Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ dc4a1883

History | View | Annotate | Download (77.2 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
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 "libavcore/imgutils.h"
37
#include "avcodec.h"
38
#include "dsputil.h"
39
#include "get_bits.h"
40

    
41
#include "vp3data.h"
42
#include "xiph.h"
43

    
44
#define FRAGMENT_PIXELS 8
45

    
46
static av_cold int vp3_decode_end(AVCodecContext *avctx);
47

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

    
55
#define SB_NOT_CODED        0
56
#define SB_PARTIALLY_CODED  1
57
#define SB_FULLY_CODED      2
58

    
59
// This is the maximum length of a single long bit run that can be encoded
60
// for superblock coding or block qps. Theora special-cases this to read a
61
// bit instead of flipping the current bit to allow for runs longer than 4129.
62
#define MAXIMUM_LONG_BIT_RUN 4129
63

    
64
#define MODE_INTER_NO_MV      0
65
#define MODE_INTRA            1
66
#define MODE_INTER_PLUS_MV    2
67
#define MODE_INTER_LAST_MV    3
68
#define MODE_INTER_PRIOR_LAST 4
69
#define MODE_USING_GOLDEN     5
70
#define MODE_GOLDEN_MV        6
71
#define MODE_INTER_FOURMV     7
72
#define CODING_MODE_COUNT     8
73

    
74
/* special internal mode */
75
#define MODE_COPY             8
76

    
77
/* There are 6 preset schemes, plus a free-form scheme */
78
static const int ModeAlphabet[6][CODING_MODE_COUNT] =
79
{
80
    /* scheme 1: Last motion vector dominates */
81
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
82
         MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
83
         MODE_INTRA,            MODE_USING_GOLDEN,
84
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
85

    
86
    /* scheme 2 */
87
    {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
88
         MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
89
         MODE_INTRA,            MODE_USING_GOLDEN,
90
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
91

    
92
    /* scheme 3 */
93
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
94
         MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
95
         MODE_INTRA,            MODE_USING_GOLDEN,
96
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
97

    
98
    /* scheme 4 */
99
    {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
100
         MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
101
         MODE_INTRA,            MODE_USING_GOLDEN,
102
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
103

    
104
    /* scheme 5: No motion vector dominates */
105
    {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
106
         MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
107
         MODE_INTRA,            MODE_USING_GOLDEN,
108
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
109

    
110
    /* scheme 6 */
111
    {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
112
         MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
113
         MODE_INTER_PLUS_MV,    MODE_INTRA,
114
         MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
115

    
116
};
117

    
118
static const uint8_t hilbert_offset[16][2] = {
119
    {0,0}, {1,0}, {1,1}, {0,1},
120
    {0,2}, {0,3}, {1,3}, {1,2},
121
    {2,2}, {2,3}, {3,3}, {3,2},
122
    {3,1}, {2,1}, {2,0}, {3,0}
123
};
124

    
125
#define MIN_DEQUANT_VAL 2
126

    
127
typedef struct Vp3DecodeContext {
128
    AVCodecContext *avctx;
129
    int theora, theora_tables;
130
    int version;
131
    int width, height;
132
    int chroma_x_shift, chroma_y_shift;
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
    int skip_loop_filter;
141

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

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

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

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

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

    
169
    int8_t (*motion_val[2])[2];
170

    
171
    ScanTable scantable;
172

    
173
    /* tables */
174
    uint16_t coded_dc_scale_factor[64];
175
    uint32_t coded_ac_scale_factor[64];
176
    uint8_t base_matrix[384][64];
177
    uint8_t qr_count[2][3];
178
    uint8_t qr_size [2][3][64];
179
    uint16_t qr_base[2][3][64];
180

    
181
    /**
182
     * This is a list of all tokens in bitstream order. Reordering takes place
183
     * by pulling from each level during IDCT. As a consequence, IDCT must be
184
     * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
185
     * otherwise. The 32 different tokens with up to 12 bits of extradata are
186
     * collapsed into 3 types, packed as follows:
187
     *   (from the low to high bits)
188
     *
189
     * 2 bits: type (0,1,2)
190
     *   0: EOB run, 14 bits for run length (12 needed)
191
     *   1: zero run, 7 bits for run length
192
     *                7 bits for the next coefficient (3 needed)
193
     *   2: coefficient, 14 bits (11 needed)
194
     *
195
     * Coefficients are signed, so are packed in the highest bits for automatic
196
     * sign extension.
197
     */
198
    int16_t *dct_tokens[3][64];
199
    int16_t *dct_tokens_base;
200
#define TOKEN_EOB(eob_run)              ((eob_run) << 2)
201
#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
202
#define TOKEN_COEFF(coeff)              (((coeff) << 2) + 2)
203

    
204
    /**
205
     * number of blocks that contain DCT coefficients at the given level or higher
206
     */
207
    int num_coded_frags[3][64];
208
    int total_num_coded_frags;
209

    
210
    /* this is a list of indexes into the all_fragments array indicating
211
     * which of the fragments are coded */
212
    int *coded_fragment_list[3];
213

    
214
    VLC dc_vlc[16];
215
    VLC ac_vlc_1[16];
216
    VLC ac_vlc_2[16];
217
    VLC ac_vlc_3[16];
218
    VLC ac_vlc_4[16];
219

    
220
    VLC superblock_run_length_vlc;
221
    VLC fragment_run_length_vlc;
222
    VLC mode_code_vlc;
223
    VLC motion_vector_vlc;
224

    
225
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
226
     * index into them */
227
    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
228

    
229
    /* This table contains superblock_count * 16 entries. Each set of 16
230
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
231
     * An entry will be -1 to indicate that no entry corresponds to that
232
     * index. */
233
    int *superblock_fragments;
234

    
235
    /* This is an array that indicates how a particular macroblock
236
     * is coded. */
237
    unsigned char *macroblock_coding;
238

    
239
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
240
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
241

    
242
    /* Huffman decode */
243
    int hti;
244
    unsigned int hbits;
245
    int entries;
246
    int huff_code_size;
247
    uint32_t huffman_table[80][32][2];
248

    
249
    uint8_t filter_limit_values[64];
250
    DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
251
} Vp3DecodeContext;
252

    
253
/************************************************************************
254
 * VP3 specific functions
255
 ************************************************************************/
256

    
257
/*
258
 * This function sets up all of the various blocks mappings:
259
 * superblocks <-> fragments, macroblocks <-> fragments,
260
 * superblocks <-> macroblocks
261
 *
262
 * @return 0 is successful; returns 1 if *anything* went wrong.
263
 */
264
static int init_block_mapping(Vp3DecodeContext *s)
265
{
266
    int sb_x, sb_y, plane;
267
    int x, y, i, j = 0;
268

    
269
    for (plane = 0; plane < 3; plane++) {
270
        int sb_width    = plane ? s->c_superblock_width  : s->y_superblock_width;
271
        int sb_height   = plane ? s->c_superblock_height : s->y_superblock_height;
272
        int frag_width  = s->fragment_width[!!plane];
273
        int frag_height = s->fragment_height[!!plane];
274

    
275
        for (sb_y = 0; sb_y < sb_height; sb_y++)
276
            for (sb_x = 0; sb_x < sb_width; sb_x++)
277
                for (i = 0; i < 16; i++) {
278
                    x = 4*sb_x + hilbert_offset[i][0];
279
                    y = 4*sb_y + hilbert_offset[i][1];
280

    
281
                    if (x < frag_width && y < frag_height)
282
                        s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
283
                    else
284
                        s->superblock_fragments[j++] = -1;
285
                }
286
    }
287

    
288
    return 0;  /* successful path out */
289
}
290

    
291
/*
292
 * This function sets up the dequantization tables used for a particular
293
 * frame.
294
 */
295
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
296
{
297
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
298
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
299
    int i, plane, inter, qri, bmi, bmj, qistart;
300

    
301
    for(inter=0; inter<2; inter++){
302
        for(plane=0; plane<3; plane++){
303
            int sum=0;
304
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
305
                sum+= s->qr_size[inter][plane][qri];
306
                if(s->qps[qpi] <= sum)
307
                    break;
308
            }
309
            qistart= sum - s->qr_size[inter][plane][qri];
310
            bmi= s->qr_base[inter][plane][qri  ];
311
            bmj= s->qr_base[inter][plane][qri+1];
312
            for(i=0; i<64; i++){
313
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
314
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
315
                            + s->qr_size[inter][plane][qri])
316
                           / (2*s->qr_size[inter][plane][qri]);
317

    
318
                int qmin= 8<<(inter + !i);
319
                int qscale= i ? ac_scale_factor : dc_scale_factor;
320

    
321
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
322
            }
323
            // all DC coefficients use the same quant so as not to interfere with DC prediction
324
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
325
        }
326
    }
327

    
328
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
329
}
330

    
331
/*
332
 * This function initializes the loop filter boundary limits if the frame's
333
 * quality index is different from the previous frame's.
334
 *
335
 * The filter_limit_values may not be larger than 127.
336
 */
337
static void init_loop_filter(Vp3DecodeContext *s)
338
{
339
    int *bounding_values= s->bounding_values_array+127;
340
    int filter_limit;
341
    int x;
342
    int value;
343

    
344
    filter_limit = s->filter_limit_values[s->qps[0]];
345

    
346
    /* set up the bounding values */
347
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
348
    for (x = 0; x < filter_limit; x++) {
349
        bounding_values[-x] = -x;
350
        bounding_values[x] = x;
351
    }
352
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
353
        bounding_values[ x] =  value;
354
        bounding_values[-x] = -value;
355
    }
356
    if (value)
357
        bounding_values[128] = value;
358
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
359
}
360

    
361
/*
362
 * This function unpacks all of the superblock/macroblock/fragment coding
363
 * information from the bitstream.
364
 */
365
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
366
{
367
    int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
368
    int bit = 0;
369
    int current_superblock = 0;
370
    int current_run = 0;
371
    int num_partial_superblocks = 0;
372

    
373
    int i, j;
374
    int current_fragment;
375
    int plane;
376

    
377
    if (s->keyframe) {
378
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
379

    
380
    } else {
381

    
382
        /* unpack the list of partially-coded superblocks */
383
        bit = get_bits1(gb) ^ 1;
384
        current_run = 0;
385

    
386
        while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
387
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
388
                bit = get_bits1(gb);
389
            else
390
                bit ^= 1;
391

    
392
                current_run = get_vlc2(gb,
393
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
394
                if (current_run == 34)
395
                    current_run += get_bits(gb, 12);
396

    
397
            if (current_superblock + current_run > s->superblock_count) {
398
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
399
                return -1;
400
            }
401

    
402
            memset(s->superblock_coding + current_superblock, bit, current_run);
403

    
404
            current_superblock += current_run;
405
            if (bit)
406
                num_partial_superblocks += current_run;
407
        }
408

    
409
        /* unpack the list of fully coded superblocks if any of the blocks were
410
         * not marked as partially coded in the previous step */
411
        if (num_partial_superblocks < s->superblock_count) {
412
            int superblocks_decoded = 0;
413

    
414
            current_superblock = 0;
415
            bit = get_bits1(gb) ^ 1;
416
            current_run = 0;
417

    
418
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks
419
                   && get_bits_left(gb) > 0) {
420

    
421
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
422
                    bit = get_bits1(gb);
423
                else
424
                    bit ^= 1;
425

    
426
                        current_run = get_vlc2(gb,
427
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
428
                        if (current_run == 34)
429
                            current_run += get_bits(gb, 12);
430

    
431
                for (j = 0; j < current_run; current_superblock++) {
432
                    if (current_superblock >= s->superblock_count) {
433
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
434
                        return -1;
435
                    }
436

    
437
                /* skip any superblocks already marked as partially coded */
438
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
439
                    s->superblock_coding[current_superblock] = 2*bit;
440
                    j++;
441
                }
442
                }
443
                superblocks_decoded += current_run;
444
            }
445
        }
446

    
447
        /* if there were partial blocks, initialize bitstream for
448
         * unpacking fragment codings */
449
        if (num_partial_superblocks) {
450

    
451
            current_run = 0;
452
            bit = get_bits1(gb);
453
            /* toggle the bit because as soon as the first run length is
454
             * fetched the bit will be toggled again */
455
            bit ^= 1;
456
        }
457
    }
458

    
459
    /* figure out which fragments are coded; iterate through each
460
     * superblock (all planes) */
461
    s->total_num_coded_frags = 0;
462
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
463

    
464
    for (plane = 0; plane < 3; plane++) {
465
        int sb_start = superblock_starts[plane];
466
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
467
        int num_coded_frags = 0;
468

    
469
    for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
470

    
471
        /* iterate through all 16 fragments in a superblock */
472
        for (j = 0; j < 16; j++) {
473

    
474
            /* if the fragment is in bounds, check its coding status */
475
            current_fragment = s->superblock_fragments[i * 16 + j];
476
            if (current_fragment != -1) {
477
                int coded = s->superblock_coding[i];
478

    
479
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
480

    
481
                    /* fragment may or may not be coded; this is the case
482
                     * that cares about the fragment coding runs */
483
                    if (current_run-- == 0) {
484
                        bit ^= 1;
485
                        current_run = get_vlc2(gb,
486
                            s->fragment_run_length_vlc.table, 5, 2);
487
                    }
488
                    coded = bit;
489
                }
490

    
491
                    if (coded) {
492
                        /* default mode; actual mode will be decoded in
493
                         * the next phase */
494
                        s->all_fragments[current_fragment].coding_method =
495
                            MODE_INTER_NO_MV;
496
                        s->coded_fragment_list[plane][num_coded_frags++] =
497
                            current_fragment;
498
                    } else {
499
                        /* not coded; copy this fragment from the prior frame */
500
                        s->all_fragments[current_fragment].coding_method =
501
                            MODE_COPY;
502
                    }
503
            }
504
        }
505
    }
506
        s->total_num_coded_frags += num_coded_frags;
507
        for (i = 0; i < 64; i++)
508
            s->num_coded_frags[plane][i] = num_coded_frags;
509
        if (plane < 2)
510
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
511
    }
512
    return 0;
513
}
514

    
515
/*
516
 * This function unpacks all the coding mode data for individual macroblocks
517
 * from the bitstream.
518
 */
519
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
520
{
521
    int i, j, k, sb_x, sb_y;
522
    int scheme;
523
    int current_macroblock;
524
    int current_fragment;
525
    int coding_mode;
526
    int custom_mode_alphabet[CODING_MODE_COUNT];
527
    const int *alphabet;
528
    Vp3Fragment *frag;
529

    
530
    if (s->keyframe) {
531
        for (i = 0; i < s->fragment_count; i++)
532
            s->all_fragments[i].coding_method = MODE_INTRA;
533

    
534
    } else {
535

    
536
        /* fetch the mode coding scheme for this frame */
537
        scheme = get_bits(gb, 3);
538

    
539
        /* is it a custom coding scheme? */
540
        if (scheme == 0) {
541
            for (i = 0; i < 8; i++)
542
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
543
            for (i = 0; i < 8; i++)
544
                custom_mode_alphabet[get_bits(gb, 3)] = i;
545
            alphabet = custom_mode_alphabet;
546
        } else
547
            alphabet = ModeAlphabet[scheme-1];
548

    
549
        /* iterate through all of the macroblocks that contain 1 or more
550
         * coded fragments */
551
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
552
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
553
                if (get_bits_left(gb) <= 0)
554
                    return -1;
555

    
556
            for (j = 0; j < 4; j++) {
557
                int mb_x = 2*sb_x +   (j>>1);
558
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
559
                current_macroblock = mb_y * s->macroblock_width + mb_x;
560

    
561
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
562
                    continue;
563

    
564
#define BLOCK_X (2*mb_x + (k&1))
565
#define BLOCK_Y (2*mb_y + (k>>1))
566
                /* coding modes are only stored if the macroblock has at least one
567
                 * luma block coded, otherwise it must be INTER_NO_MV */
568
                for (k = 0; k < 4; k++) {
569
                    current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
570
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
571
                        break;
572
                }
573
                if (k == 4) {
574
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
575
                    continue;
576
                }
577

    
578
                /* mode 7 means get 3 bits for each coding mode */
579
                if (scheme == 7)
580
                    coding_mode = get_bits(gb, 3);
581
                else
582
                    coding_mode = alphabet
583
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
584

    
585
                s->macroblock_coding[current_macroblock] = coding_mode;
586
                for (k = 0; k < 4; k++) {
587
                    frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
588
                    if (frag->coding_method != MODE_COPY)
589
                        frag->coding_method = coding_mode;
590
                }
591

    
592
#define SET_CHROMA_MODES \
593
    if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
594
        frag[s->fragment_start[1]].coding_method = coding_mode;\
595
    if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
596
        frag[s->fragment_start[2]].coding_method = coding_mode;
597

    
598
                if (s->chroma_y_shift) {
599
                    frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
600
                    SET_CHROMA_MODES
601
                } else if (s->chroma_x_shift) {
602
                    frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
603
                    for (k = 0; k < 2; k++) {
604
                        SET_CHROMA_MODES
605
                        frag += s->fragment_width[1];
606
                    }
607
                } else {
608
                    for (k = 0; k < 4; k++) {
609
                        frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
610
                        SET_CHROMA_MODES
611
                    }
612
                }
613
            }
614
            }
615
        }
616
    }
617

    
618
    return 0;
619
}
620

    
621
/*
622
 * This function unpacks all the motion vectors for the individual
623
 * macroblocks from the bitstream.
624
 */
625
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
626
{
627
    int j, k, sb_x, sb_y;
628
    int coding_mode;
629
    int motion_x[4];
630
    int motion_y[4];
631
    int last_motion_x = 0;
632
    int last_motion_y = 0;
633
    int prior_last_motion_x = 0;
634
    int prior_last_motion_y = 0;
635
    int current_macroblock;
636
    int current_fragment;
637
    int frag;
638

    
639
    if (s->keyframe)
640
        return 0;
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
            if (get_bits_left(gb) <= 0)
650
                return -1;
651

    
652
        for (j = 0; j < 4; j++) {
653
            int mb_x = 2*sb_x +   (j>>1);
654
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
655
            current_macroblock = mb_y * s->macroblock_width + mb_x;
656

    
657
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
658
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
659
                continue;
660

    
661
            switch (s->macroblock_coding[current_macroblock]) {
662

    
663
            case MODE_INTER_PLUS_MV:
664
            case MODE_GOLDEN_MV:
665
                /* all 6 fragments use the same motion vector */
666
                if (coding_mode == 0) {
667
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
669
                } else {
670
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
671
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
672
                }
673

    
674
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
675
                if (s->macroblock_coding[current_macroblock] ==
676
                    MODE_INTER_PLUS_MV) {
677
                    prior_last_motion_x = last_motion_x;
678
                    prior_last_motion_y = last_motion_y;
679
                    last_motion_x = motion_x[0];
680
                    last_motion_y = motion_y[0];
681
                }
682
                break;
683

    
684
            case MODE_INTER_FOURMV:
685
                /* vector maintenance */
686
                prior_last_motion_x = last_motion_x;
687
                prior_last_motion_y = last_motion_y;
688

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

    
710
            case MODE_INTER_LAST_MV:
711
                /* all 6 fragments use the last motion vector */
712
                motion_x[0] = last_motion_x;
713
                motion_y[0] = last_motion_y;
714

    
715
                /* no vector maintenance (last vector remains the
716
                 * last vector) */
717
                break;
718

    
719
            case MODE_INTER_PRIOR_LAST:
720
                /* all 6 fragments use the motion vector prior to the
721
                 * last motion vector */
722
                motion_x[0] = prior_last_motion_x;
723
                motion_y[0] = prior_last_motion_y;
724

    
725
                /* vector maintenance */
726
                prior_last_motion_x = last_motion_x;
727
                prior_last_motion_y = last_motion_y;
728
                last_motion_x = motion_x[0];
729
                last_motion_y = motion_y[0];
730
                break;
731

    
732
            default:
733
                /* covers intra, inter without MV, golden without MV */
734
                motion_x[0] = 0;
735
                motion_y[0] = 0;
736

    
737
                /* no vector maintenance */
738
                break;
739
            }
740

    
741
            /* assign the motion vectors to the correct fragments */
742
            for (k = 0; k < 4; k++) {
743
                current_fragment =
744
                    BLOCK_Y*s->fragment_width[0] + BLOCK_X;
745
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
746
                    s->motion_val[0][current_fragment][0] = motion_x[k];
747
                    s->motion_val[0][current_fragment][1] = motion_y[k];
748
                } else {
749
                    s->motion_val[0][current_fragment][0] = motion_x[0];
750
                    s->motion_val[0][current_fragment][1] = motion_y[0];
751
                }
752
            }
753

    
754
            if (s->chroma_y_shift) {
755
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
756
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
757
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
758
                }
759
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
760
                motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
761
                frag = mb_y*s->fragment_width[1] + mb_x;
762
                s->motion_val[1][frag][0] = motion_x[0];
763
                s->motion_val[1][frag][1] = motion_y[0];
764
            } else if (s->chroma_x_shift) {
765
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
766
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
767
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
768
                    motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
769
                    motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
770
                } else {
771
                    motion_x[1] = motion_x[0];
772
                    motion_y[1] = motion_y[0];
773
                }
774
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
775
                motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
776

    
777
                frag = 2*mb_y*s->fragment_width[1] + mb_x;
778
                for (k = 0; k < 2; k++) {
779
                    s->motion_val[1][frag][0] = motion_x[k];
780
                    s->motion_val[1][frag][1] = motion_y[k];
781
                    frag += s->fragment_width[1];
782
                }
783
            } else {
784
                for (k = 0; k < 4; k++) {
785
                    frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
786
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
787
                        s->motion_val[1][frag][0] = motion_x[k];
788
                        s->motion_val[1][frag][1] = motion_y[k];
789
                    } else {
790
                        s->motion_val[1][frag][0] = motion_x[0];
791
                        s->motion_val[1][frag][1] = motion_y[0];
792
                    }
793
                }
794
            }
795
        }
796
        }
797
    }
798

    
799
    return 0;
800
}
801

    
802
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
803
{
804
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
805
    int num_blocks = s->total_num_coded_frags;
806

    
807
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
808
        i = blocks_decoded = num_blocks_at_qpi = 0;
809

    
810
        bit = get_bits1(gb) ^ 1;
811
        run_length = 0;
812

    
813
        do {
814
            if (run_length == MAXIMUM_LONG_BIT_RUN)
815
                bit = get_bits1(gb);
816
            else
817
                bit ^= 1;
818

    
819
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
820
            if (run_length == 34)
821
                run_length += get_bits(gb, 12);
822
            blocks_decoded += run_length;
823

    
824
            if (!bit)
825
                num_blocks_at_qpi += run_length;
826

    
827
            for (j = 0; j < run_length; i++) {
828
                if (i >= s->total_num_coded_frags)
829
                    return -1;
830

    
831
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
832
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
833
                    j++;
834
                }
835
            }
836
        } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
837

    
838
        num_blocks -= num_blocks_at_qpi;
839
    }
840

    
841
    return 0;
842
}
843

    
844
/*
845
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
846
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
847
 * data. This function unpacks all the VLCs for either the Y plane or both
848
 * C planes, and is called for DC coefficients or different AC coefficient
849
 * levels (since different coefficient types require different VLC tables.
850
 *
851
 * This function returns a residual eob run. E.g, if a particular token gave
852
 * instructions to EOB the next 5 fragments and there were only 2 fragments
853
 * left in the current fragment range, 3 would be returned so that it could
854
 * be passed into the next call to this same function.
855
 */
856
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
857
                        VLC *table, int coeff_index,
858
                        int plane,
859
                        int eob_run)
860
{
861
    int i, j = 0;
862
    int token;
863
    int zero_run = 0;
864
    DCTELEM coeff = 0;
865
    int bits_to_get;
866
    int blocks_ended;
867
    int coeff_i = 0;
868
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
869
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
870

    
871
    /* local references to structure members to avoid repeated deferences */
872
    int *coded_fragment_list = s->coded_fragment_list[plane];
873
    Vp3Fragment *all_fragments = s->all_fragments;
874
    VLC_TYPE (*vlc_table)[2] = table->table;
875

    
876
    if (num_coeffs < 0)
877
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
878

    
879
    if (eob_run > num_coeffs) {
880
        coeff_i = blocks_ended = num_coeffs;
881
        eob_run -= num_coeffs;
882
    } else {
883
        coeff_i = blocks_ended = eob_run;
884
        eob_run = 0;
885
    }
886

    
887
    // insert fake EOB token to cover the split between planes or zzi
888
    if (blocks_ended)
889
        dct_tokens[j++] = blocks_ended << 2;
890

    
891
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
892
            /* decode a VLC into a token */
893
            token = get_vlc2(gb, vlc_table, 11, 3);
894
            /* use the token to get a zero run, a coefficient, and an eob run */
895
            if (token <= 6) {
896
                eob_run = eob_run_base[token];
897
                if (eob_run_get_bits[token])
898
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
899

    
900
                // record only the number of blocks ended in this plane,
901
                // any spill will be recorded in the next plane.
902
                if (eob_run > num_coeffs - coeff_i) {
903
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
904
                    blocks_ended   += num_coeffs - coeff_i;
905
                    eob_run        -= num_coeffs - coeff_i;
906
                    coeff_i         = num_coeffs;
907
                } else {
908
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
909
                    blocks_ended   += eob_run;
910
                    coeff_i        += eob_run;
911
                    eob_run = 0;
912
                }
913
            } else {
914
                bits_to_get = coeff_get_bits[token];
915
                if (bits_to_get)
916
                    bits_to_get = get_bits(gb, bits_to_get);
917
                coeff = coeff_tables[token][bits_to_get];
918

    
919
                zero_run = zero_run_base[token];
920
                if (zero_run_get_bits[token])
921
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
922

    
923
                if (zero_run) {
924
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
925
                } else {
926
                    // Save DC into the fragment structure. DC prediction is
927
                    // done in raster order, so the actual DC can't be in with
928
                    // other tokens. We still need the token in dct_tokens[]
929
                    // however, or else the structure collapses on itself.
930
                    if (!coeff_index)
931
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
932

    
933
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
934
                }
935

    
936
                if (coeff_index + zero_run > 64) {
937
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
938
                           " %d coeffs left\n", zero_run, 64-coeff_index);
939
                    zero_run = 64 - coeff_index;
940
                }
941

    
942
                // zero runs code multiple coefficients,
943
                // so don't try to decode coeffs for those higher levels
944
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
945
                    s->num_coded_frags[plane][i]--;
946
                coeff_i++;
947
            }
948
    }
949

    
950
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
951
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
952

    
953
    // decrement the number of blocks that have higher coeffecients for each
954
    // EOB run at this level
955
    if (blocks_ended)
956
        for (i = coeff_index+1; i < 64; i++)
957
            s->num_coded_frags[plane][i] -= blocks_ended;
958

    
959
    // setup the next buffer
960
    if (plane < 2)
961
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
962
    else if (coeff_index < 63)
963
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
964

    
965
    return eob_run;
966
}
967

    
968
static void reverse_dc_prediction(Vp3DecodeContext *s,
969
                                  int first_fragment,
970
                                  int fragment_width,
971
                                  int fragment_height);
972
/*
973
 * This function unpacks all of the DCT coefficient data from the
974
 * bitstream.
975
 */
976
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
977
{
978
    int i;
979
    int dc_y_table;
980
    int dc_c_table;
981
    int ac_y_table;
982
    int ac_c_table;
983
    int residual_eob_run = 0;
984
    VLC *y_tables[64];
985
    VLC *c_tables[64];
986

    
987
    s->dct_tokens[0][0] = s->dct_tokens_base;
988

    
989
    /* fetch the DC table indexes */
990
    dc_y_table = get_bits(gb, 4);
991
    dc_c_table = get_bits(gb, 4);
992

    
993
    /* unpack the Y plane DC coefficients */
994
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
995
        0, residual_eob_run);
996

    
997
    /* reverse prediction of the Y-plane DC coefficients */
998
    reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
999

    
1000
    /* unpack the C plane DC coefficients */
1001
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1002
        1, residual_eob_run);
1003
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1004
        2, residual_eob_run);
1005

    
1006
    /* reverse prediction of the C-plane DC coefficients */
1007
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1008
    {
1009
        reverse_dc_prediction(s, s->fragment_start[1],
1010
            s->fragment_width[1], s->fragment_height[1]);
1011
        reverse_dc_prediction(s, s->fragment_start[2],
1012
            s->fragment_width[1], s->fragment_height[1]);
1013
    }
1014

    
1015
    /* fetch the AC table indexes */
1016
    ac_y_table = get_bits(gb, 4);
1017
    ac_c_table = get_bits(gb, 4);
1018

    
1019
    /* build tables of AC VLC tables */
1020
    for (i = 1; i <= 5; i++) {
1021
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1022
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1023
    }
1024
    for (i = 6; i <= 14; i++) {
1025
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1026
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1027
    }
1028
    for (i = 15; i <= 27; i++) {
1029
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1030
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1031
    }
1032
    for (i = 28; i <= 63; i++) {
1033
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1034
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1035
    }
1036

    
1037
    /* decode all AC coefficents */
1038
    for (i = 1; i <= 63; i++) {
1039
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1040
                0, residual_eob_run);
1041

    
1042
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1043
                1, residual_eob_run);
1044
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1045
                2, residual_eob_run);
1046
    }
1047

    
1048
    return 0;
1049
}
1050

    
1051
/*
1052
 * This function reverses the DC prediction for each coded fragment in
1053
 * the frame. Much of this function is adapted directly from the original
1054
 * VP3 source code.
1055
 */
1056
#define COMPATIBLE_FRAME(x) \
1057
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1058
#define DC_COEFF(u) s->all_fragments[u].dc
1059

    
1060
static void reverse_dc_prediction(Vp3DecodeContext *s,
1061
                                  int first_fragment,
1062
                                  int fragment_width,
1063
                                  int fragment_height)
1064
{
1065

    
1066
#define PUL 8
1067
#define PU 4
1068
#define PUR 2
1069
#define PL 1
1070

    
1071
    int x, y;
1072
    int i = first_fragment;
1073

    
1074
    int predicted_dc;
1075

    
1076
    /* DC values for the left, up-left, up, and up-right fragments */
1077
    int vl, vul, vu, vur;
1078

    
1079
    /* indexes for the left, up-left, up, and up-right fragments */
1080
    int l, ul, u, ur;
1081

    
1082
    /*
1083
     * The 6 fields mean:
1084
     *   0: up-left multiplier
1085
     *   1: up multiplier
1086
     *   2: up-right multiplier
1087
     *   3: left multiplier
1088
     */
1089
    static const int predictor_transform[16][4] = {
1090
        {  0,  0,  0,  0},
1091
        {  0,  0,  0,128},        // PL
1092
        {  0,  0,128,  0},        // PUR
1093
        {  0,  0, 53, 75},        // PUR|PL
1094
        {  0,128,  0,  0},        // PU
1095
        {  0, 64,  0, 64},        // PU|PL
1096
        {  0,128,  0,  0},        // PU|PUR
1097
        {  0,  0, 53, 75},        // PU|PUR|PL
1098
        {128,  0,  0,  0},        // PUL
1099
        {  0,  0,  0,128},        // PUL|PL
1100
        { 64,  0, 64,  0},        // PUL|PUR
1101
        {  0,  0, 53, 75},        // PUL|PUR|PL
1102
        {  0,128,  0,  0},        // PUL|PU
1103
       {-104,116,  0,116},        // PUL|PU|PL
1104
        { 24, 80, 24,  0},        // PUL|PU|PUR
1105
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1106
    };
1107

    
1108
    /* This table shows which types of blocks can use other blocks for
1109
     * prediction. For example, INTRA is the only mode in this table to
1110
     * have a frame number of 0. That means INTRA blocks can only predict
1111
     * from other INTRA blocks. There are 2 golden frame coding types;
1112
     * blocks encoding in these modes can only predict from other blocks
1113
     * that were encoded with these 1 of these 2 modes. */
1114
    static const unsigned char compatible_frame[9] = {
1115
        1,    /* MODE_INTER_NO_MV */
1116
        0,    /* MODE_INTRA */
1117
        1,    /* MODE_INTER_PLUS_MV */
1118
        1,    /* MODE_INTER_LAST_MV */
1119
        1,    /* MODE_INTER_PRIOR_MV */
1120
        2,    /* MODE_USING_GOLDEN */
1121
        2,    /* MODE_GOLDEN_MV */
1122
        1,    /* MODE_INTER_FOUR_MV */
1123
        3     /* MODE_COPY */
1124
    };
1125
    int current_frame_type;
1126

    
1127
    /* there is a last DC predictor for each of the 3 frame types */
1128
    short last_dc[3];
1129

    
1130
    int transform = 0;
1131

    
1132
    vul = vu = vur = vl = 0;
1133
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1134

    
1135
    /* for each fragment row... */
1136
    for (y = 0; y < fragment_height; y++) {
1137

    
1138
        /* for each fragment in a row... */
1139
        for (x = 0; x < fragment_width; x++, i++) {
1140

    
1141
            /* reverse prediction if this block was coded */
1142
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1143

    
1144
                current_frame_type =
1145
                    compatible_frame[s->all_fragments[i].coding_method];
1146

    
1147
                transform= 0;
1148
                if(x){
1149
                    l= i-1;
1150
                    vl = DC_COEFF(l);
1151
                    if(COMPATIBLE_FRAME(l))
1152
                        transform |= PL;
1153
                }
1154
                if(y){
1155
                    u= i-fragment_width;
1156
                    vu = DC_COEFF(u);
1157
                    if(COMPATIBLE_FRAME(u))
1158
                        transform |= PU;
1159
                    if(x){
1160
                        ul= i-fragment_width-1;
1161
                        vul = DC_COEFF(ul);
1162
                        if(COMPATIBLE_FRAME(ul))
1163
                            transform |= PUL;
1164
                    }
1165
                    if(x + 1 < fragment_width){
1166
                        ur= i-fragment_width+1;
1167
                        vur = DC_COEFF(ur);
1168
                        if(COMPATIBLE_FRAME(ur))
1169
                            transform |= PUR;
1170
                    }
1171
                }
1172

    
1173
                if (transform == 0) {
1174

    
1175
                    /* if there were no fragments to predict from, use last
1176
                     * DC saved */
1177
                    predicted_dc = last_dc[current_frame_type];
1178
                } else {
1179

    
1180
                    /* apply the appropriate predictor transform */
1181
                    predicted_dc =
1182
                        (predictor_transform[transform][0] * vul) +
1183
                        (predictor_transform[transform][1] * vu) +
1184
                        (predictor_transform[transform][2] * vur) +
1185
                        (predictor_transform[transform][3] * vl);
1186

    
1187
                    predicted_dc /= 128;
1188

    
1189
                    /* check for outranging on the [ul u l] and
1190
                     * [ul u ur l] predictors */
1191
                    if ((transform == 15) || (transform == 13)) {
1192
                        if (FFABS(predicted_dc - vu) > 128)
1193
                            predicted_dc = vu;
1194
                        else if (FFABS(predicted_dc - vl) > 128)
1195
                            predicted_dc = vl;
1196
                        else if (FFABS(predicted_dc - vul) > 128)
1197
                            predicted_dc = vul;
1198
                    }
1199
                }
1200

    
1201
                /* at long last, apply the predictor */
1202
                DC_COEFF(i) += predicted_dc;
1203
                /* save the DC */
1204
                last_dc[current_frame_type] = DC_COEFF(i);
1205
            }
1206
        }
1207
    }
1208
}
1209

    
1210
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1211
{
1212
    int x, y;
1213
    int *bounding_values= s->bounding_values_array+127;
1214

    
1215
    int width           = s->fragment_width[!!plane];
1216
    int height          = s->fragment_height[!!plane];
1217
    int fragment        = s->fragment_start        [plane] + ystart * width;
1218
    int stride          = s->current_frame.linesize[plane];
1219
    uint8_t *plane_data = s->current_frame.data    [plane];
1220
    if (!s->flipped_image) stride = -stride;
1221
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1222

    
1223
    for (y = ystart; y < yend; y++) {
1224

    
1225
        for (x = 0; x < width; x++) {
1226
            /* This code basically just deblocks on the edges of coded blocks.
1227
             * However, it has to be much more complicated because of the
1228
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1229
             * because some pixels get filtered twice. */
1230
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1231
            {
1232
                /* do not perform left edge filter for left columns frags */
1233
                if (x > 0) {
1234
                    s->dsp.vp3_h_loop_filter(
1235
                        plane_data + 8*x,
1236
                        stride, bounding_values);
1237
                }
1238

    
1239
                /* do not perform top edge filter for top row fragments */
1240
                if (y > 0) {
1241
                    s->dsp.vp3_v_loop_filter(
1242
                        plane_data + 8*x,
1243
                        stride, bounding_values);
1244
                }
1245

    
1246
                /* do not perform right edge filter for right column
1247
                 * fragments or if right fragment neighbor is also coded
1248
                 * in this frame (it will be filtered in next iteration) */
1249
                if ((x < width - 1) &&
1250
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1251
                    s->dsp.vp3_h_loop_filter(
1252
                        plane_data + 8*x + 8,
1253
                        stride, bounding_values);
1254
                }
1255

    
1256
                /* do not perform bottom edge filter for bottom row
1257
                 * fragments or if bottom fragment neighbor is also coded
1258
                 * in this frame (it will be filtered in the next row) */
1259
                if ((y < height - 1) &&
1260
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1261
                    s->dsp.vp3_v_loop_filter(
1262
                        plane_data + 8*x + 8*stride,
1263
                        stride, bounding_values);
1264
                }
1265
            }
1266

    
1267
            fragment++;
1268
        }
1269
        plane_data += 8*stride;
1270
    }
1271
}
1272

    
1273
/**
1274
 * Pull DCT tokens from the 64 levels to decode and dequant the coefficients
1275
 * for the next block in coding order
1276
 */
1277
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1278
                              int plane, int inter, DCTELEM block[64])
1279
{
1280
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1281
    uint8_t *perm = s->scantable.permutated;
1282
    int i = 0;
1283

    
1284
    do {
1285
        int token = *s->dct_tokens[plane][i];
1286
        switch (token & 3) {
1287
        case 0: // EOB
1288
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1289
                s->dct_tokens[plane][i]++;
1290
            else
1291
                *s->dct_tokens[plane][i] = token & ~3;
1292
            goto end;
1293
        case 1: // zero run
1294
            s->dct_tokens[plane][i]++;
1295
            i += (token >> 2) & 0x7f;
1296
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1297
            i++;
1298
            break;
1299
        case 2: // coeff
1300
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1301
            s->dct_tokens[plane][i++]++;
1302
            break;
1303
        default: // shouldn't happen
1304
            return i;
1305
        }
1306
    } while (i < 64);
1307
end:
1308
    // the actual DC+prediction is in the fragment structure
1309
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1310
    return i;
1311
}
1312

    
1313
/**
1314
 * called when all pixels up to row y are complete
1315
 */
1316
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1317
{
1318
    int h, cy;
1319
    int offset[4];
1320

    
1321
    if(s->avctx->draw_horiz_band==NULL)
1322
        return;
1323

    
1324
    h= y - s->last_slice_end;
1325
    s->last_slice_end= y;
1326
    y -= h;
1327

    
1328
    if (!s->flipped_image) {
1329
        y = s->avctx->height - y - h;
1330
    }
1331

    
1332
    cy = y >> s->chroma_y_shift;
1333
    offset[0] = s->current_frame.linesize[0]*y;
1334
    offset[1] = s->current_frame.linesize[1]*cy;
1335
    offset[2] = s->current_frame.linesize[2]*cy;
1336
    offset[3] = 0;
1337

    
1338
    emms_c();
1339
    s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
1340
}
1341

    
1342
/*
1343
 * Perform the final rendering for a particular slice of data.
1344
 * The slice number ranges from 0..(c_superblock_height - 1).
1345
 */
1346
static void render_slice(Vp3DecodeContext *s, int slice)
1347
{
1348
    int x, y, i, j, fragment;
1349
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1350
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1351
    int motion_halfpel_index;
1352
    uint8_t *motion_source;
1353
    int plane, first_pixel;
1354

    
1355
    if (slice >= s->c_superblock_height)
1356
        return;
1357

    
1358
    for (plane = 0; plane < 3; plane++) {
1359
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1360
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1361
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1362
        int stride            = s->current_frame.linesize[plane];
1363
        int plane_width       = s->width  >> (plane && s->chroma_x_shift);
1364
        int plane_height      = s->height >> (plane && s->chroma_y_shift);
1365
        int8_t (*motion_val)[2] = s->motion_val[!!plane];
1366

    
1367
        int sb_x, sb_y        = slice << (!plane && s->chroma_y_shift);
1368
        int slice_height      = sb_y + 1 + (!plane && s->chroma_y_shift);
1369
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1370

    
1371
        int fragment_width    = s->fragment_width[!!plane];
1372
        int fragment_height   = s->fragment_height[!!plane];
1373
        int fragment_start    = s->fragment_start[plane];
1374

    
1375
        if (!s->flipped_image) stride = -stride;
1376
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1377
            continue;
1378

    
1379

    
1380
        if(FFABS(stride) > 2048)
1381
            return; //various tables are fixed size
1382

    
1383
        /* for each superblock row in the slice (both of them)... */
1384
        for (; sb_y < slice_height; sb_y++) {
1385

    
1386
            /* for each superblock in a row... */
1387
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1388

    
1389
                /* for each block in a superblock... */
1390
                for (j = 0; j < 16; j++) {
1391
                    x = 4*sb_x + hilbert_offset[j][0];
1392
                    y = 4*sb_y + hilbert_offset[j][1];
1393
                    fragment = y*fragment_width + x;
1394

    
1395
                    i = fragment_start + fragment;
1396

    
1397
                    // bounds check
1398
                    if (x >= fragment_width || y >= fragment_height)
1399
                        continue;
1400

    
1401
                first_pixel = 8*y*stride + 8*x;
1402

    
1403
                /* transform if this block was coded */
1404
                if (s->all_fragments[i].coding_method != MODE_COPY) {
1405
                    if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
1406
                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
1407
                        motion_source= golden_plane;
1408
                    else
1409
                        motion_source= last_plane;
1410

    
1411
                    motion_source += first_pixel;
1412
                    motion_halfpel_index = 0;
1413

    
1414
                    /* sort out the motion vector if this fragment is coded
1415
                     * using a motion vector method */
1416
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
1417
                        (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
1418
                        int src_x, src_y;
1419
                        motion_x = motion_val[fragment][0];
1420
                        motion_y = motion_val[fragment][1];
1421

    
1422
                        src_x= (motion_x>>1) + 8*x;
1423
                        src_y= (motion_y>>1) + 8*y;
1424

    
1425
                        motion_halfpel_index = motion_x & 0x01;
1426
                        motion_source += (motion_x >> 1);
1427

    
1428
                        motion_halfpel_index |= (motion_y & 0x01) << 1;
1429
                        motion_source += ((motion_y >> 1) * stride);
1430

    
1431
                        if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
1432
                            uint8_t *temp= s->edge_emu_buffer;
1433
                            if(stride<0) temp -= 9*stride;
1434
                            else temp += 9*stride;
1435

    
1436
                            s->dsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
1437
                            motion_source= temp;
1438
                        }
1439
                    }
1440

    
1441

    
1442
                    /* first, take care of copying a block from either the
1443
                     * previous or the golden frame */
1444
                    if (s->all_fragments[i].coding_method != MODE_INTRA) {
1445
                        /* Note, it is possible to implement all MC cases with
1446
                           put_no_rnd_pixels_l2 which would look more like the
1447
                           VP3 source but this would be slower as
1448
                           put_no_rnd_pixels_tab is better optimzed */
1449
                        if(motion_halfpel_index != 3){
1450
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
1451
                                output_plane + first_pixel,
1452
                                motion_source, stride, 8);
1453
                        }else{
1454
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
1455
                            s->dsp.put_no_rnd_pixels_l2[1](
1456
                                output_plane + first_pixel,
1457
                                motion_source - d,
1458
                                motion_source + stride + 1 + d,
1459
                                stride, 8);
1460
                        }
1461
                    }
1462

    
1463
                        s->dsp.clear_block(block);
1464

    
1465
                    /* invert DCT and place (or add) in final output */
1466

    
1467
                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
1468
                        vp3_dequant(s, s->all_fragments + i, plane, 0, block);
1469
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
1470
                            block[0] += 128<<3;
1471
                        s->dsp.idct_put(
1472
                            output_plane + first_pixel,
1473
                            stride,
1474
                            block);
1475
                    } else {
1476
                        if (vp3_dequant(s, s->all_fragments + i, plane, 1, block)) {
1477
                        s->dsp.idct_add(
1478
                            output_plane + first_pixel,
1479
                            stride,
1480
                            block);
1481
                        } else {
1482
                            s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
1483
                        }
1484
                    }
1485
                } else {
1486

    
1487
                    /* copy directly from the previous frame */
1488
                    s->dsp.put_pixels_tab[1][0](
1489
                        output_plane + first_pixel,
1490
                        last_plane + first_pixel,
1491
                        stride, 8);
1492

    
1493
                }
1494
                }
1495
            }
1496

    
1497
            // Filter up to the last row in the superblock row
1498
            if (!s->skip_loop_filter)
1499
                apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
1500
        }
1501
    }
1502

    
1503
     /* this looks like a good place for slice dispatch... */
1504
     /* algorithm:
1505
      *   if (slice == s->macroblock_height - 1)
1506
      *     dispatch (both last slice & 2nd-to-last slice);
1507
      *   else if (slice > 0)
1508
      *     dispatch (slice - 1);
1509
      */
1510

    
1511
    vp3_draw_horiz_band(s, FFMIN((32 << s->chroma_y_shift) * (slice + 1) -16, s->height-16));
1512
}
1513

    
1514
/*
1515
 * This is the ffmpeg/libavcodec API init function.
1516
 */
1517
static av_cold int vp3_decode_init(AVCodecContext *avctx)
1518
{
1519
    Vp3DecodeContext *s = avctx->priv_data;
1520
    int i, inter, plane;
1521
    int c_width;
1522
    int c_height;
1523
    int y_fragment_count, c_fragment_count;
1524

    
1525
    if (avctx->codec_tag == MKTAG('V','P','3','0'))
1526
        s->version = 0;
1527
    else
1528
        s->version = 1;
1529

    
1530
    s->avctx = avctx;
1531
    s->width = FFALIGN(avctx->width, 16);
1532
    s->height = FFALIGN(avctx->height, 16);
1533
    if (avctx->pix_fmt == PIX_FMT_NONE)
1534
        avctx->pix_fmt = PIX_FMT_YUV420P;
1535
    avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
1536
    if(avctx->idct_algo==FF_IDCT_AUTO)
1537
        avctx->idct_algo=FF_IDCT_VP3;
1538
    dsputil_init(&s->dsp, avctx);
1539

    
1540
    ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
1541

    
1542
    /* initialize to an impossible value which will force a recalculation
1543
     * in the first frame decode */
1544
    for (i = 0; i < 3; i++)
1545
        s->qps[i] = -1;
1546

    
1547
    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
1548

    
1549
    s->y_superblock_width = (s->width + 31) / 32;
1550
    s->y_superblock_height = (s->height + 31) / 32;
1551
    s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
1552

    
1553
    /* work out the dimensions for the C planes */
1554
    c_width = s->width >> s->chroma_x_shift;
1555
    c_height = s->height >> s->chroma_y_shift;
1556
    s->c_superblock_width = (c_width + 31) / 32;
1557
    s->c_superblock_height = (c_height + 31) / 32;
1558
    s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
1559

    
1560
    s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
1561
    s->u_superblock_start = s->y_superblock_count;
1562
    s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
1563
    s->superblock_coding = av_malloc(s->superblock_count);
1564

    
1565
    s->macroblock_width = (s->width + 15) / 16;
1566
    s->macroblock_height = (s->height + 15) / 16;
1567
    s->macroblock_count = s->macroblock_width * s->macroblock_height;
1568

    
1569
    s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
1570
    s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
1571
    s->fragment_width[1]  = s->fragment_width[0]  >> s->chroma_x_shift;
1572
    s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
1573

    
1574
    /* fragment count covers all 8x8 blocks for all 3 planes */
1575
    y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
1576
    c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
1577
    s->fragment_count    = y_fragment_count + 2*c_fragment_count;
1578
    s->fragment_start[1] = y_fragment_count;
1579
    s->fragment_start[2] = y_fragment_count + c_fragment_count;
1580

    
1581
    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
1582
    s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
1583
    s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
1584
    s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
1585
    s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
1586

    
1587
    if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
1588
        !s->coded_fragment_list[0] || !s->motion_val[0] || !s->motion_val[1]) {
1589
        vp3_decode_end(avctx);
1590
        return -1;
1591
    }
1592

    
1593
    if (!s->theora_tables)
1594
    {
1595
        for (i = 0; i < 64; i++) {
1596
            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
1597
            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
1598
            s->base_matrix[0][i] = vp31_intra_y_dequant[i];
1599
            s->base_matrix[1][i] = vp31_intra_c_dequant[i];
1600
            s->base_matrix[2][i] = vp31_inter_dequant[i];
1601
            s->filter_limit_values[i] = vp31_filter_limit_values[i];
1602
        }
1603

    
1604
        for(inter=0; inter<2; inter++){
1605
            for(plane=0; plane<3; plane++){
1606
                s->qr_count[inter][plane]= 1;
1607
                s->qr_size [inter][plane][0]= 63;
1608
                s->qr_base [inter][plane][0]=
1609
                s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
1610
            }
1611
        }
1612

    
1613
        /* init VLC tables */
1614
        for (i = 0; i < 16; i++) {
1615

    
1616
            /* DC histograms */
1617
            init_vlc(&s->dc_vlc[i], 11, 32,
1618
                &dc_bias[i][0][1], 4, 2,
1619
                &dc_bias[i][0][0], 4, 2, 0);
1620

    
1621
            /* group 1 AC histograms */
1622
            init_vlc(&s->ac_vlc_1[i], 11, 32,
1623
                &ac_bias_0[i][0][1], 4, 2,
1624
                &ac_bias_0[i][0][0], 4, 2, 0);
1625

    
1626
            /* group 2 AC histograms */
1627
            init_vlc(&s->ac_vlc_2[i], 11, 32,
1628
                &ac_bias_1[i][0][1], 4, 2,
1629
                &ac_bias_1[i][0][0], 4, 2, 0);
1630

    
1631
            /* group 3 AC histograms */
1632
            init_vlc(&s->ac_vlc_3[i], 11, 32,
1633
                &ac_bias_2[i][0][1], 4, 2,
1634
                &ac_bias_2[i][0][0], 4, 2, 0);
1635

    
1636
            /* group 4 AC histograms */
1637
            init_vlc(&s->ac_vlc_4[i], 11, 32,
1638
                &ac_bias_3[i][0][1], 4, 2,
1639
                &ac_bias_3[i][0][0], 4, 2, 0);
1640
        }
1641
    } else {
1642

    
1643
        for (i = 0; i < 16; i++) {
1644
            /* DC histograms */
1645
            if (init_vlc(&s->dc_vlc[i], 11, 32,
1646
                &s->huffman_table[i][0][1], 8, 4,
1647
                &s->huffman_table[i][0][0], 8, 4, 0) < 0)
1648
                goto vlc_fail;
1649

    
1650
            /* group 1 AC histograms */
1651
            if (init_vlc(&s->ac_vlc_1[i], 11, 32,
1652
                &s->huffman_table[i+16][0][1], 8, 4,
1653
                &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
1654
                goto vlc_fail;
1655

    
1656
            /* group 2 AC histograms */
1657
            if (init_vlc(&s->ac_vlc_2[i], 11, 32,
1658
                &s->huffman_table[i+16*2][0][1], 8, 4,
1659
                &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
1660
                goto vlc_fail;
1661

    
1662
            /* group 3 AC histograms */
1663
            if (init_vlc(&s->ac_vlc_3[i], 11, 32,
1664
                &s->huffman_table[i+16*3][0][1], 8, 4,
1665
                &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
1666
                goto vlc_fail;
1667

    
1668
            /* group 4 AC histograms */
1669
            if (init_vlc(&s->ac_vlc_4[i], 11, 32,
1670
                &s->huffman_table[i+16*4][0][1], 8, 4,
1671
                &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
1672
                goto vlc_fail;
1673
        }
1674
    }
1675

    
1676
    init_vlc(&s->superblock_run_length_vlc, 6, 34,
1677
        &superblock_run_length_vlc_table[0][1], 4, 2,
1678
        &superblock_run_length_vlc_table[0][0], 4, 2, 0);
1679

    
1680
    init_vlc(&s->fragment_run_length_vlc, 5, 30,
1681
        &fragment_run_length_vlc_table[0][1], 4, 2,
1682
        &fragment_run_length_vlc_table[0][0], 4, 2, 0);
1683

    
1684
    init_vlc(&s->mode_code_vlc, 3, 8,
1685
        &mode_code_vlc_table[0][1], 2, 1,
1686
        &mode_code_vlc_table[0][0], 2, 1, 0);
1687

    
1688
    init_vlc(&s->motion_vector_vlc, 6, 63,
1689
        &motion_vector_vlc_table[0][1], 2, 1,
1690
        &motion_vector_vlc_table[0][0], 2, 1, 0);
1691

    
1692
    /* work out the block mapping tables */
1693
    s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
1694
    s->macroblock_coding = av_malloc(s->macroblock_count + 1);
1695
    if (!s->superblock_fragments || !s->macroblock_coding) {
1696
        vp3_decode_end(avctx);
1697
        return -1;
1698
    }
1699
    init_block_mapping(s);
1700

    
1701
    for (i = 0; i < 3; i++) {
1702
        s->current_frame.data[i] = NULL;
1703
        s->last_frame.data[i] = NULL;
1704
        s->golden_frame.data[i] = NULL;
1705
    }
1706

    
1707
    return 0;
1708

    
1709
vlc_fail:
1710
    av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
1711
    return -1;
1712
}
1713

    
1714
/*
1715
 * This is the ffmpeg/libavcodec API frame decode function.
1716
 */
1717
static int vp3_decode_frame(AVCodecContext *avctx,
1718
                            void *data, int *data_size,
1719
                            AVPacket *avpkt)
1720
{
1721
    const uint8_t *buf = avpkt->data;
1722
    int buf_size = avpkt->size;
1723
    Vp3DecodeContext *s = avctx->priv_data;
1724
    GetBitContext gb;
1725
    int i;
1726

    
1727
    init_get_bits(&gb, buf, buf_size * 8);
1728

    
1729
    if (s->theora && get_bits1(&gb))
1730
    {
1731
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1732
        return -1;
1733
    }
1734

    
1735
    s->keyframe = !get_bits1(&gb);
1736
    if (!s->theora)
1737
        skip_bits(&gb, 1);
1738
    for (i = 0; i < 3; i++)
1739
        s->last_qps[i] = s->qps[i];
1740

    
1741
    s->nqps=0;
1742
    do{
1743
        s->qps[s->nqps++]= get_bits(&gb, 6);
1744
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1745
    for (i = s->nqps; i < 3; i++)
1746
        s->qps[i] = -1;
1747

    
1748
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1749
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1750
            s->keyframe?"key":"", avctx->frame_number+1, s->qps[0]);
1751

    
1752
    s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1753
        avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1754

    
1755
    if (s->qps[0] != s->last_qps[0])
1756
        init_loop_filter(s);
1757

    
1758
    for (i = 0; i < s->nqps; i++)
1759
        // reinit all dequantizers if the first one changed, because
1760
        // the DC of the first quantizer must be used for all matrices
1761
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1762
            init_dequantizer(s, i);
1763

    
1764
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1765
        return buf_size;
1766

    
1767
    s->current_frame.reference = 3;
1768
    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
1769
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1770
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1771
        goto error;
1772
    }
1773

    
1774
    if (s->keyframe) {
1775
        if (!s->theora)
1776
        {
1777
            skip_bits(&gb, 4); /* width code */
1778
            skip_bits(&gb, 4); /* height code */
1779
            if (s->version)
1780
            {
1781
                s->version = get_bits(&gb, 5);
1782
                if (avctx->frame_number == 0)
1783
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1784
            }
1785
        }
1786
        if (s->version || s->theora)
1787
        {
1788
                if (get_bits1(&gb))
1789
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1790
            skip_bits(&gb, 2); /* reserved? */
1791
        }
1792
    } else {
1793
        if (!s->golden_frame.data[0]) {
1794
            av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1795

    
1796
            s->golden_frame.reference = 3;
1797
            s->golden_frame.pict_type = FF_I_TYPE;
1798
            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1799
                av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1800
                goto error;
1801
            }
1802
            s->last_frame = s->golden_frame;
1803
            s->last_frame.type = FF_BUFFER_TYPE_COPY;
1804
        }
1805
    }
1806

    
1807
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1808
    s->current_frame.qstride= 0;
1809

    
1810
    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1811

    
1812
    if (unpack_superblocks(s, &gb)){
1813
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1814
        goto error;
1815
    }
1816
    if (unpack_modes(s, &gb)){
1817
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1818
        goto error;
1819
    }
1820
    if (unpack_vectors(s, &gb)){
1821
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1822
        goto error;
1823
    }
1824
    if (unpack_block_qpis(s, &gb)){
1825
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1826
        goto error;
1827
    }
1828
    if (unpack_dct_coeffs(s, &gb)){
1829
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1830
        goto error;
1831
    }
1832

    
1833
    for (i = 0; i < 3; i++) {
1834
        int height = s->height >> (i && s->chroma_y_shift);
1835
        if (s->flipped_image)
1836
            s->data_offset[i] = 0;
1837
        else
1838
            s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1839
    }
1840

    
1841
    s->last_slice_end = 0;
1842
    for (i = 0; i < s->c_superblock_height; i++)
1843
        render_slice(s, i);
1844

    
1845
    // filter the last row
1846
    for (i = 0; i < 3; i++) {
1847
        int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1848
        apply_loop_filter(s, i, row, row+1);
1849
    }
1850
    vp3_draw_horiz_band(s, s->avctx->height);
1851

    
1852
    *data_size=sizeof(AVFrame);
1853
    *(AVFrame*)data= s->current_frame;
1854

    
1855
    /* release the last frame, if it is allocated and if it is not the
1856
     * golden frame */
1857
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1858
        avctx->release_buffer(avctx, &s->last_frame);
1859

    
1860
    /* shuffle frames (last = current) */
1861
    s->last_frame= s->current_frame;
1862

    
1863
    if (s->keyframe) {
1864
        if (s->golden_frame.data[0])
1865
            avctx->release_buffer(avctx, &s->golden_frame);
1866
        s->golden_frame = s->current_frame;
1867
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1868
    }
1869

    
1870
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1871

    
1872
    return buf_size;
1873

    
1874
error:
1875
    if (s->current_frame.data[0])
1876
        avctx->release_buffer(avctx, &s->current_frame);
1877
    return -1;
1878
}
1879

    
1880
/*
1881
 * This is the ffmpeg/libavcodec API module cleanup function.
1882
 */
1883
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1884
{
1885
    Vp3DecodeContext *s = avctx->priv_data;
1886
    int i;
1887

    
1888
    av_free(s->superblock_coding);
1889
    av_free(s->all_fragments);
1890
    av_free(s->coded_fragment_list[0]);
1891
    av_free(s->dct_tokens_base);
1892
    av_free(s->superblock_fragments);
1893
    av_free(s->macroblock_coding);
1894
    av_free(s->motion_val[0]);
1895
    av_free(s->motion_val[1]);
1896

    
1897
    for (i = 0; i < 16; i++) {
1898
        free_vlc(&s->dc_vlc[i]);
1899
        free_vlc(&s->ac_vlc_1[i]);
1900
        free_vlc(&s->ac_vlc_2[i]);
1901
        free_vlc(&s->ac_vlc_3[i]);
1902
        free_vlc(&s->ac_vlc_4[i]);
1903
    }
1904

    
1905
    free_vlc(&s->superblock_run_length_vlc);
1906
    free_vlc(&s->fragment_run_length_vlc);
1907
    free_vlc(&s->mode_code_vlc);
1908
    free_vlc(&s->motion_vector_vlc);
1909

    
1910
    /* release all frames */
1911
    if (s->golden_frame.data[0])
1912
        avctx->release_buffer(avctx, &s->golden_frame);
1913
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1914
        avctx->release_buffer(avctx, &s->last_frame);
1915
    /* no need to release the current_frame since it will always be pointing
1916
     * to the same frame as either the golden or last frame */
1917

    
1918
    return 0;
1919
}
1920

    
1921
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1922
{
1923
    Vp3DecodeContext *s = avctx->priv_data;
1924

    
1925
    if (get_bits1(gb)) {
1926
        int token;
1927
        if (s->entries >= 32) { /* overflow */
1928
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1929
            return -1;
1930
        }
1931
        token = get_bits(gb, 5);
1932
        //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);
1933
        s->huffman_table[s->hti][token][0] = s->hbits;
1934
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1935
        s->entries++;
1936
    }
1937
    else {
1938
        if (s->huff_code_size >= 32) {/* overflow */
1939
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1940
            return -1;
1941
        }
1942
        s->huff_code_size++;
1943
        s->hbits <<= 1;
1944
        if (read_huffman_tree(avctx, gb))
1945
            return -1;
1946
        s->hbits |= 1;
1947
        if (read_huffman_tree(avctx, gb))
1948
            return -1;
1949
        s->hbits >>= 1;
1950
        s->huff_code_size--;
1951
    }
1952
    return 0;
1953
}
1954

    
1955
#if CONFIG_THEORA_DECODER
1956
static const enum PixelFormat theora_pix_fmts[4] = {
1957
    PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
1958
};
1959

    
1960
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1961
{
1962
    Vp3DecodeContext *s = avctx->priv_data;
1963
    int visible_width, visible_height, colorspace;
1964
    int offset_x = 0, offset_y = 0;
1965
    AVRational fps, aspect;
1966

    
1967
    s->theora = get_bits_long(gb, 24);
1968
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1969

    
1970
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1971
    /* but previous versions have the image flipped relative to vp3 */
1972
    if (s->theora < 0x030200)
1973
    {
1974
        s->flipped_image = 1;
1975
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1976
    }
1977

    
1978
    visible_width  = s->width  = get_bits(gb, 16) << 4;
1979
    visible_height = s->height = get_bits(gb, 16) << 4;
1980

    
1981
    if(av_image_check_size(s->width, s->height, 0, avctx)){
1982
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
1983
        s->width= s->height= 0;
1984
        return -1;
1985
    }
1986

    
1987
    if (s->theora >= 0x030200) {
1988
        visible_width  = get_bits_long(gb, 24);
1989
        visible_height = get_bits_long(gb, 24);
1990

    
1991
        offset_x = get_bits(gb, 8); /* offset x */
1992
        offset_y = get_bits(gb, 8); /* offset y, from bottom */
1993
    }
1994

    
1995
    fps.num = get_bits_long(gb, 32);
1996
    fps.den = get_bits_long(gb, 32);
1997
    if (fps.num && fps.den) {
1998
        av_reduce(&avctx->time_base.num, &avctx->time_base.den,
1999
                  fps.den, fps.num, 1<<30);
2000
    }
2001

    
2002
    aspect.num = get_bits_long(gb, 24);
2003
    aspect.den = get_bits_long(gb, 24);
2004
    if (aspect.num && aspect.den) {
2005
        av_reduce(&avctx->sample_aspect_ratio.num,
2006
                  &avctx->sample_aspect_ratio.den,
2007
                  aspect.num, aspect.den, 1<<30);
2008
    }
2009

    
2010
    if (s->theora < 0x030200)
2011
        skip_bits(gb, 5); /* keyframe frequency force */
2012
    colorspace = get_bits(gb, 8);
2013
    skip_bits(gb, 24); /* bitrate */
2014

    
2015
    skip_bits(gb, 6); /* quality hint */
2016

    
2017
    if (s->theora >= 0x030200)
2018
    {
2019
        skip_bits(gb, 5); /* keyframe frequency force */
2020
        avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2021
        skip_bits(gb, 3); /* reserved */
2022
    }
2023

    
2024
//    align_get_bits(gb);
2025

    
2026
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2027
        && visible_height <= s->height && visible_height > s->height-16
2028
        && !offset_x && (offset_y == s->height - visible_height))
2029
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2030
    else
2031
        avcodec_set_dimensions(avctx, s->width, s->height);
2032

    
2033
    if (colorspace == 1) {
2034
        avctx->color_primaries = AVCOL_PRI_BT470M;
2035
    } else if (colorspace == 2) {
2036
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2037
    }
2038
    if (colorspace == 1 || colorspace == 2) {
2039
        avctx->colorspace = AVCOL_SPC_BT470BG;
2040
        avctx->color_trc  = AVCOL_TRC_BT709;
2041
    }
2042

    
2043
    return 0;
2044
}
2045

    
2046
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2047
{
2048
    Vp3DecodeContext *s = avctx->priv_data;
2049
    int i, n, matrices, inter, plane;
2050

    
2051
    if (s->theora >= 0x030200) {
2052
        n = get_bits(gb, 3);
2053
        /* loop filter limit values table */
2054
        if (n)
2055
            for (i = 0; i < 64; i++)
2056
                s->filter_limit_values[i] = get_bits(gb, n);
2057
    }
2058

    
2059
    if (s->theora >= 0x030200)
2060
        n = get_bits(gb, 4) + 1;
2061
    else
2062
        n = 16;
2063
    /* quality threshold table */
2064
    for (i = 0; i < 64; i++)
2065
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2066

    
2067
    if (s->theora >= 0x030200)
2068
        n = get_bits(gb, 4) + 1;
2069
    else
2070
        n = 16;
2071
    /* dc scale factor table */
2072
    for (i = 0; i < 64; i++)
2073
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2074

    
2075
    if (s->theora >= 0x030200)
2076
        matrices = get_bits(gb, 9) + 1;
2077
    else
2078
        matrices = 3;
2079

    
2080
    if(matrices > 384){
2081
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2082
        return -1;
2083
    }
2084

    
2085
    for(n=0; n<matrices; n++){
2086
        for (i = 0; i < 64; i++)
2087
            s->base_matrix[n][i]= get_bits(gb, 8);
2088
    }
2089

    
2090
    for (inter = 0; inter <= 1; inter++) {
2091
        for (plane = 0; plane <= 2; plane++) {
2092
            int newqr= 1;
2093
            if (inter || plane > 0)
2094
                newqr = get_bits1(gb);
2095
            if (!newqr) {
2096
                int qtj, plj;
2097
                if(inter && get_bits1(gb)){
2098
                    qtj = 0;
2099
                    plj = plane;
2100
                }else{
2101
                    qtj= (3*inter + plane - 1) / 3;
2102
                    plj= (plane + 2) % 3;
2103
                }
2104
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2105
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2106
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2107
            } else {
2108
                int qri= 0;
2109
                int qi = 0;
2110

    
2111
                for(;;){
2112
                    i= get_bits(gb, av_log2(matrices-1)+1);
2113
                    if(i>= matrices){
2114
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2115
                        return -1;
2116
                    }
2117
                    s->qr_base[inter][plane][qri]= i;
2118
                    if(qi >= 63)
2119
                        break;
2120
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2121
                    s->qr_size[inter][plane][qri++]= i;
2122
                    qi += i;
2123
                }
2124

    
2125
                if (qi > 63) {
2126
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2127
                    return -1;
2128
                }
2129
                s->qr_count[inter][plane]= qri;
2130
            }
2131
        }
2132
    }
2133

    
2134
    /* Huffman tables */
2135
    for (s->hti = 0; s->hti < 80; s->hti++) {
2136
        s->entries = 0;
2137
        s->huff_code_size = 1;
2138
        if (!get_bits1(gb)) {
2139
            s->hbits = 0;
2140
            if(read_huffman_tree(avctx, gb))
2141
                return -1;
2142
            s->hbits = 1;
2143
            if(read_huffman_tree(avctx, gb))
2144
                return -1;
2145
        }
2146
    }
2147

    
2148
    s->theora_tables = 1;
2149

    
2150
    return 0;
2151
}
2152

    
2153
static av_cold int theora_decode_init(AVCodecContext *avctx)
2154
{
2155
    Vp3DecodeContext *s = avctx->priv_data;
2156
    GetBitContext gb;
2157
    int ptype;
2158
    uint8_t *header_start[3];
2159
    int header_len[3];
2160
    int i;
2161

    
2162
    s->theora = 1;
2163

    
2164
    if (!avctx->extradata_size)
2165
    {
2166
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2167
        return -1;
2168
    }
2169

    
2170
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2171
                              42, header_start, header_len) < 0) {
2172
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2173
        return -1;
2174
    }
2175

    
2176
  for(i=0;i<3;i++) {
2177
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2178

    
2179
    ptype = get_bits(&gb, 8);
2180

    
2181
     if (!(ptype & 0x80))
2182
     {
2183
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2184
//        return -1;
2185
     }
2186

    
2187
    // FIXME: Check for this as well.
2188
    skip_bits_long(&gb, 6*8); /* "theora" */
2189

    
2190
    switch(ptype)
2191
    {
2192
        case 0x80:
2193
            theora_decode_header(avctx, &gb);
2194
                break;
2195
        case 0x81:
2196
// FIXME: is this needed? it breaks sometimes
2197
//            theora_decode_comments(avctx, gb);
2198
            break;
2199
        case 0x82:
2200
            if (theora_decode_tables(avctx, &gb))
2201
                return -1;
2202
            break;
2203
        default:
2204
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2205
            break;
2206
    }
2207
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2208
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2209
    if (s->theora < 0x030200)
2210
        break;
2211
  }
2212

    
2213
    return vp3_decode_init(avctx);
2214
}
2215

    
2216
AVCodec ff_theora_decoder = {
2217
    "theora",
2218
    AVMEDIA_TYPE_VIDEO,
2219
    CODEC_ID_THEORA,
2220
    sizeof(Vp3DecodeContext),
2221
    theora_decode_init,
2222
    NULL,
2223
    vp3_decode_end,
2224
    vp3_decode_frame,
2225
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2226
    NULL,
2227
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2228
};
2229
#endif
2230

    
2231
AVCodec ff_vp3_decoder = {
2232
    "vp3",
2233
    AVMEDIA_TYPE_VIDEO,
2234
    CODEC_ID_VP3,
2235
    sizeof(Vp3DecodeContext),
2236
    vp3_decode_init,
2237
    NULL,
2238
    vp3_decode_end,
2239
    vp3_decode_frame,
2240
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2241
    NULL,
2242
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2243
};