1216
                 * in this frame (it will be filtered in the next row) */
1217
                if ((y < height - 1) &&
1218
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1219
                    s->dsp.vp3_v_loop_filter(
1220
                        plane_data + 8*x + 8*stride,
1221
                        stride, bounding_values);
1222
                }
1223
            }
1224

    
1225
            fragment++;
1226
        }
1227
        plane_data += 8*stride;
1228
    }
1229
}
1230

    
1231
/**
1232
 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1233
 * for the next block in coding order
1234
 */
1235
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1236
                              int plane, int inter, DCTELEM block[64])
1237
{
1238
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1239
    uint8_t *perm = s->scantable.permutated;
1240
    int i = 0;
1241

    
1242
    do {
1243
        int token = *s->dct_tokens[plane][i];
1244
        switch (token & 3) {
1245
        case 0: // EOB
1246
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1247
                s->dct_tokens[plane][i]++;
1248
            else
1249
                *s->dct_tokens[plane][i] = token & ~3;
1250
            goto end;
1251
        case 1: // zero run
1252
            s->dct_tokens[plane][i]++;
1253
            i += (token >> 2) & 0x7f;
1254
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1255
            i++;
1256
            break;
1257
        case 2: // coeff
1258
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1259
            s->dct_tokens[plane][i++]++;
1260
            break;
1261
        default:
1262
            av_log(s->avctx, AV_LOG_ERROR, "internal: invalid token type\n");
1263
            return i;
1264
        }
1265
    } while (i < 64);
1266
end:
1267
    // the actual DC+prediction is in the fragment structure
1268
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1269
    return i;
1270
}
1271

    
1272
/**
1273
 * called when all pixels up to row y are complete
1274
 */
1275
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1276
{
1277
    int h, cy;
1278
    int offset[4];
1279

    
1280
    if(s->avctx->draw_horiz_band==NULL)
1281
        return;
1282

    
1283
    h= y - s->last_slice_end;
1284
    y -= h;
1285

    
1286
    if (!s->flipped_image) {
1287
        if (y == 0)
1288
            h -= s->height - s->avctx->height;  // account for non-mod16
1289
        y = s->height - y - h;
1290
    }
1291

    
1292
    cy = y >> 1;
1293
    offset[0] = s->current_frame.linesize[0]*y;
1294
    offset[1] = s->current_frame.linesize[1]*cy;
1295
    offset[2] = s->current_frame.linesize[2]*cy;
1296
    offset[3] = 0;
1297

    
1298
    emms_c();
1299
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1300
    s->last_slice_end= y + h;
1301
}
1302

    
1303
/*
1304
 * Perform the final rendering for a particular slice of data.
1305
 * The slice number ranges from 0..(c_superblock_height - 1).
1306
 */
1307
static void render_slice(Vp3DecodeContext *s, int slice)
1308
{
1309
    int x, y, i, j;
1310
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1311
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1312
    int motion_halfpel_index;
1313
    uint8_t *motion_source;
1314
    int plane, first_pixel;
1315

    
1316
    if (slice >= s->c_superblock_height)
1317
        return;
1318

    
1319
    for (plane = 0; plane < 3; plane++) {
1320
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1321
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1322
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1323
        int stride            = s->current_frame.linesize[plane];
1324
        int plane_width       = s->width  >> !!plane;
1325
        int plane_height      = s->height >> !!plane;
1326

    
1327
        int sb_x, sb_y        = slice << !plane;
1328
        int slice_height      = sb_y + (plane ? 1 : 2);
1329
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1330

    
1331
        int fragment_width    = s->fragment_width  >> !!plane;
1332
        int fragment_height   = s->fragment_height >> !!plane;
1333
        int fragment_start    = s->fragment_start[plane];
1334

    
1335
        if (!s->flipped_image) stride = -stride;
1336
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1337
            continue;
1338

    
1339

    
1340
        if(FFABS(stride) > 2048)
1341
            return; //various tables are fixed size
1342

    
1343
        /* for each superblock row in the slice (both of them)... */
1344
        for (; sb_y < slice_height; sb_y++) {
1345

    
1346
            /* for each superblock in a row... */
1347
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1348

    
1349
                /* for each block in a superblock... */
1350
                for (j = 0; j < 16; j++) {
1351
                    x = 4*sb_x + hilbert_offset[j][0];
1352
                    y = 4*sb_y + hilbert_offset[j][1];
1353

    
1354
                    i = fragment_start + y*fragment_width + x;
1355

    
1356
                    // bounds check
1357
                    if (x >= fragment_width || y >= fragment_height)
1358
                        continue;
1359

    
1360
                first_pixel = 8*y*stride + 8*x;
1361

    
1362
                /* transform if this block was coded */
1363
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1364
                    int intra = s->all_fragments[i].coding_method == MODE_INTRA;
1365

    
1366
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1367
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1368
                        motion_source= golden_plane;
1369
                    else
1370
                        motion_source= last_plane;
1371

    
1372
                    motion_source += first_pixel;
1373
                    motion_halfpel_index = 0;
1374

    
1375
                    /* sort out the motion vector if this fragment is coded
1376
                     * using a motion vector method */
1377
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1378
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1379
                        int src_x, src_y;
1380
                        motion_x = s->all_fragments[i].motion_x;
1381
                        motion_y = s->all_fragments[i].motion_y;
1382
                        if(plane){
1383
                            motion_x= (motion_x>>1) | (motion_x&1);
1384
                            motion_y= (motion_y>>1) | (motion_y&1);
1385
                        }
1386

    
1387
                        src_x= (motion_x>>1) + 8*x;
1388
                        src_y= (motion_y>>1) + 8*y;
1389

    
1390
                        motion_halfpel_index = motion_x & 0x01;
1391
                        motion_source += (motion_x >> 1);
1392

    
1393
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1394
                        motion_source += ((motion_y >> 1) * stride);
1395

    
1396
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1397
                            uint8_t *temp= s->edge_emu_buffer;
1398
                            if(stride<0) temp -= 9*stride;
1399
                            else temp += 9*stride;
1400

    
1401
                            ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1402
                            motion_source= temp;
1403
                        }
1404
                    }
1405

    
1406

    
1407
                    /* first, take care of copying a block from either the
1408
                     * previous or the golden frame */
1409
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1410
                        /* Note, it is possible to implement all MC cases with
1411
                           put_no_rnd_pixels_l2 which would look more like the
1412
                           VP3 source but this would be slower as
1413
                           put_no_rnd_pixels_tab is better optimzed */
1414
                        if(motion_halfpel_index != 3){
1415
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1416
                                output_plane + first_pixel,
1417
                                motion_source, stride, 8);
1418
                        }else{
1419
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1420
                            s->dsp.put_no_rnd_pixels_l2[1](
1421
                                output_plane + first_pixel,
1422
                                motion_source - d,
1423
                                motion_source + stride + 1 + d,
1424
                                stride, 8);
1425
                        }
1426
                    }
1427

    
1428
                        s->dsp.clear_block(block);
1429
                        vp3_dequant(s, s->all_fragments + i, plane, !intra, block);
1430

    
1431
                    /* invert DCT and place (or add) in final output */
1432

    
1433
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1434
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1435
                            block[0] += 128<<3;
1436
                        s->dsp.idct_put(
1437
                            output_plane + first_pixel,
1438
                            stride,
1439
                            block);
1440
                    } else {
1441
                        s->dsp.idct_add(
1442
                            output_plane + first_pixel,
1443
                            stride,
1444
                            block);
1445
                    }
1446
                } else {
1447

    
1448
                    /* copy directly from the previous frame */
1449
                    s->dsp.put_pixels_tab[1][0](
1450
                        output_plane + first_pixel,
1451
                        last_plane + first_pixel,
1452
                        stride, 8);
1453

    
1454
                }
1455
                }
1456
            }
1457

    
1458
            // Filter up to the last row in the superblock row
1459
            apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1460
        }
1461
    }
1462

    
1463
     /* this looks like a good place for slice dispatch... */
1464
     /* algorithm:
1465
      *   if (slice == s->macroblock_height - 1)
1466
      *     dispatch (both last slice & 2nd-to-last slice);
1467
      *   else if (slice > 0)
1468
      *     dispatch (slice - 1);
1469
      */
1470

    
1471
    vp3_draw_horiz_band(s, FFMIN(64*slice + 64-16, s->height-16));
1472
}
1473

    
1474
/*
1475
 * This is the ffmpeg/libavcodec API init function.
1476
 */
1477
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1478
{
1479
    Vp3DecodeContext *s = avctx->priv_data;
1480
    int i, inter, plane;
1481
    int c_width;
1482
    int c_height;
1483

    
1484
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1485
        s->version = 0;
1486
    else
1487
        s->version = 1;
1488

    
1489
    s->avctx = avctx;
1490
    s->width = FFALIGN(avctx->width, 16);
1491
    s->height = FFALIGN(avctx->height, 16);
1492
    avctx->pix_fmt = PIX_FMT_YUV420P;
1493
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1494
    if(avctx->idct_algo==FF_IDCT_AUTO)
1495
        avctx->idct_algo=FF_IDCT_VP3;
1496
    dsputil_init(&s->dsp, avctx);
1497

    
1498
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1499

    
1500
    /* initialize to an impossible value which will force a recalculation
1501
     * in the first frame decode */
1502
    for (i = 0; i < 3; i++)
1503
        s->qps[i] = -1;
1504

    
1505
    s->y_superblock_width = (s->width + 31) / 32;
1506
    s->y_superblock_height = (s->height + 31) / 32;
1507
    s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1508

    
1509
    /* work out the dimensions for the C planes */
1510
    c_width = s->width / 2;
1511
    c_height = s->height / 2;
1512
    s->c_superblock_width = (c_width + 31) / 32;
1513
    s->c_superblock_height = (c_height + 31) / 32;
1514
    s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1515

    
1516
    s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1517
    s->u_superblock_start = s->y_superblock_count;
1518
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1519
    s->superblock_coding = av_malloc(s->superblock_count);
1520

    
1521
    s->macroblock_width = (s->width + 15) / 16;
1522
    s->macroblock_height = (s->height + 15) / 16;
1523
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1524

    
1525
    s->fragment_width = s->width / FRAGMENT_PIXELS;
1526
    s->fragment_height = s->height / FRAGMENT_PIXELS;
1527

    
1528
    /* fragment count covers all 8x8 blocks for all 3 planes */
1529
    s->fragment_count = s->fragment_width * s->fragment_height * 3 / 2;
1530
    s->fragment_start[1] = s->fragment_width * s->fragment_height;
1531
    s->fragment_start[2] = s->fragment_width * s->fragment_height * 5 / 4;
1532

    
1533
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1534
    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1535
    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1536
    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1537
        !s->coded_fragment_list[0]) {
1538
        vp3_decode_end(avctx);
1539
        return -1;
1540
    }
1541

    
1542
    if (!s->theora_tables)
1543
    {
1544
        for (i = 0; i < 64; i++) {
1545
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1546
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1547
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1548
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1549
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1550
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1551
        }
1552

    
1553
        for(inter=0; inter<2; inter++){
1554
            for(plane=0; plane<3; plane++){
1555
                s->qr_count[inter][plane]= 1;
1556
                s->qr_size [inter][plane][0]= 63;
1557
                s->qr_base [inter][plane][0]=
1558
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1559
            }
1560
        }
1561

    
1562
        /* init VLC tables */
1563
        for (i = 0; i < 16; i++) {
1564

    
1565
            /* DC histograms */
1566
            init_vlc(&s->dc_vlc[i], 5, 32,
1567
                &dc_bias[i][0][1], 4, 2,
1568
                &dc_bias[i][0][0], 4, 2, 0);
1569

    
1570
            /* group 1 AC histograms */
1571
            init_vlc(&s->ac_vlc_1[i], 5, 32,
1572
                &ac_bias_0[i][0][1], 4, 2,
1573
                &ac_bias_0[i][0][0], 4, 2, 0);
1574

    
1575
            /* group 2 AC histograms */
1576
            init_vlc(&s->ac_vlc_2[i], 5, 32,
1577
                &ac_bias_1[i][0][1], 4, 2,
1578
                &ac_bias_1[i][0][0], 4, 2, 0);
1579

    
1580
            /* group 3 AC histograms */
1581
            init_vlc(&s->ac_vlc_3[i], 5, 32,
1582
                &ac_bias_2[i][0][1], 4, 2,
1583
                &ac_bias_2[i][0][0], 4, 2, 0);
1584

    
1585
            /* group 4 AC histograms */
1586
            init_vlc(&s->ac_vlc_4[i], 5, 32,
1587
                &ac_bias_3[i][0][1], 4, 2,
1588
                &ac_bias_3[i][0][0], 4, 2, 0);
1589
        }
1590
    } else {
1591
        for (i = 0; i < 16; i++) {
1592

    
1593
            /* DC histograms */
1594
            if (init_vlc(&s->dc_vlc[i], 5, 32,
1595
                &s->huffman_table[i][0][1], 4, 2,
1596
                &s->huffman_table[i][0][0], 4, 2, 0) < 0)
1597
                goto vlc_fail;
1598

    
1599
            /* group 1 AC histograms */
1600
            if (init_vlc(&s->ac_vlc_1[i], 5, 32,
1601
                &s->huffman_table[i+16][0][1], 4, 2,
1602
                &s->huffman_table[i+16][0][0], 4, 2, 0) < 0)
1603
                goto vlc_fail;
1604

    
1605
            /* group 2 AC histograms */
1606
            if (init_vlc(&s->ac_vlc_2[i], 5, 32,
1607
                &s->huffman_table[i+16*2][0][1], 4, 2,
1608
                &s->huffman_table[i+16*2][0][0], 4, 2, 0) < 0)
1609
                goto vlc_fail;
1610

    
1611
            /* group 3 AC histograms */
1612
            if (init_vlc(&s->ac_vlc_3[i], 5, 32,
1613
                &s->huffman_table[i+16*3][0][1], 4, 2,
1614
                &s->huffman_table[i+16*3][0][0], 4, 2, 0) < 0)
1615
                goto vlc_fail;
1616

    
1617
            /* group 4 AC histograms */
1618
            if (init_vlc(&s->ac_vlc_4[i], 5, 32,
1619
                &s->huffman_table[i+16*4][0][1], 4, 2,
1620
                &s->huffman_table[i+16*4][0][0], 4, 2, 0) < 0)
1621
                goto vlc_fail;
1622
        }
1623
    }
1624

    
1625
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1626
        &superblock_run_length_vlc_table[0][1], 4, 2,
1627
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1628

    
1629
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1630
        &fragment_run_length_vlc_table[0][1], 4, 2,
1631
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1632

    
1633
    init_vlc(&s->mode_code_vlc, 3, 8,
1634
        &mode_code_vlc_table[0][1], 2, 1,
1635
        &mode_code_vlc_table[0][0], 2, 1, 0);
1636

    
1637
    init_vlc(&s->motion_vector_vlc, 6, 63,
1638
        &motion_vector_vlc_table[0][1], 2, 1,
1639
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1640

    
1641
    /* work out the block mapping tables */
1642
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1643
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1644
    if (!s->superblock_fragments || !s->macroblock_coding) {
1645
        vp3_decode_end(avctx);
1646
        return -1;
1647
    }
1648
    init_block_mapping(s);
1649

    
1650
    for (i = 0; i < 3; i++) {
1651
        s->current_frame.data[i] = NULL;
1652
        s->last_frame.data[i] = NULL;
1653
        s->golden_frame.data[i] = NULL;
1654
    }
1655

    
1656
    return 0;
1657

    
1658
vlc_fail:
1659
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1660
    return -1;
1661
}
1662

    
1663
/*
1664
 * This is the ffmpeg/libavcodec API frame decode function.
1665
 */
1666
static int vp3_decode_frame(AVCodecContext *avctx,
1667
                            void *data, int *data_size,
1668
                            AVPacket *avpkt)
1669
{
1670
    const uint8_t *buf = avpkt->data;
1671
    int buf_size = avpkt->size;
1672
    Vp3DecodeContext *s = avctx->priv_data;
1673
    GetBitContext gb;
1674
    static int counter = 0;
1675
    int i;
1676

    
1677
    init_get_bits(&gb, buf, buf_size * 8);
1678

    
1679
    if (s->theora && get_bits1(&gb))
1680
    {
1681
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1682
        return -1;
1683
    }
1684

    
1685
    s->keyframe = !get_bits1(&gb);
1686
    if (!s->theora)
1687
        skip_bits(&gb, 1);
1688
    for (i = 0; i < 3; i++)
1689
        s->last_qps[i] = s->qps[i];
1690

    
1691
    s->nqps=0;
1692
    do{
1693
        s->qps[s->nqps++]= get_bits(&gb, 6);
1694
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1695
    for (i = s->nqps; i < 3; i++)
1696
        s->qps[i] = -1;
1697

    
1698
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1699
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1700
            s->keyframe?"key":"", counter, s->qps[0]);
1701
    counter++;
1702

    
1703
    if (s->qps[0] != s->last_qps[0])
1704
        init_loop_filter(s);
1705

    
1706
    for (i = 0; i < s->nqps; i++)
1707
        // reinit all dequantizers if the first one changed, because
1708
        // the DC of the first quantizer must be used for all matrices
1709
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1710
            init_dequantizer(s, i);
1711

    
1712
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1713
        return buf_size;
1714

    
1715
    s->current_frame.reference = 3;
1716
    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
1717
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1718
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1719
        goto error;
1720
    }
1721

    
1722
    if (s->keyframe) {
1723
        if (!s->theora)
1724
        {
1725
            skip_bits(&gb, 4); /* width code */
1726
            skip_bits(&gb, 4); /* height code */
1727
            if (s->version)
1728
            {
1729
                s->version = get_bits(&gb, 5);
1730
                if (counter == 1)
1731
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1732
            }
1733
        }
1734
        if (s->version || s->theora)
1735
        {
1736
                if (get_bits1(&gb))
1737
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1738
            skip_bits(&gb, 2); /* reserved? */
1739
        }
1740
    } else {
1741
        if (!s->golden_frame.data[0]) {
1742
            av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1743

    
1744
            s->golden_frame.reference = 3;
1745
            s->golden_frame.pict_type = FF_I_TYPE;
1746
            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1747
                av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1748
                goto error;
1749
            }
1750
            s->last_frame = s->golden_frame;
1751
            s->last_frame.type = FF_BUFFER_TYPE_COPY;
1752
        }
1753
    }
1754

    
1755
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1756
    s->current_frame.qstride= 0;
1757

    
1758
    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1759

    
1760
    if (unpack_superblocks(s, &gb)){
1761
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1762
        goto error;
1763
    }
1764
    if (unpack_modes(s, &gb)){
1765
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1766
        goto error;
1767
    }
1768
    if (unpack_vectors(s, &gb)){
1769
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1770
        goto error;
1771
    }
1772
    if (unpack_block_qpis(s, &gb)){
1773
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1774
        goto error;
1775
    }
1776
    if (unpack_dct_coeffs(s, &gb)){
1777
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1778
        goto error;
1779
    }
1780

    
1781
    for (i = 0; i < 3; i++) {
1782
        if (s->flipped_image)
1783
            s->data_offset[i] = 0;
1784
        else
1785
            s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
1786
    }
1787

    
1788
    s->last_slice_end = 0;
1789
    for (i = 0; i < s->c_superblock_height; i++)
1790
        render_slice(s, i);
1791

    
1792
    // filter the last row
1793
    for (i = 0; i < 3; i++) {
1794
        int row = (s->height >> (3+!!i)) - 1;
1795
        apply_loop_filter(s, i, row, row+1);
1796
    }
1797
    vp3_draw_horiz_band(s, s->height);
1798

    
1799
    *data_size=sizeof(AVFrame);
1800
    *(AVFrame*)data= s->current_frame;
1801

    
1802
    /* release the last frame, if it is allocated and if it is not the
1803
     * golden frame */
1804
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1805
        avctx->release_buffer(avctx, &s->last_frame);
1806

    
1807
    /* shuffle frames (last = current) */
1808
    s->last_frame= s->current_frame;
1809

    
1810
    if (s->keyframe) {
1811
        if (s->golden_frame.data[0])
1812
            avctx->release_buffer(avctx, &s->golden_frame);
1813
        s->golden_frame = s->current_frame;
1814
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1815
    }
1816

    
1817
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1818

    
1819
    return buf_size;
1820

    
1821
error:
1822
    if (s->current_frame.data[0])
1823
        avctx->release_buffer(avctx, &s->current_frame);
1824
    return -1;
1825
}
1826

    
1827
/*
1828
 * This is the ffmpeg/libavcodec API module cleanup function.
1829
 */
1830
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1831
{
1832
    Vp3DecodeContext *s = avctx->priv_data;
1833
    int i;
1834

    
1835
    av_free(s->superblock_coding);
1836
    av_free(s->all_fragments);
1837
    av_free(s->coded_fragment_list[0]);
1838
    av_free(s->dct_tokens_base);
1839
    av_free(s->superblock_fragments);
1840
    av_free(s->macroblock_coding);
1841

    
1842
    for (i = 0; i < 16; i++) {
1843
        free_vlc(&s->dc_vlc[i]);
1844
        free_vlc(&s->ac_vlc_1[i]);
1845
        free_vlc(&s->ac_vlc_2[i]);
1846
        free_vlc(&s->ac_vlc_3[i]);
1847
        free_vlc(&s->ac_vlc_4[i]);
1848
    }
1849

    
1850
    free_vlc(&s->superblock_run_length_vlc);
1851
    free_vlc(&s->fragment_run_length_vlc);
1852
    free_vlc(&s->mode_code_vlc);
1853
    free_vlc(&s->motion_vector_vlc);
1854

    
1855
    /* release all frames */
1856
    if (s->golden_frame.data[0])
1857
        avctx->release_buffer(avctx, &s->golden_frame);
1858
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1859
        avctx->release_buffer(avctx, &s->last_frame);
1860
    /* no need to release the current_frame since it will always be pointing
1861
     * to the same frame as either the golden or last frame */
1862

    
1863
    return 0;
1864
}
1865

    
1866
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1867
{
1868
    Vp3DecodeContext *s = avctx->priv_data;
1869

    
1870
    if (get_bits1(gb)) {
1871
        int token;
1872
        if (s->entries >= 32) { /* overflow */
1873
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1874
            return -1;
1875
        }
1876
        token = get_bits(gb, 5);
1877
        //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);
1878
        s->huffman_table[s->hti][token][0] = s->hbits;
1879
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1880
        s->entries++;
1881
    }
1882
    else {
1883
        if (s->huff_code_size >= 32) {/* overflow */
1884
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1885
            return -1;
1886
        }
1887
        s->huff_code_size++;
1888
        s->hbits <<= 1;
1889
        if (read_huffman_tree(avctx, gb))
1890
            return -1;
1891
        s->hbits |= 1;
1892
        if (read_huffman_tree(avctx, gb))
1893
            return -1;
1894
        s->hbits >>= 1;
1895
        s->huff_code_size--;
1896
    }
1897
    return 0;
1898
}
1899

    
1900
#if CONFIG_THEORA_DECODER
1901
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1902
{
1903
    Vp3DecodeContext *s = avctx->priv_data;
1904
    int visible_width, visible_height, colorspace;
1905

    
1906
    s->theora = get_bits_long(gb, 24);
1907
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1908

    
1909
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1910
    /* but previous versions have the image flipped relative to vp3 */
1911
    if (s->theora < 0x030200)
1912
    {
1913
        s->flipped_image = 1;
1914
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1915
    }
1916

    
1917
    visible_width  = s->width  = get_bits(gb, 16) << 4;
1918
    visible_height = s->height = get_bits(gb, 16) << 4;
1919

    
1920
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
1921
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
1922
        s->width= s->height= 0;
1923
        return -1;
1924
    }
1925

    
1926
    if (s->theora >= 0x030200) {
1927
        visible_width  = get_bits_long(gb, 24);
1928
        visible_height = get_bits_long(gb, 24);
1929

    
1930
        skip_bits(gb, 8); /* offset x */
1931
        skip_bits(gb, 8); /* offset y */
1932
    }
1933

    
1934
    skip_bits(gb, 32); /* fps numerator */
1935
    skip_bits(gb, 32); /* fps denumerator */
1936
    skip_bits(gb, 24); /* aspect numerator */
1937
    skip_bits(gb, 24); /* aspect denumerator */
1938

    
1939
    if (s->theora < 0x030200)
1940
        skip_bits(gb, 5); /* keyframe frequency force */
1941
    colorspace = get_bits(gb, 8);
1942
    skip_bits(gb, 24); /* bitrate */
1943

    
1944
    skip_bits(gb, 6); /* quality hint */
1945

    
1946
    if (s->theora >= 0x030200)
1947
    {
1948
        skip_bits(gb, 5); /* keyframe frequency force */
1949
        skip_bits(gb, 2); /* pixel format: 420,res,422,444 */
1950
        skip_bits(gb, 3); /* reserved */
1951
    }
1952

    
1953
//    align_get_bits(gb);
1954

    
1955
    if (   visible_width  <= s->width  && visible_width  > s->width-16
1956
        && visible_height <= s->height && visible_height > s->height-16)
1957
        avcodec_set_dimensions(avctx, visible_width, visible_height);
1958
    else
1959
        avcodec_set_dimensions(avctx, s->width, s->height);
1960

    
1961
    if (colorspace == 1) {
1962
        avctx->color_primaries = AVCOL_PRI_BT470M;
1963
    } else if (colorspace == 2) {
1964
        avctx->color_primaries = AVCOL_PRI_BT470BG;
1965
    }
1966
    if (colorspace == 1 || colorspace == 2) {
1967
        avctx->colorspace = AVCOL_SPC_BT470BG;
1968
        avctx->color_trc  = AVCOL_TRC_BT709;
1969
    }
1970

    
1971
    return 0;
1972
}
1973

    
1974
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
1975
{
1976
    Vp3DecodeContext *s = avctx->priv_data;
1977
    int i, n, matrices, inter, plane;
1978

    
1979
    if (s->theora >= 0x030200) {
1980
        n = get_bits(gb, 3);
1981
        /* loop filter limit values table */
1982
        for (i = 0; i < 64; i++) {
1983
            s->filter_limit_values[i] = get_bits(gb, n);
1984
            if (s->filter_limit_values[i] > 127) {
1985
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
1986
                s->filter_limit_values[i] = 127;
1987
            }
1988
        }
1989
    }
1990

    
1991
    if (s->theora >= 0x030200)
1992
        n = get_bits(gb, 4) + 1;
1993
    else
1994
        n = 16;
1995
    /* quality threshold table */
1996
    for (i = 0; i < 64; i++)
1997
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
1998

    
1999
    if (s->theora >= 0x030200)
2000
        n = get_bits(gb, 4) + 1;
2001
    else
2002
        n = 16;
2003
    /* dc scale factor table */
2004
    for (i = 0; i < 64; i++)
2005
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2006

    
2007
    if (s->theora >= 0x030200)
2008
        matrices = get_bits(gb, 9) + 1;
2009
    else
2010
        matrices = 3;
2011

    
2012
    if(matrices > 384){
2013
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2014
        return -1;
2015
    }
2016

    
2017
    for(n=0; n<matrices; n++){
2018
        for (i = 0; i < 64; i++)
2019
            s->base_matrix[n][i]= get_bits(gb, 8);
2020
    }
2021

    
2022
    for (inter = 0; inter <= 1; inter++) {
2023
        for (plane = 0; plane <= 2; plane++) {
2024
            int newqr= 1;
2025
            if (inter || plane > 0)
2026
                newqr = get_bits1(gb);
2027
            if (!newqr) {
2028
                int qtj, plj;
2029
                if(inter && get_bits1(gb)){
2030
                    qtj = 0;
2031
                    plj = plane;
2032
                }else{
2033
                    qtj= (3*inter + plane - 1) / 3;
2034
                    plj= (plane + 2) % 3;
2035
                }
2036
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2037
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2038
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2039
            } else {
2040
                int qri= 0;
2041
                int qi = 0;
2042

    
2043
                for(;;){
2044
                    i= get_bits(gb, av_log2(matrices-1)+1);
2045
                    if(i>= matrices){
2046
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2047
                        return -1;
2048
                    }
2049
                    s->qr_base[inter][plane][qri]= i;
2050
                    if(qi >= 63)
2051
                        break;
2052
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2053
                    s->qr_size[inter][plane][qri++]= i;
2054
                    qi += i;
2055
                }
2056

    
2057
                if (qi > 63) {
2058
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2059
                    return -1;
2060
                }
2061
                s->qr_count[inter][plane]= qri;
2062
            }
2063
        }
2064
    }
2065

    
2066
    /* Huffman tables */
2067
    for (s->hti = 0; s->hti < 80; s->hti++) {
2068
        s->entries = 0;
2069
        s->huff_code_size = 1;
2070
        if (!get_bits1(gb)) {
2071
            s->hbits = 0;
2072
            if(read_huffman_tree(avctx, gb))
2073
                return -1;
2074
            s->hbits = 1;
2075
            if(read_huffman_tree(avctx, gb))
2076
                return -1;
2077
        }
2078
    }
2079

    
2080
    s->theora_tables = 1;
2081

    
2082
    return 0;
2083
}
2084

    
2085
static av_cold int theora_decode_init(AVCodecContext *avctx)
2086
{
2087
    Vp3DecodeContext *s = avctx->priv_data;
2088
    GetBitContext gb;
2089
    int ptype;
2090
    uint8_t *header_start[3];
2091
    int header_len[3];
2092
    int i;
2093

    
2094
    s->theora = 1;
2095

    
2096
    if (!avctx->extradata_size)
2097
    {
2098
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2099
        return -1;
2100
    }
2101

    
2102
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2103
                              42, header_start, header_len) < 0) {
2104
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2105
        return -1;
2106
    }
2107

    
2108
  for(i=0;i<3;i++) {
2109
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2110

    
2111
    ptype = get_bits(&gb, 8);
2112

    
2113
     if (!(ptype & 0x80))
2114
     {
2115
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2116
//        return -1;
2117
     }
2118

    
2119
    // FIXME: Check for this as well.
2120
    skip_bits_long(&gb, 6*8); /* "theora" */
2121

    
2122
    switch(ptype)
2123
    {
2124
        case 0x80:
2125
            theora_decode_header(avctx, &gb);
2126
                break;
2127
        case 0x81:
2128
// FIXME: is this needed? it breaks sometimes
2129
//            theora_decode_comments(avctx, gb);
2130
            break;
2131
        case 0x82:
2132
            if (theora_decode_tables(avctx, &gb))
2133
                return -1;
2134
            break;
2135
        default:
2136
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2137
            break;
2138
    }
2139
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2140
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2141
    if (s->theora < 0x030200)
2142
        break;
2143
  }
2144

    
2145
    return vp3_decode_init(avctx);
2146
}
2147

    
2148
AVCodec theora_decoder = {
2149
    "theora",
2150
    CODEC_TYPE_VIDEO,
2151
    CODEC_ID_THEORA,
2152
    sizeof(Vp3DecodeContext),
2153
    theora_decode_init,
2154
    NULL,
2155
    vp3_decode_end,
2156
    vp3_decode_frame,
2157
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2158
    NULL,
2159
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2160
};
2161
#endif
2162

    
2163
AVCodec vp3_decoder = {
2164
    "vp3",
2165
    CODEC_TYPE_VIDEO,
2166
    CODEC_ID_VP3,
2167
    sizeof(Vp3DecodeContext),
2168
    vp3_decode_init,
2169
    NULL,
2170
    vp3_decode_end,
2171
    vp3_decode_frame,
2172
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2173
    NULL,
2174
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2175
};