Statistics
| Branch: | Revision:

ffmpeg / libavcodec / vp3.c @ 8764389d

History | View | Annotate | Download (77.3 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 "avcodec.h"
37
#include "dsputil.h"
38
#include "get_bits.h"
39

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

    
43
#define FRAGMENT_PIXELS 8
44

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

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

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

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

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

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

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

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

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

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

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

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

    
115
};
116

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

    
124
#define MIN_DEQUANT_VAL 2
125

    
126
typedef struct Vp3DecodeContext {
127
    AVCodecContext *avctx;
128
    int theora, theora_tables;
129
    int version;
130
    int width, height;
131
    int chroma_x_shift, chroma_y_shift;
132
    AVFrame golden_frame;
133
    AVFrame last_frame;
134
    AVFrame current_frame;
135
    int keyframe;
136
    DSPContext dsp;
137
    int flipped_image;
138
    int last_slice_end;
139
    int skip_loop_filter;
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[2];
162
    int fragment_height[2];
163

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

    
168
    int8_t (*motion_val[2])[2];
169

    
170
    ScanTable scantable;
171

    
172
    /* tables */
173
    uint16_t coded_dc_scale_factor[64];
174
    uint32_t coded_ac_scale_factor[64];
175
    uint8_t base_matrix[384][64];
176
    uint8_t qr_count[2][3];
177
    uint8_t qr_size [2][3][64];
178
    uint16_t qr_base[2][3][64];
179

    
180
    /**
181
     * This is a list of all tokens in bitstream order. Reordering takes place
182
     * by pulling from each level during IDCT. As a consequence, IDCT must be
183
     * in Hilbert order, making the minimum slice height 64 for 4:2:0 and 32
184
     * otherwise. The 32 different tokens with up to 12 bits of extradata are
185
     * collapsed into 3 types, packed as follows:
186
     *   (from the low to high bits)
187
     *
188
     * 2 bits: type (0,1,2)
189
     *   0: EOB run, 14 bits for run length (12 needed)
190
     *   1: zero run, 7 bits for run length
191
     *                7 bits for the next coefficient (3 needed)
192
     *   2: coefficient, 14 bits (11 needed)
193
     *
194
     * Coefficients are signed, so are packed in the highest bits for automatic
195
     * sign extension.
196
     */
197
    int16_t *dct_tokens[3][64];
198
    int16_t *dct_tokens_base;
199
#define TOKEN_EOB(eob_run)              ((eob_run) << 2)
200
#define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
201
#define TOKEN_COEFF(coeff)              (((coeff) << 2) + 2)
202

    
203
    /**
204
     * number of blocks that contain DCT coefficients at the given level or higher
205
     */
206
    int num_coded_frags[3][64];
207
    int total_num_coded_frags;
208

    
209
    /* this is a list of indexes into the all_fragments array indicating
210
     * which of the fragments are coded */
211
    int *coded_fragment_list[3];
212

    
213
    VLC dc_vlc[16];
214
    VLC ac_vlc_1[16];
215
    VLC ac_vlc_2[16];
216
    VLC ac_vlc_3[16];
217
    VLC ac_vlc_4[16];
218

    
219
    VLC superblock_run_length_vlc;
220
    VLC fragment_run_length_vlc;
221
    VLC mode_code_vlc;
222
    VLC motion_vector_vlc;
223

    
224
    /* these arrays need to be on 16-byte boundaries since SSE2 operations
225
     * index into them */
226
    DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
227

    
228
    /* This table contains superblock_count * 16 entries. Each set of 16
229
     * numbers corresponds to the fragment indexes 0..15 of the superblock.
230
     * An entry will be -1 to indicate that no entry corresponds to that
231
     * index. */
232
    int *superblock_fragments;
233

    
234
    /* This is an array that indicates how a particular macroblock
235
     * is coded. */
236
    unsigned char *macroblock_coding;
237

    
238
    uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
239
    int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
240

    
241
    /* Huffman decode */
242
    int hti;
243
    unsigned int hbits;
244
    int entries;
245
    int huff_code_size;
246
    uint32_t huffman_table[80][32][2];
247

    
248
    uint8_t filter_limit_values[64];
249
    DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
250
} Vp3DecodeContext;
251

    
252
/************************************************************************
253
 * VP3 specific functions
254
 ************************************************************************/
255

    
256
/*
257
 * This function sets up all of the various blocks mappings:
258
 * superblocks <-> fragments, macroblocks <-> fragments,
259
 * superblocks <-> macroblocks
260
 *
261
 * Returns 0 is successful; returns 1 if *anything* went wrong.
262
 */
263
static int init_block_mapping(Vp3DecodeContext *s)
264
{
265
    int sb_x, sb_y, plane;
266
    int x, y, i, j = 0;
267

    
268
    for (plane = 0; plane < 3; plane++) {
269
        int sb_width    = plane ? s->c_superblock_width  : s->y_superblock_width;
270
        int sb_height   = plane ? s->c_superblock_height : s->y_superblock_height;
271
        int frag_width  = s->fragment_width[!!plane];
272
        int frag_height = s->fragment_height[!!plane];
273

    
274
        for (sb_y = 0; sb_y < sb_height; sb_y++)
275
            for (sb_x = 0; sb_x < sb_width; sb_x++)
276
                for (i = 0; i < 16; i++) {
277
                    x = 4*sb_x + hilbert_offset[i][0];
278
                    y = 4*sb_y + hilbert_offset[i][1];
279

    
280
                    if (x < frag_width && y < frag_height)
281
                        s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
282
                    else
283
                        s->superblock_fragments[j++] = -1;
284
                }
285
    }
286

    
287
    return 0;  /* successful path out */
288
}
289

    
290
/*
291
 * This function sets up the dequantization tables used for a particular
292
 * frame.
293
 */
294
static void init_dequantizer(Vp3DecodeContext *s, int qpi)
295
{
296
    int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
297
    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
298
    int i, plane, inter, qri, bmi, bmj, qistart;
299

    
300
    for(inter=0; inter<2; inter++){
301
        for(plane=0; plane<3; plane++){
302
            int sum=0;
303
            for(qri=0; qri<s->qr_count[inter][plane]; qri++){
304
                sum+= s->qr_size[inter][plane][qri];
305
                if(s->qps[qpi] <= sum)
306
                    break;
307
            }
308
            qistart= sum - s->qr_size[inter][plane][qri];
309
            bmi= s->qr_base[inter][plane][qri  ];
310
            bmj= s->qr_base[inter][plane][qri+1];
311
            for(i=0; i<64; i++){
312
                int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
313
                            - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
314
                            + s->qr_size[inter][plane][qri])
315
                           / (2*s->qr_size[inter][plane][qri]);
316

    
317
                int qmin= 8<<(inter + !i);
318
                int qscale= i ? ac_scale_factor : dc_scale_factor;
319

    
320
                s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
321
            }
322
            // all DC coefficients use the same quant so as not to interfere with DC prediction
323
            s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
324
        }
325
    }
326

    
327
    memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
328
}
329

    
330
/*
331
 * This function initializes the loop filter boundary limits if the frame's
332
 * quality index is different from the previous frame's.
333
 *
334
 * The filter_limit_values may not be larger than 127.
335
 */
336
static void init_loop_filter(Vp3DecodeContext *s)
337
{
338
    int *bounding_values= s->bounding_values_array+127;
339
    int filter_limit;
340
    int x;
341
    int value;
342

    
343
    filter_limit = s->filter_limit_values[s->qps[0]];
344

    
345
    /* set up the bounding values */
346
    memset(s->bounding_values_array, 0, 256 * sizeof(int));
347
    for (x = 0; x < filter_limit; x++) {
348
        bounding_values[-x] = -x;
349
        bounding_values[x] = x;
350
    }
351
    for (x = value = filter_limit; x < 128 && value; x++, value--) {
352
        bounding_values[ x] =  value;
353
        bounding_values[-x] = -value;
354
    }
355
    if (value)
356
        bounding_values[128] = value;
357
    bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
358
}
359

    
360
/*
361
 * This function unpacks all of the superblock/macroblock/fragment coding
362
 * information from the bitstream.
363
 */
364
static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
365
{
366
    int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
367
    int bit = 0;
368
    int current_superblock = 0;
369
    int current_run = 0;
370
    int num_partial_superblocks = 0;
371

    
372
    int i, j;
373
    int current_fragment;
374
    int plane;
375

    
376
    if (s->keyframe) {
377
        memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
378

    
379
    } else {
380

    
381
        /* unpack the list of partially-coded superblocks */
382
        bit = get_bits1(gb) ^ 1;
383
        current_run = 0;
384

    
385
        while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
386
            if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
387
                bit = get_bits1(gb);
388
            else
389
                bit ^= 1;
390

    
391
                current_run = get_vlc2(gb,
392
                    s->superblock_run_length_vlc.table, 6, 2) + 1;
393
                if (current_run == 34)
394
                    current_run += get_bits(gb, 12);
395

    
396
            if (current_superblock + current_run > s->superblock_count) {
397
                av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
398
                return -1;
399
            }
400

    
401
            memset(s->superblock_coding + current_superblock, bit, current_run);
402

    
403
            current_superblock += current_run;
404
            if (bit)
405
                num_partial_superblocks += current_run;
406
        }
407

    
408
        /* unpack the list of fully coded superblocks if any of the blocks were
409
         * not marked as partially coded in the previous step */
410
        if (num_partial_superblocks < s->superblock_count) {
411
            int superblocks_decoded = 0;
412

    
413
            current_superblock = 0;
414
            bit = get_bits1(gb) ^ 1;
415
            current_run = 0;
416

    
417
            while (superblocks_decoded < s->superblock_count - num_partial_superblocks
418
                   && get_bits_left(gb) > 0) {
419

    
420
                if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
421
                    bit = get_bits1(gb);
422
                else
423
                    bit ^= 1;
424

    
425
                        current_run = get_vlc2(gb,
426
                            s->superblock_run_length_vlc.table, 6, 2) + 1;
427
                        if (current_run == 34)
428
                            current_run += get_bits(gb, 12);
429

    
430
                for (j = 0; j < current_run; current_superblock++) {
431
                    if (current_superblock >= s->superblock_count) {
432
                        av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
433
                        return -1;
434
                    }
435

    
436
                /* skip any superblocks already marked as partially coded */
437
                if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
438
                    s->superblock_coding[current_superblock] = 2*bit;
439
                    j++;
440
                }
441
                }
442
                superblocks_decoded += current_run;
443
            }
444
        }
445

    
446
        /* if there were partial blocks, initialize bitstream for
447
         * unpacking fragment codings */
448
        if (num_partial_superblocks) {
449

    
450
            current_run = 0;
451
            bit = get_bits1(gb);
452
            /* toggle the bit because as soon as the first run length is
453
             * fetched the bit will be toggled again */
454
            bit ^= 1;
455
        }
456
    }
457

    
458
    /* figure out which fragments are coded; iterate through each
459
     * superblock (all planes) */
460
    s->total_num_coded_frags = 0;
461
    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
462

    
463
    for (plane = 0; plane < 3; plane++) {
464
        int sb_start = superblock_starts[plane];
465
        int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
466
        int num_coded_frags = 0;
467

    
468
    for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
469

    
470
        /* iterate through all 16 fragments in a superblock */
471
        for (j = 0; j < 16; j++) {
472

    
473
            /* if the fragment is in bounds, check its coding status */
474
            current_fragment = s->superblock_fragments[i * 16 + j];
475
            if (current_fragment != -1) {
476
                int coded = s->superblock_coding[i];
477

    
478
                if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
479

    
480
                    /* fragment may or may not be coded; this is the case
481
                     * that cares about the fragment coding runs */
482
                    if (current_run-- == 0) {
483
                        bit ^= 1;
484
                        current_run = get_vlc2(gb,
485
                            s->fragment_run_length_vlc.table, 5, 2);
486
                    }
487
                    coded = bit;
488
                }
489

    
490
                    if (coded) {
491
                        /* default mode; actual mode will be decoded in
492
                         * the next phase */
493
                        s->all_fragments[current_fragment].coding_method =
494
                            MODE_INTER_NO_MV;
495
                        s->coded_fragment_list[plane][num_coded_frags++] =
496
                            current_fragment;
497
                    } else {
498
                        /* not coded; copy this fragment from the prior frame */
499
                        s->all_fragments[current_fragment].coding_method =
500
                            MODE_COPY;
501
                    }
502
            }
503
        }
504
    }
505
        s->total_num_coded_frags += num_coded_frags;
506
        for (i = 0; i < 64; i++)
507
            s->num_coded_frags[plane][i] = num_coded_frags;
508
        if (plane < 2)
509
            s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
510
    }
511
    return 0;
512
}
513

    
514
/*
515
 * This function unpacks all the coding mode data for individual macroblocks
516
 * from the bitstream.
517
 */
518
static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
519
{
520
    int i, j, k, sb_x, sb_y;
521
    int scheme;
522
    int current_macroblock;
523
    int current_fragment;
524
    int coding_mode;
525
    int custom_mode_alphabet[CODING_MODE_COUNT];
526
    const int *alphabet;
527
    Vp3Fragment *frag;
528

    
529
    if (s->keyframe) {
530
        for (i = 0; i < s->fragment_count; i++)
531
            s->all_fragments[i].coding_method = MODE_INTRA;
532

    
533
    } else {
534

    
535
        /* fetch the mode coding scheme for this frame */
536
        scheme = get_bits(gb, 3);
537

    
538
        /* is it a custom coding scheme? */
539
        if (scheme == 0) {
540
            for (i = 0; i < 8; i++)
541
                custom_mode_alphabet[i] = MODE_INTER_NO_MV;
542
            for (i = 0; i < 8; i++)
543
                custom_mode_alphabet[get_bits(gb, 3)] = i;
544
            alphabet = custom_mode_alphabet;
545
        } else
546
            alphabet = ModeAlphabet[scheme-1];
547

    
548
        /* iterate through all of the macroblocks that contain 1 or more
549
         * coded fragments */
550
        for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
551
            for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
552
                if (get_bits_left(gb) <= 0)
553
                    return -1;
554

    
555
            for (j = 0; j < 4; j++) {
556
                int mb_x = 2*sb_x +   (j>>1);
557
                int mb_y = 2*sb_y + (((j>>1)+j)&1);
558
                current_macroblock = mb_y * s->macroblock_width + mb_x;
559

    
560
                if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
561
                    continue;
562

    
563
#define BLOCK_X (2*mb_x + (k&1))
564
#define BLOCK_Y (2*mb_y + (k>>1))
565
                /* coding modes are only stored if the macroblock has at least one
566
                 * luma block coded, otherwise it must be INTER_NO_MV */
567
                for (k = 0; k < 4; k++) {
568
                    current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
569
                    if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
570
                        break;
571
                }
572
                if (k == 4) {
573
                    s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
574
                    continue;
575
                }
576

    
577
                /* mode 7 means get 3 bits for each coding mode */
578
                if (scheme == 7)
579
                    coding_mode = get_bits(gb, 3);
580
                else
581
                    coding_mode = alphabet
582
                        [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
583

    
584
                s->macroblock_coding[current_macroblock] = coding_mode;
585
                for (k = 0; k < 4; k++) {
586
                    frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
587
                    if (frag->coding_method != MODE_COPY)
588
                        frag->coding_method = coding_mode;
589
                }
590

    
591
#define SET_CHROMA_MODES \
592
    if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
593
        frag[s->fragment_start[1]].coding_method = coding_mode;\
594
    if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
595
        frag[s->fragment_start[2]].coding_method = coding_mode;
596

    
597
                if (s->chroma_y_shift) {
598
                    frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
599
                    SET_CHROMA_MODES
600
                } else if (s->chroma_x_shift) {
601
                    frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
602
                    for (k = 0; k < 2; k++) {
603
                        SET_CHROMA_MODES
604
                        frag += s->fragment_width[1];
605
                    }
606
                } else {
607
                    for (k = 0; k < 4; k++) {
608
                        frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
609
                        SET_CHROMA_MODES
610
                    }
611
                }
612
            }
613
            }
614
        }
615
    }
616

    
617
    return 0;
618
}
619

    
620
/*
621
 * This function unpacks all the motion vectors for the individual
622
 * macroblocks from the bitstream.
623
 */
624
static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
625
{
626
    int j, k, sb_x, sb_y;
627
    int coding_mode;
628
    int motion_x[4];
629
    int motion_y[4];
630
    int last_motion_x = 0;
631
    int last_motion_y = 0;
632
    int prior_last_motion_x = 0;
633
    int prior_last_motion_y = 0;
634
    int current_macroblock;
635
    int current_fragment;
636
    int frag;
637

    
638
    if (s->keyframe)
639
        return 0;
640

    
641
    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
642
    coding_mode = get_bits1(gb);
643

    
644
    /* iterate through all of the macroblocks that contain 1 or more
645
     * coded fragments */
646
    for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
647
        for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
648
            if (get_bits_left(gb) <= 0)
649
                return -1;
650

    
651
        for (j = 0; j < 4; j++) {
652
            int mb_x = 2*sb_x +   (j>>1);
653
            int mb_y = 2*sb_y + (((j>>1)+j)&1);
654
            current_macroblock = mb_y * s->macroblock_width + mb_x;
655

    
656
            if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
657
                (s->macroblock_coding[current_macroblock] == MODE_COPY))
658
                continue;
659

    
660
            switch (s->macroblock_coding[current_macroblock]) {
661

    
662
            case MODE_INTER_PLUS_MV:
663
            case MODE_GOLDEN_MV:
664
                /* all 6 fragments use the same motion vector */
665
                if (coding_mode == 0) {
666
                    motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
667
                    motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
668
                } else {
669
                    motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
670
                    motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
671
                }
672

    
673
                /* vector maintenance, only on MODE_INTER_PLUS_MV */
674
                if (s->macroblock_coding[current_macroblock] ==
675
                    MODE_INTER_PLUS_MV) {
676
                    prior_last_motion_x = last_motion_x;
677
                    prior_last_motion_y = last_motion_y;
678
                    last_motion_x = motion_x[0];
679
                    last_motion_y = motion_y[0];
680
                }
681
                break;
682

    
683
            case MODE_INTER_FOURMV:
684
                /* vector maintenance */
685
                prior_last_motion_x = last_motion_x;
686
                prior_last_motion_y = last_motion_y;
687

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

    
709
            case MODE_INTER_LAST_MV:
710
                /* all 6 fragments use the last motion vector */
711
                motion_x[0] = last_motion_x;
712
                motion_y[0] = last_motion_y;
713

    
714
                /* no vector maintenance (last vector remains the
715
                 * last vector) */
716
                break;
717

    
718
            case MODE_INTER_PRIOR_LAST:
719
                /* all 6 fragments use the motion vector prior to the
720
                 * last motion vector */
721
                motion_x[0] = prior_last_motion_x;
722
                motion_y[0] = prior_last_motion_y;
723

    
724
                /* vector maintenance */
725
                prior_last_motion_x = last_motion_x;
726
                prior_last_motion_y = last_motion_y;
727
                last_motion_x = motion_x[0];
728
                last_motion_y = motion_y[0];
729
                break;
730

    
731
            default:
732
                /* covers intra, inter without MV, golden without MV */
733
                motion_x[0] = 0;
734
                motion_y[0] = 0;
735

    
736
                /* no vector maintenance */
737
                break;
738
            }
739

    
740
            /* assign the motion vectors to the correct fragments */
741
            for (k = 0; k < 4; k++) {
742
                current_fragment =
743
                    BLOCK_Y*s->fragment_width[0] + BLOCK_X;
744
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
745
                    s->motion_val[0][current_fragment][0] = motion_x[k];
746
                    s->motion_val[0][current_fragment][1] = motion_y[k];
747
                } else {
748
                    s->motion_val[0][current_fragment][0] = motion_x[0];
749
                    s->motion_val[0][current_fragment][1] = motion_y[0];
750
                }
751
            }
752

    
753
            if (s->chroma_y_shift) {
754
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
755
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
756
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
757
                }
758
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
759
                motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
760
                frag = mb_y*s->fragment_width[1] + mb_x;
761
                s->motion_val[1][frag][0] = motion_x[0];
762
                s->motion_val[1][frag][1] = motion_y[0];
763
            } else if (s->chroma_x_shift) {
764
                if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
765
                    motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
766
                    motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
767
                    motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
768
                    motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
769
                } else {
770
                    motion_x[1] = motion_x[0];
771
                    motion_y[1] = motion_y[0];
772
                }
773
                motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
774
                motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
775

    
776
                frag = 2*mb_y*s->fragment_width[1] + mb_x;
777
                for (k = 0; k < 2; k++) {
778
                    s->motion_val[1][frag][0] = motion_x[k];
779
                    s->motion_val[1][frag][1] = motion_y[k];
780
                    frag += s->fragment_width[1];
781
                }
782
            } else {
783
                for (k = 0; k < 4; k++) {
784
                    frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
785
                    if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
786
                        s->motion_val[1][frag][0] = motion_x[k];
787
                        s->motion_val[1][frag][1] = motion_y[k];
788
                    } else {
789
                        s->motion_val[1][frag][0] = motion_x[0];
790
                        s->motion_val[1][frag][1] = motion_y[0];
791
                    }
792
                }
793
            }
794
        }
795
        }
796
    }
797

    
798
    return 0;
799
}
800

    
801
static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
802
{
803
    int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
804
    int num_blocks = s->total_num_coded_frags;
805

    
806
    for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
807
        i = blocks_decoded = num_blocks_at_qpi = 0;
808

    
809
        bit = get_bits1(gb) ^ 1;
810
        run_length = 0;
811

    
812
        do {
813
            if (run_length == MAXIMUM_LONG_BIT_RUN)
814
                bit = get_bits1(gb);
815
            else
816
                bit ^= 1;
817

    
818
            run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
819
            if (run_length == 34)
820
                run_length += get_bits(gb, 12);
821
            blocks_decoded += run_length;
822

    
823
            if (!bit)
824
                num_blocks_at_qpi += run_length;
825

    
826
            for (j = 0; j < run_length; i++) {
827
                if (i >= s->total_num_coded_frags)
828
                    return -1;
829

    
830
                if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
831
                    s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
832
                    j++;
833
                }
834
            }
835
        } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
836

    
837
        num_blocks -= num_blocks_at_qpi;
838
    }
839

    
840
    return 0;
841
}
842

    
843
/*
844
 * This function is called by unpack_dct_coeffs() to extract the VLCs from
845
 * the bitstream. The VLCs encode tokens which are used to unpack DCT
846
 * data. This function unpacks all the VLCs for either the Y plane or both
847
 * C planes, and is called for DC coefficients or different AC coefficient
848
 * levels (since different coefficient types require different VLC tables.
849
 *
850
 * This function returns a residual eob run. E.g, if a particular token gave
851
 * instructions to EOB the next 5 fragments and there were only 2 fragments
852
 * left in the current fragment range, 3 would be returned so that it could
853
 * be passed into the next call to this same function.
854
 */
855
static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
856
                        VLC *table, int coeff_index,
857
                        int plane,
858
                        int eob_run)
859
{
860
    int i, j = 0;
861
    int token;
862
    int zero_run = 0;
863
    DCTELEM coeff = 0;
864
    int bits_to_get;
865
    int blocks_ended;
866
    int coeff_i = 0;
867
    int num_coeffs = s->num_coded_frags[plane][coeff_index];
868
    int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
869

    
870
    /* local references to structure members to avoid repeated deferences */
871
    int *coded_fragment_list = s->coded_fragment_list[plane];
872
    Vp3Fragment *all_fragments = s->all_fragments;
873
    VLC_TYPE (*vlc_table)[2] = table->table;
874

    
875
    if (num_coeffs < 0)
876
        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
877

    
878
    if (eob_run > num_coeffs) {
879
        coeff_i = blocks_ended = num_coeffs;
880
        eob_run -= num_coeffs;
881
    } else {
882
        coeff_i = blocks_ended = eob_run;
883
        eob_run = 0;
884
    }
885

    
886
    // insert fake EOB token to cover the split between planes or zzi
887
    if (blocks_ended)
888
        dct_tokens[j++] = blocks_ended << 2;
889

    
890
    while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
891
            /* decode a VLC into a token */
892
            token = get_vlc2(gb, vlc_table, 11, 3);
893
            /* use the token to get a zero run, a coefficient, and an eob run */
894
            if (token <= 6) {
895
                eob_run = eob_run_base[token];
896
                if (eob_run_get_bits[token])
897
                    eob_run += get_bits(gb, eob_run_get_bits[token]);
898

    
899
                // record only the number of blocks ended in this plane,
900
                // any spill will be recorded in the next plane.
901
                if (eob_run > num_coeffs - coeff_i) {
902
                    dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
903
                    blocks_ended   += num_coeffs - coeff_i;
904
                    eob_run        -= num_coeffs - coeff_i;
905
                    coeff_i         = num_coeffs;
906
                } else {
907
                    dct_tokens[j++] = TOKEN_EOB(eob_run);
908
                    blocks_ended   += eob_run;
909
                    coeff_i        += eob_run;
910
                    eob_run = 0;
911
                }
912
            } else {
913
                bits_to_get = coeff_get_bits[token];
914
                if (bits_to_get)
915
                    bits_to_get = get_bits(gb, bits_to_get);
916
                coeff = coeff_tables[token][bits_to_get];
917

    
918
                zero_run = zero_run_base[token];
919
                if (zero_run_get_bits[token])
920
                    zero_run += get_bits(gb, zero_run_get_bits[token]);
921

    
922
                if (zero_run) {
923
                    dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
924
                } else {
925
                    // Save DC into the fragment structure. DC prediction is
926
                    // done in raster order, so the actual DC can't be in with
927
                    // other tokens. We still need the token in dct_tokens[]
928
                    // however, or else the structure collapses on itself.
929
                    if (!coeff_index)
930
                        all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
931

    
932
                    dct_tokens[j++] = TOKEN_COEFF(coeff);
933
                }
934

    
935
                if (coeff_index + zero_run > 64) {
936
                    av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
937
                           " %d coeffs left\n", zero_run, 64-coeff_index);
938
                    zero_run = 64 - coeff_index;
939
                }
940

    
941
                // zero runs code multiple coefficients,
942
                // so don't try to decode coeffs for those higher levels
943
                for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
944
                    s->num_coded_frags[plane][i]--;
945
                coeff_i++;
946
            }
947
    }
948

    
949
    if (blocks_ended > s->num_coded_frags[plane][coeff_index])
950
        av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
951

    
952
    // decrement the number of blocks that have higher coeffecients for each
953
    // EOB run at this level
954
    if (blocks_ended)
955
        for (i = coeff_index+1; i < 64; i++)
956
            s->num_coded_frags[plane][i] -= blocks_ended;
957

    
958
    // setup the next buffer
959
    if (plane < 2)
960
        s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
961
    else if (coeff_index < 63)
962
        s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
963

    
964
    return eob_run;
965
}
966

    
967
static void reverse_dc_prediction(Vp3DecodeContext *s,
968
                                  int first_fragment,
969
                                  int fragment_width,
970
                                  int fragment_height);
971
/*
972
 * This function unpacks all of the DCT coefficient data from the
973
 * bitstream.
974
 */
975
static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
976
{
977
    int i;
978
    int dc_y_table;
979
    int dc_c_table;
980
    int ac_y_table;
981
    int ac_c_table;
982
    int residual_eob_run = 0;
983
    VLC *y_tables[64];
984
    VLC *c_tables[64];
985

    
986
    s->dct_tokens[0][0] = s->dct_tokens_base;
987

    
988
    /* fetch the DC table indexes */
989
    dc_y_table = get_bits(gb, 4);
990
    dc_c_table = get_bits(gb, 4);
991

    
992
    /* unpack the Y plane DC coefficients */
993
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
994
        0, residual_eob_run);
995

    
996
    /* reverse prediction of the Y-plane DC coefficients */
997
    reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
998

    
999
    /* unpack the C plane DC coefficients */
1000
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1001
        1, residual_eob_run);
1002
    residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
1003
        2, residual_eob_run);
1004

    
1005
    /* reverse prediction of the C-plane DC coefficients */
1006
    if (!(s->avctx->flags & CODEC_FLAG_GRAY))
1007
    {
1008
        reverse_dc_prediction(s, s->fragment_start[1],
1009
            s->fragment_width[1], s->fragment_height[1]);
1010
        reverse_dc_prediction(s, s->fragment_start[2],
1011
            s->fragment_width[1], s->fragment_height[1]);
1012
    }
1013

    
1014
    /* fetch the AC table indexes */
1015
    ac_y_table = get_bits(gb, 4);
1016
    ac_c_table = get_bits(gb, 4);
1017

    
1018
    /* build tables of AC VLC tables */
1019
    for (i = 1; i <= 5; i++) {
1020
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
1021
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
1022
    }
1023
    for (i = 6; i <= 14; i++) {
1024
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
1025
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
1026
    }
1027
    for (i = 15; i <= 27; i++) {
1028
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
1029
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
1030
    }
1031
    for (i = 28; i <= 63; i++) {
1032
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
1033
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
1034
    }
1035

    
1036
    /* decode all AC coefficents */
1037
    for (i = 1; i <= 63; i++) {
1038
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
1039
                0, residual_eob_run);
1040

    
1041
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1042
                1, residual_eob_run);
1043
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
1044
                2, residual_eob_run);
1045
    }
1046

    
1047
    return 0;
1048
}
1049

    
1050
/*
1051
 * This function reverses the DC prediction for each coded fragment in
1052
 * the frame. Much of this function is adapted directly from the original
1053
 * VP3 source code.
1054
 */
1055
#define COMPATIBLE_FRAME(x) \
1056
  (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
1057
#define DC_COEFF(u) s->all_fragments[u].dc
1058

    
1059
static void reverse_dc_prediction(Vp3DecodeContext *s,
1060
                                  int first_fragment,
1061
                                  int fragment_width,
1062
                                  int fragment_height)
1063
{
1064

    
1065
#define PUL 8
1066
#define PU 4
1067
#define PUR 2
1068
#define PL 1
1069

    
1070
    int x, y;
1071
    int i = first_fragment;
1072

    
1073
    int predicted_dc;
1074

    
1075
    /* DC values for the left, up-left, up, and up-right fragments */
1076
    int vl, vul, vu, vur;
1077

    
1078
    /* indexes for the left, up-left, up, and up-right fragments */
1079
    int l, ul, u, ur;
1080

    
1081
    /*
1082
     * The 6 fields mean:
1083
     *   0: up-left multiplier
1084
     *   1: up multiplier
1085
     *   2: up-right multiplier
1086
     *   3: left multiplier
1087
     */
1088
    static const int predictor_transform[16][4] = {
1089
        {  0,  0,  0,  0},
1090
        {  0,  0,  0,128},        // PL
1091
        {  0,  0,128,  0},        // PUR
1092
        {  0,  0, 53, 75},        // PUR|PL
1093
        {  0,128,  0,  0},        // PU
1094
        {  0, 64,  0, 64},        // PU|PL
1095
        {  0,128,  0,  0},        // PU|PUR
1096
        {  0,  0, 53, 75},        // PU|PUR|PL
1097
        {128,  0,  0,  0},        // PUL
1098
        {  0,  0,  0,128},        // PUL|PL
1099
        { 64,  0, 64,  0},        // PUL|PUR
1100
        {  0,  0, 53, 75},        // PUL|PUR|PL
1101
        {  0,128,  0,  0},        // PUL|PU
1102
       {-104,116,  0,116},        // PUL|PU|PL
1103
        { 24, 80, 24,  0},        // PUL|PU|PUR
1104
       {-104,116,  0,116}         // PUL|PU|PUR|PL
1105
    };
1106

    
1107
    /* This table shows which types of blocks can use other blocks for
1108
     * prediction. For example, INTRA is the only mode in this table to
1109
     * have a frame number of 0. That means INTRA blocks can only predict
1110
     * from other INTRA blocks. There are 2 golden frame coding types;
1111
     * blocks encoding in these modes can only predict from other blocks
1112
     * that were encoded with these 1 of these 2 modes. */
1113
    static const unsigned char compatible_frame[9] = {
1114
        1,    /* MODE_INTER_NO_MV */
1115
        0,    /* MODE_INTRA */
1116
        1,    /* MODE_INTER_PLUS_MV */
1117
        1,    /* MODE_INTER_LAST_MV */
1118
        1,    /* MODE_INTER_PRIOR_MV */
1119
        2,    /* MODE_USING_GOLDEN */
1120
        2,    /* MODE_GOLDEN_MV */
1121
        1,    /* MODE_INTER_FOUR_MV */
1122
        3     /* MODE_COPY */
1123
    };
1124
    int current_frame_type;
1125

    
1126
    /* there is a last DC predictor for each of the 3 frame types */
1127
    short last_dc[3];
1128

    
1129
    int transform = 0;
1130

    
1131
    vul = vu = vur = vl = 0;
1132
    last_dc[0] = last_dc[1] = last_dc[2] = 0;
1133

    
1134
    /* for each fragment row... */
1135
    for (y = 0; y < fragment_height; y++) {
1136

    
1137
        /* for each fragment in a row... */
1138
        for (x = 0; x < fragment_width; x++, i++) {
1139

    
1140
            /* reverse prediction if this block was coded */
1141
            if (s->all_fragments[i].coding_method != MODE_COPY) {
1142

    
1143
                current_frame_type =
1144
                    compatible_frame[s->all_fragments[i].coding_method];
1145

    
1146
                transform= 0;
1147
                if(x){
1148
                    l= i-1;
1149
                    vl = DC_COEFF(l);
1150
                    if(COMPATIBLE_FRAME(l))
1151
                        transform |= PL;
1152
                }
1153
                if(y){
1154
                    u= i-fragment_width;
1155
                    vu = DC_COEFF(u);
1156
                    if(COMPATIBLE_FRAME(u))
1157
                        transform |= PU;
1158
                    if(x){
1159
                        ul= i-fragment_width-1;
1160
                        vul = DC_COEFF(ul);
1161
                        if(COMPATIBLE_FRAME(ul))
1162
                            transform |= PUL;
1163
                    }
1164
                    if(x + 1 < fragment_width){
1165
                        ur= i-fragment_width+1;
1166
                        vur = DC_COEFF(ur);
1167
                        if(COMPATIBLE_FRAME(ur))
1168
                            transform |= PUR;
1169
                    }
1170
                }
1171

    
1172
                if (transform == 0) {
1173

    
1174
                    /* if there were no fragments to predict from, use last
1175
                     * DC saved */
1176
                    predicted_dc = last_dc[current_frame_type];
1177
                } else {
1178

    
1179
                    /* apply the appropriate predictor transform */
1180
                    predicted_dc =
1181
                        (predictor_transform[transform][0] * vul) +
1182
                        (predictor_transform[transform][1] * vu) +
1183
                        (predictor_transform[transform][2] * vur) +
1184
                        (predictor_transform[transform][3] * vl);
1185

    
1186
                    predicted_dc /= 128;
1187

    
1188
                    /* check for outranging on the [ul u l] and
1189
                     * [ul u ur l] predictors */
1190
                    if ((transform == 15) || (transform == 13)) {
1191
                        if (FFABS(predicted_dc - vu) > 128)
1192
                            predicted_dc = vu;
1193
                        else if (FFABS(predicted_dc - vl) > 128)
1194
                            predicted_dc = vl;
1195
                        else if (FFABS(predicted_dc - vul) > 128)
1196
                            predicted_dc = vul;
1197
                    }
1198
                }
1199

    
1200
                /* at long last, apply the predictor */
1201
                DC_COEFF(i) += predicted_dc;
1202
                /* save the DC */
1203
                last_dc[current_frame_type] = DC_COEFF(i);
1204
            }
1205
        }
1206
    }
1207
}
1208

    
1209
static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
1210
{
1211
    int x, y;
1212
    int *bounding_values= s->bounding_values_array+127;
1213

    
1214
    int width           = s->fragment_width[!!plane];
1215
    int height          = s->fragment_height[!!plane];
1216
    int fragment        = s->fragment_start        [plane] + ystart * width;
1217
    int stride          = s->current_frame.linesize[plane];
1218
    uint8_t *plane_data = s->current_frame.data    [plane];
1219
    if (!s->flipped_image) stride = -stride;
1220
    plane_data += s->data_offset[plane] + 8*ystart*stride;
1221

    
1222
    for (y = ystart; y < yend; y++) {
1223

    
1224
        for (x = 0; x < width; x++) {
1225
            /* This code basically just deblocks on the edges of coded blocks.
1226
             * However, it has to be much more complicated because of the
1227
             * braindamaged deblock ordering used in VP3/Theora. Order matters
1228
             * because some pixels get filtered twice. */
1229
            if( s->all_fragments[fragment].coding_method != MODE_COPY )
1230
            {
1231
                /* do not perform left edge filter for left columns frags */
1232
                if (x > 0) {
1233
                    s->dsp.vp3_h_loop_filter(
1234
                        plane_data + 8*x,
1235
                        stride, bounding_values);
1236
                }
1237

    
1238
                /* do not perform top edge filter for top row fragments */
1239
                if (y > 0) {
1240
                    s->dsp.vp3_v_loop_filter(
1241
                        plane_data + 8*x,
1242
                        stride, bounding_values);
1243
                }
1244

    
1245
                /* do not perform right edge filter for right column
1246
                 * fragments or if right fragment neighbor is also coded
1247
                 * in this frame (it will be filtered in next iteration) */
1248
                if ((x < width - 1) &&
1249
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
1250
                    s->dsp.vp3_h_loop_filter(
1251
                        plane_data + 8*x + 8,
1252
                        stride, bounding_values);
1253
                }
1254

    
1255
                /* do not perform bottom edge filter for bottom row
1256
                 * fragments or if bottom fragment neighbor is also coded
1257
                 * in this frame (it will be filtered in the next row) */
1258
                if ((y < height - 1) &&
1259
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
1260
                    s->dsp.vp3_v_loop_filter(
1261
                        plane_data + 8*x + 8*stride,
1262
                        stride, bounding_values);
1263
                }
1264
            }
1265

    
1266
            fragment++;
1267
        }
1268
        plane_data += 8*stride;
1269
    }
1270
}
1271

    
1272
/**
1273
 * Pulls DCT tokens from the 64 levels to decode and dequant the coefficients
1274
 * for the next block in coding order
1275
 */
1276
static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
1277
                              int plane, int inter, DCTELEM block[64])
1278
{
1279
    int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
1280
    uint8_t *perm = s->scantable.permutated;
1281
    int i = 0;
1282

    
1283
    do {
1284
        int token = *s->dct_tokens[plane][i];
1285
        switch (token & 3) {
1286
        case 0: // EOB
1287
            if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
1288
                s->dct_tokens[plane][i]++;
1289
            else
1290
                *s->dct_tokens[plane][i] = token & ~3;
1291
            goto end;
1292
        case 1: // zero run
1293
            s->dct_tokens[plane][i]++;
1294
            i += (token >> 2) & 0x7f;
1295
            block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
1296
            i++;
1297
            break;
1298
        case 2: // coeff
1299
            block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
1300
            s->dct_tokens[plane][i++]++;
1301
            break;
1302
        default: // shouldn't happen
1303
            return i;
1304
        }
1305
    } while (i < 64);
1306
end:
1307
    // the actual DC+prediction is in the fragment structure
1308
    block[0] = frag->dc * s->qmat[0][inter][plane][0];
1309
    return i;
1310
}
1311

    
1312
/**
1313
 * called when all pixels up to row y are complete
1314
 */
1315
static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
1316
{
1317
    int h, cy;
1318
    int offset[4];
1319

    
1320
    if(s->avctx->draw_horiz_band==NULL)
1321
        return;
1322

    
1323
    h= y - s->last_slice_end;
1324
    y -= h;
1325

    
1326
    if (!s->flipped_image) {
1327
        if (y == 0)
1328
            h -= s->height - s->avctx->height;  // account for non-mod16
1329
        y = s->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
    s->last_slice_end= y + h;
1341
}
1342

    
1343
/*
1344
 * Perform the final rendering for a particular slice of data.
1345
 * The slice number ranges from 0..(c_superblock_height - 1).
1346
 */
1347
static void render_slice(Vp3DecodeContext *s, int slice)
1348
{
1349
    int x, y, i, j;
1350
    LOCAL_ALIGNED_16(DCTELEM, block, [64]);
1351
    int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
1352
    int motion_halfpel_index;
1353
    uint8_t *motion_source;
1354
    int plane, first_pixel;
1355

    
1356
    if (slice >= s->c_superblock_height)
1357
        return;
1358

    
1359
    for (plane = 0; plane < 3; plane++) {
1360
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
1361
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
1362
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
1363
        int stride            = s->current_frame.linesize[plane];
1364
        int plane_width       = s->width  >> (plane && s->chroma_x_shift);
1365
        int plane_height      = s->height >> (plane && s->chroma_y_shift);
1366
        int8_t (*motion_val)[2] = s->motion_val[!!plane];
1367

    
1368
        int sb_x, sb_y        = slice << (!plane && s->chroma_y_shift);
1369
        int slice_height      = sb_y + 1 + (!plane && s->chroma_y_shift);
1370
        int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
1371

    
1372
        int fragment_width    = s->fragment_width[!!plane];
1373
        int fragment_height   = s->fragment_height[!!plane];
1374
        int fragment_start    = s->fragment_start[plane];
1375

    
1376
        if (!s->flipped_image) stride = -stride;
1377
        if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
1378
            continue;
1379

    
1380

    
1381
        if(FFABS(stride) > 2048)
1382
            return; //various tables are fixed size
1383

    
1384
        /* for each superblock row in the slice (both of them)... */
1385
        for (; sb_y < slice_height; sb_y++) {
1386

    
1387
            /* for each superblock in a row... */
1388
            for (sb_x = 0; sb_x < slice_width; sb_x++) {
1389

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

    
1395
                    i = fragment_start + y*fragment_width + x;
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[y*fragment_width + x][0];
1420
                        motion_y = motion_val[y*fragment_width + x][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
                            ff_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(64*slice + 64-16, 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
    static int counter = 0;
1726
    int i;
1727

    
1728
    init_get_bits(&gb, buf, buf_size * 8);
1729

    
1730
    if (s->theora && get_bits1(&gb))
1731
    {
1732
        av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
1733
        return -1;
1734
    }
1735

    
1736
    s->keyframe = !get_bits1(&gb);
1737
    if (!s->theora)
1738
        skip_bits(&gb, 1);
1739
    for (i = 0; i < 3; i++)
1740
        s->last_qps[i] = s->qps[i];
1741

    
1742
    s->nqps=0;
1743
    do{
1744
        s->qps[s->nqps++]= get_bits(&gb, 6);
1745
    } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
1746
    for (i = s->nqps; i < 3; i++)
1747
        s->qps[i] = -1;
1748

    
1749
    if (s->avctx->debug & FF_DEBUG_PICT_INFO)
1750
        av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
1751
            s->keyframe?"key":"", counter, s->qps[0]);
1752
    counter++;
1753

    
1754
    s->skip_loop_filter = !s->filter_limit_values[s->qps[0]] ||
1755
        avctx->skip_loop_filter >= (s->keyframe ? AVDISCARD_ALL : AVDISCARD_NONKEY);
1756

    
1757
    if (s->qps[0] != s->last_qps[0])
1758
        init_loop_filter(s);
1759

    
1760
    for (i = 0; i < s->nqps; i++)
1761
        // reinit all dequantizers if the first one changed, because
1762
        // the DC of the first quantizer must be used for all matrices
1763
        if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
1764
            init_dequantizer(s, i);
1765

    
1766
    if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
1767
        return buf_size;
1768

    
1769
    s->current_frame.reference = 3;
1770
    s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
1771
    if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
1772
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1773
        goto error;
1774
    }
1775

    
1776
    if (s->keyframe) {
1777
        if (!s->theora)
1778
        {
1779
            skip_bits(&gb, 4); /* width code */
1780
            skip_bits(&gb, 4); /* height code */
1781
            if (s->version)
1782
            {
1783
                s->version = get_bits(&gb, 5);
1784
                if (counter == 1)
1785
                    av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
1786
            }
1787
        }
1788
        if (s->version || s->theora)
1789
        {
1790
                if (get_bits1(&gb))
1791
                    av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
1792
            skip_bits(&gb, 2); /* reserved? */
1793
        }
1794
    } else {
1795
        if (!s->golden_frame.data[0]) {
1796
            av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
1797

    
1798
            s->golden_frame.reference = 3;
1799
            s->golden_frame.pict_type = FF_I_TYPE;
1800
            if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
1801
                av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1802
                goto error;
1803
            }
1804
            s->last_frame = s->golden_frame;
1805
            s->last_frame.type = FF_BUFFER_TYPE_COPY;
1806
        }
1807
    }
1808

    
1809
    s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
1810
    s->current_frame.qstride= 0;
1811

    
1812
    memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
1813

    
1814
    if (unpack_superblocks(s, &gb)){
1815
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
1816
        goto error;
1817
    }
1818
    if (unpack_modes(s, &gb)){
1819
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
1820
        goto error;
1821
    }
1822
    if (unpack_vectors(s, &gb)){
1823
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
1824
        goto error;
1825
    }
1826
    if (unpack_block_qpis(s, &gb)){
1827
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
1828
        goto error;
1829
    }
1830
    if (unpack_dct_coeffs(s, &gb)){
1831
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
1832
        goto error;
1833
    }
1834

    
1835
    for (i = 0; i < 3; i++) {
1836
        int height = s->height >> (i && s->chroma_y_shift);
1837
        if (s->flipped_image)
1838
            s->data_offset[i] = 0;
1839
        else
1840
            s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
1841
    }
1842

    
1843
    s->last_slice_end = 0;
1844
    for (i = 0; i < s->c_superblock_height; i++)
1845
        render_slice(s, i);
1846

    
1847
    // filter the last row
1848
    for (i = 0; i < 3; i++) {
1849
        int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
1850
        apply_loop_filter(s, i, row, row+1);
1851
    }
1852
    vp3_draw_horiz_band(s, s->height);
1853

    
1854
    *data_size=sizeof(AVFrame);
1855
    *(AVFrame*)data= s->current_frame;
1856

    
1857
    /* release the last frame, if it is allocated and if it is not the
1858
     * golden frame */
1859
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1860
        avctx->release_buffer(avctx, &s->last_frame);
1861

    
1862
    /* shuffle frames (last = current) */
1863
    s->last_frame= s->current_frame;
1864

    
1865
    if (s->keyframe) {
1866
        if (s->golden_frame.data[0])
1867
            avctx->release_buffer(avctx, &s->golden_frame);
1868
        s->golden_frame = s->current_frame;
1869
        s->last_frame.type = FF_BUFFER_TYPE_COPY;
1870
    }
1871

    
1872
    s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
1873

    
1874
    return buf_size;
1875

    
1876
error:
1877
    if (s->current_frame.data[0])
1878
        avctx->release_buffer(avctx, &s->current_frame);
1879
    return -1;
1880
}
1881

    
1882
/*
1883
 * This is the ffmpeg/libavcodec API module cleanup function.
1884
 */
1885
static av_cold int vp3_decode_end(AVCodecContext *avctx)
1886
{
1887
    Vp3DecodeContext *s = avctx->priv_data;
1888
    int i;
1889

    
1890
    av_free(s->superblock_coding);
1891
    av_free(s->all_fragments);
1892
    av_free(s->coded_fragment_list[0]);
1893
    av_free(s->dct_tokens_base);
1894
    av_free(s->superblock_fragments);
1895
    av_free(s->macroblock_coding);
1896
    av_free(s->motion_val[0]);
1897
    av_free(s->motion_val[1]);
1898

    
1899
    for (i = 0; i < 16; i++) {
1900
        free_vlc(&s->dc_vlc[i]);
1901
        free_vlc(&s->ac_vlc_1[i]);
1902
        free_vlc(&s->ac_vlc_2[i]);
1903
        free_vlc(&s->ac_vlc_3[i]);
1904
        free_vlc(&s->ac_vlc_4[i]);
1905
    }
1906

    
1907
    free_vlc(&s->superblock_run_length_vlc);
1908
    free_vlc(&s->fragment_run_length_vlc);
1909
    free_vlc(&s->mode_code_vlc);
1910
    free_vlc(&s->motion_vector_vlc);
1911

    
1912
    /* release all frames */
1913
    if (s->golden_frame.data[0])
1914
        avctx->release_buffer(avctx, &s->golden_frame);
1915
    if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
1916
        avctx->release_buffer(avctx, &s->last_frame);
1917
    /* no need to release the current_frame since it will always be pointing
1918
     * to the same frame as either the golden or last frame */
1919

    
1920
    return 0;
1921
}
1922

    
1923
static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
1924
{
1925
    Vp3DecodeContext *s = avctx->priv_data;
1926

    
1927
    if (get_bits1(gb)) {
1928
        int token;
1929
        if (s->entries >= 32) { /* overflow */
1930
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1931
            return -1;
1932
        }
1933
        token = get_bits(gb, 5);
1934
        //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);
1935
        s->huffman_table[s->hti][token][0] = s->hbits;
1936
        s->huffman_table[s->hti][token][1] = s->huff_code_size;
1937
        s->entries++;
1938
    }
1939
    else {
1940
        if (s->huff_code_size >= 32) {/* overflow */
1941
            av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
1942
            return -1;
1943
        }
1944
        s->huff_code_size++;
1945
        s->hbits <<= 1;
1946
        if (read_huffman_tree(avctx, gb))
1947
            return -1;
1948
        s->hbits |= 1;
1949
        if (read_huffman_tree(avctx, gb))
1950
            return -1;
1951
        s->hbits >>= 1;
1952
        s->huff_code_size--;
1953
    }
1954
    return 0;
1955
}
1956

    
1957
#if CONFIG_THEORA_DECODER
1958
static const enum PixelFormat theora_pix_fmts[4] = {
1959
    PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
1960
};
1961

    
1962
static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
1963
{
1964
    Vp3DecodeContext *s = avctx->priv_data;
1965
    int visible_width, visible_height, colorspace;
1966
    int offset_x = 0, offset_y = 0;
1967
    AVRational fps;
1968

    
1969
    s->theora = get_bits_long(gb, 24);
1970
    av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
1971

    
1972
    /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
1973
    /* but previous versions have the image flipped relative to vp3 */
1974
    if (s->theora < 0x030200)
1975
    {
1976
        s->flipped_image = 1;
1977
        av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
1978
    }
1979

    
1980
    visible_width  = s->width  = get_bits(gb, 16) << 4;
1981
    visible_height = s->height = get_bits(gb, 16) << 4;
1982

    
1983
    if(avcodec_check_dimensions(avctx, s->width, s->height)){
1984
        av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
1985
        s->width= s->height= 0;
1986
        return -1;
1987
    }
1988

    
1989
    if (s->theora >= 0x030200) {
1990
        visible_width  = get_bits_long(gb, 24);
1991
        visible_height = get_bits_long(gb, 24);
1992

    
1993
        offset_x = get_bits(gb, 8); /* offset x */
1994
        offset_y = get_bits(gb, 8); /* offset y, from bottom */
1995
    }
1996

    
1997
    fps.num = get_bits_long(gb, 32);
1998
    fps.den = get_bits_long(gb, 32);
1999
    if (fps.num && fps.den) {
2000
        av_reduce(&avctx->time_base.num, &avctx->time_base.den,
2001
                  fps.den, fps.num, 1<<30);
2002
    }
2003

    
2004
    avctx->sample_aspect_ratio.num = get_bits_long(gb, 24);
2005
    avctx->sample_aspect_ratio.den = get_bits_long(gb, 24);
2006

    
2007
    if (s->theora < 0x030200)
2008
        skip_bits(gb, 5); /* keyframe frequency force */
2009
    colorspace = get_bits(gb, 8);
2010
    skip_bits(gb, 24); /* bitrate */
2011

    
2012
    skip_bits(gb, 6); /* quality hint */
2013

    
2014
    if (s->theora >= 0x030200)
2015
    {
2016
        skip_bits(gb, 5); /* keyframe frequency force */
2017
        avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
2018
        skip_bits(gb, 3); /* reserved */
2019
    }
2020

    
2021
//    align_get_bits(gb);
2022

    
2023
    if (   visible_width  <= s->width  && visible_width  > s->width-16
2024
        && visible_height <= s->height && visible_height > s->height-16
2025
        && !offset_x && (offset_y == s->height - visible_height))
2026
        avcodec_set_dimensions(avctx, visible_width, visible_height);
2027
    else
2028
        avcodec_set_dimensions(avctx, s->width, s->height);
2029

    
2030
    if (colorspace == 1) {
2031
        avctx->color_primaries = AVCOL_PRI_BT470M;
2032
    } else if (colorspace == 2) {
2033
        avctx->color_primaries = AVCOL_PRI_BT470BG;
2034
    }
2035
    if (colorspace == 1 || colorspace == 2) {
2036
        avctx->colorspace = AVCOL_SPC_BT470BG;
2037
        avctx->color_trc  = AVCOL_TRC_BT709;
2038
    }
2039

    
2040
    return 0;
2041
}
2042

    
2043
static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
2044
{
2045
    Vp3DecodeContext *s = avctx->priv_data;
2046
    int i, n, matrices, inter, plane;
2047

    
2048
    if (s->theora >= 0x030200) {
2049
        n = get_bits(gb, 3);
2050
        /* loop filter limit values table */
2051
        for (i = 0; i < 64; i++) {
2052
            s->filter_limit_values[i] = get_bits(gb, n);
2053
            if (s->filter_limit_values[i] > 127) {
2054
                av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
2055
                s->filter_limit_values[i] = 127;
2056
            }
2057
        }
2058
    }
2059

    
2060
    if (s->theora >= 0x030200)
2061
        n = get_bits(gb, 4) + 1;
2062
    else
2063
        n = 16;
2064
    /* quality threshold table */
2065
    for (i = 0; i < 64; i++)
2066
        s->coded_ac_scale_factor[i] = get_bits(gb, n);
2067

    
2068
    if (s->theora >= 0x030200)
2069
        n = get_bits(gb, 4) + 1;
2070
    else
2071
        n = 16;
2072
    /* dc scale factor table */
2073
    for (i = 0; i < 64; i++)
2074
        s->coded_dc_scale_factor[i] = get_bits(gb, n);
2075

    
2076
    if (s->theora >= 0x030200)
2077
        matrices = get_bits(gb, 9) + 1;
2078
    else
2079
        matrices = 3;
2080

    
2081
    if(matrices > 384){
2082
        av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
2083
        return -1;
2084
    }
2085

    
2086
    for(n=0; n<matrices; n++){
2087
        for (i = 0; i < 64; i++)
2088
            s->base_matrix[n][i]= get_bits(gb, 8);
2089
    }
2090

    
2091
    for (inter = 0; inter <= 1; inter++) {
2092
        for (plane = 0; plane <= 2; plane++) {
2093
            int newqr= 1;
2094
            if (inter || plane > 0)
2095
                newqr = get_bits1(gb);
2096
            if (!newqr) {
2097
                int qtj, plj;
2098
                if(inter && get_bits1(gb)){
2099
                    qtj = 0;
2100
                    plj = plane;
2101
                }else{
2102
                    qtj= (3*inter + plane - 1) / 3;
2103
                    plj= (plane + 2) % 3;
2104
                }
2105
                s->qr_count[inter][plane]= s->qr_count[qtj][plj];
2106
                memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
2107
                memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
2108
            } else {
2109
                int qri= 0;
2110
                int qi = 0;
2111

    
2112
                for(;;){
2113
                    i= get_bits(gb, av_log2(matrices-1)+1);
2114
                    if(i>= matrices){
2115
                        av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
2116
                        return -1;
2117
                    }
2118
                    s->qr_base[inter][plane][qri]= i;
2119
                    if(qi >= 63)
2120
                        break;
2121
                    i = get_bits(gb, av_log2(63-qi)+1) + 1;
2122
                    s->qr_size[inter][plane][qri++]= i;
2123
                    qi += i;
2124
                }
2125

    
2126
                if (qi > 63) {
2127
                    av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
2128
                    return -1;
2129
                }
2130
                s->qr_count[inter][plane]= qri;
2131
            }
2132
        }
2133
    }
2134

    
2135
    /* Huffman tables */
2136
    for (s->hti = 0; s->hti < 80; s->hti++) {
2137
        s->entries = 0;
2138
        s->huff_code_size = 1;
2139
        if (!get_bits1(gb)) {
2140
            s->hbits = 0;
2141
            if(read_huffman_tree(avctx, gb))
2142
                return -1;
2143
            s->hbits = 1;
2144
            if(read_huffman_tree(avctx, gb))
2145
                return -1;
2146
        }
2147
    }
2148

    
2149
    s->theora_tables = 1;
2150

    
2151
    return 0;
2152
}
2153

    
2154
static av_cold int theora_decode_init(AVCodecContext *avctx)
2155
{
2156
    Vp3DecodeContext *s = avctx->priv_data;
2157
    GetBitContext gb;
2158
    int ptype;
2159
    uint8_t *header_start[3];
2160
    int header_len[3];
2161
    int i;
2162

    
2163
    s->theora = 1;
2164

    
2165
    if (!avctx->extradata_size)
2166
    {
2167
        av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
2168
        return -1;
2169
    }
2170

    
2171
    if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
2172
                              42, header_start, header_len) < 0) {
2173
        av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
2174
        return -1;
2175
    }
2176

    
2177
  for(i=0;i<3;i++) {
2178
    init_get_bits(&gb, header_start[i], header_len[i] * 8);
2179

    
2180
    ptype = get_bits(&gb, 8);
2181

    
2182
     if (!(ptype & 0x80))
2183
     {
2184
        av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
2185
//        return -1;
2186
     }
2187

    
2188
    // FIXME: Check for this as well.
2189
    skip_bits_long(&gb, 6*8); /* "theora" */
2190

    
2191
    switch(ptype)
2192
    {
2193
        case 0x80:
2194
            theora_decode_header(avctx, &gb);
2195
                break;
2196
        case 0x81:
2197
// FIXME: is this needed? it breaks sometimes
2198
//            theora_decode_comments(avctx, gb);
2199
            break;
2200
        case 0x82:
2201
            if (theora_decode_tables(avctx, &gb))
2202
                return -1;
2203
            break;
2204
        default:
2205
            av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
2206
            break;
2207
    }
2208
    if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
2209
        av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
2210
    if (s->theora < 0x030200)
2211
        break;
2212
  }
2213

    
2214
    return vp3_decode_init(avctx);
2215
}
2216

    
2217
AVCodec theora_decoder = {
2218
    "theora",
2219
    AVMEDIA_TYPE_VIDEO,
2220
    CODEC_ID_THEORA,
2221
    sizeof(Vp3DecodeContext),
2222
    theora_decode_init,
2223
    NULL,
2224
    vp3_decode_end,
2225
    vp3_decode_frame,
2226
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2227
    NULL,
2228
    .long_name = NULL_IF_CONFIG_SMALL("Theora"),
2229
};
2230
#endif
2231

    
2232
AVCodec vp3_decoder = {
2233
    "vp3",
2234
    AVMEDIA_TYPE_VIDEO,
2235
    CODEC_ID_VP3,
2236
    sizeof(Vp3DecodeContext),
2237
    vp3_decode_init,
2238
    NULL,
2239
    vp3_decode_end,
2240
    vp3_decode_frame,
2241
    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
2242
    NULL,
2243
    .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
2244
};